diff options
23 files changed, 1778 insertions, 40 deletions
diff --git a/MAINTAINERS b/MAINTAINERS index 781e099495d3..d68df93ef637 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -631,7 +631,7 @@ F: drivers/iommu/amd_iommu*.[ch] F: include/linux/amd-iommu.h AMD KFD -M: Oded Gabbay <oded.gabbay@amd.com> +M: Oded Gabbay <oded.gabbay@gmail.com> L: dri-devel@lists.freedesktop.org T: git git://people.freedesktop.org/~gabbayo/linux.git S: Supported diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile index 0f4960148126..652d25478fd5 100644 --- a/drivers/gpu/drm/amd/amdkfd/Makefile +++ b/drivers/gpu/drm/amd/amdkfd/Makefile @@ -12,5 +12,6 @@ amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \ kfd_kernel_queue_vi.o kfd_packet_manager.o \ kfd_process_queue_manager.o kfd_device_queue_manager.o \ kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \ + kfd_interrupt.o kfd_events.o cik_event_interrupt.o obj-$(CONFIG_HSA_AMD) += amdkfd.o diff --git a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c new file mode 100644 index 000000000000..211fc48697fa --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c @@ -0,0 +1,66 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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 "kfd_priv.h" +#include "kfd_events.h" +#include "cik_int.h" + +static bool cik_event_interrupt_isr(struct kfd_dev *dev, + const uint32_t *ih_ring_entry) +{ + unsigned int pasid; + const struct cik_ih_ring_entry *ihre = + (const struct cik_ih_ring_entry *)ih_ring_entry; + + pasid = (ihre->ring_id & 0xffff0000) >> 16; + + /* Do not process in ISR, just request it to be forwarded to WQ. */ + return (pasid != 0) && + (ihre->source_id == CIK_INTSRC_CP_END_OF_PIPE || + ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG || + ihre->source_id == CIK_INTSRC_CP_BAD_OPCODE); +} + +static void cik_event_interrupt_wq(struct kfd_dev *dev, + const uint32_t *ih_ring_entry) +{ + unsigned int pasid; + const struct cik_ih_ring_entry *ihre = + (const struct cik_ih_ring_entry *)ih_ring_entry; + + pasid = (ihre->ring_id & 0xffff0000) >> 16; + + if (pasid == 0) + return; + + if (ihre->source_id == CIK_INTSRC_CP_END_OF_PIPE) + kfd_signal_event_interrupt(pasid, 0, 0); + else if (ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG) + kfd_signal_event_interrupt(pasid, ihre->data & 0xFF, 8); + else if (ihre->source_id == CIK_INTSRC_CP_BAD_OPCODE) + kfd_signal_hw_exception_event(pasid); +} + +const struct kfd_event_interrupt_class event_interrupt_class_cik = { + .interrupt_isr = cik_event_interrupt_isr, + .interrupt_wq = cik_event_interrupt_wq, +}; diff --git a/drivers/gpu/drm/amd/amdkfd/cik_int.h b/drivers/gpu/drm/amd/amdkfd/cik_int.h new file mode 100644 index 000000000000..79a16d24c1b8 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/cik_int.h @@ -0,0 +1,41 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +#ifndef HSA_RADEON_CIK_INT_H_INCLUDED +#define HSA_RADEON_CIK_INT_H_INCLUDED + +#include <linux/types.h> + +struct cik_ih_ring_entry { + uint32_t source_id; + uint32_t data; + uint32_t ring_id; + uint32_t reserved; +}; + +#define CIK_INTSRC_DEQUEUE_COMPLETE 0xC6 +#define CIK_INTSRC_CP_END_OF_PIPE 0xB5 +#define CIK_INTSRC_CP_BAD_OPCODE 0xB7 +#define CIK_INTSRC_SQ_INTERRUPT_MSG 0xEF + +#endif + diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 19a4fba46e4e..b2c6109bd7af 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -289,8 +289,10 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, args->queue_id = queue_id; + /* Return gpu_id as doorbell offset for mmap usage */ - args->doorbell_offset = args->gpu_id << PAGE_SHIFT; + args->doorbell_offset = (KFD_MMAP_DOORBELL_MASK | args->gpu_id); + args->doorbell_offset <<= PAGE_SHIFT; mutex_unlock(&p->mutex); @@ -514,6 +516,62 @@ static int kfd_ioctl_get_process_apertures(struct file *filp, return 0; } +static int kfd_ioctl_create_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_create_event_args *args = data; + int err; + + err = kfd_event_create(filp, p, args->event_type, + args->auto_reset != 0, args->node_id, + &args->event_id, &args->event_trigger_data, + &args->event_page_offset, + &args->event_slot_index); + + return err; +} + +static int kfd_ioctl_destroy_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_destroy_event_args *args = data; + + return kfd_event_destroy(p, args->event_id); +} + +static int kfd_ioctl_set_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_set_event_args *args = data; + + return kfd_set_event(p, args->event_id); +} + +static int kfd_ioctl_reset_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_reset_event_args *args = data; + + return kfd_reset_event(p, args->event_id); +} + +static int kfd_ioctl_wait_events(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_wait_events_args *args = data; + enum kfd_event_wait_result wait_result; + int err; + + err = kfd_wait_on_events(p, args->num_events, + (void __user *)args->events_ptr, + (args->wait_for_all != 0), + args->timeout, &wait_result); + + args->wait_result = wait_result; + + return err; +} + #define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \ [_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, .cmd_drv = 0, .name = #ioctl} @@ -539,6 +597,21 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = { AMDKFD_IOCTL_DEF(AMDKFD_IOC_UPDATE_QUEUE, kfd_ioctl_update_queue, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_CREATE_EVENT, + kfd_ioctl_create_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_DESTROY_EVENT, + kfd_ioctl_destroy_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_SET_EVENT, + kfd_ioctl_set_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_RESET_EVENT, + kfd_ioctl_reset_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_WAIT_EVENTS, + kfd_ioctl_wait_events, 0), }; #define AMDKFD_CORE_IOCTL_COUNT ARRAY_SIZE(amdkfd_ioctls) @@ -639,5 +712,15 @@ static int kfd_mmap(struct file *filp, struct vm_area_struct *vma) if (IS_ERR(process)) return PTR_ERR(process); - return kfd_doorbell_mmap(process, vma); + if ((vma->vm_pgoff & KFD_MMAP_DOORBELL_MASK) == + KFD_MMAP_DOORBELL_MASK) { + vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_DOORBELL_MASK; + return kfd_doorbell_mmap(process, vma); + } else if ((vma->vm_pgoff & KFD_MMAP_EVENTS_MASK) == + KFD_MMAP_EVENTS_MASK) { + vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_EVENTS_MASK; + return kfd_event_mmap(process, vma); + } + + return -EFAULT; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index ca7f2d3af2ff..52cab0f53ebc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -34,6 +34,7 @@ static const struct kfd_device_info kaveri_device_info = { .asic_family = CHIP_KAVERI, .max_pasid_bits = 16, .ih_ring_entry_size = 4 * sizeof(uint32_t), + .event_interrupt_class = &event_interrupt_class_cik, .mqd_size_aligned = MQD_SIZE_ALIGNED }; @@ -181,6 +182,32 @@ static void iommu_pasid_shutdown_callback(struct pci_dev *pdev, int pasid) kfd_unbind_process_from_device(dev, pasid); } +/* + * This function called by IOMMU driver on PPR failure + */ +static int iommu_invalid_ppr_cb(struct pci_dev *pdev, int pasid, + unsigned long address, u16 flags) +{ + struct kfd_dev *dev; + + dev_warn(kfd_device, + "Invalid PPR device %x:%x.%x pasid %d address 0x%lX flags 0x%X", + PCI_BUS_NUM(pdev->devfn), + PCI_SLOT(pdev->devfn), + PCI_FUNC(pdev->devfn), + pasid, + address, + flags); + + dev = kfd_device_by_pci_dev(pdev); + BUG_ON(dev == NULL); + + kfd_signal_iommu_event(dev, pasid, address, + flags & PPR_FAULT_WRITE, flags & PPR_FAULT_EXEC); + + return AMD_IOMMU_INV_PRI_RSP_INVALID; +} + bool kgd2kfd_device_init(struct kfd_dev *kfd, const struct kgd2kfd_shared_resources *gpu_resources) { @@ -235,6 +262,13 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, goto kfd_topology_add_device_error; } + if (kfd_interrupt_init(kfd)) { + dev_err(kfd_device, + "Error initializing interrupts for device (%x:%x)\n", + kfd->pdev->vendor, kfd->pdev->device); + goto kfd_interrupt_error; + } + if (!device_iommu_pasid_init(kfd)) { dev_err(kfd_device, "Error initializing iommuv2 for device (%x:%x)\n", @@ -243,6 +277,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, } amd_iommu_set_invalidate_ctx_cb(kfd->pdev, iommu_pasid_shutdown_callback); + amd_iommu_set_invalid_ppr_cb(kfd->pdev, iommu_invalid_ppr_cb); kfd->dqm = device_queue_manager_init(kfd); if (!kfd->dqm) { @@ -273,6 +308,8 @@ dqm_start_error: device_queue_manager_error: amd_iommu_free_device(kfd->pdev); device_iommu_pasid_error: + kfd_interrupt_exit(kfd); +kfd_interrupt_error: kfd_topology_remove_device(kfd); kfd_topology_add_device_error: kfd_gtt_sa_fini(kfd); @@ -290,6 +327,7 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd) if (kfd->init_complete) { device_queue_manager_uninit(kfd->dqm); amd_iommu_free_device(kfd->pdev); + kfd_interrupt_exit(kfd); kfd_topology_remove_device(kfd); kfd_gtt_sa_fini(kfd); kfd->kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem); @@ -305,6 +343,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd) if (kfd->init_complete) { kfd->dqm->ops.stop(kfd->dqm); amd_iommu_set_invalidate_ctx_cb(kfd->pdev, NULL); + amd_iommu_set_invalid_ppr_cb(kfd->pdev, NULL); amd_iommu_free_device(kfd->pdev); } } @@ -324,6 +363,7 @@ int kgd2kfd_resume(struct kfd_dev *kfd) return -ENXIO; amd_iommu_set_invalidate_ctx_cb(kfd->pdev, iommu_pasid_shutdown_callback); + amd_iommu_set_invalid_ppr_cb(kfd->pdev, iommu_invalid_ppr_cb); kfd->dqm->ops.start(kfd->dqm); } @@ -333,7 +373,17 @@ int kgd2kfd_resume(struct kfd_dev *kfd) /* This is called directly from KGD at ISR. */ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) { - /* Process interrupts / schedule work as necessary */ + if (!kfd->init_complete) + return; + + spin_lock(&kfd->interrupt_lock); + + if (kfd->interrupts_active + && interrupt_is_wanted(kfd, ih_ring_entry) + && enqueue_ih_ring_entry(kfd, ih_ring_entry)) + schedule_work(&kfd->interrupt_work); + + spin_unlock(&kfd->interrupt_lock); } static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 69af73f15310..078092c5a59f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -522,6 +522,17 @@ int init_pipelines(struct device_queue_manager *dqm, return 0; } +static void init_interrupts(struct device_queue_manager *dqm) +{ + unsigned int i; + + BUG_ON(dqm == NULL); + + for (i = 0 ; i < get_pipes_num(dqm) ; i++) + dqm->dev->kfd2kgd->init_interrupts(dqm->dev->kgd, + i + get_first_pipe(dqm)); +} + static int init_scheduler(struct device_queue_manager *dqm) { int retval; @@ -581,6 +592,7 @@ static void uninitialize_nocpsch(struct device_queue_manager *dqm) static int start_nocpsch(struct device_queue_manager *dqm) { + init_interrupts(dqm); return 0; } @@ -614,19 +626,6 @@ static void deallocate_sdma_queue(struct device_queue_manager *dqm, set_bit(sdma_queue_id, (unsigned long *)&dqm->sdma_bitmap); } -static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, - struct qcm_process_device *qpd) -{ - uint32_t value = SDMA_ATC; - - if (q->process->is_32bit_user_mode) - value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd)); - else - value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64( - qpd_to_pdd(qpd))); - q->properties.sdma_vm_addr = value; -} - static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, struct queue *q, struct qcm_process_device *qpd) @@ -649,7 +648,7 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, pr_debug(" sdma queue id: %d\n", q->properties.sdma_queue_id); pr_debug(" sdma engine id: %d\n", q->properties.sdma_engine_id); - init_sdma_vm(dqm, q, qpd); + dqm->ops_asic_specific.init_sdma_vm(dqm, q, qpd); retval = mqd->init_mqd(mqd, &q->mqd, &q->mqd_mem_obj, &q->gart_mqd_addr, &q->properties); if (retval != 0) { @@ -750,6 +749,9 @@ static int start_cpsch(struct device_queue_manager *dqm) dqm->fence_addr = dqm->fence_mem->cpu_ptr; dqm->fence_gpu_addr = dqm->fence_mem->gpu_addr; + + init_interrupts(dqm); + list_for_each_entry(node, &dqm->queues, list) if (node->qpd->pqm->process && dqm->dev) kfd_bind_process_to_device(dqm->dev, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h index 488f51d19427..57278e2d72e0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h @@ -120,6 +120,21 @@ struct device_queue_manager_ops { uint64_t alternate_aperture_size); }; +struct device_queue_manager_asic_ops { + int (*register_process)(struct device_queue_manager *dqm, + struct qcm_process_device *qpd); + int (*initialize)(struct device_queue_manager *dqm); + bool (*set_cache_memory_policy)(struct device_queue_manager *dqm, + struct qcm_process_device *qpd, + enum cache_policy default_policy, + enum cache_policy alternate_policy, + void __user *alternate_aperture_base, + uint64_t alternate_aperture_size); + void (*init_sdma_vm)(struct device_queue_manager *dqm, + struct queue *q, + struct qcm_process_device *qpd); +}; + /** * struct device_queue_manager * @@ -134,7 +149,7 @@ struct device_queue_manager_ops { struct device_queue_manager { struct device_queue_manager_ops ops; - struct device_queue_manager_ops ops_asic_specific; + struct device_queue_manager_asic_ops ops_asic_specific; struct mqd_manager *mqds[KFD_MQD_TYPE_MAX]; struct packet_manager packets; @@ -157,8 +172,8 @@ struct device_queue_manager { bool active_runlist; }; -void device_queue_manager_init_cik(struct device_queue_manager_ops *ops); -void device_queue_manager_init_vi(struct device_queue_manager_ops *ops); +void device_queue_manager_init_cik(struct device_queue_manager_asic_ops *ops); +void device_queue_manager_init_vi(struct device_queue_manager_asic_ops *ops); void program_sh_mem_settings(struct device_queue_manager *dqm, struct qcm_process_device *qpd); int init_pipelines(struct device_queue_manager *dqm, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c index 5469efe0523e..9ce8a20a7aff 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c @@ -33,12 +33,15 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm, static int register_process_cik(struct device_queue_manager *dqm, struct qcm_process_device *qpd); static int initialize_cpsch_cik(struct device_queue_manager *dqm); +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd); -void device_queue_manager_init_cik(struct device_queue_manager_ops *ops) +void device_queue_manager_init_cik(struct device_queue_manager_asic_ops *ops) { ops->set_cache_memory_policy = set_cache_memory_policy_cik; ops->register_process = register_process_cik; ops->initialize = initialize_cpsch_cik; + ops->init_sdma_vm = init_sdma_vm; } static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble) @@ -129,6 +132,19 @@ static int register_process_cik(struct device_queue_manager *dqm, return 0; } +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd) +{ + uint32_t value = SDMA_ATC; + + if (q->process->is_32bit_user_mode) + value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd)); + else + value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64( + qpd_to_pdd(qpd))); + q->properties.sdma_vm_addr = value; +} + static int initialize_cpsch_cik(struct device_queue_manager *dqm) { return init_pipelines(dqm, get_pipes_num(dqm), get_first_pipe(dqm)); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c index 20553dcd257d..4c15212a3899 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c @@ -32,14 +32,17 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm, static int register_process_vi(struct device_queue_manager *dqm, struct qcm_process_device *qpd); static int initialize_cpsch_vi(struct device_queue_manager *dqm); +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd); -void device_queue_manager_init_vi(struct device_queue_manager_ops *ops) +void device_queue_manager_init_vi(struct device_queue_manager_asic_ops *ops) { pr_warn("amdkfd: VI DQM is not currently supported\n"); ops->set_cache_memory_policy = set_cache_memory_policy_vi; ops->register_process = register_process_vi; ops->initialize = initialize_cpsch_vi; + ops->init_sdma_vm = init_sdma_vm; } static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm, @@ -58,6 +61,11 @@ static int register_process_vi(struct device_queue_manager *dqm, return -1; } +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd) +{ +} + static int initialize_cpsch_vi(struct device_queue_manager *dqm) { return 0; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c index 17e56dcc8540..e621eba63126 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c @@ -142,14 +142,13 @@ int kfd_doorbell_mmap(struct kfd_process *process, struct vm_area_struct *vma) vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); - pr_debug("kfd: mapping doorbell page in kfd_doorbell_mmap\n" - " target user address == 0x%08llX\n" - " physical address == 0x%08llX\n" - " vm_flags == 0x%04lX\n" - " size == 0x%04lX\n", - (unsigned long long) vma->vm_start, address, vma->vm_flags, - doorbell_process_allocation()); - + pr_debug("mapping doorbell page:\n"); + pr_debug(" target user address == 0x%08llX\n", + (unsigned long long) vma->vm_start); + pr_debug(" physical address == 0x%08llX\n", address); + pr_debug(" vm_flags == 0x%04lX\n", vma->vm_flags); + pr_debug(" size == 0x%04lX\n", + doorbell_process_allocation()); return io_remap_pfn_range(vma, vma->vm_start, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c new file mode 100644 index 000000000000..3cb37d220f60 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -0,0 +1,965 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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/mm_types.h> +#include <linux/slab.h> +#include <linux/types.h> +#include <linux/sched.h> +#include <linux/uaccess.h> +#include <linux/mm.h> +#include <linux/mman.h> +#include <linux/memory.h> +#include "kfd_priv.h" +#include "kfd_events.h" +#include <linux/device.h> + +/* + * A task can only be on a single wait_queue at a time, but we need to support + * waiting on multiple events (any/all). + * Instead of each event simply having a wait_queue with sleeping tasks, it + * has a singly-linked list of tasks. + * A thread that wants to sleep creates an array of these, one for each event + * and adds one to each event's waiter chain. + */ +struct kfd_event_waiter { + struct list_head waiters; + struct task_struct *sleeping_task; + + /* Transitions to true when the event this belongs to is signaled. */ + bool activated; + + /* Event */ + struct kfd_event *event; + uint32_t input_index; +}; + +/* + * Over-complicated pooled allocator for event notification slots. + * + * Each signal event needs a 64-bit signal slot where the signaler will write + * a 1 before sending an interrupt.l (This is needed because some interrupts + * do not contain enough spare data bits to identify an event.) + * We get whole pages from vmalloc and map them to the process VA. + * Individual signal events are then allocated a slot in a page. + */ + +struct signal_page { + struct list_head event_pages; /* kfd_process.signal_event_pages */ + uint64_t *kernel_address; + uint64_t __user *user_address; + uint32_t page_index; /* Index into the mmap aperture. */ + unsigned int free_slots; + unsigned long used_slot_bitmap[0]; +}; + +#define SLOTS_PER_PAGE KFD_SIGNAL_EVENT_LIMIT +#define SLOT_BITMAP_SIZE BITS_TO_LONGS(SLOTS_PER_PAGE) +#define BITS_PER_PAGE (ilog2(SLOTS_PER_PAGE)+1) +#define SIGNAL_PAGE_SIZE (sizeof(struct signal_page) + \ + SLOT_BITMAP_SIZE * sizeof(long)) + +/* + * For signal events, the event ID is used as the interrupt user data. + * For SQ s_sendmsg interrupts, this is limited to 8 bits. + */ + +#define INTERRUPT_DATA_BITS 8 +#define SIGNAL_EVENT_ID_SLOT_SHIFT 0 + +static uint64_t *page_slots(struct signal_page *page) +{ + return page->kernel_address; +} + +static bool allocate_free_slot(struct kfd_process *process, + struct signal_page **out_page, + unsigned int *out_slot_index) +{ + struct signal_page *page; + + list_for_each_entry(page, &process->signal_event_pages, event_pages) { + if (page->free_slots > 0) { + unsigned int slot = + find_first_zero_bit(page->used_slot_bitmap, + SLOTS_PER_PAGE); + + __set_bit(slot, page->used_slot_bitmap); + page->free_slots--; + + page_slots(page)[slot] = UNSIGNALED_EVENT_SLOT; + + *out_page = page; + *out_slot_index = slot; + + pr_debug("allocated event signal slot in page %p, slot %d\n", + page, slot); + + return true; + } + } + + pr_debug("No free event signal slots were found for process %p\n", + process); + + return false; +} + +#define list_tail_entry(head, type, member) \ + list_entry((head)->prev, type, member) + +static bool allocate_signal_page(struct file *devkfd, struct kfd_process *p) +{ + void *backing_store; + struct signal_page *page; + + page = kzalloc(SIGNAL_PAGE_SIZE, GFP_KERNEL); + if (!page) + goto fail_alloc_signal_page; + + page->free_slots = SLOTS_PER_PAGE; + + backing_store = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO, + get_order(KFD_SIGNAL_EVENT_LIMIT * 8)); + if (!backing_store) + goto fail_alloc_signal_store; + + /* prevent user-mode info leaks */ + memset(backing_store, (uint8_t) UNSIGNALED_EVENT_SLOT, + KFD_SIGNAL_EVENT_LIMIT * 8); + + page->kernel_address = backing_store; + + if (list_empty(&p->signal_event_pages)) + page->page_index = 0; + else + page->page_index = list_tail_entry(&p->signal_event_pages, + struct signal_page, + event_pages)->page_index + 1; + + pr_debug("allocated new event signal page at %p, for process %p\n", + page, p); + pr_debug("page index is %d\n", page->page_index); + + list_add(&page->event_pages, &p->signal_event_pages); + + return true; + +fail_alloc_signal_store: + kfree(page); +fail_alloc_signal_page: + return false; +} + +static bool allocate_event_notification_slot(struct file *devkfd, + struct kfd_process *p, + struct signal_page **page, + unsigned int *signal_slot_index) +{ + bool ret; + + ret = allocate_free_slot(p, page, signal_slot_index); + if (ret == false) { + ret = allocate_signal_page(devkfd, p); + if (ret == true) + ret = allocate_free_slot(p, page, signal_slot_index); + } + + return ret; +} + +/* Assumes that the process's event_mutex is locked. */ +static void release_event_notification_slot(struct signal_page *page, + size_t slot_index) +{ + __clear_bit(slot_index, page->used_slot_bitmap); + page->free_slots++; + + /* We don't free signal pages, they are retained by the process + * and reused until it exits. */ +} + +static struct signal_page *lookup_signal_page_by_index(struct kfd_process *p, + unsigned int page_index) +{ + struct signal_page *page; + + /* + * This is safe because we don't delete signal pages until the + * process exits. + */ + list_for_each_entry(page, &p->signal_event_pages, event_pages) + if (page->page_index == page_index) + return page; + + return NULL; +} + +/* + * Assumes that p->event_mutex is held and of course that p is not going + * away (current or locked). + */ +static struct kfd_event *lookup_event_by_id(struct kfd_process *p, uint32_t id) +{ + struct kfd_event *ev; + + hash_for_each_possible(p->events, ev, events, id) + if (ev->event_id == id) + return ev; + + return NULL; +} + +static u32 make_signal_event_id(struct signal_page *page, + unsigned int signal_slot_index) +{ + return page->page_index | + (signal_slot_index << SIGNAL_EVENT_ID_SLOT_SHIFT); +} + +/* + * Produce a kfd event id for a nonsignal event. + * These are arbitrary numbers, so we do a sequential search through + * the hash table for an unused number. + */ +static u32 make_nonsignal_event_id(struct kfd_process *p) +{ + u32 id; + + for (id = p->next_nonsignal_event_id; + id < KFD_LAST_NONSIGNAL_EVENT_ID && + lookup_event_by_id(p, id) != NULL; + id++) + ; + + if (id < KFD_LAST_NONSIGNAL_EVENT_ID) { + + /* + * What if id == LAST_NONSIGNAL_EVENT_ID - 1? + * Then next_nonsignal_event_id = LAST_NONSIGNAL_EVENT_ID so + * the first loop fails immediately and we proceed with the + * wraparound loop below. + */ + p->next_nonsignal_event_id = id + 1; + + return id; + } + + for (id = KFD_FIRST_NONSIGNAL_EVENT_ID; + id < KFD_LAST_NONSIGNAL_EVENT_ID && + lookup_event_by_id(p, id) != NULL; + id++) + ; + + + if (id < KFD_LAST_NONSIGNAL_EVENT_ID) { + p->next_nonsignal_event_id = id + 1; + return id; + } + + p->next_nonsignal_event_id = KFD_FIRST_NONSIGNAL_EVENT_ID; + return 0; +} + +static struct kfd_event *lookup_event_by_page_slot(struct kfd_process *p, + struct signal_page *page, + unsigned int signal_slot) +{ + return lookup_event_by_id(p, make_signal_event_id(page, signal_slot)); +} + +static int create_signal_event(struct file *devkfd, + struct kfd_process *p, + struct kfd_event *ev) +{ + if (p->signal_event_count == KFD_SIGNAL_EVENT_LIMIT) { + pr_warn("amdkfd: Signal event wasn't created because limit was reached\n"); + return -ENOMEM; + } + + if (!allocate_event_notification_slot(devkfd, p, &ev->signal_page, + &ev->signal_slot_index)) { + pr_warn("amdkfd: Signal event wasn't created because out of kernel memory\n"); + return -ENOMEM; + } + + p->signal_event_count++; + + ev->user_signal_address = + &ev->signal_page->user_address[ev->signal_slot_index]; + + ev->event_id = make_signal_event_id(ev->signal_page, + ev->signal_slot_index); + + pr_debug("signal event number %zu created with id %d, address %p\n", + p->signal_event_count, ev->event_id, + ev->user_signal_address); + + return 0; +} + +/* + * No non-signal events are supported yet. + * We create them as events that never signal. + * Set event calls from user-mode are failed. + */ +static int create_other_event(struct kfd_process *p, struct kfd_event *ev) +{ + ev->event_id = make_nonsignal_event_id(p); + if (ev->event_id == 0) + return -ENOMEM; + + return 0; +} + +void kfd_event_init_process(struct kfd_process *p) +{ + mutex_init(&p->event_mutex); + hash_init(p->events); + INIT_LIST_HEAD(&p->signal_event_pages); + p->next_nonsignal_event_id = KFD_FIRST_NONSIGNAL_EVENT_ID; + p->signal_event_count = 0; +} + +static void destroy_event(struct kfd_process *p, struct kfd_event *ev) +{ + if (ev->signal_page != NULL) { + release_event_notification_slot(ev->signal_page, + ev->signal_slot_index); + p->signal_event_count--; + } + + /* + * Abandon the list of waiters. Individual waiting threads will + * clean up their own data. + */ + list_del(&ev->waiters); + + hash_del(&ev->events); + kfree(ev); +} + +static void destroy_events(struct kfd_process *p) +{ + struct kfd_event *ev; + struct hlist_node *tmp; + unsigned int hash_bkt; + + hash_for_each_safe(p->events, hash_bkt, tmp, ev, events) + destroy_event(p, ev); +} + +/* + * We assume that the process is being destroyed and there is no need to + * unmap the pages or keep bookkeeping data in order. + */ +static void shutdown_signal_pages(struct kfd_process *p) +{ + struct signal_page *page, *tmp; + + list_for_each_entry_safe(page, tmp, &p->signal_event_pages, + event_pages) { + free_pages((unsigned long)page->kernel_address, + get_order(KFD_SIGNAL_EVENT_LIMIT * 8)); + kfree(page); + } +} + +void kfd_event_free_process(struct kfd_process *p) +{ + destroy_events(p); + shutdown_signal_pages(p); +} + +static bool event_can_be_gpu_signaled(const struct kfd_event *ev) +{ + return ev->type == KFD_EVENT_TYPE_SIGNAL || + ev->type == KFD_EVENT_TYPE_DEBUG; +} + +static bool event_can_be_cpu_signaled(const struct kfd_event *ev) +{ + return ev->type == KFD_EVENT_TYPE_SIGNAL; +} + +int kfd_event_create(struct file *devkfd, struct kfd_process *p, + uint32_t event_type, bool auto_reset, uint32_t node_id, + uint32_t *event_id, uint32_t *event_trigger_data, + uint64_t *event_page_offset, uint32_t *event_slot_index) +{ + int ret = 0; + struct kfd_event *ev = kzalloc(sizeof(*ev), GFP_KERNEL); + + if (!ev) + return -ENOMEM; + + ev->type = event_type; + ev->auto_reset = auto_reset; + ev->signaled = false; + + INIT_LIST_HEAD(&ev->waiters); + + *event_page_offset = 0; + + mutex_lock(&p->event_mutex); + + switch (event_type) { + case KFD_EVENT_TYPE_SIGNAL: + case KFD_EVENT_TYPE_DEBUG: + ret = create_signal_event(devkfd, p, ev); + if (!ret) { + *event_page_offset = (ev->signal_page->page_index | + KFD_MMAP_EVENTS_MASK); + *event_page_offset <<= PAGE_SHIFT; + *event_slot_index = ev->signal_slot_index; + } + break; + default: + ret = create_other_event(p, ev); + break; + } + + if (!ret) { + hash_add(p->events, &ev->events, ev->event_id); + + *event_id = ev->event_id; + *event_trigger_data = ev->event_id; + } else { + kfree(ev); + } + + mutex_unlock(&p->event_mutex); + + return ret; +} + +/* Assumes that p is current. */ +int kfd_event_destroy(struct kfd_process *p, uint32_t event_id) +{ + struct kfd_event *ev; + int ret = 0; + + mutex_lock(&p->event_mutex); + + ev = lookup_event_by_id(p, event_id); + + if (ev) + destroy_event(p, ev); + else + ret = -EINVAL; + + mutex_unlock(&p->event_mutex); + return ret; +} + +static void set_event(struct kfd_event *ev) +{ + struct kfd_event_waiter *waiter; + struct kfd_event_waiter *next; + + /* Auto reset if the list is non-empty and we're waking someone. */ + ev->signaled = !ev->auto_reset || list_empty(&ev->waiters); + + list_for_each_entry_safe(waiter, next, &ev->waiters, waiters) { + waiter->activated = true; + + /* _init because free_waiters will call list_del */ + list_del_init(&waiter->waiters); + + wake_up_process(waiter->sleeping_task); + } +} + +/* Assumes that p is current. */ +int kfd_set_event(struct kfd_process *p, uint32_t event_id) +{ + int ret = 0; + struct kfd_event *ev; + + mutex_lock(&p->event_mutex); + + ev = lookup_event_by_id(p, event_id); + + if (ev && event_can_be_cpu_signaled(ev)) + set_event(ev); + else + ret = -EINVAL; + + mutex_unlock(&p->event_mutex); + return ret; +} + +static void reset_event(struct kfd_event *ev) +{ + ev->signaled = false; +} + +/* Assumes that p is current. */ +int kfd_reset_event(struct kfd_process *p, uint32_t event_id) +{ + int ret = 0; + struct kfd_event *ev; + + mutex_lock(&p->event_mutex); + + ev = lookup_event_by_id(p, event_id); + + if (ev && event_can_be_cpu_signaled(ev)) + reset_event(ev); + else + ret = -EINVAL; + + mutex_unlock(&p->event_mutex); + return ret; + +} + +static void acknowledge_signal(struct kfd_process *p, struct kfd_event *ev) +{ + page_slots(ev->signal_page)[ev->signal_slot_index] = + UNSIGNALED_EVENT_SLOT; +} + +static bool is_slot_signaled(struct signal_page *page, unsigned int index) +{ + return page_slots(page)[index] != UNSIGNALED_EVENT_SLOT; +} + +static void set_event_from_interrupt(struct kfd_process *p, + struct kfd_event *ev) +{ + if (ev && event_can_be_gpu_signaled(ev)) { + acknowledge_signal(p, ev); + set_event(ev); + } +} + +void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, + uint32_t valid_id_bits) +{ + struct kfd_event *ev; + + /* + * Because we are called from arbitrary context (workqueue) as opposed + * to process context, kfd_process could attempt to exit while we are + * running so the lookup function returns a locked process. + */ + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; /* Presumably process exited. */ + + mutex_lock(&p->event_mutex); + + if (valid_id_bits >= INTERRUPT_DATA_BITS) { + /* Partial ID is a full ID. */ + ev = lookup_event_by_id(p, partial_id); + set_event_from_interrupt(p, ev); + } else { + /* + * Partial ID is in fact partial. For now we completely + * ignore it, but we could use any bits we did receive to + * search faster. + */ + struct signal_page *page; + unsigned i; + + list_for_each_entry(page, &p->signal_event_pages, event_pages) + for (i = 0; i < SLOTS_PER_PAGE; i++) + if (is_slot_signaled(page, i)) { + ev = lookup_event_by_page_slot(p, + page, i); + set_event_from_interrupt(p, ev); + } + } + + mutex_unlock(&p->event_mutex); + mutex_unlock(&p->mutex); +} + +static struct kfd_event_waiter *alloc_event_waiters(uint32_t num_events) +{ + struct kfd_event_waiter *event_waiters; + uint32_t i; + + event_waiters = kmalloc_array(num_events, + sizeof(struct kfd_event_waiter), + GFP_KERNEL); + + for (i = 0; (event_waiters) && (i < num_events) ; i++) { + INIT_LIST_HEAD(&event_waiters[i].waiters); + event_waiters[i].sleeping_task = current; + event_waiters[i].activated = false; + } + + return event_waiters; +} + +static int init_event_waiter(struct kfd_process *p, + struct kfd_event_waiter *waiter, + uint32_t event_id, + uint32_t input_index) +{ + struct kfd_event *ev = lookup_event_by_id(p, event_id); + + if (!ev) + return -EINVAL; + + waiter->event = ev; + waiter->input_index = input_index; + waiter->activated = ev->signaled; + ev->signaled = ev->signaled && !ev->auto_reset; + + list_add(&waiter->waiters, &ev->waiters); + + return 0; +} + +static bool test_event_condition(bool all, uint32_t num_events, + struct kfd_event_waiter *event_waiters) +{ + uint32_t i; + uint32_t activated_count = 0; + + for (i = 0; i < num_events; i++) { + if (event_waiters[i].activated) { + if (!all) + return true; + + activated_count++; + } + } + + return activated_count == num_events; +} + +/* + * Copy event specific data, if defined. + * Currently only memory exception events have additional data to copy to user + */ +static bool copy_signaled_event_data(uint32_t num_events, + struct kfd_event_waiter *event_waiters, + struct kfd_event_data __user *data) +{ + struct kfd_hsa_memory_exception_data *src; + struct kfd_hsa_memory_exception_data __user *dst; + struct kfd_event_waiter *waiter; + struct kfd_event *event; + uint32_t i; + + for (i = 0; i < num_events; i++) { + waiter = &event_waiters[i]; + event = waiter->event; + if (waiter->activated && event->type == KFD_EVENT_TYPE_MEMORY) { + dst = &data[waiter->input_index].memory_exception_data; + src = &event->memory_exception_data; + if (copy_to_user(dst, src, + sizeof(struct kfd_hsa_memory_exception_data))) + return false; + } + } + + return true; + +} + + + +static long user_timeout_to_jiffies(uint32_t user_timeout_ms) +{ + if (user_timeout_ms == KFD_EVENT_TIMEOUT_IMMEDIATE) + return 0; + + if (user_timeout_ms == KFD_EVENT_TIMEOUT_INFINITE) + return MAX_SCHEDULE_TIMEOUT; + + /* + * msecs_to_jiffies interprets all values above 2^31-1 as infinite, + * but we consider them finite. + * This hack is wrong, but nobody is likely to notice. + */ + user_timeout_ms = min_t(uint32_t, user_timeout_ms, 0x7FFFFFFF); + + return msecs_to_jiffies(user_timeout_ms) + 1; +} + +static void free_waiters(uint32_t num_events, struct kfd_event_waiter *waiters) +{ + uint32_t i; + + for (i = 0; i < num_events; i++) + list_del(&waiters[i].waiters); + + kfree(waiters); +} + +int kfd_wait_on_events(struct kfd_process *p, + uint32_t num_events, void __user *data, + bool all, uint32_t user_timeout_ms, + enum kfd_event_wait_result *wait_result) +{ + struct kfd_event_data __user *events = + (struct kfd_event_data __user *) data; + uint32_t i; + int ret = 0; + struct kfd_event_waiter *event_waiters = NULL; + long timeout = user_timeout_to_jiffies(user_timeout_ms); + + mutex_lock(&p->event_mutex); + + event_waiters = alloc_event_waiters(num_events); + if (!event_waiters) { + ret = -ENOMEM; + goto fail; + } + + for (i = 0; i < num_events; i++) { + struct kfd_event_data event_data; + + if (copy_from_user(&event_data, &events[i], + sizeof(struct kfd_event_data))) + goto fail; + + ret = init_event_waiter(p, &event_waiters[i], + event_data.event_id, i); + if (ret) + goto fail; + } + + mutex_unlock(&p->event_mutex); + + while (true) { + if (fatal_signal_pending(current)) { + ret = -EINTR; + break; + } + + if (signal_pending(current)) { + /* + * This is wrong when a nonzero, non-infinite timeout + * is specified. We need to use + * ERESTARTSYS_RESTARTBLOCK, but struct restart_block + * contains a union with data for each user and it's + * in generic kernel code that I don't want to + * touch yet. + */ + ret = -ERESTARTSYS; + break; + } + + if (test_event_condition(all, num_events, event_waiters)) { + if (copy_signaled_event_data(num_events, + event_waiters, events)) + *wait_result = KFD_WAIT_COMPLETE; + else + *wait_result = KFD_WAIT_ERROR; + break; + } + + if (timeout <= 0) { + *wait_result = KFD_WAIT_TIMEOUT; + break; + } + + timeout = schedule_timeout_interruptible(timeout); + } + __set_current_state(TASK_RUNNING); + + mutex_lock(&p->event_mutex); + free_waiters(num_events, event_waiters); + mutex_unlock(&p->event_mutex); + + return ret; + +fail: + if (event_waiters) + free_waiters(num_events, event_waiters); + + mutex_unlock(&p->event_mutex); + + *wait_result = KFD_WAIT_ERROR; + + return ret; +} + +int kfd_event_mmap(struct kfd_process *p, struct vm_area_struct *vma) +{ + + unsigned int page_index; + unsigned long pfn; + struct signal_page *page; + + /* check required size is logical */ + if (get_order(KFD_SIGNAL_EVENT_LIMIT * 8) != + get_order(vma->vm_end - vma->vm_start)) { + pr_err("amdkfd: event page mmap requested illegal size\n"); + return -EINVAL; + } + + page_index = vma->vm_pgoff; + + page = lookup_signal_page_by_index(p, page_index); + if (!page) { + /* Probably KFD bug, but mmap is user-accessible. */ + pr_debug("signal page could not be found for page_index %u\n", + page_index); + return -EINVAL; + } + + pfn = __pa(page->kernel_address); + pfn >>= PAGE_SHIFT; + + vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_NORESERVE + | VM_DONTDUMP | VM_PFNMAP; + + pr_debug("mapping signal page\n"); + pr_debug(" start user address == 0x%08lx\n", vma->vm_start); + pr_debug(" end user address == 0x%08lx\n", vma->vm_end); + pr_debug(" pfn == 0x%016lX\n", pfn); + pr_debug(" vm_flags == 0x%08lX\n", vma->vm_flags); + pr_debug(" size == 0x%08lX\n", + vma->vm_end - vma->vm_start); + + page->user_address = (uint64_t __user *)vma->vm_start; + + /* mapping the page to user process */ + return remap_pfn_range(vma, vma->vm_start, pfn, + vma->vm_end - vma->vm_start, vma->vm_page_prot); +} + +/* + * Assumes that p->event_mutex is held and of course + * that p is not going away (current or locked). + */ +static void lookup_events_by_type_and_signal(struct kfd_process *p, + int type, void *event_data) +{ + struct kfd_hsa_memory_exception_data *ev_data; + struct kfd_event *ev; + int bkt; + bool send_signal = true; + + ev_data = (struct kfd_hsa_memory_exception_data *) event_data; + + hash_for_each(p->events, bkt, ev, events) + if (ev->type == type) { + send_signal = false; + dev_dbg(kfd_device, + "Event found: id %X type %d", + ev->event_id, ev->type); + set_event(ev); + if (ev->type == KFD_EVENT_TYPE_MEMORY && ev_data) + ev->memory_exception_data = *ev_data; + } + + /* Send SIGTERM no event of type "type" has been found*/ + if (send_signal) { + if (send_sigterm) { + dev_warn(kfd_device, + "Sending SIGTERM to HSA Process with PID %d ", + p->lead_thread->pid); + send_sig(SIGTERM, p->lead_thread, 0); + } else { + dev_err(kfd_device, + "HSA Process (PID %d) got unhandled exception", + p->lead_thread->pid); + } + } +} + +void kfd_signal_iommu_event(struct kfd_dev *dev, unsigned int pasid, + unsigned long address, bool is_write_requested, + bool is_execute_requested) +{ + struct kfd_hsa_memory_exception_data memory_exception_data; + struct vm_area_struct *vma; + + /* + * Because we are called from arbitrary context (workqueue) as opposed + * to process context, kfd_process could attempt to exit while we are + * running so the lookup function returns a locked process. + */ + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; /* Presumably process exited. */ + + memset(&memory_exception_data, 0, sizeof(memory_exception_data)); + + down_read(&p->mm->mmap_sem); + vma = find_vma(p->mm, address); + + memory_exception_data.gpu_id = dev->id; + memory_exception_data.va = address; + /* Set failure reason */ + memory_exception_data.failure.NotPresent = 1; + memory_exception_data.failure.NoExecute = 0; + memory_exception_data.failure.ReadOnly = 0; + if (vma) { + if (vma->vm_start > address) { + memory_exception_data.failure.NotPresent = 1; + memory_exception_data.failure.NoExecute = 0; + memory_exception_data.failure.ReadOnly = 0; + } else { + memory_exception_data.failure.NotPresent = 0; + if (is_write_requested && !(vma->vm_flags & VM_WRITE)) + memory_exception_data.failure.ReadOnly = 1; + else + memory_exception_data.failure.ReadOnly = 0; + if (is_execute_requested && !(vma->vm_flags & VM_EXEC)) + memory_exception_data.failure.NoExecute = 1; + else + memory_exception_data.failure.NoExecute = 0; + } + } + + up_read(&p->mm->mmap_sem); + + mutex_lock(&p->event_mutex); + + /* Lookup events by type and signal them */ + lookup_events_by_type_and_signal(p, KFD_EVENT_TYPE_MEMORY, + &memory_exception_data); + + mutex_unlock(&p->event_mutex); + mutex_unlock(&p->mutex); +} + +void kfd_signal_hw_exception_event(unsigned int pasid) +{ + /* + * Because we are called from arbitrary context (workqueue) as opposed + * to process context, kfd_process could attempt to exit while we are + * running so the lookup function returns a locked process. + */ + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; /* Presumably process exited. */ + + mutex_lock(&p->event_mutex); + + /* Lookup events by type and signal them */ + lookup_events_by_type_and_signal(p, KFD_EVENT_TYPE_HW_EXCEPTION, NULL); + + mutex_unlock(&p->event_mutex); + mutex_unlock(&p->mutex); +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.h b/drivers/gpu/drm/amd/amdkfd/kfd_events.h new file mode 100644 index 000000000000..28f6838b1f4c --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.h @@ -0,0 +1,84 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +#ifndef KFD_EVENTS_H_INCLUDED +#define KFD_EVENTS_H_INCLUDED + +#include <linux/kernel.h> +#include <linux/hashtable.h> +#include <linux/types.h> +#include <linux/list.h> +#include "kfd_priv.h" +#include <uapi/linux/kfd_ioctl.h> + +#define KFD_EVENT_ID_NONSIGNAL_MASK 0x80000000U +#define KFD_FIRST_NONSIGNAL_EVENT_ID KFD_EVENT_ID_NONSIGNAL_MASK +#define KFD_LAST_NONSIGNAL_EVENT_ID UINT_MAX + +/* + * Written into kfd_signal_slot_t to indicate that the event is not signaled. + * Since the event protocol may need to write the event ID into memory, this + * must not be a valid event ID. + * For the sake of easy memset-ing, this must be a byte pattern. + */ +#define UNSIGNALED_EVENT_SLOT ((uint64_t)-1) + +struct kfd_event_waiter; +struct signal_page; + +struct kfd_event { + /* All events in process, rooted at kfd_process.events. */ + struct hlist_node events; + + u32 event_id; + + bool signaled; + bool auto_reset; + + int type; + + struct list_head waiters; /* List of kfd_event_waiter by waiters. */ + + /* Only for signal events. */ + struct signal_page *signal_page; + unsigned int signal_slot_index; + uint64_t __user *user_signal_address; + + /* type specific data */ + union { + struct kfd_hsa_memory_exception_data memory_exception_data; + }; +}; + +#define KFD_EVENT_TIMEOUT_IMMEDIATE 0 +#define KFD_EVENT_TIMEOUT_INFINITE 0xFFFFFFFFu + +/* Matching HSA_EVENTTYPE */ +#define KFD_EVENT_TYPE_SIGNAL 0 +#define KFD_EVENT_TYPE_HW_EXCEPTION 3 +#define KFD_EVENT_TYPE_DEBUG 5 +#define KFD_EVENT_TYPE_MEMORY 8 + +extern void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, + uint32_t valid_id_bits); + +#endif diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c new file mode 100644 index 000000000000..7f134aa9bfd3 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -0,0 +1,188 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +/* + * KFD Interrupts. + * + * AMD GPUs deliver interrupts by pushing an interrupt description onto the + * interrupt ring and then sending an interrupt. KGD receives the interrupt + * in ISR and sends us a pointer to each new entry on the interrupt ring. + * + * We generally can't process interrupt-signaled events from ISR, so we call + * out to each interrupt client module (currently only the scheduler) to ask if + * each interrupt is interesting. If they return true, then it requires further + * processing so we copy it to an internal interrupt ring and call each + * interrupt client again from a work-queue. + * + * There's no acknowledgment for the interrupts we use. The hardware simply + * queues a new interrupt each time without waiting. + * + * The fixed-size internal queue means that it's possible for us to lose + * interrupts because we have no back-pressure to the hardware. + */ + +#include <linux/slab.h> +#include <linux/device.h> +#include "kfd_priv.h" + +#define KFD_INTERRUPT_RING_SIZE 1024 + +static void interrupt_wq(struct work_struct *); + +int kfd_interrupt_init(struct kfd_dev *kfd) +{ + void *interrupt_ring = kmalloc_array(KFD_INTERRUPT_RING_SIZE, + kfd->device_info->ih_ring_entry_size, + GFP_KERNEL); + if (!interrupt_ring) + return -ENOMEM; + + kfd->interrupt_ring = interrupt_ring; + kfd->interrupt_ring_size = + KFD_INTERRUPT_RING_SIZE * kfd->device_info->ih_ring_entry_size; + atomic_set(&kfd->interrupt_ring_wptr, 0); + atomic_set(&kfd->interrupt_ring_rptr, 0); + + spin_lock_init(&kfd->interrupt_lock); + + INIT_WORK(&kfd->interrupt_work, interrupt_wq); + + kfd->interrupts_active = true; + + /* + * After this function returns, the interrupt will be enabled. This + * barrier ensures that the interrupt running on a different processor + * sees all the above writes. + */ + smp_wmb(); + + return 0; +} + +void kfd_interrupt_exit(struct kfd_dev *kfd) +{ + /* + * Stop the interrupt handler from writing to the ring and scheduling + * workqueue items. The spinlock ensures that any interrupt running + * after we have unlocked sees interrupts_active = false. + */ + unsigned long flags; + + spin_lock_irqsave(&kfd->interrupt_lock, flags); + kfd->interrupts_active = false; + spin_unlock_irqrestore(&kfd->interrupt_lock, flags); + + /* + * Flush_scheduled_work ensures that there are no outstanding + * work-queue items that will access interrupt_ring. New work items + * can't be created because we stopped interrupt handling above. + */ + flush_scheduled_work(); + + kfree(kfd->interrupt_ring); +} + +/* + * This assumes that it can't be called concurrently with itself + * but only with dequeue_ih_ring_entry. + */ +bool enqueue_ih_ring_entry(struct kfd_dev *kfd, const void *ih_ring_entry) +{ + unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr); + unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr); + + if ((rptr - wptr) % kfd->interrupt_ring_size == + kfd->device_info->ih_ring_entry_size) { + /* This is very bad, the system is likely to hang. */ + dev_err_ratelimited(kfd_chardev(), + "Interrupt ring overflow, dropping interrupt.\n"); + return false; + } + + memcpy(kfd->interrupt_ring + wptr, ih_ring_entry, + kfd->device_info->ih_ring_entry_size); + + wptr = (wptr + kfd->device_info->ih_ring_entry_size) % + kfd->interrupt_ring_size; + smp_wmb(); /* Ensure memcpy'd data is visible before wptr update. */ + atomic_set(&kfd->interrupt_ring_wptr, wptr); + + return true; +} + +/* + * This assumes that it can't be called concurrently with itself + * but only with enqueue_ih_ring_entry. + */ +static bool dequeue_ih_ring_entry(struct kfd_dev *kfd, void *ih_ring_entry) +{ + /* + * Assume that wait queues have an implicit barrier, i.e. anything that + * happened in the ISR before it queued work is visible. + */ + + unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr); + unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr); + + if (rptr == wptr) + return false; + + memcpy(ih_ring_entry, kfd->interrupt_ring + rptr, + kfd->device_info->ih_ring_entry_size); + + rptr = (rptr + kfd->device_info->ih_ring_entry_size) % + kfd->interrupt_ring_size; + + /* + * Ensure the rptr write update is not visible until + * memcpy has finished reading. + */ + smp_mb(); + atomic_set(&kfd->interrupt_ring_rptr, rptr); + + return true; +} + +static void interrupt_wq(struct work_struct *work) +{ + struct kfd_dev *dev = container_of(work, struct kfd_dev, + interrupt_work); + + uint32_t ih_ring_entry[DIV_ROUND_UP( + dev->device_info->ih_ring_entry_size, + sizeof(uint32_t))]; + + while (dequeue_ih_ring_entry(dev, ih_ring_entry)) + dev->device_info->event_interrupt_class->interrupt_wq(dev, + ih_ring_entry); +} + +bool interrupt_is_wanted(struct kfd_dev *dev, const uint32_t *ih_ring_entry) +{ + /* integer and bitwise OR so there is no boolean short-circuiting */ + unsigned wanted = 0; + + wanted |= dev->device_info->event_interrupt_class->interrupt_isr(dev, + ih_ring_entry); + + return wanted != 0; +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c index c7d298e62c96..8fa894100290 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c @@ -215,8 +215,9 @@ static int acquire_packet_buffer(struct kernel_queue *kq, queue_address = (unsigned int *)kq->pq_kernel_addr; queue_size_dwords = kq->queue->properties.queue_size / sizeof(uint32_t); - pr_debug("amdkfd: In func %s\nrptr: %d\nwptr: %d\nqueue_address 0x%p\n", - __func__, rptr, wptr, queue_address); + pr_debug("rptr: %d\n", rptr); + pr_debug("wptr: %d\n", wptr); + pr_debug("queue_address 0x%p\n", queue_address); available_size = (rptr - 1 - wptr + queue_size_dwords) % queue_size_dwords; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_module.c b/drivers/gpu/drm/amd/amdkfd/kfd_module.c index 4e0a68f13a77..ca8410e8683d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_module.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_module.c @@ -29,10 +29,10 @@ #define KFD_DRIVER_AUTHOR "AMD Inc. and others" #define KFD_DRIVER_DESC "Standalone HSA driver for AMD's GPUs" -#define KFD_DRIVER_DATE "20150122" +#define KFD_DRIVER_DATE "20150421" #define KFD_DRIVER_MAJOR 0 #define KFD_DRIVER_MINOR 7 -#define KFD_DRIVER_PATCHLEVEL 1 +#define KFD_DRIVER_PATCHLEVEL 2 static const struct kgd2kfd_calls kgd2kfd = { .exit = kgd2kfd_exit, @@ -54,6 +54,11 @@ module_param(max_num_of_queues_per_device, int, 0444); MODULE_PARM_DESC(max_num_of_queues_per_device, "Maximum number of supported queues per device (1 = Minimum, 4096 = default)"); +int send_sigterm; +module_param(send_sigterm, int, 0444); +MODULE_PARM_DESC(send_sigterm, + "Send sigterm to HSA process on unhandled exception (0 = disable, 1 = enable)"); + bool kgd2kfd_init(unsigned interface_version, const struct kgd2kfd_calls **g2f) { /* diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index f21fccebd75b..b6f838f56589 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -35,6 +35,9 @@ #define KFD_SYSFS_FILE_MODE 0444 +#define KFD_MMAP_DOORBELL_MASK 0x8000000000000 +#define KFD_MMAP_EVENTS_MASK 0x4000000000000 + /* * When working with cp scheduler we should assign the HIQ manually or via * the radeon driver to a fixed hqd slot, here are the fixed HIQ hqd slot @@ -71,6 +74,12 @@ extern int max_num_of_queues_per_device; /* Kernel module parameter to specify the scheduling policy */ extern int sched_policy; +/* + * Kernel module parameter to specify whether to send sigterm to HSA process on + * unhandled exception + */ +extern int send_sigterm; + /** * enum kfd_sched_policy * @@ -108,8 +117,16 @@ enum asic_family_type { CHIP_CARRIZO }; +struct kfd_event_interrupt_class { + bool (*interrupt_isr)(struct kfd_dev *dev, + const uint32_t *ih_ring_entry); + void (*interrupt_wq)(struct kfd_dev *dev, + const uint32_t *ih_ring_entry); +}; + struct kfd_device_info { unsigned int asic_family; + const struct kfd_event_interrupt_class *event_interrupt_class; unsigned int max_pasid_bits; size_t ih_ring_entry_size; uint8_t num_of_watch_points; @@ -161,10 +178,23 @@ struct kfd_dev { unsigned int gtt_sa_chunk_size; unsigned int gtt_sa_num_of_chunks; + /* Interrupts */ + void *interrupt_ring; + size_t interrupt_ring_size; + atomic_t interrupt_ring_rptr; + atomic_t interrupt_ring_wptr; + struct work_struct interrupt_work; + spinlock_t interrupt_lock; + /* QCM Device instance */ struct device_queue_manager *dqm; bool init_complete; + /* + * Interrupts of interest to KFD are copied + * from the HW ring into a SW ring. + */ + bool interrupts_active; }; /* KGD2KFD callbacks */ @@ -477,6 +507,15 @@ struct kfd_process { /*Is the user space process 32 bit?*/ bool is_32bit_user_mode; + + /* Event-related data */ + struct mutex event_mutex; + /* All events in process hashed by ID, linked on kfd_event.events. */ + DECLARE_HASHTABLE(events, 4); + struct list_head signal_event_pages; /* struct slot_page_header. + event_pages */ + u32 next_nonsignal_event_id; + size_t signal_event_count; }; /** @@ -501,6 +540,7 @@ void kfd_process_create_wq(void); void kfd_process_destroy_wq(void); struct kfd_process *kfd_create_process(const struct task_struct *); struct kfd_process *kfd_get_process(const struct task_struct *); +struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid); struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev, struct kfd_process *p); @@ -555,7 +595,11 @@ struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev); struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx); /* Interrupts */ +int kfd_interrupt_init(struct kfd_dev *dev); +void kfd_interrupt_exit(struct kfd_dev *dev); void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry); +bool enqueue_ih_ring_entry(struct kfd_dev *kfd, const void *ih_ring_entry); +bool interrupt_is_wanted(struct kfd_dev *dev, const uint32_t *ih_ring_entry); /* Power Management */ void kgd2kfd_suspend(struct kfd_dev *kfd); @@ -642,4 +686,35 @@ uint64_t kfd_get_number_elems(struct kfd_dev *kfd); phys_addr_t kfd_get_process_doorbells(struct kfd_dev *dev, struct kfd_process *process); +/* Events */ +extern const struct kfd_event_interrupt_class event_interrupt_class_cik; +extern const struct kfd_device_global_init_class device_global_init_class_cik; + +enum kfd_event_wait_result { + KFD_WAIT_COMPLETE, + KFD_WAIT_TIMEOUT, + KFD_WAIT_ERROR +}; + +void kfd_event_init_process(struct kfd_process *p); +void kfd_event_free_process(struct kfd_process *p); +int kfd_event_mmap(struct kfd_process *process, struct vm_area_struct *vma); +int kfd_wait_on_events(struct kfd_process *p, + uint32_t num_events, void __user *data, + bool all, uint32_t user_timeout_ms, + enum kfd_event_wait_result *wait_result); +void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, + uint32_t valid_id_bits); +void kfd_signal_iommu_event(struct kfd_dev *dev, + unsigned int pasid, unsigned long address, + bool is_write_requested, bool is_execute_requested); +void kfd_signal_hw_exception_event(unsigned int pasid); +int kfd_set_event(struct kfd_process *p, uint32_t event_id); +int kfd_reset_event(struct kfd_process *p, uint32_t event_id); +int kfd_event_create(struct file *devkfd, struct kfd_process *p, + uint32_t event_type, bool auto_reset, uint32_t node_id, + uint32_t *event_id, uint32_t *event_trigger_data, + uint64_t *event_page_offset, uint32_t *event_slot_index); +int kfd_event_destroy(struct kfd_process *p, uint32_t event_id); + #endif diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 945d6226dc51..dc910af2bb3c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -178,6 +178,8 @@ static void kfd_process_wq_release(struct work_struct *work) kfree(pdd); } + kfd_event_free_process(p); + kfd_pasid_free(p->pasid); mutex_unlock(&p->mutex); @@ -203,8 +205,7 @@ static void kfd_process_destroy_delayed(struct rcu_head *rcu) mmdrop(p->mm); - work = (struct kfd_process_release_work *) - kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC); + work = kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC); if (work) { INIT_WORK((struct work_struct *) work, kfd_process_wq_release); @@ -289,6 +290,8 @@ static struct kfd_process *create_process(const struct task_struct *thread) INIT_LIST_HEAD(&process->per_device_data); + kfd_event_init_process(process); + err = pqm_init(&process->pqm, process); if (err != 0) goto err_process_pqm_init; @@ -431,3 +434,23 @@ bool kfd_has_process_device_data(struct kfd_process *p) { return !(list_empty(&p->per_device_data)); } + +/* This returns with process->mutex locked. */ +struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid) +{ + struct kfd_process *p; + unsigned int temp; + + int idx = srcu_read_lock(&kfd_processes_srcu); + + hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) { + if (p->pasid == pasid) { + mutex_lock(&p->mutex); + break; + } + } + + srcu_read_unlock(&kfd_processes_srcu, idx); + + return p; +} diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h index dabd94446b7b..4ea21ae88b07 100644 --- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h +++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h @@ -144,6 +144,8 @@ struct kfd2kgd_calls { int (*init_pipeline)(struct kgd_dev *kgd, uint32_t pipe_id, uint32_t hpd_size, uint64_t hpd_gpu_addr); + int (*init_interrupts)(struct kgd_dev *kgd, uint32_t pipe_id); + int (*hqd_load)(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, uint32_t queue_id, uint32_t __user *wptr); diff --git a/drivers/gpu/drm/radeon/cik_reg.h b/drivers/gpu/drm/radeon/cik_reg.h index f667347d8157..0ec5d53eb6fb 100644 --- a/drivers/gpu/drm/radeon/cik_reg.h +++ b/drivers/gpu/drm/radeon/cik_reg.h @@ -149,6 +149,8 @@ #define KFD_CIK_SDMA_QUEUE_OFFSET 0x200 +#define CPC_INT_CNTL 0xC2D0 + #define CP_HQD_IQ_RPTR 0xC970u #define AQL_ENABLE (1U << 0) #define SDMA0_RLC0_RB_CNTL 0xD400u diff --git a/drivers/gpu/drm/radeon/cikd.h b/drivers/gpu/drm/radeon/cikd.h index 0089d837a8e3..b33ba3b0808b 100644 --- a/drivers/gpu/drm/radeon/cikd.h +++ b/drivers/gpu/drm/radeon/cikd.h @@ -1335,6 +1335,7 @@ # define CNTX_EMPTY_INT_ENABLE (1 << 20) # define PRIV_INSTR_INT_ENABLE (1 << 22) # define PRIV_REG_INT_ENABLE (1 << 23) +# define OPCODE_ERROR_INT_ENABLE (1 << 24) # define TIME_STAMP_INT_ENABLE (1 << 26) # define CP_RINGID2_INT_ENABLE (1 << 29) # define CP_RINGID1_INT_ENABLE (1 << 30) diff --git a/drivers/gpu/drm/radeon/radeon_kfd.c b/drivers/gpu/drm/radeon/radeon_kfd.c index 3db23007cdf4..813a416db538 100644 --- a/drivers/gpu/drm/radeon/radeon_kfd.c +++ b/drivers/gpu/drm/radeon/radeon_kfd.c @@ -66,7 +66,7 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id, uint32_t hpd_size, uint64_t hpd_gpu_addr); - +static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id); static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, uint32_t queue_id, uint32_t __user *wptr); static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd); @@ -89,6 +89,7 @@ static const struct kfd2kgd_calls kfd2kgd = { .program_sh_mem_settings = kgd_program_sh_mem_settings, .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, .init_pipeline = kgd_init_pipeline, + .init_interrupts = kgd_init_interrupts, .hqd_load = kgd_hqd_load, .hqd_sdma_load = kgd_hqd_sdma_load, .hqd_is_occupied = kgd_hqd_is_occupied, @@ -407,6 +408,24 @@ static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id, return 0; } +static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) +{ + uint32_t mec; + uint32_t pipe; + + mec = (pipe_id / CIK_PIPE_PER_MEC) + 1; + pipe = (pipe_id % CIK_PIPE_PER_MEC); + + lock_srbm(kgd, mec, pipe, 0, 0); + + write_register(kgd, CPC_INT_CNTL, + TIME_STAMP_INT_ENABLE | OPCODE_ERROR_INT_ENABLE); + + unlock_srbm(kgd); + + return 0; +} + static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m) { uint32_t retval; diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h index af94f31e33ac..4ca35a8f9891 100644 --- a/include/uapi/linux/kfd_ioctl.h +++ b/include/uapi/linux/kfd_ioctl.h @@ -27,7 +27,7 @@ #include <linux/ioctl.h> #define KFD_IOCTL_MAJOR_VERSION 1 -#define KFD_IOCTL_MINOR_VERSION 0 +#define KFD_IOCTL_MINOR_VERSION 1 struct kfd_ioctl_get_version_args { uint32_t major_version; /* from KFD */ @@ -128,6 +128,83 @@ struct kfd_ioctl_get_process_apertures_args { uint32_t pad; }; +/* Matching HSA_EVENTTYPE */ +#define KFD_IOC_EVENT_SIGNAL 0 +#define KFD_IOC_EVENT_NODECHANGE 1 +#define KFD_IOC_EVENT_DEVICESTATECHANGE 2 +#define KFD_IOC_EVENT_HW_EXCEPTION 3 +#define KFD_IOC_EVENT_SYSTEM_EVENT 4 +#define KFD_IOC_EVENT_DEBUG_EVENT 5 +#define KFD_IOC_EVENT_PROFILE_EVENT 6 +#define KFD_IOC_EVENT_QUEUE_EVENT 7 +#define KFD_IOC_EVENT_MEMORY 8 + +#define KFD_IOC_WAIT_RESULT_COMPLETE 0 +#define KFD_IOC_WAIT_RESULT_TIMEOUT 1 +#define KFD_IOC_WAIT_RESULT_FAIL 2 + +#define KFD_SIGNAL_EVENT_LIMIT 256 + +struct kfd_ioctl_create_event_args { + uint64_t event_page_offset; /* from KFD */ + uint32_t event_trigger_data; /* from KFD - signal events only */ + uint32_t event_type; /* to KFD */ + uint32_t auto_reset; /* to KFD */ + uint32_t node_id; /* to KFD - only valid for certain + event types */ + uint32_t event_id; /* from KFD */ + uint32_t event_slot_index; /* from KFD */ +}; + +struct kfd_ioctl_destroy_event_args { + uint32_t event_id; /* to KFD */ + uint32_t pad; +}; + +struct kfd_ioctl_set_event_args { + uint32_t event_id; /* to KFD */ + uint32_t pad; +}; + +struct kfd_ioctl_reset_event_args { + uint32_t event_id; /* to KFD */ + uint32_t pad; +}; + +struct kfd_memory_exception_failure { + uint32_t NotPresent; /* Page not present or supervisor privilege */ + uint32_t ReadOnly; /* Write access to a read-only page */ + uint32_t NoExecute; /* Execute access to a page marked NX */ + uint32_t pad; +}; + +/* memory exception data*/ +struct kfd_hsa_memory_exception_data { + struct kfd_memory_exception_failure failure; + uint64_t va; + uint32_t gpu_id; + uint32_t pad; +}; + +/* Event data*/ +struct kfd_event_data { + union { + struct kfd_hsa_memory_exception_data memory_exception_data; + }; /* From KFD */ + uint64_t kfd_event_data_ext; /* pointer to an extension structure + for future exception types */ + uint32_t event_id; /* to KFD */ + uint32_t pad; +}; + +struct kfd_ioctl_wait_events_args { + uint64_t events_ptr; /* to KFD */ + uint32_t num_events; /* to KFD */ + uint32_t wait_for_all; /* to KFD */ + uint32_t timeout; /* to KFD */ + uint32_t wait_result; /* from KFD */ +}; + #define AMDKFD_IOCTL_BASE 'K' #define AMDKFD_IO(nr) _IO(AMDKFD_IOCTL_BASE, nr) #define AMDKFD_IOR(nr, type) _IOR(AMDKFD_IOCTL_BASE, nr, type) @@ -155,7 +232,22 @@ struct kfd_ioctl_get_process_apertures_args { #define AMDKFD_IOC_UPDATE_QUEUE \ AMDKFD_IOW(0x07, struct kfd_ioctl_update_queue_args) +#define AMDKFD_IOC_CREATE_EVENT \ + AMDKFD_IOWR(0x08, struct kfd_ioctl_create_event_args) + +#define AMDKFD_IOC_DESTROY_EVENT \ + AMDKFD_IOW(0x09, struct kfd_ioctl_destroy_event_args) + +#define AMDKFD_IOC_SET_EVENT \ + AMDKFD_IOW(0x0A, struct kfd_ioctl_set_event_args) + +#define AMDKFD_IOC_RESET_EVENT \ + AMDKFD_IOW(0x0B, struct kfd_ioctl_reset_event_args) + +#define AMDKFD_IOC_WAIT_EVENTS \ + AMDKFD_IOWR(0x0C, struct kfd_ioctl_wait_events_args) + #define AMDKFD_COMMAND_START 0x01 -#define AMDKFD_COMMAND_END 0x08 +#define AMDKFD_COMMAND_END 0x0D #endif |