From aeec6cdad6cda5fdf26f937c0d15f101539d8185 Mon Sep 17 00:00:00 2001 From: Viresh Kumar Date: Thu, 16 Jul 2015 16:56:14 +0530 Subject: [PATCH 01/15] ARC/time: Migrate to new 'set-state' interface Migrate arc driver to the new 'set-state' interface provided by clockevents core, the earlier 'set-mode' interface is marked obsolete now. This also enables us to implement callbacks for new states of clockevent devices, for example: ONESHOT_STOPPED. Cc: Vineet Gupta Signed-off-by: Viresh Kumar Signed-off-by: Vineet Gupta --- arch/arc/kernel/time.c | 40 +++++++++++++++------------------------- 1 file changed, 15 insertions(+), 25 deletions(-) diff --git a/arch/arc/kernel/time.c b/arch/arc/kernel/time.c index 3364d2bbc515..4294761a2b3e 100644 --- a/arch/arc/kernel/time.c +++ b/arch/arc/kernel/time.c @@ -203,34 +203,24 @@ static int arc_clkevent_set_next_event(unsigned long delta, return 0; } -static void arc_clkevent_set_mode(enum clock_event_mode mode, - struct clock_event_device *dev) +static int arc_clkevent_set_periodic(struct clock_event_device *dev) { - switch (mode) { - case CLOCK_EVT_MODE_PERIODIC: - /* - * At X Hz, 1 sec = 1000ms -> X cycles; - * 10ms -> X / 100 cycles - */ - arc_timer_event_setup(arc_get_core_freq() / HZ); - break; - case CLOCK_EVT_MODE_ONESHOT: - break; - default: - break; - } - - return; + /* + * At X Hz, 1 sec = 1000ms -> X cycles; + * 10ms -> X / 100 cycles + */ + arc_timer_event_setup(arc_get_core_freq() / HZ); + return 0; } static DEFINE_PER_CPU(struct clock_event_device, arc_clockevent_device) = { - .name = "ARC Timer0", - .features = CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_PERIODIC, - .mode = CLOCK_EVT_MODE_UNUSED, - .rating = 300, - .irq = TIMER0_IRQ, /* hardwired, no need for resources */ - .set_next_event = arc_clkevent_set_next_event, - .set_mode = arc_clkevent_set_mode, + .name = "ARC Timer0", + .features = CLOCK_EVT_FEAT_ONESHOT | + CLOCK_EVT_FEAT_PERIODIC, + .rating = 300, + .irq = TIMER0_IRQ, /* hardwired, no need for resources */ + .set_next_event = arc_clkevent_set_next_event, + .set_state_periodic = arc_clkevent_set_periodic, }; static irqreturn_t timer_irq_handler(int irq, void *dev_id) @@ -240,7 +230,7 @@ static irqreturn_t timer_irq_handler(int irq, void *dev_id) * irq_set_chip_and_handler() asked for handle_percpu_devid_irq() */ struct clock_event_device *evt = this_cpu_ptr(&arc_clockevent_device); - int irq_reenable = evt->mode == CLOCK_EVT_MODE_PERIODIC; + int irq_reenable = clockevent_state_periodic(evt); /* * Any write to CTRL reg ACks the interrupt, we rewrite the From d05a76ab4d9ece9bd987a8c29e3f384ec9f7b211 Mon Sep 17 00:00:00 2001 From: Alexey Brodkin Date: Thu, 16 Jul 2015 21:45:38 +0300 Subject: [PATCH 02/15] ARCv2: add knob for DIV_REV in Kconfig Being highly configurable core ARC HS among other features might be configured with or without DIV_REM_OPTION (hardware divider). That option when enabled adds following instructions: div, divu, rem, remu. By default ARC HS38 has this option enabled. So we add here possibility to disable usage of hardware divider by compiler. Signed-off-by: Alexey Brodkin Signed-off-by: Vineet Gupta --- arch/arc/Kconfig | 4 ++++ arch/arc/Makefile | 10 +++++++++- 2 files changed, 13 insertions(+), 1 deletion(-) diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig index 91cf4055acab..f7cfa7d574f6 100644 --- a/arch/arc/Kconfig +++ b/arch/arc/Kconfig @@ -379,6 +379,10 @@ config ARC_HAS_LL64 dest operands with 2 possible source operands. default y +config ARC_HAS_DIV_REM + bool "Insn: div, divu, rem, remu" + default y + config ARC_HAS_RTC bool "Local 64-bit r/o cycle counter" default n diff --git a/arch/arc/Makefile b/arch/arc/Makefile index 46d87310220d..8a27a48304a4 100644 --- a/arch/arc/Makefile +++ b/arch/arc/Makefile @@ -36,8 +36,16 @@ cflags-$(atleast_gcc44) += -fsection-anchors cflags-$(CONFIG_ARC_HAS_LLSC) += -mlock cflags-$(CONFIG_ARC_HAS_SWAPE) += -mswape +ifdef CONFIG_ISA_ARCV2 + ifndef CONFIG_ARC_HAS_LL64 -cflags-$(CONFIG_ISA_ARCV2) += -mno-ll64 +cflags-y += -mno-ll64 +endif + +ifndef CONFIG_ARC_HAS_DIV_REM +cflags-y += -mno-div-rem +endif + endif cflags-$(CONFIG_ARC_DW2_UNWIND) += -fasynchronous-unwind-tables From 21481f2cfef9a79d3676916ef9424b1a7794776c Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 20 Jul 2015 17:19:17 +0300 Subject: [PATCH 03/15] ARCv2: lib: memcpy: Missing PREFETCHW Signed-off-by: Vineet Gupta --- arch/arc/lib/memcpy-archs.S | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/arch/arc/lib/memcpy-archs.S b/arch/arc/lib/memcpy-archs.S index 1b2b3acfed52..0cab0b8a57c5 100644 --- a/arch/arc/lib/memcpy-archs.S +++ b/arch/arc/lib/memcpy-archs.S @@ -206,7 +206,7 @@ unalignedOffby3: ld.ab r6, [r1, 4] prefetch [r1, 28] ;Prefetch the next read location ld.ab r8, [r1,4] - prefetch [r3, 32] ;Prefetch the next write location + prefetchw [r3, 32] ;Prefetch the next write location SHIFT_1 (r7, r6, 8) or r7, r7, r5 From 262137bca7cfbe690f8e904822b68f720998324a Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 20 Jul 2015 12:05:03 +0300 Subject: [PATCH 04/15] ARCv2: lib: memset: Don't assume 64-bit load/stores There are configurations which may not have LDD/STD Signed-off-by: Claudiu Zissulescu Signed-off-by: Alexey Brodkin Signed-off-by: Vineet Gupta --- arch/arc/lib/memset-archs.S | 43 +++++++++++++++++++++++++++++++------ 1 file changed, 36 insertions(+), 7 deletions(-) diff --git a/arch/arc/lib/memset-archs.S b/arch/arc/lib/memset-archs.S index 92d573c734b5..365b18364815 100644 --- a/arch/arc/lib/memset-archs.S +++ b/arch/arc/lib/memset-archs.S @@ -10,12 +10,6 @@ #undef PREALLOC_NOT_AVAIL -#ifdef PREALLOC_NOT_AVAIL -#define PREWRITE(A,B) prefetchw [(A),(B)] -#else -#define PREWRITE(A,B) prealloc [(A),(B)] -#endif - ENTRY(memset) prefetchw [r0] ; Prefetch the write location mov.f 0, r2 @@ -51,9 +45,15 @@ ENTRY(memset) ;;; Convert len to Dwords, unfold x8 lsr.f lp_count, lp_count, 6 + lpnz @.Lset64bytes ;; LOOP START - PREWRITE(r3, 64) ;Prefetch the next write location +#ifdef PREALLOC_NOT_AVAIL + prefetchw [r3, 64] ;Prefetch the next write location +#else + prealloc [r3, 64] +#endif +#ifdef CONFIG_ARC_HAS_LL64 std.ab r4, [r3, 8] std.ab r4, [r3, 8] std.ab r4, [r3, 8] @@ -62,16 +62,45 @@ ENTRY(memset) std.ab r4, [r3, 8] std.ab r4, [r3, 8] std.ab r4, [r3, 8] +#else + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] +#endif .Lset64bytes: lsr.f lp_count, r2, 5 ;Last remaining max 124 bytes lpnz .Lset32bytes ;; LOOP START prefetchw [r3, 32] ;Prefetch the next write location +#ifdef CONFIG_ARC_HAS_LL64 std.ab r4, [r3, 8] std.ab r4, [r3, 8] std.ab r4, [r3, 8] std.ab r4, [r3, 8] +#else + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] +#endif .Lset32bytes: and.f lp_count, r2, 0x1F ;Last remaining 31 bytes From 450ed0db011d200a691591a75fdbcd7cae1a9b4a Mon Sep 17 00:00:00 2001 From: Alexey Brodkin Date: Thu, 16 Jul 2015 21:45:17 +0300 Subject: [PATCH 05/15] ARCv2: allow selection of page size for MMUv4 MMUv4 also supports the configurable page size as MMUv3. Signed-off-by: Alexey Brodkin Signed-off-by: Vineet Gupta --- arch/arc/Kconfig | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig index f7cfa7d574f6..a5fccdfbfc8f 100644 --- a/arch/arc/Kconfig +++ b/arch/arc/Kconfig @@ -313,11 +313,11 @@ config ARC_PAGE_SIZE_8K config ARC_PAGE_SIZE_16K bool "16KB" - depends on ARC_MMU_V3 + depends on ARC_MMU_V3 || ARC_MMU_V4 config ARC_PAGE_SIZE_4K bool "4KB" - depends on ARC_MMU_V3 + depends on ARC_MMU_V3 || ARC_MMU_V4 endchoice From e13c42ecbe580509451e021ba2586871e5b47640 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 3 Aug 2015 15:37:24 +0530 Subject: [PATCH 06/15] ARCv2: Fix the peripheral address space detection With HS 2.1 release, the peripheral space register no longer contains the uncached space specifics, causing the kernel to panic early on. So read the newer NON VOLATILE AUX register to get that info. Signed-off-by: Vineet Gupta --- arch/arc/include/asm/arcregs.h | 7 +++---- arch/arc/kernel/setup.c | 8 +++++++- 2 files changed, 10 insertions(+), 5 deletions(-) diff --git a/arch/arc/include/asm/arcregs.h b/arch/arc/include/asm/arcregs.h index 070f58827a5c..c8f57b8449dc 100644 --- a/arch/arc/include/asm/arcregs.h +++ b/arch/arc/include/asm/arcregs.h @@ -89,11 +89,10 @@ #define ECR_C_BIT_DTLB_LD_MISS 8 #define ECR_C_BIT_DTLB_ST_MISS 9 - /* Auxiliary registers */ #define AUX_IDENTITY 4 #define AUX_INTR_VEC_BASE 0x25 - +#define AUX_NON_VOL 0x5e /* * Floating Pt Registers @@ -240,9 +239,9 @@ struct bcr_extn_xymem { struct bcr_perip { #ifdef CONFIG_CPU_BIG_ENDIAN - unsigned int start:8, pad2:8, sz:8, pad:8; + unsigned int start:8, pad2:8, sz:8, ver:8; #else - unsigned int pad:8, sz:8, pad2:8, start:8; + unsigned int ver:8, sz:8, pad2:8, start:8; #endif }; diff --git a/arch/arc/kernel/setup.c b/arch/arc/kernel/setup.c index 18cc01591c96..f2f771bd3ede 100644 --- a/arch/arc/kernel/setup.c +++ b/arch/arc/kernel/setup.c @@ -47,6 +47,7 @@ static void read_arc_build_cfg_regs(void) struct bcr_perip uncached_space; struct bcr_generic bcr; struct cpuinfo_arc *cpu = &cpuinfo_arc700[smp_processor_id()]; + unsigned long perip_space; FIX_PTR(cpu); READ_BCR(AUX_IDENTITY, cpu->core); @@ -56,7 +57,12 @@ static void read_arc_build_cfg_regs(void) cpu->vec_base = read_aux_reg(AUX_INTR_VEC_BASE); READ_BCR(ARC_REG_D_UNCACH_BCR, uncached_space); - BUG_ON((uncached_space.start << 24) != ARC_UNCACHED_ADDR_SPACE); + if (uncached_space.ver < 3) + perip_space = uncached_space.start << 24; + else + perip_space = read_aux_reg(AUX_NON_VOL) & 0xF0000000; + + BUG_ON(perip_space != ARC_UNCACHED_ADDR_SPACE); READ_BCR(ARC_REG_MUL_BCR, cpu->extn_mpy); From 6de7abfbad1c6a45893a47a17c2ac91b551aa90d Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Mon, 3 Aug 2015 18:27:56 +0530 Subject: [PATCH 07/15] ARCv2: [axs103_smp] Reduce clk for Quad FPGA configs Signed-off-by: Vineet Gupta --- arch/arc/plat-axs10x/axs10x.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/arch/arc/plat-axs10x/axs10x.c b/arch/arc/plat-axs10x/axs10x.c index 99f7da513a48..e7769c3ab5f2 100644 --- a/arch/arc/plat-axs10x/axs10x.c +++ b/arch/arc/plat-axs10x/axs10x.c @@ -389,6 +389,21 @@ axs103_set_freq(unsigned int id, unsigned int fd, unsigned int od) static void __init axs103_early_init(void) { + /* + * AXS103 configurations for SMP/QUAD configurations share device tree + * which defaults to 90 MHz. However recent failures of Quad config + * revealed P&R timing violations so clamp it down to safe 50 MHz + * Instead of duplicating defconfig/DT for SMP/QUAD, add a small hack + * + * This hack is really hacky as of now. Fix it properly by getting the + * number of cores as return value of platform's early SMP callback + */ +#ifdef CONFIG_ARC_MCIP + unsigned int num_cores = (read_aux_reg(ARC_REG_MCIP_BCR) >> 16) & 0x3F; + if (num_cores > 2) + arc_set_core_freq(50 * 1000000); +#endif + switch (arc_get_core_freq()/1000000) { case 33: axs103_set_freq(1, 1, 1); From f5959cb0c34c5c7f9f387e5d8fe1ec831ccb42ac Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Wed, 29 Jul 2015 19:20:58 +0530 Subject: [PATCH 08/15] Revert "ARCv2: STAR 9000837815 workaround hardware exclusive transactions livelock" Extended testing of quad core configuration revealed that this fix was insufficient. Specifically LTP open posix shm_op/23-1 would cause the hardware livelock in llock/scond loop in update_cpu_load_active() So remove this and make way for a proper workaround This reverts commit a5c8b52abe677977883655166796f167ef1e0084. Signed-off-by: Vineet Gupta --- arch/arc/include/asm/atomic.h | 14 ++------------ 1 file changed, 2 insertions(+), 12 deletions(-) diff --git a/arch/arc/include/asm/atomic.h b/arch/arc/include/asm/atomic.h index 03484cb4d16d..20b7dc17979e 100644 --- a/arch/arc/include/asm/atomic.h +++ b/arch/arc/include/asm/atomic.h @@ -23,21 +23,13 @@ #define atomic_set(v, i) (((v)->counter) = (i)) -#ifdef CONFIG_ISA_ARCV2 -#define PREFETCHW " prefetchw [%1] \n" -#else -#define PREFETCHW -#endif - #define ATOMIC_OP(op, c_op, asm_op) \ static inline void atomic_##op(int i, atomic_t *v) \ { \ unsigned int temp; \ \ __asm__ __volatile__( \ - "1: \n" \ - PREFETCHW \ - " llock %0, [%1] \n" \ + "1: llock %0, [%1] \n" \ " " #asm_op " %0, %0, %2 \n" \ " scond %0, [%1] \n" \ " bnz 1b \n" \ @@ -58,9 +50,7 @@ static inline int atomic_##op##_return(int i, atomic_t *v) \ smp_mb(); \ \ __asm__ __volatile__( \ - "1: \n" \ - PREFETCHW \ - " llock %0, [%1] \n" \ + "1: llock %0, [%1] \n" \ " " #asm_op " %0, %0, %2 \n" \ " scond %0, [%1] \n" \ " bnz 1b \n" \ From 8ac0665fb6ff125efc4df08772317f7311a95ce5 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Tue, 21 Jul 2015 12:05:42 +0300 Subject: [PATCH 09/15] ARC: refactor atomic inline asm operands with symbolic names This reduces the diff in forth-coming patches and also helps understand better the incremental changes to inline asm. Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/include/asm/atomic.h | 32 +++++++++++++++++--------------- 1 file changed, 17 insertions(+), 15 deletions(-) diff --git a/arch/arc/include/asm/atomic.h b/arch/arc/include/asm/atomic.h index 20b7dc17979e..3dd36c1efee1 100644 --- a/arch/arc/include/asm/atomic.h +++ b/arch/arc/include/asm/atomic.h @@ -26,22 +26,23 @@ #define ATOMIC_OP(op, c_op, asm_op) \ static inline void atomic_##op(int i, atomic_t *v) \ { \ - unsigned int temp; \ + unsigned int val; \ \ __asm__ __volatile__( \ - "1: llock %0, [%1] \n" \ - " " #asm_op " %0, %0, %2 \n" \ - " scond %0, [%1] \n" \ - " bnz 1b \n" \ - : "=&r"(temp) /* Early clobber, to prevent reg reuse */ \ - : "r"(&v->counter), "ir"(i) \ + "1: llock %[val], [%[ctr]] \n" \ + " " #asm_op " %[val], %[val], %[i] \n" \ + " scond %[val], [%[ctr]] \n" \ + " bnz 1b \n" \ + : [val] "=&r" (val) /* Early clobber to prevent reg reuse */ \ + : [ctr] "r" (&v->counter), /* Not "m": llock only supports reg direct addr mode */ \ + [i] "ir" (i) \ : "cc"); \ } \ #define ATOMIC_OP_RETURN(op, c_op, asm_op) \ static inline int atomic_##op##_return(int i, atomic_t *v) \ { \ - unsigned int temp; \ + unsigned int val; \ \ /* \ * Explicit full memory barrier needed before/after as \ @@ -50,17 +51,18 @@ static inline int atomic_##op##_return(int i, atomic_t *v) \ smp_mb(); \ \ __asm__ __volatile__( \ - "1: llock %0, [%1] \n" \ - " " #asm_op " %0, %0, %2 \n" \ - " scond %0, [%1] \n" \ - " bnz 1b \n" \ - : "=&r"(temp) \ - : "r"(&v->counter), "ir"(i) \ + "1: llock %[val], [%[ctr]] \n" \ + " " #asm_op " %[val], %[val], %[i] \n" \ + " scond %[val], [%[ctr]] \n" \ + " bnz 1b \n" \ + : [val] "=&r" (val) \ + : [ctr] "r" (&v->counter), \ + [i] "ir" (i) \ : "cc"); \ \ smp_mb(); \ \ - return temp; \ + return val; \ } #else /* !CONFIG_ARC_HAS_LLSC */ From ae7eae9e031206bde8645d038c76ad89e7d3fbcb Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Tue, 14 Jul 2015 17:55:05 +0530 Subject: [PATCH 10/15] ARC: LLOCK/SCOND based spin_lock Current spin_lock uses EXchange instruction to implement the atomic test and set of lock location (reads orig value and ST 1). This however forces the cacheline into exclusive state (because of the ST) and concurrent loops in multiple cores will bounce the line around between cores. Instead, use LLOCK/SCOND to implement the atomic test and set which is better as line is in shared state while lock is spinning on LLOCK The real motivation of this change however is to make way for future changes in atomics to implement delayed retry (with backoff). Initial experiment with delayed retry in atomics combined with orig EX based spinlock was a total disaster (broke even LMBench) as struct sock has a cache line sharing an atomic_t and spinlock. The tight spinning on lock, caused the atomic retry to keep backing off such that it would never finish. Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/include/asm/spinlock.h | 76 ++++++++++++++++++++++++++++++--- 1 file changed, 69 insertions(+), 7 deletions(-) diff --git a/arch/arc/include/asm/spinlock.h b/arch/arc/include/asm/spinlock.h index e1651df6a93d..4f6c90a0a68a 100644 --- a/arch/arc/include/asm/spinlock.h +++ b/arch/arc/include/asm/spinlock.h @@ -18,9 +18,68 @@ #define arch_spin_unlock_wait(x) \ do { while (arch_spin_is_locked(x)) cpu_relax(); } while (0) +#ifdef CONFIG_ARC_HAS_LLSC + static inline void arch_spin_lock(arch_spinlock_t *lock) { - unsigned int tmp = __ARCH_SPIN_LOCK_LOCKED__; + unsigned int val; + + smp_mb(); + + __asm__ __volatile__( + "1: llock %[val], [%[slock]] \n" + " breq %[val], %[LOCKED], 1b \n" /* spin while LOCKED */ + " scond %[LOCKED], [%[slock]] \n" /* acquire */ + " bnz 1b \n" + " \n" + : [val] "=&r" (val) + : [slock] "r" (&(lock->slock)), + [LOCKED] "r" (__ARCH_SPIN_LOCK_LOCKED__) + : "memory", "cc"); + + smp_mb(); +} + +/* 1 - lock taken successfully */ +static inline int arch_spin_trylock(arch_spinlock_t *lock) +{ + unsigned int val, got_it = 0; + + smp_mb(); + + __asm__ __volatile__( + "1: llock %[val], [%[slock]] \n" + " breq %[val], %[LOCKED], 4f \n" /* already LOCKED, just bail */ + " scond %[LOCKED], [%[slock]] \n" /* acquire */ + " bnz 1b \n" + " mov %[got_it], 1 \n" + "4: \n" + " \n" + : [val] "=&r" (val), + [got_it] "+&r" (got_it) + : [slock] "r" (&(lock->slock)), + [LOCKED] "r" (__ARCH_SPIN_LOCK_LOCKED__) + : "memory", "cc"); + + smp_mb(); + + return got_it; +} + +static inline void arch_spin_unlock(arch_spinlock_t *lock) +{ + smp_mb(); + + lock->slock = __ARCH_SPIN_LOCK_UNLOCKED__; + + smp_mb(); +} + +#else /* !CONFIG_ARC_HAS_LLSC */ + +static inline void arch_spin_lock(arch_spinlock_t *lock) +{ + unsigned int val = __ARCH_SPIN_LOCK_LOCKED__; /* * This smp_mb() is technically superfluous, we only need the one @@ -33,7 +92,7 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) __asm__ __volatile__( "1: ex %0, [%1] \n" " breq %0, %2, 1b \n" - : "+&r" (tmp) + : "+&r" (val) : "r"(&(lock->slock)), "ir"(__ARCH_SPIN_LOCK_LOCKED__) : "memory"); @@ -48,26 +107,27 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) smp_mb(); } +/* 1 - lock taken successfully */ static inline int arch_spin_trylock(arch_spinlock_t *lock) { - unsigned int tmp = __ARCH_SPIN_LOCK_LOCKED__; + unsigned int val = __ARCH_SPIN_LOCK_LOCKED__; smp_mb(); __asm__ __volatile__( "1: ex %0, [%1] \n" - : "+r" (tmp) + : "+r" (val) : "r"(&(lock->slock)) : "memory"); smp_mb(); - return (tmp == __ARCH_SPIN_LOCK_UNLOCKED__); + return (val == __ARCH_SPIN_LOCK_UNLOCKED__); } static inline void arch_spin_unlock(arch_spinlock_t *lock) { - unsigned int tmp = __ARCH_SPIN_LOCK_UNLOCKED__; + unsigned int val = __ARCH_SPIN_LOCK_UNLOCKED__; /* * RELEASE barrier: given the instructions avail on ARCv2, full barrier @@ -77,7 +137,7 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock) __asm__ __volatile__( " ex %0, [%1] \n" - : "+r" (tmp) + : "+r" (val) : "r"(&(lock->slock)) : "memory"); @@ -88,6 +148,8 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock) smp_mb(); } +#endif + /* * Read-write spinlocks, allowing multiple readers but only one writer. * From 69cbe630f54ec02efe47fdb9e257e617161da370 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Thu, 16 Jul 2015 10:31:45 +0530 Subject: [PATCH 11/15] ARC: LLOCK/SCOND based rwlock With LLOCK/SCOND, the rwlock counter can be atomically updated w/o need for a guarding spin lock. This in turn elides the EXchange instruction based spinning which causes the cacheline transition to exclusive state and concurrent spinning across cores would cause the line to keep bouncing around. LLOCK/SCOND based implementation is superior as spinning on LLOCK keeps the cacheline in shared state. Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/include/asm/spinlock.h | 174 ++++++++++++++++++++++++-- arch/arc/include/asm/spinlock_types.h | 2 + 2 files changed, 166 insertions(+), 10 deletions(-) diff --git a/arch/arc/include/asm/spinlock.h b/arch/arc/include/asm/spinlock.h index 4f6c90a0a68a..9fd5a0221671 100644 --- a/arch/arc/include/asm/spinlock.h +++ b/arch/arc/include/asm/spinlock.h @@ -75,6 +75,164 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock) smp_mb(); } +/* + * Read-write spinlocks, allowing multiple readers but only one writer. + * Unfair locking as Writers could be starved indefinitely by Reader(s) + */ + +static inline void arch_read_lock(arch_rwlock_t *rw) +{ + unsigned int val; + + smp_mb(); + + /* + * zero means writer holds the lock exclusively, deny Reader. + * Otherwise grant lock to first/subseq reader + * + * if (rw->counter > 0) { + * rw->counter--; + * ret = 1; + * } + */ + + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " brls %[val], %[WR_LOCKED], 1b\n" /* <= 0: spin while write locked */ + " sub %[val], %[val], 1 \n" /* reader lock */ + " scond %[val], [%[rwlock]] \n" + " bnz 1b \n" + " \n" + : [val] "=&r" (val) + : [rwlock] "r" (&(rw->counter)), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); +} + +/* 1 - lock taken successfully */ +static inline int arch_read_trylock(arch_rwlock_t *rw) +{ + unsigned int val, got_it = 0; + + smp_mb(); + + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " brls %[val], %[WR_LOCKED], 4f\n" /* <= 0: already write locked, bail */ + " sub %[val], %[val], 1 \n" /* counter-- */ + " scond %[val], [%[rwlock]] \n" + " bnz 1b \n" /* retry if collided with someone */ + " mov %[got_it], 1 \n" + " \n" + "4: ; --- done --- \n" + + : [val] "=&r" (val), + [got_it] "+&r" (got_it) + : [rwlock] "r" (&(rw->counter)), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); + + return got_it; +} + +static inline void arch_write_lock(arch_rwlock_t *rw) +{ + unsigned int val; + + smp_mb(); + + /* + * If reader(s) hold lock (lock < __ARCH_RW_LOCK_UNLOCKED__), + * deny writer. Otherwise if unlocked grant to writer + * Hence the claim that Linux rwlocks are unfair to writers. + * (can be starved for an indefinite time by readers). + * + * if (rw->counter == __ARCH_RW_LOCK_UNLOCKED__) { + * rw->counter = 0; + * ret = 1; + * } + */ + + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " brne %[val], %[UNLOCKED], 1b \n" /* while !UNLOCKED spin */ + " mov %[val], %[WR_LOCKED] \n" + " scond %[val], [%[rwlock]] \n" + " bnz 1b \n" + " \n" + : [val] "=&r" (val) + : [rwlock] "r" (&(rw->counter)), + [UNLOCKED] "ir" (__ARCH_RW_LOCK_UNLOCKED__), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); +} + +/* 1 - lock taken successfully */ +static inline int arch_write_trylock(arch_rwlock_t *rw) +{ + unsigned int val, got_it = 0; + + smp_mb(); + + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " brne %[val], %[UNLOCKED], 4f \n" /* !UNLOCKED, bail */ + " mov %[val], %[WR_LOCKED] \n" + " scond %[val], [%[rwlock]] \n" + " bnz 1b \n" /* retry if collided with someone */ + " mov %[got_it], 1 \n" + " \n" + "4: ; --- done --- \n" + + : [val] "=&r" (val), + [got_it] "+&r" (got_it) + : [rwlock] "r" (&(rw->counter)), + [UNLOCKED] "ir" (__ARCH_RW_LOCK_UNLOCKED__), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); + + return got_it; +} + +static inline void arch_read_unlock(arch_rwlock_t *rw) +{ + unsigned int val; + + smp_mb(); + + /* + * rw->counter++; + */ + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " add %[val], %[val], 1 \n" + " scond %[val], [%[rwlock]] \n" + " bnz 1b \n" + " \n" + : [val] "=&r" (val) + : [rwlock] "r" (&(rw->counter)) + : "memory", "cc"); + + smp_mb(); +} + +static inline void arch_write_unlock(arch_rwlock_t *rw) +{ + smp_mb(); + + rw->counter = __ARCH_RW_LOCK_UNLOCKED__; + + smp_mb(); +} + #else /* !CONFIG_ARC_HAS_LLSC */ static inline void arch_spin_lock(arch_spinlock_t *lock) @@ -148,23 +306,14 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock) smp_mb(); } -#endif - /* * Read-write spinlocks, allowing multiple readers but only one writer. + * Unfair locking as Writers could be starved indefinitely by Reader(s) * * The spinlock itself is contained in @counter and access to it is * serialized with @lock_mutex. - * - * Unfair locking as Writers could be starved indefinitely by Reader(s) */ -/* Would read_trylock() succeed? */ -#define arch_read_can_lock(x) ((x)->counter > 0) - -/* Would write_trylock() succeed? */ -#define arch_write_can_lock(x) ((x)->counter == __ARCH_RW_LOCK_UNLOCKED__) - /* 1 - lock taken successfully */ static inline int arch_read_trylock(arch_rwlock_t *rw) { @@ -235,6 +384,11 @@ static inline void arch_write_unlock(arch_rwlock_t *rw) arch_spin_unlock(&(rw->lock_mutex)); } +#endif + +#define arch_read_can_lock(x) ((x)->counter > 0) +#define arch_write_can_lock(x) ((x)->counter == __ARCH_RW_LOCK_UNLOCKED__) + #define arch_read_lock_flags(lock, flags) arch_read_lock(lock) #define arch_write_lock_flags(lock, flags) arch_write_lock(lock) diff --git a/arch/arc/include/asm/spinlock_types.h b/arch/arc/include/asm/spinlock_types.h index 662627ced4f2..4e1ef5f650c6 100644 --- a/arch/arc/include/asm/spinlock_types.h +++ b/arch/arc/include/asm/spinlock_types.h @@ -26,7 +26,9 @@ typedef struct { */ typedef struct { volatile unsigned int counter; +#ifndef CONFIG_ARC_HAS_LLSC arch_spinlock_t lock_mutex; +#endif } arch_rwlock_t; #define __ARCH_RW_LOCK_UNLOCKED__ 0x01000000 From e78fdfef84be13a5c2b8276e12203cdf24778596 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Tue, 14 Jul 2015 19:50:18 +0530 Subject: [PATCH 12/15] ARCv2: spinlock/rwlock/atomics: Delayed retry of failed SCOND with exponential backoff This is to workaround the llock/scond livelock HS38x4 could get into a LLOCK/SCOND livelock in case of multiple overlapping coherency transactions in the SCU. The exclusive line state keeps rotating among contenting cores leading to a never ending cycle. So break the cycle by deferring the retry of failed exclusive access (SCOND). The actual delay needed is function of number of contending cores as well as the unrelated coherency traffic from other cores. To keep the code simple, start off with small delay of 1 which would suffice most cases and in case of contention double the delay. Eventually the delay is sufficient such that the coherency pipeline is drained, thus a subsequent exclusive access would succeed. Link: http://lkml.kernel.org/r/1438612568-28265-1-git-send-email-vgupta@synopsys.com Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/Kconfig | 5 + arch/arc/include/asm/atomic.h | 49 +++++- arch/arc/include/asm/spinlock.h | 293 ++++++++++++++++++++++++++++++++ arch/arc/kernel/setup.c | 4 + 4 files changed, 347 insertions(+), 4 deletions(-) diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig index a5fccdfbfc8f..bd4670d1b89b 100644 --- a/arch/arc/Kconfig +++ b/arch/arc/Kconfig @@ -365,6 +365,11 @@ config ARC_HAS_LLSC default y depends on !ARC_CANT_LLSC +config ARC_STAR_9000923308 + bool "Workaround for llock/scond livelock" + default y + depends on ISA_ARCV2 && SMP && ARC_HAS_LLSC + config ARC_HAS_SWAPE bool "Insn: SWAPE (endian-swap)" default y diff --git a/arch/arc/include/asm/atomic.h b/arch/arc/include/asm/atomic.h index 3dd36c1efee1..629dfd0a0c6b 100644 --- a/arch/arc/include/asm/atomic.h +++ b/arch/arc/include/asm/atomic.h @@ -23,17 +23,51 @@ #define atomic_set(v, i) (((v)->counter) = (i)) +#ifdef CONFIG_ARC_STAR_9000923308 + +#define SCOND_FAIL_RETRY_VAR_DEF \ + unsigned int delay = 1, tmp; \ + +#define SCOND_FAIL_RETRY_ASM \ + " bz 4f \n" \ + " ; --- scond fail delay --- \n" \ + " mov %[tmp], %[delay] \n" /* tmp = delay */ \ + "2: brne.d %[tmp], 0, 2b \n" /* while (tmp != 0) */ \ + " sub %[tmp], %[tmp], 1 \n" /* tmp-- */ \ + " asl.f %[delay], %[delay], 1 \n" /* delay *= 2 */ \ + " mov.z %[delay], 1 \n" /* handle overflow */ \ + " b 1b \n" /* start over */ \ + "4: ; --- success --- \n" \ + +#define SCOND_FAIL_RETRY_VARS \ + ,[delay] "+&r" (delay),[tmp] "=&r" (tmp) \ + +#else /* !CONFIG_ARC_STAR_9000923308 */ + +#define SCOND_FAIL_RETRY_VAR_DEF + +#define SCOND_FAIL_RETRY_ASM \ + " bnz 1b \n" \ + +#define SCOND_FAIL_RETRY_VARS + +#endif + #define ATOMIC_OP(op, c_op, asm_op) \ static inline void atomic_##op(int i, atomic_t *v) \ { \ - unsigned int val; \ + unsigned int val; \ + SCOND_FAIL_RETRY_VAR_DEF \ \ __asm__ __volatile__( \ "1: llock %[val], [%[ctr]] \n" \ " " #asm_op " %[val], %[val], %[i] \n" \ " scond %[val], [%[ctr]] \n" \ - " bnz 1b \n" \ + " \n" \ + SCOND_FAIL_RETRY_ASM \ + \ : [val] "=&r" (val) /* Early clobber to prevent reg reuse */ \ + SCOND_FAIL_RETRY_VARS \ : [ctr] "r" (&v->counter), /* Not "m": llock only supports reg direct addr mode */ \ [i] "ir" (i) \ : "cc"); \ @@ -42,7 +76,8 @@ static inline void atomic_##op(int i, atomic_t *v) \ #define ATOMIC_OP_RETURN(op, c_op, asm_op) \ static inline int atomic_##op##_return(int i, atomic_t *v) \ { \ - unsigned int val; \ + unsigned int val; \ + SCOND_FAIL_RETRY_VAR_DEF \ \ /* \ * Explicit full memory barrier needed before/after as \ @@ -54,8 +89,11 @@ static inline int atomic_##op##_return(int i, atomic_t *v) \ "1: llock %[val], [%[ctr]] \n" \ " " #asm_op " %[val], %[val], %[i] \n" \ " scond %[val], [%[ctr]] \n" \ - " bnz 1b \n" \ + " \n" \ + SCOND_FAIL_RETRY_ASM \ + \ : [val] "=&r" (val) \ + SCOND_FAIL_RETRY_VARS \ : [ctr] "r" (&v->counter), \ [i] "ir" (i) \ : "cc"); \ @@ -142,6 +180,9 @@ ATOMIC_OP(and, &=, and) #undef ATOMIC_OPS #undef ATOMIC_OP_RETURN #undef ATOMIC_OP +#undef SCOND_FAIL_RETRY_VAR_DEF +#undef SCOND_FAIL_RETRY_ASM +#undef SCOND_FAIL_RETRY_VARS /** * __atomic_add_unless - add unless the number is a given value diff --git a/arch/arc/include/asm/spinlock.h b/arch/arc/include/asm/spinlock.h index 9fd5a0221671..2a84525b5aa4 100644 --- a/arch/arc/include/asm/spinlock.h +++ b/arch/arc/include/asm/spinlock.h @@ -20,6 +20,11 @@ #ifdef CONFIG_ARC_HAS_LLSC +/* + * A normal LLOCK/SCOND based system, w/o need for livelock workaround + */ +#ifndef CONFIG_ARC_STAR_9000923308 + static inline void arch_spin_lock(arch_spinlock_t *lock) { unsigned int val; @@ -233,6 +238,294 @@ static inline void arch_write_unlock(arch_rwlock_t *rw) smp_mb(); } +#else /* CONFIG_ARC_STAR_9000923308 */ + +/* + * HS38x4 could get into a LLOCK/SCOND livelock in case of multiple overlapping + * coherency transactions in the SCU. The exclusive line state keeps rotating + * among contenting cores leading to a never ending cycle. So break the cycle + * by deferring the retry of failed exclusive access (SCOND). The actual delay + * needed is function of number of contending cores as well as the unrelated + * coherency traffic from other cores. To keep the code simple, start off with + * small delay of 1 which would suffice most cases and in case of contention + * double the delay. Eventually the delay is sufficient such that the coherency + * pipeline is drained, thus a subsequent exclusive access would succeed. + */ + +#define SCOND_FAIL_RETRY_VAR_DEF \ + unsigned int delay, tmp; \ + +#define SCOND_FAIL_RETRY_ASM \ + " ; --- scond fail delay --- \n" \ + " mov %[tmp], %[delay] \n" /* tmp = delay */ \ + "2: brne.d %[tmp], 0, 2b \n" /* while (tmp != 0) */ \ + " sub %[tmp], %[tmp], 1 \n" /* tmp-- */ \ + " asl.f %[delay], %[delay], 1 \n" /* delay *= 2 */ \ + " mov.z %[delay], 1 \n" /* handle overflow */ \ + " b 1b \n" /* start over */ \ + " \n" \ + "4: ; --- done --- \n" \ + +#define SCOND_FAIL_RETRY_VARS \ + ,[delay] "=&r" (delay), [tmp] "=&r" (tmp) \ + +static inline void arch_spin_lock(arch_spinlock_t *lock) +{ + unsigned int val; + SCOND_FAIL_RETRY_VAR_DEF; + + smp_mb(); + + __asm__ __volatile__( + "0: mov %[delay], 1 \n" + "1: llock %[val], [%[slock]] \n" + " breq %[val], %[LOCKED], 1b \n" /* spin while LOCKED */ + " scond %[LOCKED], [%[slock]] \n" /* acquire */ + " bz 4f \n" /* done */ + " \n" + SCOND_FAIL_RETRY_ASM + + : [val] "=&r" (val) + SCOND_FAIL_RETRY_VARS + : [slock] "r" (&(lock->slock)), + [LOCKED] "r" (__ARCH_SPIN_LOCK_LOCKED__) + : "memory", "cc"); + + smp_mb(); +} + +/* 1 - lock taken successfully */ +static inline int arch_spin_trylock(arch_spinlock_t *lock) +{ + unsigned int val, got_it = 0; + SCOND_FAIL_RETRY_VAR_DEF; + + smp_mb(); + + __asm__ __volatile__( + "0: mov %[delay], 1 \n" + "1: llock %[val], [%[slock]] \n" + " breq %[val], %[LOCKED], 4f \n" /* already LOCKED, just bail */ + " scond %[LOCKED], [%[slock]] \n" /* acquire */ + " bz.d 4f \n" + " mov.z %[got_it], 1 \n" /* got it */ + " \n" + SCOND_FAIL_RETRY_ASM + + : [val] "=&r" (val), + [got_it] "+&r" (got_it) + SCOND_FAIL_RETRY_VARS + : [slock] "r" (&(lock->slock)), + [LOCKED] "r" (__ARCH_SPIN_LOCK_LOCKED__) + : "memory", "cc"); + + smp_mb(); + + return got_it; +} + +static inline void arch_spin_unlock(arch_spinlock_t *lock) +{ + smp_mb(); + + lock->slock = __ARCH_SPIN_LOCK_UNLOCKED__; + + smp_mb(); +} + +/* + * Read-write spinlocks, allowing multiple readers but only one writer. + * Unfair locking as Writers could be starved indefinitely by Reader(s) + */ + +static inline void arch_read_lock(arch_rwlock_t *rw) +{ + unsigned int val; + SCOND_FAIL_RETRY_VAR_DEF; + + smp_mb(); + + /* + * zero means writer holds the lock exclusively, deny Reader. + * Otherwise grant lock to first/subseq reader + * + * if (rw->counter > 0) { + * rw->counter--; + * ret = 1; + * } + */ + + __asm__ __volatile__( + "0: mov %[delay], 1 \n" + "1: llock %[val], [%[rwlock]] \n" + " brls %[val], %[WR_LOCKED], 1b\n" /* <= 0: spin while write locked */ + " sub %[val], %[val], 1 \n" /* reader lock */ + " scond %[val], [%[rwlock]] \n" + " bz 4f \n" /* done */ + " \n" + SCOND_FAIL_RETRY_ASM + + : [val] "=&r" (val) + SCOND_FAIL_RETRY_VARS + : [rwlock] "r" (&(rw->counter)), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); +} + +/* 1 - lock taken successfully */ +static inline int arch_read_trylock(arch_rwlock_t *rw) +{ + unsigned int val, got_it = 0; + SCOND_FAIL_RETRY_VAR_DEF; + + smp_mb(); + + __asm__ __volatile__( + "0: mov %[delay], 1 \n" + "1: llock %[val], [%[rwlock]] \n" + " brls %[val], %[WR_LOCKED], 4f\n" /* <= 0: already write locked, bail */ + " sub %[val], %[val], 1 \n" /* counter-- */ + " scond %[val], [%[rwlock]] \n" + " bz.d 4f \n" + " mov.z %[got_it], 1 \n" /* got it */ + " \n" + SCOND_FAIL_RETRY_ASM + + : [val] "=&r" (val), + [got_it] "+&r" (got_it) + SCOND_FAIL_RETRY_VARS + : [rwlock] "r" (&(rw->counter)), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); + + return got_it; +} + +static inline void arch_write_lock(arch_rwlock_t *rw) +{ + unsigned int val; + SCOND_FAIL_RETRY_VAR_DEF; + + smp_mb(); + + /* + * If reader(s) hold lock (lock < __ARCH_RW_LOCK_UNLOCKED__), + * deny writer. Otherwise if unlocked grant to writer + * Hence the claim that Linux rwlocks are unfair to writers. + * (can be starved for an indefinite time by readers). + * + * if (rw->counter == __ARCH_RW_LOCK_UNLOCKED__) { + * rw->counter = 0; + * ret = 1; + * } + */ + + __asm__ __volatile__( + "0: mov %[delay], 1 \n" + "1: llock %[val], [%[rwlock]] \n" + " brne %[val], %[UNLOCKED], 1b \n" /* while !UNLOCKED spin */ + " mov %[val], %[WR_LOCKED] \n" + " scond %[val], [%[rwlock]] \n" + " bz 4f \n" + " \n" + SCOND_FAIL_RETRY_ASM + + : [val] "=&r" (val) + SCOND_FAIL_RETRY_VARS + : [rwlock] "r" (&(rw->counter)), + [UNLOCKED] "ir" (__ARCH_RW_LOCK_UNLOCKED__), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); +} + +/* 1 - lock taken successfully */ +static inline int arch_write_trylock(arch_rwlock_t *rw) +{ + unsigned int val, got_it = 0; + SCOND_FAIL_RETRY_VAR_DEF; + + smp_mb(); + + __asm__ __volatile__( + "0: mov %[delay], 1 \n" + "1: llock %[val], [%[rwlock]] \n" + " brne %[val], %[UNLOCKED], 4f \n" /* !UNLOCKED, bail */ + " mov %[val], %[WR_LOCKED] \n" + " scond %[val], [%[rwlock]] \n" + " bz.d 4f \n" + " mov.z %[got_it], 1 \n" /* got it */ + " \n" + SCOND_FAIL_RETRY_ASM + + : [val] "=&r" (val), + [got_it] "+&r" (got_it) + SCOND_FAIL_RETRY_VARS + : [rwlock] "r" (&(rw->counter)), + [UNLOCKED] "ir" (__ARCH_RW_LOCK_UNLOCKED__), + [WR_LOCKED] "ir" (0) + : "memory", "cc"); + + smp_mb(); + + return got_it; +} + +static inline void arch_read_unlock(arch_rwlock_t *rw) +{ + unsigned int val; + + smp_mb(); + + /* + * rw->counter++; + */ + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " add %[val], %[val], 1 \n" + " scond %[val], [%[rwlock]] \n" + " bnz 1b \n" + " \n" + : [val] "=&r" (val) + : [rwlock] "r" (&(rw->counter)) + : "memory", "cc"); + + smp_mb(); +} + +static inline void arch_write_unlock(arch_rwlock_t *rw) +{ + unsigned int val; + + smp_mb(); + + /* + * rw->counter = __ARCH_RW_LOCK_UNLOCKED__; + */ + __asm__ __volatile__( + "1: llock %[val], [%[rwlock]] \n" + " scond %[UNLOCKED], [%[rwlock]]\n" + " bnz 1b \n" + " \n" + : [val] "=&r" (val) + : [rwlock] "r" (&(rw->counter)), + [UNLOCKED] "r" (__ARCH_RW_LOCK_UNLOCKED__) + : "memory", "cc"); + + smp_mb(); +} + +#undef SCOND_FAIL_RETRY_VAR_DEF +#undef SCOND_FAIL_RETRY_ASM +#undef SCOND_FAIL_RETRY_VARS + +#endif /* CONFIG_ARC_STAR_9000923308 */ + #else /* !CONFIG_ARC_HAS_LLSC */ static inline void arch_spin_lock(arch_spinlock_t *lock) diff --git a/arch/arc/kernel/setup.c b/arch/arc/kernel/setup.c index f2f771bd3ede..cabde9dc0696 100644 --- a/arch/arc/kernel/setup.c +++ b/arch/arc/kernel/setup.c @@ -336,6 +336,10 @@ static void arc_chk_core_config(void) pr_warn("CONFIG_ARC_FPU_SAVE_RESTORE needed for working apps\n"); else if (!cpu->extn.fpu_dp && fpu_enabled) panic("FPU non-existent, disable CONFIG_ARC_FPU_SAVE_RESTORE\n"); + + if (is_isa_arcv2() && IS_ENABLED(CONFIG_SMP) && cpu->isa.atomic && + !IS_ENABLED(CONFIG_ARC_STAR_9000923308)) + panic("llock/scond livelock workaround missing\n"); } /* From b89aa12c177477e34caa722818536fb5d0bffd76 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Tue, 21 Jul 2015 18:46:53 +0300 Subject: [PATCH 13/15] ARCv2: spinlock/rwlock: Reset retry delay when starting a new spin-wait cycle The previous commit for delayed retry of SCOND needs some fine tuning for spin locks. The backoff from delayed retry in conjunction with spin looping of lock itself can potentially cause the delay counter to reach high values. So to provide fairness to any lock operation, after a lock "seems" available (i.e. just before first SCOND try0, reset the delay counter back to starting value of 1 Essentially reset delay to 1 for a new spin-wait-loop-acquire cycle. Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/include/asm/spinlock.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/arch/arc/include/asm/spinlock.h b/arch/arc/include/asm/spinlock.h index 2a84525b5aa4..7071fc0da56a 100644 --- a/arch/arc/include/asm/spinlock.h +++ b/arch/arc/include/asm/spinlock.h @@ -279,7 +279,7 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) __asm__ __volatile__( "0: mov %[delay], 1 \n" "1: llock %[val], [%[slock]] \n" - " breq %[val], %[LOCKED], 1b \n" /* spin while LOCKED */ + " breq %[val], %[LOCKED], 0b \n" /* spin while LOCKED */ " scond %[LOCKED], [%[slock]] \n" /* acquire */ " bz 4f \n" /* done */ " \n" @@ -358,7 +358,7 @@ static inline void arch_read_lock(arch_rwlock_t *rw) __asm__ __volatile__( "0: mov %[delay], 1 \n" "1: llock %[val], [%[rwlock]] \n" - " brls %[val], %[WR_LOCKED], 1b\n" /* <= 0: spin while write locked */ + " brls %[val], %[WR_LOCKED], 0b\n" /* <= 0: spin while write locked */ " sub %[val], %[val], 1 \n" /* reader lock */ " scond %[val], [%[rwlock]] \n" " bz 4f \n" /* done */ @@ -427,7 +427,7 @@ static inline void arch_write_lock(arch_rwlock_t *rw) __asm__ __volatile__( "0: mov %[delay], 1 \n" "1: llock %[val], [%[rwlock]] \n" - " brne %[val], %[UNLOCKED], 1b \n" /* while !UNLOCKED spin */ + " brne %[val], %[UNLOCKED], 0b \n" /* while !UNLOCKED spin */ " mov %[val], %[WR_LOCKED] \n" " scond %[val], [%[rwlock]] \n" " bz 4f \n" From 87ce62802f7a3553234ebf1ae7cd52c8bf272fb9 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Wed, 5 Aug 2015 11:12:00 +0530 Subject: [PATCH 14/15] ARC: Make pt_regs regs unsigned KGDB fails to build after f51e2f191112 ("ARC: make sure instruction_pointer() returns unsigned value") The hack to force one specific reg to unsigned backfired. There's no reason to keep the regs signed after all. | CC arch/arc/kernel/kgdb.o |../arch/arc/kernel/kgdb.c: In function 'kgdb_trap': | ../arch/arc/kernel/kgdb.c:180:29: error: lvalue required as left operand of assignment | instruction_pointer(regs) -= BREAK_INSTR_SIZE; Reported-by: Yuriy Kolerov Fixes: f51e2f191112 ("ARC: make sure instruction_pointer() returns unsigned value") Cc: Alexey Brodkin Signed-off-by: Vineet Gupta --- arch/arc/include/asm/ptrace.h | 54 +++++++++++++++--------------- arch/arc/include/uapi/asm/ptrace.h | 20 +++++------ 2 files changed, 37 insertions(+), 37 deletions(-) diff --git a/arch/arc/include/asm/ptrace.h b/arch/arc/include/asm/ptrace.h index 91694ec1ce95..69095da1fcfd 100644 --- a/arch/arc/include/asm/ptrace.h +++ b/arch/arc/include/asm/ptrace.h @@ -20,20 +20,20 @@ struct pt_regs { /* Real registers */ - long bta; /* bta_l1, bta_l2, erbta */ + unsigned long bta; /* bta_l1, bta_l2, erbta */ - long lp_start, lp_end, lp_count; + unsigned long lp_start, lp_end, lp_count; - long status32; /* status32_l1, status32_l2, erstatus */ - long ret; /* ilink1, ilink2 or eret */ - long blink; - long fp; - long r26; /* gp */ + unsigned long status32; /* status32_l1, status32_l2, erstatus */ + unsigned long ret; /* ilink1, ilink2 or eret */ + unsigned long blink; + unsigned long fp; + unsigned long r26; /* gp */ - long r12, r11, r10, r9, r8, r7, r6, r5, r4, r3, r2, r1, r0; + unsigned long r12, r11, r10, r9, r8, r7, r6, r5, r4, r3, r2, r1, r0; - long sp; /* user/kernel sp depending on where we came from */ - long orig_r0; + unsigned long sp; /* User/Kernel depending on where we came from */ + unsigned long orig_r0; /* * To distinguish bet excp, syscall, irq @@ -55,13 +55,13 @@ struct pt_regs { unsigned long event; }; - long user_r25; + unsigned long user_r25; }; #else struct pt_regs { - long orig_r0; + unsigned long orig_r0; union { struct { @@ -76,26 +76,26 @@ struct pt_regs { unsigned long event; }; - long bta; /* bta_l1, bta_l2, erbta */ + unsigned long bta; /* bta_l1, bta_l2, erbta */ - long user_r25; + unsigned long user_r25; - long r26; /* gp */ - long fp; - long sp; /* user/kernel sp depending on where we came from */ + unsigned long r26; /* gp */ + unsigned long fp; + unsigned long sp; /* user/kernel sp depending on where we came from */ - long r12; + unsigned long r12; /*------- Below list auto saved by h/w -----------*/ - long r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11; + unsigned long r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11; - long blink; - long lp_end, lp_start, lp_count; + unsigned long blink; + unsigned long lp_end, lp_start, lp_count; - long ei, ldi, jli; + unsigned long ei, ldi, jli; - long ret; - long status32; + unsigned long ret; + unsigned long status32; }; #endif @@ -103,10 +103,10 @@ struct pt_regs { /* Callee saved registers - need to be saved only when you are scheduled out */ struct callee_regs { - long r25, r24, r23, r22, r21, r20, r19, r18, r17, r16, r15, r14, r13; + unsigned long r25, r24, r23, r22, r21, r20, r19, r18, r17, r16, r15, r14, r13; }; -#define instruction_pointer(regs) (unsigned long)((regs)->ret) +#define instruction_pointer(regs) ((regs)->ret) #define profile_pc(regs) instruction_pointer(regs) /* return 1 if user mode or 0 if kernel mode */ @@ -142,7 +142,7 @@ struct callee_regs { static inline long regs_return_value(struct pt_regs *regs) { - return regs->r0; + return (long)regs->r0; } #endif /* !__ASSEMBLY__ */ diff --git a/arch/arc/include/uapi/asm/ptrace.h b/arch/arc/include/uapi/asm/ptrace.h index 76a7739aab1c..0b3ef63d4a03 100644 --- a/arch/arc/include/uapi/asm/ptrace.h +++ b/arch/arc/include/uapi/asm/ptrace.h @@ -32,20 +32,20 @@ */ struct user_regs_struct { - long pad; + unsigned long pad; struct { - long bta, lp_start, lp_end, lp_count; - long status32, ret, blink, fp, gp; - long r12, r11, r10, r9, r8, r7, r6, r5, r4, r3, r2, r1, r0; - long sp; + unsigned long bta, lp_start, lp_end, lp_count; + unsigned long status32, ret, blink, fp, gp; + unsigned long r12, r11, r10, r9, r8, r7, r6, r5, r4, r3, r2, r1, r0; + unsigned long sp; } scratch; - long pad2; + unsigned long pad2; struct { - long r25, r24, r23, r22, r21, r20; - long r19, r18, r17, r16, r15, r14, r13; + unsigned long r25, r24, r23, r22, r21, r20; + unsigned long r19, r18, r17, r16, r15, r14, r13; } callee; - long efa; /* break pt addr, for break points in delay slots */ - long stop_pc; /* give dbg stop_pc after ensuring brkpt trap */ + unsigned long efa; /* break pt addr, for break points in delay slots */ + unsigned long stop_pc; /* give dbg stop_pc after ensuring brkpt trap */ }; #endif /* !__ASSEMBLY__ */ From 10971638701dedadb58c88ce4d31c9375b224ed6 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Fri, 7 Aug 2015 13:01:39 +0530 Subject: [PATCH 15/15] ARCv2: spinlock/rwlock/atomics: reduce 1 instruction in exponential backoff The increment of delay counter was 2 instructions: Arithmatic Shfit Left (ASL) + set to 1 on overflow This can be done in 1 using ROtate Left (ROL) Suggested-by: Nigel Topham Cc: Peter Zijlstra (Intel) Cc: linux-kernel@vger.kernel.org Signed-off-by: Vineet Gupta --- arch/arc/include/asm/atomic.h | 3 +-- arch/arc/include/asm/spinlock.h | 3 +-- 2 files changed, 2 insertions(+), 4 deletions(-) diff --git a/arch/arc/include/asm/atomic.h b/arch/arc/include/asm/atomic.h index 629dfd0a0c6b..87d18ae53115 100644 --- a/arch/arc/include/asm/atomic.h +++ b/arch/arc/include/asm/atomic.h @@ -34,8 +34,7 @@ " mov %[tmp], %[delay] \n" /* tmp = delay */ \ "2: brne.d %[tmp], 0, 2b \n" /* while (tmp != 0) */ \ " sub %[tmp], %[tmp], 1 \n" /* tmp-- */ \ - " asl.f %[delay], %[delay], 1 \n" /* delay *= 2 */ \ - " mov.z %[delay], 1 \n" /* handle overflow */ \ + " rol %[delay], %[delay] \n" /* delay *= 2 */ \ " b 1b \n" /* start over */ \ "4: ; --- success --- \n" \ diff --git a/arch/arc/include/asm/spinlock.h b/arch/arc/include/asm/spinlock.h index 7071fc0da56a..db8c59d1eaeb 100644 --- a/arch/arc/include/asm/spinlock.h +++ b/arch/arc/include/asm/spinlock.h @@ -260,8 +260,7 @@ static inline void arch_write_unlock(arch_rwlock_t *rw) " mov %[tmp], %[delay] \n" /* tmp = delay */ \ "2: brne.d %[tmp], 0, 2b \n" /* while (tmp != 0) */ \ " sub %[tmp], %[tmp], 1 \n" /* tmp-- */ \ - " asl.f %[delay], %[delay], 1 \n" /* delay *= 2 */ \ - " mov.z %[delay], 1 \n" /* handle overflow */ \ + " rol %[delay], %[delay] \n" /* delay *= 2 */ \ " b 1b \n" /* start over */ \ " \n" \ "4: ; --- done --- \n" \