From 99dd5497e5be4fe4194cad181d45fd6569a930db Mon Sep 17 00:00:00 2001 From: "Liu, Chuansheng" Date: Mon, 26 Mar 2012 07:11:50 +0000 Subject: [PATCH 1/4] x86: Preserve lazy irq disable semantics in fixup_irqs() The default irq_disable() sematics are to mark the interrupt disabled, but keep it unmasked. If the interrupt is delivered while marked disabled, the low level interrupt handler masks it and marks it pending. This is important for detecting wakeup interrupts during suspend and for edge type interrupts to avoid losing interrupts. fixup_irqs() moves the interrupts away from an offlined cpu. For certain interrupt types it needs to mask the interrupt line before changing the affinity. After affinity has changed the interrupt line is unmasked again, but only if it is not marked disabled. This breaks the lazy irq disable semantics and causes problems in suspend as the interrupt can be lost or wakeup functionality is broken. Check irqd_irq_masked() instead of irqd_irq_disabled() because irqd_irq_masked() is only set, when the core code actually masked the interrupt line. If it's not set, we unmask the interrupt and let the lazy irq disable logic deal with an eventually incoming interrupt. [ tglx: Massaged changelog and added a comment ] Signed-off-by: liu chuansheng Cc: Yanmin Zhang Link: http://lkml.kernel.org/r/27240C0AC20F114CBF8149A2696CBE4A05DFB3@SHSMSX101.ccr.corp.intel.com Signed-off-by: Thomas Gleixner --- arch/x86/kernel/irq.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/arch/x86/kernel/irq.c b/arch/x86/kernel/irq.c index 7943e0c21bde..3dafc6003b7c 100644 --- a/arch/x86/kernel/irq.c +++ b/arch/x86/kernel/irq.c @@ -282,8 +282,13 @@ void fixup_irqs(void) else if (!(warned++)) set_affinity = 0; + /* + * We unmask if the irq was not marked masked by the + * core code. That respects the lazy irq disable + * behaviour. + */ if (!irqd_can_move_in_process_context(data) && - !irqd_irq_disabled(data) && chip->irq_unmask) + !irqd_irq_masked(data) && chip->irq_unmask) chip->irq_unmask(data); raw_spin_unlock(&desc->lock); From f6365201d8a21fb347260f89d6e9b3e718d63c70 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Thu, 29 Mar 2012 14:49:17 -0700 Subject: [PATCH 2/4] x86: Remove the ancient and deprecated disable_hlt() and enable_hlt() facility The X86_32-only disable_hlt/enable_hlt mechanism was used by the 32-bit floppy driver. Its effect was to replace the use of the HLT instruction inside default_idle() with cpu_relax() - essentially it turned off the use of HLT. This workaround was commented in the code as: "disable hlt during certain critical i/o operations" "This halt magic was a workaround for ancient floppy DMA wreckage. It should be safe to remove." H. Peter Anvin additionally adds: "To the best of my knowledge, no-hlt only existed because of flaky power distributions on 386/486 systems which were sold to run DOS. Since DOS did no power management of any kind, including HLT, the power draw was fairly uniform; when exposed to the much hhigher noise levels you got when Linux used HLT caused some of these systems to fail. They were by far in the minority even back then." Alan Cox further says: "Also for the Cyrix 5510 which tended to go castors up if a HLT occurred during a DMA cycle and on a few other boxes HLT during DMA tended to go astray. Do we care ? I doubt it. The 5510 was pretty obscure, the 5520 fixed it, the 5530 is probably the oldest still in any kind of use." So, let's finally drop this. Signed-off-by: Len Brown Signed-off-by: Josh Boyer Signed-off-by: Andrew Morton Acked-by: "H. Peter Anvin" Acked-by: Alan Cox Cc: Stephen Hemminger Cc: Link: http://lkml.kernel.org/n/tip-3rhk9bzf0x9rljkv488tloib@git.kernel.org [ If anyone cares then alternative instruction patching could be used to replace HLT with a one-byte NOP instruction. Much simpler. ] Signed-off-by: Ingo Molnar --- Documentation/feature-removal-schedule.txt | 8 ----- arch/x86/include/asm/processor.h | 10 ------ arch/x86/kernel/process.c | 24 --------------- drivers/block/floppy.c | 36 ---------------------- 4 files changed, 78 deletions(-) diff --git a/Documentation/feature-removal-schedule.txt b/Documentation/feature-removal-schedule.txt index 0cad4803ffac..7c950d48d76e 100644 --- a/Documentation/feature-removal-schedule.txt +++ b/Documentation/feature-removal-schedule.txt @@ -6,14 +6,6 @@ be removed from this file. --------------------------- -What: x86 floppy disable_hlt -When: 2012 -Why: ancient workaround of dubious utility clutters the - code used by everybody else. -Who: Len Brown - ---------------------------- - What: CONFIG_APM_CPU_IDLE, and its ability to call APM BIOS in idle When: 2012 Why: This optional sub-feature of APM is of dubious reliability, diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h index 7284c9a6a0b5..4fa7dcceb6c0 100644 --- a/arch/x86/include/asm/processor.h +++ b/arch/x86/include/asm/processor.h @@ -974,16 +974,6 @@ extern bool cpu_has_amd_erratum(const int *); #define cpu_has_amd_erratum(x) (false) #endif /* CONFIG_CPU_SUP_AMD */ -#ifdef CONFIG_X86_32 -/* - * disable hlt during certain critical i/o operations - */ -#define HAVE_DISABLE_HLT -#endif - -void disable_hlt(void); -void enable_hlt(void); - void cpu_idle_wait(void); extern unsigned long arch_align_stack(unsigned long sp); diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c index a33afaa5ddb7..1d92a5ab6e8b 100644 --- a/arch/x86/kernel/process.c +++ b/arch/x86/kernel/process.c @@ -362,34 +362,10 @@ void (*pm_idle)(void); EXPORT_SYMBOL(pm_idle); #endif -#ifdef CONFIG_X86_32 -/* - * This halt magic was a workaround for ancient floppy DMA - * wreckage. It should be safe to remove. - */ -static int hlt_counter; -void disable_hlt(void) -{ - hlt_counter++; -} -EXPORT_SYMBOL(disable_hlt); - -void enable_hlt(void) -{ - hlt_counter--; -} -EXPORT_SYMBOL(enable_hlt); - -static inline int hlt_use_halt(void) -{ - return (!hlt_counter && boot_cpu_data.hlt_works_ok); -} -#else static inline int hlt_use_halt(void) { return 1; } -#endif #ifndef CONFIG_SMP static inline void play_dead(void) diff --git a/drivers/block/floppy.c b/drivers/block/floppy.c index 76a08236430a..b0b00d70c166 100644 --- a/drivers/block/floppy.c +++ b/drivers/block/floppy.c @@ -1030,37 +1030,6 @@ static int fd_wait_for_completion(unsigned long delay, timeout_fn function) return 0; } -static DEFINE_SPINLOCK(floppy_hlt_lock); -static int hlt_disabled; -static void floppy_disable_hlt(void) -{ - unsigned long flags; - - WARN_ONCE(1, "floppy_disable_hlt() scheduled for removal in 2012"); - spin_lock_irqsave(&floppy_hlt_lock, flags); - if (!hlt_disabled) { - hlt_disabled = 1; -#ifdef HAVE_DISABLE_HLT - disable_hlt(); -#endif - } - spin_unlock_irqrestore(&floppy_hlt_lock, flags); -} - -static void floppy_enable_hlt(void) -{ - unsigned long flags; - - spin_lock_irqsave(&floppy_hlt_lock, flags); - if (hlt_disabled) { - hlt_disabled = 0; -#ifdef HAVE_DISABLE_HLT - enable_hlt(); -#endif - } - spin_unlock_irqrestore(&floppy_hlt_lock, flags); -} - static void setup_DMA(void) { unsigned long f; @@ -1105,7 +1074,6 @@ static void setup_DMA(void) fd_enable_dma(); release_dma_lock(f); #endif - floppy_disable_hlt(); } static void show_floppy(void); @@ -1707,7 +1675,6 @@ irqreturn_t floppy_interrupt(int irq, void *dev_id) fd_disable_dma(); release_dma_lock(f); - floppy_enable_hlt(); do_floppy = NULL; if (fdc >= N_FDC || FDCS->address == -1) { /* we don't even know which FDC is the culprit */ @@ -1856,8 +1823,6 @@ static void floppy_shutdown(unsigned long data) show_floppy(); cancel_activity(); - floppy_enable_hlt(); - flags = claim_dma_lock(); fd_disable_dma(); release_dma_lock(flags); @@ -4508,7 +4473,6 @@ static void floppy_release_irq_and_dma(void) #if N_FDC > 1 set_dor(1, ~8, 0); #endif - floppy_enable_hlt(); if (floppy_track_buffer && max_buffer_sectors) { tmpsize = max_buffer_sectors * 1024; From c0e9afc0da6cb0f11497e5ea83377b3c451450e0 Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Wed, 28 Mar 2012 11:51:17 -0700 Subject: [PATCH 3/4] x86: Use -mno-avx when available On gccs that support AVX it's a good idea to disable that too, similar to how SSE2, SSE1 etc. are already disabled. This prevents the compiler from generating AVX ever implicitely. No failure observed, just from review. [ hpa: Marking this for urgent and stable, simply because the patch will either have absolutely no effect *or* it will avoid potentially very hard to debug failures. ] Signed-off-by: Andi Kleen Link: http://lkml.kernel.org/r/1332960678-11879-1-git-send-email-andi@firstfloor.org Signed-off-by: H. Peter Anvin Cc: --- arch/x86/Makefile | 1 + 1 file changed, 1 insertion(+) diff --git a/arch/x86/Makefile b/arch/x86/Makefile index 968dbe24a255..41a7237606a3 100644 --- a/arch/x86/Makefile +++ b/arch/x86/Makefile @@ -129,6 +129,7 @@ KBUILD_CFLAGS += -Wno-sign-compare KBUILD_CFLAGS += -fno-asynchronous-unwind-tables # prevent gcc from generating any FP code by mistake KBUILD_CFLAGS += $(call cc-option,-mno-sse -mno-mmx -mno-sse2 -mno-3dnow,) +KBUILD_CFLAGS += $(call cc-option,-mno-avx,) KBUILD_CFLAGS += $(mflags-y) KBUILD_AFLAGS += $(mflags-y) From dba69d1092e291e257fb5673a3ad0e4c87878ebc Mon Sep 17 00:00:00 2001 From: Marcelo Tosatti Date: Sun, 1 Apr 2012 13:53:36 -0300 Subject: [PATCH 4/4] x86, kvm: Call restore_sched_clock_state() only after %gs is initialized s2ram broke due to this KVM commit: b74f05d61b73 x86: kvmclock: abstract save/restore sched_clock_state restore_sched_clock_state() methods use percpu data, therefore they must run after %gs is initialized, but before mtrr_bp_restore() (due to lockstat using sched_clock). Move it to the correct place. Reported-and-tested-by: Konstantin Khlebnikov Signed-off-by: Marcelo Tosatti Cc: Avi Kivity Signed-off-by: Ingo Molnar --- arch/x86/power/cpu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/arch/x86/power/cpu.c b/arch/x86/power/cpu.c index 47936830968c..218cdb16163c 100644 --- a/arch/x86/power/cpu.c +++ b/arch/x86/power/cpu.c @@ -225,13 +225,13 @@ static void __restore_processor_state(struct saved_context *ctxt) fix_processor_context(); do_fpu_end(); + x86_platform.restore_sched_clock_state(); mtrr_bp_restore(); } /* Needed by apm.c */ void restore_processor_state(void) { - x86_platform.restore_sched_clock_state(); __restore_processor_state(&saved_context); } #ifdef CONFIG_X86_32