From 5a425294ee7d4ab5a374248e85838dfd450caf75 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 Jun 2012 15:30:31 +0200 Subject: perf/x86: Fix Intel shared extra MSR allocation Zheng Yan reported that event group validation can wreck event state when Intel extra_reg allocation changes event state. Validation shouldn't change any persistent state. Cloning events in validate_{event,group}() isn't really pretty either, so add a few special cases to avoid modifying the event state. The code is restructured to minimize the special case impact. Reported-by: Zheng Yan Acked-by: Stephane Eranian Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1338903031.28282.175.camel@twins Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 1 + 1 file changed, 1 insertion(+) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 6638aaf54493..83794d8e6af0 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -117,6 +117,7 @@ struct cpu_hw_events { struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */ unsigned int group_flag; + int is_fake; /* * Intel DebugStore bits -- cgit From 0780c927a02492f917a74f51f3c801c76a637c57 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 Jun 2012 10:26:43 +0200 Subject: perf/x86: Implement cycles:p for SNB/IVB Now that there's finally a chip with working PEBS (IvyBridge), we can enable the hardware and implement cycles:p for SNB/IVB. Cc: Stephane Eranian Requested-and-tested-by: Linus Torvalds Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1338884803.28282.153.camel@twins Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 1 + 1 file changed, 1 insertion(+) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 83794d8e6af0..7241e2fc3c17 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -365,6 +365,7 @@ struct x86_pmu { int pebs_record_size; void (*drain_pebs)(struct pt_regs *regs); struct event_constraint *pebs_constraints; + void (*pebs_aliases)(struct perf_event *event); /* * Intel LBR -- cgit From 1c2ac3fde3e35279958e7b0408e2dcf866465301 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 14 May 2012 15:25:34 +0200 Subject: perf/x86: Fix wrmsrl() debug wrapper Move the wrmslr() debug wrapper to the common header now that all the include games are gone. Also clean it up a bit to avoid multiple evaluation of the argument. Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/n/tip-l4gkfnivwv4yi5mqxjlovymx@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 7241e2fc3c17..23b5710b1747 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -14,6 +14,18 @@ #include +#if 0 +#undef wrmsrl +#define wrmsrl(msr, val) \ +do { \ + unsigned int _msr = (msr); \ + u64 _val = (val); \ + trace_printk("wrmsrl(%x, %Lx)\n", (unsigned int)(_msr), \ + (unsigned long long)(_val)); \ + native_write_msr((_msr), (u32)(_val), (u32)(_val >> 32)); \ +} while (0) +#endif + /* * | NHM/WSM | SNB | * register ------------------------------- -- cgit From 70ab7003dec58afeae7f5d681dfa309b3a259f03 Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Tue, 5 Jun 2012 17:56:48 -0700 Subject: perf/x86: Don't assume there can be only 4 PEBS events On Sandy Bridge in non HT mode there are 8 counters available. Since every counter can write a PEBS record assuming there are 4 max is incorrect. Use the reported counter number -- with an upper limit for a static array -- instead. Also I made the warning messages a bit more informational. Signed-off-by: Andi Kleen Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1338944211-28275-2-git-send-email-andi@firstfloor.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 23b5710b1747..3df3de9452a9 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -69,7 +69,7 @@ struct amd_nb { }; /* The maximal number of PEBS events: */ -#define MAX_PEBS_EVENTS 4 +#define MAX_PEBS_EVENTS 8 /* * A debug store configuration. @@ -378,6 +378,7 @@ struct x86_pmu { void (*drain_pebs)(struct pt_regs *regs); struct event_constraint *pebs_constraints; void (*pebs_aliases)(struct perf_event *event); + int max_pebs_events; /* * Intel LBR -- cgit From 4b4969b14490a4f65b572b8f180164181104b5e1 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Fri, 15 Jun 2012 14:31:30 +0800 Subject: perf: Export perf_assign_events() Export perf_assign_events() so the uncore code can use it to schedule events. Signed-off-by: Zheng Yan Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1339741902-8449-2-git-send-email-zheng.z.yan@intel.com Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 3df3de9452a9..83238f2a12b2 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -481,6 +481,8 @@ static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc, void x86_pmu_enable_all(int added); +int perf_assign_events(struct event_constraint **constraints, int n, + int wmin, int wmax, int *assign); int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign); void x86_pmu_stop(struct perf_event *event, int flags); -- cgit From c93dc84cbe32435be3ffa2fbde355eff94955c32 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 8 Jun 2012 14:50:50 +0200 Subject: perf/x86: Add a microcode revision check for SNB-PEBS Recent Intel microcode resolved the SNB-PEBS issues, so conditionally enable PEBS on SNB hardware depending on the microcode revision. Thanks to Stephane for figuring out the various microcode revisions. Suggested-by: Stephane Eranian Acked-by: Borislav Petkov Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/n/tip-v3672ziwh9damwqwh1uz3krm@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 83238f2a12b2..3f5c66904355 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -361,6 +361,8 @@ struct x86_pmu { void (*cpu_starting)(int cpu); void (*cpu_dying)(int cpu); void (*cpu_dead)(int cpu); + + void (*check_microcode)(void); void (*flush_branch_stack)(void); /* @@ -373,7 +375,7 @@ struct x86_pmu { * Intel DebugStore bits */ int bts, pebs; - int bts_active, pebs_active; + int bts_active, pebs_active, pebs_broken; int pebs_record_size; void (*drain_pebs)(struct pt_regs *regs); struct event_constraint *pebs_constraints; -- cgit From 3e0091e2b6f8cd59e567f247e345a3a6ad1f6e7e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 26 Jun 2012 23:38:39 +0200 Subject: perf/x86: Save a few bytes in 'struct x86_pmu' All these are basically boolean flags, use a bitfield to save a few bytes. Suggested-by: Borislav Petkov Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/n/tip-vsevd5g8lhcn129n3s7trl7r@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 3f5c66904355..a15df4be151f 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -374,8 +374,11 @@ struct x86_pmu { /* * Intel DebugStore bits */ - int bts, pebs; - int bts_active, pebs_active, pebs_broken; + int bts :1, + bts_active :1, + pebs :1, + pebs_active :1, + pebs_broken :1; int pebs_record_size; void (*drain_pebs)(struct pt_regs *regs); struct event_constraint *pebs_constraints; -- cgit From 597ed953d7db28528b4687e46388c1aa905c14bd Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 9 Jul 2012 13:50:23 +0200 Subject: perf/x86: Make bitfield unsigned Fix: arch/x86/kernel/cpu/perf_event.h:377:43: sparse: dubious one-bit signed bitfield Cc: Borislav Petkov Reported-by: Fengguang Wu Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/n/tip-2jxkmktkppkclj1qe6qxd7ah@git.kernel.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index a15df4be151f..821d53b696d1 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -374,7 +374,7 @@ struct x86_pmu { /* * Intel DebugStore bits */ - int bts :1, + unsigned int bts :1, bts_active :1, pebs :1, pebs_active :1, -- cgit From d07bdfd322d307789f15b427dbcc39257665356f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 10 Jul 2012 09:42:15 +0200 Subject: perf/x86: Fix USER/KERNEL tagging of samples properly Some PMUs don't provide a full register set for their sample, specifically 'advanced' PMUs like AMD IBS and Intel PEBS which provide 'better' than regular interrupt accuracy. In this case we use the interrupt regs as basis and over-write some fields (typically IP) with different information. The perf core however uses user_mode() to distinguish user/kernel samples, user_mode() relies on regs->cs. If the interrupt skid pushed us over a boundary the new IP might not be in the same domain as the interrupt. Commit ce5c1fe9a9e ("perf/x86: Fix USER/KERNEL tagging of samples") tried to fix this by making the perf core use kernel_ip(). This however is wrong (TM), as pointed out by Linus, since it doesn't allow for VM86 and non-zero based segments in IA32 mode. Therefore, provide a new helper to set the regs->ip field, set_linear_ip(), which massages the regs into a suitable state assuming the provided IP is in fact a linear address. Also modify perf_instruction_pointer() and perf_callchain_user() to deal with segments base offsets. Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1341910954.3462.102.camel@twins Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event.h | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) (limited to 'arch/x86/kernel/cpu/perf_event.h') diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index 821d53b696d1..6605a81ba339 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -516,6 +516,26 @@ static inline bool kernel_ip(unsigned long ip) #endif } +/* + * Not all PMUs provide the right context information to place the reported IP + * into full context. Specifically segment registers are typically not + * supplied. + * + * Assuming the address is a linear address (it is for IBS), we fake the CS and + * vm86 mode using the known zero-based code segment and 'fix up' the registers + * to reflect this. + * + * Intel PEBS/LBR appear to typically provide the effective address, nothing + * much we can do about that but pray and treat it like a linear address. + */ +static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip) +{ + regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS; + if (regs->flags & X86_VM_MASK) + regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK); + regs->ip = ip; +} + #ifdef CONFIG_CPU_SUP_AMD int amd_pmu_init(void); -- cgit