From b0f82b81fe6bbcf78d478071f33e44554726bc81 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Thu, 20 May 2010 07:47:21 +0200 Subject: perf: Drop the skip argument from perf_arch_fetch_regs_caller Drop this argument now that we always want to rewind only to the state of the first caller. It means frame pointers are not necessary anymore to reliably get the source of an event. But this also means we need this helper to be a macro now, as an inline function is not an option since we need to know when to provide a default implentation. Signed-off-by: Frederic Weisbecker Signed-off-by: Paul Mackerras Cc: David Miller Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo --- include/linux/perf_event.h | 32 +++++++------------------------- 1 file changed, 7 insertions(+), 25 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index fb6c91eac7e3..bea785cef493 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -905,8 +905,10 @@ extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64); -extern void -perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip); +#ifndef perf_arch_fetch_caller_regs +static inline void +perf_arch_fetch_caller_regs(struct regs *regs, unsigned long ip) { } +#endif /* * Take a snapshot of the regs. Skip ip and frame pointer to @@ -916,31 +918,11 @@ perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip); * - bp for callchains * - eflags, for future purposes, just in case */ -static inline void perf_fetch_caller_regs(struct pt_regs *regs, int skip) +static inline void perf_fetch_caller_regs(struct pt_regs *regs) { - unsigned long ip; - memset(regs, 0, sizeof(*regs)); - switch (skip) { - case 1 : - ip = CALLER_ADDR0; - break; - case 2 : - ip = CALLER_ADDR1; - break; - case 3 : - ip = CALLER_ADDR2; - break; - case 4: - ip = CALLER_ADDR3; - break; - /* No need to support further for now */ - default: - ip = 0; - } - - return perf_arch_fetch_caller_regs(regs, ip, skip); + perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); } static inline void @@ -950,7 +932,7 @@ perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) struct pt_regs hot_regs; if (!regs) { - perf_fetch_caller_regs(&hot_regs, 1); + perf_fetch_caller_regs(&hot_regs); regs = &hot_regs; } __perf_sw_event(event_id, nr, nmi, regs, addr); -- cgit v1.2.3 From ecc55f84b2e9741f29daa787ded93986df6cbe17 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 21 May 2010 15:11:34 +0200 Subject: perf, trace: Inline perf_swevent_put_recursion_context() Inline perf_swevent_put_recursion_context into perf_tp_event(), this shrinks the per trace template code footprint and saves a function call. Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/ftrace_event.h | 3 +-- include/linux/perf_event.h | 2 +- kernel/perf_event.c | 8 ++++---- 3 files changed, 6 insertions(+), 7 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index 3167f2df4126..0af31cd335d6 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -257,8 +257,7 @@ static inline void perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr, u64 count, struct pt_regs *regs, void *head) { - perf_tp_event(addr, count, raw_data, size, regs, head); - perf_swevent_put_recursion_context(rctx); + perf_tp_event(addr, count, raw_data, size, regs, head, rctx); } #endif diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 5d0266d94985..c691a0b27bcd 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -1001,7 +1001,7 @@ static inline bool perf_paranoid_kernel(void) extern void perf_event_init(void); extern void perf_tp_event(u64 addr, u64 count, void *record, int entry_size, struct pt_regs *regs, - struct hlist_head *head); + struct hlist_head *head, int rctx); extern void perf_bp_event(struct perf_event *event, void *data); #ifndef perf_misc_flags diff --git a/kernel/perf_event.c b/kernel/perf_event.c index ff86c558af4c..4bd3b597bcca 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -4213,14 +4213,12 @@ int perf_swevent_get_recursion_context(void) } EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context); -void perf_swevent_put_recursion_context(int rctx) +void inline perf_swevent_put_recursion_context(int rctx) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); barrier(); cpuctx->recursion[rctx]--; } -EXPORT_SYMBOL_GPL(perf_swevent_put_recursion_context); - void __perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) @@ -4601,7 +4599,7 @@ static int perf_tp_event_match(struct perf_event *event, } void perf_tp_event(u64 addr, u64 count, void *record, int entry_size, - struct pt_regs *regs, struct hlist_head *head) + struct pt_regs *regs, struct hlist_head *head, int rctx) { struct perf_sample_data data; struct perf_event *event; @@ -4621,6 +4619,8 @@ void perf_tp_event(u64 addr, u64 count, void *record, int entry_size, perf_swevent_add(event, count, 1, &data, regs); } rcu_read_unlock(); + + perf_swevent_put_recursion_context(rctx); } EXPORT_SYMBOL_GPL(perf_tp_event); -- cgit v1.2.3 From 3af9e859281bda7eb7c20b51879cf43aa788ac2e Mon Sep 17 00:00:00 2001 From: Eric B Munson Date: Tue, 18 May 2010 15:30:49 +0100 Subject: perf: Add non-exec mmap() tracking Add the capacility to track data mmap()s. This can be used together with PERF_SAMPLE_ADDR for data profiling. Signed-off-by: Anton Blanchard [Updated code for stable perf ABI] Signed-off-by: Eric B Munson Signed-off-by: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker Cc: Paul Mackerras Cc: Mike Galbraith Cc: Steven Rostedt LKML-Reference: <1274193049-25997-1-git-send-email-ebmunson@us.ibm.com> Signed-off-by: Ingo Molnar --- fs/exec.c | 1 + include/linux/perf_event.h | 12 +++--------- kernel/perf_event.c | 34 +++++++++++++++++++++++----------- mm/mmap.c | 6 +++++- tools/perf/builtin-record.c | 4 +++- 5 files changed, 35 insertions(+), 22 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/fs/exec.c b/fs/exec.c index e19de6a80339..97d91a03fb13 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -653,6 +653,7 @@ int setup_arg_pages(struct linux_binprm *bprm, else stack_base = vma->vm_start - stack_expand; #endif + current->mm->start_stack = bprm->p; ret = expand_stack(vma, stack_base); if (ret) ret = -EFAULT; diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index c691a0b27bcd..36efad90cd43 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -214,8 +214,9 @@ struct perf_event_attr { * See also PERF_RECORD_MISC_EXACT_IP */ precise_ip : 2, /* skid constraint */ + mmap_data : 1, /* non-exec mmap data */ - __reserved_1 : 47; + __reserved_1 : 46; union { __u32 wakeup_events; /* wakeup every n events */ @@ -962,14 +963,7 @@ perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) } } -extern void __perf_event_mmap(struct vm_area_struct *vma); - -static inline void perf_event_mmap(struct vm_area_struct *vma) -{ - if (vma->vm_flags & VM_EXEC) - __perf_event_mmap(vma); -} - +extern void perf_event_mmap(struct vm_area_struct *vma); extern struct perf_guest_info_callbacks *perf_guest_cbs; extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); diff --git a/kernel/perf_event.c b/kernel/perf_event.c index b39bec346e80..227ed9c8ec34 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1891,7 +1891,7 @@ static void free_event(struct perf_event *event) if (!event->parent) { atomic_dec(&nr_events); - if (event->attr.mmap) + if (event->attr.mmap || event->attr.mmap_data) atomic_dec(&nr_mmap_events); if (event->attr.comm) atomic_dec(&nr_comm_events); @@ -3491,7 +3491,7 @@ perf_event_read_event(struct perf_event *event, /* * task tracking -- fork/exit * - * enabled by: attr.comm | attr.mmap | attr.task + * enabled by: attr.comm | attr.mmap | attr.mmap_data | attr.task */ struct perf_task_event { @@ -3541,7 +3541,8 @@ static int perf_event_task_match(struct perf_event *event) if (event->cpu != -1 && event->cpu != smp_processor_id()) return 0; - if (event->attr.comm || event->attr.mmap || event->attr.task) + if (event->attr.comm || event->attr.mmap || + event->attr.mmap_data || event->attr.task) return 1; return 0; @@ -3766,7 +3767,8 @@ static void perf_event_mmap_output(struct perf_event *event, } static int perf_event_mmap_match(struct perf_event *event, - struct perf_mmap_event *mmap_event) + struct perf_mmap_event *mmap_event, + int executable) { if (event->state < PERF_EVENT_STATE_INACTIVE) return 0; @@ -3774,19 +3776,21 @@ static int perf_event_mmap_match(struct perf_event *event, if (event->cpu != -1 && event->cpu != smp_processor_id()) return 0; - if (event->attr.mmap) + if ((!executable && event->attr.mmap_data) || + (executable && event->attr.mmap)) return 1; return 0; } static void perf_event_mmap_ctx(struct perf_event_context *ctx, - struct perf_mmap_event *mmap_event) + struct perf_mmap_event *mmap_event, + int executable) { struct perf_event *event; list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { - if (perf_event_mmap_match(event, mmap_event)) + if (perf_event_mmap_match(event, mmap_event, executable)) perf_event_mmap_output(event, mmap_event); } } @@ -3830,6 +3834,14 @@ static void perf_event_mmap_event(struct perf_mmap_event *mmap_event) if (!vma->vm_mm) { name = strncpy(tmp, "[vdso]", sizeof(tmp)); goto got_name; + } else if (vma->vm_start <= vma->vm_mm->start_brk && + vma->vm_end >= vma->vm_mm->brk) { + name = strncpy(tmp, "[heap]", sizeof(tmp)); + goto got_name; + } else if (vma->vm_start <= vma->vm_mm->start_stack && + vma->vm_end >= vma->vm_mm->start_stack) { + name = strncpy(tmp, "[stack]", sizeof(tmp)); + goto got_name; } name = strncpy(tmp, "//anon", sizeof(tmp)); @@ -3846,17 +3858,17 @@ got_name: rcu_read_lock(); cpuctx = &get_cpu_var(perf_cpu_context); - perf_event_mmap_ctx(&cpuctx->ctx, mmap_event); + perf_event_mmap_ctx(&cpuctx->ctx, mmap_event, vma->vm_flags & VM_EXEC); ctx = rcu_dereference(current->perf_event_ctxp); if (ctx) - perf_event_mmap_ctx(ctx, mmap_event); + perf_event_mmap_ctx(ctx, mmap_event, vma->vm_flags & VM_EXEC); put_cpu_var(perf_cpu_context); rcu_read_unlock(); kfree(buf); } -void __perf_event_mmap(struct vm_area_struct *vma) +void perf_event_mmap(struct vm_area_struct *vma) { struct perf_mmap_event mmap_event; @@ -4911,7 +4923,7 @@ done: if (!event->parent) { atomic_inc(&nr_events); - if (event->attr.mmap) + if (event->attr.mmap || event->attr.mmap_data) atomic_inc(&nr_mmap_events); if (event->attr.comm) atomic_inc(&nr_comm_events); diff --git a/mm/mmap.c b/mm/mmap.c index 456ec6f27889..e38e910cb756 100644 --- a/mm/mmap.c +++ b/mm/mmap.c @@ -1734,8 +1734,10 @@ int expand_upwards(struct vm_area_struct *vma, unsigned long address) grow = (address - vma->vm_end) >> PAGE_SHIFT; error = acct_stack_growth(vma, size, grow); - if (!error) + if (!error) { vma->vm_end = address; + perf_event_mmap(vma); + } } anon_vma_unlock(vma); return error; @@ -1781,6 +1783,7 @@ static int expand_downwards(struct vm_area_struct *vma, if (!error) { vma->vm_start = address; vma->vm_pgoff -= grow; + perf_event_mmap(vma); } } anon_vma_unlock(vma); @@ -2208,6 +2211,7 @@ unsigned long do_brk(unsigned long addr, unsigned long len) vma->vm_page_prot = vm_get_page_prot(flags); vma_link(mm, vma, prev, rb_link, rb_parent); out: + perf_event_mmap(vma); mm->total_vm += len >> PAGE_SHIFT; if (flags & VM_LOCKED) { if (!mlock_vma_pages_range(vma, addr, addr + len)) diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 5e5c6403a315..39c7247bc54a 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -268,8 +268,10 @@ static void create_counter(int counter, int cpu) if (inherit_stat) attr->inherit_stat = 1; - if (sample_address) + if (sample_address) { attr->sample_type |= PERF_SAMPLE_ADDR; + attr->mmap_data = track; + } if (call_graph) attr->sample_type |= PERF_SAMPLE_CALLCHAIN; -- cgit v1.2.3 From 8d2cacbbb8deadfae78aa16e4e1ee619bdd7019e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 25 May 2010 17:49:05 +0200 Subject: perf: Cleanup {start,commit,cancel}_txn details Clarify some of the transactional group scheduling API details and change it so that a successfull ->commit_txn also closes the transaction. Signed-off-by: Peter Zijlstra Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker Cc: Paul Mackerras Cc: Mike Galbraith Cc: Steven Rostedt LKML-Reference: <1274803086.5882.1752.camel@twins> Signed-off-by: Ingo Molnar --- arch/powerpc/kernel/perf_event.c | 7 ++++--- arch/sparc/kernel/perf_event.c | 7 ++++--- arch/x86/kernel/cpu/perf_event.c | 14 +++++--------- include/linux/perf_event.h | 27 ++++++++++++++++++++++----- kernel/perf_event.c | 9 +-------- 5 files changed, 36 insertions(+), 28 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/arch/powerpc/kernel/perf_event.c b/arch/powerpc/kernel/perf_event.c index 43b83c35cf54..ac2a8c2554d9 100644 --- a/arch/powerpc/kernel/perf_event.c +++ b/arch/powerpc/kernel/perf_event.c @@ -754,7 +754,7 @@ static int power_pmu_enable(struct perf_event *event) * skip the schedulability test here, it will be peformed * at commit time(->commit_txn) as a whole */ - if (cpuhw->group_flag & PERF_EVENT_TXN_STARTED) + if (cpuhw->group_flag & PERF_EVENT_TXN) goto nocheck; if (check_excludes(cpuhw->event, cpuhw->flags, n0, 1)) @@ -858,7 +858,7 @@ void power_pmu_start_txn(const struct pmu *pmu) { struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); - cpuhw->group_flag |= PERF_EVENT_TXN_STARTED; + cpuhw->group_flag |= PERF_EVENT_TXN; cpuhw->n_txn_start = cpuhw->n_events; } @@ -871,7 +871,7 @@ void power_pmu_cancel_txn(const struct pmu *pmu) { struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); - cpuhw->group_flag &= ~PERF_EVENT_TXN_STARTED; + cpuhw->group_flag &= ~PERF_EVENT_TXN; } /* @@ -897,6 +897,7 @@ int power_pmu_commit_txn(const struct pmu *pmu) for (i = cpuhw->n_txn_start; i < n; ++i) cpuhw->event[i]->hw.config = cpuhw->events[i]; + cpuhw->group_flag &= ~PERF_EVENT_TXN; return 0; } diff --git a/arch/sparc/kernel/perf_event.c b/arch/sparc/kernel/perf_event.c index 0ec92c8861dd..beeb92fa3acd 100644 --- a/arch/sparc/kernel/perf_event.c +++ b/arch/sparc/kernel/perf_event.c @@ -1005,7 +1005,7 @@ static int sparc_pmu_enable(struct perf_event *event) * skip the schedulability test here, it will be peformed * at commit time(->commit_txn) as a whole */ - if (cpuc->group_flag & PERF_EVENT_TXN_STARTED) + if (cpuc->group_flag & PERF_EVENT_TXN) goto nocheck; if (check_excludes(cpuc->event, n0, 1)) @@ -1102,7 +1102,7 @@ static void sparc_pmu_start_txn(const struct pmu *pmu) { struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); - cpuhw->group_flag |= PERF_EVENT_TXN_STARTED; + cpuhw->group_flag |= PERF_EVENT_TXN; } /* @@ -1114,7 +1114,7 @@ static void sparc_pmu_cancel_txn(const struct pmu *pmu) { struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); - cpuhw->group_flag &= ~PERF_EVENT_TXN_STARTED; + cpuhw->group_flag &= ~PERF_EVENT_TXN; } /* @@ -1137,6 +1137,7 @@ static int sparc_pmu_commit_txn(const struct pmu *pmu) if (sparc_check_constraints(cpuc->event, cpuc->events, n)) return -EAGAIN; + cpuc->group_flag &= ~PERF_EVENT_TXN; return 0; } diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 5db5b7d65a18..af04c6fa59cb 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -969,7 +969,7 @@ static int x86_pmu_enable(struct perf_event *event) * skip the schedulability test here, it will be peformed * at commit time(->commit_txn) as a whole */ - if (cpuc->group_flag & PERF_EVENT_TXN_STARTED) + if (cpuc->group_flag & PERF_EVENT_TXN) goto out; ret = x86_pmu.schedule_events(cpuc, n, assign); @@ -1096,7 +1096,7 @@ static void x86_pmu_disable(struct perf_event *event) * The events never got scheduled and ->cancel_txn will truncate * the event_list. */ - if (cpuc->group_flag & PERF_EVENT_TXN_STARTED) + if (cpuc->group_flag & PERF_EVENT_TXN) return; x86_pmu_stop(event); @@ -1388,7 +1388,7 @@ static void x86_pmu_start_txn(const struct pmu *pmu) { struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); - cpuc->group_flag |= PERF_EVENT_TXN_STARTED; + cpuc->group_flag |= PERF_EVENT_TXN; cpuc->n_txn = 0; } @@ -1401,7 +1401,7 @@ static void x86_pmu_cancel_txn(const struct pmu *pmu) { struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); - cpuc->group_flag &= ~PERF_EVENT_TXN_STARTED; + cpuc->group_flag &= ~PERF_EVENT_TXN; /* * Truncate the collected events. */ @@ -1435,11 +1435,7 @@ static int x86_pmu_commit_txn(const struct pmu *pmu) */ memcpy(cpuc->assign, assign, n*sizeof(int)); - /* - * Clear out the txn count so that ->cancel_txn() which gets - * run after ->commit_txn() doesn't undo things. - */ - cpuc->n_txn = 0; + cpuc->group_flag &= ~PERF_EVENT_TXN; return 0; } diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 36efad90cd43..f1b6ba0770e0 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -549,7 +549,10 @@ struct hw_perf_event { struct perf_event; -#define PERF_EVENT_TXN_STARTED 1 +/* + * Common implementation detail of pmu::{start,commit,cancel}_txn + */ +#define PERF_EVENT_TXN 0x1 /** * struct pmu - generic performance monitoring unit @@ -563,14 +566,28 @@ struct pmu { void (*unthrottle) (struct perf_event *event); /* - * group events scheduling is treated as a transaction, - * add group events as a whole and perform one schedulability test. - * If test fails, roll back the whole group + * Group events scheduling is treated as a transaction, add group + * events as a whole and perform one schedulability test. If the test + * fails, roll back the whole group */ + /* + * Start the transaction, after this ->enable() doesn't need + * to do schedulability tests. + */ void (*start_txn) (const struct pmu *pmu); - void (*cancel_txn) (const struct pmu *pmu); + /* + * If ->start_txn() disabled the ->enable() schedulability test + * then ->commit_txn() is required to perform one. On success + * the transaction is closed. On error the transaction is kept + * open until ->cancel_txn() is called. + */ int (*commit_txn) (const struct pmu *pmu); + /* + * Will cancel the transaction, assumes ->disable() is called for + * each successfull ->enable() during the transaction. + */ + void (*cancel_txn) (const struct pmu *pmu); }; /** diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 227ed9c8ec34..6f60920772b3 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -675,7 +675,6 @@ group_sched_in(struct perf_event *group_event, struct perf_event *event, *partial_group = NULL; const struct pmu *pmu = group_event->pmu; bool txn = false; - int ret; if (group_event->state == PERF_EVENT_STATE_OFF) return 0; @@ -703,15 +702,9 @@ group_sched_in(struct perf_event *group_event, } } - if (!txn) + if (!txn || !pmu->commit_txn(pmu)) return 0; - ret = pmu->commit_txn(pmu); - if (!ret) { - pmu->cancel_txn(pmu); - return 0; - } - group_error: /* * Groups can be scheduled in as one unit only, so undo any -- cgit v1.2.3 From ca5135e6b4a3cbc7e187737520fbc4b508f6f7a2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 28 May 2010 19:33:23 +0200 Subject: perf: Rename perf_mmap_data to perf_buffer Rename to clarify code. s/perf_mmap_data/perf_buffer/g and selective s/data/buffer/g Signed-off-by: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker Cc: Paul Mackerras Cc: Mike Galbraith Cc: Steven Rostedt LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 6 +- kernel/perf_event.c | 308 ++++++++++++++++++++++----------------------- 2 files changed, 157 insertions(+), 157 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index f1b6ba0770e0..2a0da021c23f 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -602,7 +602,7 @@ enum perf_event_active_state { struct file; -struct perf_mmap_data { +struct perf_buffer { atomic_t refcount; struct rcu_head rcu_head; #ifdef CONFIG_PERF_USE_VMALLOC @@ -727,7 +727,7 @@ struct perf_event { atomic_t mmap_count; int mmap_locked; struct user_struct *mmap_user; - struct perf_mmap_data *data; + struct perf_buffer *buffer; /* poll related */ wait_queue_head_t waitq; @@ -825,7 +825,7 @@ struct perf_cpu_context { struct perf_output_handle { struct perf_event *event; - struct perf_mmap_data *data; + struct perf_buffer *buffer; unsigned long wakeup; unsigned long size; void *addr; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 6f60920772b3..93d545801e43 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1876,7 +1876,7 @@ static void free_event_rcu(struct rcu_head *head) } static void perf_pending_sync(struct perf_event *event); -static void perf_mmap_data_put(struct perf_mmap_data *data); +static void perf_buffer_put(struct perf_buffer *buffer); static void free_event(struct perf_event *event) { @@ -1892,9 +1892,9 @@ static void free_event(struct perf_event *event) atomic_dec(&nr_task_events); } - if (event->data) { - perf_mmap_data_put(event->data); - event->data = NULL; + if (event->buffer) { + perf_buffer_put(event->buffer); + event->buffer = NULL; } if (event->destroy) @@ -2119,13 +2119,13 @@ perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) static unsigned int perf_poll(struct file *file, poll_table *wait) { struct perf_event *event = file->private_data; - struct perf_mmap_data *data; + struct perf_buffer *buffer; unsigned int events = POLL_HUP; rcu_read_lock(); - data = rcu_dereference(event->data); - if (data) - events = atomic_xchg(&data->poll, 0); + buffer = rcu_dereference(event->buffer); + if (buffer) + events = atomic_xchg(&buffer->poll, 0); rcu_read_unlock(); poll_wait(file, &event->waitq, wait); @@ -2335,14 +2335,14 @@ static int perf_event_index(struct perf_event *event) void perf_event_update_userpage(struct perf_event *event) { struct perf_event_mmap_page *userpg; - struct perf_mmap_data *data; + struct perf_buffer *buffer; rcu_read_lock(); - data = rcu_dereference(event->data); - if (!data) + buffer = rcu_dereference(event->buffer); + if (!buffer) goto unlock; - userpg = data->user_page; + userpg = buffer->user_page; /* * Disable preemption so as to not let the corresponding user-space @@ -2376,15 +2376,15 @@ unlock: */ static struct page * -perf_mmap_to_page(struct perf_mmap_data *data, unsigned long pgoff) +perf_mmap_to_page(struct perf_buffer *buffer, unsigned long pgoff) { - if (pgoff > data->nr_pages) + if (pgoff > buffer->nr_pages) return NULL; if (pgoff == 0) - return virt_to_page(data->user_page); + return virt_to_page(buffer->user_page); - return virt_to_page(data->data_pages[pgoff - 1]); + return virt_to_page(buffer->data_pages[pgoff - 1]); } static void *perf_mmap_alloc_page(int cpu) @@ -2400,42 +2400,42 @@ static void *perf_mmap_alloc_page(int cpu) return page_address(page); } -static struct perf_mmap_data * -perf_mmap_data_alloc(struct perf_event *event, int nr_pages) +static struct perf_buffer * +perf_buffer_alloc(struct perf_event *event, int nr_pages) { - struct perf_mmap_data *data; + struct perf_buffer *buffer; unsigned long size; int i; - size = sizeof(struct perf_mmap_data); + size = sizeof(struct perf_buffer); size += nr_pages * sizeof(void *); - data = kzalloc(size, GFP_KERNEL); - if (!data) + buffer = kzalloc(size, GFP_KERNEL); + if (!buffer) goto fail; - data->user_page = perf_mmap_alloc_page(event->cpu); - if (!data->user_page) + buffer->user_page = perf_mmap_alloc_page(event->cpu); + if (!buffer->user_page) goto fail_user_page; for (i = 0; i < nr_pages; i++) { - data->data_pages[i] = perf_mmap_alloc_page(event->cpu); - if (!data->data_pages[i]) + buffer->data_pages[i] = perf_mmap_alloc_page(event->cpu); + if (!buffer->data_pages[i]) goto fail_data_pages; } - data->nr_pages = nr_pages; + buffer->nr_pages = nr_pages; - return data; + return buffer; fail_data_pages: for (i--; i >= 0; i--) - free_page((unsigned long)data->data_pages[i]); + free_page((unsigned long)buffer->data_pages[i]); - free_page((unsigned long)data->user_page); + free_page((unsigned long)buffer->user_page); fail_user_page: - kfree(data); + kfree(buffer); fail: return NULL; @@ -2449,17 +2449,17 @@ static void perf_mmap_free_page(unsigned long addr) __free_page(page); } -static void perf_mmap_data_free(struct perf_mmap_data *data) +static void perf_buffer_free(struct perf_buffer *buffer) { int i; - perf_mmap_free_page((unsigned long)data->user_page); - for (i = 0; i < data->nr_pages; i++) - perf_mmap_free_page((unsigned long)data->data_pages[i]); - kfree(data); + perf_mmap_free_page((unsigned long)buffer->user_page); + for (i = 0; i < buffer->nr_pages; i++) + perf_mmap_free_page((unsigned long)buffer->data_pages[i]); + kfree(buffer); } -static inline int page_order(struct perf_mmap_data *data) +static inline int page_order(struct perf_buffer *buffer) { return 0; } @@ -2472,18 +2472,18 @@ static inline int page_order(struct perf_mmap_data *data) * Required for architectures that have d-cache aliasing issues. */ -static inline int page_order(struct perf_mmap_data *data) +static inline int page_order(struct perf_buffer *buffer) { - return data->page_order; + return buffer->page_order; } static struct page * -perf_mmap_to_page(struct perf_mmap_data *data, unsigned long pgoff) +perf_mmap_to_page(struct perf_buffer *buffer, unsigned long pgoff) { - if (pgoff > (1UL << page_order(data))) + if (pgoff > (1UL << page_order(buffer))) return NULL; - return vmalloc_to_page((void *)data->user_page + pgoff * PAGE_SIZE); + return vmalloc_to_page((void *)buffer->user_page + pgoff * PAGE_SIZE); } static void perf_mmap_unmark_page(void *addr) @@ -2493,57 +2493,57 @@ static void perf_mmap_unmark_page(void *addr) page->mapping = NULL; } -static void perf_mmap_data_free_work(struct work_struct *work) +static void perf_buffer_free_work(struct work_struct *work) { - struct perf_mmap_data *data; + struct perf_buffer *buffer; void *base; int i, nr; - data = container_of(work, struct perf_mmap_data, work); - nr = 1 << page_order(data); + buffer = container_of(work, struct perf_buffer, work); + nr = 1 << page_order(buffer); - base = data->user_page; + base = buffer->user_page; for (i = 0; i < nr + 1; i++) perf_mmap_unmark_page(base + (i * PAGE_SIZE)); vfree(base); - kfree(data); + kfree(buffer); } -static void perf_mmap_data_free(struct perf_mmap_data *data) +static void perf_buffer_free(struct perf_buffer *buffer) { - schedule_work(&data->work); + schedule_work(&buffer->work); } -static struct perf_mmap_data * -perf_mmap_data_alloc(struct perf_event *event, int nr_pages) +static struct perf_buffer * +perf_buffer_alloc(struct perf_event *event, int nr_pages) { - struct perf_mmap_data *data; + struct perf_buffer *buffer; unsigned long size; void *all_buf; - size = sizeof(struct perf_mmap_data); + size = sizeof(struct perf_buffer); size += sizeof(void *); - data = kzalloc(size, GFP_KERNEL); - if (!data) + buffer = kzalloc(size, GFP_KERNEL); + if (!buffer) goto fail; - INIT_WORK(&data->work, perf_mmap_data_free_work); + INIT_WORK(&buffer->work, perf_buffer_free_work); all_buf = vmalloc_user((nr_pages + 1) * PAGE_SIZE); if (!all_buf) goto fail_all_buf; - data->user_page = all_buf; - data->data_pages[0] = all_buf + PAGE_SIZE; - data->page_order = ilog2(nr_pages); - data->nr_pages = 1; + buffer->user_page = all_buf; + buffer->data_pages[0] = all_buf + PAGE_SIZE; + buffer->page_order = ilog2(nr_pages); + buffer->nr_pages = 1; - return data; + return buffer; fail_all_buf: - kfree(data); + kfree(buffer); fail: return NULL; @@ -2551,15 +2551,15 @@ fail: #endif -static unsigned long perf_data_size(struct perf_mmap_data *data) +static unsigned long perf_data_size(struct perf_buffer *buffer) { - return data->nr_pages << (PAGE_SHIFT + page_order(data)); + return buffer->nr_pages << (PAGE_SHIFT + page_order(buffer)); } static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) { struct perf_event *event = vma->vm_file->private_data; - struct perf_mmap_data *data; + struct perf_buffer *buffer; int ret = VM_FAULT_SIGBUS; if (vmf->flags & FAULT_FLAG_MKWRITE) { @@ -2569,14 +2569,14 @@ static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) } rcu_read_lock(); - data = rcu_dereference(event->data); - if (!data) + buffer = rcu_dereference(event->buffer); + if (!buffer) goto unlock; if (vmf->pgoff && (vmf->flags & FAULT_FLAG_WRITE)) goto unlock; - vmf->page = perf_mmap_to_page(data, vmf->pgoff); + vmf->page = perf_mmap_to_page(buffer, vmf->pgoff); if (!vmf->page) goto unlock; @@ -2592,51 +2592,51 @@ unlock: } static void -perf_mmap_data_init(struct perf_event *event, struct perf_mmap_data *data) +perf_buffer_init(struct perf_event *event, struct perf_buffer *buffer) { - long max_size = perf_data_size(data); + long max_size = perf_data_size(buffer); if (event->attr.watermark) { - data->watermark = min_t(long, max_size, + buffer->watermark = min_t(long, max_size, event->attr.wakeup_watermark); } - if (!data->watermark) - data->watermark = max_size / 2; + if (!buffer->watermark) + buffer->watermark = max_size / 2; - atomic_set(&data->refcount, 1); - rcu_assign_pointer(event->data, data); + atomic_set(&buffer->refcount, 1); + rcu_assign_pointer(event->buffer, buffer); } -static void perf_mmap_data_free_rcu(struct rcu_head *rcu_head) +static void perf_buffer_free_rcu(struct rcu_head *rcu_head) { - struct perf_mmap_data *data; + struct perf_buffer *buffer; - data = container_of(rcu_head, struct perf_mmap_data, rcu_head); - perf_mmap_data_free(data); + buffer = container_of(rcu_head, struct perf_buffer, rcu_head); + perf_buffer_free(buffer); } -static struct perf_mmap_data *perf_mmap_data_get(struct perf_event *event) +static struct perf_buffer *perf_buffer_get(struct perf_event *event) { - struct perf_mmap_data *data; + struct perf_buffer *buffer; rcu_read_lock(); - data = rcu_dereference(event->data); - if (data) { - if (!atomic_inc_not_zero(&data->refcount)) - data = NULL; + buffer = rcu_dereference(event->buffer); + if (buffer) { + if (!atomic_inc_not_zero(&buffer->refcount)) + buffer = NULL; } rcu_read_unlock(); - return data; + return buffer; } -static void perf_mmap_data_put(struct perf_mmap_data *data) +static void perf_buffer_put(struct perf_buffer *buffer) { - if (!atomic_dec_and_test(&data->refcount)) + if (!atomic_dec_and_test(&buffer->refcount)) return; - call_rcu(&data->rcu_head, perf_mmap_data_free_rcu); + call_rcu(&buffer->rcu_head, perf_buffer_free_rcu); } static void perf_mmap_open(struct vm_area_struct *vma) @@ -2651,16 +2651,16 @@ static void perf_mmap_close(struct vm_area_struct *vma) struct perf_event *event = vma->vm_file->private_data; if (atomic_dec_and_mutex_lock(&event->mmap_count, &event->mmap_mutex)) { - unsigned long size = perf_data_size(event->data); + unsigned long size = perf_data_size(event->buffer); struct user_struct *user = event->mmap_user; - struct perf_mmap_data *data = event->data; + struct perf_buffer *buffer = event->buffer; atomic_long_sub((size >> PAGE_SHIFT) + 1, &user->locked_vm); vma->vm_mm->locked_vm -= event->mmap_locked; - rcu_assign_pointer(event->data, NULL); + rcu_assign_pointer(event->buffer, NULL); mutex_unlock(&event->mmap_mutex); - perf_mmap_data_put(data); + perf_buffer_put(buffer); free_uid(user); } } @@ -2678,7 +2678,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) unsigned long user_locked, user_lock_limit; struct user_struct *user = current_user(); unsigned long locked, lock_limit; - struct perf_mmap_data *data; + struct perf_buffer *buffer; unsigned long vma_size; unsigned long nr_pages; long user_extra, extra; @@ -2699,7 +2699,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) nr_pages = (vma_size / PAGE_SIZE) - 1; /* - * If we have data pages ensure they're a power-of-two number, so we + * If we have buffer pages ensure they're a power-of-two number, so we * can do bitmasks instead of modulo. */ if (nr_pages != 0 && !is_power_of_2(nr_pages)) @@ -2713,9 +2713,9 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) WARN_ON_ONCE(event->ctx->parent_ctx); mutex_lock(&event->mmap_mutex); - if (event->data) { - if (event->data->nr_pages == nr_pages) - atomic_inc(&event->data->refcount); + if (event->buffer) { + if (event->buffer->nr_pages == nr_pages) + atomic_inc(&event->buffer->refcount); else ret = -EINVAL; goto unlock; @@ -2745,17 +2745,17 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; } - WARN_ON(event->data); + WARN_ON(event->buffer); - data = perf_mmap_data_alloc(event, nr_pages); - if (!data) { + buffer = perf_buffer_alloc(event, nr_pages); + if (!buffer) { ret = -ENOMEM; goto unlock; } - perf_mmap_data_init(event, data); + perf_buffer_init(event, buffer); if (vma->vm_flags & VM_WRITE) - event->data->writable = 1; + event->buffer->writable = 1; atomic_long_add(user_extra, &user->locked_vm); event->mmap_locked = extra; @@ -2964,15 +2964,15 @@ EXPORT_SYMBOL_GPL(perf_unregister_guest_info_callbacks); /* * Output */ -static bool perf_output_space(struct perf_mmap_data *data, unsigned long tail, +static bool perf_output_space(struct perf_buffer *buffer, unsigned long tail, unsigned long offset, unsigned long head) { unsigned long mask; - if (!data->writable) + if (!buffer->writable) return true; - mask = perf_data_size(data) - 1; + mask = perf_data_size(buffer) - 1; offset = (offset - tail) & mask; head = (head - tail) & mask; @@ -2985,7 +2985,7 @@ static bool perf_output_space(struct perf_mmap_data *data, unsigned long tail, static void perf_output_wakeup(struct perf_output_handle *handle) { - atomic_set(&handle->data->poll, POLL_IN); + atomic_set(&handle->buffer->poll, POLL_IN); if (handle->nmi) { handle->event->pending_wakeup = 1; @@ -3005,45 +3005,45 @@ static void perf_output_wakeup(struct perf_output_handle *handle) */ static void perf_output_get_handle(struct perf_output_handle *handle) { - struct perf_mmap_data *data = handle->data; + struct perf_buffer *buffer = handle->buffer; preempt_disable(); - local_inc(&data->nest); - handle->wakeup = local_read(&data->wakeup); + local_inc(&buffer->nest); + handle->wakeup = local_read(&buffer->wakeup); } static void perf_output_put_handle(struct perf_output_handle *handle) { - struct perf_mmap_data *data = handle->data; + struct perf_buffer *buffer = handle->buffer; unsigned long head; again: - head = local_read(&data->head); + head = local_read(&buffer->head); /* * IRQ/NMI can happen here, which means we can miss a head update. */ - if (!local_dec_and_test(&data->nest)) + if (!local_dec_and_test(&buffer->nest)) goto out; /* * Publish the known good head. Rely on the full barrier implied - * by atomic_dec_and_test() order the data->head read and this + * by atomic_dec_and_test() order the buffer->head read and this * write. */ - data->user_page->data_head = head; + buffer->user_page->data_head = head; /* * Now check if we missed an update, rely on the (compiler) - * barrier in atomic_dec_and_test() to re-read data->head. + * barrier in atomic_dec_and_test() to re-read buffer->head. */ - if (unlikely(head != local_read(&data->head))) { - local_inc(&data->nest); + if (unlikely(head != local_read(&buffer->head))) { + local_inc(&buffer->nest); goto again; } - if (handle->wakeup != local_read(&data->wakeup)) + if (handle->wakeup != local_read(&buffer->wakeup)) perf_output_wakeup(handle); out: @@ -3063,12 +3063,12 @@ __always_inline void perf_output_copy(struct perf_output_handle *handle, buf += size; handle->size -= size; if (!handle->size) { - struct perf_mmap_data *data = handle->data; + struct perf_buffer *buffer = handle->buffer; handle->page++; - handle->page &= data->nr_pages - 1; - handle->addr = data->data_pages[handle->page]; - handle->size = PAGE_SIZE << page_order(data); + handle->page &= buffer->nr_pages - 1; + handle->addr = buffer->data_pages[handle->page]; + handle->size = PAGE_SIZE << page_order(buffer); } } while (len); } @@ -3077,7 +3077,7 @@ int perf_output_begin(struct perf_output_handle *handle, struct perf_event *event, unsigned int size, int nmi, int sample) { - struct perf_mmap_data *data; + struct perf_buffer *buffer; unsigned long tail, offset, head; int have_lost; struct { @@ -3093,19 +3093,19 @@ int perf_output_begin(struct perf_output_handle *handle, if (event->parent) event = event->parent; - data = rcu_dereference(event->data); - if (!data) + buffer = rcu_dereference(event->buffer); + if (!buffer) goto out; - handle->data = data; + handle->buffer = buffer; handle->event = event; handle->nmi = nmi; handle->sample = sample; - if (!data->nr_pages) + if (!buffer->nr_pages) goto out; - have_lost = local_read(&data->lost); + have_lost = local_read(&buffer->lost); if (have_lost) size += sizeof(lost_event); @@ -3117,30 +3117,30 @@ int perf_output_begin(struct perf_output_handle *handle, * tail pointer. So that all reads will be completed before the * write is issued. */ - tail = ACCESS_ONCE(data->user_page->data_tail); + tail = ACCESS_ONCE(buffer->user_page->data_tail); smp_rmb(); - offset = head = local_read(&data->head); + offset = head = local_read(&buffer->head); head += size; - if (unlikely(!perf_output_space(data, tail, offset, head))) + if (unlikely(!perf_output_space(buffer, tail, offset, head))) goto fail; - } while (local_cmpxchg(&data->head, offset, head) != offset); + } while (local_cmpxchg(&buffer->head, offset, head) != offset); - if (head - local_read(&data->wakeup) > data->watermark) - local_add(data->watermark, &data->wakeup); + if (head - local_read(&buffer->wakeup) > buffer->watermark) + local_add(buffer->watermark, &buffer->wakeup); - handle->page = offset >> (PAGE_SHIFT + page_order(data)); - handle->page &= data->nr_pages - 1; - handle->size = offset & ((PAGE_SIZE << page_order(data)) - 1); - handle->addr = data->data_pages[handle->page]; + handle->page = offset >> (PAGE_SHIFT + page_order(buffer)); + handle->page &= buffer->nr_pages - 1; + handle->size = offset & ((PAGE_SIZE << page_order(buffer)) - 1); + handle->addr = buffer->data_pages[handle->page]; handle->addr += handle->size; - handle->size = (PAGE_SIZE << page_order(data)) - handle->size; + handle->size = (PAGE_SIZE << page_order(buffer)) - handle->size; if (have_lost) { lost_event.header.type = PERF_RECORD_LOST; lost_event.header.misc = 0; lost_event.header.size = sizeof(lost_event); lost_event.id = event->id; - lost_event.lost = local_xchg(&data->lost, 0); + lost_event.lost = local_xchg(&buffer->lost, 0); perf_output_put(handle, lost_event); } @@ -3148,7 +3148,7 @@ int perf_output_begin(struct perf_output_handle *handle, return 0; fail: - local_inc(&data->lost); + local_inc(&buffer->lost); perf_output_put_handle(handle); out: rcu_read_unlock(); @@ -3159,15 +3159,15 @@ out: void perf_output_end(struct perf_output_handle *handle) { struct perf_event *event = handle->event; - struct perf_mmap_data *data = handle->data; + struct perf_buffer *buffer = handle->buffer; int wakeup_events = event->attr.wakeup_events; if (handle->sample && wakeup_events) { - int events = local_inc_return(&data->events); + int events = local_inc_return(&buffer->events); if (events >= wakeup_events) { - local_sub(wakeup_events, &data->events); - local_inc(&data->wakeup); + local_sub(wakeup_events, &buffer->events); + local_inc(&buffer->wakeup); } } @@ -5010,7 +5010,7 @@ err_size: static int perf_event_set_output(struct perf_event *event, struct perf_event *output_event) { - struct perf_mmap_data *data = NULL, *old_data = NULL; + struct perf_buffer *buffer = NULL, *old_buffer = NULL; int ret = -EINVAL; if (!output_event) @@ -5040,19 +5040,19 @@ set: if (output_event) { /* get the buffer we want to redirect to */ - data = perf_mmap_data_get(output_event); - if (!data) + buffer = perf_buffer_get(output_event); + if (!buffer) goto unlock; } - old_data = event->data; - rcu_assign_pointer(event->data, data); + old_buffer = event->buffer; + rcu_assign_pointer(event->buffer, buffer); ret = 0; unlock: mutex_unlock(&event->mmap_mutex); - if (old_data) - perf_mmap_data_put(old_data); + if (old_buffer) + perf_buffer_put(old_buffer); out: return ret; } -- cgit v1.2.3 From d57e34fdd60be7ffd0b1d86bfa1a553df86b7172 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 28 May 2010 19:41:35 +0200 Subject: perf: Simplify the ring-buffer logic: make perf_buffer_alloc() do everything needed Currently there are perf_buffer_alloc() + perf_buffer_init() + some separate bits, fold it all into a single perf_buffer_alloc() and only leave the attachment to the event separate. Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 2 ++ kernel/perf_event.c | 61 ++++++++++++++++++++++++++-------------------- 2 files changed, 36 insertions(+), 27 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 2a0da021c23f..441992a9775c 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -602,6 +602,8 @@ enum perf_event_active_state { struct file; +#define PERF_BUFFER_WRITABLE 0x01 + struct perf_buffer { atomic_t refcount; struct rcu_head rcu_head; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 93d545801e43..f75c9c9c8177 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -2369,6 +2369,25 @@ unlock: rcu_read_unlock(); } +static unsigned long perf_data_size(struct perf_buffer *buffer); + +static void +perf_buffer_init(struct perf_buffer *buffer, long watermark, int flags) +{ + long max_size = perf_data_size(buffer); + + if (watermark) + buffer->watermark = min(max_size, watermark); + + if (!buffer->watermark) + buffer->watermark = max_size / 2; + + if (flags & PERF_BUFFER_WRITABLE) + buffer->writable = 1; + + atomic_set(&buffer->refcount, 1); +} + #ifndef CONFIG_PERF_USE_VMALLOC /* @@ -2401,7 +2420,7 @@ static void *perf_mmap_alloc_page(int cpu) } static struct perf_buffer * -perf_buffer_alloc(struct perf_event *event, int nr_pages) +perf_buffer_alloc(int nr_pages, long watermark, int cpu, int flags) { struct perf_buffer *buffer; unsigned long size; @@ -2414,18 +2433,20 @@ perf_buffer_alloc(struct perf_event *event, int nr_pages) if (!buffer) goto fail; - buffer->user_page = perf_mmap_alloc_page(event->cpu); + buffer->user_page = perf_mmap_alloc_page(cpu); if (!buffer->user_page) goto fail_user_page; for (i = 0; i < nr_pages; i++) { - buffer->data_pages[i] = perf_mmap_alloc_page(event->cpu); + buffer->data_pages[i] = perf_mmap_alloc_page(cpu); if (!buffer->data_pages[i]) goto fail_data_pages; } buffer->nr_pages = nr_pages; + perf_buffer_init(buffer, watermark, flags); + return buffer; fail_data_pages: @@ -2516,7 +2537,7 @@ static void perf_buffer_free(struct perf_buffer *buffer) } static struct perf_buffer * -perf_buffer_alloc(struct perf_event *event, int nr_pages) +perf_buffer_alloc(int nr_pages, long watermark, int cpu, int flags) { struct perf_buffer *buffer; unsigned long size; @@ -2540,6 +2561,8 @@ perf_buffer_alloc(struct perf_event *event, int nr_pages) buffer->page_order = ilog2(nr_pages); buffer->nr_pages = 1; + perf_buffer_init(buffer, watermark, flags); + return buffer; fail_all_buf: @@ -2591,23 +2614,6 @@ unlock: return ret; } -static void -perf_buffer_init(struct perf_event *event, struct perf_buffer *buffer) -{ - long max_size = perf_data_size(buffer); - - if (event->attr.watermark) { - buffer->watermark = min_t(long, max_size, - event->attr.wakeup_watermark); - } - - if (!buffer->watermark) - buffer->watermark = max_size / 2; - - atomic_set(&buffer->refcount, 1); - rcu_assign_pointer(event->buffer, buffer); -} - static void perf_buffer_free_rcu(struct rcu_head *rcu_head) { struct perf_buffer *buffer; @@ -2682,7 +2688,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) unsigned long vma_size; unsigned long nr_pages; long user_extra, extra; - int ret = 0; + int ret = 0, flags = 0; /* * Don't allow mmap() of inherited per-task counters. This would @@ -2747,15 +2753,16 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) WARN_ON(event->buffer); - buffer = perf_buffer_alloc(event, nr_pages); + if (vma->vm_flags & VM_WRITE) + flags |= PERF_BUFFER_WRITABLE; + + buffer = perf_buffer_alloc(nr_pages, event->attr.wakeup_watermark, + event->cpu, flags); if (!buffer) { ret = -ENOMEM; goto unlock; } - - perf_buffer_init(event, buffer); - if (vma->vm_flags & VM_WRITE) - event->buffer->writable = 1; + rcu_assign_pointer(event->buffer, buffer); atomic_long_add(user_extra, &user->locked_vm); event->mmap_locked = extra; -- cgit v1.2.3 From a6e6dea68c18f705957573ee5596097c7e82d0e5 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 21 May 2010 14:27:58 +0200 Subject: perf: Add perf_event::child_count Only child counters adding back their values into the parent counter are responsible for cross-cpu updates to event->count. So if we pull that out into a new child_count variable, we get an event->count that is only modified locally. Signed-off-by: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker Cc: Paul Mackerras Cc: Mike Galbraith Cc: Steven Rostedt LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 1 + kernel/perf_event.c | 4 ++-- 2 files changed, 3 insertions(+), 2 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 441992a9775c..f34dab9b275e 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -671,6 +671,7 @@ struct perf_event { enum perf_event_active_state state; unsigned int attach_state; atomic64_t count; + atomic64_t child_count; /* * These are the total time in nanoseconds that the event diff --git a/kernel/perf_event.c b/kernel/perf_event.c index ab4c0ffc271c..a395fda2d94c 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1738,7 +1738,7 @@ static void __perf_event_read(void *info) static inline u64 perf_event_count(struct perf_event *event) { - return atomic64_read(&event->count); + return atomic64_read(&event->count) + atomic64_read(&event->child_count); } static u64 perf_event_read(struct perf_event *event) @@ -5379,7 +5379,7 @@ static void sync_child_event(struct perf_event *child_event, /* * Add back the child's count to the parent's count: */ - atomic64_add(child_val, &parent_event->count); + atomic64_add(child_val, &parent_event->child_count); atomic64_add(child_event->total_time_enabled, &parent_event->child_total_time_enabled); atomic64_add(child_event->total_time_running, -- cgit v1.2.3 From e78505958cf123048fb48cb56b79cebb8edd15fb Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 21 May 2010 14:43:08 +0200 Subject: perf: Convert perf_event to local_t Since now all modification to event->count (and ->prev_count and ->period_left) are local to a cpu, change then to local64_t so we avoid the LOCK'ed ops. Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- arch/arm/kernel/perf_event.c | 18 ++++++++--------- arch/powerpc/kernel/perf_event.c | 34 ++++++++++++++++---------------- arch/sh/kernel/perf_event.c | 6 +++--- arch/sparc/kernel/perf_event.c | 18 ++++++++--------- arch/x86/kernel/cpu/perf_event.c | 18 ++++++++--------- include/linux/perf_event.h | 7 ++++--- kernel/perf_event.c | 42 ++++++++++++++++++++-------------------- 7 files changed, 72 insertions(+), 71 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c index c45768614c8a..5b7cfafc0720 100644 --- a/arch/arm/kernel/perf_event.c +++ b/arch/arm/kernel/perf_event.c @@ -164,20 +164,20 @@ armpmu_event_set_period(struct perf_event *event, struct hw_perf_event *hwc, int idx) { - s64 left = atomic64_read(&hwc->period_left); + s64 left = local64_read(&hwc->period_left); s64 period = hwc->sample_period; int ret = 0; if (unlikely(left <= -period)) { left = period; - atomic64_set(&hwc->period_left, left); + local64_set(&hwc->period_left, left); hwc->last_period = period; ret = 1; } if (unlikely(left <= 0)) { left += period; - atomic64_set(&hwc->period_left, left); + local64_set(&hwc->period_left, left); hwc->last_period = period; ret = 1; } @@ -185,7 +185,7 @@ armpmu_event_set_period(struct perf_event *event, if (left > (s64)armpmu->max_period) left = armpmu->max_period; - atomic64_set(&hwc->prev_count, (u64)-left); + local64_set(&hwc->prev_count, (u64)-left); armpmu->write_counter(idx, (u64)(-left) & 0xffffffff); @@ -204,18 +204,18 @@ armpmu_event_update(struct perf_event *event, s64 delta; again: - prev_raw_count = atomic64_read(&hwc->prev_count); + prev_raw_count = local64_read(&hwc->prev_count); new_raw_count = armpmu->read_counter(idx); - if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + if (local64_cmpxchg(&hwc->prev_count, prev_raw_count, new_raw_count) != prev_raw_count) goto again; delta = (new_raw_count << shift) - (prev_raw_count << shift); delta >>= shift; - atomic64_add(delta, &event->count); - atomic64_sub(delta, &hwc->period_left); + local64_add(delta, &event->count); + local64_sub(delta, &hwc->period_left); return new_raw_count; } @@ -478,7 +478,7 @@ __hw_perf_event_init(struct perf_event *event) if (!hwc->sample_period) { hwc->sample_period = armpmu->max_period; hwc->last_period = hwc->sample_period; - atomic64_set(&hwc->period_left, hwc->sample_period); + local64_set(&hwc->period_left, hwc->sample_period); } err = 0; diff --git a/arch/powerpc/kernel/perf_event.c b/arch/powerpc/kernel/perf_event.c index ac2a8c2554d9..af1d9a7c65d1 100644 --- a/arch/powerpc/kernel/perf_event.c +++ b/arch/powerpc/kernel/perf_event.c @@ -410,15 +410,15 @@ static void power_pmu_read(struct perf_event *event) * Therefore we treat them like NMIs. */ do { - prev = atomic64_read(&event->hw.prev_count); + prev = local64_read(&event->hw.prev_count); barrier(); val = read_pmc(event->hw.idx); - } while (atomic64_cmpxchg(&event->hw.prev_count, prev, val) != prev); + } while (local64_cmpxchg(&event->hw.prev_count, prev, val) != prev); /* The counters are only 32 bits wide */ delta = (val - prev) & 0xfffffffful; - atomic64_add(delta, &event->count); - atomic64_sub(delta, &event->hw.period_left); + local64_add(delta, &event->count); + local64_sub(delta, &event->hw.period_left); } /* @@ -444,10 +444,10 @@ static void freeze_limited_counters(struct cpu_hw_events *cpuhw, if (!event->hw.idx) continue; val = (event->hw.idx == 5) ? pmc5 : pmc6; - prev = atomic64_read(&event->hw.prev_count); + prev = local64_read(&event->hw.prev_count); event->hw.idx = 0; delta = (val - prev) & 0xfffffffful; - atomic64_add(delta, &event->count); + local64_add(delta, &event->count); } } @@ -462,7 +462,7 @@ static void thaw_limited_counters(struct cpu_hw_events *cpuhw, event = cpuhw->limited_counter[i]; event->hw.idx = cpuhw->limited_hwidx[i]; val = (event->hw.idx == 5) ? pmc5 : pmc6; - atomic64_set(&event->hw.prev_count, val); + local64_set(&event->hw.prev_count, val); perf_event_update_userpage(event); } } @@ -666,11 +666,11 @@ void hw_perf_enable(void) } val = 0; if (event->hw.sample_period) { - left = atomic64_read(&event->hw.period_left); + left = local64_read(&event->hw.period_left); if (left < 0x80000000L) val = 0x80000000L - left; } - atomic64_set(&event->hw.prev_count, val); + local64_set(&event->hw.prev_count, val); event->hw.idx = idx; write_pmc(idx, val); perf_event_update_userpage(event); @@ -842,8 +842,8 @@ static void power_pmu_unthrottle(struct perf_event *event) if (left < 0x80000000L) val = 0x80000000L - left; write_pmc(event->hw.idx, val); - atomic64_set(&event->hw.prev_count, val); - atomic64_set(&event->hw.period_left, left); + local64_set(&event->hw.prev_count, val); + local64_set(&event->hw.period_left, left); perf_event_update_userpage(event); perf_enable(); local_irq_restore(flags); @@ -1109,7 +1109,7 @@ const struct pmu *hw_perf_event_init(struct perf_event *event) event->hw.config = events[n]; event->hw.event_base = cflags[n]; event->hw.last_period = event->hw.sample_period; - atomic64_set(&event->hw.period_left, event->hw.last_period); + local64_set(&event->hw.period_left, event->hw.last_period); /* * See if we need to reserve the PMU. @@ -1147,16 +1147,16 @@ static void record_and_restart(struct perf_event *event, unsigned long val, int record = 0; /* we don't have to worry about interrupts here */ - prev = atomic64_read(&event->hw.prev_count); + prev = local64_read(&event->hw.prev_count); delta = (val - prev) & 0xfffffffful; - atomic64_add(delta, &event->count); + local64_add(delta, &event->count); /* * See if the total period for this event has expired, * and update for the next period. */ val = 0; - left = atomic64_read(&event->hw.period_left) - delta; + left = local64_read(&event->hw.period_left) - delta; if (period) { if (left <= 0) { left += period; @@ -1194,8 +1194,8 @@ static void record_and_restart(struct perf_event *event, unsigned long val, } write_pmc(event->hw.idx, val); - atomic64_set(&event->hw.prev_count, val); - atomic64_set(&event->hw.period_left, left); + local64_set(&event->hw.prev_count, val); + local64_set(&event->hw.period_left, left); perf_event_update_userpage(event); } diff --git a/arch/sh/kernel/perf_event.c b/arch/sh/kernel/perf_event.c index 81b6de41ae5d..7a3dc3567258 100644 --- a/arch/sh/kernel/perf_event.c +++ b/arch/sh/kernel/perf_event.c @@ -185,10 +185,10 @@ static void sh_perf_event_update(struct perf_event *event, * this is the simplest approach for maintaining consistency. */ again: - prev_raw_count = atomic64_read(&hwc->prev_count); + prev_raw_count = local64_read(&hwc->prev_count); new_raw_count = sh_pmu->read(idx); - if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + if (local64_cmpxchg(&hwc->prev_count, prev_raw_count, new_raw_count) != prev_raw_count) goto again; @@ -203,7 +203,7 @@ again: delta = (new_raw_count << shift) - (prev_raw_count << shift); delta >>= shift; - atomic64_add(delta, &event->count); + local64_add(delta, &event->count); } static void sh_pmu_disable(struct perf_event *event) diff --git a/arch/sparc/kernel/perf_event.c b/arch/sparc/kernel/perf_event.c index beeb92fa3acd..8a6660da8e08 100644 --- a/arch/sparc/kernel/perf_event.c +++ b/arch/sparc/kernel/perf_event.c @@ -572,18 +572,18 @@ static u64 sparc_perf_event_update(struct perf_event *event, s64 delta; again: - prev_raw_count = atomic64_read(&hwc->prev_count); + prev_raw_count = local64_read(&hwc->prev_count); new_raw_count = read_pmc(idx); - if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + if (local64_cmpxchg(&hwc->prev_count, prev_raw_count, new_raw_count) != prev_raw_count) goto again; delta = (new_raw_count << shift) - (prev_raw_count << shift); delta >>= shift; - atomic64_add(delta, &event->count); - atomic64_sub(delta, &hwc->period_left); + local64_add(delta, &event->count); + local64_sub(delta, &hwc->period_left); return new_raw_count; } @@ -591,27 +591,27 @@ again: static int sparc_perf_event_set_period(struct perf_event *event, struct hw_perf_event *hwc, int idx) { - s64 left = atomic64_read(&hwc->period_left); + s64 left = local64_read(&hwc->period_left); s64 period = hwc->sample_period; int ret = 0; if (unlikely(left <= -period)) { left = period; - atomic64_set(&hwc->period_left, left); + local64_set(&hwc->period_left, left); hwc->last_period = period; ret = 1; } if (unlikely(left <= 0)) { left += period; - atomic64_set(&hwc->period_left, left); + local64_set(&hwc->period_left, left); hwc->last_period = period; ret = 1; } if (left > MAX_PERIOD) left = MAX_PERIOD; - atomic64_set(&hwc->prev_count, (u64)-left); + local64_set(&hwc->prev_count, (u64)-left); write_pmc(idx, (u64)(-left) & 0xffffffff); @@ -1087,7 +1087,7 @@ static int __hw_perf_event_init(struct perf_event *event) if (!hwc->sample_period) { hwc->sample_period = MAX_PERIOD; hwc->last_period = hwc->sample_period; - atomic64_set(&hwc->period_left, hwc->sample_period); + local64_set(&hwc->period_left, hwc->sample_period); } return 0; diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 79e199843db6..2d0d29069275 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -296,10 +296,10 @@ x86_perf_event_update(struct perf_event *event) * count to the generic event atomically: */ again: - prev_raw_count = atomic64_read(&hwc->prev_count); + prev_raw_count = local64_read(&hwc->prev_count); rdmsrl(hwc->event_base + idx, new_raw_count); - if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + if (local64_cmpxchg(&hwc->prev_count, prev_raw_count, new_raw_count) != prev_raw_count) goto again; @@ -314,8 +314,8 @@ again: delta = (new_raw_count << shift) - (prev_raw_count << shift); delta >>= shift; - atomic64_add(delta, &event->count); - atomic64_sub(delta, &hwc->period_left); + local64_add(delta, &event->count); + local64_sub(delta, &hwc->period_left); return new_raw_count; } @@ -439,7 +439,7 @@ static int x86_setup_perfctr(struct perf_event *event) if (!hwc->sample_period) { hwc->sample_period = x86_pmu.max_period; hwc->last_period = hwc->sample_period; - atomic64_set(&hwc->period_left, hwc->sample_period); + local64_set(&hwc->period_left, hwc->sample_period); } else { /* * If we have a PMU initialized but no APIC @@ -886,7 +886,7 @@ static int x86_perf_event_set_period(struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; - s64 left = atomic64_read(&hwc->period_left); + s64 left = local64_read(&hwc->period_left); s64 period = hwc->sample_period; int ret = 0, idx = hwc->idx; @@ -898,14 +898,14 @@ x86_perf_event_set_period(struct perf_event *event) */ if (unlikely(left <= -period)) { left = period; - atomic64_set(&hwc->period_left, left); + local64_set(&hwc->period_left, left); hwc->last_period = period; ret = 1; } if (unlikely(left <= 0)) { left += period; - atomic64_set(&hwc->period_left, left); + local64_set(&hwc->period_left, left); hwc->last_period = period; ret = 1; } @@ -924,7 +924,7 @@ x86_perf_event_set_period(struct perf_event *event) * The hw event starts counting from this event offset, * mark it to be able to extra future deltas: */ - atomic64_set(&hwc->prev_count, (u64)-left); + local64_set(&hwc->prev_count, (u64)-left); wrmsrl(hwc->event_base + idx, (u64)(-left) & x86_pmu.cntval_mask); diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index f34dab9b275e..7342979f95f2 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -487,6 +487,7 @@ struct perf_guest_info_callbacks { #include #include #include +#include #define PERF_MAX_STACK_DEPTH 255 @@ -536,10 +537,10 @@ struct hw_perf_event { struct arch_hw_breakpoint info; #endif }; - atomic64_t prev_count; + local64_t prev_count; u64 sample_period; u64 last_period; - atomic64_t period_left; + local64_t period_left; u64 interrupts; u64 freq_time_stamp; @@ -670,7 +671,7 @@ struct perf_event { enum perf_event_active_state state; unsigned int attach_state; - atomic64_t count; + local64_t count; atomic64_t child_count; /* diff --git a/kernel/perf_event.c b/kernel/perf_event.c index a395fda2d94c..97c73018592e 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1148,9 +1148,9 @@ static void __perf_event_sync_stat(struct perf_event *event, * In order to keep per-task stats reliable we need to flip the event * values when we flip the contexts. */ - value = atomic64_read(&next_event->count); - value = atomic64_xchg(&event->count, value); - atomic64_set(&next_event->count, value); + value = local64_read(&next_event->count); + value = local64_xchg(&event->count, value); + local64_set(&next_event->count, value); swap(event->total_time_enabled, next_event->total_time_enabled); swap(event->total_time_running, next_event->total_time_running); @@ -1540,10 +1540,10 @@ static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count) hwc->sample_period = sample_period; - if (atomic64_read(&hwc->period_left) > 8*sample_period) { + if (local64_read(&hwc->period_left) > 8*sample_period) { perf_disable(); perf_event_stop(event); - atomic64_set(&hwc->period_left, 0); + local64_set(&hwc->period_left, 0); perf_event_start(event); perf_enable(); } @@ -1584,7 +1584,7 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx) perf_disable(); event->pmu->read(event); - now = atomic64_read(&event->count); + now = local64_read(&event->count); delta = now - hwc->freq_count_stamp; hwc->freq_count_stamp = now; @@ -1738,7 +1738,7 @@ static void __perf_event_read(void *info) static inline u64 perf_event_count(struct perf_event *event) { - return atomic64_read(&event->count) + atomic64_read(&event->child_count); + return local64_read(&event->count) + atomic64_read(&event->child_count); } static u64 perf_event_read(struct perf_event *event) @@ -2141,7 +2141,7 @@ static unsigned int perf_poll(struct file *file, poll_table *wait) static void perf_event_reset(struct perf_event *event) { (void)perf_event_read(event); - atomic64_set(&event->count, 0); + local64_set(&event->count, 0); perf_event_update_userpage(event); } @@ -2359,7 +2359,7 @@ void perf_event_update_userpage(struct perf_event *event) userpg->index = perf_event_index(event); userpg->offset = perf_event_count(event); if (event->state == PERF_EVENT_STATE_ACTIVE) - userpg->offset -= atomic64_read(&event->hw.prev_count); + userpg->offset -= local64_read(&event->hw.prev_count); userpg->time_enabled = event->total_time_enabled + atomic64_read(&event->child_total_time_enabled); @@ -4035,14 +4035,14 @@ static u64 perf_swevent_set_period(struct perf_event *event) hwc->last_period = hwc->sample_period; again: - old = val = atomic64_read(&hwc->period_left); + old = val = local64_read(&hwc->period_left); if (val < 0) return 0; nr = div64_u64(period + val, period); offset = nr * period; val -= offset; - if (atomic64_cmpxchg(&hwc->period_left, old, val) != old) + if (local64_cmpxchg(&hwc->period_left, old, val) != old) goto again; return nr; @@ -4081,7 +4081,7 @@ static void perf_swevent_add(struct perf_event *event, u64 nr, { struct hw_perf_event *hwc = &event->hw; - atomic64_add(nr, &event->count); + local64_add(nr, &event->count); if (!regs) return; @@ -4092,7 +4092,7 @@ static void perf_swevent_add(struct perf_event *event, u64 nr, if (nr == 1 && hwc->sample_period == 1 && !event->attr.freq) return perf_swevent_overflow(event, 1, nmi, data, regs); - if (atomic64_add_negative(nr, &hwc->period_left)) + if (local64_add_negative(nr, &hwc->period_left)) return; perf_swevent_overflow(event, 0, nmi, data, regs); @@ -4383,8 +4383,8 @@ static void cpu_clock_perf_event_update(struct perf_event *event) u64 now; now = cpu_clock(cpu); - prev = atomic64_xchg(&event->hw.prev_count, now); - atomic64_add(now - prev, &event->count); + prev = local64_xchg(&event->hw.prev_count, now); + local64_add(now - prev, &event->count); } static int cpu_clock_perf_event_enable(struct perf_event *event) @@ -4392,7 +4392,7 @@ static int cpu_clock_perf_event_enable(struct perf_event *event) struct hw_perf_event *hwc = &event->hw; int cpu = raw_smp_processor_id(); - atomic64_set(&hwc->prev_count, cpu_clock(cpu)); + local64_set(&hwc->prev_count, cpu_clock(cpu)); perf_swevent_start_hrtimer(event); return 0; @@ -4424,9 +4424,9 @@ static void task_clock_perf_event_update(struct perf_event *event, u64 now) u64 prev; s64 delta; - prev = atomic64_xchg(&event->hw.prev_count, now); + prev = local64_xchg(&event->hw.prev_count, now); delta = now - prev; - atomic64_add(delta, &event->count); + local64_add(delta, &event->count); } static int task_clock_perf_event_enable(struct perf_event *event) @@ -4436,7 +4436,7 @@ static int task_clock_perf_event_enable(struct perf_event *event) now = event->ctx->time; - atomic64_set(&hwc->prev_count, now); + local64_set(&hwc->prev_count, now); perf_swevent_start_hrtimer(event); @@ -4879,7 +4879,7 @@ perf_event_alloc(struct perf_event_attr *attr, hwc->sample_period = 1; hwc->last_period = hwc->sample_period; - atomic64_set(&hwc->period_left, hwc->sample_period); + local64_set(&hwc->period_left, hwc->sample_period); /* * we currently do not support PERF_FORMAT_GROUP on inherited events @@ -5313,7 +5313,7 @@ inherit_event(struct perf_event *parent_event, hwc->sample_period = sample_period; hwc->last_period = sample_period; - atomic64_set(&hwc->period_left, sample_period); + local64_set(&hwc->period_left, sample_period); } child_event->overflow_handler = parent_event->overflow_handler; -- cgit v1.2.3 From 7be7923633a142402266d642ccebf74f556a649b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 9 Jun 2010 11:57:23 +0200 Subject: perf: Fix build breakage for architecutes without atomic64_t The local64.h include dependency was not dependent on PERF_EVENT=y, which meant that arch's without atomic64_t support ended up including it and failed to build. Signed-off-by: Peter Zijlstra LKML-Reference: --- include/linux/perf_event.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 7342979f95f2..1218d05728b9 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -462,6 +462,7 @@ enum perf_callchain_context { #ifdef CONFIG_PERF_EVENTS # include +# include #endif struct perf_guest_info_callbacks { @@ -487,7 +488,6 @@ struct perf_guest_info_callbacks { #include #include #include -#include #define PERF_MAX_STACK_DEPTH 255 -- cgit v1.2.3 From 45a73372efe4a63f44aa2e1125d4a777c2fdc8d8 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 23 Jun 2010 23:00:37 +0200 Subject: hw_breakpoints: Fix per task breakpoint tracking Freeing a perf event can happen in several ways. A task calls perf_event_exit_task() right before exiting. This helper will detach all the events from the task context and queue their removal through free_event() if they are child tasks. The task also loses its context reference there. Releasing the breakpoint slot from the constraint table is made from free_event() that calls release_bp_slot(). We count the number of breakpoints this task is running by looking at the task's perf_event_ctxp and iterating through its attached events. But at this time, the reference to this context has been cleaned up already. So looking at the event->ctx instead of task->perf_event_ctxp to count the remaining breakpoints should solve the problem. At least it would for child breakpoints, but not for parent ones. If the parent exits before the child, it will remove all its events from the context but free_event() will be called later, on fd release time. And checking the number of breakpoints the task has attached to its context at this time is unreliable as all events have been removed from the context. To solve this, we keep track of the list of per task breakpoints. On top of it, we maintain our array of numbers of breakpoints used by the tasks. We use the context address as a task id. So, instead of looking at the number of events attached to a context, we walk through our list of per task breakpoints and count the number of breakpoints that use the same ctx than the one to be reserved or released from the constraint table, and update the count on top of this result. In the meantime it solves a bad refcounting, it also solves a warning, reported by Paul. Badness at /home/paulus/kernel/perf/kernel/hw_breakpoint.c:114 NIP: c0000000000cb470 LR: c0000000000cb46c CTR: c00000000032d9b8 REGS: c000000118e7b570 TRAP: 0700 Not tainted (2.6.35-rc3-perf-00008-g76b0f13 ) MSR: 9000000000029032 CR: 44004424 XER: 000fffff TASK = c0000001187dcad0[3143] 'perf' THREAD: c000000118e78000 CPU: 1 GPR00: c0000000000cb46c c000000118e7b7f0 c0000000009866a0 0000000000000020 GPR04: 0000000000000000 000000000000001d 0000000000000000 0000000000000001 GPR08: c0000000009bed68 c00000000086dff8 c000000000a5bf10 0000000000000001 GPR12: 0000000024004422 c00000000ffff200 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000018 00000000101150f4 GPR20: 0000000010206b40 0000000000000000 0000000000000000 00000000101150f4 GPR24: c0000001199090c0 0000000000000001 0000000000000000 0000000000000001 GPR28: 0000000000000000 0000000000000000 c0000000008ec290 0000000000000000 NIP [c0000000000cb470] .task_bp_pinned+0x5c/0x12c LR [c0000000000cb46c] .task_bp_pinned+0x58/0x12c Call Trace: [c000000118e7b7f0] [c0000000000cb46c] .task_bp_pinned+0x58/0x12c (unreliable) [c000000118e7b8a0] [c0000000000cb584] .toggle_bp_task_slot+0x44/0xe4 [c000000118e7b940] [c0000000000cb6c8] .toggle_bp_slot+0xa4/0x164 [c000000118e7b9f0] [c0000000000cbafc] .release_bp_slot+0x44/0x6c [c000000118e7ba80] [c0000000000c4178] .bp_perf_event_destroy+0x10/0x24 [c000000118e7bb00] [c0000000000c4aec] .free_event+0x180/0x1bc [c000000118e7bbc0] [c0000000000c54c4] .perf_event_release_kernel+0x14c/0x170 Reported-by: Paul Mackerras Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Prasad Cc: Mahesh Salgaonkar Cc: Will Deacon Cc: Jason Wessel --- include/linux/perf_event.h | 6 ++-- kernel/hw_breakpoint.c | 78 ++++++++++++++++++++++++---------------------- 2 files changed, 45 insertions(+), 39 deletions(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 63b5aa5dce69..0dd5f8ad77ac 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -533,8 +533,10 @@ struct hw_perf_event { struct hrtimer hrtimer; }; #ifdef CONFIG_HAVE_HW_BREAKPOINT - /* breakpoint */ - struct arch_hw_breakpoint info; + struct { /* breakpoint */ + struct arch_hw_breakpoint info; + struct list_head bp_list; + }; #endif }; local64_t prev_count; diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c index 7a56b22e0602..e34d94d50924 100644 --- a/kernel/hw_breakpoint.c +++ b/kernel/hw_breakpoint.c @@ -41,6 +41,7 @@ #include #include #include +#include #include #include @@ -62,6 +63,9 @@ static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]); static int nr_slots[TYPE_MAX]; +/* Keep track of the breakpoints attached to tasks */ +static LIST_HEAD(bp_task_head); + static int constraints_initialized; /* Gather the number of total pinned and un-pinned bp in a cpuset */ @@ -103,33 +107,21 @@ static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type) return 0; } -static int task_bp_pinned(struct task_struct *tsk, enum bp_type_idx type) +/* + * Count the number of breakpoints of the same type and same task. + * The given event must be not on the list. + */ +static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type) { - struct perf_event_context *ctx = tsk->perf_event_ctxp; - struct list_head *list; - struct perf_event *bp; - unsigned long flags; + struct perf_event_context *ctx = bp->ctx; + struct perf_event *iter; int count = 0; - if (WARN_ONCE(!ctx, "No perf context for this task")) - return 0; - - list = &ctx->event_list; - - raw_spin_lock_irqsave(&ctx->lock, flags); - - /* - * The current breakpoint counter is not included in the list - * at the open() callback time - */ - list_for_each_entry(bp, list, event_entry) { - if (bp->attr.type == PERF_TYPE_BREAKPOINT) - if (find_slot_idx(bp) == type) - count += hw_breakpoint_weight(bp); + list_for_each_entry(iter, &bp_task_head, hw.bp_list) { + if (iter->ctx == ctx && find_slot_idx(iter) == type) + count += hw_breakpoint_weight(iter); } - raw_spin_unlock_irqrestore(&ctx->lock, flags); - return count; } @@ -149,7 +141,7 @@ fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp, if (!tsk) slots->pinned += max_task_bp_pinned(cpu, type); else - slots->pinned += task_bp_pinned(tsk, type); + slots->pinned += task_bp_pinned(bp, type); slots->flexible = per_cpu(nr_bp_flexible[type], cpu); return; @@ -162,7 +154,7 @@ fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp, if (!tsk) nr += max_task_bp_pinned(cpu, type); else - nr += task_bp_pinned(tsk, type); + nr += task_bp_pinned(bp, type); if (nr > slots->pinned) slots->pinned = nr; @@ -188,7 +180,7 @@ fetch_this_slot(struct bp_busy_slots *slots, int weight) /* * Add a pinned breakpoint for the given task in our constraint table */ -static void toggle_bp_task_slot(struct task_struct *tsk, int cpu, bool enable, +static void toggle_bp_task_slot(struct perf_event *bp, int cpu, bool enable, enum bp_type_idx type, int weight) { unsigned int *tsk_pinned; @@ -196,10 +188,11 @@ static void toggle_bp_task_slot(struct task_struct *tsk, int cpu, bool enable, int old_idx = 0; int idx = 0; - old_count = task_bp_pinned(tsk, type); + old_count = task_bp_pinned(bp, type); old_idx = old_count - 1; idx = old_idx + weight; + /* tsk_pinned[n] is the number of tasks having n breakpoints */ tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu); if (enable) { tsk_pinned[idx]++; @@ -222,23 +215,30 @@ toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type, int cpu = bp->cpu; struct task_struct *tsk = bp->ctx->task; + /* Pinned counter cpu profiling */ + if (!tsk) { + + if (enable) + per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight; + else + per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight; + return; + } + /* Pinned counter task profiling */ - if (tsk) { - if (cpu >= 0) { - toggle_bp_task_slot(tsk, cpu, enable, type, weight); - return; - } + if (!enable) + list_del(&bp->hw.bp_list); + + if (cpu >= 0) { + toggle_bp_task_slot(bp, cpu, enable, type, weight); + } else { for_each_online_cpu(cpu) - toggle_bp_task_slot(tsk, cpu, enable, type, weight); - return; + toggle_bp_task_slot(bp, cpu, enable, type, weight); } - /* Pinned counter cpu profiling */ if (enable) - per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight; - else - per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight; + list_add_tail(&bp->hw.bp_list, &bp_task_head); } /* @@ -301,6 +301,10 @@ static int __reserve_bp_slot(struct perf_event *bp) weight = hw_breakpoint_weight(bp); fetch_bp_busy_slots(&slots, bp, type); + /* + * Simulate the addition of this breakpoint to the constraints + * and see the result. + */ fetch_this_slot(&slots, weight); /* Flexible counters need to keep at least one slot */ -- cgit v1.2.3 From 5cfaf214856eb934759ae500a0b812dd06a00bd9 Mon Sep 17 00:00:00 2001 From: Nobuhiro Iwamatsu Date: Wed, 23 Jun 2010 09:17:53 +0900 Subject: perf: Fix argument of perf_arch_fetch_caller_regs "struct regs" was set to argument of perf_arch_fetch_caller_regs off-case. It should be "struct pt_regs". This fixes various build errors in archs that have CONFIG_PERF_EVENTS=y but no overriden implementation of perf_arch_fetch_caller_regs. cc1: warnings being treated as errors In file included from include/linux/ftrace_event.h:8, from include/trace/syscall.h:6, from include/linux/syscalls.h:75, from arch/sh/kernel/sys_sh32.c:9: include/linux/perf_event.h:937: error: 'struct regs' declared inside parameter list include/linux/perf_event.h:937: error: its scope is only this definition or declaration, which is probably not what you want include/linux/perf_event.h: In function 'perf_fetch_caller_regs': include/linux/perf_event.h:952: error: passing argument 1 of 'perf_arch_fetch_caller_regs' from incompatible pointer type Signed-off-by: Nobuhiro Iwamatsu Reported-by: Stephen Rothwell Cc: Paul Mackerras Cc: David Miller Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Frederic Weisbecker --- include/linux/perf_event.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include/linux/perf_event.h') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 0dd5f8ad77ac..937495c25073 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -936,7 +936,7 @@ extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64); #ifndef perf_arch_fetch_caller_regs static inline void -perf_arch_fetch_caller_regs(struct regs *regs, unsigned long ip) { } +perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } #endif /* -- cgit v1.2.3