aboutsummaryrefslogtreecommitdiff
path: root/arch/arc/include
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2019-12-04 19:06:18 -0800
committerLinus Torvalds <torvalds@linux-foundation.org>2019-12-04 19:06:18 -0800
commit056df578c2dcac1e624254567f5df5ddaa223234 (patch)
treea19197bded5e2d2ee6219d502d6e9a3e42dd1e6f /arch/arc/include
parentaedc0650f9135f3b92b39cbed1a8fe98d8088825 (diff)
parent9fbea0b7e842890a76acffce9be9e430b9e11194 (diff)
Merge tag 'arc-5.5-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
Pull ARC updates from Vineet Gupta - Jump Label support for ARC - kmemleak enabled - arc mm backend TLB Miss / flush optimizations - nSIM platform switching to dwuart (vs. arcuart) and ensuing defconfig updates and cleanups - axs platform pll / video-mode updates * tag 'arc-5.5-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc: ARC: add kmemleak support ARC: [plat-axs10x]: remove hardcoded video mode from bootargs ARC: [plat-axs10x]: use pgu pll instead of fixed clock ARC: ARCv2: jump label: implement jump label patching ARC: mm: tlb flush optim: elide redundant uTLB invalidates for MMUv3 ARC: mm: tlb flush optim: elide repeated uTLB invalidate in loop ARC: mm: tlb flush optim: Make TLBWriteNI fallback to TLBWrite if not available ARC: mm: TLB Miss optim: avoid re-reading ECR ARCv2: mm: TLB Miss optim: Use double world load/stores LDD/STD ARCv2: mm: TLB Miss optim: SMP builds can cache pgd pointer in mmu scratch reg ARC: nSIM_700: remove unused network options ARC: nSIM_700: switch to DW UART usage ARC: merge HAPS-HS with nSIM-HS configs ARC: HAPS: cleanup defconfigs from unused ETH drivers ARC: HAPS: add HIGHMEM memory zone to DTS ARC: HAPS: use same UART configuration everywhere ARC: HAPS: cleanup defconfigs from unused IO-related options ARC: regenerate nSIM and HAPS defconfigs
Diffstat (limited to 'arch/arc/include')
-rw-r--r--arch/arc/include/asm/cache.h2
-rw-r--r--arch/arc/include/asm/entry-compact.h4
-rw-r--r--arch/arc/include/asm/jump_label.h72
-rw-r--r--arch/arc/include/asm/mmu.h6
-rw-r--r--arch/arc/include/asm/mmu_context.h2
-rw-r--r--arch/arc/include/asm/pgtable.h2
6 files changed, 84 insertions, 4 deletions
diff --git a/arch/arc/include/asm/cache.h b/arch/arc/include/asm/cache.h
index 918804c7c1a4..d8ece4292388 100644
--- a/arch/arc/include/asm/cache.h
+++ b/arch/arc/include/asm/cache.h
@@ -25,6 +25,8 @@
#ifndef __ASSEMBLY__
+#include <linux/build_bug.h>
+
/* Uncached access macros */
#define arc_read_uncached_32(ptr) \
({ \
diff --git a/arch/arc/include/asm/entry-compact.h b/arch/arc/include/asm/entry-compact.h
index 66a292335ee6..c3aa775878dc 100644
--- a/arch/arc/include/asm/entry-compact.h
+++ b/arch/arc/include/asm/entry-compact.h
@@ -130,7 +130,7 @@
* to be saved again on kernel mode stack, as part of pt_regs.
*-------------------------------------------------------------*/
.macro PROLOG_FREEUP_REG reg, mem
-#ifdef CONFIG_SMP
+#ifndef ARC_USE_SCRATCH_REG
sr \reg, [ARC_REG_SCRATCH_DATA0]
#else
st \reg, [\mem]
@@ -138,7 +138,7 @@
.endm
.macro PROLOG_RESTORE_REG reg, mem
-#ifdef CONFIG_SMP
+#ifndef ARC_USE_SCRATCH_REG
lr \reg, [ARC_REG_SCRATCH_DATA0]
#else
ld \reg, [\mem]
diff --git a/arch/arc/include/asm/jump_label.h b/arch/arc/include/asm/jump_label.h
new file mode 100644
index 000000000000..9d9618079739
--- /dev/null
+++ b/arch/arc/include/asm/jump_label.h
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_ARC_JUMP_LABEL_H
+#define _ASM_ARC_JUMP_LABEL_H
+
+#ifndef __ASSEMBLY__
+
+#include <linux/stringify.h>
+#include <linux/types.h>
+
+#define JUMP_LABEL_NOP_SIZE 4
+
+/*
+ * NOTE about '.balign 4':
+ *
+ * To make atomic update of patched instruction available we need to guarantee
+ * that this instruction doesn't cross L1 cache line boundary.
+ *
+ * As of today we simply align instruction which can be patched by 4 byte using
+ * ".balign 4" directive. In that case patched instruction is aligned with one
+ * 16-bit NOP_S if this is required.
+ * However 'align by 4' directive is much stricter than it actually required.
+ * It's enough that our 32-bit instruction don't cross L1 cache line boundary /
+ * L1 I$ fetch block boundary which can be achieved by using
+ * ".bundle_align_mode" assembler directive. That will save us from adding
+ * useless NOP_S padding in most of the cases.
+ *
+ * TODO: switch to ".bundle_align_mode" directive using whin it will be
+ * supported by ARC toolchain.
+ */
+
+static __always_inline bool arch_static_branch(struct static_key *key,
+ bool branch)
+{
+ asm_volatile_goto(".balign "__stringify(JUMP_LABEL_NOP_SIZE)" \n"
+ "1: \n"
+ "nop \n"
+ ".pushsection __jump_table, \"aw\" \n"
+ ".word 1b, %l[l_yes], %c0 \n"
+ ".popsection \n"
+ : : "i" (&((char *)key)[branch]) : : l_yes);
+
+ return false;
+l_yes:
+ return true;
+}
+
+static __always_inline bool arch_static_branch_jump(struct static_key *key,
+ bool branch)
+{
+ asm_volatile_goto(".balign "__stringify(JUMP_LABEL_NOP_SIZE)" \n"
+ "1: \n"
+ "b %l[l_yes] \n"
+ ".pushsection __jump_table, \"aw\" \n"
+ ".word 1b, %l[l_yes], %c0 \n"
+ ".popsection \n"
+ : : "i" (&((char *)key)[branch]) : : l_yes);
+
+ return false;
+l_yes:
+ return true;
+}
+
+typedef u32 jump_label_t;
+
+struct jump_entry {
+ jump_label_t code;
+ jump_label_t target;
+ jump_label_t key;
+};
+
+#endif /* __ASSEMBLY__ */
+#endif
diff --git a/arch/arc/include/asm/mmu.h b/arch/arc/include/asm/mmu.h
index 98cadf1a09ac..26b731d32a2b 100644
--- a/arch/arc/include/asm/mmu.h
+++ b/arch/arc/include/asm/mmu.h
@@ -40,6 +40,10 @@
#define ARC_REG_SCRATCH_DATA0 0x46c
#endif
+#if defined(CONFIG_ISA_ARCV2) || !defined(CONFIG_SMP)
+#define ARC_USE_SCRATCH_REG
+#endif
+
/* Bits in MMU PID register */
#define __TLB_ENABLE (1 << 31)
#define __PROG_ENABLE (1 << 30)
@@ -63,6 +67,8 @@
#if (CONFIG_ARC_MMU_VER >= 2)
#define TLBWriteNI 0x5 /* write JTLB without inv uTLBs */
#define TLBIVUTLB 0x6 /* explicitly inv uTLBs */
+#else
+#define TLBWriteNI TLBWrite /* Not present in hardware, fallback */
#endif
#if (CONFIG_ARC_MMU_VER >= 4)
diff --git a/arch/arc/include/asm/mmu_context.h b/arch/arc/include/asm/mmu_context.h
index 035470816be5..3a5e6a5b9ed6 100644
--- a/arch/arc/include/asm/mmu_context.h
+++ b/arch/arc/include/asm/mmu_context.h
@@ -144,7 +144,7 @@ static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
*/
cpumask_set_cpu(cpu, mm_cpumask(next));
-#ifndef CONFIG_SMP
+#ifdef ARC_USE_SCRATCH_REG
/* PGD cached in MMU reg to avoid 3 mem lookups: task->mm->pgd */
write_aux_reg(ARC_REG_SCRATCH_DATA0, next->pgd);
#endif
diff --git a/arch/arc/include/asm/pgtable.h b/arch/arc/include/asm/pgtable.h
index b917b596f7fb..9019ed9f9c94 100644
--- a/arch/arc/include/asm/pgtable.h
+++ b/arch/arc/include/asm/pgtable.h
@@ -350,7 +350,7 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
* Thus use this macro only when you are certain that "current" is current
* e.g. when dealing with signal frame setup code etc
*/
-#ifndef CONFIG_SMP
+#ifdef ARC_USE_SCRATCH_REG
#define pgd_offset_fast(mm, addr) \
({ \
pgd_t *pgd_base = (pgd_t *) read_aux_reg(ARC_REG_SCRATCH_DATA0); \