diff options
Diffstat (limited to 'drivers/infiniband')
45 files changed, 3808 insertions, 1244 deletions
diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig index 77089399359b..b899531498eb 100644 --- a/drivers/infiniband/Kconfig +++ b/drivers/infiniband/Kconfig @@ -38,6 +38,17 @@ config INFINIBAND_USER_MEM depends on INFINIBAND_USER_ACCESS != n default y +config INFINIBAND_ON_DEMAND_PAGING + bool "InfiniBand on-demand paging support" + depends on INFINIBAND_USER_MEM + select MMU_NOTIFIER + default y + ---help--- + On demand paging support for the InfiniBand subsystem. + Together with driver support this allows registration of + memory regions without pinning their pages, fetching the + pages on demand instead. + config INFINIBAND_ADDR_TRANS bool depends on INFINIBAND diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile index ffd0af6734af..acf736764445 100644 --- a/drivers/infiniband/core/Makefile +++ b/drivers/infiniband/core/Makefile @@ -11,6 +11,7 @@ obj-$(CONFIG_INFINIBAND_USER_ACCESS) += ib_uverbs.o ib_ucm.o \ ib_core-y := packer.o ud_header.o verbs.o sysfs.o \ device.o fmr_pool.o cache.o netlink.o ib_core-$(CONFIG_INFINIBAND_USER_MEM) += umem.o +ib_core-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += umem_odp.o umem_rbtree.o ib_mad-y := mad.o smi.o agent.o mad_rmpp.o diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c index 8172d37f9add..f80da50d84a5 100644 --- a/drivers/infiniband/core/addr.c +++ b/drivers/infiniband/core/addr.c @@ -176,8 +176,8 @@ static void set_timeout(unsigned long time) unsigned long delay; delay = time - jiffies; - if ((long)delay <= 0) - delay = 1; + if ((long)delay < 0) + delay = 0; mod_delayed_work(addr_wq, &work, delay); } diff --git a/drivers/infiniband/core/multicast.c b/drivers/infiniband/core/multicast.c index d2360a8ef0b2..fa17b552ff78 100644 --- a/drivers/infiniband/core/multicast.c +++ b/drivers/infiniband/core/multicast.c @@ -525,17 +525,22 @@ static void join_handler(int status, struct ib_sa_mcmember_rec *rec, if (status) process_join_error(group, status); else { + int mgids_changed, is_mgid0; ib_find_pkey(group->port->dev->device, group->port->port_num, be16_to_cpu(rec->pkey), &pkey_index); spin_lock_irq(&group->port->lock); - group->rec = *rec; if (group->state == MCAST_BUSY && group->pkey_index == MCAST_INVALID_PKEY_INDEX) group->pkey_index = pkey_index; - if (!memcmp(&mgid0, &group->rec.mgid, sizeof mgid0)) { + mgids_changed = memcmp(&rec->mgid, &group->rec.mgid, + sizeof(group->rec.mgid)); + group->rec = *rec; + if (mgids_changed) { rb_erase(&group->node, &group->port->table); - mcast_insert(group->port, group, 1); + is_mgid0 = !memcmp(&mgid0, &group->rec.mgid, + sizeof(mgid0)); + mcast_insert(group->port, group, is_mgid0); } spin_unlock_irq(&group->port->lock); } diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c index df0c4f605a21..aec7a6aa2951 100644 --- a/drivers/infiniband/core/umem.c +++ b/drivers/infiniband/core/umem.c @@ -39,6 +39,7 @@ #include <linux/hugetlb.h> #include <linux/dma-attrs.h> #include <linux/slab.h> +#include <rdma/ib_umem_odp.h> #include "uverbs.h" @@ -69,6 +70,10 @@ static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int d /** * ib_umem_get - Pin and DMA map userspace memory. + * + * If access flags indicate ODP memory, avoid pinning. Instead, stores + * the mm for future page fault handling in conjunction with MMU notifiers. + * * @context: userspace context to pin memory for * @addr: userspace virtual address to start at * @size: length of region to pin @@ -103,17 +108,30 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, umem->context = context; umem->length = size; - umem->offset = addr & ~PAGE_MASK; + umem->address = addr; umem->page_size = PAGE_SIZE; umem->pid = get_task_pid(current, PIDTYPE_PID); /* - * We ask for writable memory if any access flags other than - * "remote read" are set. "Local write" and "remote write" + * We ask for writable memory if any of the following + * access flags are set. "Local write" and "remote write" * obviously require write access. "Remote atomic" can do * things like fetch and add, which will modify memory, and * "MW bind" can change permissions by binding a window. */ - umem->writable = !!(access & ~IB_ACCESS_REMOTE_READ); + umem->writable = !!(access & + (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | + IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND)); + + if (access & IB_ACCESS_ON_DEMAND) { + ret = ib_umem_odp_get(context, umem); + if (ret) { + kfree(umem); + return ERR_PTR(ret); + } + return umem; + } + + umem->odp_data = NULL; /* We assume the memory is from hugetlb until proved otherwise */ umem->hugetlb = 1; @@ -132,7 +150,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, if (!vma_list) umem->hugetlb = 0; - npages = PAGE_ALIGN(size + umem->offset) >> PAGE_SHIFT; + npages = ib_umem_num_pages(umem); down_write(¤t->mm->mmap_sem); @@ -235,6 +253,11 @@ void ib_umem_release(struct ib_umem *umem) struct task_struct *task; unsigned long diff; + if (umem->odp_data) { + ib_umem_odp_release(umem); + return; + } + __ib_umem_release(umem->context->device, umem, 1); task = get_pid_task(umem->pid, PIDTYPE_PID); @@ -246,7 +269,7 @@ void ib_umem_release(struct ib_umem *umem) if (!mm) goto out; - diff = PAGE_ALIGN(umem->length + umem->offset) >> PAGE_SHIFT; + diff = ib_umem_num_pages(umem); /* * We may be called with the mm's mmap_sem already held. This @@ -283,6 +306,9 @@ int ib_umem_page_count(struct ib_umem *umem) int n; struct scatterlist *sg; + if (umem->odp_data) + return ib_umem_num_pages(umem); + shift = ilog2(umem->page_size); n = 0; @@ -292,3 +318,37 @@ int ib_umem_page_count(struct ib_umem *umem) return n; } EXPORT_SYMBOL(ib_umem_page_count); + +/* + * Copy from the given ib_umem's pages to the given buffer. + * + * umem - the umem to copy from + * offset - offset to start copying from + * dst - destination buffer + * length - buffer length + * + * Returns 0 on success, or an error code. + */ +int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset, + size_t length) +{ + size_t end = offset + length; + int ret; + + if (offset > umem->length || length > umem->length - offset) { + pr_err("ib_umem_copy_from not in range. offset: %zd umem length: %zd end: %zd\n", + offset, umem->length, end); + return -EINVAL; + } + + ret = sg_pcopy_to_buffer(umem->sg_head.sgl, umem->nmap, dst, length, + offset + ib_umem_offset(umem)); + + if (ret < 0) + return ret; + else if (ret != length) + return -EINVAL; + else + return 0; +} +EXPORT_SYMBOL(ib_umem_copy_from); diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c new file mode 100644 index 000000000000..6095872549e7 --- /dev/null +++ b/drivers/infiniband/core/umem_odp.c @@ -0,0 +1,668 @@ +/* + * Copyright (c) 2014 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/types.h> +#include <linux/sched.h> +#include <linux/pid.h> +#include <linux/slab.h> +#include <linux/export.h> +#include <linux/vmalloc.h> + +#include <rdma/ib_verbs.h> +#include <rdma/ib_umem.h> +#include <rdma/ib_umem_odp.h> + +static void ib_umem_notifier_start_account(struct ib_umem *item) +{ + mutex_lock(&item->odp_data->umem_mutex); + + /* Only update private counters for this umem if it has them. + * Otherwise skip it. All page faults will be delayed for this umem. */ + if (item->odp_data->mn_counters_active) { + int notifiers_count = item->odp_data->notifiers_count++; + + if (notifiers_count == 0) + /* Initialize the completion object for waiting on + * notifiers. Since notifier_count is zero, no one + * should be waiting right now. */ + reinit_completion(&item->odp_data->notifier_completion); + } + mutex_unlock(&item->odp_data->umem_mutex); +} + +static void ib_umem_notifier_end_account(struct ib_umem *item) +{ + mutex_lock(&item->odp_data->umem_mutex); + + /* Only update private counters for this umem if it has them. + * Otherwise skip it. All page faults will be delayed for this umem. */ + if (item->odp_data->mn_counters_active) { + /* + * This sequence increase will notify the QP page fault that + * the page that is going to be mapped in the spte could have + * been freed. + */ + ++item->odp_data->notifiers_seq; + if (--item->odp_data->notifiers_count == 0) + complete_all(&item->odp_data->notifier_completion); + } + mutex_unlock(&item->odp_data->umem_mutex); +} + +/* Account for a new mmu notifier in an ib_ucontext. */ +static void ib_ucontext_notifier_start_account(struct ib_ucontext *context) +{ + atomic_inc(&context->notifier_count); +} + +/* Account for a terminating mmu notifier in an ib_ucontext. + * + * Must be called with the ib_ucontext->umem_rwsem semaphore unlocked, since + * the function takes the semaphore itself. */ +static void ib_ucontext_notifier_end_account(struct ib_ucontext *context) +{ + int zero_notifiers = atomic_dec_and_test(&context->notifier_count); + + if (zero_notifiers && + !list_empty(&context->no_private_counters)) { + /* No currently running mmu notifiers. Now is the chance to + * add private accounting to all previously added umems. */ + struct ib_umem_odp *odp_data, *next; + + /* Prevent concurrent mmu notifiers from working on the + * no_private_counters list. */ + down_write(&context->umem_rwsem); + + /* Read the notifier_count again, with the umem_rwsem + * semaphore taken for write. */ + if (!atomic_read(&context->notifier_count)) { + list_for_each_entry_safe(odp_data, next, + &context->no_private_counters, + no_private_counters) { + mutex_lock(&odp_data->umem_mutex); + odp_data->mn_counters_active = true; + list_del(&odp_data->no_private_counters); + complete_all(&odp_data->notifier_completion); + mutex_unlock(&odp_data->umem_mutex); + } + } + + up_write(&context->umem_rwsem); + } +} + +static int ib_umem_notifier_release_trampoline(struct ib_umem *item, u64 start, + u64 end, void *cookie) { + /* + * Increase the number of notifiers running, to + * prevent any further fault handling on this MR. + */ + ib_umem_notifier_start_account(item); + item->odp_data->dying = 1; + /* Make sure that the fact the umem is dying is out before we release + * all pending page faults. */ + smp_wmb(); + complete_all(&item->odp_data->notifier_completion); + item->context->invalidate_range(item, ib_umem_start(item), + ib_umem_end(item)); + return 0; +} + +static void ib_umem_notifier_release(struct mmu_notifier *mn, + struct mm_struct *mm) +{ + struct ib_ucontext *context = container_of(mn, struct ib_ucontext, mn); + + if (!context->invalidate_range) + return; + + ib_ucontext_notifier_start_account(context); + down_read(&context->umem_rwsem); + rbt_ib_umem_for_each_in_range(&context->umem_tree, 0, + ULLONG_MAX, + ib_umem_notifier_release_trampoline, + NULL); + up_read(&context->umem_rwsem); +} + +static int invalidate_page_trampoline(struct ib_umem *item, u64 start, + u64 end, void *cookie) +{ + ib_umem_notifier_start_account(item); + item->context->invalidate_range(item, start, start + PAGE_SIZE); + ib_umem_notifier_end_account(item); + return 0; +} + +static void ib_umem_notifier_invalidate_page(struct mmu_notifier *mn, + struct mm_struct *mm, + unsigned long address) +{ + struct ib_ucontext *context = container_of(mn, struct ib_ucontext, mn); + + if (!context->invalidate_range) + return; + + ib_ucontext_notifier_start_account(context); + down_read(&context->umem_rwsem); + rbt_ib_umem_for_each_in_range(&context->umem_tree, address, + address + PAGE_SIZE, + invalidate_page_trampoline, NULL); + up_read(&context->umem_rwsem); + ib_ucontext_notifier_end_account(context); +} + +static int invalidate_range_start_trampoline(struct ib_umem *item, u64 start, + u64 end, void *cookie) +{ + ib_umem_notifier_start_account(item); + item->context->invalidate_range(item, start, end); + return 0; +} + +static void ib_umem_notifier_invalidate_range_start(struct mmu_notifier *mn, + struct mm_struct *mm, + unsigned long start, + unsigned long end) +{ + struct ib_ucontext *context = container_of(mn, struct ib_ucontext, mn); + + if (!context->invalidate_range) + return; + + ib_ucontext_notifier_start_account(context); + down_read(&context->umem_rwsem); + rbt_ib_umem_for_each_in_range(&context->umem_tree, start, + end, + invalidate_range_start_trampoline, NULL); + up_read(&context->umem_rwsem); +} + +static int invalidate_range_end_trampoline(struct ib_umem *item, u64 start, + u64 end, void *cookie) +{ + ib_umem_notifier_end_account(item); + return 0; +} + +static void ib_umem_notifier_invalidate_range_end(struct mmu_notifier *mn, + struct mm_struct *mm, + unsigned long start, + unsigned long end) +{ + struct ib_ucontext *context = container_of(mn, struct ib_ucontext, mn); + + if (!context->invalidate_range) + return; + + down_read(&context->umem_rwsem); + rbt_ib_umem_for_each_in_range(&context->umem_tree, start, + end, + invalidate_range_end_trampoline, NULL); + up_read(&context->umem_rwsem); + ib_ucontext_notifier_end_account(context); +} + +static struct mmu_notifier_ops ib_umem_notifiers = { + .release = ib_umem_notifier_release, + .invalidate_page = ib_umem_notifier_invalidate_page, + .invalidate_range_start = ib_umem_notifier_invalidate_range_start, + .invalidate_range_end = ib_umem_notifier_invalidate_range_end, +}; + +int ib_umem_odp_get(struct ib_ucontext *context, struct ib_umem *umem) +{ + int ret_val; + struct pid *our_pid; + struct mm_struct *mm = get_task_mm(current); + + if (!mm) + return -EINVAL; + + /* Prevent creating ODP MRs in child processes */ + rcu_read_lock(); + our_pid = get_task_pid(current->group_leader, PIDTYPE_PID); + rcu_read_unlock(); + put_pid(our_pid); + if (context->tgid != our_pid) { + ret_val = -EINVAL; + goto out_mm; + } + + umem->hugetlb = 0; + umem->odp_data = kzalloc(sizeof(*umem->odp_data), GFP_KERNEL); + if (!umem->odp_data) { + ret_val = -ENOMEM; + goto out_mm; + } + umem->odp_data->umem = umem; + + mutex_init(&umem->odp_data->umem_mutex); + + init_completion(&umem->odp_data->notifier_completion); + + umem->odp_data->page_list = vzalloc(ib_umem_num_pages(umem) * + sizeof(*umem->odp_data->page_list)); + if (!umem->odp_data->page_list) { + ret_val = -ENOMEM; + goto out_odp_data; + } + + umem->odp_data->dma_list = vzalloc(ib_umem_num_pages(umem) * + sizeof(*umem->odp_data->dma_list)); + if (!umem->odp_data->dma_list) { + ret_val = -ENOMEM; + goto out_page_list; + } + + /* + * When using MMU notifiers, we will get a + * notification before the "current" task (and MM) is + * destroyed. We use the umem_rwsem semaphore to synchronize. + */ + down_write(&context->umem_rwsem); + context->odp_mrs_count++; + if (likely(ib_umem_start(umem) != ib_umem_end(umem))) + rbt_ib_umem_insert(&umem->odp_data->interval_tree, + &context->umem_tree); + if (likely(!atomic_read(&context->notifier_count))) + umem->odp_data->mn_counters_active = true; + else + list_add(&umem->odp_data->no_private_counters, + &context->no_private_counters); + downgrade_write(&context->umem_rwsem); + + if (context->odp_mrs_count == 1) { + /* + * Note that at this point, no MMU notifier is running + * for this context! + */ + atomic_set(&context->notifier_count, 0); + INIT_HLIST_NODE(&context->mn.hlist); + context->mn.ops = &ib_umem_notifiers; + /* + * Lock-dep detects a false positive for mmap_sem vs. + * umem_rwsem, due to not grasping downgrade_write correctly. + */ + lockdep_off(); + ret_val = mmu_notifier_register(&context->mn, mm); + lockdep_on(); + if (ret_val) { + pr_err("Failed to register mmu_notifier %d\n", ret_val); + ret_val = -EBUSY; + goto out_mutex; + } + } + + up_read(&context->umem_rwsem); + + /* + * Note that doing an mmput can cause a notifier for the relevant mm. + * If the notifier is called while we hold the umem_rwsem, this will + * cause a deadlock. Therefore, we release the reference only after we + * released the semaphore. + */ + mmput(mm); + return 0; + +out_mutex: + up_read(&context->umem_rwsem); + vfree(umem->odp_data->dma_list); +out_page_list: + vfree(umem->odp_data->page_list); +out_odp_data: + kfree(umem->odp_data); +out_mm: + mmput(mm); + return ret_val; +} + +void ib_umem_odp_release(struct ib_umem *umem) +{ + struct ib_ucontext *context = umem->context; + + /* + * Ensure that no more pages are mapped in the umem. + * + * It is the driver's responsibility to ensure, before calling us, + * that the hardware will not attempt to access the MR any more. + */ + ib_umem_odp_unmap_dma_pages(umem, ib_umem_start(umem), + ib_umem_end(umem)); + + down_write(&context->umem_rwsem); + if (likely(ib_umem_start(umem) != ib_umem_end(umem))) + rbt_ib_umem_remove(&umem->odp_data->interval_tree, + &context->umem_tree); + context->odp_mrs_count--; + if (!umem->odp_data->mn_counters_active) { + list_del(&umem->odp_data->no_private_counters); + complete_all(&umem->odp_data->notifier_completion); + } + + /* + * Downgrade the lock to a read lock. This ensures that the notifiers + * (who lock the mutex for reading) will be able to finish, and we + * will be able to enventually obtain the mmu notifiers SRCU. Note + * that since we are doing it atomically, no other user could register + * and unregister while we do the check. + */ + downgrade_write(&context->umem_rwsem); + if (!context->odp_mrs_count) { + struct task_struct *owning_process = NULL; + struct mm_struct *owning_mm = NULL; + + owning_process = get_pid_task(context->tgid, + PIDTYPE_PID); + if (owning_process == NULL) + /* + * The process is already dead, notifier were removed + * already. + */ + goto out; + + owning_mm = get_task_mm(owning_process); + if (owning_mm == NULL) + /* + * The process' mm is already dead, notifier were + * removed already. + */ + goto out_put_task; + mmu_notifier_unregister(&context->mn, owning_mm); + + mmput(owning_mm); + +out_put_task: + put_task_struct(owning_process); + } +out: + up_read(&context->umem_rwsem); + + vfree(umem->odp_data->dma_list); + vfree(umem->odp_data->page_list); + kfree(umem->odp_data); + kfree(umem); +} + +/* + * Map for DMA and insert a single page into the on-demand paging page tables. + * + * @umem: the umem to insert the page to. + * @page_index: index in the umem to add the page to. + * @page: the page struct to map and add. + * @access_mask: access permissions needed for this page. + * @current_seq: sequence number for synchronization with invalidations. + * the sequence number is taken from + * umem->odp_data->notifiers_seq. + * + * The function returns -EFAULT if the DMA mapping operation fails. It returns + * -EAGAIN if a concurrent invalidation prevents us from updating the page. + * + * The page is released via put_page even if the operation failed. For + * on-demand pinning, the page is released whenever it isn't stored in the + * umem. + */ +static int ib_umem_odp_map_dma_single_page( + struct ib_umem *umem, + int page_index, + u64 base_virt_addr, + struct page *page, + u64 access_mask, + unsigned long current_seq) +{ + struct ib_device *dev = umem->context->device; + dma_addr_t dma_addr; + int stored_page = 0; + int remove_existing_mapping = 0; + int ret = 0; + + mutex_lock(&umem->odp_data->umem_mutex); + /* + * Note: we avoid writing if seq is different from the initial seq, to + * handle case of a racing notifier. This check also allows us to bail + * early if we have a notifier running in parallel with us. + */ + if (ib_umem_mmu_notifier_retry(umem, current_seq)) { + ret = -EAGAIN; + goto out; + } + if (!(umem->odp_data->dma_list[page_index])) { + dma_addr = ib_dma_map_page(dev, + page, + 0, PAGE_SIZE, + DMA_BIDIRECTIONAL); + if (ib_dma_mapping_error(dev, dma_addr)) { + ret = -EFAULT; + goto out; + } + umem->odp_data->dma_list[page_index] = dma_addr | access_mask; + umem->odp_data->page_list[page_index] = page; + stored_page = 1; + } else if (umem->odp_data->page_list[page_index] == page) { + umem->odp_data->dma_list[page_index] |= access_mask; + } else { + pr_err("error: got different pages in IB device and from get_user_pages. IB device page: %p, gup page: %p\n", + umem->odp_data->page_list[page_index], page); + /* Better remove the mapping now, to prevent any further + * damage. */ + remove_existing_mapping = 1; + } + +out: + mutex_unlock(&umem->odp_data->umem_mutex); + + /* On Demand Paging - avoid pinning the page */ + if (umem->context->invalidate_range || !stored_page) + put_page(page); + + if (remove_existing_mapping && umem->context->invalidate_range) { + invalidate_page_trampoline( + umem, + base_virt_addr + (page_index * PAGE_SIZE), + base_virt_addr + ((page_index+1)*PAGE_SIZE), + NULL); + ret = -EAGAIN; + } + + return ret; +} + +/** + * ib_umem_odp_map_dma_pages - Pin and DMA map userspace memory in an ODP MR. + * + * Pins the range of pages passed in the argument, and maps them to + * DMA addresses. The DMA addresses of the mapped pages is updated in + * umem->odp_data->dma_list. + * + * Returns the number of pages mapped in success, negative error code + * for failure. + * An -EAGAIN error code is returned when a concurrent mmu notifier prevents + * the function from completing its task. + * + * @umem: the umem to map and pin + * @user_virt: the address from which we need to map. + * @bcnt: the minimal number of bytes to pin and map. The mapping might be + * bigger due to alignment, and may also be smaller in case of an error + * pinning or mapping a page. The actual pages mapped is returned in + * the return value. + * @access_mask: bit mask of the requested access permissions for the given + * range. + * @current_seq: the MMU notifiers sequance value for synchronization with + * invalidations. the sequance number is read from + * umem->odp_data->notifiers_seq before calling this function + */ +int ib_umem_odp_map_dma_pages(struct ib_umem *umem, u64 user_virt, u64 bcnt, + u64 access_mask, unsigned long current_seq) +{ + struct task_struct *owning_process = NULL; + struct mm_struct *owning_mm = NULL; + struct page **local_page_list = NULL; + u64 off; + int j, k, ret = 0, start_idx, npages = 0; + u64 base_virt_addr; + + if (access_mask == 0) + return -EINVAL; + + if (user_virt < ib_umem_start(umem) || + user_virt + bcnt > ib_umem_end(umem)) + return -EFAULT; + + local_page_list = (struct page **)__get_free_page(GFP_KERNEL); + if (!local_page_list) + return -ENOMEM; + + off = user_virt & (~PAGE_MASK); + user_virt = user_virt & PAGE_MASK; + base_virt_addr = user_virt; + bcnt += off; /* Charge for the first page offset as well. */ + + owning_process = get_pid_task(umem->context->tgid, PIDTYPE_PID); + if (owning_process == NULL) { + ret = -EINVAL; + goto out_no_task; + } + + owning_mm = get_task_mm(owning_process); + if (owning_mm == NULL) { + ret = -EINVAL; + goto out_put_task; + } + + start_idx = (user_virt - ib_umem_start(umem)) >> PAGE_SHIFT; + k = start_idx; + + while (bcnt > 0) { + const size_t gup_num_pages = + min_t(size_t, ALIGN(bcnt, PAGE_SIZE) / PAGE_SIZE, + PAGE_SIZE / sizeof(struct page *)); + + down_read(&owning_mm->mmap_sem); + /* + * Note: this might result in redundent page getting. We can + * avoid this by checking dma_list to be 0 before calling + * get_user_pages. However, this make the code much more + * complex (and doesn't gain us much performance in most use + * cases). + */ + npages = get_user_pages(owning_process, owning_mm, user_virt, + gup_num_pages, + access_mask & ODP_WRITE_ALLOWED_BIT, 0, + local_page_list, NULL); + up_read(&owning_mm->mmap_sem); + + if (npages < 0) + break; + + bcnt -= min_t(size_t, npages << PAGE_SHIFT, bcnt); + user_virt += npages << PAGE_SHIFT; + for (j = 0; j < npages; ++j) { + ret = ib_umem_odp_map_dma_single_page( + umem, k, base_virt_addr, local_page_list[j], + access_mask, current_seq); + if (ret < 0) + break; + k++; + } + + if (ret < 0) { + /* Release left over pages when handling errors. */ + for (++j; j < npages; ++j) + put_page(local_page_list[j]); + break; + } + } + + if (ret >= 0) { + if (npages < 0 && k == start_idx) + ret = npages; + else + ret = k - start_idx; + } + + mmput(owning_mm); +out_put_task: + put_task_struct(owning_process); +out_no_task: + free_page((unsigned long)local_page_list); + return ret; +} +EXPORT_SYMBOL(ib_umem_odp_map_dma_pages); + +void ib_umem_odp_unmap_dma_pages(struct ib_umem *umem, u64 virt, + u64 bound) +{ + int idx; + u64 addr; + struct ib_device *dev = umem->context->device; + + virt = max_t(u64, virt, ib_umem_start(umem)); + bound = min_t(u64, bound, ib_umem_end(umem)); + /* Note that during the run of this function, the + * notifiers_count of the MR is > 0, preventing any racing + * faults from completion. We might be racing with other + * invalidations, so we must make sure we free each page only + * once. */ + for (addr = virt; addr < bound; addr += (u64)umem->page_size) { + idx = (addr - ib_umem_start(umem)) / PAGE_SIZE; + mutex_lock(&umem->odp_data->umem_mutex); + if (umem->odp_data->page_list[idx]) { + struct page *page = umem->odp_data->page_list[idx]; + struct page *head_page = compound_head(page); + dma_addr_t dma = umem->odp_data->dma_list[idx]; + dma_addr_t dma_addr = dma & ODP_DMA_ADDR_MASK; + + WARN_ON(!dma_addr); + + ib_dma_unmap_page(dev, dma_addr, PAGE_SIZE, + DMA_BIDIRECTIONAL); + if (dma & ODP_WRITE_ALLOWED_BIT) + /* + * set_page_dirty prefers being called with + * the page lock. However, MMU notifiers are + * called sometimes with and sometimes without + * the lock. We rely on the umem_mutex instead + * to prevent other mmu notifiers from + * continuing and allowing the page mapping to + * be removed. + */ + set_page_dirty(head_page); + /* on demand pinning support */ + if (!umem->context->invalidate_range) + put_page(page); + umem->odp_data->page_list[idx] = NULL; + umem->odp_data->dma_list[idx] = 0; + } + mutex_unlock(&umem->odp_data->umem_mutex); + } +} +EXPORT_SYMBOL(ib_umem_odp_unmap_dma_pages); diff --git a/drivers/infiniband/core/umem_rbtree.c b/drivers/infiniband/core/umem_rbtree.c new file mode 100644 index 000000000000..727d788448f5 --- /dev/null +++ b/drivers/infiniband/core/umem_rbtree.c @@ -0,0 +1,94 @@ +/* + * Copyright (c) 2014 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/interval_tree_generic.h> +#include <linux/sched.h> +#include <linux/gfp.h> +#include <rdma/ib_umem_odp.h> + +/* + * The ib_umem list keeps track of memory regions for which the HW + * device request to receive notification when the related memory + * mapping is changed. + * + * ib_umem_lock protects the list. + */ + +static inline u64 node_start(struct umem_odp_node *n) +{ + struct ib_umem_odp *umem_odp = + container_of(n, struct ib_umem_odp, interval_tree); + + return ib_umem_start(umem_odp->umem); +} + +/* Note that the representation of the intervals in the interval tree + * considers the ending point as contained in the interval, while the + * function ib_umem_end returns the first address which is not contained + * in the umem. + */ +static inline u64 node_last(struct umem_odp_node *n) +{ + struct ib_umem_odp *umem_odp = + container_of(n, struct ib_umem_odp, interval_tree); + + return ib_umem_end(umem_odp->umem) - 1; +} + +INTERVAL_TREE_DEFINE(struct umem_odp_node, rb, u64, __subtree_last, + node_start, node_last, , rbt_ib_umem) + +/* @last is not a part of the interval. See comment for function + * node_last. + */ +int rbt_ib_umem_for_each_in_range(struct rb_root *root, + u64 start, u64 last, + umem_call_back cb, + void *cookie) +{ + int ret_val = 0; + struct umem_odp_node *node; + struct ib_umem_odp *umem; + + if (unlikely(start == last)) + return ret_val; + + for (node = rbt_ib_umem_iter_first(root, start, last - 1); node; + node = rbt_ib_umem_iter_next(node, start, last - 1)) { + umem = container_of(node, struct ib_umem_odp, interval_tree); + ret_val = cb(umem->umem, start, last, cookie) || ret_val; + } + + return ret_val; +} diff --git a/drivers/infiniband/core/uverbs.h b/drivers/infiniband/core/uverbs.h index 643c08a025a5..b716b0815644 100644 --- a/drivers/infiniband/core/uverbs.h +++ b/drivers/infiniband/core/uverbs.h @@ -258,5 +258,6 @@ IB_UVERBS_DECLARE_CMD(close_xrcd); IB_UVERBS_DECLARE_EX_CMD(create_flow); IB_UVERBS_DECLARE_EX_CMD(destroy_flow); +IB_UVERBS_DECLARE_EX_CMD(query_device); #endif /* UVERBS_H */ diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c index 5ba2a86aab6a..532d8eba8b02 100644 --- a/drivers/infiniband/core/uverbs_cmd.c +++ b/drivers/infiniband/core/uverbs_cmd.c @@ -36,6 +36,7 @@ #include <linux/file.h> #include <linux/fs.h> #include <linux/slab.h> +#include <linux/sched.h> #include <asm/uaccess.h> @@ -288,6 +289,9 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file, struct ib_uverbs_get_context_resp resp; struct ib_udata udata; struct ib_device *ibdev = file->device->ib_dev; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + struct ib_device_attr dev_attr; +#endif struct ib_ucontext *ucontext; struct file *filp; int ret; @@ -325,8 +329,25 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file, INIT_LIST_HEAD(&ucontext->ah_list); INIT_LIST_HEAD(&ucontext->xrcd_list); INIT_LIST_HEAD(&ucontext->rule_list); + rcu_read_lock(); + ucontext->tgid = get_task_pid(current->group_leader, PIDTYPE_PID); + rcu_read_unlock(); ucontext->closing = 0; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + ucontext->umem_tree = RB_ROOT; + init_rwsem(&ucontext->umem_rwsem); + ucontext->odp_mrs_count = 0; + INIT_LIST_HEAD(&ucontext->no_private_counters); + + ret = ib_query_device(ibdev, &dev_attr); + if (ret) + goto err_free; + if (!(dev_attr.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING)) + ucontext->invalidate_range = NULL; + +#endif + resp.num_comp_vectors = file->device->num_comp_vectors; ret = get_unused_fd_flags(O_CLOEXEC); @@ -371,6 +392,7 @@ err_fd: put_unused_fd(resp.async_fd); err_free: + put_pid(ucontext->tgid); ibdev->dealloc_ucontext(ucontext); err: @@ -378,6 +400,52 @@ err: return ret; } +static void copy_query_dev_fields(struct ib_uverbs_file *file, + struct ib_uverbs_query_device_resp *resp, + struct ib_device_attr *attr) +{ + resp->fw_ver = attr->fw_ver; + resp->node_guid = file->device->ib_dev->node_guid; + resp->sys_image_guid = attr->sys_image_guid; + resp->max_mr_size = attr->max_mr_size; + resp->page_size_cap = attr->page_size_cap; + resp->vendor_id = attr->vendor_id; + resp->vendor_part_id = attr->vendor_part_id; + resp->hw_ver = attr->hw_ver; + resp->max_qp = attr->max_qp; + resp->max_qp_wr = attr->max_qp_wr; + resp->device_cap_flags = attr->device_cap_flags; + resp->max_sge = attr->max_sge; + resp->max_sge_rd = attr->max_sge_rd; + resp->max_cq = attr->max_cq; + resp->max_cqe = attr->max_cqe; + resp->max_mr = attr->max_mr; + resp->max_pd = attr->max_pd; + resp->max_qp_rd_atom = attr->max_qp_rd_atom; + resp->max_ee_rd_atom = attr->max_ee_rd_atom; + resp->max_res_rd_atom = attr->max_res_rd_atom; + resp->max_qp_init_rd_atom = attr->max_qp_init_rd_atom; + resp->max_ee_init_rd_atom = attr->max_ee_init_rd_atom; + resp->atomic_cap = attr->atomic_cap; + resp->max_ee = attr->max_ee; + resp->max_rdd = attr->max_rdd; + resp->max_mw = attr->max_mw; + resp->max_raw_ipv6_qp = attr->max_raw_ipv6_qp; + resp->max_raw_ethy_qp = attr->max_raw_ethy_qp; + resp->max_mcast_grp = attr->max_mcast_grp; + resp->max_mcast_qp_attach = attr->max_mcast_qp_attach; + resp->max_total_mcast_qp_attach = attr->max_total_mcast_qp_attach; + resp->max_ah = attr->max_ah; + resp->max_fmr = attr->max_fmr; + resp->max_map_per_fmr = attr->max_map_per_fmr; + resp->max_srq = attr->max_srq; + resp->max_srq_wr = attr->max_srq_wr; + resp->max_srq_sge = attr->max_srq_sge; + resp->max_pkeys = attr->max_pkeys; + resp->local_ca_ack_delay = attr->local_ca_ack_delay; + resp->phys_port_cnt = file->device->ib_dev->phys_port_cnt; +} + ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file, const char __user *buf, int in_len, int out_len) @@ -398,47 +466,7 @@ ssize_t ib_uverbs_query_device(struct ib_uverbs_file *file, return ret; memset(&resp, 0, sizeof resp); - - resp.fw_ver = attr.fw_ver; - resp.node_guid = file->device->ib_dev->node_guid; - resp.sys_image_guid = attr.sys_image_guid; - resp.max_mr_size = attr.max_mr_size; - resp.page_size_cap = attr.page_size_cap; - resp.vendor_id = attr.vendor_id; - resp.vendor_part_id = attr.vendor_part_id; - resp.hw_ver = attr.hw_ver; - resp.max_qp = attr.max_qp; - resp.max_qp_wr = attr.max_qp_wr; - resp.device_cap_flags = attr.device_cap_flags; - resp.max_sge = attr.max_sge; - resp.max_sge_rd = attr.max_sge_rd; - resp.max_cq = attr.max_cq; - resp.max_cqe = attr.max_cqe; - resp.max_mr = attr.max_mr; - resp.max_pd = attr.max_pd; - resp.max_qp_rd_atom = attr.max_qp_rd_atom; - resp.max_ee_rd_atom = attr.max_ee_rd_atom; - resp.max_res_rd_atom = attr.max_res_rd_atom; - resp.max_qp_init_rd_atom = attr.max_qp_init_rd_atom; - resp.max_ee_init_rd_atom = attr.max_ee_init_rd_atom; - resp.atomic_cap = attr.atomic_cap; - resp.max_ee = attr.max_ee; - resp.max_rdd = attr.max_rdd; - resp.max_mw = attr.max_mw; - resp.max_raw_ipv6_qp = attr.max_raw_ipv6_qp; - resp.max_raw_ethy_qp = attr.max_raw_ethy_qp; - resp.max_mcast_grp = attr.max_mcast_grp; - resp.max_mcast_qp_attach = attr.max_mcast_qp_attach; - resp.max_total_mcast_qp_attach = attr.max_total_mcast_qp_attach; - resp.max_ah = attr.max_ah; - resp.max_fmr = attr.max_fmr; - resp.max_map_per_fmr = attr.max_map_per_fmr; - resp.max_srq = attr.max_srq; - resp.max_srq_wr = attr.max_srq_wr; - resp.max_srq_sge = attr.max_srq_sge; - resp.max_pkeys = attr.max_pkeys; - resp.local_ca_ack_delay = attr.local_ca_ack_delay; - resp.phys_port_cnt = file->device->ib_dev->phys_port_cnt; + copy_query_dev_fields(file, &resp, &attr); if (copy_to_user((void __user *) (unsigned long) cmd.response, &resp, sizeof resp)) @@ -947,6 +975,18 @@ ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file, goto err_free; } + if (cmd.access_flags & IB_ACCESS_ON_DEMAND) { + struct ib_device_attr attr; + + ret = ib_query_device(pd->device, &attr); + if (ret || !(attr.device_cap_flags & + IB_DEVICE_ON_DEMAND_PAGING)) { + pr_debug("ODP support not available\n"); + ret = -EINVAL; + goto err_put; + } + } + mr = pd->device->reg_user_mr(pd, cmd.start, cmd.length, cmd.hca_va, cmd.access_flags, &udata); if (IS_ERR(mr)) { @@ -3253,3 +3293,52 @@ ssize_t ib_uverbs_destroy_srq(struct ib_uverbs_file *file, return ret ? ret : in_len; } + +int ib_uverbs_ex_query_device(struct ib_uverbs_file *file, + struct ib_udata *ucore, + struct ib_udata *uhw) +{ + struct ib_uverbs_ex_query_device_resp resp; + struct ib_uverbs_ex_query_device cmd; + struct ib_device_attr attr; + struct ib_device *device; + int err; + + device = file->device->ib_dev; + if (ucore->inlen < sizeof(cmd)) + return -EINVAL; + + err = ib_copy_from_udata(&cmd, ucore, sizeof(cmd)); + if (err) + return err; + + if (cmd.reserved) + return -EINVAL; + + err = device->query_device(device, &attr); + if (err) + return err; + + memset(&resp, 0, sizeof(resp)); + copy_query_dev_fields(file, &resp.base, &attr); + resp.comp_mask = 0; + +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + if (cmd.comp_mask & IB_USER_VERBS_EX_QUERY_DEVICE_ODP) { + resp.odp_caps.general_caps = attr.odp_caps.general_caps; + resp.odp_caps.per_transport_caps.rc_odp_caps = + attr.odp_caps.per_transport_caps.rc_odp_caps; + resp.odp_caps.per_transport_caps.uc_odp_caps = + attr.odp_caps.per_transport_caps.uc_odp_caps; + resp.odp_caps.per_transport_caps.ud_odp_caps = + attr.odp_caps.per_transport_caps.ud_odp_caps; + resp.comp_mask |= IB_USER_VERBS_EX_QUERY_DEVICE_ODP; + } +#endif + + err = ib_copy_to_udata(ucore, &resp, sizeof(resp)); + if (err) + return err; + + return 0; +} diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c index 71ab83fde472..e6c23b9eab33 100644 --- a/drivers/infiniband/core/uverbs_main.c +++ b/drivers/infiniband/core/uverbs_main.c @@ -122,7 +122,8 @@ static int (*uverbs_ex_cmd_table[])(struct ib_uverbs_file *file, struct ib_udata *ucore, struct ib_udata *uhw) = { [IB_USER_VERBS_EX_CMD_CREATE_FLOW] = ib_uverbs_ex_create_flow, - [IB_USER_VERBS_EX_CMD_DESTROY_FLOW] = ib_uverbs_ex_destroy_flow + [IB_USER_VERBS_EX_CMD_DESTROY_FLOW] = ib_uverbs_ex_destroy_flow, + [IB_USER_VERBS_EX_CMD_QUERY_DEVICE] = ib_uverbs_ex_query_device }; static void ib_uverbs_add_one(struct ib_device *device); @@ -296,6 +297,8 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file, kfree(uobj); } + put_pid(context->tgid); + return context->device->dealloc_ucontext(context); } diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c index c2b89cc5dbca..f93eb8da7b5a 100644 --- a/drivers/infiniband/core/verbs.c +++ b/drivers/infiniband/core/verbs.c @@ -879,7 +879,8 @@ int ib_resolve_eth_l2_attrs(struct ib_qp *qp, if (rdma_link_local_addr((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw)) { rdma_get_ll_mac((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw, qp_attr->ah_attr.dmac); rdma_get_ll_mac((struct in6_addr *)sgid.raw, qp_attr->smac); - qp_attr->vlan_id = rdma_get_vlan_id(&sgid); + if (!(*qp_attr_mask & IB_QP_VID)) + qp_attr->vlan_id = rdma_get_vlan_id(&sgid); } else { ret = rdma_addr_find_dmac_by_grh(&sgid, &qp_attr->ah_attr.grh.dgid, qp_attr->ah_attr.dmac, &qp_attr->vlan_id); diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c index 2d5cbf4363e4..bdf3507810cb 100644 --- a/drivers/infiniband/hw/amso1100/c2_provider.c +++ b/drivers/infiniband/hw/amso1100/c2_provider.c @@ -476,7 +476,7 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, c2mr->umem->page_size, i, length, - c2mr->umem->offset, + ib_umem_offset(c2mr->umem), &kva, c2_convert_access(acc), c2mr); diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index 4b8c6116c058..9edc200b311d 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -1640,7 +1640,8 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb) __state_set(&ep->com, MPA_REQ_RCVD); /* drive upcall */ - mutex_lock(&ep->parent_ep->com.mutex); + mutex_lock_nested(&ep->parent_ep->com.mutex, + SINGLE_DEPTH_NESTING); if (ep->parent_ep->com.state != DEAD) { if (connect_request_upcall(ep)) abort_connection(ep, skb, GFP_KERNEL); @@ -3126,6 +3127,8 @@ static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0, __func__); + else if (err > 0) + err = net_xmit_errno(err); if (err) pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n", err, ep->stid, @@ -3159,6 +3162,8 @@ static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 0, 0, __func__); + else if (err > 0) + err = net_xmit_errno(err); } if (err) pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n" diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c index 72f1f052e88c..eb5df4e62703 100644 --- a/drivers/infiniband/hw/cxgb4/device.c +++ b/drivers/infiniband/hw/cxgb4/device.c @@ -670,7 +670,7 @@ static int ep_open(struct inode *inode, struct file *file) idr_for_each(&epd->devp->stid_idr, count_idrs, &count); spin_unlock_irq(&epd->devp->lock); - epd->bufsize = count * 160; + epd->bufsize = count * 240; epd->buf = vmalloc(epd->bufsize); if (!epd->buf) { ret = -ENOMEM; diff --git a/drivers/infiniband/hw/cxgb4/mem.c b/drivers/infiniband/hw/cxgb4/mem.c index 0744455cd88b..cb43c2299ac0 100644 --- a/drivers/infiniband/hw/cxgb4/mem.c +++ b/drivers/infiniband/hw/cxgb4/mem.c @@ -50,6 +50,13 @@ static int inline_threshold = C4IW_INLINE_THRESHOLD; module_param(inline_threshold, int, 0644); MODULE_PARM_DESC(inline_threshold, "inline vs dsgl threshold (default=128)"); +static int mr_exceeds_hw_limits(struct c4iw_dev *dev, u64 length) +{ + return (is_t4(dev->rdev.lldi.adapter_type) || + is_t5(dev->rdev.lldi.adapter_type)) && + length >= 8*1024*1024*1024ULL; +} + static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr, u32 len, dma_addr_t data, int wait) { @@ -369,9 +376,11 @@ static int register_mem(struct c4iw_dev *rhp, struct c4iw_pd *php, int ret; ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, mhp->attr.pdid, - FW_RI_STAG_NSMR, mhp->attr.perms, + FW_RI_STAG_NSMR, mhp->attr.len ? + mhp->attr.perms : 0, mhp->attr.mw_bind_enable, mhp->attr.zbva, - mhp->attr.va_fbo, mhp->attr.len, shift - 12, + mhp->attr.va_fbo, mhp->attr.len ? + mhp->attr.len : -1, shift - 12, mhp->attr.pbl_size, mhp->attr.pbl_addr); if (ret) return ret; @@ -536,6 +545,11 @@ int c4iw_reregister_phys_mem(struct ib_mr *mr, int mr_rereg_mask, return ret; } + if (mr_exceeds_hw_limits(rhp, total_size)) { + kfree(page_list); + return -EINVAL; + } + ret = reregister_mem(rhp, php, &mh, shift, npages); kfree(page_list); if (ret) @@ -596,6 +610,12 @@ struct ib_mr *c4iw_register_phys_mem(struct ib_pd *pd, if (ret) goto err; + if (mr_exceeds_hw_limits(rhp, total_size)) { + kfree(page_list); + ret = -EINVAL; + goto err; + } + ret = alloc_pbl(mhp, npages); if (ret) { kfree(page_list); @@ -699,6 +719,10 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, php = to_c4iw_pd(pd); rhp = php->rhp; + + if (mr_exceeds_hw_limits(rhp, length)) + return ERR_PTR(-EINVAL); + mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); if (!mhp) return ERR_PTR(-ENOMEM); diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c index 2ed3ece2b2ee..bb85d479e66e 100644 --- a/drivers/infiniband/hw/cxgb4/qp.c +++ b/drivers/infiniband/hw/cxgb4/qp.c @@ -1538,9 +1538,9 @@ err: set_state(qhp, C4IW_QP_STATE_ERROR); free = 1; abort = 1; - wake_up(&qhp->wait); BUG_ON(!ep); flush_qp(qhp); + wake_up(&qhp->wait); out: mutex_unlock(&qhp->mutex); diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c index 3488e8c9fcb4..f914b30999f8 100644 --- a/drivers/infiniband/hw/ehca/ehca_mrmw.c +++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c @@ -399,7 +399,7 @@ reg_user_mr_fallback: pginfo.num_kpages = num_kpages; pginfo.num_hwpages = num_hwpages; pginfo.u.usr.region = e_mr->umem; - pginfo.next_hwpage = e_mr->umem->offset / hwpage_size; + pginfo.next_hwpage = ib_umem_offset(e_mr->umem) / hwpage_size; pginfo.u.usr.next_sg = pginfo.u.usr.region->sg_head.sgl; ret = ehca_reg_mr(shca, e_mr, (u64 *)virt, length, mr_access_flags, e_pd, &pginfo, &e_mr->ib.ib_mr.lkey, diff --git a/drivers/infiniband/hw/ipath/ipath_mr.c b/drivers/infiniband/hw/ipath/ipath_mr.c index 5e61e9bff697..c7278f6a8217 100644 --- a/drivers/infiniband/hw/ipath/ipath_mr.c +++ b/drivers/infiniband/hw/ipath/ipath_mr.c @@ -214,7 +214,7 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mr->mr.user_base = start; mr->mr.iova = virt_addr; mr->mr.length = length; - mr->mr.offset = umem->offset; + mr->mr.offset = ib_umem_offset(umem); mr->mr.access_flags = mr_access_flags; mr->mr.max_segs = n; mr->umem = umem; diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c index 8f9325cfc85d..c36ccbd9a644 100644 --- a/drivers/infiniband/hw/mlx4/mr.c +++ b/drivers/infiniband/hw/mlx4/mr.c @@ -223,7 +223,6 @@ int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags, if (flags & IB_MR_REREG_TRANS) { int shift; - int err; int n; mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr); diff --git a/drivers/infiniband/hw/mlx5/Makefile b/drivers/infiniband/hw/mlx5/Makefile index 4ea0135af484..27a70159e2ea 100644 --- a/drivers/infiniband/hw/mlx5/Makefile +++ b/drivers/infiniband/hw/mlx5/Makefile @@ -1,3 +1,4 @@ obj-$(CONFIG_MLX5_INFINIBAND) += mlx5_ib.o mlx5_ib-y := main.o cq.o doorbell.o qp.o mem.o srq.o mr.o ah.o mad.o +mlx5_ib-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += odp.o diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 1ba6c42e4df8..8a87404e9c76 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -244,6 +244,12 @@ static int mlx5_ib_query_device(struct ib_device *ibdev, props->max_mcast_grp; props->max_map_per_fmr = INT_MAX; /* no limit in ConnectIB */ +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + if (dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_ON_DMND_PG) + props->device_cap_flags |= IB_DEVICE_ON_DEMAND_PAGING; + props->odp_caps = dev->odp_caps; +#endif + out: kfree(in_mad); kfree(out_mad); @@ -568,6 +574,10 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, goto out_count; } +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + context->ibucontext.invalidate_range = &mlx5_ib_invalidate_range; +#endif + INIT_LIST_HEAD(&context->db_page_list); mutex_init(&context->db_page_mutex); @@ -858,7 +868,7 @@ static ssize_t show_reg_pages(struct device *device, struct mlx5_ib_dev *dev = container_of(device, struct mlx5_ib_dev, ib_dev.dev); - return sprintf(buf, "%d\n", dev->mdev->priv.reg_pages); + return sprintf(buf, "%d\n", atomic_read(&dev->mdev->priv.reg_pages)); } static ssize_t show_hca(struct device *device, struct device_attribute *attr, @@ -1321,6 +1331,8 @@ static void *mlx5_ib_add(struct mlx5_core_dev *mdev) (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) | (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) | (1ull << IB_USER_VERBS_CMD_OPEN_QP); + dev->ib_dev.uverbs_ex_cmd_mask = + (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE); dev->ib_dev.query_device = mlx5_ib_query_device; dev->ib_dev.query_port = mlx5_ib_query_port; @@ -1366,6 +1378,8 @@ static void *mlx5_ib_add(struct mlx5_core_dev *mdev) dev->ib_dev.free_fast_reg_page_list = mlx5_ib_free_fast_reg_page_list; dev->ib_dev.check_mr_status = mlx5_ib_check_mr_status; + mlx5_ib_internal_query_odp_caps(dev); + if (mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_XRC) { dev->ib_dev.alloc_xrcd = mlx5_ib_alloc_xrcd; dev->ib_dev.dealloc_xrcd = mlx5_ib_dealloc_xrcd; @@ -1379,16 +1393,19 @@ static void *mlx5_ib_add(struct mlx5_core_dev *mdev) goto err_eqs; mutex_init(&dev->cap_mask_mutex); - spin_lock_init(&dev->mr_lock); err = create_dev_resources(&dev->devr); if (err) goto err_eqs; - err = ib_register_device(&dev->ib_dev, NULL); + err = mlx5_ib_odp_init_one(dev); if (err) goto err_rsrc; + err = ib_register_device(&dev->ib_dev, NULL); + if (err) + goto err_odp; + err = create_umr_res(dev); if (err) goto err_dev; @@ -1410,6 +1427,9 @@ err_umrc: err_dev: ib_unregister_device(&dev->ib_dev); +err_odp: + mlx5_ib_odp_remove_one(dev); + err_rsrc: destroy_dev_resources(&dev->devr); @@ -1425,8 +1445,10 @@ err_dealloc: static void mlx5_ib_remove(struct mlx5_core_dev *mdev, void *context) { struct mlx5_ib_dev *dev = context; + ib_unregister_device(&dev->ib_dev); destroy_umrc_res(dev); + mlx5_ib_odp_remove_one(dev); destroy_dev_resources(&dev->devr); free_comp_eqs(dev); ib_dealloc_device(&dev->ib_dev); @@ -1440,15 +1462,30 @@ static struct mlx5_interface mlx5_ib_interface = { static int __init mlx5_ib_init(void) { + int err; + if (deprecated_prof_sel != 2) pr_warn("prof_sel is deprecated for mlx5_ib, set it for mlx5_core\n"); - return mlx5_register_interface(&mlx5_ib_interface); + err = mlx5_ib_odp_init(); + if (err) + return err; + + err = mlx5_register_interface(&mlx5_ib_interface); + if (err) + goto clean_odp; + + return err; + +clean_odp: + mlx5_ib_odp_cleanup(); + return err; } static void __exit mlx5_ib_cleanup(void) { mlx5_unregister_interface(&mlx5_ib_interface); + mlx5_ib_odp_cleanup(); } module_init(mlx5_ib_init); diff --git a/drivers/infiniband/hw/mlx5/mem.c b/drivers/infiniband/hw/mlx5/mem.c index dae07eae9507..b56e4c5593ee 100644 --- a/drivers/infiniband/hw/mlx5/mem.c +++ b/drivers/infiniband/hw/mlx5/mem.c @@ -32,6 +32,7 @@ #include <linux/module.h> #include <rdma/ib_umem.h> +#include <rdma/ib_umem_odp.h> #include "mlx5_ib.h" /* @umem: umem object to scan @@ -57,6 +58,17 @@ void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift, int entry; unsigned long page_shift = ilog2(umem->page_size); + /* With ODP we must always match OS page size. */ + if (umem->odp_data) { + *count = ib_umem_page_count(umem); + *shift = PAGE_SHIFT; + *ncont = *count; + if (order) + *order = ilog2(roundup_pow_of_two(*count)); + + return; + } + addr = addr >> page_shift; tmp = (unsigned long)addr; m = find_first_bit(&tmp, sizeof(tmp)); @@ -108,8 +120,36 @@ void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift, *count = i; } -void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, - int page_shift, __be64 *pas, int umr) +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING +static u64 umem_dma_to_mtt(dma_addr_t umem_dma) +{ + u64 mtt_entry = umem_dma & ODP_DMA_ADDR_MASK; + + if (umem_dma & ODP_READ_ALLOWED_BIT) + mtt_entry |= MLX5_IB_MTT_READ; + if (umem_dma & ODP_WRITE_ALLOWED_BIT) + mtt_entry |= MLX5_IB_MTT_WRITE; + + return mtt_entry; +} +#endif + +/* + * Populate the given array with bus addresses from the umem. + * + * dev - mlx5_ib device + * umem - umem to use to fill the pages + * page_shift - determines the page size used in the resulting array + * offset - offset into the umem to start from, + * only implemented for ODP umems + * num_pages - total number of pages to fill + * pas - bus addresses array to fill + * access_flags - access flags to set on all present pages. + use enum mlx5_ib_mtt_access_flags for this. + */ +void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, + int page_shift, size_t offset, size_t num_pages, + __be64 *pas, int access_flags) { unsigned long umem_page_shift = ilog2(umem->page_size); int shift = page_shift - umem_page_shift; @@ -120,6 +160,21 @@ void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, int len; struct scatterlist *sg; int entry; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + const bool odp = umem->odp_data != NULL; + + if (odp) { + WARN_ON(shift != 0); + WARN_ON(access_flags != (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)); + + for (i = 0; i < num_pages; ++i) { + dma_addr_t pa = umem->odp_data->dma_list[offset + i]; + + pas[i] = cpu_to_be64(umem_dma_to_mtt(pa)); + } + return; + } +#endif i = 0; for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { @@ -128,8 +183,7 @@ void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, for (k = 0; k < len; k++) { if (!(i & mask)) { cur = base + (k << umem_page_shift); - if (umr) - cur |= 3; + cur |= access_flags; pas[i >> shift] = cpu_to_be64(cur); mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n", @@ -142,6 +196,13 @@ void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, } } +void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, + int page_shift, __be64 *pas, int access_flags) +{ + return __mlx5_ib_populate_pas(dev, umem, page_shift, 0, + ib_umem_num_pages(umem), pas, + access_flags); +} int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset) { u64 page_size; diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 386780f0d1e1..83f22fe297c8 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -111,6 +111,8 @@ struct mlx5_ib_pd { */ #define MLX5_IB_SEND_UMR_UNREG IB_SEND_RESERVED_START +#define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1) +#define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2) #define MLX5_IB_QPT_REG_UMR IB_QPT_RESERVED1 #define MLX5_IB_WR_UMR IB_WR_RESERVED1 @@ -147,6 +149,29 @@ enum { MLX5_QP_EMPTY }; +/* + * Connect-IB can trigger up to four concurrent pagefaults + * per-QP. + */ +enum mlx5_ib_pagefault_context { + MLX5_IB_PAGEFAULT_RESPONDER_READ, + MLX5_IB_PAGEFAULT_REQUESTOR_READ, + MLX5_IB_PAGEFAULT_RESPONDER_WRITE, + MLX5_IB_PAGEFAULT_REQUESTOR_WRITE, + MLX5_IB_PAGEFAULT_CONTEXTS +}; + +static inline enum mlx5_ib_pagefault_context + mlx5_ib_get_pagefault_context(struct mlx5_pagefault *pagefault) +{ + return pagefault->flags & (MLX5_PFAULT_REQUESTOR | MLX5_PFAULT_WRITE); +} + +struct mlx5_ib_pfault { + struct work_struct work; + struct mlx5_pagefault mpfault; +}; + struct mlx5_ib_qp { struct ib_qp ibqp; struct mlx5_core_qp mqp; @@ -192,6 +217,21 @@ struct mlx5_ib_qp { /* Store signature errors */ bool signature_en; + +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + /* + * A flag that is true for QP's that are in a state that doesn't + * allow page faults, and shouldn't schedule any more faults. + */ + int disable_page_faults; + /* + * The disable_page_faults_lock protects a QP's disable_page_faults + * field, allowing for a thread to atomically check whether the QP + * allows page faults, and if so schedule a page fault. + */ + spinlock_t disable_page_faults_lock; + struct mlx5_ib_pfault pagefaults[MLX5_IB_PAGEFAULT_CONTEXTS]; +#endif }; struct mlx5_ib_cq_buf { @@ -206,6 +246,19 @@ enum mlx5_ib_qp_flags { MLX5_IB_QP_SIGNATURE_HANDLING = 1 << 1, }; +struct mlx5_umr_wr { + union { + u64 virt_addr; + u64 offset; + } target; + struct ib_pd *pd; + unsigned int page_shift; + unsigned int npages; + u32 length; + int access_flags; + u32 mkey; +}; + struct mlx5_shared_mr_info { int mr_id; struct ib_umem *umem; @@ -253,6 +306,13 @@ struct mlx5_ib_xrcd { u32 xrcdn; }; +enum mlx5_ib_mtt_access_flags { + MLX5_IB_MTT_READ = (1 << 0), + MLX5_IB_MTT_WRITE = (1 << 1), +}; + +#define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE) + struct mlx5_ib_mr { struct ib_mr ibmr; struct mlx5_core_mr mmr; @@ -261,12 +321,11 @@ struct mlx5_ib_mr { struct list_head list; int order; int umred; - __be64 *pas; - dma_addr_t dma; int npages; struct mlx5_ib_dev *dev; struct mlx5_create_mkey_mbox_out out; struct mlx5_core_sig_ctx *sig; + int live; }; struct mlx5_ib_fast_reg_page_list { @@ -372,11 +431,18 @@ struct mlx5_ib_dev { struct umr_common umrc; /* sync used page count stats */ - spinlock_t mr_lock; struct mlx5_ib_resources devr; struct mlx5_mr_cache cache; struct timer_list delay_timer; int fill_delay; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + struct ib_odp_caps odp_caps; + /* + * Sleepable RCU that prevents destruction of MRs while they are still + * being used by a page fault handler. + */ + struct srcu_struct mr_srcu; +#endif }; static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq) @@ -490,6 +556,8 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, int mlx5_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, struct ib_recv_wr **bad_wr); void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n); +int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index, + void *buffer, u32 length); struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, int entries, int vector, struct ib_ucontext *context, struct ib_udata *udata); @@ -502,6 +570,8 @@ struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc); struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_udata *udata); +int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index, + int npages, int zap); int mlx5_ib_dereg_mr(struct ib_mr *ibmr); int mlx5_ib_destroy_mr(struct ib_mr *ibmr); struct ib_mr *mlx5_ib_create_mr(struct ib_pd *pd, @@ -533,8 +603,11 @@ int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev); void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev); void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift, int *ncont, int *order); +void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, + int page_shift, size_t offset, size_t num_pages, + __be64 *pas, int access_flags); void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, - int page_shift, __be64 *pas, int umr); + int page_shift, __be64 *pas, int access_flags); void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num); int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq); int mlx5_mr_cache_init(struct mlx5_ib_dev *dev); @@ -544,6 +617,38 @@ void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context); int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask, struct ib_mr_status *mr_status); +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING +extern struct workqueue_struct *mlx5_ib_page_fault_wq; + +int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev); +void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault); +void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp); +int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev); +void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev); +int __init mlx5_ib_odp_init(void); +void mlx5_ib_odp_cleanup(void); +void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp); +void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp); +void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start, + unsigned long end); + +#else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ +static inline int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev) +{ + return 0; +} + +static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp) {} +static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; } +static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev) {} +static inline int mlx5_ib_odp_init(void) { return 0; } +static inline void mlx5_ib_odp_cleanup(void) {} +static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) {} +static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp) {} + +#endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ + static inline void init_query_mad(struct ib_smp *mad) { mad->base_version = 1; @@ -561,4 +666,7 @@ static inline u8 convert_access(int acc) MLX5_PERM_LOCAL_READ; } +#define MLX5_MAX_UMR_SHIFT 16 +#define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT) + #endif /* MLX5_IB_H */ diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 5a80dd993761..32a28bd50b20 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -37,21 +37,34 @@ #include <linux/export.h> #include <linux/delay.h> #include <rdma/ib_umem.h> +#include <rdma/ib_umem_odp.h> +#include <rdma/ib_verbs.h> #include "mlx5_ib.h" enum { MAX_PENDING_REG_MR = 8, }; -enum { - MLX5_UMR_ALIGN = 2048 -}; +#define MLX5_UMR_ALIGN 2048 +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING +static __be64 mlx5_ib_update_mtt_emergency_buffer[ + MLX5_UMR_MTT_MIN_CHUNK_SIZE/sizeof(__be64)] + __aligned(MLX5_UMR_ALIGN); +static DEFINE_MUTEX(mlx5_ib_update_mtt_emergency_buffer_mutex); +#endif + +static int clean_mr(struct mlx5_ib_mr *mr); -static __be64 *mr_align(__be64 *ptr, int align) +static int destroy_mkey(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) { - unsigned long mask = align - 1; + int err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmr); - return (__be64 *)(((unsigned long)ptr + mask) & ~mask); +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + /* Wait until all page fault handlers using the mr complete. */ + synchronize_srcu(&dev->mr_srcu); +#endif + + return err; } static int order2idx(struct mlx5_ib_dev *dev, int order) @@ -146,7 +159,7 @@ static int add_keys(struct mlx5_ib_dev *dev, int c, int num) mr->order = ent->order; mr->umred = 1; mr->dev = dev; - in->seg.status = 1 << 6; + in->seg.status = MLX5_MKEY_STATUS_FREE; in->seg.xlt_oct_size = cpu_to_be32((npages + 1) / 2); in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8); in->seg.flags = MLX5_ACCESS_MODE_MTT | MLX5_PERM_UMR_EN; @@ -191,7 +204,7 @@ static void remove_keys(struct mlx5_ib_dev *dev, int c, int num) ent->cur--; ent->size--; spin_unlock_irq(&ent->lock); - err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmr); + err = destroy_mkey(dev, mr); if (err) mlx5_ib_warn(dev, "failed destroy mkey\n"); else @@ -482,7 +495,7 @@ static void clean_keys(struct mlx5_ib_dev *dev, int c) ent->cur--; ent->size--; spin_unlock_irq(&ent->lock); - err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmr); + err = destroy_mkey(dev, mr); if (err) mlx5_ib_warn(dev, "failed destroy mkey\n"); else @@ -668,7 +681,7 @@ static int get_octo_len(u64 addr, u64 len, int page_size) static int use_umr(int order) { - return order <= 17; + return order <= MLX5_MAX_UMR_SHIFT; } static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr, @@ -678,6 +691,7 @@ static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr, { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct ib_mr *mr = dev->umrc.mr; + struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg; sg->addr = dma; sg->length = ALIGN(sizeof(u64) * n, 64); @@ -692,21 +706,24 @@ static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr, wr->num_sge = 0; wr->opcode = MLX5_IB_WR_UMR; - wr->wr.fast_reg.page_list_len = n; - wr->wr.fast_reg.page_shift = page_shift; - wr->wr.fast_reg.rkey = key; - wr->wr.fast_reg.iova_start = virt_addr; - wr->wr.fast_reg.length = len; - wr->wr.fast_reg.access_flags = access_flags; - wr->wr.fast_reg.page_list = (struct ib_fast_reg_page_list *)pd; + + umrwr->npages = n; + umrwr->page_shift = page_shift; + umrwr->mkey = key; + umrwr->target.virt_addr = virt_addr; + umrwr->length = len; + umrwr->access_flags = access_flags; + umrwr->pd = pd; } static void prep_umr_unreg_wqe(struct mlx5_ib_dev *dev, struct ib_send_wr *wr, u32 key) { - wr->send_flags = MLX5_IB_SEND_UMR_UNREG; + struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg; + + wr->send_flags = MLX5_IB_SEND_UMR_UNREG | MLX5_IB_SEND_UMR_FAIL_IF_FREE; wr->opcode = MLX5_IB_WR_UMR; - wr->wr.fast_reg.rkey = key; + umrwr->mkey = key; } void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context) @@ -742,7 +759,10 @@ static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem, struct ib_send_wr wr, *bad; struct mlx5_ib_mr *mr; struct ib_sge sg; - int size = sizeof(u64) * npages; + int size; + __be64 *mr_pas; + __be64 *pas; + dma_addr_t dma; int err = 0; int i; @@ -761,25 +781,31 @@ static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem, if (!mr) return ERR_PTR(-EAGAIN); - mr->pas = kmalloc(size + MLX5_UMR_ALIGN - 1, GFP_KERNEL); - if (!mr->pas) { + /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes. + * To avoid copying garbage after the pas array, we allocate + * a little more. */ + size = ALIGN(sizeof(u64) * npages, MLX5_UMR_MTT_ALIGNMENT); + mr_pas = kmalloc(size + MLX5_UMR_ALIGN - 1, GFP_KERNEL); + if (!mr_pas) { err = -ENOMEM; goto free_mr; } - mlx5_ib_populate_pas(dev, umem, page_shift, - mr_align(mr->pas, MLX5_UMR_ALIGN), 1); + pas = PTR_ALIGN(mr_pas, MLX5_UMR_ALIGN); + mlx5_ib_populate_pas(dev, umem, page_shift, pas, MLX5_IB_MTT_PRESENT); + /* Clear padding after the actual pages. */ + memset(pas + npages, 0, size - npages * sizeof(u64)); - mr->dma = dma_map_single(ddev, mr_align(mr->pas, MLX5_UMR_ALIGN), size, - DMA_TO_DEVICE); - if (dma_mapping_error(ddev, mr->dma)) { + dma = dma_map_single(ddev, pas, size, DMA_TO_DEVICE); + if (dma_mapping_error(ddev, dma)) { err = -ENOMEM; goto free_pas; } memset(&wr, 0, sizeof(wr)); wr.wr_id = (u64)(unsigned long)&umr_context; - prep_umr_reg_wqe(pd, &wr, &sg, mr->dma, npages, mr->mmr.key, page_shift, virt_addr, len, access_flags); + prep_umr_reg_wqe(pd, &wr, &sg, dma, npages, mr->mmr.key, page_shift, + virt_addr, len, access_flags); mlx5_ib_init_umr_context(&umr_context); down(&umrc->sem); @@ -799,12 +825,14 @@ static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem, mr->mmr.size = len; mr->mmr.pd = to_mpd(pd)->pdn; + mr->live = 1; + unmap_dma: up(&umrc->sem); - dma_unmap_single(ddev, mr->dma, size, DMA_TO_DEVICE); + dma_unmap_single(ddev, dma, size, DMA_TO_DEVICE); free_pas: - kfree(mr->pas); + kfree(mr_pas); free_mr: if (err) { @@ -815,6 +843,128 @@ free_mr: return mr; } +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING +int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index, int npages, + int zap) +{ + struct mlx5_ib_dev *dev = mr->dev; + struct device *ddev = dev->ib_dev.dma_device; + struct umr_common *umrc = &dev->umrc; + struct mlx5_ib_umr_context umr_context; + struct ib_umem *umem = mr->umem; + int size; + __be64 *pas; + dma_addr_t dma; + struct ib_send_wr wr, *bad; + struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr.wr.fast_reg; + struct ib_sge sg; + int err = 0; + const int page_index_alignment = MLX5_UMR_MTT_ALIGNMENT / sizeof(u64); + const int page_index_mask = page_index_alignment - 1; + size_t pages_mapped = 0; + size_t pages_to_map = 0; + size_t pages_iter = 0; + int use_emergency_buf = 0; + + /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes, + * so we need to align the offset and length accordingly */ + if (start_page_index & page_index_mask) { + npages += start_page_index & page_index_mask; + start_page_index &= ~page_index_mask; + } + + pages_to_map = ALIGN(npages, page_index_alignment); + + if (start_page_index + pages_to_map > MLX5_MAX_UMR_PAGES) + return -EINVAL; + + size = sizeof(u64) * pages_to_map; + size = min_t(int, PAGE_SIZE, size); + /* We allocate with GFP_ATOMIC to avoid recursion into page-reclaim + * code, when we are called from an invalidation. The pas buffer must + * be 2k-aligned for Connect-IB. */ + pas = (__be64 *)get_zeroed_page(GFP_ATOMIC); + if (!pas) { + mlx5_ib_warn(dev, "unable to allocate memory during MTT update, falling back to slower chunked mechanism.\n"); + pas = mlx5_ib_update_mtt_emergency_buffer; + size = MLX5_UMR_MTT_MIN_CHUNK_SIZE; + use_emergency_buf = 1; + mutex_lock(&mlx5_ib_update_mtt_emergency_buffer_mutex); + memset(pas, 0, size); + } + pages_iter = size / sizeof(u64); + dma = dma_map_single(ddev, pas, size, DMA_TO_DEVICE); + if (dma_mapping_error(ddev, dma)) { + mlx5_ib_err(dev, "unable to map DMA during MTT update.\n"); + err = -ENOMEM; + goto free_pas; + } + + for (pages_mapped = 0; + pages_mapped < pages_to_map && !err; + pages_mapped += pages_iter, start_page_index += pages_iter) { + dma_sync_single_for_cpu(ddev, dma, size, DMA_TO_DEVICE); + + npages = min_t(size_t, + pages_iter, + ib_umem_num_pages(umem) - start_page_index); + + if (!zap) { + __mlx5_ib_populate_pas(dev, umem, PAGE_SHIFT, + start_page_index, npages, pas, + MLX5_IB_MTT_PRESENT); + /* Clear padding after the pages brought from the + * umem. */ + memset(pas + npages, 0, size - npages * sizeof(u64)); + } + + dma_sync_single_for_device(ddev, dma, size, DMA_TO_DEVICE); + + memset(&wr, 0, sizeof(wr)); + wr.wr_id = (u64)(unsigned long)&umr_context; + + sg.addr = dma; + sg.length = ALIGN(npages * sizeof(u64), + MLX5_UMR_MTT_ALIGNMENT); + sg.lkey = dev->umrc.mr->lkey; + + wr.send_flags = MLX5_IB_SEND_UMR_FAIL_IF_FREE | + MLX5_IB_SEND_UMR_UPDATE_MTT; + wr.sg_list = &sg; + wr.num_sge = 1; + wr.opcode = MLX5_IB_WR_UMR; + umrwr->npages = sg.length / sizeof(u64); + umrwr->page_shift = PAGE_SHIFT; + umrwr->mkey = mr->mmr.key; + umrwr->target.offset = start_page_index; + + mlx5_ib_init_umr_context(&umr_context); + down(&umrc->sem); + err = ib_post_send(umrc->qp, &wr, &bad); + if (err) { + mlx5_ib_err(dev, "UMR post send failed, err %d\n", err); + } else { + wait_for_completion(&umr_context.done); + if (umr_context.status != IB_WC_SUCCESS) { + mlx5_ib_err(dev, "UMR completion failed, code %d\n", + umr_context.status); + err = -EFAULT; + } + } + up(&umrc->sem); + } + dma_unmap_single(ddev, dma, size, DMA_TO_DEVICE); + +free_pas: + if (!use_emergency_buf) + free_page((unsigned long)pas); + else + mutex_unlock(&mlx5_ib_update_mtt_emergency_buffer_mutex); + + return err; +} +#endif + static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, u64 virt_addr, u64 length, struct ib_umem *umem, int npages, int page_shift, @@ -825,6 +975,8 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, u64 virt_addr, struct mlx5_ib_mr *mr; int inlen; int err; + bool pg_cap = !!(dev->mdev->caps.gen.flags & + MLX5_DEV_CAP_FLAG_ON_DMND_PG); mr = kzalloc(sizeof(*mr), GFP_KERNEL); if (!mr) @@ -836,8 +988,12 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, u64 virt_addr, err = -ENOMEM; goto err_1; } - mlx5_ib_populate_pas(dev, umem, page_shift, in->pas, 0); + mlx5_ib_populate_pas(dev, umem, page_shift, in->pas, + pg_cap ? MLX5_IB_MTT_PRESENT : 0); + /* The MLX5_MKEY_INBOX_PG_ACCESS bit allows setting the access flags + * in the page list submitted with the command. */ + in->flags = pg_cap ? cpu_to_be32(MLX5_MKEY_INBOX_PG_ACCESS) : 0; in->seg.flags = convert_access(access_flags) | MLX5_ACCESS_MODE_MTT; in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn); @@ -856,6 +1012,7 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, u64 virt_addr, goto err_2; } mr->umem = umem; + mr->live = 1; kvfree(in); mlx5_ib_dbg(dev, "mkey = 0x%x\n", mr->mmr.key); @@ -910,6 +1067,10 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mlx5_ib_dbg(dev, "cache empty for order %d", order); mr = NULL; } + } else if (access_flags & IB_ACCESS_ON_DEMAND) { + err = -EINVAL; + pr_err("Got MR registration for ODP MR > 512MB, not supported for Connect-IB"); + goto error; } if (!mr) @@ -925,16 +1086,51 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mr->umem = umem; mr->npages = npages; - spin_lock(&dev->mr_lock); - dev->mdev->priv.reg_pages += npages; - spin_unlock(&dev->mr_lock); + atomic_add(npages, &dev->mdev->priv.reg_pages); mr->ibmr.lkey = mr->mmr.key; mr->ibmr.rkey = mr->mmr.key; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + if (umem->odp_data) { + /* + * This barrier prevents the compiler from moving the + * setting of umem->odp_data->private to point to our + * MR, before reg_umr finished, to ensure that the MR + * initialization have finished before starting to + * handle invalidations. + */ + smp_wmb(); + mr->umem->odp_data->private = mr; + /* + * Make sure we will see the new + * umem->odp_data->private value in the invalidation + * routines, before we can get page faults on the + * MR. Page faults can happen once we put the MR in + * the tree, below this line. Without the barrier, + * there can be a fault handling and an invalidation + * before umem->odp_data->private == mr is visible to + * the invalidation handler. + */ + smp_wmb(); + } +#endif + return &mr->ibmr; error: + /* + * Destroy the umem *before* destroying the MR, to ensure we + * will not have any in-flight notifiers when destroying the + * MR. + * + * As the MR is completely invalid to begin with, and this + * error path is only taken if we can't push the mr entry into + * the pagefault tree, this is safe. + */ + ib_umem_release(umem); + /* Kill the MR, and return an error code. */ + clean_mr(mr); return ERR_PTR(err); } @@ -971,17 +1167,14 @@ error: return err; } -int mlx5_ib_dereg_mr(struct ib_mr *ibmr) +static int clean_mr(struct mlx5_ib_mr *mr) { - struct mlx5_ib_dev *dev = to_mdev(ibmr->device); - struct mlx5_ib_mr *mr = to_mmr(ibmr); - struct ib_umem *umem = mr->umem; - int npages = mr->npages; + struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device); int umred = mr->umred; int err; if (!umred) { - err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmr); + err = destroy_mkey(dev, mr); if (err) { mlx5_ib_warn(dev, "failed to destroy mkey 0x%x (%d)\n", mr->mmr.key, err); @@ -996,15 +1189,47 @@ int mlx5_ib_dereg_mr(struct ib_mr *ibmr) free_cached_mr(dev, mr); } - if (umem) { + if (!umred) + kfree(mr); + + return 0; +} + +int mlx5_ib_dereg_mr(struct ib_mr *ibmr) +{ + struct mlx5_ib_dev *dev = to_mdev(ibmr->device); + struct mlx5_ib_mr *mr = to_mmr(ibmr); + int npages = mr->npages; + struct ib_umem *umem = mr->umem; + +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + if (umem && umem->odp_data) { + /* Prevent new page faults from succeeding */ + mr->live = 0; + /* Wait for all running page-fault handlers to finish. */ + synchronize_srcu(&dev->mr_srcu); + /* Destroy all page mappings */ + mlx5_ib_invalidate_range(umem, ib_umem_start(umem), + ib_umem_end(umem)); + /* + * We kill the umem before the MR for ODP, + * so that there will not be any invalidations in + * flight, looking at the *mr struct. + */ ib_umem_release(umem); - spin_lock(&dev->mr_lock); - dev->mdev->priv.reg_pages -= npages; - spin_unlock(&dev->mr_lock); + atomic_sub(npages, &dev->mdev->priv.reg_pages); + + /* Avoid double-freeing the umem. */ + umem = NULL; } +#endif - if (!umred) - kfree(mr); + clean_mr(mr); + + if (umem) { + ib_umem_release(umem); + atomic_sub(npages, &dev->mdev->priv.reg_pages); + } return 0; } @@ -1028,7 +1253,7 @@ struct ib_mr *mlx5_ib_create_mr(struct ib_pd *pd, goto err_free; } - in->seg.status = 1 << 6; /* free */ + in->seg.status = MLX5_MKEY_STATUS_FREE; in->seg.xlt_oct_size = cpu_to_be32(ndescs); in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8); in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn); @@ -1113,7 +1338,7 @@ int mlx5_ib_destroy_mr(struct ib_mr *ibmr) kfree(mr->sig); } - err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmr); + err = destroy_mkey(dev, mr); if (err) { mlx5_ib_warn(dev, "failed to destroy mkey 0x%x (%d)\n", mr->mmr.key, err); @@ -1143,7 +1368,7 @@ struct ib_mr *mlx5_ib_alloc_fast_reg_mr(struct ib_pd *pd, goto err_free; } - in->seg.status = 1 << 6; /* free */ + in->seg.status = MLX5_MKEY_STATUS_FREE; in->seg.xlt_oct_size = cpu_to_be32((max_page_list_len + 1) / 2); in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8); in->seg.flags = MLX5_PERM_UMR_EN | MLX5_ACCESS_MODE_MTT; diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c new file mode 100644 index 000000000000..a2c541c4809a --- /dev/null +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -0,0 +1,798 @@ +/* + * Copyright (c) 2014 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <rdma/ib_umem.h> +#include <rdma/ib_umem_odp.h> + +#include "mlx5_ib.h" + +#define MAX_PREFETCH_LEN (4*1024*1024U) + +/* Timeout in ms to wait for an active mmu notifier to complete when handling + * a pagefault. */ +#define MMU_NOTIFIER_TIMEOUT 1000 + +struct workqueue_struct *mlx5_ib_page_fault_wq; + +void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start, + unsigned long end) +{ + struct mlx5_ib_mr *mr; + const u64 umr_block_mask = (MLX5_UMR_MTT_ALIGNMENT / sizeof(u64)) - 1; + u64 idx = 0, blk_start_idx = 0; + int in_block = 0; + u64 addr; + + if (!umem || !umem->odp_data) { + pr_err("invalidation called on NULL umem or non-ODP umem\n"); + return; + } + + mr = umem->odp_data->private; + + if (!mr || !mr->ibmr.pd) + return; + + start = max_t(u64, ib_umem_start(umem), start); + end = min_t(u64, ib_umem_end(umem), end); + + /* + * Iteration one - zap the HW's MTTs. The notifiers_count ensures that + * while we are doing the invalidation, no page fault will attempt to + * overwrite the same MTTs. Concurent invalidations might race us, + * but they will write 0s as well, so no difference in the end result. + */ + + for (addr = start; addr < end; addr += (u64)umem->page_size) { + idx = (addr - ib_umem_start(umem)) / PAGE_SIZE; + /* + * Strive to write the MTTs in chunks, but avoid overwriting + * non-existing MTTs. The huristic here can be improved to + * estimate the cost of another UMR vs. the cost of bigger + * UMR. + */ + if (umem->odp_data->dma_list[idx] & + (ODP_READ_ALLOWED_BIT | ODP_WRITE_ALLOWED_BIT)) { + if (!in_block) { + blk_start_idx = idx; + in_block = 1; + } + } else { + u64 umr_offset = idx & umr_block_mask; + + if (in_block && umr_offset == 0) { + mlx5_ib_update_mtt(mr, blk_start_idx, + idx - blk_start_idx, 1); + in_block = 0; + } + } + } + if (in_block) + mlx5_ib_update_mtt(mr, blk_start_idx, idx - blk_start_idx + 1, + 1); + + /* + * We are now sure that the device will not access the + * memory. We can safely unmap it, and mark it as dirty if + * needed. + */ + + ib_umem_odp_unmap_dma_pages(umem, start, end); +} + +#define COPY_ODP_BIT_MLX_TO_IB(reg, ib_caps, field_name, bit_name) do { \ + if (be32_to_cpu(reg.field_name) & MLX5_ODP_SUPPORT_##bit_name) \ + ib_caps->field_name |= IB_ODP_SUPPORT_##bit_name; \ +} while (0) + +int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev) +{ + int err; + struct mlx5_odp_caps hw_caps; + struct ib_odp_caps *caps = &dev->odp_caps; + + memset(caps, 0, sizeof(*caps)); + + if (!(dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_ON_DMND_PG)) + return 0; + + err = mlx5_query_odp_caps(dev->mdev, &hw_caps); + if (err) + goto out; + + caps->general_caps = IB_ODP_SUPPORT; + COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.ud_odp_caps, + SEND); + COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps, + SEND); + COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps, + RECV); + COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps, + WRITE); + COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps, + READ); + +out: + return err; +} + +static struct mlx5_ib_mr *mlx5_ib_odp_find_mr_lkey(struct mlx5_ib_dev *dev, + u32 key) +{ + u32 base_key = mlx5_base_mkey(key); + struct mlx5_core_mr *mmr = __mlx5_mr_lookup(dev->mdev, base_key); + struct mlx5_ib_mr *mr = container_of(mmr, struct mlx5_ib_mr, mmr); + + if (!mmr || mmr->key != key || !mr->live) + return NULL; + + return container_of(mmr, struct mlx5_ib_mr, mmr); +} + +static void mlx5_ib_page_fault_resume(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault, + int error) { + struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device); + int ret = mlx5_core_page_fault_resume(dev->mdev, qp->mqp.qpn, + pfault->mpfault.flags, + error); + if (ret) + pr_err("Failed to resolve the page fault on QP 0x%x\n", + qp->mqp.qpn); +} + +/* + * Handle a single data segment in a page-fault WQE. + * + * Returns number of pages retrieved on success. The caller will continue to + * the next data segment. + * Can return the following error codes: + * -EAGAIN to designate a temporary error. The caller will abort handling the + * page fault and resolve it. + * -EFAULT when there's an error mapping the requested pages. The caller will + * abort the page fault handling and possibly move the QP to an error state. + * On other errors the QP should also be closed with an error. + */ +static int pagefault_single_data_segment(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault, + u32 key, u64 io_virt, size_t bcnt, + u32 *bytes_mapped) +{ + struct mlx5_ib_dev *mib_dev = to_mdev(qp->ibqp.pd->device); + int srcu_key; + unsigned int current_seq; + u64 start_idx; + int npages = 0, ret = 0; + struct mlx5_ib_mr *mr; + u64 access_mask = ODP_READ_ALLOWED_BIT; + + srcu_key = srcu_read_lock(&mib_dev->mr_srcu); + mr = mlx5_ib_odp_find_mr_lkey(mib_dev, key); + /* + * If we didn't find the MR, it means the MR was closed while we were + * handling the ODP event. In this case we return -EFAULT so that the + * QP will be closed. + */ + if (!mr || !mr->ibmr.pd) { + pr_err("Failed to find relevant mr for lkey=0x%06x, probably the MR was destroyed\n", + key); + ret = -EFAULT; + goto srcu_unlock; + } + if (!mr->umem->odp_data) { + pr_debug("skipping non ODP MR (lkey=0x%06x) in page fault handler.\n", + key); + if (bytes_mapped) + *bytes_mapped += + (bcnt - pfault->mpfault.bytes_committed); + goto srcu_unlock; + } + if (mr->ibmr.pd != qp->ibqp.pd) { + pr_err("Page-fault with different PDs for QP and MR.\n"); + ret = -EFAULT; + goto srcu_unlock; + } + + current_seq = ACCESS_ONCE(mr->umem->odp_data->notifiers_seq); + /* + * Ensure the sequence number is valid for some time before we call + * gup. + */ + smp_rmb(); + + /* + * Avoid branches - this code will perform correctly + * in all iterations (in iteration 2 and above, + * bytes_committed == 0). + */ + io_virt += pfault->mpfault.bytes_committed; + bcnt -= pfault->mpfault.bytes_committed; + + start_idx = (io_virt - (mr->mmr.iova & PAGE_MASK)) >> PAGE_SHIFT; + + if (mr->umem->writable) + access_mask |= ODP_WRITE_ALLOWED_BIT; + npages = ib_umem_odp_map_dma_pages(mr->umem, io_virt, bcnt, + access_mask, current_seq); + if (npages < 0) { + ret = npages; + goto srcu_unlock; + } + + if (npages > 0) { + mutex_lock(&mr->umem->odp_data->umem_mutex); + if (!ib_umem_mmu_notifier_retry(mr->umem, current_seq)) { + /* + * No need to check whether the MTTs really belong to + * this MR, since ib_umem_odp_map_dma_pages already + * checks this. + */ + ret = mlx5_ib_update_mtt(mr, start_idx, npages, 0); + } else { + ret = -EAGAIN; + } + mutex_unlock(&mr->umem->odp_data->umem_mutex); + if (ret < 0) { + if (ret != -EAGAIN) + pr_err("Failed to update mkey page tables\n"); + goto srcu_unlock; + } + + if (bytes_mapped) { + u32 new_mappings = npages * PAGE_SIZE - + (io_virt - round_down(io_virt, PAGE_SIZE)); + *bytes_mapped += min_t(u32, new_mappings, bcnt); + } + } + +srcu_unlock: + if (ret == -EAGAIN) { + if (!mr->umem->odp_data->dying) { + struct ib_umem_odp *odp_data = mr->umem->odp_data; + unsigned long timeout = + msecs_to_jiffies(MMU_NOTIFIER_TIMEOUT); + + if (!wait_for_completion_timeout( + &odp_data->notifier_completion, + timeout)) { + pr_warn("timeout waiting for mmu notifier completion\n"); + } + } else { + /* The MR is being killed, kill the QP as well. */ + ret = -EFAULT; + } + } + srcu_read_unlock(&mib_dev->mr_srcu, srcu_key); + pfault->mpfault.bytes_committed = 0; + return ret ? ret : npages; +} + +/** + * Parse a series of data segments for page fault handling. + * + * @qp the QP on which the fault occurred. + * @pfault contains page fault information. + * @wqe points at the first data segment in the WQE. + * @wqe_end points after the end of the WQE. + * @bytes_mapped receives the number of bytes that the function was able to + * map. This allows the caller to decide intelligently whether + * enough memory was mapped to resolve the page fault + * successfully (e.g. enough for the next MTU, or the entire + * WQE). + * @total_wqe_bytes receives the total data size of this WQE in bytes (minus + * the committed bytes). + * + * Returns the number of pages loaded if positive, zero for an empty WQE, or a + * negative error code. + */ +static int pagefault_data_segments(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault, void *wqe, + void *wqe_end, u32 *bytes_mapped, + u32 *total_wqe_bytes, int receive_queue) +{ + int ret = 0, npages = 0; + u64 io_virt; + u32 key; + u32 byte_count; + size_t bcnt; + int inline_segment; + + /* Skip SRQ next-WQE segment. */ + if (receive_queue && qp->ibqp.srq) + wqe += sizeof(struct mlx5_wqe_srq_next_seg); + + if (bytes_mapped) + *bytes_mapped = 0; + if (total_wqe_bytes) + *total_wqe_bytes = 0; + + while (wqe < wqe_end) { + struct mlx5_wqe_data_seg *dseg = wqe; + + io_virt = be64_to_cpu(dseg->addr); + key = be32_to_cpu(dseg->lkey); + byte_count = be32_to_cpu(dseg->byte_count); + inline_segment = !!(byte_count & MLX5_INLINE_SEG); + bcnt = byte_count & ~MLX5_INLINE_SEG; + + if (inline_segment) { + bcnt = bcnt & MLX5_WQE_INLINE_SEG_BYTE_COUNT_MASK; + wqe += ALIGN(sizeof(struct mlx5_wqe_inline_seg) + bcnt, + 16); + } else { + wqe += sizeof(*dseg); + } + + /* receive WQE end of sg list. */ + if (receive_queue && bcnt == 0 && key == MLX5_INVALID_LKEY && + io_virt == 0) + break; + + if (!inline_segment && total_wqe_bytes) { + *total_wqe_bytes += bcnt - min_t(size_t, bcnt, + pfault->mpfault.bytes_committed); + } + + /* A zero length data segment designates a length of 2GB. */ + if (bcnt == 0) + bcnt = 1U << 31; + + if (inline_segment || bcnt <= pfault->mpfault.bytes_committed) { + pfault->mpfault.bytes_committed -= + min_t(size_t, bcnt, + pfault->mpfault.bytes_committed); + continue; + } + + ret = pagefault_single_data_segment(qp, pfault, key, io_virt, + bcnt, bytes_mapped); + if (ret < 0) + break; + npages += ret; + } + + return ret < 0 ? ret : npages; +} + +/* + * Parse initiator WQE. Advances the wqe pointer to point at the + * scatter-gather list, and set wqe_end to the end of the WQE. + */ +static int mlx5_ib_mr_initiator_pfault_handler( + struct mlx5_ib_qp *qp, struct mlx5_ib_pfault *pfault, + void **wqe, void **wqe_end, int wqe_length) +{ + struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device); + struct mlx5_wqe_ctrl_seg *ctrl = *wqe; + u16 wqe_index = pfault->mpfault.wqe.wqe_index; + unsigned ds, opcode; +#if defined(DEBUG) + u32 ctrl_wqe_index, ctrl_qpn; +#endif + + ds = be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_DS_MASK; + if (ds * MLX5_WQE_DS_UNITS > wqe_length) { + mlx5_ib_err(dev, "Unable to read the complete WQE. ds = 0x%x, ret = 0x%x\n", + ds, wqe_length); + return -EFAULT; + } + + if (ds == 0) { + mlx5_ib_err(dev, "Got WQE with zero DS. wqe_index=%x, qpn=%x\n", + wqe_index, qp->mqp.qpn); + return -EFAULT; + } + +#if defined(DEBUG) + ctrl_wqe_index = (be32_to_cpu(ctrl->opmod_idx_opcode) & + MLX5_WQE_CTRL_WQE_INDEX_MASK) >> + MLX5_WQE_CTRL_WQE_INDEX_SHIFT; + if (wqe_index != ctrl_wqe_index) { + mlx5_ib_err(dev, "Got WQE with invalid wqe_index. wqe_index=0x%x, qpn=0x%x ctrl->wqe_index=0x%x\n", + wqe_index, qp->mqp.qpn, + ctrl_wqe_index); + return -EFAULT; + } + + ctrl_qpn = (be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_QPN_MASK) >> + MLX5_WQE_CTRL_QPN_SHIFT; + if (qp->mqp.qpn != ctrl_qpn) { + mlx5_ib_err(dev, "Got WQE with incorrect QP number. wqe_index=0x%x, qpn=0x%x ctrl->qpn=0x%x\n", + wqe_index, qp->mqp.qpn, + ctrl_qpn); + return -EFAULT; + } +#endif /* DEBUG */ + + *wqe_end = *wqe + ds * MLX5_WQE_DS_UNITS; + *wqe += sizeof(*ctrl); + + opcode = be32_to_cpu(ctrl->opmod_idx_opcode) & + MLX5_WQE_CTRL_OPCODE_MASK; + switch (qp->ibqp.qp_type) { + case IB_QPT_RC: + switch (opcode) { + case MLX5_OPCODE_SEND: + case MLX5_OPCODE_SEND_IMM: + case MLX5_OPCODE_SEND_INVAL: + if (!(dev->odp_caps.per_transport_caps.rc_odp_caps & + IB_ODP_SUPPORT_SEND)) + goto invalid_transport_or_opcode; + break; + case MLX5_OPCODE_RDMA_WRITE: + case MLX5_OPCODE_RDMA_WRITE_IMM: + if (!(dev->odp_caps.per_transport_caps.rc_odp_caps & + IB_ODP_SUPPORT_WRITE)) + goto invalid_transport_or_opcode; + *wqe += sizeof(struct mlx5_wqe_raddr_seg); + break; + case MLX5_OPCODE_RDMA_READ: + if (!(dev->odp_caps.per_transport_caps.rc_odp_caps & + IB_ODP_SUPPORT_READ)) + goto invalid_transport_or_opcode; + *wqe += sizeof(struct mlx5_wqe_raddr_seg); + break; + default: + goto invalid_transport_or_opcode; + } + break; + case IB_QPT_UD: + switch (opcode) { + case MLX5_OPCODE_SEND: + case MLX5_OPCODE_SEND_IMM: + if (!(dev->odp_caps.per_transport_caps.ud_odp_caps & + IB_ODP_SUPPORT_SEND)) + goto invalid_transport_or_opcode; + *wqe += sizeof(struct mlx5_wqe_datagram_seg); + break; + default: + goto invalid_transport_or_opcode; + } + break; + default: +invalid_transport_or_opcode: + mlx5_ib_err(dev, "ODP fault on QP of an unsupported opcode or transport. transport: 0x%x opcode: 0x%x.\n", + qp->ibqp.qp_type, opcode); + return -EFAULT; + } + + return 0; +} + +/* + * Parse responder WQE. Advances the wqe pointer to point at the + * scatter-gather list, and set wqe_end to the end of the WQE. + */ +static int mlx5_ib_mr_responder_pfault_handler( + struct mlx5_ib_qp *qp, struct mlx5_ib_pfault *pfault, + void **wqe, void **wqe_end, int wqe_length) +{ + struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device); + struct mlx5_ib_wq *wq = &qp->rq; + int wqe_size = 1 << wq->wqe_shift; + + if (qp->ibqp.srq) { + mlx5_ib_err(dev, "ODP fault on SRQ is not supported\n"); + return -EFAULT; + } + + if (qp->wq_sig) { + mlx5_ib_err(dev, "ODP fault with WQE signatures is not supported\n"); + return -EFAULT; + } + + if (wqe_size > wqe_length) { + mlx5_ib_err(dev, "Couldn't read all of the receive WQE's content\n"); + return -EFAULT; + } + + switch (qp->ibqp.qp_type) { + case IB_QPT_RC: + if (!(dev->odp_caps.per_transport_caps.rc_odp_caps & + IB_ODP_SUPPORT_RECV)) + goto invalid_transport_or_opcode; + break; + default: +invalid_transport_or_opcode: + mlx5_ib_err(dev, "ODP fault on QP of an unsupported transport. transport: 0x%x\n", + qp->ibqp.qp_type); + return -EFAULT; + } + + *wqe_end = *wqe + wqe_size; + + return 0; +} + +static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault) +{ + struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device); + int ret; + void *wqe, *wqe_end; + u32 bytes_mapped, total_wqe_bytes; + char *buffer = NULL; + int resume_with_error = 0; + u16 wqe_index = pfault->mpfault.wqe.wqe_index; + int requestor = pfault->mpfault.flags & MLX5_PFAULT_REQUESTOR; + + buffer = (char *)__get_free_page(GFP_KERNEL); + if (!buffer) { + mlx5_ib_err(dev, "Error allocating memory for IO page fault handling.\n"); + resume_with_error = 1; + goto resolve_page_fault; + } + + ret = mlx5_ib_read_user_wqe(qp, requestor, wqe_index, buffer, + PAGE_SIZE); + if (ret < 0) { + mlx5_ib_err(dev, "Failed reading a WQE following page fault, error=%x, wqe_index=%x, qpn=%x\n", + -ret, wqe_index, qp->mqp.qpn); + resume_with_error = 1; + goto resolve_page_fault; + } + + wqe = buffer; + if (requestor) + ret = mlx5_ib_mr_initiator_pfault_handler(qp, pfault, &wqe, + &wqe_end, ret); + else + ret = mlx5_ib_mr_responder_pfault_handler(qp, pfault, &wqe, + &wqe_end, ret); + if (ret < 0) { + resume_with_error = 1; + goto resolve_page_fault; + } + + if (wqe >= wqe_end) { + mlx5_ib_err(dev, "ODP fault on invalid WQE.\n"); + resume_with_error = 1; + goto resolve_page_fault; + } + + ret = pagefault_data_segments(qp, pfault, wqe, wqe_end, &bytes_mapped, + &total_wqe_bytes, !requestor); + if (ret == -EAGAIN) { + goto resolve_page_fault; + } else if (ret < 0 || total_wqe_bytes > bytes_mapped) { + mlx5_ib_err(dev, "Error getting user pages for page fault. Error: 0x%x\n", + -ret); + resume_with_error = 1; + goto resolve_page_fault; + } + +resolve_page_fault: + mlx5_ib_page_fault_resume(qp, pfault, resume_with_error); + mlx5_ib_dbg(dev, "PAGE FAULT completed. QP 0x%x resume_with_error=%d, flags: 0x%x\n", + qp->mqp.qpn, resume_with_error, pfault->mpfault.flags); + + free_page((unsigned long)buffer); +} + +static int pages_in_range(u64 address, u32 length) +{ + return (ALIGN(address + length, PAGE_SIZE) - + (address & PAGE_MASK)) >> PAGE_SHIFT; +} + +static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault) +{ + struct mlx5_pagefault *mpfault = &pfault->mpfault; + u64 address; + u32 length; + u32 prefetch_len = mpfault->bytes_committed; + int prefetch_activated = 0; + u32 rkey = mpfault->rdma.r_key; + int ret; + + /* The RDMA responder handler handles the page fault in two parts. + * First it brings the necessary pages for the current packet + * (and uses the pfault context), and then (after resuming the QP) + * prefetches more pages. The second operation cannot use the pfault + * context and therefore uses the dummy_pfault context allocated on + * the stack */ + struct mlx5_ib_pfault dummy_pfault = {}; + + dummy_pfault.mpfault.bytes_committed = 0; + + mpfault->rdma.rdma_va += mpfault->bytes_committed; + mpfault->rdma.rdma_op_len -= min(mpfault->bytes_committed, + mpfault->rdma.rdma_op_len); + mpfault->bytes_committed = 0; + + address = mpfault->rdma.rdma_va; + length = mpfault->rdma.rdma_op_len; + + /* For some operations, the hardware cannot tell the exact message + * length, and in those cases it reports zero. Use prefetch + * logic. */ + if (length == 0) { + prefetch_activated = 1; + length = mpfault->rdma.packet_size; + prefetch_len = min(MAX_PREFETCH_LEN, prefetch_len); + } + + ret = pagefault_single_data_segment(qp, pfault, rkey, address, length, + NULL); + if (ret == -EAGAIN) { + /* We're racing with an invalidation, don't prefetch */ + prefetch_activated = 0; + } else if (ret < 0 || pages_in_range(address, length) > ret) { + mlx5_ib_page_fault_resume(qp, pfault, 1); + return; + } + + mlx5_ib_page_fault_resume(qp, pfault, 0); + + /* At this point, there might be a new pagefault already arriving in + * the eq, switch to the dummy pagefault for the rest of the + * processing. We're still OK with the objects being alive as the + * work-queue is being fenced. */ + + if (prefetch_activated) { + ret = pagefault_single_data_segment(qp, &dummy_pfault, rkey, + address, + prefetch_len, + NULL); + if (ret < 0) { + pr_warn("Prefetch failed (ret = %d, prefetch_activated = %d) for QPN %d, address: 0x%.16llx, length = 0x%.16x\n", + ret, prefetch_activated, + qp->ibqp.qp_num, address, prefetch_len); + } + } +} + +void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp, + struct mlx5_ib_pfault *pfault) +{ + u8 event_subtype = pfault->mpfault.event_subtype; + + switch (event_subtype) { + case MLX5_PFAULT_SUBTYPE_WQE: + mlx5_ib_mr_wqe_pfault_handler(qp, pfault); + break; + case MLX5_PFAULT_SUBTYPE_RDMA: + mlx5_ib_mr_rdma_pfault_handler(qp, pfault); + break; + default: + pr_warn("Invalid page fault event subtype: 0x%x\n", + event_subtype); + mlx5_ib_page_fault_resume(qp, pfault, 1); + break; + } +} + +static void mlx5_ib_qp_pfault_action(struct work_struct *work) +{ + struct mlx5_ib_pfault *pfault = container_of(work, + struct mlx5_ib_pfault, + work); + enum mlx5_ib_pagefault_context context = + mlx5_ib_get_pagefault_context(&pfault->mpfault); + struct mlx5_ib_qp *qp = container_of(pfault, struct mlx5_ib_qp, + pagefaults[context]); + mlx5_ib_mr_pfault_handler(qp, pfault); +} + +void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) +{ + unsigned long flags; + + spin_lock_irqsave(&qp->disable_page_faults_lock, flags); + qp->disable_page_faults = 1; + spin_unlock_irqrestore(&qp->disable_page_faults_lock, flags); + + /* + * Note that at this point, we are guarenteed that no more + * work queue elements will be posted to the work queue with + * the QP we are closing. + */ + flush_workqueue(mlx5_ib_page_fault_wq); +} + +void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp) +{ + unsigned long flags; + + spin_lock_irqsave(&qp->disable_page_faults_lock, flags); + qp->disable_page_faults = 0; + spin_unlock_irqrestore(&qp->disable_page_faults_lock, flags); +} + +static void mlx5_ib_pfault_handler(struct mlx5_core_qp *qp, + struct mlx5_pagefault *pfault) +{ + /* + * Note that we will only get one fault event per QP per context + * (responder/initiator, read/write), until we resolve the page fault + * with the mlx5_ib_page_fault_resume command. Since this function is + * called from within the work element, there is no risk of missing + * events. + */ + struct mlx5_ib_qp *mibqp = to_mibqp(qp); + enum mlx5_ib_pagefault_context context = + mlx5_ib_get_pagefault_context(pfault); + struct mlx5_ib_pfault *qp_pfault = &mibqp->pagefaults[context]; + + qp_pfault->mpfault = *pfault; + + /* No need to stop interrupts here since we are in an interrupt */ + spin_lock(&mibqp->disable_page_faults_lock); + if (!mibqp->disable_page_faults) + queue_work(mlx5_ib_page_fault_wq, &qp_pfault->work); + spin_unlock(&mibqp->disable_page_faults_lock); +} + +void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp) +{ + int i; + + qp->disable_page_faults = 1; + spin_lock_init(&qp->disable_page_faults_lock); + + qp->mqp.pfault_handler = mlx5_ib_pfault_handler; + + for (i = 0; i < MLX5_IB_PAGEFAULT_CONTEXTS; ++i) + INIT_WORK(&qp->pagefaults[i].work, mlx5_ib_qp_pfault_action); +} + +int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) +{ + int ret; + + ret = init_srcu_struct(&ibdev->mr_srcu); + if (ret) + return ret; + + return 0; +} + +void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev) +{ + cleanup_srcu_struct(&ibdev->mr_srcu); +} + +int __init mlx5_ib_odp_init(void) +{ + mlx5_ib_page_fault_wq = + create_singlethread_workqueue("mlx5_ib_page_faults"); + if (!mlx5_ib_page_fault_wq) + return -ENOMEM; + + return 0; +} + +void mlx5_ib_odp_cleanup(void) +{ + destroy_workqueue(mlx5_ib_page_fault_wq); +} diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index 1cae1c7132b4..be0cd358b080 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -70,15 +70,6 @@ static const u32 mlx5_ib_opcode[] = { [MLX5_IB_WR_UMR] = MLX5_OPCODE_UMR, }; -struct umr_wr { - u64 virt_addr; - struct ib_pd *pd; - unsigned int page_shift; - unsigned int npages; - u32 length; - int access_flags; - u32 mkey; -}; static int is_qp0(enum ib_qp_type qp_type) { @@ -110,6 +101,77 @@ void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n) return get_wqe(qp, qp->sq.offset + (n << MLX5_IB_SQ_STRIDE)); } +/** + * mlx5_ib_read_user_wqe() - Copy a user-space WQE to kernel space. + * + * @qp: QP to copy from. + * @send: copy from the send queue when non-zero, use the receive queue + * otherwise. + * @wqe_index: index to start copying from. For send work queues, the + * wqe_index is in units of MLX5_SEND_WQE_BB. + * For receive work queue, it is the number of work queue + * element in the queue. + * @buffer: destination buffer. + * @length: maximum number of bytes to copy. + * + * Copies at least a single WQE, but may copy more data. + * + * Return: the number of bytes copied, or an error code. + */ +int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index, + void *buffer, u32 length) +{ + struct ib_device *ibdev = qp->ibqp.device; + struct mlx5_ib_dev *dev = to_mdev(ibdev); + struct mlx5_ib_wq *wq = send ? &qp->sq : &qp->rq; + size_t offset; + size_t wq_end; + struct ib_umem *umem = qp->umem; + u32 first_copy_length; + int wqe_length; + int ret; + + if (wq->wqe_cnt == 0) { + mlx5_ib_dbg(dev, "mlx5_ib_read_user_wqe for a QP with wqe_cnt == 0. qp_type: 0x%x\n", + qp->ibqp.qp_type); + return -EINVAL; + } + + offset = wq->offset + ((wqe_index % wq->wqe_cnt) << wq->wqe_shift); + wq_end = wq->offset + (wq->wqe_cnt << wq->wqe_shift); + + if (send && length < sizeof(struct mlx5_wqe_ctrl_seg)) + return -EINVAL; + + if (offset > umem->length || + (send && offset + sizeof(struct mlx5_wqe_ctrl_seg) > umem->length)) + return -EINVAL; + + first_copy_length = min_t(u32, offset + length, wq_end) - offset; + ret = ib_umem_copy_from(buffer, umem, offset, first_copy_length); + if (ret) + return ret; + + if (send) { + struct mlx5_wqe_ctrl_seg *ctrl = buffer; + int ds = be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_DS_MASK; + + wqe_length = ds * MLX5_WQE_DS_UNITS; + } else { + wqe_length = 1 << wq->wqe_shift; + } + + if (wqe_length <= first_copy_length) + return first_copy_length; + + ret = ib_umem_copy_from(buffer + first_copy_length, umem, wq->offset, + wqe_length - first_copy_length); + if (ret) + return ret; + + return wqe_length; +} + static void mlx5_ib_qp_event(struct mlx5_core_qp *qp, int type) { struct ib_qp *ibqp = &to_mibqp(qp)->ibqp; @@ -814,6 +876,8 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd, int inlen = sizeof(*in); int err; + mlx5_ib_odp_create_qp(qp); + gen = &dev->mdev->caps.gen; mutex_init(&qp->mutex); spin_lock_init(&qp->sq.lock); @@ -1098,11 +1162,13 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp) in = kzalloc(sizeof(*in), GFP_KERNEL); if (!in) return; - if (qp->state != IB_QPS_RESET) + if (qp->state != IB_QPS_RESET) { + mlx5_ib_qp_disable_pagefaults(qp); if (mlx5_core_qp_modify(dev->mdev, to_mlx5_state(qp->state), MLX5_QP_STATE_RST, in, sizeof(*in), &qp->mqp)) mlx5_ib_warn(dev, "mlx5_ib: modify QP %06x to RESET failed\n", qp->mqp.qpn); + } get_cqs(qp, &send_cq, &recv_cq); @@ -1650,6 +1716,15 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, if (mlx5_st < 0) goto out; + /* If moving to a reset or error state, we must disable page faults on + * this QP and flush all current page faults. Otherwise a stale page + * fault may attempt to work on this QP after it is reset and moved + * again to RTS, and may cause the driver and the device to get out of + * sync. */ + if (cur_state != IB_QPS_RESET && cur_state != IB_QPS_ERR && + (new_state == IB_QPS_RESET || new_state == IB_QPS_ERR)) + mlx5_ib_qp_disable_pagefaults(qp); + optpar = ib_mask_to_mlx5_opt(attr_mask); optpar &= opt_mask[mlx5_cur][mlx5_new][mlx5_st]; in->optparam = cpu_to_be32(optpar); @@ -1659,6 +1734,9 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp, if (err) goto out; + if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) + mlx5_ib_qp_enable_pagefaults(qp); + qp->state = new_state; if (attr_mask & IB_QP_ACCESS_FLAGS) @@ -1848,37 +1926,70 @@ static void set_frwr_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr, umr->mkey_mask = frwr_mkey_mask(); } +static __be64 get_umr_reg_mr_mask(void) +{ + u64 result; + + result = MLX5_MKEY_MASK_LEN | + MLX5_MKEY_MASK_PAGE_SIZE | + MLX5_MKEY_MASK_START_ADDR | + MLX5_MKEY_MASK_PD | + MLX5_MKEY_MASK_LR | + MLX5_MKEY_MASK_LW | + MLX5_MKEY_MASK_KEY | + MLX5_MKEY_MASK_RR | + MLX5_MKEY_MASK_RW | + MLX5_MKEY_MASK_A | + MLX5_MKEY_MASK_FREE; + + return cpu_to_be64(result); +} + +static __be64 get_umr_unreg_mr_mask(void) +{ + u64 result; + + result = MLX5_MKEY_MASK_FREE; + + return cpu_to_be64(result); +} + +static __be64 get_umr_update_mtt_mask(void) +{ + u64 result; + + result = MLX5_MKEY_MASK_FREE; + + return cpu_to_be64(result); +} + static void set_reg_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr, struct ib_send_wr *wr) { - struct umr_wr *umrwr = (struct umr_wr *)&wr->wr.fast_reg; - u64 mask; + struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg; memset(umr, 0, sizeof(*umr)); + if (wr->send_flags & MLX5_IB_SEND_UMR_FAIL_IF_FREE) + umr->flags = MLX5_UMR_CHECK_FREE; /* fail if free */ + else + umr->flags = MLX5_UMR_CHECK_NOT_FREE; /* fail if not free */ + if (!(wr->send_flags & MLX5_IB_SEND_UMR_UNREG)) { - umr->flags = 1 << 5; /* fail if not free */ umr->klm_octowords = get_klm_octo(umrwr->npages); - mask = MLX5_MKEY_MASK_LEN | - MLX5_MKEY_MASK_PAGE_SIZE | - MLX5_MKEY_MASK_START_ADDR | - MLX5_MKEY_MASK_PD | - MLX5_MKEY_MASK_LR | - MLX5_MKEY_MASK_LW | - MLX5_MKEY_MASK_KEY | - MLX5_MKEY_MASK_RR | - MLX5_MKEY_MASK_RW | - MLX5_MKEY_MASK_A | - MLX5_MKEY_MASK_FREE; - umr->mkey_mask = cpu_to_be64(mask); + if (wr->send_flags & MLX5_IB_SEND_UMR_UPDATE_MTT) { + umr->mkey_mask = get_umr_update_mtt_mask(); + umr->bsf_octowords = get_klm_octo(umrwr->target.offset); + umr->flags |= MLX5_UMR_TRANSLATION_OFFSET_EN; + } else { + umr->mkey_mask = get_umr_reg_mr_mask(); + } } else { - umr->flags = 2 << 5; /* fail if free */ - mask = MLX5_MKEY_MASK_FREE; - umr->mkey_mask = cpu_to_be64(mask); + umr->mkey_mask = get_umr_unreg_mr_mask(); } if (!wr->num_sge) - umr->flags |= (1 << 7); /* inline */ + umr->flags |= MLX5_UMR_INLINE; } static u8 get_umr_flags(int acc) @@ -1895,7 +2006,7 @@ static void set_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr, { memset(seg, 0, sizeof(*seg)); if (li) { - seg->status = 1 << 6; + seg->status = MLX5_MKEY_STATUS_FREE; return; } @@ -1912,19 +2023,23 @@ static void set_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr, static void set_reg_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr) { + struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg; + memset(seg, 0, sizeof(*seg)); if (wr->send_flags & MLX5_IB_SEND_UMR_UNREG) { - seg->status = 1 << 6; + seg->status = MLX5_MKEY_STATUS_FREE; return; } - seg->flags = convert_access(wr->wr.fast_reg.access_flags); - seg->flags_pd = cpu_to_be32(to_mpd((struct ib_pd *)wr->wr.fast_reg.page_list)->pdn); - seg->start_addr = cpu_to_be64(wr->wr.fast_reg.iova_start); - seg->len = cpu_to_be64(wr->wr.fast_reg.length); - seg->log2_page_size = wr->wr.fast_reg.page_shift; + seg->flags = convert_access(umrwr->access_flags); + if (!(wr->send_flags & MLX5_IB_SEND_UMR_UPDATE_MTT)) { + seg->flags_pd = cpu_to_be32(to_mpd(umrwr->pd)->pdn); + seg->start_addr = cpu_to_be64(umrwr->target.virt_addr); + } + seg->len = cpu_to_be64(umrwr->length); + seg->log2_page_size = umrwr->page_shift; seg->qpn_mkey7_0 = cpu_to_be32(0xffffff00 | - mlx5_mkey_variant(wr->wr.fast_reg.rkey)); + mlx5_mkey_variant(umrwr->mkey)); } static void set_frwr_pages(struct mlx5_wqe_data_seg *dseg, @@ -2927,6 +3042,14 @@ int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr int mlx5_state; int err = 0; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + /* + * Wait for any outstanding page faults, in case the user frees memory + * based upon this query's result. + */ + flush_workqueue(mlx5_ib_page_fault_wq); +#endif + mutex_lock(&qp->mutex); outb = kzalloc(sizeof(*outb), GFP_KERNEL); if (!outb) { diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c index fef067c959fc..c0d0296e7a00 100644 --- a/drivers/infiniband/hw/nes/nes_verbs.c +++ b/drivers/infiniband/hw/nes/nes_verbs.c @@ -2341,9 +2341,9 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, nes_debug(NES_DBG_MR, "User base = 0x%lX, Virt base = 0x%lX, length = %u," " offset = %u, page size = %u.\n", (unsigned long int)start, (unsigned long int)virt, (u32)length, - region->offset, region->page_size); + ib_umem_offset(region), region->page_size); - skip_pages = ((u32)region->offset) >> 12; + skip_pages = ((u32)ib_umem_offset(region)) >> 12; if (ib_copy_from_udata(&req, udata, sizeof(req))) { ib_umem_release(region); @@ -2408,7 +2408,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, region_length -= skip_pages << 12; for (page_index = skip_pages; page_index < chunk_pages; page_index++) { skip_pages = 0; - if ((page_count != 0) && (page_count<<12)-(region->offset&(4096-1)) >= region->length) + if ((page_count != 0) && (page_count << 12) - (ib_umem_offset(region) & (4096 - 1)) >= region->length) goto enough_pages; if ((page_count&0x01FF) == 0) { if (page_count >= 1024 * 512) { diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c index ac02ce4e8040..f3cc8c9e65ae 100644 --- a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c +++ b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c @@ -96,7 +96,6 @@ struct ib_ah *ocrdma_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr) struct ocrdma_pd *pd = get_ocrdma_pd(ibpd); struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device); union ib_gid sgid; - u8 zmac[ETH_ALEN]; if (!(attr->ah_flags & IB_AH_GRH)) return ERR_PTR(-EINVAL); @@ -118,9 +117,7 @@ struct ib_ah *ocrdma_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr) goto av_conf_err; } - memset(&zmac, 0, ETH_ALEN); - if (pd->uctx && - memcmp(attr->dmac, &zmac, ETH_ALEN)) { + if (pd->uctx) { status = rdma_addr_find_dmac_by_grh(&sgid, &attr->grh.dgid, attr->dmac, &attr->vlan_id); if (status) { diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c index 4c68305ee781..fb8d8c4dfbb9 100644 --- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c +++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c @@ -805,7 +805,7 @@ struct ib_mr *ocrdma_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len, goto umem_err; mr->hwmr.pbe_size = mr->umem->page_size; - mr->hwmr.fbo = mr->umem->offset; + mr->hwmr.fbo = ib_umem_offset(mr->umem); mr->hwmr.va = usr_addr; mr->hwmr.len = len; mr->hwmr.remote_wr = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0; @@ -1410,6 +1410,8 @@ int ocrdma_query_qp(struct ib_qp *ibqp, mutex_unlock(&dev->dev_lock); if (status) goto mbx_err; + if (qp->qp_type == IB_QPT_UD) + qp_attr->qkey = params.qkey; qp_attr->qp_state = get_ibqp_state(IB_QPS_INIT); qp_attr->cur_qp_state = get_ibqp_state(IB_QPS_INIT); qp_attr->path_mtu = diff --git a/drivers/infiniband/hw/qib/qib_mr.c b/drivers/infiniband/hw/qib/qib_mr.c index 9bbb55347cc1..a77fb4fb14e4 100644 --- a/drivers/infiniband/hw/qib/qib_mr.c +++ b/drivers/infiniband/hw/qib/qib_mr.c @@ -258,7 +258,7 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mr->mr.user_base = start; mr->mr.iova = virt_addr; mr->mr.length = length; - mr->mr.offset = umem->offset; + mr->mr.offset = ib_umem_offset(umem); mr->mr.access_flags = mr_access_flags; mr->umem = umem; diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h index d7562beb5423..8ba80a6d3a46 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib.h +++ b/drivers/infiniband/ulp/ipoib/ipoib.h @@ -98,9 +98,15 @@ enum { IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */ IPOIB_MCAST_FLAG_SENDONLY = 1, - IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */ + /* + * For IPOIB_MCAST_FLAG_BUSY + * When set, in flight join and mcast->mc is unreliable + * When clear and mcast->mc IS_ERR_OR_NULL, need to restart or + * haven't started yet + * When clear and mcast->mc is valid pointer, join was successful + */ + IPOIB_MCAST_FLAG_BUSY = 2, IPOIB_MCAST_FLAG_ATTACHED = 3, - IPOIB_MCAST_JOIN_STARTED = 4, MAX_SEND_CQE = 16, IPOIB_CM_COPYBREAK = 256, @@ -317,6 +323,7 @@ struct ipoib_dev_priv { struct list_head multicast_list; struct rb_root multicast_tree; + struct workqueue_struct *wq; struct delayed_work mcast_task; struct work_struct carrier_on_task; struct work_struct flush_light; @@ -477,10 +484,10 @@ void ipoib_ib_dev_flush_heavy(struct work_struct *work); void ipoib_pkey_event(struct work_struct *work); void ipoib_ib_dev_cleanup(struct net_device *dev); -int ipoib_ib_dev_open(struct net_device *dev, int flush); +int ipoib_ib_dev_open(struct net_device *dev); int ipoib_ib_dev_up(struct net_device *dev); -int ipoib_ib_dev_down(struct net_device *dev, int flush); -int ipoib_ib_dev_stop(struct net_device *dev, int flush); +int ipoib_ib_dev_down(struct net_device *dev); +int ipoib_ib_dev_stop(struct net_device *dev); void ipoib_pkey_dev_check_presence(struct net_device *dev); int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port); @@ -492,7 +499,7 @@ void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb); void ipoib_mcast_restart_task(struct work_struct *work); int ipoib_mcast_start_thread(struct net_device *dev); -int ipoib_mcast_stop_thread(struct net_device *dev, int flush); +int ipoib_mcast_stop_thread(struct net_device *dev); void ipoib_mcast_dev_down(struct net_device *dev); void ipoib_mcast_dev_flush(struct net_device *dev); diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c index 933efcea0d03..56959adb6c7d 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c @@ -474,7 +474,7 @@ static int ipoib_cm_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *even } spin_lock_irq(&priv->lock); - queue_delayed_work(ipoib_workqueue, + queue_delayed_work(priv->wq, &priv->cm.stale_task, IPOIB_CM_RX_DELAY); /* Add this entry to passive ids list head, but do not re-add it * if IB_EVENT_QP_LAST_WQE_REACHED has moved it to flush list. */ @@ -576,7 +576,7 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) spin_lock_irqsave(&priv->lock, flags); list_splice_init(&priv->cm.rx_drain_list, &priv->cm.rx_reap_list); ipoib_cm_start_rx_drain(priv); - queue_work(ipoib_workqueue, &priv->cm.rx_reap_task); + queue_work(priv->wq, &priv->cm.rx_reap_task); spin_unlock_irqrestore(&priv->lock, flags); } else ipoib_warn(priv, "cm recv completion event with wrid %d (> %d)\n", @@ -603,7 +603,7 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) spin_lock_irqsave(&priv->lock, flags); list_move(&p->list, &priv->cm.rx_reap_list); spin_unlock_irqrestore(&priv->lock, flags); - queue_work(ipoib_workqueue, &priv->cm.rx_reap_task); + queue_work(priv->wq, &priv->cm.rx_reap_task); } return; } @@ -827,7 +827,7 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) { list_move(&tx->list, &priv->cm.reap_list); - queue_work(ipoib_workqueue, &priv->cm.reap_task); + queue_work(priv->wq, &priv->cm.reap_task); } clear_bit(IPOIB_FLAG_OPER_UP, &tx->flags); @@ -1255,7 +1255,7 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id, if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) { list_move(&tx->list, &priv->cm.reap_list); - queue_work(ipoib_workqueue, &priv->cm.reap_task); + queue_work(priv->wq, &priv->cm.reap_task); } spin_unlock_irqrestore(&priv->lock, flags); @@ -1284,7 +1284,7 @@ struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path tx->dev = dev; list_add(&tx->list, &priv->cm.start_list); set_bit(IPOIB_FLAG_INITIALIZED, &tx->flags); - queue_work(ipoib_workqueue, &priv->cm.start_task); + queue_work(priv->wq, &priv->cm.start_task); return tx; } @@ -1295,7 +1295,7 @@ void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx) if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) { spin_lock_irqsave(&priv->lock, flags); list_move(&tx->list, &priv->cm.reap_list); - queue_work(ipoib_workqueue, &priv->cm.reap_task); + queue_work(priv->wq, &priv->cm.reap_task); ipoib_dbg(priv, "Reap connection for gid %pI6\n", tx->neigh->daddr + 4); tx->neigh = NULL; @@ -1417,7 +1417,7 @@ void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb, skb_queue_tail(&priv->cm.skb_queue, skb); if (e) - queue_work(ipoib_workqueue, &priv->cm.skb_task); + queue_work(priv->wq, &priv->cm.skb_task); } static void ipoib_cm_rx_reap(struct work_struct *work) @@ -1450,7 +1450,7 @@ static void ipoib_cm_stale_task(struct work_struct *work) } if (!list_empty(&priv->cm.passive_ids)) - queue_delayed_work(ipoib_workqueue, + queue_delayed_work(priv->wq, &priv->cm.stale_task, IPOIB_CM_RX_DELAY); spin_unlock_irq(&priv->lock); } diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c index 72626c348174..fe65abb5150c 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c @@ -655,7 +655,7 @@ void ipoib_reap_ah(struct work_struct *work) __ipoib_reap_ah(dev); if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) - queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, + queue_delayed_work(priv->wq, &priv->ah_reap_task, round_jiffies_relative(HZ)); } @@ -664,7 +664,7 @@ static void ipoib_ib_tx_timer_func(unsigned long ctx) drain_tx_cq((struct net_device *)ctx); } -int ipoib_ib_dev_open(struct net_device *dev, int flush) +int ipoib_ib_dev_open(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); int ret; @@ -696,7 +696,7 @@ int ipoib_ib_dev_open(struct net_device *dev, int flush) } clear_bit(IPOIB_STOP_REAPER, &priv->flags); - queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, + queue_delayed_work(priv->wq, &priv->ah_reap_task, round_jiffies_relative(HZ)); if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) @@ -706,7 +706,7 @@ int ipoib_ib_dev_open(struct net_device *dev, int flush) dev_stop: if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) napi_enable(&priv->napi); - ipoib_ib_dev_stop(dev, flush); + ipoib_ib_dev_stop(dev); return -1; } @@ -738,7 +738,7 @@ int ipoib_ib_dev_up(struct net_device *dev) return ipoib_mcast_start_thread(dev); } -int ipoib_ib_dev_down(struct net_device *dev, int flush) +int ipoib_ib_dev_down(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); @@ -747,7 +747,7 @@ int ipoib_ib_dev_down(struct net_device *dev, int flush) clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); netif_carrier_off(dev); - ipoib_mcast_stop_thread(dev, flush); + ipoib_mcast_stop_thread(dev); ipoib_mcast_dev_flush(dev); ipoib_flush_paths(dev); @@ -807,7 +807,7 @@ void ipoib_drain_cq(struct net_device *dev) local_bh_enable(); } -int ipoib_ib_dev_stop(struct net_device *dev, int flush) +int ipoib_ib_dev_stop(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct ib_qp_attr qp_attr; @@ -880,8 +880,7 @@ timeout: /* Wait for all AHs to be reaped */ set_bit(IPOIB_STOP_REAPER, &priv->flags); cancel_delayed_work(&priv->ah_reap_task); - if (flush) - flush_workqueue(ipoib_workqueue); + flush_workqueue(priv->wq); begin = jiffies; @@ -918,7 +917,7 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) (unsigned long) dev); if (dev->flags & IFF_UP) { - if (ipoib_ib_dev_open(dev, 1)) { + if (ipoib_ib_dev_open(dev)) { ipoib_transport_dev_cleanup(dev); return -ENODEV; } @@ -1040,12 +1039,12 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, } if (level >= IPOIB_FLUSH_NORMAL) - ipoib_ib_dev_down(dev, 0); + ipoib_ib_dev_down(dev); if (level == IPOIB_FLUSH_HEAVY) { if (test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) - ipoib_ib_dev_stop(dev, 0); - if (ipoib_ib_dev_open(dev, 0) != 0) + ipoib_ib_dev_stop(dev); + if (ipoib_ib_dev_open(dev) != 0) return; if (netif_queue_stopped(dev)) netif_start_queue(dev); @@ -1097,7 +1096,7 @@ void ipoib_ib_dev_cleanup(struct net_device *dev) */ ipoib_flush_paths(dev); - ipoib_mcast_stop_thread(dev, 1); + ipoib_mcast_stop_thread(dev); ipoib_mcast_dev_flush(dev); ipoib_transport_dev_cleanup(dev); diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index 58b5aa3b6f2d..6bad17d4d588 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -108,7 +108,7 @@ int ipoib_open(struct net_device *dev) set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); - if (ipoib_ib_dev_open(dev, 1)) { + if (ipoib_ib_dev_open(dev)) { if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) return 0; goto err_disable; @@ -139,7 +139,7 @@ int ipoib_open(struct net_device *dev) return 0; err_stop: - ipoib_ib_dev_stop(dev, 1); + ipoib_ib_dev_stop(dev); err_disable: clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); @@ -157,8 +157,8 @@ static int ipoib_stop(struct net_device *dev) netif_stop_queue(dev); - ipoib_ib_dev_down(dev, 1); - ipoib_ib_dev_stop(dev, 0); + ipoib_ib_dev_down(dev); + ipoib_ib_dev_stop(dev); if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { struct ipoib_dev_priv *cpriv; @@ -839,7 +839,7 @@ static void ipoib_set_mcast_list(struct net_device *dev) return; } - queue_work(ipoib_workqueue, &priv->restart_task); + queue_work(priv->wq, &priv->restart_task); } static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr) @@ -954,7 +954,7 @@ static void ipoib_reap_neigh(struct work_struct *work) __ipoib_reap_neigh(priv); if (!test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags)) - queue_delayed_work(ipoib_workqueue, &priv->neigh_reap_task, + queue_delayed_work(priv->wq, &priv->neigh_reap_task, arp_tbl.gc_interval); } @@ -1133,7 +1133,7 @@ static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv) /* start garbage collection */ clear_bit(IPOIB_STOP_NEIGH_GC, &priv->flags); - queue_delayed_work(ipoib_workqueue, &priv->neigh_reap_task, + queue_delayed_work(priv->wq, &priv->neigh_reap_task, arp_tbl.gc_interval); return 0; @@ -1262,15 +1262,13 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port) { struct ipoib_dev_priv *priv = netdev_priv(dev); - if (ipoib_neigh_hash_init(priv) < 0) - goto out; /* Allocate RX/TX "rings" to hold queued skbs */ priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring, GFP_KERNEL); if (!priv->rx_ring) { printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n", ca->name, ipoib_recvq_size); - goto out_neigh_hash_cleanup; + goto out; } priv->tx_ring = vzalloc(ipoib_sendq_size * sizeof *priv->tx_ring); @@ -1285,16 +1283,24 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port) if (ipoib_ib_dev_init(dev, ca, port)) goto out_tx_ring_cleanup; + /* + * Must be after ipoib_ib_dev_init so we can allocate a per + * device wq there and use it here + */ + if (ipoib_neigh_hash_init(priv) < 0) + goto out_dev_uninit; + return 0; +out_dev_uninit: + ipoib_ib_dev_cleanup(dev); + out_tx_ring_cleanup: vfree(priv->tx_ring); out_rx_ring_cleanup: kfree(priv->rx_ring); -out_neigh_hash_cleanup: - ipoib_neigh_hash_uninit(dev); out: return -ENOMEM; } @@ -1317,6 +1323,12 @@ void ipoib_dev_cleanup(struct net_device *dev) } unregister_netdevice_many(&head); + /* + * Must be before ipoib_ib_dev_cleanup or we delete an in use + * work queue + */ + ipoib_neigh_hash_uninit(dev); + ipoib_ib_dev_cleanup(dev); kfree(priv->rx_ring); @@ -1324,8 +1336,6 @@ void ipoib_dev_cleanup(struct net_device *dev) priv->rx_ring = NULL; priv->tx_ring = NULL; - - ipoib_neigh_hash_uninit(dev); } static const struct header_ops ipoib_header_ops = { @@ -1636,7 +1646,7 @@ register_failed: /* Stop GC if started before flush */ set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags); cancel_delayed_work(&priv->neigh_reap_task); - flush_workqueue(ipoib_workqueue); + flush_workqueue(priv->wq); event_failed: ipoib_dev_cleanup(priv->dev); @@ -1707,7 +1717,7 @@ static void ipoib_remove_one(struct ib_device *device) /* Stop GC */ set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags); cancel_delayed_work(&priv->neigh_reap_task); - flush_workqueue(ipoib_workqueue); + flush_workqueue(priv->wq); unregister_netdev(priv->dev); free_netdev(priv->dev); @@ -1748,8 +1758,13 @@ static int __init ipoib_init_module(void) * unregister_netdev() and linkwatch_event take the rtnl lock, * so flush_scheduled_work() can deadlock during device * removal. + * + * In addition, bringing one device up and another down at the + * same time can deadlock a single workqueue, so we have this + * global fallback workqueue, but we also attempt to open a + * per device workqueue each time we bring an interface up */ - ipoib_workqueue = create_singlethread_workqueue("ipoib"); + ipoib_workqueue = create_singlethread_workqueue("ipoib_flush"); if (!ipoib_workqueue) { ret = -ENOMEM; goto err_fs; diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c index ffb83b5f7e80..bc50dd0d0e4d 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c @@ -190,12 +190,6 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, spin_unlock_irq(&priv->lock); priv->tx_wr.wr.ud.remote_qkey = priv->qkey; set_qkey = 1; - - if (!ipoib_cm_admin_enabled(dev)) { - rtnl_lock(); - dev_set_mtu(dev, min(priv->mcast_mtu, priv->admin_mtu)); - rtnl_unlock(); - } } if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { @@ -277,16 +271,27 @@ ipoib_mcast_sendonly_join_complete(int status, struct ipoib_mcast *mcast = multicast->context; struct net_device *dev = mcast->dev; + /* + * We have to take the mutex to force mcast_sendonly_join to + * return from ib_sa_multicast_join and set mcast->mc to a + * valid value. Otherwise we were racing with ourselves in + * that we might fail here, but get a valid return from + * ib_sa_multicast_join after we had cleared mcast->mc here, + * resulting in mis-matched joins and leaves and a deadlock + */ + mutex_lock(&mcast_mutex); + /* We trap for port events ourselves. */ if (status == -ENETRESET) - return 0; + goto out; if (!status) status = ipoib_mcast_join_finish(mcast, &multicast->rec); if (status) { if (mcast->logcount++ < 20) - ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for %pI6, status %d\n", + ipoib_dbg_mcast(netdev_priv(dev), "sendonly multicast " + "join failed for %pI6, status %d\n", mcast->mcmember.mgid.raw, status); /* Flush out any queued packets */ @@ -296,11 +301,15 @@ ipoib_mcast_sendonly_join_complete(int status, dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); } netif_tx_unlock_bh(dev); - - /* Clear the busy flag so we try again */ - status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, - &mcast->flags); } +out: + clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); + if (status) + mcast->mc = NULL; + complete(&mcast->done); + if (status == -ENETRESET) + status = 0; + mutex_unlock(&mcast_mutex); return status; } @@ -318,12 +327,14 @@ static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast) int ret = 0; if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) { - ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n"); + ipoib_dbg_mcast(priv, "device shutting down, no sendonly " + "multicast joins\n"); return -ENODEV; } - if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) { - ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n"); + if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) { + ipoib_dbg_mcast(priv, "multicast entry busy, skipping " + "sendonly join\n"); return -EBUSY; } @@ -331,6 +342,9 @@ static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast) rec.port_gid = priv->local_gid; rec.pkey = cpu_to_be16(priv->pkey); + mutex_lock(&mcast_mutex); + init_completion(&mcast->done); + set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port, &rec, IB_SA_MCMEMBER_REC_MGID | @@ -343,12 +357,14 @@ static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast) if (IS_ERR(mcast->mc)) { ret = PTR_ERR(mcast->mc); clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); - ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n", - ret); + complete(&mcast->done); + ipoib_warn(priv, "ib_sa_join_multicast for sendonly join " + "failed (ret = %d)\n", ret); } else { - ipoib_dbg_mcast(priv, "no multicast record for %pI6, starting join\n", - mcast->mcmember.mgid.raw); + ipoib_dbg_mcast(priv, "no multicast record for %pI6, starting " + "sendonly join\n", mcast->mcmember.mgid.raw); } + mutex_unlock(&mcast_mutex); return ret; } @@ -359,18 +375,29 @@ void ipoib_mcast_carrier_on_task(struct work_struct *work) carrier_on_task); struct ib_port_attr attr; - /* - * Take rtnl_lock to avoid racing with ipoib_stop() and - * turning the carrier back on while a device is being - * removed. - */ if (ib_query_port(priv->ca, priv->port, &attr) || attr.state != IB_PORT_ACTIVE) { ipoib_dbg(priv, "Keeping carrier off until IB port is active\n"); return; } - rtnl_lock(); + /* + * Take rtnl_lock to avoid racing with ipoib_stop() and + * turning the carrier back on while a device is being + * removed. However, ipoib_stop() will attempt to flush + * the workqueue while holding the rtnl lock, so loop + * on trylock until either we get the lock or we see + * FLAG_ADMIN_UP go away as that signals that we are bailing + * and can safely ignore the carrier on work. + */ + while (!rtnl_trylock()) { + if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) + return; + else + msleep(20); + } + if (!ipoib_cm_admin_enabled(priv->dev)) + dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu)); netif_carrier_on(priv->dev); rtnl_unlock(); } @@ -385,60 +412,63 @@ static int ipoib_mcast_join_complete(int status, ipoib_dbg_mcast(priv, "join completion for %pI6 (status %d)\n", mcast->mcmember.mgid.raw, status); + /* + * We have to take the mutex to force mcast_join to + * return from ib_sa_multicast_join and set mcast->mc to a + * valid value. Otherwise we were racing with ourselves in + * that we might fail here, but get a valid return from + * ib_sa_multicast_join after we had cleared mcast->mc here, + * resulting in mis-matched joins and leaves and a deadlock + */ + mutex_lock(&mcast_mutex); + /* We trap for port events ourselves. */ - if (status == -ENETRESET) { - status = 0; + if (status == -ENETRESET) goto out; - } if (!status) status = ipoib_mcast_join_finish(mcast, &multicast->rec); if (!status) { mcast->backoff = 1; - mutex_lock(&mcast_mutex); if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) - queue_delayed_work(ipoib_workqueue, - &priv->mcast_task, 0); - mutex_unlock(&mcast_mutex); + queue_delayed_work(priv->wq, &priv->mcast_task, 0); /* - * Defer carrier on work to ipoib_workqueue to avoid a + * Defer carrier on work to priv->wq to avoid a * deadlock on rtnl_lock here. */ if (mcast == priv->broadcast) - queue_work(ipoib_workqueue, &priv->carrier_on_task); - - status = 0; - goto out; - } - - if (mcast->logcount++ < 20) { - if (status == -ETIMEDOUT || status == -EAGAIN) { - ipoib_dbg_mcast(priv, "multicast join failed for %pI6, status %d\n", - mcast->mcmember.mgid.raw, status); - } else { - ipoib_warn(priv, "multicast join failed for %pI6, status %d\n", - mcast->mcmember.mgid.raw, status); + queue_work(priv->wq, &priv->carrier_on_task); + } else { + if (mcast->logcount++ < 20) { + if (status == -ETIMEDOUT || status == -EAGAIN) { + ipoib_dbg_mcast(priv, "multicast join failed for %pI6, status %d\n", + mcast->mcmember.mgid.raw, status); + } else { + ipoib_warn(priv, "multicast join failed for %pI6, status %d\n", + mcast->mcmember.mgid.raw, status); + } } - } - - mcast->backoff *= 2; - if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) - mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; - /* Clear the busy flag so we try again */ - status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); - - mutex_lock(&mcast_mutex); + mcast->backoff *= 2; + if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) + mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; + } +out: spin_lock_irq(&priv->lock); - if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) - queue_delayed_work(ipoib_workqueue, &priv->mcast_task, + clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); + if (status) + mcast->mc = NULL; + complete(&mcast->done); + if (status == -ENETRESET) + status = 0; + if (status && test_bit(IPOIB_MCAST_RUN, &priv->flags)) + queue_delayed_work(priv->wq, &priv->mcast_task, mcast->backoff * HZ); spin_unlock_irq(&priv->lock); mutex_unlock(&mcast_mutex); -out: - complete(&mcast->done); + return status; } @@ -487,10 +517,9 @@ static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast, rec.hop_limit = priv->broadcast->mcmember.hop_limit; } - set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); + mutex_lock(&mcast_mutex); init_completion(&mcast->done); - set_bit(IPOIB_MCAST_JOIN_STARTED, &mcast->flags); - + set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port, &rec, comp_mask, GFP_KERNEL, ipoib_mcast_join_complete, mcast); @@ -504,13 +533,11 @@ static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast, if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; - mutex_lock(&mcast_mutex); if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) - queue_delayed_work(ipoib_workqueue, - &priv->mcast_task, + queue_delayed_work(priv->wq, &priv->mcast_task, mcast->backoff * HZ); - mutex_unlock(&mcast_mutex); } + mutex_unlock(&mcast_mutex); } void ipoib_mcast_join_task(struct work_struct *work) @@ -547,8 +574,8 @@ void ipoib_mcast_join_task(struct work_struct *work) ipoib_warn(priv, "failed to allocate broadcast group\n"); mutex_lock(&mcast_mutex); if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) - queue_delayed_work(ipoib_workqueue, - &priv->mcast_task, HZ); + queue_delayed_work(priv->wq, &priv->mcast_task, + HZ); mutex_unlock(&mcast_mutex); return; } @@ -563,7 +590,8 @@ void ipoib_mcast_join_task(struct work_struct *work) } if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) { - if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) + if (IS_ERR_OR_NULL(priv->broadcast->mc) && + !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) ipoib_mcast_join(dev, priv->broadcast, 0); return; } @@ -571,23 +599,33 @@ void ipoib_mcast_join_task(struct work_struct *work) while (1) { struct ipoib_mcast *mcast = NULL; + /* + * Need the mutex so our flags are consistent, need the + * priv->lock so we don't race with list removals in either + * mcast_dev_flush or mcast_restart_task + */ + mutex_lock(&mcast_mutex); spin_lock_irq(&priv->lock); list_for_each_entry(mcast, &priv->multicast_list, list) { - if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) - && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) - && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { + if (IS_ERR_OR_NULL(mcast->mc) && + !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) && + !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { /* Found the next unjoined group */ break; } } spin_unlock_irq(&priv->lock); + mutex_unlock(&mcast_mutex); if (&mcast->list == &priv->multicast_list) { /* All done */ break; } - ipoib_mcast_join(dev, mcast, 1); + if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) + ipoib_mcast_sendonly_join(mcast); + else + ipoib_mcast_join(dev, mcast, 1); return; } @@ -604,13 +642,13 @@ int ipoib_mcast_start_thread(struct net_device *dev) mutex_lock(&mcast_mutex); if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags)) - queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0); + queue_delayed_work(priv->wq, &priv->mcast_task, 0); mutex_unlock(&mcast_mutex); return 0; } -int ipoib_mcast_stop_thread(struct net_device *dev, int flush) +int ipoib_mcast_stop_thread(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); @@ -621,8 +659,7 @@ int ipoib_mcast_stop_thread(struct net_device *dev, int flush) cancel_delayed_work(&priv->mcast_task); mutex_unlock(&mcast_mutex); - if (flush) - flush_workqueue(ipoib_workqueue); + flush_workqueue(priv->wq); return 0; } @@ -633,6 +670,9 @@ static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast) int ret = 0; if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) + ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n"); + + if (!IS_ERR_OR_NULL(mcast->mc)) ib_sa_free_multicast(mcast->mc); if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { @@ -685,6 +725,8 @@ void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb) memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid)); __ipoib_mcast_add(dev, mcast); list_add_tail(&mcast->list, &priv->multicast_list); + if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags)) + queue_delayed_work(priv->wq, &priv->mcast_task, 0); } if (!mcast->ah) { @@ -698,8 +740,6 @@ void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb) if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) ipoib_dbg_mcast(priv, "no address vector, " "but multicast join already started\n"); - else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) - ipoib_mcast_sendonly_join(mcast); /* * If lookup completes between here and out:, don't @@ -759,9 +799,12 @@ void ipoib_mcast_dev_flush(struct net_device *dev) spin_unlock_irqrestore(&priv->lock, flags); - /* seperate between the wait to the leave*/ + /* + * make sure the in-flight joins have finished before we attempt + * to leave + */ list_for_each_entry_safe(mcast, tmcast, &remove_list, list) - if (test_bit(IPOIB_MCAST_JOIN_STARTED, &mcast->flags)) + if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) wait_for_completion(&mcast->done); list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { @@ -794,8 +837,6 @@ void ipoib_mcast_restart_task(struct work_struct *work) ipoib_dbg_mcast(priv, "restarting multicast task\n"); - ipoib_mcast_stop_thread(dev, 0); - local_irq_save(flags); netif_addr_lock(dev); spin_lock(&priv->lock); @@ -880,14 +921,38 @@ void ipoib_mcast_restart_task(struct work_struct *work) netif_addr_unlock(dev); local_irq_restore(flags); - /* We have to cancel outside of the spinlock */ + /* + * make sure the in-flight joins have finished before we attempt + * to leave + */ + list_for_each_entry_safe(mcast, tmcast, &remove_list, list) + if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) + wait_for_completion(&mcast->done); + + /* + * We have to cancel outside of the spinlock, but we have to + * take the rtnl lock or else we race with the removal of + * entries from the remove list in mcast_dev_flush as part + * of ipoib_stop(). We detect the drop of the ADMIN_UP flag + * to signal that we have hit this particular race, and we + * return since we know we don't need to do anything else + * anyway. + */ + while (!rtnl_trylock()) { + if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) + return; + else + msleep(20); + } list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { ipoib_mcast_leave(mcast->dev, mcast); ipoib_mcast_free(mcast); } - - if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) - ipoib_mcast_start_thread(dev); + /* + * Restart our join task if needed + */ + ipoib_mcast_start_thread(dev); + rtnl_unlock(); } #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c index c56d5d44c53b..b72a753eb41d 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c @@ -145,10 +145,20 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca) int ret, size; int i; + /* + * the various IPoIB tasks assume they will never race against + * themselves, so always use a single thread workqueue + */ + priv->wq = create_singlethread_workqueue("ipoib_wq"); + if (!priv->wq) { + printk(KERN_WARNING "ipoib: failed to allocate device WQ\n"); + return -ENODEV; + } + priv->pd = ib_alloc_pd(priv->ca); if (IS_ERR(priv->pd)) { printk(KERN_WARNING "%s: failed to allocate PD\n", ca->name); - return -ENODEV; + goto out_free_wq; } priv->mr = ib_get_dma_mr(priv->pd, IB_ACCESS_LOCAL_WRITE); @@ -242,6 +252,10 @@ out_free_mr: out_free_pd: ib_dealloc_pd(priv->pd); + +out_free_wq: + destroy_workqueue(priv->wq); + priv->wq = NULL; return -ENODEV; } @@ -270,6 +284,12 @@ void ipoib_transport_dev_cleanup(struct net_device *dev) if (ib_dealloc_pd(priv->pd)) ipoib_warn(priv, "ib_dealloc_pd failed\n"); + + if (priv->wq) { + flush_workqueue(priv->wq); + destroy_workqueue(priv->wq); + priv->wq = NULL; + } } void ipoib_event(struct ib_event_handler *handler, diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c index 20ca6a619476..6a594aac2290 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.c +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c @@ -97,7 +97,7 @@ module_param_named(pi_enable, iser_pi_enable, bool, 0644); MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)"); module_param_named(pi_guard, iser_pi_guard, int, 0644); -MODULE_PARM_DESC(pi_guard, "T10-PI guard_type, 0:CRC|1:IP_CSUM (default:IP_CSUM)"); +MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]"); static struct workqueue_struct *release_wq; struct iser_global ig; @@ -164,18 +164,42 @@ iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode) return 0; } -int iser_initialize_task_headers(struct iscsi_task *task, - struct iser_tx_desc *tx_desc) +/** + * iser_initialize_task_headers() - Initialize task headers + * @task: iscsi task + * @tx_desc: iser tx descriptor + * + * Notes: + * This routine may race with iser teardown flow for scsi + * error handling TMFs. So for TMF we should acquire the + * state mutex to avoid dereferencing the IB device which + * may have already been terminated. + */ +int +iser_initialize_task_headers(struct iscsi_task *task, + struct iser_tx_desc *tx_desc) { - struct iser_conn *iser_conn = task->conn->dd_data; + struct iser_conn *iser_conn = task->conn->dd_data; struct iser_device *device = iser_conn->ib_conn.device; struct iscsi_iser_task *iser_task = task->dd_data; u64 dma_addr; + const bool mgmt_task = !task->sc && !in_interrupt(); + int ret = 0; + + if (unlikely(mgmt_task)) + mutex_lock(&iser_conn->state_mutex); + + if (unlikely(iser_conn->state != ISER_CONN_UP)) { + ret = -ENODEV; + goto out; + } dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc, ISER_HEADERS_LEN, DMA_TO_DEVICE); - if (ib_dma_mapping_error(device->ib_device, dma_addr)) - return -ENOMEM; + if (ib_dma_mapping_error(device->ib_device, dma_addr)) { + ret = -ENOMEM; + goto out; + } tx_desc->dma_addr = dma_addr; tx_desc->tx_sg[0].addr = tx_desc->dma_addr; @@ -183,7 +207,11 @@ int iser_initialize_task_headers(struct iscsi_task *task, tx_desc->tx_sg[0].lkey = device->mr->lkey; iser_task->iser_conn = iser_conn; - return 0; +out: + if (unlikely(mgmt_task)) + mutex_unlock(&iser_conn->state_mutex); + + return ret; } /** @@ -199,9 +227,14 @@ static int iscsi_iser_task_init(struct iscsi_task *task) { struct iscsi_iser_task *iser_task = task->dd_data; + int ret; - if (iser_initialize_task_headers(task, &iser_task->desc)) - return -ENOMEM; + ret = iser_initialize_task_headers(task, &iser_task->desc); + if (ret) { + iser_err("Failed to init task %p, err = %d\n", + iser_task, ret); + return ret; + } /* mgmt task */ if (!task->sc) @@ -508,8 +541,8 @@ iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag) */ if (iser_conn) { mutex_lock(&iser_conn->state_mutex); - iscsi_conn_stop(cls_conn, flag); iser_conn_terminate(iser_conn); + iscsi_conn_stop(cls_conn, flag); /* unbind */ iser_conn->iscsi_conn = NULL; @@ -541,12 +574,13 @@ iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session) static inline unsigned int iser_dif_prot_caps(int prot_caps) { - return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? SHOST_DIF_TYPE1_PROTECTION | - SHOST_DIX_TYPE1_PROTECTION : 0) | - ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? SHOST_DIF_TYPE2_PROTECTION | - SHOST_DIX_TYPE2_PROTECTION : 0) | - ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? SHOST_DIF_TYPE3_PROTECTION | - SHOST_DIX_TYPE3_PROTECTION : 0); + return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ? + SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION | + SHOST_DIX_TYPE1_PROTECTION : 0) | + ((prot_caps & IB_PROT_T10DIF_TYPE_2) ? + SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) | + ((prot_caps & IB_PROT_T10DIF_TYPE_3) ? + SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0); } /** @@ -569,6 +603,7 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep, struct Scsi_Host *shost; struct iser_conn *iser_conn = NULL; struct ib_conn *ib_conn; + u16 max_cmds; shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0); if (!shost) @@ -586,26 +621,41 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep, */ if (ep) { iser_conn = ep->dd_data; + max_cmds = iser_conn->max_cmds; + + mutex_lock(&iser_conn->state_mutex); + if (iser_conn->state != ISER_CONN_UP) { + iser_err("iser conn %p already started teardown\n", + iser_conn); + mutex_unlock(&iser_conn->state_mutex); + goto free_host; + } + ib_conn = &iser_conn->ib_conn; if (ib_conn->pi_support) { u32 sig_caps = ib_conn->device->dev_attr.sig_prot_cap; scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps)); - if (iser_pi_guard) - scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP); - else - scsi_host_set_guard(shost, SHOST_DIX_GUARD_CRC); + scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP | + SHOST_DIX_GUARD_CRC); } - } - if (iscsi_host_add(shost, ep ? - ib_conn->device->ib_device->dma_device : NULL)) - goto free_host; + if (iscsi_host_add(shost, + ib_conn->device->ib_device->dma_device)) { + mutex_unlock(&iser_conn->state_mutex); + goto free_host; + } + mutex_unlock(&iser_conn->state_mutex); + } else { + max_cmds = ISER_DEF_XMIT_CMDS_MAX; + if (iscsi_host_add(shost, NULL)) + goto free_host; + } - if (cmds_max > ISER_DEF_XMIT_CMDS_MAX) { + if (cmds_max > max_cmds) { iser_info("cmds_max changed from %u to %u\n", - cmds_max, ISER_DEF_XMIT_CMDS_MAX); - cmds_max = ISER_DEF_XMIT_CMDS_MAX; + cmds_max, max_cmds); + cmds_max = max_cmds; } cls_session = iscsi_session_setup(&iscsi_iser_transport, shost, diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h index cd4174ca9a76..5ce26817e7e1 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.h +++ b/drivers/infiniband/ulp/iser/iscsi_iser.h @@ -69,34 +69,31 @@ #define DRV_NAME "iser" #define PFX DRV_NAME ": " -#define DRV_VER "1.4.8" +#define DRV_VER "1.5" #define iser_dbg(fmt, arg...) \ do { \ - if (iser_debug_level > 2) \ + if (unlikely(iser_debug_level > 2)) \ printk(KERN_DEBUG PFX "%s: " fmt,\ __func__ , ## arg); \ } while (0) #define iser_warn(fmt, arg...) \ do { \ - if (iser_debug_level > 0) \ + if (unlikely(iser_debug_level > 0)) \ pr_warn(PFX "%s: " fmt, \ __func__ , ## arg); \ } while (0) #define iser_info(fmt, arg...) \ do { \ - if (iser_debug_level > 1) \ + if (unlikely(iser_debug_level > 1)) \ pr_info(PFX "%s: " fmt, \ __func__ , ## arg); \ } while (0) -#define iser_err(fmt, arg...) \ - do { \ - printk(KERN_ERR PFX "%s: " fmt, \ - __func__ , ## arg); \ - } while (0) +#define iser_err(fmt, arg...) \ + pr_err(PFX "%s: " fmt, __func__ , ## arg) #define SHIFT_4K 12 #define SIZE_4K (1ULL << SHIFT_4K) @@ -144,6 +141,11 @@ ISER_MAX_TX_MISC_PDUS + \ ISER_MAX_RX_MISC_PDUS) +#define ISER_GET_MAX_XMIT_CMDS(send_wr) ((send_wr \ + - ISER_MAX_TX_MISC_PDUS \ + - ISER_MAX_RX_MISC_PDUS) / \ + (1 + ISER_INFLIGHT_DATAOUTS)) + #define ISER_WC_BATCH_COUNT 16 #define ISER_SIGNAL_CMD_COUNT 32 @@ -247,7 +249,6 @@ struct iscsi_endpoint; * @va: MR start address (buffer va) * @len: MR length * @mem_h: pointer to registration context (FMR/Fastreg) - * @is_mr: indicates weather we registered the buffer */ struct iser_mem_reg { u32 lkey; @@ -255,7 +256,6 @@ struct iser_mem_reg { u64 va; u64 len; void *mem_h; - int is_mr; }; /** @@ -323,8 +323,6 @@ struct iser_rx_desc { char pad[ISER_RX_PAD_SIZE]; } __attribute__((packed)); -#define ISER_MAX_CQ 4 - struct iser_conn; struct ib_conn; struct iscsi_iser_task; @@ -375,7 +373,7 @@ struct iser_device { struct list_head ig_list; int refcount; int comps_used; - struct iser_comp comps[ISER_MAX_CQ]; + struct iser_comp *comps; int (*iser_alloc_rdma_reg_res)(struct ib_conn *ib_conn, unsigned cmds_max); void (*iser_free_rdma_reg_res)(struct ib_conn *ib_conn); @@ -432,6 +430,7 @@ struct fast_reg_descriptor { * @cma_id: rdma_cm connection maneger handle * @qp: Connection Queue-pair * @post_recv_buf_count: post receive counter + * @sig_count: send work request signal count * @rx_wr: receive work request for batch posts * @device: reference to iser device * @comp: iser completion context @@ -452,6 +451,7 @@ struct ib_conn { struct rdma_cm_id *cma_id; struct ib_qp *qp; int post_recv_buf_count; + u8 sig_count; struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX]; struct iser_device *device; struct iser_comp *comp; @@ -482,6 +482,7 @@ struct ib_conn { * to max number of post recvs * @qp_max_recv_dtos_mask: (qp_max_recv_dtos - 1) * @min_posted_rx: (qp_max_recv_dtos >> 2) + * @max_cmds: maximum cmds allowed for this connection * @name: connection peer portal * @release_work: deffered work for release job * @state_mutex: protects iser onnection state @@ -507,6 +508,7 @@ struct iser_conn { unsigned qp_max_recv_dtos; unsigned qp_max_recv_dtos_mask; unsigned min_posted_rx; + u16 max_cmds; char name[ISER_OBJECT_NAME_SIZE]; struct work_struct release_work; struct mutex state_mutex; diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c index 5a489ea63732..3821633f1065 100644 --- a/drivers/infiniband/ulp/iser/iser_initiator.c +++ b/drivers/infiniband/ulp/iser/iser_initiator.c @@ -369,7 +369,7 @@ static int iser_post_rx_bufs(struct iscsi_conn *conn, struct iscsi_hdr *req) return 0; } -static inline bool iser_signal_comp(int sig_count) +static inline bool iser_signal_comp(u8 sig_count) { return ((sig_count % ISER_SIGNAL_CMD_COUNT) == 0); } @@ -388,7 +388,7 @@ int iser_send_command(struct iscsi_conn *conn, struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)task->hdr; struct scsi_cmnd *sc = task->sc; struct iser_tx_desc *tx_desc = &iser_task->desc; - static unsigned sig_count; + u8 sig_count = ++iser_conn->ib_conn.sig_count; edtl = ntohl(hdr->data_length); @@ -435,7 +435,7 @@ int iser_send_command(struct iscsi_conn *conn, iser_task->status = ISER_TASK_STATUS_STARTED; err = iser_post_send(&iser_conn->ib_conn, tx_desc, - iser_signal_comp(++sig_count)); + iser_signal_comp(sig_count)); if (!err) return 0; diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c index 6c5ce357fba6..abce9339333f 100644 --- a/drivers/infiniband/ulp/iser/iser_memory.c +++ b/drivers/infiniband/ulp/iser/iser_memory.c @@ -73,7 +73,6 @@ static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task, if (cmd_dir == ISER_DIR_OUT) { /* copy the unaligned sg the buffer which is used for RDMA */ - int i; char *p, *from; sgl = (struct scatterlist *)data->buf; @@ -409,7 +408,6 @@ int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *iser_task, regd_buf->reg.rkey = device->mr->rkey; regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]); regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]); - regd_buf->reg.is_mr = 0; iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X " "va: 0x%08lX sz: %ld]\n", @@ -440,13 +438,13 @@ int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *iser_task, return 0; } -static inline void +static void iser_set_dif_domain(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs, struct ib_sig_domain *domain) { domain->sig_type = IB_SIG_TYPE_T10_DIF; - domain->sig.dif.pi_interval = sc->device->sector_size; - domain->sig.dif.ref_tag = scsi_get_lba(sc) & 0xffffffff; + domain->sig.dif.pi_interval = scsi_prot_interval(sc); + domain->sig.dif.ref_tag = scsi_prot_ref_tag(sc); /* * At the moment we hard code those, but in the future * we will take them from sc. @@ -454,8 +452,7 @@ iser_set_dif_domain(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs, domain->sig.dif.apptag_check_mask = 0xffff; domain->sig.dif.app_escape = true; domain->sig.dif.ref_escape = true; - if (scsi_get_prot_type(sc) == SCSI_PROT_DIF_TYPE1 || - scsi_get_prot_type(sc) == SCSI_PROT_DIF_TYPE2) + if (sc->prot_flags & SCSI_PROT_REF_INCREMENT) domain->sig.dif.ref_remap = true; }; @@ -473,26 +470,16 @@ iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs) case SCSI_PROT_WRITE_STRIP: sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE; iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem); - /* - * At the moment we use this modparam to tell what is - * the memory bg_type, in the future we will take it - * from sc. - */ - sig_attrs->mem.sig.dif.bg_type = iser_pi_guard ? IB_T10DIF_CSUM : - IB_T10DIF_CRC; + sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ? + IB_T10DIF_CSUM : IB_T10DIF_CRC; break; case SCSI_PROT_READ_PASS: case SCSI_PROT_WRITE_PASS: iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire); sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC; iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem); - /* - * At the moment we use this modparam to tell what is - * the memory bg_type, in the future we will take it - * from sc. - */ - sig_attrs->mem.sig.dif.bg_type = iser_pi_guard ? IB_T10DIF_CSUM : - IB_T10DIF_CRC; + sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ? + IB_T10DIF_CSUM : IB_T10DIF_CRC; break; default: iser_err("Unsupported PI operation %d\n", @@ -503,26 +490,28 @@ iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs) return 0; } -static int +static inline void iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask) { - switch (scsi_get_prot_type(sc)) { - case SCSI_PROT_DIF_TYPE0: - break; - case SCSI_PROT_DIF_TYPE1: - case SCSI_PROT_DIF_TYPE2: - *mask = ISER_CHECK_GUARD | ISER_CHECK_REFTAG; - break; - case SCSI_PROT_DIF_TYPE3: - *mask = ISER_CHECK_GUARD; - break; - default: - iser_err("Unsupported protection type %d\n", - scsi_get_prot_type(sc)); - return -EINVAL; - } + *mask = 0; + if (sc->prot_flags & SCSI_PROT_REF_CHECK) + *mask |= ISER_CHECK_REFTAG; + if (sc->prot_flags & SCSI_PROT_GUARD_CHECK) + *mask |= ISER_CHECK_GUARD; +} - return 0; +static void +iser_inv_rkey(struct ib_send_wr *inv_wr, struct ib_mr *mr) +{ + u32 rkey; + + memset(inv_wr, 0, sizeof(*inv_wr)); + inv_wr->opcode = IB_WR_LOCAL_INV; + inv_wr->wr_id = ISER_FASTREG_LI_WRID; + inv_wr->ex.invalidate_rkey = mr->rkey; + + rkey = ib_inc_rkey(mr->rkey); + ib_update_fast_reg_key(mr, rkey); } static int @@ -536,26 +525,17 @@ iser_reg_sig_mr(struct iscsi_iser_task *iser_task, struct ib_send_wr *bad_wr, *wr = NULL; struct ib_sig_attrs sig_attrs; int ret; - u32 key; memset(&sig_attrs, 0, sizeof(sig_attrs)); ret = iser_set_sig_attrs(iser_task->sc, &sig_attrs); if (ret) goto err; - ret = iser_set_prot_checks(iser_task->sc, &sig_attrs.check_mask); - if (ret) - goto err; + iser_set_prot_checks(iser_task->sc, &sig_attrs.check_mask); if (!(desc->reg_indicators & ISER_SIG_KEY_VALID)) { - memset(&inv_wr, 0, sizeof(inv_wr)); - inv_wr.opcode = IB_WR_LOCAL_INV; - inv_wr.wr_id = ISER_FASTREG_LI_WRID; - inv_wr.ex.invalidate_rkey = pi_ctx->sig_mr->rkey; + iser_inv_rkey(&inv_wr, pi_ctx->sig_mr); wr = &inv_wr; - /* Bump the key */ - key = (u8)(pi_ctx->sig_mr->rkey & 0x000000FF); - ib_update_fast_reg_key(pi_ctx->sig_mr, ++key); } memset(&sig_wr, 0, sizeof(sig_wr)); @@ -585,12 +565,7 @@ iser_reg_sig_mr(struct iscsi_iser_task *iser_task, sig_sge->lkey = pi_ctx->sig_mr->lkey; sig_sge->addr = 0; - sig_sge->length = data_sge->length + prot_sge->length; - if (scsi_get_prot_op(iser_task->sc) == SCSI_PROT_WRITE_INSERT || - scsi_get_prot_op(iser_task->sc) == SCSI_PROT_READ_STRIP) { - sig_sge->length += (data_sge->length / - iser_task->sc->device->sector_size) * 8; - } + sig_sge->length = scsi_transfer_length(iser_task->sc); iser_dbg("sig_sge: addr: 0x%llx length: %u lkey: 0x%x\n", sig_sge->addr, sig_sge->length, @@ -613,7 +588,6 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, struct ib_fast_reg_page_list *frpl; struct ib_send_wr fastreg_wr, inv_wr; struct ib_send_wr *bad_wr, *wr = NULL; - u8 key; int ret, offset, size, plen; /* if there a single dma entry, dma mr suffices */ @@ -645,14 +619,8 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, } if (!(desc->reg_indicators & ind)) { - memset(&inv_wr, 0, sizeof(inv_wr)); - inv_wr.wr_id = ISER_FASTREG_LI_WRID; - inv_wr.opcode = IB_WR_LOCAL_INV; - inv_wr.ex.invalidate_rkey = mr->rkey; + iser_inv_rkey(&inv_wr, mr); wr = &inv_wr; - /* Bump the key */ - key = (u8)(mr->rkey & 0x000000FF); - ib_update_fast_reg_key(mr, ++key); } /* Prepare FASTREG WR */ @@ -770,15 +738,11 @@ int iser_reg_rdma_mem_fastreg(struct iscsi_iser_task *iser_task, regd_buf->reg.rkey = desc->pi_ctx->sig_mr->rkey; regd_buf->reg.va = sig_sge.addr; regd_buf->reg.len = sig_sge.length; - regd_buf->reg.is_mr = 1; } else { - if (desc) { + if (desc) regd_buf->reg.rkey = desc->data_mr->rkey; - regd_buf->reg.is_mr = 1; - } else { + else regd_buf->reg.rkey = device->mr->rkey; - regd_buf->reg.is_mr = 0; - } regd_buf->reg.lkey = data_sge.lkey; regd_buf->reg.va = data_sge.addr; diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c index 67225bb82bb5..695a2704bd43 100644 --- a/drivers/infiniband/ulp/iser/iser_verbs.c +++ b/drivers/infiniband/ulp/iser/iser_verbs.c @@ -76,7 +76,7 @@ static void iser_event_handler(struct ib_event_handler *handler, static int iser_create_device_ib_res(struct iser_device *device) { struct ib_device_attr *dev_attr = &device->dev_attr; - int ret, i; + int ret, i, max_cqe; ret = ib_query_device(device->ib_device, dev_attr); if (ret) { @@ -104,11 +104,19 @@ static int iser_create_device_ib_res(struct iser_device *device) return -1; } - device->comps_used = min(ISER_MAX_CQ, + device->comps_used = min_t(int, num_online_cpus(), device->ib_device->num_comp_vectors); - iser_info("using %d CQs, device %s supports %d vectors\n", + + device->comps = kcalloc(device->comps_used, sizeof(*device->comps), + GFP_KERNEL); + if (!device->comps) + goto comps_err; + + max_cqe = min(ISER_MAX_CQ_LEN, dev_attr->max_cqe); + + iser_info("using %d CQs, device %s supports %d vectors max_cqe %d\n", device->comps_used, device->ib_device->name, - device->ib_device->num_comp_vectors); + device->ib_device->num_comp_vectors, max_cqe); device->pd = ib_alloc_pd(device->ib_device); if (IS_ERR(device->pd)) @@ -122,7 +130,7 @@ static int iser_create_device_ib_res(struct iser_device *device) iser_cq_callback, iser_cq_event_callback, (void *)comp, - ISER_MAX_CQ_LEN, i); + max_cqe, i); if (IS_ERR(comp->cq)) { comp->cq = NULL; goto cq_err; @@ -162,6 +170,8 @@ cq_err: } ib_dealloc_pd(device->pd); pd_err: + kfree(device->comps); +comps_err: iser_err("failed to allocate an IB resource\n"); return -1; } @@ -187,6 +197,9 @@ static void iser_free_device_ib_res(struct iser_device *device) (void)ib_dereg_mr(device->mr); (void)ib_dealloc_pd(device->pd); + kfree(device->comps); + device->comps = NULL; + device->mr = NULL; device->pd = NULL; } @@ -425,7 +438,10 @@ void iser_free_fastreg_pool(struct ib_conn *ib_conn) */ static int iser_create_ib_conn_res(struct ib_conn *ib_conn) { + struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, + ib_conn); struct iser_device *device; + struct ib_device_attr *dev_attr; struct ib_qp_init_attr init_attr; int ret = -ENOMEM; int index, min_index = 0; @@ -433,6 +449,7 @@ static int iser_create_ib_conn_res(struct ib_conn *ib_conn) BUG_ON(ib_conn->device == NULL); device = ib_conn->device; + dev_attr = &device->dev_attr; memset(&init_attr, 0, sizeof init_attr); @@ -460,8 +477,20 @@ static int iser_create_ib_conn_res(struct ib_conn *ib_conn) if (ib_conn->pi_support) { init_attr.cap.max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1; init_attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; + iser_conn->max_cmds = + ISER_GET_MAX_XMIT_CMDS(ISER_QP_SIG_MAX_REQ_DTOS); } else { - init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS + 1; + if (dev_attr->max_qp_wr > ISER_QP_MAX_REQ_DTOS) { + init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS + 1; + iser_conn->max_cmds = + ISER_GET_MAX_XMIT_CMDS(ISER_QP_MAX_REQ_DTOS); + } else { + init_attr.cap.max_send_wr = dev_attr->max_qp_wr; + iser_conn->max_cmds = + ISER_GET_MAX_XMIT_CMDS(dev_attr->max_qp_wr); + iser_dbg("device %s supports max_send_wr %d\n", + device->ib_device->name, dev_attr->max_qp_wr); + } } ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); @@ -475,7 +504,11 @@ static int iser_create_ib_conn_res(struct ib_conn *ib_conn) return ret; out_err: + mutex_lock(&ig.connlist_mutex); + ib_conn->comp->active_qps--; + mutex_unlock(&ig.connlist_mutex); iser_err("unable to alloc mem or create resource, err %d\n", ret); + return ret; } @@ -610,9 +643,11 @@ void iser_conn_release(struct iser_conn *iser_conn) mutex_unlock(&ig.connlist_mutex); mutex_lock(&iser_conn->state_mutex); - if (iser_conn->state != ISER_CONN_DOWN) + if (iser_conn->state != ISER_CONN_DOWN) { iser_warn("iser conn %p state %d, expected state down.\n", iser_conn, iser_conn->state); + iser_conn->state = ISER_CONN_DOWN; + } /* * In case we never got to bind stage, we still need to * release IB resources (which is safe to call more than once). @@ -662,8 +697,10 @@ int iser_conn_terminate(struct iser_conn *iser_conn) /* post an indication that all flush errors were consumed */ err = ib_post_send(ib_conn->qp, &ib_conn->beacon, &bad_wr); - if (err) + if (err) { iser_err("conn %p failed to post beacon", ib_conn); + return 1; + } wait_for_completion(&ib_conn->flush_comp); } @@ -846,20 +883,21 @@ static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *eve break; case RDMA_CM_EVENT_DISCONNECTED: case RDMA_CM_EVENT_ADDR_CHANGE: - iser_disconnected_handler(cma_id); + case RDMA_CM_EVENT_TIMEWAIT_EXIT: + iser_cleanup_handler(cma_id, false); break; case RDMA_CM_EVENT_DEVICE_REMOVAL: /* * we *must* destroy the device as we cannot rely * on iscsid to be around to initiate error handling. - * also implicitly destroy the cma_id. + * also if we are not in state DOWN implicitly destroy + * the cma_id. */ iser_cleanup_handler(cma_id, true); - iser_conn->ib_conn.cma_id = NULL; - ret = 1; - break; - case RDMA_CM_EVENT_TIMEWAIT_EXIT: - iser_cleanup_handler(cma_id, false); + if (iser_conn->state != ISER_CONN_DOWN) { + iser_conn->ib_conn.cma_id = NULL; + ret = 1; + } break; default: iser_err("Unexpected RDMA CM event (%d)\n", event->event); @@ -981,7 +1019,6 @@ int iser_reg_page_vec(struct ib_conn *ib_conn, mem_reg->rkey = mem->fmr->rkey; mem_reg->len = page_vec->length * SIZE_4K; mem_reg->va = io_addr; - mem_reg->is_mr = 1; mem_reg->mem_h = (void *)mem; mem_reg->va += page_vec->offset; @@ -1008,7 +1045,7 @@ void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; int ret; - if (!reg->is_mr) + if (!reg->mem_h) return; iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h); @@ -1028,11 +1065,10 @@ void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task, struct ib_conn *ib_conn = &iser_conn->ib_conn; struct fast_reg_descriptor *desc = reg->mem_h; - if (!reg->is_mr) + if (!desc) return; reg->mem_h = NULL; - reg->is_mr = 0; spin_lock_bh(&ib_conn->lock); list_add_tail(&desc->list, &ib_conn->fastreg.pool); spin_unlock_bh(&ib_conn->lock); @@ -1049,7 +1085,7 @@ int iser_post_recvl(struct iser_conn *iser_conn) sge.length = ISER_RX_LOGIN_SIZE; sge.lkey = ib_conn->device->mr->lkey; - rx_wr.wr_id = (unsigned long)iser_conn->login_resp_buf; + rx_wr.wr_id = (uintptr_t)iser_conn->login_resp_buf; rx_wr.sg_list = &sge; rx_wr.num_sge = 1; rx_wr.next = NULL; @@ -1073,7 +1109,7 @@ int iser_post_recvm(struct iser_conn *iser_conn, int count) for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { rx_desc = &iser_conn->rx_descs[my_rx_head]; - rx_wr->wr_id = (unsigned long)rx_desc; + rx_wr->wr_id = (uintptr_t)rx_desc; rx_wr->sg_list = &rx_desc->rx_sg; rx_wr->num_sge = 1; rx_wr->next = rx_wr + 1; @@ -1110,7 +1146,7 @@ int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc, DMA_TO_DEVICE); send_wr.next = NULL; - send_wr.wr_id = (unsigned long)tx_desc; + send_wr.wr_id = (uintptr_t)tx_desc; send_wr.sg_list = tx_desc->tx_sg; send_wr.num_sge = tx_desc->num_sge; send_wr.opcode = IB_WR_SEND; @@ -1160,6 +1196,7 @@ static void iser_handle_comp_error(struct ib_conn *ib_conn, struct ib_wc *wc) { + void *wr_id = (void *)(uintptr_t)wc->wr_id; struct iser_conn *iser_conn = container_of(ib_conn, struct iser_conn, ib_conn); @@ -1168,8 +1205,8 @@ iser_handle_comp_error(struct ib_conn *ib_conn, iscsi_conn_failure(iser_conn->iscsi_conn, ISCSI_ERR_CONN_FAILED); - if (is_iser_tx_desc(iser_conn, (void *)wc->wr_id)) { - struct iser_tx_desc *desc = (struct iser_tx_desc *)wc->wr_id; + if (is_iser_tx_desc(iser_conn, wr_id)) { + struct iser_tx_desc *desc = wr_id; if (desc->type == ISCSI_TX_DATAOUT) kmem_cache_free(ig.desc_cache, desc); @@ -1193,14 +1230,14 @@ static void iser_handle_wc(struct ib_wc *wc) struct iser_rx_desc *rx_desc; ib_conn = wc->qp->qp_context; - if (wc->status == IB_WC_SUCCESS) { + if (likely(wc->status == IB_WC_SUCCESS)) { if (wc->opcode == IB_WC_RECV) { - rx_desc = (struct iser_rx_desc *)wc->wr_id; + rx_desc = (struct iser_rx_desc *)(uintptr_t)wc->wr_id; iser_rcv_completion(rx_desc, wc->byte_len, ib_conn); } else if (wc->opcode == IB_WC_SEND) { - tx_desc = (struct iser_tx_desc *)wc->wr_id; + tx_desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id; iser_snd_completion(tx_desc, ib_conn); } else { iser_err("Unknown wc opcode %d\n", wc->opcode); diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 10641b7816f4..dafb3c531f96 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -22,7 +22,6 @@ #include <linux/socket.h> #include <linux/in.h> #include <linux/in6.h> -#include <linux/llist.h> #include <rdma/ib_verbs.h> #include <rdma/rdma_cm.h> #include <target/target_core_base.h> @@ -36,11 +35,17 @@ #define ISERT_MAX_CONN 8 #define ISER_MAX_RX_CQ_LEN (ISERT_QP_MAX_RECV_DTOS * ISERT_MAX_CONN) #define ISER_MAX_TX_CQ_LEN (ISERT_QP_MAX_REQ_DTOS * ISERT_MAX_CONN) +#define ISER_MAX_CQ_LEN (ISER_MAX_RX_CQ_LEN + ISER_MAX_TX_CQ_LEN + \ + ISERT_MAX_CONN) + +int isert_debug_level = 0; +module_param_named(debug_level, isert_debug_level, int, 0644); +MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:0)"); static DEFINE_MUTEX(device_list_mutex); static LIST_HEAD(device_list); -static struct workqueue_struct *isert_rx_wq; static struct workqueue_struct *isert_comp_wq; +static struct workqueue_struct *isert_release_wq; static void isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn); @@ -54,19 +59,32 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, struct isert_rdma_wr *wr); static int isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd); +static int +isert_rdma_post_recvl(struct isert_conn *isert_conn); +static int +isert_rdma_accept(struct isert_conn *isert_conn); +struct rdma_cm_id *isert_setup_id(struct isert_np *isert_np); + +static inline bool +isert_prot_cmd(struct isert_conn *conn, struct se_cmd *cmd) +{ + return (conn->pi_support && + cmd->prot_op != TARGET_PROT_NORMAL); +} + static void isert_qp_event_callback(struct ib_event *e, void *context) { struct isert_conn *isert_conn = (struct isert_conn *)context; - pr_err("isert_qp_event_callback event: %d\n", e->event); + isert_err("conn %p event: %d\n", isert_conn, e->event); switch (e->event) { case IB_EVENT_COMM_EST: rdma_notify(isert_conn->conn_cm_id, IB_EVENT_COMM_EST); break; case IB_EVENT_QP_LAST_WQE_REACHED: - pr_warn("Reached TX IB_EVENT_QP_LAST_WQE_REACHED:\n"); + isert_warn("Reached TX IB_EVENT_QP_LAST_WQE_REACHED\n"); break; default: break; @@ -80,39 +98,41 @@ isert_query_device(struct ib_device *ib_dev, struct ib_device_attr *devattr) ret = ib_query_device(ib_dev, devattr); if (ret) { - pr_err("ib_query_device() failed: %d\n", ret); + isert_err("ib_query_device() failed: %d\n", ret); return ret; } - pr_debug("devattr->max_sge: %d\n", devattr->max_sge); - pr_debug("devattr->max_sge_rd: %d\n", devattr->max_sge_rd); + isert_dbg("devattr->max_sge: %d\n", devattr->max_sge); + isert_dbg("devattr->max_sge_rd: %d\n", devattr->max_sge_rd); return 0; } static int -isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id, - u8 protection) +isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id) { struct isert_device *device = isert_conn->conn_device; struct ib_qp_init_attr attr; - int ret, index, min_index = 0; + struct isert_comp *comp; + int ret, i, min = 0; mutex_lock(&device_list_mutex); - for (index = 0; index < device->cqs_used; index++) - if (device->cq_active_qps[index] < - device->cq_active_qps[min_index]) - min_index = index; - device->cq_active_qps[min_index]++; - pr_debug("isert_conn_setup_qp: Using min_index: %d\n", min_index); + for (i = 0; i < device->comps_used; i++) + if (device->comps[i].active_qps < + device->comps[min].active_qps) + min = i; + comp = &device->comps[min]; + comp->active_qps++; + isert_info("conn %p, using comp %p min_index: %d\n", + isert_conn, comp, min); mutex_unlock(&device_list_mutex); memset(&attr, 0, sizeof(struct ib_qp_init_attr)); attr.event_handler = isert_qp_event_callback; attr.qp_context = isert_conn; - attr.send_cq = device->dev_tx_cq[min_index]; - attr.recv_cq = device->dev_rx_cq[min_index]; + attr.send_cq = comp->cq; + attr.recv_cq = comp->cq; attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS; - attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS; + attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1; /* * FIXME: Use devattr.max_sge - 2 for max_send_sge as * work-around for RDMA_READs with ConnectX-2. @@ -126,29 +146,29 @@ isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id, attr.cap.max_recv_sge = 1; attr.sq_sig_type = IB_SIGNAL_REQ_WR; attr.qp_type = IB_QPT_RC; - if (protection) + if (device->pi_capable) attr.create_flags |= IB_QP_CREATE_SIGNATURE_EN; - pr_debug("isert_conn_setup_qp cma_id->device: %p\n", - cma_id->device); - pr_debug("isert_conn_setup_qp conn_pd->device: %p\n", - isert_conn->conn_pd->device); - ret = rdma_create_qp(cma_id, isert_conn->conn_pd, &attr); if (ret) { - pr_err("rdma_create_qp failed for cma_id %d\n", ret); - return ret; + isert_err("rdma_create_qp failed for cma_id %d\n", ret); + goto err; } isert_conn->conn_qp = cma_id->qp; - pr_debug("rdma_create_qp() returned success >>>>>>>>>>>>>>>>>>>>>>>>>.\n"); return 0; +err: + mutex_lock(&device_list_mutex); + comp->active_qps--; + mutex_unlock(&device_list_mutex); + + return ret; } static void isert_cq_event_callback(struct ib_event *e, void *context) { - pr_debug("isert_cq_event_callback event: %d\n", e->event); + isert_dbg("event: %d\n", e->event); } static int @@ -182,6 +202,7 @@ isert_alloc_rx_descriptors(struct isert_conn *isert_conn) } isert_conn->conn_rx_desc_head = 0; + return 0; dma_map_fail: @@ -193,6 +214,8 @@ dma_map_fail: kfree(isert_conn->conn_rx_descs); isert_conn->conn_rx_descs = NULL; fail: + isert_err("conn %p failed to allocate rx descriptors\n", isert_conn); + return -ENOMEM; } @@ -216,27 +239,23 @@ isert_free_rx_descriptors(struct isert_conn *isert_conn) isert_conn->conn_rx_descs = NULL; } -static void isert_cq_tx_work(struct work_struct *); -static void isert_cq_tx_callback(struct ib_cq *, void *); -static void isert_cq_rx_work(struct work_struct *); -static void isert_cq_rx_callback(struct ib_cq *, void *); +static void isert_cq_work(struct work_struct *); +static void isert_cq_callback(struct ib_cq *, void *); static int isert_create_device_ib_res(struct isert_device *device) { struct ib_device *ib_dev = device->ib_device; - struct isert_cq_desc *cq_desc; struct ib_device_attr *dev_attr; - int ret = 0, i, j; - int max_rx_cqe, max_tx_cqe; + int ret = 0, i; + int max_cqe; dev_attr = &device->dev_attr; ret = isert_query_device(ib_dev, dev_attr); if (ret) return ret; - max_rx_cqe = min(ISER_MAX_RX_CQ_LEN, dev_attr->max_cqe); - max_tx_cqe = min(ISER_MAX_TX_CQ_LEN, dev_attr->max_cqe); + max_cqe = min(ISER_MAX_CQ_LEN, dev_attr->max_cqe); /* asign function handlers */ if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS && @@ -254,55 +273,38 @@ isert_create_device_ib_res(struct isert_device *device) device->pi_capable = dev_attr->device_cap_flags & IB_DEVICE_SIGNATURE_HANDOVER ? true : false; - device->cqs_used = min_t(int, num_online_cpus(), - device->ib_device->num_comp_vectors); - device->cqs_used = min(ISERT_MAX_CQ, device->cqs_used); - pr_debug("Using %d CQs, device %s supports %d vectors support " - "Fast registration %d pi_capable %d\n", - device->cqs_used, device->ib_device->name, - device->ib_device->num_comp_vectors, device->use_fastreg, - device->pi_capable); - device->cq_desc = kzalloc(sizeof(struct isert_cq_desc) * - device->cqs_used, GFP_KERNEL); - if (!device->cq_desc) { - pr_err("Unable to allocate device->cq_desc\n"); + device->comps_used = min(ISERT_MAX_CQ, min_t(int, num_online_cpus(), + device->ib_device->num_comp_vectors)); + isert_info("Using %d CQs, %s supports %d vectors support " + "Fast registration %d pi_capable %d\n", + device->comps_used, device->ib_device->name, + device->ib_device->num_comp_vectors, device->use_fastreg, + device->pi_capable); + + device->comps = kcalloc(device->comps_used, sizeof(struct isert_comp), + GFP_KERNEL); + if (!device->comps) { + isert_err("Unable to allocate completion contexts\n"); return -ENOMEM; } - cq_desc = device->cq_desc; - - for (i = 0; i < device->cqs_used; i++) { - cq_desc[i].device = device; - cq_desc[i].cq_index = i; - - INIT_WORK(&cq_desc[i].cq_rx_work, isert_cq_rx_work); - device->dev_rx_cq[i] = ib_create_cq(device->ib_device, - isert_cq_rx_callback, - isert_cq_event_callback, - (void *)&cq_desc[i], - max_rx_cqe, i); - if (IS_ERR(device->dev_rx_cq[i])) { - ret = PTR_ERR(device->dev_rx_cq[i]); - device->dev_rx_cq[i] = NULL; - goto out_cq; - } - INIT_WORK(&cq_desc[i].cq_tx_work, isert_cq_tx_work); - device->dev_tx_cq[i] = ib_create_cq(device->ib_device, - isert_cq_tx_callback, - isert_cq_event_callback, - (void *)&cq_desc[i], - max_tx_cqe, i); - if (IS_ERR(device->dev_tx_cq[i])) { - ret = PTR_ERR(device->dev_tx_cq[i]); - device->dev_tx_cq[i] = NULL; - goto out_cq; - } + for (i = 0; i < device->comps_used; i++) { + struct isert_comp *comp = &device->comps[i]; - ret = ib_req_notify_cq(device->dev_rx_cq[i], IB_CQ_NEXT_COMP); - if (ret) + comp->device = device; + INIT_WORK(&comp->work, isert_cq_work); + comp->cq = ib_create_cq(device->ib_device, + isert_cq_callback, + isert_cq_event_callback, + (void *)comp, + max_cqe, i); + if (IS_ERR(comp->cq)) { + ret = PTR_ERR(comp->cq); + comp->cq = NULL; goto out_cq; + } - ret = ib_req_notify_cq(device->dev_tx_cq[i], IB_CQ_NEXT_COMP); + ret = ib_req_notify_cq(comp->cq, IB_CQ_NEXT_COMP); if (ret) goto out_cq; } @@ -310,19 +312,15 @@ isert_create_device_ib_res(struct isert_device *device) return 0; out_cq: - for (j = 0; j < i; j++) { - cq_desc = &device->cq_desc[j]; + for (i = 0; i < device->comps_used; i++) { + struct isert_comp *comp = &device->comps[i]; - if (device->dev_rx_cq[j]) { - cancel_work_sync(&cq_desc->cq_rx_work); - ib_destroy_cq(device->dev_rx_cq[j]); - } - if (device->dev_tx_cq[j]) { - cancel_work_sync(&cq_desc->cq_tx_work); - ib_destroy_cq(device->dev_tx_cq[j]); + if (comp->cq) { + cancel_work_sync(&comp->work); + ib_destroy_cq(comp->cq); } } - kfree(device->cq_desc); + kfree(device->comps); return ret; } @@ -330,21 +328,18 @@ out_cq: static void isert_free_device_ib_res(struct isert_device *device) { - struct isert_cq_desc *cq_desc; int i; - for (i = 0; i < device->cqs_used; i++) { - cq_desc = &device->cq_desc[i]; + isert_info("device %p\n", device); - cancel_work_sync(&cq_desc->cq_rx_work); - cancel_work_sync(&cq_desc->cq_tx_work); - ib_destroy_cq(device->dev_rx_cq[i]); - ib_destroy_cq(device->dev_tx_cq[i]); - device->dev_rx_cq[i] = NULL; - device->dev_tx_cq[i] = NULL; - } + for (i = 0; i < device->comps_used; i++) { + struct isert_comp *comp = &device->comps[i]; - kfree(device->cq_desc); + cancel_work_sync(&comp->work); + ib_destroy_cq(comp->cq); + comp->cq = NULL; + } + kfree(device->comps); } static void @@ -352,6 +347,7 @@ isert_device_try_release(struct isert_device *device) { mutex_lock(&device_list_mutex); device->refcount--; + isert_info("device %p refcount %d\n", device, device->refcount); if (!device->refcount) { isert_free_device_ib_res(device); list_del(&device->dev_node); @@ -370,6 +366,8 @@ isert_device_find_by_ib_dev(struct rdma_cm_id *cma_id) list_for_each_entry(device, &device_list, dev_node) { if (device->ib_device->node_guid == cma_id->device->node_guid) { device->refcount++; + isert_info("Found iser device %p refcount %d\n", + device, device->refcount); mutex_unlock(&device_list_mutex); return device; } @@ -393,6 +391,8 @@ isert_device_find_by_ib_dev(struct rdma_cm_id *cma_id) device->refcount++; list_add_tail(&device->dev_node, &device_list); + isert_info("Created a new iser device %p refcount %d\n", + device, device->refcount); mutex_unlock(&device_list_mutex); return device; @@ -407,7 +407,7 @@ isert_conn_free_fastreg_pool(struct isert_conn *isert_conn) if (list_empty(&isert_conn->conn_fr_pool)) return; - pr_debug("Freeing conn %p fastreg pool", isert_conn); + isert_info("Freeing conn %p fastreg pool", isert_conn); list_for_each_entry_safe(fr_desc, tmp, &isert_conn->conn_fr_pool, list) { @@ -425,87 +425,97 @@ isert_conn_free_fastreg_pool(struct isert_conn *isert_conn) } if (i < isert_conn->conn_fr_pool_size) - pr_warn("Pool still has %d regions registered\n", + isert_warn("Pool still has %d regions registered\n", isert_conn->conn_fr_pool_size - i); } static int +isert_create_pi_ctx(struct fast_reg_descriptor *desc, + struct ib_device *device, + struct ib_pd *pd) +{ + struct ib_mr_init_attr mr_init_attr; + struct pi_context *pi_ctx; + int ret; + + pi_ctx = kzalloc(sizeof(*desc->pi_ctx), GFP_KERNEL); + if (!pi_ctx) { + isert_err("Failed to allocate pi context\n"); + return -ENOMEM; + } + + pi_ctx->prot_frpl = ib_alloc_fast_reg_page_list(device, + ISCSI_ISER_SG_TABLESIZE); + if (IS_ERR(pi_ctx->prot_frpl)) { + isert_err("Failed to allocate prot frpl err=%ld\n", + PTR_ERR(pi_ctx->prot_frpl)); + ret = PTR_ERR(pi_ctx->prot_frpl); + goto err_pi_ctx; + } + + pi_ctx->prot_mr = ib_alloc_fast_reg_mr(pd, ISCSI_ISER_SG_TABLESIZE); + if (IS_ERR(pi_ctx->prot_mr)) { + isert_err("Failed to allocate prot frmr err=%ld\n", + PTR_ERR(pi_ctx->prot_mr)); + ret = PTR_ERR(pi_ctx->prot_mr); + goto err_prot_frpl; + } + desc->ind |= ISERT_PROT_KEY_VALID; + + memset(&mr_init_attr, 0, sizeof(mr_init_attr)); + mr_init_attr.max_reg_descriptors = 2; + mr_init_attr.flags |= IB_MR_SIGNATURE_EN; + pi_ctx->sig_mr = ib_create_mr(pd, &mr_init_attr); + if (IS_ERR(pi_ctx->sig_mr)) { + isert_err("Failed to allocate signature enabled mr err=%ld\n", + PTR_ERR(pi_ctx->sig_mr)); + ret = PTR_ERR(pi_ctx->sig_mr); + goto err_prot_mr; + } + + desc->pi_ctx = pi_ctx; + desc->ind |= ISERT_SIG_KEY_VALID; + desc->ind &= ~ISERT_PROTECTED; + + return 0; + +err_prot_mr: + ib_dereg_mr(desc->pi_ctx->prot_mr); +err_prot_frpl: + ib_free_fast_reg_page_list(desc->pi_ctx->prot_frpl); +err_pi_ctx: + kfree(desc->pi_ctx); + + return ret; +} + +static int isert_create_fr_desc(struct ib_device *ib_device, struct ib_pd *pd, - struct fast_reg_descriptor *fr_desc, u8 protection) + struct fast_reg_descriptor *fr_desc) { int ret; fr_desc->data_frpl = ib_alloc_fast_reg_page_list(ib_device, ISCSI_ISER_SG_TABLESIZE); if (IS_ERR(fr_desc->data_frpl)) { - pr_err("Failed to allocate data frpl err=%ld\n", - PTR_ERR(fr_desc->data_frpl)); + isert_err("Failed to allocate data frpl err=%ld\n", + PTR_ERR(fr_desc->data_frpl)); return PTR_ERR(fr_desc->data_frpl); } fr_desc->data_mr = ib_alloc_fast_reg_mr(pd, ISCSI_ISER_SG_TABLESIZE); if (IS_ERR(fr_desc->data_mr)) { - pr_err("Failed to allocate data frmr err=%ld\n", - PTR_ERR(fr_desc->data_mr)); + isert_err("Failed to allocate data frmr err=%ld\n", + PTR_ERR(fr_desc->data_mr)); ret = PTR_ERR(fr_desc->data_mr); goto err_data_frpl; } - pr_debug("Create fr_desc %p page_list %p\n", - fr_desc, fr_desc->data_frpl->page_list); fr_desc->ind |= ISERT_DATA_KEY_VALID; - if (protection) { - struct ib_mr_init_attr mr_init_attr = {0}; - struct pi_context *pi_ctx; - - fr_desc->pi_ctx = kzalloc(sizeof(*fr_desc->pi_ctx), GFP_KERNEL); - if (!fr_desc->pi_ctx) { - pr_err("Failed to allocate pi context\n"); - ret = -ENOMEM; - goto err_data_mr; - } - pi_ctx = fr_desc->pi_ctx; - - pi_ctx->prot_frpl = ib_alloc_fast_reg_page_list(ib_device, - ISCSI_ISER_SG_TABLESIZE); - if (IS_ERR(pi_ctx->prot_frpl)) { - pr_err("Failed to allocate prot frpl err=%ld\n", - PTR_ERR(pi_ctx->prot_frpl)); - ret = PTR_ERR(pi_ctx->prot_frpl); - goto err_pi_ctx; - } - - pi_ctx->prot_mr = ib_alloc_fast_reg_mr(pd, ISCSI_ISER_SG_TABLESIZE); - if (IS_ERR(pi_ctx->prot_mr)) { - pr_err("Failed to allocate prot frmr err=%ld\n", - PTR_ERR(pi_ctx->prot_mr)); - ret = PTR_ERR(pi_ctx->prot_mr); - goto err_prot_frpl; - } - fr_desc->ind |= ISERT_PROT_KEY_VALID; - - mr_init_attr.max_reg_descriptors = 2; - mr_init_attr.flags |= IB_MR_SIGNATURE_EN; - pi_ctx->sig_mr = ib_create_mr(pd, &mr_init_attr); - if (IS_ERR(pi_ctx->sig_mr)) { - pr_err("Failed to allocate signature enabled mr err=%ld\n", - PTR_ERR(pi_ctx->sig_mr)); - ret = PTR_ERR(pi_ctx->sig_mr); - goto err_prot_mr; - } - fr_desc->ind |= ISERT_SIG_KEY_VALID; - } - fr_desc->ind &= ~ISERT_PROTECTED; + isert_dbg("Created fr_desc %p\n", fr_desc); return 0; -err_prot_mr: - ib_dereg_mr(fr_desc->pi_ctx->prot_mr); -err_prot_frpl: - ib_free_fast_reg_page_list(fr_desc->pi_ctx->prot_frpl); -err_pi_ctx: - kfree(fr_desc->pi_ctx); -err_data_mr: - ib_dereg_mr(fr_desc->data_mr); + err_data_frpl: ib_free_fast_reg_page_list(fr_desc->data_frpl); @@ -513,7 +523,7 @@ err_data_frpl: } static int -isert_conn_create_fastreg_pool(struct isert_conn *isert_conn, u8 pi_support) +isert_conn_create_fastreg_pool(struct isert_conn *isert_conn) { struct fast_reg_descriptor *fr_desc; struct isert_device *device = isert_conn->conn_device; @@ -531,16 +541,15 @@ isert_conn_create_fastreg_pool(struct isert_conn *isert_conn, u8 pi_support) for (i = 0; i < tag_num; i++) { fr_desc = kzalloc(sizeof(*fr_desc), GFP_KERNEL); if (!fr_desc) { - pr_err("Failed to allocate fast_reg descriptor\n"); + isert_err("Failed to allocate fast_reg descriptor\n"); ret = -ENOMEM; goto err; } ret = isert_create_fr_desc(device->ib_device, - isert_conn->conn_pd, fr_desc, - pi_support); + isert_conn->conn_pd, fr_desc); if (ret) { - pr_err("Failed to create fastreg descriptor err=%d\n", + isert_err("Failed to create fastreg descriptor err=%d\n", ret); kfree(fr_desc); goto err; @@ -550,7 +559,7 @@ isert_conn_create_fastreg_pool(struct isert_conn *isert_conn, u8 pi_support) isert_conn->conn_fr_pool_size++; } - pr_debug("Creating conn %p fastreg pool size=%d", + isert_dbg("Creating conn %p fastreg pool size=%d", isert_conn, isert_conn->conn_fr_pool_size); return 0; @@ -563,47 +572,45 @@ err: static int isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) { - struct iscsi_np *np = cma_id->context; - struct isert_np *isert_np = np->np_context; + struct isert_np *isert_np = cma_id->context; + struct iscsi_np *np = isert_np->np; struct isert_conn *isert_conn; struct isert_device *device; struct ib_device *ib_dev = cma_id->device; int ret = 0; - u8 pi_support; spin_lock_bh(&np->np_thread_lock); if (!np->enabled) { spin_unlock_bh(&np->np_thread_lock); - pr_debug("iscsi_np is not enabled, reject connect request\n"); + isert_dbg("iscsi_np is not enabled, reject connect request\n"); return rdma_reject(cma_id, NULL, 0); } spin_unlock_bh(&np->np_thread_lock); - pr_debug("Entering isert_connect_request cma_id: %p, context: %p\n", + isert_dbg("cma_id: %p, portal: %p\n", cma_id, cma_id->context); isert_conn = kzalloc(sizeof(struct isert_conn), GFP_KERNEL); if (!isert_conn) { - pr_err("Unable to allocate isert_conn\n"); + isert_err("Unable to allocate isert_conn\n"); return -ENOMEM; } isert_conn->state = ISER_CONN_INIT; INIT_LIST_HEAD(&isert_conn->conn_accept_node); init_completion(&isert_conn->conn_login_comp); + init_completion(&isert_conn->login_req_comp); init_completion(&isert_conn->conn_wait); - init_completion(&isert_conn->conn_wait_comp_err); kref_init(&isert_conn->conn_kref); mutex_init(&isert_conn->conn_mutex); spin_lock_init(&isert_conn->conn_lock); INIT_LIST_HEAD(&isert_conn->conn_fr_pool); - cma_id->context = isert_conn; isert_conn->conn_cm_id = cma_id; isert_conn->login_buf = kzalloc(ISCSI_DEF_MAX_RECV_SEG_LEN + ISER_RX_LOGIN_SIZE, GFP_KERNEL); if (!isert_conn->login_buf) { - pr_err("Unable to allocate isert_conn->login_buf\n"); + isert_err("Unable to allocate isert_conn->login_buf\n"); ret = -ENOMEM; goto out; } @@ -611,7 +618,7 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) isert_conn->login_req_buf = isert_conn->login_buf; isert_conn->login_rsp_buf = isert_conn->login_buf + ISCSI_DEF_MAX_RECV_SEG_LEN; - pr_debug("Set login_buf: %p login_req_buf: %p login_rsp_buf: %p\n", + isert_dbg("Set login_buf: %p login_req_buf: %p login_rsp_buf: %p\n", isert_conn->login_buf, isert_conn->login_req_buf, isert_conn->login_rsp_buf); @@ -621,7 +628,7 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) ret = ib_dma_mapping_error(ib_dev, isert_conn->login_req_dma); if (ret) { - pr_err("ib_dma_mapping_error failed for login_req_dma: %d\n", + isert_err("ib_dma_mapping_error failed for login_req_dma: %d\n", ret); isert_conn->login_req_dma = 0; goto out_login_buf; @@ -633,7 +640,7 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) ret = ib_dma_mapping_error(ib_dev, isert_conn->login_rsp_dma); if (ret) { - pr_err("ib_dma_mapping_error failed for login_rsp_dma: %d\n", + isert_err("ib_dma_mapping_error failed for login_rsp_dma: %d\n", ret); isert_conn->login_rsp_dma = 0; goto out_req_dma_map; @@ -649,13 +656,13 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) isert_conn->initiator_depth = min_t(u8, event->param.conn.initiator_depth, device->dev_attr.max_qp_init_rd_atom); - pr_debug("Using initiator_depth: %u\n", isert_conn->initiator_depth); + isert_dbg("Using initiator_depth: %u\n", isert_conn->initiator_depth); isert_conn->conn_device = device; isert_conn->conn_pd = ib_alloc_pd(isert_conn->conn_device->ib_device); if (IS_ERR(isert_conn->conn_pd)) { ret = PTR_ERR(isert_conn->conn_pd); - pr_err("ib_alloc_pd failed for conn %p: ret=%d\n", + isert_err("ib_alloc_pd failed for conn %p: ret=%d\n", isert_conn, ret); goto out_pd; } @@ -664,20 +671,20 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) IB_ACCESS_LOCAL_WRITE); if (IS_ERR(isert_conn->conn_mr)) { ret = PTR_ERR(isert_conn->conn_mr); - pr_err("ib_get_dma_mr failed for conn %p: ret=%d\n", + isert_err("ib_get_dma_mr failed for conn %p: ret=%d\n", isert_conn, ret); goto out_mr; } - pi_support = np->tpg_np->tpg->tpg_attrib.t10_pi; - if (pi_support && !device->pi_capable) { - pr_err("Protection information requested but not supported, " - "rejecting connect request\n"); - ret = rdma_reject(cma_id, NULL, 0); - goto out_mr; - } + ret = isert_conn_setup_qp(isert_conn, cma_id); + if (ret) + goto out_conn_dev; - ret = isert_conn_setup_qp(isert_conn, cma_id, pi_support); + ret = isert_rdma_post_recvl(isert_conn); + if (ret) + goto out_conn_dev; + + ret = isert_rdma_accept(isert_conn); if (ret) goto out_conn_dev; @@ -685,7 +692,7 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) list_add_tail(&isert_conn->conn_accept_node, &isert_np->np_accept_list); mutex_unlock(&isert_np->np_accept_mutex); - pr_debug("isert_connect_request() up np_sem np: %p\n", np); + isert_info("np %p: Allow accept_np to continue\n", np); up(&isert_np->np_sem); return 0; @@ -705,6 +712,7 @@ out_login_buf: kfree(isert_conn->login_buf); out: kfree(isert_conn); + rdma_reject(cma_id, NULL, 0); return ret; } @@ -713,24 +721,25 @@ isert_connect_release(struct isert_conn *isert_conn) { struct ib_device *ib_dev = isert_conn->conn_cm_id->device; struct isert_device *device = isert_conn->conn_device; - int cq_index; - pr_debug("Entering isert_connect_release(): >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("conn %p\n", isert_conn); if (device && device->use_fastreg) isert_conn_free_fastreg_pool(isert_conn); + isert_free_rx_descriptors(isert_conn); + rdma_destroy_id(isert_conn->conn_cm_id); + if (isert_conn->conn_qp) { - cq_index = ((struct isert_cq_desc *) - isert_conn->conn_qp->recv_cq->cq_context)->cq_index; - pr_debug("isert_connect_release: cq_index: %d\n", cq_index); - isert_conn->conn_device->cq_active_qps[cq_index]--; + struct isert_comp *comp = isert_conn->conn_qp->recv_cq->cq_context; - rdma_destroy_qp(isert_conn->conn_cm_id); - } + isert_dbg("dec completion context %p active_qps\n", comp); + mutex_lock(&device_list_mutex); + comp->active_qps--; + mutex_unlock(&device_list_mutex); - isert_free_rx_descriptors(isert_conn); - rdma_destroy_id(isert_conn->conn_cm_id); + ib_destroy_qp(isert_conn->conn_qp); + } ib_dereg_mr(isert_conn->conn_mr); ib_dealloc_pd(isert_conn->conn_pd); @@ -747,16 +756,24 @@ isert_connect_release(struct isert_conn *isert_conn) if (device) isert_device_try_release(device); - - pr_debug("Leaving isert_connect_release >>>>>>>>>>>>\n"); } static void isert_connected_handler(struct rdma_cm_id *cma_id) { - struct isert_conn *isert_conn = cma_id->context; + struct isert_conn *isert_conn = cma_id->qp->qp_context; - kref_get(&isert_conn->conn_kref); + isert_info("conn %p\n", isert_conn); + + if (!kref_get_unless_zero(&isert_conn->conn_kref)) { + isert_warn("conn %p connect_release is running\n", isert_conn); + return; + } + + mutex_lock(&isert_conn->conn_mutex); + if (isert_conn->state != ISER_CONN_FULL_FEATURE) + isert_conn->state = ISER_CONN_UP; + mutex_unlock(&isert_conn->conn_mutex); } static void @@ -765,8 +782,8 @@ isert_release_conn_kref(struct kref *kref) struct isert_conn *isert_conn = container_of(kref, struct isert_conn, conn_kref); - pr_debug("Calling isert_connect_release for final kref %s/%d\n", - current->comm, current->pid); + isert_info("conn %p final kref %s/%d\n", isert_conn, current->comm, + current->pid); isert_connect_release(isert_conn); } @@ -777,75 +794,111 @@ isert_put_conn(struct isert_conn *isert_conn) kref_put(&isert_conn->conn_kref, isert_release_conn_kref); } +/** + * isert_conn_terminate() - Initiate connection termination + * @isert_conn: isert connection struct + * + * Notes: + * In case the connection state is FULL_FEATURE, move state + * to TEMINATING and start teardown sequence (rdma_disconnect). + * In case the connection state is UP, complete flush as well. + * + * This routine must be called with conn_mutex held. Thus it is + * safe to call multiple times. + */ static void -isert_disconnect_work(struct work_struct *work) +isert_conn_terminate(struct isert_conn *isert_conn) { - struct isert_conn *isert_conn = container_of(work, - struct isert_conn, conn_logout_work); + int err; - pr_debug("isert_disconnect_work(): >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n"); - mutex_lock(&isert_conn->conn_mutex); - if (isert_conn->state == ISER_CONN_UP) + switch (isert_conn->state) { + case ISER_CONN_TERMINATING: + break; + case ISER_CONN_UP: + case ISER_CONN_FULL_FEATURE: /* FALLTHRU */ + isert_info("Terminating conn %p state %d\n", + isert_conn, isert_conn->state); isert_conn->state = ISER_CONN_TERMINATING; - - if (isert_conn->post_recv_buf_count == 0 && - atomic_read(&isert_conn->post_send_buf_count) == 0) { - mutex_unlock(&isert_conn->conn_mutex); - goto wake_up; - } - if (!isert_conn->conn_cm_id) { - mutex_unlock(&isert_conn->conn_mutex); - isert_put_conn(isert_conn); - return; + err = rdma_disconnect(isert_conn->conn_cm_id); + if (err) + isert_warn("Failed rdma_disconnect isert_conn %p\n", + isert_conn); + break; + default: + isert_warn("conn %p teminating in state %d\n", + isert_conn, isert_conn->state); } +} - if (isert_conn->disconnect) { - /* Send DREQ/DREP towards our initiator */ - rdma_disconnect(isert_conn->conn_cm_id); - } +static int +isert_np_cma_handler(struct isert_np *isert_np, + enum rdma_cm_event_type event) +{ + isert_dbg("isert np %p, handling event %d\n", isert_np, event); - mutex_unlock(&isert_conn->conn_mutex); + switch (event) { + case RDMA_CM_EVENT_DEVICE_REMOVAL: + isert_np->np_cm_id = NULL; + break; + case RDMA_CM_EVENT_ADDR_CHANGE: + isert_np->np_cm_id = isert_setup_id(isert_np); + if (IS_ERR(isert_np->np_cm_id)) { + isert_err("isert np %p setup id failed: %ld\n", + isert_np, PTR_ERR(isert_np->np_cm_id)); + isert_np->np_cm_id = NULL; + } + break; + default: + isert_err("isert np %p Unexpected event %d\n", + isert_np, event); + } -wake_up: - complete(&isert_conn->conn_wait); + return -1; } static int -isert_disconnected_handler(struct rdma_cm_id *cma_id, bool disconnect) +isert_disconnected_handler(struct rdma_cm_id *cma_id, + enum rdma_cm_event_type event) { + struct isert_np *isert_np = cma_id->context; struct isert_conn *isert_conn; - if (!cma_id->qp) { - struct isert_np *isert_np = cma_id->context; + if (isert_np->np_cm_id == cma_id) + return isert_np_cma_handler(cma_id->context, event); - isert_np->np_cm_id = NULL; - return -1; - } + isert_conn = cma_id->qp->qp_context; - isert_conn = (struct isert_conn *)cma_id->context; + mutex_lock(&isert_conn->conn_mutex); + isert_conn_terminate(isert_conn); + mutex_unlock(&isert_conn->conn_mutex); - isert_conn->disconnect = disconnect; - INIT_WORK(&isert_conn->conn_logout_work, isert_disconnect_work); - schedule_work(&isert_conn->conn_logout_work); + isert_info("conn %p completing conn_wait\n", isert_conn); + complete(&isert_conn->conn_wait); return 0; } +static void +isert_connect_error(struct rdma_cm_id *cma_id) +{ + struct isert_conn *isert_conn = cma_id->qp->qp_context; + + isert_put_conn(isert_conn); +} + static int isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) { int ret = 0; - bool disconnect = false; - pr_debug("isert_cma_handler: event %d status %d conn %p id %p\n", - event->event, event->status, cma_id->context, cma_id); + isert_info("event %d status %d id %p np %p\n", event->event, + event->status, cma_id, cma_id->context); switch (event->event) { case RDMA_CM_EVENT_CONNECT_REQUEST: ret = isert_connect_request(cma_id, event); if (ret) - pr_err("isert_cma_handler failed RDMA_CM_EVENT: 0x%08x %d\n", - event->event, ret); + isert_err("failed handle connect request %d\n", ret); break; case RDMA_CM_EVENT_ESTABLISHED: isert_connected_handler(cma_id); @@ -853,13 +906,16 @@ isert_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) case RDMA_CM_EVENT_ADDR_CHANGE: /* FALLTHRU */ case RDMA_CM_EVENT_DISCONNECTED: /* FALLTHRU */ case RDMA_CM_EVENT_DEVICE_REMOVAL: /* FALLTHRU */ - disconnect = true; case RDMA_CM_EVENT_TIMEWAIT_EXIT: /* FALLTHRU */ - ret = isert_disconnected_handler(cma_id, disconnect); + ret = isert_disconnected_handler(cma_id, event->event); break; + case RDMA_CM_EVENT_REJECTED: /* FALLTHRU */ + case RDMA_CM_EVENT_UNREACHABLE: /* FALLTHRU */ case RDMA_CM_EVENT_CONNECT_ERROR: + isert_connect_error(cma_id); + break; default: - pr_err("Unhandled RDMA CMA event: %d\n", event->event); + isert_err("Unhandled RDMA CMA event: %d\n", event->event); break; } @@ -876,7 +932,7 @@ isert_post_recv(struct isert_conn *isert_conn, u32 count) for (rx_wr = isert_conn->conn_rx_wr, i = 0; i < count; i++, rx_wr++) { rx_desc = &isert_conn->conn_rx_descs[rx_head]; - rx_wr->wr_id = (unsigned long)rx_desc; + rx_wr->wr_id = (uintptr_t)rx_desc; rx_wr->sg_list = &rx_desc->rx_sg; rx_wr->num_sge = 1; rx_wr->next = rx_wr + 1; @@ -890,10 +946,10 @@ isert_post_recv(struct isert_conn *isert_conn, u32 count) ret = ib_post_recv(isert_conn->conn_qp, isert_conn->conn_rx_wr, &rx_wr_failed); if (ret) { - pr_err("ib_post_recv() failed with ret: %d\n", ret); + isert_err("ib_post_recv() failed with ret: %d\n", ret); isert_conn->post_recv_buf_count -= count; } else { - pr_debug("isert_post_recv(): Posted %d RX buffers\n", count); + isert_dbg("isert_post_recv(): Posted %d RX buffers\n", count); isert_conn->conn_rx_desc_head = rx_head; } return ret; @@ -910,19 +966,15 @@ isert_post_send(struct isert_conn *isert_conn, struct iser_tx_desc *tx_desc) ISER_HEADERS_LEN, DMA_TO_DEVICE); send_wr.next = NULL; - send_wr.wr_id = (unsigned long)tx_desc; + send_wr.wr_id = (uintptr_t)tx_desc; send_wr.sg_list = tx_desc->tx_sg; send_wr.num_sge = tx_desc->num_sge; send_wr.opcode = IB_WR_SEND; send_wr.send_flags = IB_SEND_SIGNALED; - atomic_inc(&isert_conn->post_send_buf_count); - ret = ib_post_send(isert_conn->conn_qp, &send_wr, &send_wr_failed); - if (ret) { - pr_err("ib_post_send() failed, ret: %d\n", ret); - atomic_dec(&isert_conn->post_send_buf_count); - } + if (ret) + isert_err("ib_post_send() failed, ret: %d\n", ret); return ret; } @@ -945,7 +997,7 @@ isert_create_send_desc(struct isert_conn *isert_conn, if (tx_desc->tx_sg[0].lkey != isert_conn->conn_mr->lkey) { tx_desc->tx_sg[0].lkey = isert_conn->conn_mr->lkey; - pr_debug("tx_desc %p lkey mismatch, fixing\n", tx_desc); + isert_dbg("tx_desc %p lkey mismatch, fixing\n", tx_desc); } } @@ -959,7 +1011,7 @@ isert_init_tx_hdrs(struct isert_conn *isert_conn, dma_addr = ib_dma_map_single(ib_dev, (void *)tx_desc, ISER_HEADERS_LEN, DMA_TO_DEVICE); if (ib_dma_mapping_error(ib_dev, dma_addr)) { - pr_err("ib_dma_mapping_error() failed\n"); + isert_err("ib_dma_mapping_error() failed\n"); return -ENOMEM; } @@ -968,40 +1020,24 @@ isert_init_tx_hdrs(struct isert_conn *isert_conn, tx_desc->tx_sg[0].length = ISER_HEADERS_LEN; tx_desc->tx_sg[0].lkey = isert_conn->conn_mr->lkey; - pr_debug("isert_init_tx_hdrs: Setup tx_sg[0].addr: 0x%llx length: %u" - " lkey: 0x%08x\n", tx_desc->tx_sg[0].addr, - tx_desc->tx_sg[0].length, tx_desc->tx_sg[0].lkey); + isert_dbg("Setup tx_sg[0].addr: 0x%llx length: %u lkey: 0x%x\n", + tx_desc->tx_sg[0].addr, tx_desc->tx_sg[0].length, + tx_desc->tx_sg[0].lkey); return 0; } static void isert_init_send_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, - struct ib_send_wr *send_wr, bool coalesce) + struct ib_send_wr *send_wr) { struct iser_tx_desc *tx_desc = &isert_cmd->tx_desc; isert_cmd->rdma_wr.iser_ib_op = ISER_IB_SEND; - send_wr->wr_id = (unsigned long)&isert_cmd->tx_desc; + send_wr->wr_id = (uintptr_t)&isert_cmd->tx_desc; send_wr->opcode = IB_WR_SEND; send_wr->sg_list = &tx_desc->tx_sg[0]; send_wr->num_sge = isert_cmd->tx_desc.num_sge; - /* - * Coalesce send completion interrupts by only setting IB_SEND_SIGNALED - * bit for every ISERT_COMP_BATCH_COUNT number of ib_post_send() calls. - */ - mutex_lock(&isert_conn->conn_mutex); - if (coalesce && isert_conn->state == ISER_CONN_UP && - ++isert_conn->conn_comp_batch < ISERT_COMP_BATCH_COUNT) { - tx_desc->llnode_active = true; - llist_add(&tx_desc->comp_llnode, &isert_conn->conn_comp_llist); - mutex_unlock(&isert_conn->conn_mutex); - return; - } - isert_conn->conn_comp_batch = 0; - tx_desc->comp_llnode_batch = llist_del_all(&isert_conn->conn_comp_llist); - mutex_unlock(&isert_conn->conn_mutex); - send_wr->send_flags = IB_SEND_SIGNALED; } @@ -1017,22 +1053,21 @@ isert_rdma_post_recvl(struct isert_conn *isert_conn) sge.length = ISER_RX_LOGIN_SIZE; sge.lkey = isert_conn->conn_mr->lkey; - pr_debug("Setup sge: addr: %llx length: %d 0x%08x\n", + isert_dbg("Setup sge: addr: %llx length: %d 0x%08x\n", sge.addr, sge.length, sge.lkey); memset(&rx_wr, 0, sizeof(struct ib_recv_wr)); - rx_wr.wr_id = (unsigned long)isert_conn->login_req_buf; + rx_wr.wr_id = (uintptr_t)isert_conn->login_req_buf; rx_wr.sg_list = &sge; rx_wr.num_sge = 1; isert_conn->post_recv_buf_count++; ret = ib_post_recv(isert_conn->conn_qp, &rx_wr, &rx_wr_fail); if (ret) { - pr_err("ib_post_recv() failed: %d\n", ret); + isert_err("ib_post_recv() failed: %d\n", ret); isert_conn->post_recv_buf_count--; } - pr_debug("ib_post_recv(): returned success >>>>>>>>>>>>>>>>>>>>>>>>\n"); return ret; } @@ -1072,13 +1107,9 @@ isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login, if (login->login_complete) { if (!conn->sess->sess_ops->SessionType && isert_conn->conn_device->use_fastreg) { - /* Normal Session and fastreg is used */ - u8 pi_support = login->np->tpg_np->tpg->tpg_attrib.t10_pi; - - ret = isert_conn_create_fastreg_pool(isert_conn, - pi_support); + ret = isert_conn_create_fastreg_pool(isert_conn); if (ret) { - pr_err("Conn: %p failed to create" + isert_err("Conn: %p failed to create" " fastreg pool\n", isert_conn); return ret; } @@ -1092,7 +1123,10 @@ isert_put_login_tx(struct iscsi_conn *conn, struct iscsi_login *login, if (ret) return ret; - isert_conn->state = ISER_CONN_UP; + /* Now we are in FULL_FEATURE phase */ + mutex_lock(&isert_conn->conn_mutex); + isert_conn->state = ISER_CONN_FULL_FEATURE; + mutex_unlock(&isert_conn->conn_mutex); goto post_send; } @@ -1109,18 +1143,17 @@ post_send: } static void -isert_rx_login_req(struct iser_rx_desc *rx_desc, int rx_buflen, - struct isert_conn *isert_conn) +isert_rx_login_req(struct isert_conn *isert_conn) { + struct iser_rx_desc *rx_desc = (void *)isert_conn->login_req_buf; + int rx_buflen = isert_conn->login_req_len; struct iscsi_conn *conn = isert_conn->conn; struct iscsi_login *login = conn->conn_login; int size; - if (!login) { - pr_err("conn->conn_login is NULL\n"); - dump_stack(); - return; - } + isert_info("conn %p\n", isert_conn); + + WARN_ON_ONCE(!login); if (login->first_request) { struct iscsi_login_req *login_req = @@ -1146,8 +1179,9 @@ isert_rx_login_req(struct iser_rx_desc *rx_desc, int rx_buflen, memcpy(&login->req[0], (void *)&rx_desc->iscsi_header, ISCSI_HDR_LEN); size = min(rx_buflen, MAX_KEY_VALUE_PAIRS); - pr_debug("Using login payload size: %d, rx_buflen: %d MAX_KEY_VALUE_PAIRS: %d\n", - size, rx_buflen, MAX_KEY_VALUE_PAIRS); + isert_dbg("Using login payload size: %d, rx_buflen: %d " + "MAX_KEY_VALUE_PAIRS: %d\n", size, rx_buflen, + MAX_KEY_VALUE_PAIRS); memcpy(login->req_buf, &rx_desc->data[0], size); if (login->first_request) { @@ -1166,7 +1200,7 @@ static struct iscsi_cmd cmd = iscsit_allocate_cmd(conn, TASK_INTERRUPTIBLE); if (!cmd) { - pr_err("Unable to allocate iscsi_cmd + isert_cmd\n"); + isert_err("Unable to allocate iscsi_cmd + isert_cmd\n"); return NULL; } isert_cmd = iscsit_priv_cmd(cmd); @@ -1209,8 +1243,8 @@ isert_handle_scsi_cmd(struct isert_conn *isert_conn, sg = &cmd->se_cmd.t_data_sg[0]; sg_nents = max(1UL, DIV_ROUND_UP(imm_data_len, PAGE_SIZE)); - pr_debug("Copying Immediate SG: %p sg_nents: %u from %p imm_data_len: %d\n", - sg, sg_nents, &rx_desc->data[0], imm_data_len); + isert_dbg("Copying Immediate SG: %p sg_nents: %u from %p imm_data_len: %d\n", + sg, sg_nents, &rx_desc->data[0], imm_data_len); sg_copy_from_buffer(sg, sg_nents, &rx_desc->data[0], imm_data_len); @@ -1254,13 +1288,15 @@ isert_handle_iscsi_dataout(struct isert_conn *isert_conn, * FIXME: Unexpected unsolicited_data out */ if (!cmd->unsolicited_data) { - pr_err("Received unexpected solicited data payload\n"); + isert_err("Received unexpected solicited data payload\n"); dump_stack(); return -1; } - pr_debug("Unsolicited DataOut unsol_data_len: %u, write_data_done: %u, data_length: %u\n", - unsol_data_len, cmd->write_data_done, cmd->se_cmd.data_length); + isert_dbg("Unsolicited DataOut unsol_data_len: %u, " + "write_data_done: %u, data_length: %u\n", + unsol_data_len, cmd->write_data_done, + cmd->se_cmd.data_length); sg_off = cmd->write_data_done / PAGE_SIZE; sg_start = &cmd->se_cmd.t_data_sg[sg_off]; @@ -1270,12 +1306,13 @@ isert_handle_iscsi_dataout(struct isert_conn *isert_conn, * FIXME: Non page-aligned unsolicited_data out */ if (page_off) { - pr_err("Received unexpected non-page aligned data payload\n"); + isert_err("unexpected non-page aligned data payload\n"); dump_stack(); return -1; } - pr_debug("Copying DataOut: sg_start: %p, sg_off: %u sg_nents: %u from %p %u\n", - sg_start, sg_off, sg_nents, &rx_desc->data[0], unsol_data_len); + isert_dbg("Copying DataOut: sg_start: %p, sg_off: %u " + "sg_nents: %u from %p %u\n", sg_start, sg_off, + sg_nents, &rx_desc->data[0], unsol_data_len); sg_copy_from_buffer(sg_start, sg_nents, &rx_desc->data[0], unsol_data_len); @@ -1322,8 +1359,8 @@ isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd text_in = kzalloc(payload_length, GFP_KERNEL); if (!text_in) { - pr_err("Unable to allocate text_in of payload_length: %u\n", - payload_length); + isert_err("Unable to allocate text_in of payload_length: %u\n", + payload_length); return -ENOMEM; } cmd->text_in_ptr = text_in; @@ -1348,8 +1385,8 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, if (sess->sess_ops->SessionType && (!(opcode & ISCSI_OP_TEXT) || !(opcode & ISCSI_OP_LOGOUT))) { - pr_err("Got illegal opcode: 0x%02x in SessionType=Discovery," - " ignoring\n", opcode); + isert_err("Got illegal opcode: 0x%02x in SessionType=Discovery," + " ignoring\n", opcode); return 0; } @@ -1395,10 +1432,6 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, break; ret = iscsit_handle_logout_cmd(conn, cmd, (unsigned char *)hdr); - if (ret > 0) - wait_for_completion_timeout(&conn->conn_logout_comp, - SECONDS_FOR_LOGOUT_COMP * - HZ); break; case ISCSI_OP_TEXT: cmd = isert_allocate_cmd(conn); @@ -1410,7 +1443,7 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, rx_desc, (struct iscsi_text *)hdr); break; default: - pr_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode); + isert_err("Got unknown iSCSI OpCode: 0x%02x\n", opcode); dump_stack(); break; } @@ -1431,23 +1464,23 @@ isert_rx_do_work(struct iser_rx_desc *rx_desc, struct isert_conn *isert_conn) if (iser_hdr->flags & ISER_RSV) { read_stag = be32_to_cpu(iser_hdr->read_stag); read_va = be64_to_cpu(iser_hdr->read_va); - pr_debug("ISER_RSV: read_stag: 0x%08x read_va: 0x%16llx\n", - read_stag, (unsigned long long)read_va); + isert_dbg("ISER_RSV: read_stag: 0x%x read_va: 0x%llx\n", + read_stag, (unsigned long long)read_va); } if (iser_hdr->flags & ISER_WSV) { write_stag = be32_to_cpu(iser_hdr->write_stag); write_va = be64_to_cpu(iser_hdr->write_va); - pr_debug("ISER_WSV: write__stag: 0x%08x write_va: 0x%16llx\n", - write_stag, (unsigned long long)write_va); + isert_dbg("ISER_WSV: write_stag: 0x%x write_va: 0x%llx\n", + write_stag, (unsigned long long)write_va); } - pr_debug("ISER ISCSI_CTRL PDU\n"); + isert_dbg("ISER ISCSI_CTRL PDU\n"); break; case ISER_HELLO: - pr_err("iSER Hello message\n"); + isert_err("iSER Hello message\n"); break; default: - pr_warn("Unknown iSER hdr flags: 0x%02x\n", iser_hdr->flags); + isert_warn("Unknown iSER hdr flags: 0x%02x\n", iser_hdr->flags); break; } @@ -1457,7 +1490,7 @@ isert_rx_do_work(struct iser_rx_desc *rx_desc, struct isert_conn *isert_conn) static void isert_rx_completion(struct iser_rx_desc *desc, struct isert_conn *isert_conn, - unsigned long xfer_len) + u32 xfer_len) { struct ib_device *ib_dev = isert_conn->conn_cm_id->device; struct iscsi_hdr *hdr; @@ -1467,34 +1500,43 @@ isert_rx_completion(struct iser_rx_desc *desc, struct isert_conn *isert_conn, if ((char *)desc == isert_conn->login_req_buf) { rx_dma = isert_conn->login_req_dma; rx_buflen = ISER_RX_LOGIN_SIZE; - pr_debug("ISER login_buf: Using rx_dma: 0x%llx, rx_buflen: %d\n", + isert_dbg("login_buf: Using rx_dma: 0x%llx, rx_buflen: %d\n", rx_dma, rx_buflen); } else { rx_dma = desc->dma_addr; rx_buflen = ISER_RX_PAYLOAD_SIZE; - pr_debug("ISER req_buf: Using rx_dma: 0x%llx, rx_buflen: %d\n", + isert_dbg("req_buf: Using rx_dma: 0x%llx, rx_buflen: %d\n", rx_dma, rx_buflen); } ib_dma_sync_single_for_cpu(ib_dev, rx_dma, rx_buflen, DMA_FROM_DEVICE); hdr = &desc->iscsi_header; - pr_debug("iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n", + isert_dbg("iSCSI opcode: 0x%02x, ITT: 0x%08x, flags: 0x%02x dlen: %d\n", hdr->opcode, hdr->itt, hdr->flags, (int)(xfer_len - ISER_HEADERS_LEN)); - if ((char *)desc == isert_conn->login_req_buf) - isert_rx_login_req(desc, xfer_len - ISER_HEADERS_LEN, - isert_conn); - else + if ((char *)desc == isert_conn->login_req_buf) { + isert_conn->login_req_len = xfer_len - ISER_HEADERS_LEN; + if (isert_conn->conn) { + struct iscsi_login *login = isert_conn->conn->conn_login; + + if (login && !login->first_request) + isert_rx_login_req(isert_conn); + } + mutex_lock(&isert_conn->conn_mutex); + complete(&isert_conn->login_req_comp); + mutex_unlock(&isert_conn->conn_mutex); + } else { isert_rx_do_work(desc, isert_conn); + } ib_dma_sync_single_for_device(ib_dev, rx_dma, rx_buflen, DMA_FROM_DEVICE); isert_conn->post_recv_buf_count--; - pr_debug("iSERT: Decremented post_recv_buf_count: %d\n", - isert_conn->post_recv_buf_count); + isert_dbg("Decremented post_recv_buf_count: %d\n", + isert_conn->post_recv_buf_count); if ((char *)desc == isert_conn->login_req_buf) return; @@ -1505,7 +1547,7 @@ isert_rx_completion(struct iser_rx_desc *desc, struct isert_conn *isert_conn, ISERT_MIN_POSTED_RX); err = isert_post_recv(isert_conn, count); if (err) { - pr_err("isert_post_recv() count: %d failed, %d\n", + isert_err("isert_post_recv() count: %d failed, %d\n", count, err); } } @@ -1534,12 +1576,12 @@ isert_map_data_buf(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, data->dma_nents = ib_dma_map_sg(ib_dev, data->sg, data->nents, data->dma_dir); if (unlikely(!data->dma_nents)) { - pr_err("Cmd: unable to dma map SGs %p\n", sg); + isert_err("Cmd: unable to dma map SGs %p\n", sg); return -EINVAL; } - pr_debug("Mapped cmd: %p count: %u sg: %p sg_nents: %u rdma_len %d\n", - isert_cmd, data->dma_nents, data->sg, data->nents, data->len); + isert_dbg("Mapped cmd: %p count: %u sg: %p sg_nents: %u rdma_len %d\n", + isert_cmd, data->dma_nents, data->sg, data->nents, data->len); return 0; } @@ -1560,21 +1602,21 @@ isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn) { struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; - pr_debug("isert_unmap_cmd: %p\n", isert_cmd); + isert_dbg("Cmd %p\n", isert_cmd); if (wr->data.sg) { - pr_debug("isert_unmap_cmd: %p unmap_sg op\n", isert_cmd); + isert_dbg("Cmd %p unmap_sg op\n", isert_cmd); isert_unmap_data_buf(isert_conn, &wr->data); } if (wr->send_wr) { - pr_debug("isert_unmap_cmd: %p free send_wr\n", isert_cmd); + isert_dbg("Cmd %p free send_wr\n", isert_cmd); kfree(wr->send_wr); wr->send_wr = NULL; } if (wr->ib_sge) { - pr_debug("isert_unmap_cmd: %p free ib_sge\n", isert_cmd); + isert_dbg("Cmd %p free ib_sge\n", isert_cmd); kfree(wr->ib_sge); wr->ib_sge = NULL; } @@ -1586,11 +1628,10 @@ isert_unreg_rdma(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn) struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; LIST_HEAD(unmap_list); - pr_debug("unreg_fastreg_cmd: %p\n", isert_cmd); + isert_dbg("Cmd %p\n", isert_cmd); if (wr->fr_desc) { - pr_debug("unreg_fastreg_cmd: %p free fr_desc %p\n", - isert_cmd, wr->fr_desc); + isert_dbg("Cmd %p free fr_desc %p\n", isert_cmd, wr->fr_desc); if (wr->fr_desc->ind & ISERT_PROTECTED) { isert_unmap_data_buf(isert_conn, &wr->prot); wr->fr_desc->ind &= ~ISERT_PROTECTED; @@ -1602,7 +1643,7 @@ isert_unreg_rdma(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn) } if (wr->data.sg) { - pr_debug("unreg_fastreg_cmd: %p unmap_sg op\n", isert_cmd); + isert_dbg("Cmd %p unmap_sg op\n", isert_cmd); isert_unmap_data_buf(isert_conn, &wr->data); } @@ -1618,7 +1659,7 @@ isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err) struct iscsi_conn *conn = isert_conn->conn; struct isert_device *device = isert_conn->conn_device; - pr_debug("Entering isert_put_cmd: %p\n", isert_cmd); + isert_dbg("Cmd %p\n", isert_cmd); switch (cmd->iscsi_opcode) { case ISCSI_OP_SCSI_CMD: @@ -1668,7 +1709,7 @@ isert_put_cmd(struct isert_cmd *isert_cmd, bool comp_err) * associated cmd->se_cmd needs to be released. */ if (cmd->se_cmd.se_tfo != NULL) { - pr_debug("Calling transport_generic_free_cmd from" + isert_dbg("Calling transport_generic_free_cmd from" " isert_put_cmd for 0x%02x\n", cmd->iscsi_opcode); transport_generic_free_cmd(&cmd->se_cmd, 0); @@ -1687,7 +1728,7 @@ static void isert_unmap_tx_desc(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev) { if (tx_desc->dma_addr != 0) { - pr_debug("Calling ib_dma_unmap_single for tx_desc->dma_addr\n"); + isert_dbg("unmap single for tx_desc->dma_addr\n"); ib_dma_unmap_single(ib_dev, tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE); tx_desc->dma_addr = 0; @@ -1699,7 +1740,7 @@ isert_completion_put(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd, struct ib_device *ib_dev, bool comp_err) { if (isert_cmd->pdu_buf_dma != 0) { - pr_debug("Calling ib_dma_unmap_single for isert_cmd->pdu_buf_dma\n"); + isert_dbg("unmap single for isert_cmd->pdu_buf_dma\n"); ib_dma_unmap_single(ib_dev, isert_cmd->pdu_buf_dma, isert_cmd->pdu_buf_len, DMA_TO_DEVICE); isert_cmd->pdu_buf_dma = 0; @@ -1717,7 +1758,7 @@ isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr) ret = ib_check_mr_status(sig_mr, IB_MR_CHECK_SIG_STATUS, &mr_status); if (ret) { - pr_err("ib_check_mr_status failed, ret %d\n", ret); + isert_err("ib_check_mr_status failed, ret %d\n", ret); goto fail_mr_status; } @@ -1740,12 +1781,12 @@ isert_check_pi_status(struct se_cmd *se_cmd, struct ib_mr *sig_mr) do_div(sec_offset_err, block_size); se_cmd->bad_sector = sec_offset_err + se_cmd->t_task_lba; - pr_err("isert: PI error found type %d at sector 0x%llx " - "expected 0x%x vs actual 0x%x\n", - mr_status.sig_err.err_type, - (unsigned long long)se_cmd->bad_sector, - mr_status.sig_err.expected, - mr_status.sig_err.actual); + isert_err("PI error found type %d at sector 0x%llx " + "expected 0x%x vs actual 0x%x\n", + mr_status.sig_err.err_type, + (unsigned long long)se_cmd->bad_sector, + mr_status.sig_err.expected, + mr_status.sig_err.actual); ret = 1; } @@ -1801,7 +1842,7 @@ isert_completion_rdma_read(struct iser_tx_desc *tx_desc, cmd->write_data_done = wr->data.len; wr->send_wr_num = 0; - pr_debug("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd); + isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd); spin_lock_bh(&cmd->istate_lock); cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT; cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT; @@ -1823,36 +1864,22 @@ isert_do_control_comp(struct work_struct *work) struct ib_device *ib_dev = isert_conn->conn_cm_id->device; struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; + isert_dbg("Cmd %p i_state %d\n", isert_cmd, cmd->i_state); + switch (cmd->i_state) { case ISTATE_SEND_TASKMGTRSP: - pr_debug("Calling iscsit_tmr_post_handler >>>>>>>>>>>>>>>>>\n"); - - atomic_dec(&isert_conn->post_send_buf_count); iscsit_tmr_post_handler(cmd, cmd->conn); - - cmd->i_state = ISTATE_SENT_STATUS; - isert_completion_put(&isert_cmd->tx_desc, isert_cmd, ib_dev, false); - break; - case ISTATE_SEND_REJECT: - pr_debug("Got isert_do_control_comp ISTATE_SEND_REJECT: >>>\n"); - atomic_dec(&isert_conn->post_send_buf_count); - + case ISTATE_SEND_REJECT: /* FALLTHRU */ + case ISTATE_SEND_TEXTRSP: /* FALLTHRU */ cmd->i_state = ISTATE_SENT_STATUS; - isert_completion_put(&isert_cmd->tx_desc, isert_cmd, ib_dev, false); + isert_completion_put(&isert_cmd->tx_desc, isert_cmd, + ib_dev, false); break; case ISTATE_SEND_LOGOUTRSP: - pr_debug("Calling iscsit_logout_post_handler >>>>>>>>>>>>>>\n"); - - atomic_dec(&isert_conn->post_send_buf_count); iscsit_logout_post_handler(cmd, cmd->conn); break; - case ISTATE_SEND_TEXTRSP: - atomic_dec(&isert_conn->post_send_buf_count); - cmd->i_state = ISTATE_SENT_STATUS; - isert_completion_put(&isert_cmd->tx_desc, isert_cmd, ib_dev, false); - break; default: - pr_err("Unknown do_control_comp i_state %d\n", cmd->i_state); + isert_err("Unknown i_state %d\n", cmd->i_state); dump_stack(); break; } @@ -1865,7 +1892,6 @@ isert_response_completion(struct iser_tx_desc *tx_desc, struct ib_device *ib_dev) { struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; - struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; if (cmd->i_state == ISTATE_SEND_TASKMGTRSP || cmd->i_state == ISTATE_SEND_LOGOUTRSP || @@ -1878,267 +1904,151 @@ isert_response_completion(struct iser_tx_desc *tx_desc, return; } - /** - * If send_wr_num is 0 this means that we got - * RDMA completion and we cleared it and we should - * simply decrement the response post. else the - * response is incorporated in send_wr_num, just - * sub it. - **/ - if (wr->send_wr_num) - atomic_sub(wr->send_wr_num, &isert_conn->post_send_buf_count); - else - atomic_dec(&isert_conn->post_send_buf_count); - cmd->i_state = ISTATE_SENT_STATUS; isert_completion_put(tx_desc, isert_cmd, ib_dev, false); } static void -__isert_send_completion(struct iser_tx_desc *tx_desc, - struct isert_conn *isert_conn) +isert_send_completion(struct iser_tx_desc *tx_desc, + struct isert_conn *isert_conn) { struct ib_device *ib_dev = isert_conn->conn_cm_id->device; struct isert_cmd *isert_cmd = tx_desc->isert_cmd; struct isert_rdma_wr *wr; if (!isert_cmd) { - atomic_dec(&isert_conn->post_send_buf_count); isert_unmap_tx_desc(tx_desc, ib_dev); return; } wr = &isert_cmd->rdma_wr; + isert_dbg("Cmd %p iser_ib_op %d\n", isert_cmd, wr->iser_ib_op); + switch (wr->iser_ib_op) { case ISER_IB_RECV: - pr_err("isert_send_completion: Got ISER_IB_RECV\n"); + isert_err("Got ISER_IB_RECV\n"); dump_stack(); break; case ISER_IB_SEND: - pr_debug("isert_send_completion: Got ISER_IB_SEND\n"); isert_response_completion(tx_desc, isert_cmd, isert_conn, ib_dev); break; case ISER_IB_RDMA_WRITE: - pr_debug("isert_send_completion: Got ISER_IB_RDMA_WRITE\n"); - atomic_sub(wr->send_wr_num, &isert_conn->post_send_buf_count); isert_completion_rdma_write(tx_desc, isert_cmd); break; case ISER_IB_RDMA_READ: - pr_debug("isert_send_completion: Got ISER_IB_RDMA_READ:\n"); - - atomic_sub(wr->send_wr_num, &isert_conn->post_send_buf_count); isert_completion_rdma_read(tx_desc, isert_cmd); break; default: - pr_err("Unknown wr->iser_ib_op: 0x%02x\n", wr->iser_ib_op); + isert_err("Unknown wr->iser_ib_op: 0x%x\n", wr->iser_ib_op); dump_stack(); break; } } -static void -isert_send_completion(struct iser_tx_desc *tx_desc, - struct isert_conn *isert_conn) -{ - struct llist_node *llnode = tx_desc->comp_llnode_batch; - struct iser_tx_desc *t; - /* - * Drain coalesced completion llist starting from comp_llnode_batch - * setup in isert_init_send_wr(), and then complete trailing tx_desc. - */ - while (llnode) { - t = llist_entry(llnode, struct iser_tx_desc, comp_llnode); - llnode = llist_next(llnode); - __isert_send_completion(t, isert_conn); - } - __isert_send_completion(tx_desc, isert_conn); -} - -static void -isert_cq_drain_comp_llist(struct isert_conn *isert_conn, struct ib_device *ib_dev) +/** + * is_isert_tx_desc() - Indicate if the completion wr_id + * is a TX descriptor or not. + * @isert_conn: iser connection + * @wr_id: completion WR identifier + * + * Since we cannot rely on wc opcode in FLUSH errors + * we must work around it by checking if the wr_id address + * falls in the iser connection rx_descs buffer. If so + * it is an RX descriptor, otherwize it is a TX. + */ +static inline bool +is_isert_tx_desc(struct isert_conn *isert_conn, void *wr_id) { - struct llist_node *llnode; - struct isert_rdma_wr *wr; - struct iser_tx_desc *t; + void *start = isert_conn->conn_rx_descs; + int len = ISERT_QP_MAX_RECV_DTOS * sizeof(*isert_conn->conn_rx_descs); - mutex_lock(&isert_conn->conn_mutex); - llnode = llist_del_all(&isert_conn->conn_comp_llist); - isert_conn->conn_comp_batch = 0; - mutex_unlock(&isert_conn->conn_mutex); - - while (llnode) { - t = llist_entry(llnode, struct iser_tx_desc, comp_llnode); - llnode = llist_next(llnode); - wr = &t->isert_cmd->rdma_wr; - - /** - * If send_wr_num is 0 this means that we got - * RDMA completion and we cleared it and we should - * simply decrement the response post. else the - * response is incorporated in send_wr_num, just - * sub it. - **/ - if (wr->send_wr_num) - atomic_sub(wr->send_wr_num, - &isert_conn->post_send_buf_count); - else - atomic_dec(&isert_conn->post_send_buf_count); + if (wr_id >= start && wr_id < start + len) + return false; - isert_completion_put(t, t->isert_cmd, ib_dev, true); - } + return true; } static void -isert_cq_tx_comp_err(struct iser_tx_desc *tx_desc, struct isert_conn *isert_conn) +isert_cq_comp_err(struct isert_conn *isert_conn, struct ib_wc *wc) { - struct ib_device *ib_dev = isert_conn->conn_cm_id->device; - struct isert_cmd *isert_cmd = tx_desc->isert_cmd; - struct llist_node *llnode = tx_desc->comp_llnode_batch; - struct isert_rdma_wr *wr; - struct iser_tx_desc *t; - - while (llnode) { - t = llist_entry(llnode, struct iser_tx_desc, comp_llnode); - llnode = llist_next(llnode); - wr = &t->isert_cmd->rdma_wr; + if (wc->wr_id == ISER_BEACON_WRID) { + isert_info("conn %p completing conn_wait_comp_err\n", + isert_conn); + complete(&isert_conn->conn_wait_comp_err); + } else if (is_isert_tx_desc(isert_conn, (void *)(uintptr_t)wc->wr_id)) { + struct ib_device *ib_dev = isert_conn->conn_cm_id->device; + struct isert_cmd *isert_cmd; + struct iser_tx_desc *desc; - /** - * If send_wr_num is 0 this means that we got - * RDMA completion and we cleared it and we should - * simply decrement the response post. else the - * response is incorporated in send_wr_num, just - * sub it. - **/ - if (wr->send_wr_num) - atomic_sub(wr->send_wr_num, - &isert_conn->post_send_buf_count); + desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id; + isert_cmd = desc->isert_cmd; + if (!isert_cmd) + isert_unmap_tx_desc(desc, ib_dev); else - atomic_dec(&isert_conn->post_send_buf_count); - - isert_completion_put(t, t->isert_cmd, ib_dev, true); - } - tx_desc->comp_llnode_batch = NULL; - - if (!isert_cmd) - isert_unmap_tx_desc(tx_desc, ib_dev); - else - isert_completion_put(tx_desc, isert_cmd, ib_dev, true); -} - -static void -isert_cq_rx_comp_err(struct isert_conn *isert_conn) -{ - struct ib_device *ib_dev = isert_conn->conn_cm_id->device; - struct iscsi_conn *conn = isert_conn->conn; - - if (isert_conn->post_recv_buf_count) - return; - - isert_cq_drain_comp_llist(isert_conn, ib_dev); - - if (conn->sess) { - target_sess_cmd_list_set_waiting(conn->sess->se_sess); - target_wait_for_sess_cmds(conn->sess->se_sess); + isert_completion_put(desc, isert_cmd, ib_dev, true); + } else { + isert_conn->post_recv_buf_count--; + if (!isert_conn->post_recv_buf_count) + iscsit_cause_connection_reinstatement(isert_conn->conn, 0); } - - while (atomic_read(&isert_conn->post_send_buf_count)) - msleep(3000); - - mutex_lock(&isert_conn->conn_mutex); - isert_conn->state = ISER_CONN_DOWN; - mutex_unlock(&isert_conn->conn_mutex); - - iscsit_cause_connection_reinstatement(isert_conn->conn, 0); - - complete(&isert_conn->conn_wait_comp_err); } static void -isert_cq_tx_work(struct work_struct *work) +isert_handle_wc(struct ib_wc *wc) { - struct isert_cq_desc *cq_desc = container_of(work, - struct isert_cq_desc, cq_tx_work); - struct isert_device *device = cq_desc->device; - int cq_index = cq_desc->cq_index; - struct ib_cq *tx_cq = device->dev_tx_cq[cq_index]; struct isert_conn *isert_conn; struct iser_tx_desc *tx_desc; - struct ib_wc wc; - - while (ib_poll_cq(tx_cq, 1, &wc) == 1) { - tx_desc = (struct iser_tx_desc *)(unsigned long)wc.wr_id; - isert_conn = wc.qp->qp_context; + struct iser_rx_desc *rx_desc; - if (wc.status == IB_WC_SUCCESS) { - isert_send_completion(tx_desc, isert_conn); + isert_conn = wc->qp->qp_context; + if (likely(wc->status == IB_WC_SUCCESS)) { + if (wc->opcode == IB_WC_RECV) { + rx_desc = (struct iser_rx_desc *)(uintptr_t)wc->wr_id; + isert_rx_completion(rx_desc, isert_conn, wc->byte_len); } else { - pr_debug("TX wc.status != IB_WC_SUCCESS >>>>>>>>>>>>>>\n"); - pr_debug("TX wc.status: 0x%08x\n", wc.status); - pr_debug("TX wc.vendor_err: 0x%08x\n", wc.vendor_err); - - if (wc.wr_id != ISER_FASTREG_LI_WRID) { - if (tx_desc->llnode_active) - continue; - - atomic_dec(&isert_conn->post_send_buf_count); - isert_cq_tx_comp_err(tx_desc, isert_conn); - } + tx_desc = (struct iser_tx_desc *)(uintptr_t)wc->wr_id; + isert_send_completion(tx_desc, isert_conn); } - } - - ib_req_notify_cq(tx_cq, IB_CQ_NEXT_COMP); -} - -static void -isert_cq_tx_callback(struct ib_cq *cq, void *context) -{ - struct isert_cq_desc *cq_desc = (struct isert_cq_desc *)context; + } else { + if (wc->status != IB_WC_WR_FLUSH_ERR) + isert_err("wr id %llx status %d vend_err %x\n", + wc->wr_id, wc->status, wc->vendor_err); + else + isert_dbg("flush error: wr id %llx\n", wc->wr_id); - queue_work(isert_comp_wq, &cq_desc->cq_tx_work); + if (wc->wr_id != ISER_FASTREG_LI_WRID) + isert_cq_comp_err(isert_conn, wc); + } } static void -isert_cq_rx_work(struct work_struct *work) +isert_cq_work(struct work_struct *work) { - struct isert_cq_desc *cq_desc = container_of(work, - struct isert_cq_desc, cq_rx_work); - struct isert_device *device = cq_desc->device; - int cq_index = cq_desc->cq_index; - struct ib_cq *rx_cq = device->dev_rx_cq[cq_index]; - struct isert_conn *isert_conn; - struct iser_rx_desc *rx_desc; - struct ib_wc wc; - unsigned long xfer_len; + enum { isert_poll_budget = 65536 }; + struct isert_comp *comp = container_of(work, struct isert_comp, + work); + struct ib_wc *const wcs = comp->wcs; + int i, n, completed = 0; - while (ib_poll_cq(rx_cq, 1, &wc) == 1) { - rx_desc = (struct iser_rx_desc *)(unsigned long)wc.wr_id; - isert_conn = wc.qp->qp_context; + while ((n = ib_poll_cq(comp->cq, ARRAY_SIZE(comp->wcs), wcs)) > 0) { + for (i = 0; i < n; i++) + isert_handle_wc(&wcs[i]); - if (wc.status == IB_WC_SUCCESS) { - xfer_len = (unsigned long)wc.byte_len; - isert_rx_completion(rx_desc, isert_conn, xfer_len); - } else { - pr_debug("RX wc.status != IB_WC_SUCCESS >>>>>>>>>>>>>>\n"); - if (wc.status != IB_WC_WR_FLUSH_ERR) { - pr_debug("RX wc.status: 0x%08x\n", wc.status); - pr_debug("RX wc.vendor_err: 0x%08x\n", - wc.vendor_err); - } - isert_conn->post_recv_buf_count--; - isert_cq_rx_comp_err(isert_conn); - } + completed += n; + if (completed >= isert_poll_budget) + break; } - ib_req_notify_cq(rx_cq, IB_CQ_NEXT_COMP); + ib_req_notify_cq(comp->cq, IB_CQ_NEXT_COMP); } static void -isert_cq_rx_callback(struct ib_cq *cq, void *context) +isert_cq_callback(struct ib_cq *cq, void *context) { - struct isert_cq_desc *cq_desc = (struct isert_cq_desc *)context; + struct isert_comp *comp = context; - queue_work(isert_rx_wq, &cq_desc->cq_rx_work); + queue_work(isert_comp_wq, &comp->work); } static int @@ -2147,13 +2057,10 @@ isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd) struct ib_send_wr *wr_failed; int ret; - atomic_inc(&isert_conn->post_send_buf_count); - ret = ib_post_send(isert_conn->conn_qp, &isert_cmd->tx_desc.send_wr, &wr_failed); if (ret) { - pr_err("ib_post_send failed with %d\n", ret); - atomic_dec(&isert_conn->post_send_buf_count); + isert_err("ib_post_send failed with %d\n", ret); return ret; } return ret; @@ -2200,9 +2107,9 @@ isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd) isert_cmd->tx_desc.num_sge = 2; } - isert_init_send_wr(isert_conn, isert_cmd, send_wr, false); + isert_init_send_wr(isert_conn, isert_cmd, send_wr); - pr_debug("Posting SCSI Response IB_WR_SEND >>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("Posting SCSI Response\n"); return isert_post_response(isert_conn, isert_cmd); } @@ -2231,8 +2138,16 @@ isert_get_sup_prot_ops(struct iscsi_conn *conn) struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct isert_device *device = isert_conn->conn_device; - if (device->pi_capable) - return TARGET_PROT_ALL; + if (conn->tpg->tpg_attrib.t10_pi) { + if (device->pi_capable) { + isert_info("conn %p PI offload enabled\n", isert_conn); + isert_conn->pi_support = true; + return TARGET_PROT_ALL; + } + } + + isert_info("conn %p PI offload disabled\n", isert_conn); + isert_conn->pi_support = false; return TARGET_PROT_NORMAL; } @@ -2250,9 +2165,9 @@ isert_put_nopin(struct iscsi_cmd *cmd, struct iscsi_conn *conn, &isert_cmd->tx_desc.iscsi_header, nopout_response); isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc); - isert_init_send_wr(isert_conn, isert_cmd, send_wr, false); + isert_init_send_wr(isert_conn, isert_cmd, send_wr); - pr_debug("Posting NOPIN Response IB_WR_SEND >>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("conn %p Posting NOPIN Response\n", isert_conn); return isert_post_response(isert_conn, isert_cmd); } @@ -2268,9 +2183,9 @@ isert_put_logout_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) iscsit_build_logout_rsp(cmd, conn, (struct iscsi_logout_rsp *) &isert_cmd->tx_desc.iscsi_header); isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc); - isert_init_send_wr(isert_conn, isert_cmd, send_wr, false); + isert_init_send_wr(isert_conn, isert_cmd, send_wr); - pr_debug("Posting Logout Response IB_WR_SEND >>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("conn %p Posting Logout Response\n", isert_conn); return isert_post_response(isert_conn, isert_cmd); } @@ -2286,9 +2201,9 @@ isert_put_tm_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) iscsit_build_task_mgt_rsp(cmd, conn, (struct iscsi_tm_rsp *) &isert_cmd->tx_desc.iscsi_header); isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc); - isert_init_send_wr(isert_conn, isert_cmd, send_wr, false); + isert_init_send_wr(isert_conn, isert_cmd, send_wr); - pr_debug("Posting Task Management Response IB_WR_SEND >>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("conn %p Posting Task Management Response\n", isert_conn); return isert_post_response(isert_conn, isert_cmd); } @@ -2318,9 +2233,9 @@ isert_put_reject(struct iscsi_cmd *cmd, struct iscsi_conn *conn) tx_dsg->lkey = isert_conn->conn_mr->lkey; isert_cmd->tx_desc.num_sge = 2; - isert_init_send_wr(isert_conn, isert_cmd, send_wr, false); + isert_init_send_wr(isert_conn, isert_cmd, send_wr); - pr_debug("Posting Reject IB_WR_SEND >>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("conn %p Posting Reject\n", isert_conn); return isert_post_response(isert_conn, isert_cmd); } @@ -2358,9 +2273,9 @@ isert_put_text_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) tx_dsg->lkey = isert_conn->conn_mr->lkey; isert_cmd->tx_desc.num_sge = 2; } - isert_init_send_wr(isert_conn, isert_cmd, send_wr, false); + isert_init_send_wr(isert_conn, isert_cmd, send_wr); - pr_debug("Posting Text Response IB_WR_SEND >>>>>>>>>>>>>>>>>>>>>>\n"); + isert_dbg("conn %p Text Reject\n", isert_conn); return isert_post_response(isert_conn, isert_cmd); } @@ -2383,30 +2298,31 @@ isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, send_wr->sg_list = ib_sge; send_wr->num_sge = sg_nents; - send_wr->wr_id = (unsigned long)&isert_cmd->tx_desc; + send_wr->wr_id = (uintptr_t)&isert_cmd->tx_desc; /* * Perform mapping of TCM scatterlist memory ib_sge dma_addr. */ for_each_sg(sg_start, tmp_sg, sg_nents, i) { - pr_debug("ISER RDMA from SGL dma_addr: 0x%16llx dma_len: %u, page_off: %u\n", - (unsigned long long)tmp_sg->dma_address, - tmp_sg->length, page_off); + isert_dbg("RDMA from SGL dma_addr: 0x%llx dma_len: %u, " + "page_off: %u\n", + (unsigned long long)tmp_sg->dma_address, + tmp_sg->length, page_off); ib_sge->addr = ib_sg_dma_address(ib_dev, tmp_sg) + page_off; ib_sge->length = min_t(u32, data_left, ib_sg_dma_len(ib_dev, tmp_sg) - page_off); ib_sge->lkey = isert_conn->conn_mr->lkey; - pr_debug("RDMA ib_sge: addr: 0x%16llx length: %u lkey: %08x\n", - ib_sge->addr, ib_sge->length, ib_sge->lkey); + isert_dbg("RDMA ib_sge: addr: 0x%llx length: %u lkey: %x\n", + ib_sge->addr, ib_sge->length, ib_sge->lkey); page_off = 0; data_left -= ib_sge->length; ib_sge++; - pr_debug("Incrementing ib_sge pointer to %p\n", ib_sge); + isert_dbg("Incrementing ib_sge pointer to %p\n", ib_sge); } - pr_debug("Set outgoing sg_list: %p num_sg: %u from TCM SGLs\n", - send_wr->sg_list, send_wr->num_sge); + isert_dbg("Set outgoing sg_list: %p num_sg: %u from TCM SGLs\n", + send_wr->sg_list, send_wr->num_sge); return sg_nents; } @@ -2438,7 +2354,7 @@ isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, ib_sge = kzalloc(sizeof(struct ib_sge) * data->nents, GFP_KERNEL); if (!ib_sge) { - pr_warn("Unable to allocate ib_sge\n"); + isert_warn("Unable to allocate ib_sge\n"); ret = -ENOMEM; goto unmap_cmd; } @@ -2448,7 +2364,7 @@ isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, wr->send_wr = kzalloc(sizeof(struct ib_send_wr) * wr->send_wr_num, GFP_KERNEL); if (!wr->send_wr) { - pr_debug("Unable to allocate wr->send_wr\n"); + isert_dbg("Unable to allocate wr->send_wr\n"); ret = -ENOMEM; goto unmap_cmd; } @@ -2512,9 +2428,9 @@ isert_map_fr_pagelist(struct ib_device *ib_dev, chunk_start = start_addr; end_addr = start_addr + ib_sg_dma_len(ib_dev, tmp_sg); - pr_debug("SGL[%d] dma_addr: 0x%16llx len: %u\n", - i, (unsigned long long)tmp_sg->dma_address, - tmp_sg->length); + isert_dbg("SGL[%d] dma_addr: 0x%llx len: %u\n", + i, (unsigned long long)tmp_sg->dma_address, + tmp_sg->length); if ((end_addr & ~PAGE_MASK) && i < last_ent) { new_chunk = 0; @@ -2525,8 +2441,8 @@ isert_map_fr_pagelist(struct ib_device *ib_dev, page = chunk_start & PAGE_MASK; do { fr_pl[n_pages++] = page; - pr_debug("Mapped page_list[%d] page_addr: 0x%16llx\n", - n_pages - 1, page); + isert_dbg("Mapped page_list[%d] page_addr: 0x%llx\n", + n_pages - 1, page); page += PAGE_SIZE; } while (page < end_addr); } @@ -2534,6 +2450,21 @@ isert_map_fr_pagelist(struct ib_device *ib_dev, return n_pages; } +static inline void +isert_inv_rkey(struct ib_send_wr *inv_wr, struct ib_mr *mr) +{ + u32 rkey; + + memset(inv_wr, 0, sizeof(*inv_wr)); + inv_wr->wr_id = ISER_FASTREG_LI_WRID; + inv_wr->opcode = IB_WR_LOCAL_INV; + inv_wr->ex.invalidate_rkey = mr->rkey; + + /* Bump the key */ + rkey = ib_inc_rkey(mr->rkey); + ib_update_fast_reg_key(mr, rkey); +} + static int isert_fast_reg_mr(struct isert_conn *isert_conn, struct fast_reg_descriptor *fr_desc, @@ -2548,15 +2479,13 @@ isert_fast_reg_mr(struct isert_conn *isert_conn, struct ib_send_wr *bad_wr, *wr = NULL; int ret, pagelist_len; u32 page_off; - u8 key; if (mem->dma_nents == 1) { sge->lkey = isert_conn->conn_mr->lkey; sge->addr = ib_sg_dma_address(ib_dev, &mem->sg[0]); sge->length = ib_sg_dma_len(ib_dev, &mem->sg[0]); - pr_debug("%s:%d sge: addr: 0x%llx length: %u lkey: %x\n", - __func__, __LINE__, sge->addr, sge->length, - sge->lkey); + isert_dbg("sge: addr: 0x%llx length: %u lkey: %x\n", + sge->addr, sge->length, sge->lkey); return 0; } @@ -2572,21 +2501,15 @@ isert_fast_reg_mr(struct isert_conn *isert_conn, page_off = mem->offset % PAGE_SIZE; - pr_debug("Use fr_desc %p sg_nents %d offset %u\n", - fr_desc, mem->nents, mem->offset); + isert_dbg("Use fr_desc %p sg_nents %d offset %u\n", + fr_desc, mem->nents, mem->offset); pagelist_len = isert_map_fr_pagelist(ib_dev, mem->sg, mem->nents, &frpl->page_list[0]); - if (!(fr_desc->ind & ISERT_DATA_KEY_VALID)) { - memset(&inv_wr, 0, sizeof(inv_wr)); - inv_wr.wr_id = ISER_FASTREG_LI_WRID; - inv_wr.opcode = IB_WR_LOCAL_INV; - inv_wr.ex.invalidate_rkey = mr->rkey; + if (!(fr_desc->ind & ind)) { + isert_inv_rkey(&inv_wr, mr); wr = &inv_wr; - /* Bump the key */ - key = (u8)(mr->rkey & 0x000000FF); - ib_update_fast_reg_key(mr, ++key); } /* Prepare FASTREG WR */ @@ -2608,7 +2531,7 @@ isert_fast_reg_mr(struct isert_conn *isert_conn, ret = ib_post_send(isert_conn->conn_qp, wr, &bad_wr); if (ret) { - pr_err("fast registration failed, ret:%d\n", ret); + isert_err("fast registration failed, ret:%d\n", ret); return ret; } fr_desc->ind &= ~ind; @@ -2617,9 +2540,8 @@ isert_fast_reg_mr(struct isert_conn *isert_conn, sge->addr = frpl->page_list[0] + page_off; sge->length = mem->len; - pr_debug("%s:%d sge: addr: 0x%llx length: %u lkey: %x\n", - __func__, __LINE__, sge->addr, sge->length, - sge->lkey); + isert_dbg("sge: addr: 0x%llx length: %u lkey: %x\n", + sge->addr, sge->length, sge->lkey); return ret; } @@ -2665,7 +2587,7 @@ isert_set_sig_attrs(struct se_cmd *se_cmd, struct ib_sig_attrs *sig_attrs) isert_set_dif_domain(se_cmd, sig_attrs, &sig_attrs->mem); break; default: - pr_err("Unsupported PI operation %d\n", se_cmd->prot_op); + isert_err("Unsupported PI operation %d\n", se_cmd->prot_op); return -EINVAL; } @@ -2681,17 +2603,16 @@ isert_set_prot_checks(u8 prot_checks) } static int -isert_reg_sig_mr(struct isert_conn *isert_conn, struct se_cmd *se_cmd, - struct fast_reg_descriptor *fr_desc, - struct ib_sge *data_sge, struct ib_sge *prot_sge, - struct ib_sge *sig_sge) +isert_reg_sig_mr(struct isert_conn *isert_conn, + struct se_cmd *se_cmd, + struct isert_rdma_wr *rdma_wr, + struct fast_reg_descriptor *fr_desc) { struct ib_send_wr sig_wr, inv_wr; struct ib_send_wr *bad_wr, *wr = NULL; struct pi_context *pi_ctx = fr_desc->pi_ctx; struct ib_sig_attrs sig_attrs; int ret; - u32 key; memset(&sig_attrs, 0, sizeof(sig_attrs)); ret = isert_set_sig_attrs(se_cmd, &sig_attrs); @@ -2701,26 +2622,20 @@ isert_reg_sig_mr(struct isert_conn *isert_conn, struct se_cmd *se_cmd, sig_attrs.check_mask = isert_set_prot_checks(se_cmd->prot_checks); if (!(fr_desc->ind & ISERT_SIG_KEY_VALID)) { - memset(&inv_wr, 0, sizeof(inv_wr)); - inv_wr.opcode = IB_WR_LOCAL_INV; - inv_wr.wr_id = ISER_FASTREG_LI_WRID; - inv_wr.ex.invalidate_rkey = pi_ctx->sig_mr->rkey; + isert_inv_rkey(&inv_wr, pi_ctx->sig_mr); wr = &inv_wr; - /* Bump the key */ - key = (u8)(pi_ctx->sig_mr->rkey & 0x000000FF); - ib_update_fast_reg_key(pi_ctx->sig_mr, ++key); } memset(&sig_wr, 0, sizeof(sig_wr)); sig_wr.opcode = IB_WR_REG_SIG_MR; sig_wr.wr_id = ISER_FASTREG_LI_WRID; - sig_wr.sg_list = data_sge; + sig_wr.sg_list = &rdma_wr->ib_sg[DATA]; sig_wr.num_sge = 1; sig_wr.wr.sig_handover.access_flags = IB_ACCESS_LOCAL_WRITE; sig_wr.wr.sig_handover.sig_attrs = &sig_attrs; sig_wr.wr.sig_handover.sig_mr = pi_ctx->sig_mr; if (se_cmd->t_prot_sg) - sig_wr.wr.sig_handover.prot = prot_sge; + sig_wr.wr.sig_handover.prot = &rdma_wr->ib_sg[PROT]; if (!wr) wr = &sig_wr; @@ -2729,39 +2644,98 @@ isert_reg_sig_mr(struct isert_conn *isert_conn, struct se_cmd *se_cmd, ret = ib_post_send(isert_conn->conn_qp, wr, &bad_wr); if (ret) { - pr_err("fast registration failed, ret:%d\n", ret); + isert_err("fast registration failed, ret:%d\n", ret); goto err; } fr_desc->ind &= ~ISERT_SIG_KEY_VALID; - sig_sge->lkey = pi_ctx->sig_mr->lkey; - sig_sge->addr = 0; - sig_sge->length = se_cmd->data_length; + rdma_wr->ib_sg[SIG].lkey = pi_ctx->sig_mr->lkey; + rdma_wr->ib_sg[SIG].addr = 0; + rdma_wr->ib_sg[SIG].length = se_cmd->data_length; if (se_cmd->prot_op != TARGET_PROT_DIN_STRIP && se_cmd->prot_op != TARGET_PROT_DOUT_INSERT) /* * We have protection guards on the wire * so we need to set a larget transfer */ - sig_sge->length += se_cmd->prot_length; + rdma_wr->ib_sg[SIG].length += se_cmd->prot_length; - pr_debug("sig_sge: addr: 0x%llx length: %u lkey: %x\n", - sig_sge->addr, sig_sge->length, - sig_sge->lkey); + isert_dbg("sig_sge: addr: 0x%llx length: %u lkey: %x\n", + rdma_wr->ib_sg[SIG].addr, rdma_wr->ib_sg[SIG].length, + rdma_wr->ib_sg[SIG].lkey); err: return ret; } static int +isert_handle_prot_cmd(struct isert_conn *isert_conn, + struct isert_cmd *isert_cmd, + struct isert_rdma_wr *wr) +{ + struct isert_device *device = isert_conn->conn_device; + struct se_cmd *se_cmd = &isert_cmd->iscsi_cmd->se_cmd; + int ret; + + if (!wr->fr_desc->pi_ctx) { + ret = isert_create_pi_ctx(wr->fr_desc, + device->ib_device, + isert_conn->conn_pd); + if (ret) { + isert_err("conn %p failed to allocate pi_ctx\n", + isert_conn); + return ret; + } + } + + if (se_cmd->t_prot_sg) { + ret = isert_map_data_buf(isert_conn, isert_cmd, + se_cmd->t_prot_sg, + se_cmd->t_prot_nents, + se_cmd->prot_length, + 0, wr->iser_ib_op, &wr->prot); + if (ret) { + isert_err("conn %p failed to map protection buffer\n", + isert_conn); + return ret; + } + + memset(&wr->ib_sg[PROT], 0, sizeof(wr->ib_sg[PROT])); + ret = isert_fast_reg_mr(isert_conn, wr->fr_desc, &wr->prot, + ISERT_PROT_KEY_VALID, &wr->ib_sg[PROT]); + if (ret) { + isert_err("conn %p failed to fast reg mr\n", + isert_conn); + goto unmap_prot_cmd; + } + } + + ret = isert_reg_sig_mr(isert_conn, se_cmd, wr, wr->fr_desc); + if (ret) { + isert_err("conn %p failed to fast reg mr\n", + isert_conn); + goto unmap_prot_cmd; + } + wr->fr_desc->ind |= ISERT_PROTECTED; + + return 0; + +unmap_prot_cmd: + if (se_cmd->t_prot_sg) + isert_unmap_data_buf(isert_conn, &wr->prot); + + return ret; +} + +static int isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, struct isert_rdma_wr *wr) { struct se_cmd *se_cmd = &cmd->se_cmd; struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = conn->context; - struct ib_sge data_sge; - struct ib_send_wr *send_wr; struct fast_reg_descriptor *fr_desc = NULL; + struct ib_send_wr *send_wr; + struct ib_sge *ib_sg; u32 offset; int ret = 0; unsigned long flags; @@ -2775,8 +2749,7 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, if (ret) return ret; - if (wr->data.dma_nents != 1 || - se_cmd->prot_op != TARGET_PROT_NORMAL) { + if (wr->data.dma_nents != 1 || isert_prot_cmd(isert_conn, se_cmd)) { spin_lock_irqsave(&isert_conn->conn_lock, flags); fr_desc = list_first_entry(&isert_conn->conn_fr_pool, struct fast_reg_descriptor, list); @@ -2786,38 +2759,21 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, } ret = isert_fast_reg_mr(isert_conn, fr_desc, &wr->data, - ISERT_DATA_KEY_VALID, &data_sge); + ISERT_DATA_KEY_VALID, &wr->ib_sg[DATA]); if (ret) goto unmap_cmd; - if (se_cmd->prot_op != TARGET_PROT_NORMAL) { - struct ib_sge prot_sge, sig_sge; - - if (se_cmd->t_prot_sg) { - ret = isert_map_data_buf(isert_conn, isert_cmd, - se_cmd->t_prot_sg, - se_cmd->t_prot_nents, - se_cmd->prot_length, - 0, wr->iser_ib_op, &wr->prot); - if (ret) - goto unmap_cmd; - - ret = isert_fast_reg_mr(isert_conn, fr_desc, &wr->prot, - ISERT_PROT_KEY_VALID, &prot_sge); - if (ret) - goto unmap_prot_cmd; - } - - ret = isert_reg_sig_mr(isert_conn, se_cmd, fr_desc, - &data_sge, &prot_sge, &sig_sge); + if (isert_prot_cmd(isert_conn, se_cmd)) { + ret = isert_handle_prot_cmd(isert_conn, isert_cmd, wr); if (ret) - goto unmap_prot_cmd; + goto unmap_cmd; - fr_desc->ind |= ISERT_PROTECTED; - memcpy(&wr->s_ib_sge, &sig_sge, sizeof(sig_sge)); - } else - memcpy(&wr->s_ib_sge, &data_sge, sizeof(data_sge)); + ib_sg = &wr->ib_sg[SIG]; + } else { + ib_sg = &wr->ib_sg[DATA]; + } + memcpy(&wr->s_ib_sge, ib_sg, sizeof(*ib_sg)); wr->ib_sge = &wr->s_ib_sge; wr->send_wr_num = 1; memset(&wr->s_send_wr, 0, sizeof(*send_wr)); @@ -2827,12 +2783,12 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, send_wr = &isert_cmd->rdma_wr.s_send_wr; send_wr->sg_list = &wr->s_ib_sge; send_wr->num_sge = 1; - send_wr->wr_id = (unsigned long)&isert_cmd->tx_desc; + send_wr->wr_id = (uintptr_t)&isert_cmd->tx_desc; if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) { send_wr->opcode = IB_WR_RDMA_WRITE; send_wr->wr.rdma.remote_addr = isert_cmd->read_va; send_wr->wr.rdma.rkey = isert_cmd->read_stag; - send_wr->send_flags = se_cmd->prot_op == TARGET_PROT_NORMAL ? + send_wr->send_flags = !isert_prot_cmd(isert_conn, se_cmd) ? 0 : IB_SEND_SIGNALED; } else { send_wr->opcode = IB_WR_RDMA_READ; @@ -2842,9 +2798,7 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, } return 0; -unmap_prot_cmd: - if (se_cmd->t_prot_sg) - isert_unmap_data_buf(isert_conn, &wr->prot); + unmap_cmd: if (fr_desc) { spin_lock_irqsave(&isert_conn->conn_lock, flags); @@ -2867,16 +2821,17 @@ isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd) struct ib_send_wr *wr_failed; int rc; - pr_debug("Cmd: %p RDMA_WRITE data_length: %u\n", + isert_dbg("Cmd: %p RDMA_WRITE data_length: %u\n", isert_cmd, se_cmd->data_length); + wr->iser_ib_op = ISER_IB_RDMA_WRITE; rc = device->reg_rdma_mem(conn, cmd, wr); if (rc) { - pr_err("Cmd: %p failed to prepare RDMA res\n", isert_cmd); + isert_err("Cmd: %p failed to prepare RDMA res\n", isert_cmd); return rc; } - if (se_cmd->prot_op == TARGET_PROT_NORMAL) { + if (!isert_prot_cmd(isert_conn, se_cmd)) { /* * Build isert_conn->tx_desc for iSCSI response PDU and attach */ @@ -2886,24 +2841,20 @@ isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd) &isert_cmd->tx_desc.iscsi_header); isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc); isert_init_send_wr(isert_conn, isert_cmd, - &isert_cmd->tx_desc.send_wr, false); + &isert_cmd->tx_desc.send_wr); isert_cmd->rdma_wr.s_send_wr.next = &isert_cmd->tx_desc.send_wr; wr->send_wr_num += 1; } - atomic_add(wr->send_wr_num, &isert_conn->post_send_buf_count); - rc = ib_post_send(isert_conn->conn_qp, wr->send_wr, &wr_failed); - if (rc) { - pr_warn("ib_post_send() failed for IB_WR_RDMA_WRITE\n"); - atomic_sub(wr->send_wr_num, &isert_conn->post_send_buf_count); - } + if (rc) + isert_warn("ib_post_send() failed for IB_WR_RDMA_WRITE\n"); - if (se_cmd->prot_op == TARGET_PROT_NORMAL) - pr_debug("Cmd: %p posted RDMA_WRITE + Response for iSER Data " + if (!isert_prot_cmd(isert_conn, se_cmd)) + isert_dbg("Cmd: %p posted RDMA_WRITE + Response for iSER Data " "READ\n", isert_cmd); else - pr_debug("Cmd: %p posted RDMA_WRITE for iSER Data READ\n", + isert_dbg("Cmd: %p posted RDMA_WRITE for iSER Data READ\n", isert_cmd); return 1; @@ -2920,23 +2871,20 @@ isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery) struct ib_send_wr *wr_failed; int rc; - pr_debug("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n", + isert_dbg("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n", isert_cmd, se_cmd->data_length, cmd->write_data_done); wr->iser_ib_op = ISER_IB_RDMA_READ; rc = device->reg_rdma_mem(conn, cmd, wr); if (rc) { - pr_err("Cmd: %p failed to prepare RDMA res\n", isert_cmd); + isert_err("Cmd: %p failed to prepare RDMA res\n", isert_cmd); return rc; } - atomic_add(wr->send_wr_num, &isert_conn->post_send_buf_count); - rc = ib_post_send(isert_conn->conn_qp, wr->send_wr, &wr_failed); - if (rc) { - pr_warn("ib_post_send() failed for IB_WR_RDMA_READ\n"); - atomic_sub(wr->send_wr_num, &isert_conn->post_send_buf_count); - } - pr_debug("Cmd: %p posted RDMA_READ memory for ISER Data WRITE\n", + if (rc) + isert_warn("ib_post_send() failed for IB_WR_RDMA_READ\n"); + + isert_dbg("Cmd: %p posted RDMA_READ memory for ISER Data WRITE\n", isert_cmd); return 0; @@ -2952,7 +2900,7 @@ isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state) ret = isert_put_nopin(cmd, conn, false); break; default: - pr_err("Unknown immediate state: 0x%02x\n", state); + isert_err("Unknown immediate state: 0x%02x\n", state); ret = -EINVAL; break; } @@ -2963,15 +2911,14 @@ isert_immediate_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state) static int isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state) { + struct isert_conn *isert_conn = conn->context; int ret; switch (state) { case ISTATE_SEND_LOGOUTRSP: ret = isert_put_logout_rsp(cmd, conn); - if (!ret) { - pr_debug("Returning iSER Logout -EAGAIN\n"); - ret = -EAGAIN; - } + if (!ret) + isert_conn->logout_posted = true; break; case ISTATE_SEND_NOPIN: ret = isert_put_nopin(cmd, conn, true); @@ -2993,7 +2940,7 @@ isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state) ret = isert_put_response(conn, cmd); break; default: - pr_err("Unknown response state: 0x%02x\n", state); + isert_err("Unknown response state: 0x%02x\n", state); ret = -EINVAL; break; } @@ -3001,27 +2948,64 @@ isert_response_queue(struct iscsi_conn *conn, struct iscsi_cmd *cmd, int state) return ret; } +struct rdma_cm_id * +isert_setup_id(struct isert_np *isert_np) +{ + struct iscsi_np *np = isert_np->np; + struct rdma_cm_id *id; + struct sockaddr *sa; + int ret; + + sa = (struct sockaddr *)&np->np_sockaddr; + isert_dbg("ksockaddr: %p, sa: %p\n", &np->np_sockaddr, sa); + + id = rdma_create_id(isert_cma_handler, isert_np, + RDMA_PS_TCP, IB_QPT_RC); + if (IS_ERR(id)) { + isert_err("rdma_create_id() failed: %ld\n", PTR_ERR(id)); + ret = PTR_ERR(id); + goto out; + } + isert_dbg("id %p context %p\n", id, id->context); + + ret = rdma_bind_addr(id, sa); + if (ret) { + isert_err("rdma_bind_addr() failed: %d\n", ret); + goto out_id; + } + + ret = rdma_listen(id, ISERT_RDMA_LISTEN_BACKLOG); + if (ret) { + isert_err("rdma_listen() failed: %d\n", ret); + goto out_id; + } + + return id; +out_id: + rdma_destroy_id(id); +out: + return ERR_PTR(ret); +} + static int isert_setup_np(struct iscsi_np *np, struct __kernel_sockaddr_storage *ksockaddr) { struct isert_np *isert_np; struct rdma_cm_id *isert_lid; - struct sockaddr *sa; int ret; isert_np = kzalloc(sizeof(struct isert_np), GFP_KERNEL); if (!isert_np) { - pr_err("Unable to allocate struct isert_np\n"); + isert_err("Unable to allocate struct isert_np\n"); return -ENOMEM; } sema_init(&isert_np->np_sem, 0); mutex_init(&isert_np->np_accept_mutex); INIT_LIST_HEAD(&isert_np->np_accept_list); init_completion(&isert_np->np_login_comp); + isert_np->np = np; - sa = (struct sockaddr *)ksockaddr; - pr_debug("ksockaddr: %p, sa: %p\n", ksockaddr, sa); /* * Setup the np->np_sockaddr from the passed sockaddr setup * in iscsi_target_configfs.c code.. @@ -3029,37 +3013,20 @@ isert_setup_np(struct iscsi_np *np, memcpy(&np->np_sockaddr, ksockaddr, sizeof(struct __kernel_sockaddr_storage)); - isert_lid = rdma_create_id(isert_cma_handler, np, RDMA_PS_TCP, - IB_QPT_RC); + isert_lid = isert_setup_id(isert_np); if (IS_ERR(isert_lid)) { - pr_err("rdma_create_id() for isert_listen_handler failed: %ld\n", - PTR_ERR(isert_lid)); ret = PTR_ERR(isert_lid); goto out; } - ret = rdma_bind_addr(isert_lid, sa); - if (ret) { - pr_err("rdma_bind_addr() for isert_lid failed: %d\n", ret); - goto out_lid; - } - - ret = rdma_listen(isert_lid, ISERT_RDMA_LISTEN_BACKLOG); - if (ret) { - pr_err("rdma_listen() for isert_lid failed: %d\n", ret); - goto out_lid; - } - isert_np->np_cm_id = isert_lid; np->np_context = isert_np; - pr_debug("Setup isert_lid->context: %p\n", isert_lid->context); return 0; -out_lid: - rdma_destroy_id(isert_lid); out: kfree(isert_np); + return ret; } @@ -3075,16 +3042,12 @@ isert_rdma_accept(struct isert_conn *isert_conn) cp.retry_count = 7; cp.rnr_retry_count = 7; - pr_debug("Before rdma_accept >>>>>>>>>>>>>>>>>>>>.\n"); - ret = rdma_accept(cm_id, &cp); if (ret) { - pr_err("rdma_accept() failed with: %d\n", ret); + isert_err("rdma_accept() failed with: %d\n", ret); return ret; } - pr_debug("After rdma_accept >>>>>>>>>>>>>>>>>>>>>.\n"); - return 0; } @@ -3094,7 +3057,15 @@ isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login) struct isert_conn *isert_conn = (struct isert_conn *)conn->context; int ret; - pr_debug("isert_get_login_rx before conn_login_comp conn: %p\n", conn); + isert_info("before login_req comp conn: %p\n", isert_conn); + ret = wait_for_completion_interruptible(&isert_conn->login_req_comp); + if (ret) { + isert_err("isert_conn %p interrupted before got login req\n", + isert_conn); + return ret; + } + reinit_completion(&isert_conn->login_req_comp); + /* * For login requests after the first PDU, isert_rx_login_req() will * kick schedule_delayed_work(&conn->login_work) as the packet is @@ -3104,11 +3075,15 @@ isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login) if (!login->first_request) return 0; + isert_rx_login_req(isert_conn); + + isert_info("before conn_login_comp conn: %p\n", conn); ret = wait_for_completion_interruptible(&isert_conn->conn_login_comp); if (ret) return ret; - pr_debug("isert_get_login_rx processing login->req: %p\n", login->req); + isert_info("processing login->req: %p\n", login->req); + return 0; } @@ -3161,7 +3136,7 @@ accept_wait: spin_lock_bh(&np->np_thread_lock); if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) { spin_unlock_bh(&np->np_thread_lock); - pr_debug("np_thread_state %d for isert_accept_np\n", + isert_dbg("np_thread_state %d for isert_accept_np\n", np->np_thread_state); /** * No point in stalling here when np_thread @@ -3186,17 +3161,10 @@ accept_wait: isert_conn->conn = conn; max_accept = 0; - ret = isert_rdma_post_recvl(isert_conn); - if (ret) - return ret; - - ret = isert_rdma_accept(isert_conn); - if (ret) - return ret; - isert_set_conn_info(np, conn, isert_conn); - pr_debug("Processing isert_accept_np: isert_conn: %p\n", isert_conn); + isert_dbg("Processing isert_conn: %p\n", isert_conn); + return 0; } @@ -3204,25 +3172,103 @@ static void isert_free_np(struct iscsi_np *np) { struct isert_np *isert_np = (struct isert_np *)np->np_context; + struct isert_conn *isert_conn, *n; if (isert_np->np_cm_id) rdma_destroy_id(isert_np->np_cm_id); + /* + * FIXME: At this point we don't have a good way to insure + * that at this point we don't have hanging connections that + * completed RDMA establishment but didn't start iscsi login + * process. So work-around this by cleaning up what ever piled + * up in np_accept_list. + */ + mutex_lock(&isert_np->np_accept_mutex); + if (!list_empty(&isert_np->np_accept_list)) { + isert_info("Still have isert connections, cleaning up...\n"); + list_for_each_entry_safe(isert_conn, n, + &isert_np->np_accept_list, + conn_accept_node) { + isert_info("cleaning isert_conn %p state (%d)\n", + isert_conn, isert_conn->state); + isert_connect_release(isert_conn); + } + } + mutex_unlock(&isert_np->np_accept_mutex); + np->np_context = NULL; kfree(isert_np); } +static void isert_release_work(struct work_struct *work) +{ + struct isert_conn *isert_conn = container_of(work, + struct isert_conn, + release_work); + + isert_info("Starting release conn %p\n", isert_conn); + + wait_for_completion(&isert_conn->conn_wait); + + mutex_lock(&isert_conn->conn_mutex); + isert_conn->state = ISER_CONN_DOWN; + mutex_unlock(&isert_conn->conn_mutex); + + isert_info("Destroying conn %p\n", isert_conn); + isert_put_conn(isert_conn); +} + +static void +isert_wait4logout(struct isert_conn *isert_conn) +{ + struct iscsi_conn *conn = isert_conn->conn; + + isert_info("conn %p\n", isert_conn); + + if (isert_conn->logout_posted) { + isert_info("conn %p wait for conn_logout_comp\n", isert_conn); + wait_for_completion_timeout(&conn->conn_logout_comp, + SECONDS_FOR_LOGOUT_COMP * HZ); + } +} + +static void +isert_wait4cmds(struct iscsi_conn *conn) +{ + isert_info("iscsi_conn %p\n", conn); + + if (conn->sess) { + target_sess_cmd_list_set_waiting(conn->sess->se_sess); + target_wait_for_sess_cmds(conn->sess->se_sess); + } +} + +static void +isert_wait4flush(struct isert_conn *isert_conn) +{ + struct ib_recv_wr *bad_wr; + + isert_info("conn %p\n", isert_conn); + + init_completion(&isert_conn->conn_wait_comp_err); + isert_conn->beacon.wr_id = ISER_BEACON_WRID; + /* post an indication that all flush errors were consumed */ + if (ib_post_recv(isert_conn->conn_qp, &isert_conn->beacon, &bad_wr)) { + isert_err("conn %p failed to post beacon", isert_conn); + return; + } + + wait_for_completion(&isert_conn->conn_wait_comp_err); +} + static void isert_wait_conn(struct iscsi_conn *conn) { struct isert_conn *isert_conn = conn->context; - pr_debug("isert_wait_conn: Starting \n"); + isert_info("Starting conn %p\n", isert_conn); mutex_lock(&isert_conn->conn_mutex); - if (isert_conn->conn_cm_id && !isert_conn->disconnect) { - pr_debug("Calling rdma_disconnect from isert_wait_conn\n"); - rdma_disconnect(isert_conn->conn_cm_id); - } /* * Only wait for conn_wait_comp_err if the isert_conn made it * into full feature phase.. @@ -3231,14 +3277,15 @@ static void isert_wait_conn(struct iscsi_conn *conn) mutex_unlock(&isert_conn->conn_mutex); return; } - if (isert_conn->state == ISER_CONN_UP) - isert_conn->state = ISER_CONN_TERMINATING; + isert_conn_terminate(isert_conn); mutex_unlock(&isert_conn->conn_mutex); - wait_for_completion(&isert_conn->conn_wait_comp_err); + isert_wait4cmds(conn); + isert_wait4flush(isert_conn); + isert_wait4logout(isert_conn); - wait_for_completion(&isert_conn->conn_wait); - isert_put_conn(isert_conn); + INIT_WORK(&isert_conn->release_work, isert_release_work); + queue_work(isert_release_wq, &isert_conn->release_work); } static void isert_free_conn(struct iscsi_conn *conn) @@ -3273,35 +3320,39 @@ static int __init isert_init(void) { int ret; - isert_rx_wq = alloc_workqueue("isert_rx_wq", 0, 0); - if (!isert_rx_wq) { - pr_err("Unable to allocate isert_rx_wq\n"); + isert_comp_wq = alloc_workqueue("isert_comp_wq", 0, 0); + if (!isert_comp_wq) { + isert_err("Unable to allocate isert_comp_wq\n"); + ret = -ENOMEM; return -ENOMEM; } - isert_comp_wq = alloc_workqueue("isert_comp_wq", 0, 0); - if (!isert_comp_wq) { - pr_err("Unable to allocate isert_comp_wq\n"); + isert_release_wq = alloc_workqueue("isert_release_wq", WQ_UNBOUND, + WQ_UNBOUND_MAX_ACTIVE); + if (!isert_release_wq) { + isert_err("Unable to allocate isert_release_wq\n"); ret = -ENOMEM; - goto destroy_rx_wq; + goto destroy_comp_wq; } iscsit_register_transport(&iser_target_transport); - pr_debug("iSER_TARGET[0] - Loaded iser_target_transport\n"); + isert_info("iSER_TARGET[0] - Loaded iser_target_transport\n"); + return 0; -destroy_rx_wq: - destroy_workqueue(isert_rx_wq); +destroy_comp_wq: + destroy_workqueue(isert_comp_wq); + return ret; } static void __exit isert_exit(void) { flush_scheduled_work(); + destroy_workqueue(isert_release_wq); destroy_workqueue(isert_comp_wq); - destroy_workqueue(isert_rx_wq); iscsit_unregister_transport(&iser_target_transport); - pr_debug("iSER_TARGET[0] - Released iser_target_transport\n"); + isert_info("iSER_TARGET[0] - Released iser_target_transport\n"); } MODULE_DESCRIPTION("iSER-Target for mainline target infrastructure"); diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index 04f51f7bf614..8dc8415d152d 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -4,9 +4,37 @@ #include <rdma/ib_verbs.h> #include <rdma/rdma_cm.h> +#define DRV_NAME "isert" +#define PFX DRV_NAME ": " + +#define isert_dbg(fmt, arg...) \ + do { \ + if (unlikely(isert_debug_level > 2)) \ + printk(KERN_DEBUG PFX "%s: " fmt,\ + __func__ , ## arg); \ + } while (0) + +#define isert_warn(fmt, arg...) \ + do { \ + if (unlikely(isert_debug_level > 0)) \ + pr_warn(PFX "%s: " fmt, \ + __func__ , ## arg); \ + } while (0) + +#define isert_info(fmt, arg...) \ + do { \ + if (unlikely(isert_debug_level > 1)) \ + pr_info(PFX "%s: " fmt, \ + __func__ , ## arg); \ + } while (0) + +#define isert_err(fmt, arg...) \ + pr_err(PFX "%s: " fmt, __func__ , ## arg) + #define ISERT_RDMA_LISTEN_BACKLOG 10 #define ISCSI_ISER_SG_TABLESIZE 256 #define ISER_FASTREG_LI_WRID 0xffffffffffffffffULL +#define ISER_BEACON_WRID 0xfffffffffffffffeULL enum isert_desc_type { ISCSI_TX_CONTROL, @@ -23,6 +51,7 @@ enum iser_ib_op_code { enum iser_conn_state { ISER_CONN_INIT, ISER_CONN_UP, + ISER_CONN_FULL_FEATURE, ISER_CONN_TERMINATING, ISER_CONN_DOWN, }; @@ -44,9 +73,6 @@ struct iser_tx_desc { struct ib_sge tx_sg[2]; int num_sge; struct isert_cmd *isert_cmd; - struct llist_node *comp_llnode_batch; - struct llist_node comp_llnode; - bool llnode_active; struct ib_send_wr send_wr; } __packed; @@ -81,6 +107,12 @@ struct isert_data_buf { enum dma_data_direction dma_dir; }; +enum { + DATA = 0, + PROT = 1, + SIG = 2, +}; + struct isert_rdma_wr { struct list_head wr_list; struct isert_cmd *isert_cmd; @@ -90,6 +122,7 @@ struct isert_rdma_wr { int send_wr_num; struct ib_send_wr *send_wr; struct ib_send_wr s_send_wr; + struct ib_sge ib_sg[3]; struct isert_data_buf data; struct isert_data_buf prot; struct fast_reg_descriptor *fr_desc; @@ -117,14 +150,15 @@ struct isert_device; struct isert_conn { enum iser_conn_state state; int post_recv_buf_count; - atomic_t post_send_buf_count; u32 responder_resources; u32 initiator_depth; + bool pi_support; u32 max_sge; char *login_buf; char *login_req_buf; char *login_rsp_buf; u64 login_req_dma; + int login_req_len; u64 login_rsp_dma; unsigned int conn_rx_desc_head; struct iser_rx_desc *conn_rx_descs; @@ -132,13 +166,13 @@ struct isert_conn { struct iscsi_conn *conn; struct list_head conn_accept_node; struct completion conn_login_comp; + struct completion login_req_comp; struct iser_tx_desc conn_login_tx_desc; struct rdma_cm_id *conn_cm_id; struct ib_pd *conn_pd; struct ib_mr *conn_mr; struct ib_qp *conn_qp; struct isert_device *conn_device; - struct work_struct conn_logout_work; struct mutex conn_mutex; struct completion conn_wait; struct completion conn_wait_comp_err; @@ -147,31 +181,38 @@ struct isert_conn { int conn_fr_pool_size; /* lock to protect fastreg pool */ spinlock_t conn_lock; -#define ISERT_COMP_BATCH_COUNT 8 - int conn_comp_batch; - struct llist_head conn_comp_llist; - bool disconnect; + struct work_struct release_work; + struct ib_recv_wr beacon; + bool logout_posted; }; #define ISERT_MAX_CQ 64 -struct isert_cq_desc { - struct isert_device *device; - int cq_index; - struct work_struct cq_rx_work; - struct work_struct cq_tx_work; +/** + * struct isert_comp - iSER completion context + * + * @device: pointer to device handle + * @cq: completion queue + * @wcs: work completion array + * @active_qps: Number of active QPs attached + * to completion context + * @work: completion work handle + */ +struct isert_comp { + struct isert_device *device; + struct ib_cq *cq; + struct ib_wc wcs[16]; + int active_qps; + struct work_struct work; }; struct isert_device { int use_fastreg; bool pi_capable; - int cqs_used; int refcount; - int cq_active_qps[ISERT_MAX_CQ]; struct ib_device *ib_device; - struct ib_cq *dev_rx_cq[ISERT_MAX_CQ]; - struct ib_cq *dev_tx_cq[ISERT_MAX_CQ]; - struct isert_cq_desc *cq_desc; + struct isert_comp *comps; + int comps_used; struct list_head dev_node; struct ib_device_attr dev_attr; int (*reg_rdma_mem)(struct iscsi_conn *conn, @@ -182,6 +223,7 @@ struct isert_device { }; struct isert_np { + struct iscsi_np *np; struct semaphore np_sem; struct rdma_cm_id *np_cm_id; struct mutex np_accept_mutex; diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 5461924c9f10..0747c0595a9d 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -2740,7 +2740,6 @@ static struct scsi_host_template srp_template = { .info = srp_target_info, .queuecommand = srp_queuecommand, .change_queue_depth = srp_change_queue_depth, - .change_queue_type = scsi_change_queue_type, .eh_abort_handler = srp_abort, .eh_device_reset_handler = srp_reset_device, .eh_host_reset_handler = srp_reset_host, @@ -2929,7 +2928,7 @@ static int srp_parse_options(const char *buf, struct srp_target_port *target) return -ENOMEM; sep_opt = options; - while ((p = strsep(&sep_opt, ",")) != NULL) { + while ((p = strsep(&sep_opt, ",\n")) != NULL) { if (!*p) continue; diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c index dc829682701a..eb694ddad79f 100644 --- a/drivers/infiniband/ulp/srpt/ib_srpt.c +++ b/drivers/infiniband/ulp/srpt/ib_srpt.c @@ -1708,17 +1708,17 @@ static int srpt_handle_cmd(struct srpt_rdma_ch *ch, switch (srp_cmd->task_attr) { case SRP_CMD_SIMPLE_Q: - cmd->sam_task_attr = MSG_SIMPLE_TAG; + cmd->sam_task_attr = TCM_SIMPLE_TAG; break; case SRP_CMD_ORDERED_Q: default: - cmd->sam_task_attr = MSG_ORDERED_TAG; + cmd->sam_task_attr = TCM_ORDERED_TAG; break; case SRP_CMD_HEAD_OF_Q: - cmd->sam_task_attr = MSG_HEAD_TAG; + cmd->sam_task_attr = TCM_HEAD_TAG; break; case SRP_CMD_ACA: - cmd->sam_task_attr = MSG_ACA_TAG; + cmd->sam_task_attr = TCM_ACA_TAG; break; } @@ -1733,7 +1733,7 @@ static int srpt_handle_cmd(struct srpt_rdma_ch *ch, sizeof(srp_cmd->lun)); rc = target_submit_cmd(cmd, ch->sess, srp_cmd->cdb, &send_ioctx->sense_data[0], unpacked_lun, data_len, - MSG_SIMPLE_TAG, dir, TARGET_SCF_ACK_KREF); + TCM_SIMPLE_TAG, dir, TARGET_SCF_ACK_KREF); if (rc != 0) { ret = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE; goto send_sense; |