From fe4b04fa31a6dcf4358aa84cf81e5a7fd079469b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 2 Feb 2011 13:19:09 +0100 Subject: perf: Cure task_oncpu_function_call() races Oleg reported that on architectures with __ARCH_WANT_INTERRUPTS_ON_CTXSW the IPI from task_oncpu_function_call() can land before perf_event_task_sched_in() and cause interesting situations for eg. perf_install_in_context(). This patch reworks the task_oncpu_function_call() interface to give a more usable primitive as well as rework all its users to hopefully be more obvious as well as remove the races. While looking at the code I also found a number of races against perf_event_task_sched_out() which can flip contexts between tasks so plug those too. Reported-and-reviewed-by: Oleg Nesterov Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/sched.h | 7 ------- 1 file changed, 7 deletions(-) (limited to 'include/linux') diff --git a/include/linux/sched.h b/include/linux/sched.h index d747f948b34e..0b40ee3f6d7a 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -2578,13 +2578,6 @@ static inline void inc_syscw(struct task_struct *tsk) #define TASK_SIZE_OF(tsk) TASK_SIZE #endif -/* - * Call the function if the target task is executing on a CPU right now: - */ -extern void task_oncpu_function_call(struct task_struct *p, - void (*func) (void *info), void *info); - - #ifdef CONFIG_MM_OWNER extern void mm_update_next_owner(struct mm_struct *mm); extern void mm_init_owner(struct mm_struct *mm, struct task_struct *p); -- cgit v1.2.3 From 4a3d27e98a7f2682e96d6f863752e0424b00d691 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Thu, 27 Jan 2011 23:19:49 -0500 Subject: tracing/filter: Move MAX_FILTER_PRED to local tracing directory The MAX_FILTER_PRED is only needed by the kernel/trace/*.c files. Move it to kernel/trace/trace.h. Cc: Tom Zanussi Signed-off-by: Steven Rostedt --- include/linux/ftrace_event.h | 1 - kernel/trace/trace.h | 2 ++ 2 files changed, 2 insertions(+), 1 deletion(-) (limited to 'include/linux') diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index 47e3997f7b5c..1a99e7939c2b 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -208,7 +208,6 @@ struct ftrace_event_call { #define PERF_MAX_TRACE_SIZE 2048 -#define MAX_FILTER_PRED 32 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */ extern void destroy_preds(struct ftrace_event_call *call); diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index d754330934bb..fbff872f8db1 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -680,6 +680,8 @@ struct event_subsystem { #define FILTER_PRED_IS_RIGHT (1 << 15) #define FILTER_PRED_FOLD (1 << 15) +#define MAX_FILTER_PRED 32 + struct filter_pred; struct regex; -- cgit v1.2.3 From ba976970c79fd2fbfe1a4b3b6766a318f4eb9d4c Mon Sep 17 00:00:00 2001 From: Ian Munsie Date: Thu, 3 Feb 2011 14:27:20 +1100 Subject: tracing/syscalls: Don't add events for unmapped syscalls FTRACE_SYSCALLS would create events for each and every system call, even if it had failed to map the system call's name with it's number. This resulted in a number of events being created that would not behave as expected. This could happen, for example, on architectures who's symbol names are unusual and will not match the system call name. It could also happen with system calls which were mapped to sys_ni_syscall. This patch changes the default system call number in the metadata to -1. If the system call name from the metadata is not successfully mapped to a system call number during boot, than the event initialisation routine will now return an error, preventing the event from being created. Signed-off-by: Ian Munsie LKML-Reference: <1296703645-18718-2-git-send-email-imunsie@au1.ibm.com> Signed-off-by: Steven Rostedt --- include/linux/syscalls.h | 2 ++ kernel/trace/trace_syscalls.c | 8 ++++++++ 2 files changed, 10 insertions(+) (limited to 'include/linux') diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h index 98664db1be47..8e8968e74544 100644 --- a/include/linux/syscalls.h +++ b/include/linux/syscalls.h @@ -158,6 +158,7 @@ extern struct trace_event_functions exit_syscall_print_funcs; static struct syscall_metadata __used \ __syscall_meta_##sname = { \ .name = "sys"#sname, \ + .syscall_nr = -1, /* Filled in at boot */ \ .nb_args = nb, \ .types = types_##sname, \ .args = args_##sname, \ @@ -175,6 +176,7 @@ extern struct trace_event_functions exit_syscall_print_funcs; static struct syscall_metadata __used \ __syscall_meta__##sname = { \ .name = "sys_"#sname, \ + .syscall_nr = -1, /* Filled in at boot */ \ .nb_args = 0, \ .enter_event = &event_enter__##sname, \ .exit_event = &event_exit__##sname, \ diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c index 5c9fe08d2093..a9ceabd52247 100644 --- a/kernel/trace/trace_syscalls.c +++ b/kernel/trace/trace_syscalls.c @@ -424,6 +424,14 @@ void unreg_event_syscall_exit(struct ftrace_event_call *call) int init_syscall_trace(struct ftrace_event_call *call) { int id; + int num; + + num = ((struct syscall_metadata *)call->data)->syscall_nr; + if (num < 0 || num >= NR_syscalls) { + pr_debug("syscall %s metadata not mapped, disabling ftrace event\n", + ((struct syscall_metadata *)call->data)->name); + return -ENOSYS; + } if (set_syscall_print_fmt(call) < 0) return -ENOMEM; -- cgit v1.2.3 From f4d5c029bd6731baac0937324cef0f746e7d5ea7 Mon Sep 17 00:00:00 2001 From: Lai Jiangshan Date: Wed, 26 Jan 2011 16:49:00 +0800 Subject: tracing: Compile time initialization for event flags value Compile time initialization is better than runtime initialization. Remove many early_initcall()s and many trace_init_flags_##name()s. Acked-by: Frederic Weisbecker Signed-off-by: Lai Jiangshan LKML-Reference: <4D3FDFFC.6030304@cn.fujitsu.com> Signed-off-by: Steven Rostedt --- include/linux/syscalls.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'include/linux') diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h index 8e8968e74544..a17fcea2ca58 100644 --- a/include/linux/syscalls.h +++ b/include/linux/syscalls.h @@ -132,11 +132,11 @@ extern struct trace_event_functions exit_syscall_print_funcs; .class = &event_class_syscall_enter, \ .event.funcs = &enter_syscall_print_funcs, \ .data = (void *)&__syscall_meta_##sname,\ + .flags = TRACE_EVENT_FL_CAP_ANY, \ }; \ static struct ftrace_event_call __used \ __attribute__((section("_ftrace_events"))) \ - *__event_enter_##sname = &event_enter_##sname; \ - __TRACE_EVENT_FLAGS(enter_##sname, TRACE_EVENT_FL_CAP_ANY) + *__event_enter_##sname = &event_enter_##sname; #define SYSCALL_TRACE_EXIT_EVENT(sname) \ static struct syscall_metadata __syscall_meta_##sname; \ @@ -146,11 +146,11 @@ extern struct trace_event_functions exit_syscall_print_funcs; .class = &event_class_syscall_exit, \ .event.funcs = &exit_syscall_print_funcs, \ .data = (void *)&__syscall_meta_##sname,\ + .flags = TRACE_EVENT_FL_CAP_ANY, \ }; \ static struct ftrace_event_call __used \ __attribute__((section("_ftrace_events"))) \ - *__event_exit_##sname = &event_exit_##sname; \ - __TRACE_EVENT_FLAGS(exit_##sname, TRACE_EVENT_FL_CAP_ANY) + *__event_exit_##sname = &event_exit_##sname; #define SYSCALL_METADATA(sname, nb) \ SYSCALL_TRACE_ENTER_EVENT(sname); \ -- cgit v1.2.3 From 868baf07b1a259f5f3803c1dc2777b6c358f83cf Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Thu, 10 Feb 2011 21:26:13 -0500 Subject: ftrace: Fix memory leak with function graph and cpu hotplug When the fuction graph tracer starts, it needs to make a special stack for each task to save the real return values of the tasks. All running tasks have this stack created, as well as any new tasks. On CPU hot plug, the new idle task will allocate a stack as well when init_idle() is called. The problem is that cpu hotplug does not create a new idle_task. Instead it uses the idle task that existed when the cpu went down. ftrace_graph_init_task() will add a new ret_stack to the task that is given to it. Because a clone will make the task have a stack of its parent it does not check if the task's ret_stack is already NULL or not. When the CPU hotplug code starts a CPU up again, it will allocate a new stack even though one already existed for it. The solution is to treat the idle_task specially. In fact, the function_graph code already does, just not at init_idle(). Instead of using the ftrace_graph_init_task() for the idle task, which that function expects the task to be a clone, have a separate ftrace_graph_init_idle_task(). Also, we will create a per_cpu ret_stack that is used by the idle task. When we call ftrace_graph_init_idle_task() it will check if the idle task's ret_stack is NULL, if it is, then it will assign it the per_cpu ret_stack. Reported-by: Benjamin Herrenschmidt Suggested-by: Peter Zijlstra Cc: Stable Tree Signed-off-by: Steven Rostedt --- include/linux/ftrace.h | 2 ++ kernel/sched.c | 2 +- kernel/trace/ftrace.c | 52 +++++++++++++++++++++++++++++++++++++++++++------- 3 files changed, 48 insertions(+), 8 deletions(-) (limited to 'include/linux') diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h index dcd6a7c3a435..ca29e03c1fac 100644 --- a/include/linux/ftrace.h +++ b/include/linux/ftrace.h @@ -428,6 +428,7 @@ extern void unregister_ftrace_graph(void); extern void ftrace_graph_init_task(struct task_struct *t); extern void ftrace_graph_exit_task(struct task_struct *t); +extern void ftrace_graph_init_idle_task(struct task_struct *t, int cpu); static inline int task_curr_ret_stack(struct task_struct *t) { @@ -451,6 +452,7 @@ static inline void unpause_graph_tracing(void) static inline void ftrace_graph_init_task(struct task_struct *t) { } static inline void ftrace_graph_exit_task(struct task_struct *t) { } +static inline void ftrace_graph_init_idle_task(struct task_struct *t, int cpu) { } static inline int register_ftrace_graph(trace_func_graph_ret_t retfunc, trace_func_graph_ent_t entryfunc) diff --git a/kernel/sched.c b/kernel/sched.c index 18d38e4ec7ba..fbe86cb04b61 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -5571,7 +5571,7 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu) * The idle tasks have their own, simple scheduling class: */ idle->sched_class = &idle_sched_class; - ftrace_graph_init_task(idle); + ftrace_graph_init_idle_task(idle, cpu); } /* diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index f3dadae83883..888b611897d3 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -3328,7 +3328,7 @@ static int start_graph_tracing(void) /* The cpu_boot init_task->ret_stack will never be freed */ for_each_online_cpu(cpu) { if (!idle_task(cpu)->ret_stack) - ftrace_graph_init_task(idle_task(cpu)); + ftrace_graph_init_idle_task(idle_task(cpu), cpu); } do { @@ -3418,6 +3418,49 @@ void unregister_ftrace_graph(void) mutex_unlock(&ftrace_lock); } +static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack); + +static void +graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack) +{ + atomic_set(&t->tracing_graph_pause, 0); + atomic_set(&t->trace_overrun, 0); + t->ftrace_timestamp = 0; + /* make curr_ret_stack visable before we add the ret_stack */ + smp_wmb(); + t->ret_stack = ret_stack; +} + +/* + * Allocate a return stack for the idle task. May be the first + * time through, or it may be done by CPU hotplug online. + */ +void ftrace_graph_init_idle_task(struct task_struct *t, int cpu) +{ + t->curr_ret_stack = -1; + /* + * The idle task has no parent, it either has its own + * stack or no stack at all. + */ + if (t->ret_stack) + WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu)); + + if (ftrace_graph_active) { + struct ftrace_ret_stack *ret_stack; + + ret_stack = per_cpu(idle_ret_stack, cpu); + if (!ret_stack) { + ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH + * sizeof(struct ftrace_ret_stack), + GFP_KERNEL); + if (!ret_stack) + return; + per_cpu(idle_ret_stack, cpu) = ret_stack; + } + graph_init_task(t, ret_stack); + } +} + /* Allocate a return stack for newly created task */ void ftrace_graph_init_task(struct task_struct *t) { @@ -3433,12 +3476,7 @@ void ftrace_graph_init_task(struct task_struct *t) GFP_KERNEL); if (!ret_stack) return; - atomic_set(&t->tracing_graph_pause, 0); - atomic_set(&t->trace_overrun, 0); - t->ftrace_timestamp = 0; - /* make curr_ret_stack visable before we add the ret_stack */ - smp_wmb(); - t->ret_stack = ret_stack; + graph_init_task(t, ret_stack); } } -- cgit v1.2.3 From d41d5a01631af821d3a3447e6613a316f5ee6c25 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 7 Feb 2011 17:02:20 +0100 Subject: cgroup: Fix cgroup_subsys::exit callback Make the ::exit method act like ::attach, it is after all very nearly the same thing. The bug had no effect on correctness - fixing it is an optimization for the scheduler. Also, later perf-cgroups patches rely on it. Signed-off-by: Peter Zijlstra Acked-by: Paul Menage LKML-Reference: <1297160655.13327.92.camel@laptop> Signed-off-by: Ingo Molnar --- include/linux/cgroup.h | 3 ++- kernel/cgroup.c | 31 ++++++++++++++++++------------- kernel/sched.c | 6 ++---- 3 files changed, 22 insertions(+), 18 deletions(-) (limited to 'include/linux') diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index ce104e33cd22..38117d937332 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -474,7 +474,8 @@ struct cgroup_subsys { struct cgroup *old_cgrp, struct task_struct *tsk, bool threadgroup); void (*fork)(struct cgroup_subsys *ss, struct task_struct *task); - void (*exit)(struct cgroup_subsys *ss, struct task_struct *task); + void (*exit)(struct cgroup_subsys *ss, struct cgroup *cgrp, + struct cgroup *old_cgrp, struct task_struct *task); int (*populate)(struct cgroup_subsys *ss, struct cgroup *cgrp); void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp); diff --git a/kernel/cgroup.c b/kernel/cgroup.c index b24d7027b83c..f6495f33a355 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c @@ -4230,20 +4230,8 @@ void cgroup_post_fork(struct task_struct *child) */ void cgroup_exit(struct task_struct *tsk, int run_callbacks) { - int i; struct css_set *cg; - - if (run_callbacks && need_forkexit_callback) { - /* - * modular subsystems can't use callbacks, so no need to lock - * the subsys array - */ - for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { - struct cgroup_subsys *ss = subsys[i]; - if (ss->exit) - ss->exit(ss, tsk); - } - } + int i; /* * Unlink from the css_set task list if necessary. @@ -4261,7 +4249,24 @@ void cgroup_exit(struct task_struct *tsk, int run_callbacks) task_lock(tsk); cg = tsk->cgroups; tsk->cgroups = &init_css_set; + + if (run_callbacks && need_forkexit_callback) { + /* + * modular subsystems can't use callbacks, so no need to lock + * the subsys array + */ + for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { + struct cgroup_subsys *ss = subsys[i]; + if (ss->exit) { + struct cgroup *old_cgrp = + rcu_dereference_raw(cg->subsys[i])->cgroup; + struct cgroup *cgrp = task_cgroup(tsk, i); + ss->exit(ss, cgrp, old_cgrp, tsk); + } + } + } task_unlock(tsk); + if (cg) put_css_set_taskexit(cg); } diff --git a/kernel/sched.c b/kernel/sched.c index e142e92f38da..79e611cd83dd 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -606,9 +606,6 @@ static inline struct task_group *task_group(struct task_struct *p) struct task_group *tg; struct cgroup_subsys_state *css; - if (p->flags & PF_EXITING) - return &root_task_group; - css = task_subsys_state_check(p, cpu_cgroup_subsys_id, lockdep_is_held(&task_rq(p)->lock)); tg = container_of(css, struct task_group, css); @@ -8863,7 +8860,8 @@ cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, } static void -cpu_cgroup_exit(struct cgroup_subsys *ss, struct task_struct *task) +cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, + struct cgroup *old_cgrp, struct task_struct *task) { /* * cgroup_exit() is called in the copy_process() failure path. -- cgit v1.2.3 From e5d1367f17ba6a6fed5fd8b74e4d5720923e0c25 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Mon, 14 Feb 2011 11:20:01 +0200 Subject: perf: Add cgroup support This kernel patch adds the ability to filter monitoring based on container groups (cgroups). This is for use in per-cpu mode only. The cgroup to monitor is passed as a file descriptor in the pid argument to the syscall. The file descriptor must be opened to the cgroup name in the cgroup filesystem. For instance, if the cgroup name is foo and cgroupfs is mounted in /cgroup, then the file descriptor is opened to /cgroup/foo. Cgroup mode is activated by passing PERF_FLAG_PID_CGROUP in the flags argument to the syscall. For instance to measure in cgroup foo on CPU1 assuming cgroupfs is mounted under /cgroup: struct perf_event_attr attr; int cgroup_fd, fd; cgroup_fd = open("/cgroup/foo", O_RDONLY); fd = perf_event_open(&attr, cgroup_fd, 1, -1, PERF_FLAG_PID_CGROUP); close(cgroup_fd); Signed-off-by: Stephane Eranian [ added perf_cgroup_{exit,attach} ] Signed-off-by: Peter Zijlstra LKML-Reference: <4d590250.114ddf0a.689e.4482@mx.google.com> Signed-off-by: Ingo Molnar --- include/linux/cgroup.h | 1 + include/linux/cgroup_subsys.h | 4 + include/linux/perf_event.h | 33 ++- init/Kconfig | 10 + kernel/cgroup.c | 23 ++ kernel/perf_event.c | 638 +++++++++++++++++++++++++++++++++++++++--- 6 files changed, 671 insertions(+), 38 deletions(-) (limited to 'include/linux') diff --git a/include/linux/cgroup.h b/include/linux/cgroup.h index 38117d937332..e654fa239916 100644 --- a/include/linux/cgroup.h +++ b/include/linux/cgroup.h @@ -627,6 +627,7 @@ bool css_is_ancestor(struct cgroup_subsys_state *cg, /* Get id and depth of css */ unsigned short css_id(struct cgroup_subsys_state *css); unsigned short css_depth(struct cgroup_subsys_state *css); +struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id); #else /* !CONFIG_CGROUPS */ diff --git a/include/linux/cgroup_subsys.h b/include/linux/cgroup_subsys.h index ccefff02b6cb..cdbfcb8780ec 100644 --- a/include/linux/cgroup_subsys.h +++ b/include/linux/cgroup_subsys.h @@ -65,4 +65,8 @@ SUBSYS(net_cls) SUBSYS(blkio) #endif +#ifdef CONFIG_CGROUP_PERF +SUBSYS(perf) +#endif + /* */ diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index dda5b0a3ff60..38c8b2554842 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -464,6 +464,7 @@ enum perf_callchain_context { #define PERF_FLAG_FD_NO_GROUP (1U << 0) #define PERF_FLAG_FD_OUTPUT (1U << 1) +#define PERF_FLAG_PID_CGROUP (1U << 2) /* pid=cgroup id, per-cpu mode only */ #ifdef __KERNEL__ /* @@ -471,6 +472,7 @@ enum perf_callchain_context { */ #ifdef CONFIG_PERF_EVENTS +# include # include # include #endif @@ -716,6 +718,22 @@ struct swevent_hlist { #define PERF_ATTACH_GROUP 0x02 #define PERF_ATTACH_TASK 0x04 +#ifdef CONFIG_CGROUP_PERF +/* + * perf_cgroup_info keeps track of time_enabled for a cgroup. + * This is a per-cpu dynamically allocated data structure. + */ +struct perf_cgroup_info { + u64 time; + u64 timestamp; +}; + +struct perf_cgroup { + struct cgroup_subsys_state css; + struct perf_cgroup_info *info; /* timing info, one per cpu */ +}; +#endif + /** * struct perf_event - performance event kernel representation: */ @@ -832,6 +850,11 @@ struct perf_event { struct event_filter *filter; #endif +#ifdef CONFIG_CGROUP_PERF + struct perf_cgroup *cgrp; /* cgroup event is attach to */ + int cgrp_defer_enabled; +#endif + #endif /* CONFIG_PERF_EVENTS */ }; @@ -886,6 +909,7 @@ struct perf_event_context { u64 generation; int pin_count; struct rcu_head rcu_head; + int nr_cgroups; /* cgroup events present */ }; /* @@ -905,6 +929,9 @@ struct perf_cpu_context { struct list_head rotation_list; int jiffies_interval; struct pmu *active_pmu; +#ifdef CONFIG_CGROUP_PERF + struct perf_cgroup *cgrp; +#endif }; struct perf_output_handle { @@ -1040,11 +1067,11 @@ have_event: __perf_sw_event(event_id, nr, nmi, regs, addr); } -extern atomic_t perf_task_events; +extern atomic_t perf_sched_events; static inline void perf_event_task_sched_in(struct task_struct *task) { - COND_STMT(&perf_task_events, __perf_event_task_sched_in(task)); + COND_STMT(&perf_sched_events, __perf_event_task_sched_in(task)); } static inline @@ -1052,7 +1079,7 @@ void perf_event_task_sched_out(struct task_struct *task, struct task_struct *nex { perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0); - COND_STMT(&perf_task_events, __perf_event_task_sched_out(task, next)); + COND_STMT(&perf_sched_events, __perf_event_task_sched_out(task, next)); } extern void perf_event_mmap(struct vm_area_struct *vma); diff --git a/init/Kconfig b/init/Kconfig index be788c0957d4..20d6bd919b8d 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -683,6 +683,16 @@ config CGROUP_MEM_RES_CTLR_SWAP_ENABLED select this option (if, for some reason, they need to disable it then noswapaccount does the trick). +config CGROUP_PERF + bool "Enable perf_event per-cpu per-container group (cgroup) monitoring" + depends on PERF_EVENTS && CGROUPS + help + This option extends the per-cpu mode to restrict monitoring to + threads which belong to the cgroup specificied and run on the + designated cpu. + + Say N if unsure. + menuconfig CGROUP_SCHED bool "Group CPU scheduler" depends on EXPERIMENTAL diff --git a/kernel/cgroup.c b/kernel/cgroup.c index f6495f33a355..95362d15128c 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c @@ -4818,6 +4818,29 @@ css_get_next(struct cgroup_subsys *ss, int id, return ret; } +/* + * get corresponding css from file open on cgroupfs directory + */ +struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id) +{ + struct cgroup *cgrp; + struct inode *inode; + struct cgroup_subsys_state *css; + + inode = f->f_dentry->d_inode; + /* check in cgroup filesystem dir */ + if (inode->i_op != &cgroup_dir_inode_operations) + return ERR_PTR(-EBADF); + + if (id < 0 || id >= CGROUP_SUBSYS_COUNT) + return ERR_PTR(-EINVAL); + + /* get cgroup */ + cgrp = __d_cgrp(f->f_dentry); + css = cgrp->subsys[id]; + return css ? css : ERR_PTR(-ENOENT); +} + #ifdef CONFIG_CGROUP_DEBUG static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss, struct cgroup *cont) diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 3d3f282fa50e..65dcdc76d709 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -111,13 +111,23 @@ static int cpu_function_call(int cpu, int (*func) (void *info), void *info) return data.ret; } +#define PERF_FLAG_ALL (PERF_FLAG_FD_NO_GROUP |\ + PERF_FLAG_FD_OUTPUT |\ + PERF_FLAG_PID_CGROUP) + enum event_type_t { EVENT_FLEXIBLE = 0x1, EVENT_PINNED = 0x2, EVENT_ALL = EVENT_FLEXIBLE | EVENT_PINNED, }; -atomic_t perf_task_events __read_mostly; +/* + * perf_sched_events : >0 events exist + * perf_cgroup_events: >0 per-cpu cgroup events exist on this cpu + */ +atomic_t perf_sched_events __read_mostly; +static DEFINE_PER_CPU(atomic_t, perf_cgroup_events); + static atomic_t nr_mmap_events __read_mostly; static atomic_t nr_comm_events __read_mostly; static atomic_t nr_task_events __read_mostly; @@ -148,7 +158,11 @@ static void cpu_ctx_sched_out(struct perf_cpu_context *cpuctx, enum event_type_t event_type); static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx, - enum event_type_t event_type); + enum event_type_t event_type, + struct task_struct *task); + +static void update_context_time(struct perf_event_context *ctx); +static u64 perf_event_time(struct perf_event *event); void __weak perf_event_print_debug(void) { } @@ -162,6 +176,338 @@ static inline u64 perf_clock(void) return local_clock(); } +static inline struct perf_cpu_context * +__get_cpu_context(struct perf_event_context *ctx) +{ + return this_cpu_ptr(ctx->pmu->pmu_cpu_context); +} + +#ifdef CONFIG_CGROUP_PERF + +static inline struct perf_cgroup * +perf_cgroup_from_task(struct task_struct *task) +{ + return container_of(task_subsys_state(task, perf_subsys_id), + struct perf_cgroup, css); +} + +static inline bool +perf_cgroup_match(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + struct perf_cpu_context *cpuctx = __get_cpu_context(ctx); + + return !event->cgrp || event->cgrp == cpuctx->cgrp; +} + +static inline void perf_get_cgroup(struct perf_event *event) +{ + css_get(&event->cgrp->css); +} + +static inline void perf_put_cgroup(struct perf_event *event) +{ + css_put(&event->cgrp->css); +} + +static inline void perf_detach_cgroup(struct perf_event *event) +{ + perf_put_cgroup(event); + event->cgrp = NULL; +} + +static inline int is_cgroup_event(struct perf_event *event) +{ + return event->cgrp != NULL; +} + +static inline u64 perf_cgroup_event_time(struct perf_event *event) +{ + struct perf_cgroup_info *t; + + t = per_cpu_ptr(event->cgrp->info, event->cpu); + return t->time; +} + +static inline void __update_cgrp_time(struct perf_cgroup *cgrp) +{ + struct perf_cgroup_info *info; + u64 now; + + now = perf_clock(); + + info = this_cpu_ptr(cgrp->info); + + info->time += now - info->timestamp; + info->timestamp = now; +} + +static inline void update_cgrp_time_from_cpuctx(struct perf_cpu_context *cpuctx) +{ + struct perf_cgroup *cgrp_out = cpuctx->cgrp; + if (cgrp_out) + __update_cgrp_time(cgrp_out); +} + +static inline void update_cgrp_time_from_event(struct perf_event *event) +{ + struct perf_cgroup *cgrp = perf_cgroup_from_task(current); + /* + * do not update time when cgroup is not active + */ + if (!event->cgrp || cgrp != event->cgrp) + return; + + __update_cgrp_time(event->cgrp); +} + +static inline void +perf_cgroup_set_timestamp(struct task_struct *task, u64 now) +{ + struct perf_cgroup *cgrp; + struct perf_cgroup_info *info; + + if (!task) + return; + + cgrp = perf_cgroup_from_task(task); + info = this_cpu_ptr(cgrp->info); + info->timestamp = now; +} + +#define PERF_CGROUP_SWOUT 0x1 /* cgroup switch out every event */ +#define PERF_CGROUP_SWIN 0x2 /* cgroup switch in events based on task */ + +/* + * reschedule events based on the cgroup constraint of task. + * + * mode SWOUT : schedule out everything + * mode SWIN : schedule in based on cgroup for next + */ +void perf_cgroup_switch(struct task_struct *task, int mode) +{ + struct perf_cpu_context *cpuctx; + struct pmu *pmu; + unsigned long flags; + + /* + * disable interrupts to avoid geting nr_cgroup + * changes via __perf_event_disable(). Also + * avoids preemption. + */ + local_irq_save(flags); + + /* + * we reschedule only in the presence of cgroup + * constrained events. + */ + rcu_read_lock(); + + list_for_each_entry_rcu(pmu, &pmus, entry) { + + cpuctx = this_cpu_ptr(pmu->pmu_cpu_context); + + perf_pmu_disable(cpuctx->ctx.pmu); + + /* + * perf_cgroup_events says at least one + * context on this CPU has cgroup events. + * + * ctx->nr_cgroups reports the number of cgroup + * events for a context. + */ + if (cpuctx->ctx.nr_cgroups > 0) { + + if (mode & PERF_CGROUP_SWOUT) { + cpu_ctx_sched_out(cpuctx, EVENT_ALL); + /* + * must not be done before ctxswout due + * to event_filter_match() in event_sched_out() + */ + cpuctx->cgrp = NULL; + } + + if (mode & PERF_CGROUP_SWIN) { + /* set cgrp before ctxsw in to + * allow event_filter_match() to not + * have to pass task around + */ + cpuctx->cgrp = perf_cgroup_from_task(task); + cpu_ctx_sched_in(cpuctx, EVENT_ALL, task); + } + } + + perf_pmu_enable(cpuctx->ctx.pmu); + } + + rcu_read_unlock(); + + local_irq_restore(flags); +} + +static inline void perf_cgroup_sched_out(struct task_struct *task) +{ + perf_cgroup_switch(task, PERF_CGROUP_SWOUT); +} + +static inline void perf_cgroup_sched_in(struct task_struct *task) +{ + perf_cgroup_switch(task, PERF_CGROUP_SWIN); +} + +static inline int perf_cgroup_connect(int fd, struct perf_event *event, + struct perf_event_attr *attr, + struct perf_event *group_leader) +{ + struct perf_cgroup *cgrp; + struct cgroup_subsys_state *css; + struct file *file; + int ret = 0, fput_needed; + + file = fget_light(fd, &fput_needed); + if (!file) + return -EBADF; + + css = cgroup_css_from_dir(file, perf_subsys_id); + if (IS_ERR(css)) + return PTR_ERR(css); + + cgrp = container_of(css, struct perf_cgroup, css); + event->cgrp = cgrp; + + /* + * all events in a group must monitor + * the same cgroup because a task belongs + * to only one perf cgroup at a time + */ + if (group_leader && group_leader->cgrp != cgrp) { + perf_detach_cgroup(event); + ret = -EINVAL; + } else { + /* must be done before we fput() the file */ + perf_get_cgroup(event); + } + fput_light(file, fput_needed); + return ret; +} + +static inline void +perf_cgroup_set_shadow_time(struct perf_event *event, u64 now) +{ + struct perf_cgroup_info *t; + t = per_cpu_ptr(event->cgrp->info, event->cpu); + event->shadow_ctx_time = now - t->timestamp; +} + +static inline void +perf_cgroup_defer_enabled(struct perf_event *event) +{ + /* + * when the current task's perf cgroup does not match + * the event's, we need to remember to call the + * perf_mark_enable() function the first time a task with + * a matching perf cgroup is scheduled in. + */ + if (is_cgroup_event(event) && !perf_cgroup_match(event)) + event->cgrp_defer_enabled = 1; +} + +static inline void +perf_cgroup_mark_enabled(struct perf_event *event, + struct perf_event_context *ctx) +{ + struct perf_event *sub; + u64 tstamp = perf_event_time(event); + + if (!event->cgrp_defer_enabled) + return; + + event->cgrp_defer_enabled = 0; + + event->tstamp_enabled = tstamp - event->total_time_enabled; + list_for_each_entry(sub, &event->sibling_list, group_entry) { + if (sub->state >= PERF_EVENT_STATE_INACTIVE) { + sub->tstamp_enabled = tstamp - sub->total_time_enabled; + sub->cgrp_defer_enabled = 0; + } + } +} +#else /* !CONFIG_CGROUP_PERF */ + +static inline bool +perf_cgroup_match(struct perf_event *event) +{ + return true; +} + +static inline void perf_detach_cgroup(struct perf_event *event) +{} + +static inline int is_cgroup_event(struct perf_event *event) +{ + return 0; +} + +static inline u64 perf_cgroup_event_cgrp_time(struct perf_event *event) +{ + return 0; +} + +static inline void update_cgrp_time_from_event(struct perf_event *event) +{ +} + +static inline void update_cgrp_time_from_cpuctx(struct perf_cpu_context *cpuctx) +{ +} + +static inline void perf_cgroup_sched_out(struct task_struct *task) +{ +} + +static inline void perf_cgroup_sched_in(struct task_struct *task) +{ +} + +static inline int perf_cgroup_connect(pid_t pid, struct perf_event *event, + struct perf_event_attr *attr, + struct perf_event *group_leader) +{ + return -EINVAL; +} + +static inline void +perf_cgroup_set_timestamp(struct task_struct *task, u64 now) +{ +} + +void +perf_cgroup_switch(struct task_struct *task, struct task_struct *next) +{ +} + +static inline void +perf_cgroup_set_shadow_time(struct perf_event *event, u64 now) +{ +} + +static inline u64 perf_cgroup_event_time(struct perf_event *event) +{ + return 0; +} + +static inline void +perf_cgroup_defer_enabled(struct perf_event *event) +{ +} + +static inline void +perf_cgroup_mark_enabled(struct perf_event *event, + struct perf_event_context *ctx) +{ +} +#endif + void perf_pmu_disable(struct pmu *pmu) { int *count = this_cpu_ptr(pmu->pmu_disable_count); @@ -343,6 +689,10 @@ static void update_context_time(struct perf_event_context *ctx) static u64 perf_event_time(struct perf_event *event) { struct perf_event_context *ctx = event->ctx; + + if (is_cgroup_event(event)) + return perf_cgroup_event_time(event); + return ctx ? ctx->time : 0; } @@ -357,9 +707,20 @@ static void update_event_times(struct perf_event *event) if (event->state < PERF_EVENT_STATE_INACTIVE || event->group_leader->state < PERF_EVENT_STATE_INACTIVE) return; - - if (ctx->is_active) + /* + * in cgroup mode, time_enabled represents + * the time the event was enabled AND active + * tasks were in the monitored cgroup. This is + * independent of the activity of the context as + * there may be a mix of cgroup and non-cgroup events. + * + * That is why we treat cgroup events differently + * here. + */ + if (is_cgroup_event(event)) run_end = perf_event_time(event); + else if (ctx->is_active) + run_end = ctx->time; else run_end = event->tstamp_stopped; @@ -371,6 +732,7 @@ static void update_event_times(struct perf_event *event) run_end = perf_event_time(event); event->total_time_running = run_end - event->tstamp_running; + } /* @@ -419,6 +781,17 @@ list_add_event(struct perf_event *event, struct perf_event_context *ctx) list_add_tail(&event->group_entry, list); } + if (is_cgroup_event(event)) { + ctx->nr_cgroups++; + /* + * one more event: + * - that has cgroup constraint on event->cpu + * - that may need work on context switch + */ + atomic_inc(&per_cpu(perf_cgroup_events, event->cpu)); + jump_label_inc(&perf_sched_events); + } + list_add_rcu(&event->event_entry, &ctx->event_list); if (!ctx->nr_events) perf_pmu_rotate_start(ctx->pmu); @@ -545,6 +918,12 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx) event->attach_state &= ~PERF_ATTACH_CONTEXT; + if (is_cgroup_event(event)) { + ctx->nr_cgroups--; + atomic_dec(&per_cpu(perf_cgroup_events, event->cpu)); + jump_label_dec(&perf_sched_events); + } + ctx->nr_events--; if (event->attr.inherit_stat) ctx->nr_stat--; @@ -616,7 +995,8 @@ out: static inline int event_filter_match(struct perf_event *event) { - return event->cpu == -1 || event->cpu == smp_processor_id(); + return (event->cpu == -1 || event->cpu == smp_processor_id()) + && perf_cgroup_match(event); } static void @@ -634,7 +1014,7 @@ event_sched_out(struct perf_event *event, */ if (event->state == PERF_EVENT_STATE_INACTIVE && !event_filter_match(event)) { - delta = ctx->time - event->tstamp_stopped; + delta = tstamp - event->tstamp_stopped; event->tstamp_running += delta; event->tstamp_stopped = tstamp; } @@ -678,12 +1058,6 @@ group_sched_out(struct perf_event *group_event, cpuctx->exclusive = 0; } -static inline struct perf_cpu_context * -__get_cpu_context(struct perf_event_context *ctx) -{ - return this_cpu_ptr(ctx->pmu->pmu_cpu_context); -} - /* * Cross CPU call to remove a performance event * @@ -783,6 +1157,7 @@ static int __perf_event_disable(void *info) */ if (event->state >= PERF_EVENT_STATE_INACTIVE) { update_context_time(ctx); + update_cgrp_time_from_event(event); update_group_times(event); if (event == event->group_leader) group_sched_out(event, cpuctx, ctx); @@ -851,6 +1226,41 @@ retry: raw_spin_unlock_irq(&ctx->lock); } +static void perf_set_shadow_time(struct perf_event *event, + struct perf_event_context *ctx, + u64 tstamp) +{ + /* + * use the correct time source for the time snapshot + * + * We could get by without this by leveraging the + * fact that to get to this function, the caller + * has most likely already called update_context_time() + * and update_cgrp_time_xx() and thus both timestamp + * are identical (or very close). Given that tstamp is, + * already adjusted for cgroup, we could say that: + * tstamp - ctx->timestamp + * is equivalent to + * tstamp - cgrp->timestamp. + * + * Then, in perf_output_read(), the calculation would + * work with no changes because: + * - event is guaranteed scheduled in + * - no scheduled out in between + * - thus the timestamp would be the same + * + * But this is a bit hairy. + * + * So instead, we have an explicit cgroup call to remain + * within the time time source all along. We believe it + * is cleaner and simpler to understand. + */ + if (is_cgroup_event(event)) + perf_cgroup_set_shadow_time(event, tstamp); + else + event->shadow_ctx_time = tstamp - ctx->timestamp; +} + #define MAX_INTERRUPTS (~0ULL) static void perf_log_throttle(struct perf_event *event, int enable); @@ -891,7 +1301,7 @@ event_sched_in(struct perf_event *event, event->tstamp_running += tstamp - event->tstamp_stopped; - event->shadow_ctx_time = tstamp - ctx->timestamp; + perf_set_shadow_time(event, ctx, tstamp); if (!is_software_event(event)) cpuctx->active_oncpu++; @@ -1012,7 +1422,8 @@ static void add_event_to_ctx(struct perf_event *event, event->tstamp_stopped = tstamp; } -static void perf_event_context_sched_in(struct perf_event_context *ctx); +static void perf_event_context_sched_in(struct perf_event_context *ctx, + struct task_struct *tsk); /* * Cross CPU call to install and enable a performance event @@ -1033,11 +1444,17 @@ static int __perf_install_in_context(void *info) * which do context switches with IRQs enabled. */ if (ctx->task && !cpuctx->task_ctx) - perf_event_context_sched_in(ctx); + perf_event_context_sched_in(ctx, ctx->task); raw_spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); + /* + * update cgrp time only if current cgrp + * matches event->cgrp. Must be done before + * calling add_event_to_ctx() + */ + update_cgrp_time_from_event(event); add_event_to_ctx(event, ctx); @@ -1175,10 +1592,19 @@ static int __perf_event_enable(void *info) if (event->state >= PERF_EVENT_STATE_INACTIVE) goto unlock; + + /* + * set current task's cgroup time reference point + */ + perf_cgroup_set_timestamp(current, perf_clock()); + __perf_event_mark_enabled(event, ctx); - if (!event_filter_match(event)) + if (!event_filter_match(event)) { + if (is_cgroup_event(event)) + perf_cgroup_defer_enabled(event); goto unlock; + } /* * If the event is in a group and isn't the group leader, @@ -1307,6 +1733,7 @@ static void ctx_sched_out(struct perf_event_context *ctx, if (likely(!ctx->nr_events)) goto out; update_context_time(ctx); + update_cgrp_time_from_cpuctx(cpuctx); if (!ctx->nr_active) goto out; @@ -1496,6 +1923,14 @@ void __perf_event_task_sched_out(struct task_struct *task, for_each_task_context_nr(ctxn) perf_event_context_sched_out(task, ctxn, next); + + /* + * if cgroup events exist on this CPU, then we need + * to check if we have to switch out PMU state. + * cgroup event are system-wide mode only + */ + if (atomic_read(&__get_cpu_var(perf_cgroup_events))) + perf_cgroup_sched_out(task); } static void task_ctx_sched_out(struct perf_event_context *ctx, @@ -1534,6 +1969,10 @@ ctx_pinned_sched_in(struct perf_event_context *ctx, if (!event_filter_match(event)) continue; + /* may need to reset tstamp_enabled */ + if (is_cgroup_event(event)) + perf_cgroup_mark_enabled(event, ctx); + if (group_can_go_on(event, cpuctx, 1)) group_sched_in(event, cpuctx, ctx); @@ -1566,6 +2005,10 @@ ctx_flexible_sched_in(struct perf_event_context *ctx, if (!event_filter_match(event)) continue; + /* may need to reset tstamp_enabled */ + if (is_cgroup_event(event)) + perf_cgroup_mark_enabled(event, ctx); + if (group_can_go_on(event, cpuctx, can_add_hw)) { if (group_sched_in(event, cpuctx, ctx)) can_add_hw = 0; @@ -1576,15 +2019,19 @@ ctx_flexible_sched_in(struct perf_event_context *ctx, static void ctx_sched_in(struct perf_event_context *ctx, struct perf_cpu_context *cpuctx, - enum event_type_t event_type) + enum event_type_t event_type, + struct task_struct *task) { + u64 now; + raw_spin_lock(&ctx->lock); ctx->is_active = 1; if (likely(!ctx->nr_events)) goto out; - ctx->timestamp = perf_clock(); - + now = perf_clock(); + ctx->timestamp = now; + perf_cgroup_set_timestamp(task, now); /* * First go through the list and put on any pinned groups * in order to give them the best chance of going on. @@ -1601,11 +2048,12 @@ out: } static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx, - enum event_type_t event_type) + enum event_type_t event_type, + struct task_struct *task) { struct perf_event_context *ctx = &cpuctx->ctx; - ctx_sched_in(ctx, cpuctx, event_type); + ctx_sched_in(ctx, cpuctx, event_type, task); } static void task_ctx_sched_in(struct perf_event_context *ctx, @@ -1617,11 +2065,12 @@ static void task_ctx_sched_in(struct perf_event_context *ctx, if (cpuctx->task_ctx == ctx) return; - ctx_sched_in(ctx, cpuctx, event_type); + ctx_sched_in(ctx, cpuctx, event_type, NULL); cpuctx->task_ctx = ctx; } -static void perf_event_context_sched_in(struct perf_event_context *ctx) +static void perf_event_context_sched_in(struct perf_event_context *ctx, + struct task_struct *task) { struct perf_cpu_context *cpuctx; @@ -1637,9 +2086,9 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx) */ cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE); - ctx_sched_in(ctx, cpuctx, EVENT_PINNED); - cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE); - ctx_sched_in(ctx, cpuctx, EVENT_FLEXIBLE); + ctx_sched_in(ctx, cpuctx, EVENT_PINNED, task); + cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE, task); + ctx_sched_in(ctx, cpuctx, EVENT_FLEXIBLE, task); cpuctx->task_ctx = ctx; @@ -1672,8 +2121,15 @@ void __perf_event_task_sched_in(struct task_struct *task) if (likely(!ctx)) continue; - perf_event_context_sched_in(ctx); + perf_event_context_sched_in(ctx, task); } + /* + * if cgroup events exist on this CPU, then we need + * to check if we have to switch in PMU state. + * cgroup event are system-wide mode only + */ + if (atomic_read(&__get_cpu_var(perf_cgroup_events))) + perf_cgroup_sched_in(task); } static u64 perf_calculate_period(struct perf_event *event, u64 nsec, u64 count) @@ -1873,7 +2329,7 @@ static void perf_rotate_context(struct perf_cpu_context *cpuctx) if (ctx) rotate_ctx(ctx); - cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE); + cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE, current); if (ctx) task_ctx_sched_in(ctx, EVENT_FLEXIBLE); @@ -1952,7 +2408,7 @@ static void perf_event_enable_on_exec(struct perf_event_context *ctx) raw_spin_unlock(&ctx->lock); - perf_event_context_sched_in(ctx); + perf_event_context_sched_in(ctx, ctx->task); out: local_irq_restore(flags); } @@ -1977,8 +2433,10 @@ static void __perf_event_read(void *info) return; raw_spin_lock(&ctx->lock); - if (ctx->is_active) + if (ctx->is_active) { update_context_time(ctx); + update_cgrp_time_from_event(event); + } update_event_times(event); if (event->state == PERF_EVENT_STATE_ACTIVE) event->pmu->read(event); @@ -2009,8 +2467,10 @@ static u64 perf_event_read(struct perf_event *event) * (e.g., thread is blocked), in that case * we cannot update context time */ - if (ctx->is_active) + if (ctx->is_active) { update_context_time(ctx); + update_cgrp_time_from_event(event); + } update_event_times(event); raw_spin_unlock_irqrestore(&ctx->lock, flags); } @@ -2395,7 +2855,7 @@ static void free_event(struct perf_event *event) if (!event->parent) { if (event->attach_state & PERF_ATTACH_TASK) - jump_label_dec(&perf_task_events); + jump_label_dec(&perf_sched_events); if (event->attr.mmap || event->attr.mmap_data) atomic_dec(&nr_mmap_events); if (event->attr.comm) @@ -2411,6 +2871,9 @@ static void free_event(struct perf_event *event) event->buffer = NULL; } + if (is_cgroup_event(event)) + perf_detach_cgroup(event); + if (event->destroy) event->destroy(event); @@ -5300,6 +5763,7 @@ static void task_clock_event_read(struct perf_event *event) if (!in_nmi()) { update_context_time(event->ctx); + update_cgrp_time_from_event(event); time = event->ctx->time; } else { u64 now = perf_clock(); @@ -5725,7 +6189,7 @@ done: if (!event->parent) { if (event->attach_state & PERF_ATTACH_TASK) - jump_label_inc(&perf_task_events); + jump_label_inc(&perf_sched_events); if (event->attr.mmap || event->attr.mmap_data) atomic_inc(&nr_mmap_events); if (event->attr.comm) @@ -5900,7 +6364,7 @@ SYSCALL_DEFINE5(perf_event_open, int err; /* for future expandability... */ - if (flags & ~(PERF_FLAG_FD_NO_GROUP | PERF_FLAG_FD_OUTPUT)) + if (flags & ~PERF_FLAG_ALL) return -EINVAL; err = perf_copy_attr(attr_uptr, &attr); @@ -5917,6 +6381,15 @@ SYSCALL_DEFINE5(perf_event_open, return -EINVAL; } + /* + * In cgroup mode, the pid argument is used to pass the fd + * opened to the cgroup directory in cgroupfs. The cpu argument + * designates the cpu on which to monitor threads from that + * cgroup. + */ + if ((flags & PERF_FLAG_PID_CGROUP) && (pid == -1 || cpu == -1)) + return -EINVAL; + event_fd = get_unused_fd_flags(O_RDWR); if (event_fd < 0) return event_fd; @@ -5934,7 +6407,7 @@ SYSCALL_DEFINE5(perf_event_open, group_leader = NULL; } - if (pid != -1) { + if (pid != -1 && !(flags & PERF_FLAG_PID_CGROUP)) { task = find_lively_task_by_vpid(pid); if (IS_ERR(task)) { err = PTR_ERR(task); @@ -5948,6 +6421,12 @@ SYSCALL_DEFINE5(perf_event_open, goto err_task; } + if (flags & PERF_FLAG_PID_CGROUP) { + err = perf_cgroup_connect(pid, event, &attr, group_leader); + if (err) + goto err_alloc; + } + /* * Special case software events and allow them to be part of * any hardware group. @@ -6808,3 +7287,92 @@ unlock: return ret; } device_initcall(perf_event_sysfs_init); + +#ifdef CONFIG_CGROUP_PERF +static struct cgroup_subsys_state *perf_cgroup_create( + struct cgroup_subsys *ss, struct cgroup *cont) +{ + struct perf_cgroup *jc; + struct perf_cgroup_info *t; + int c; + + jc = kmalloc(sizeof(*jc), GFP_KERNEL); + if (!jc) + return ERR_PTR(-ENOMEM); + + memset(jc, 0, sizeof(*jc)); + + jc->info = alloc_percpu(struct perf_cgroup_info); + if (!jc->info) { + kfree(jc); + return ERR_PTR(-ENOMEM); + } + + for_each_possible_cpu(c) { + t = per_cpu_ptr(jc->info, c); + t->time = 0; + t->timestamp = 0; + } + return &jc->css; +} + +static void perf_cgroup_destroy(struct cgroup_subsys *ss, + struct cgroup *cont) +{ + struct perf_cgroup *jc; + jc = container_of(cgroup_subsys_state(cont, perf_subsys_id), + struct perf_cgroup, css); + free_percpu(jc->info); + kfree(jc); +} + +static int __perf_cgroup_move(void *info) +{ + struct task_struct *task = info; + perf_cgroup_switch(task, PERF_CGROUP_SWOUT | PERF_CGROUP_SWIN); + return 0; +} + +static void perf_cgroup_move(struct task_struct *task) +{ + task_function_call(task, __perf_cgroup_move, task); +} + +static void perf_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, + struct cgroup *old_cgrp, struct task_struct *task, + bool threadgroup) +{ + perf_cgroup_move(task); + if (threadgroup) { + struct task_struct *c; + rcu_read_lock(); + list_for_each_entry_rcu(c, &task->thread_group, thread_group) { + perf_cgroup_move(c); + } + rcu_read_unlock(); + } +} + +static void perf_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, + struct cgroup *old_cgrp, struct task_struct *task) +{ + /* + * cgroup_exit() is called in the copy_process() failure path. + * Ignore this case since the task hasn't ran yet, this avoids + * trying to poke a half freed task state from generic code. + */ + if (!(task->flags & PF_EXITING)) + return; + + perf_cgroup_move(task); +} + +struct cgroup_subsys perf_subsys = { + .name = "perf_event", + .subsys_id = perf_subsys_id, + .create = perf_cgroup_create, + .destroy = perf_cgroup_destroy, + .exit = perf_cgroup_exit, + .attach = perf_cgroup_attach, +}; +#endif /* CONFIG_CGROUP_PERF */ -- cgit v1.2.3 From 163ec4354a5135c6c38c3f4a9b46a31900ebdf48 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 16 Feb 2011 11:22:34 +0100 Subject: perf: Optimize throttling code By pre-computing the maximum number of samples per tick we can avoid a multiplication and a conditional since MAX_INTERRUPTS > max_samples_per_tick. Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 4 ++++ kernel/perf_event.c | 43 ++++++++++++++++++++++++------------------- kernel/sysctl.c | 2 +- 3 files changed, 29 insertions(+), 20 deletions(-) (limited to 'include/linux') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 38c8b2554842..8ceb5a6fd9c9 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -1110,6 +1110,10 @@ extern int sysctl_perf_event_paranoid; extern int sysctl_perf_event_mlock; extern int sysctl_perf_event_sample_rate; +extern int perf_proc_update_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, + loff_t *ppos); + static inline bool perf_paranoid_tracepoint_raw(void) { return sysctl_perf_event_paranoid > -1; diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 65dcdc76d709..e03be08d0ddf 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -150,7 +150,24 @@ int sysctl_perf_event_mlock __read_mostly = 512; /* 'free' kb per user */ /* * max perf event sample rate */ -int sysctl_perf_event_sample_rate __read_mostly = 100000; +#define DEFAULT_MAX_SAMPLE_RATE 100000 +int sysctl_perf_event_sample_rate __read_mostly = DEFAULT_MAX_SAMPLE_RATE; +static int max_samples_per_tick __read_mostly = + DIV_ROUND_UP(DEFAULT_MAX_SAMPLE_RATE, HZ); + +int perf_proc_update_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, + loff_t *ppos) +{ + int ret = proc_dointvec(table, write, buffer, lenp, ppos); + + if (ret || !write) + return ret; + + max_samples_per_tick = DIV_ROUND_UP(sysctl_perf_event_sample_rate, HZ); + + return 0; +} static atomic64_t perf_event_id; @@ -4941,26 +4958,14 @@ static int __perf_event_overflow(struct perf_event *event, int nmi, if (unlikely(!is_sampling_event(event))) return 0; - if (!throttle) { - hwc->interrupts++; - } else { - if (hwc->interrupts != MAX_INTERRUPTS) { - hwc->interrupts++; - if (HZ * hwc->interrupts > - (u64)sysctl_perf_event_sample_rate) { - hwc->interrupts = MAX_INTERRUPTS; - perf_log_throttle(event, 0); - ret = 1; - } - } else { - /* - * Keep re-disabling events even though on the previous - * pass we disabled it - just in case we raced with a - * sched-in and the event got enabled again: - */ + if (unlikely(hwc->interrupts >= max_samples_per_tick)) { + if (throttle) { + hwc->interrupts = MAX_INTERRUPTS; + perf_log_throttle(event, 0); ret = 1; } - } + } else + hwc->interrupts++; if (event->attr.freq) { u64 now = perf_clock(); diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 0f1bd83db985..daef911cbadb 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -948,7 +948,7 @@ static struct ctl_table kern_table[] = { .data = &sysctl_perf_event_sample_rate, .maxlen = sizeof(sysctl_perf_event_sample_rate), .mode = 0644, - .proc_handler = proc_dointvec, + .proc_handler = perf_proc_update_handler, }, #endif #ifdef CONFIG_KMEMCHECK -- cgit v1.2.3 From a7e3ed1e470116c9d12c2f778431a481a6be8ab6 Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Thu, 3 Mar 2011 10:34:47 +0800 Subject: perf: Add support for supplementary event registers Change logs against Andi's original version: - Extends perf_event_attr:config to config{,1,2} (Peter Zijlstra) - Fixed a major event scheduling issue. There cannot be a ref++ on an event that has already done ref++ once and without calling put_constraint() in between. (Stephane Eranian) - Use thread_cpumask for percore allocation. (Lin Ming) - Use MSR names in the extra reg lists. (Lin Ming) - Remove redundant "c = NULL" in intel_percore_constraints - Fix comment of perf_event_attr::config1 Intel Nehalem/Westmere have a special OFFCORE_RESPONSE event that can be used to monitor any offcore accesses from a core. This is a very useful event for various tunings, and it's also needed to implement the generic LLC-* events correctly. Unfortunately this event requires programming a mask in a separate register. And worse this separate register is per core, not per CPU thread. This patch: - Teaches perf_events that OFFCORE_RESPONSE needs extra parameters. The extra parameters are passed by user space in the perf_event_attr::config1 field. - Adds support to the Intel perf_event core to schedule per core resources. This adds fairly generic infrastructure that can be also used for other per core resources. The basic code has is patterned after the similar AMD northbridge constraints code. Thanks to Stephane Eranian who pointed out some problems in the original version and suggested improvements. Signed-off-by: Andi Kleen Signed-off-by: Lin Ming Signed-off-by: Peter Zijlstra LKML-Reference: <1299119690-13991-2-git-send-email-ming.m.lin@intel.com> Signed-off-by: Ingo Molnar --- arch/x86/include/asm/msr-index.h | 3 + arch/x86/kernel/cpu/perf_event.c | 64 +++++++++++ arch/x86/kernel/cpu/perf_event_intel.c | 198 +++++++++++++++++++++++++++++++++ include/linux/perf_event.h | 13 ++- 4 files changed, 276 insertions(+), 2 deletions(-) (limited to 'include/linux') diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h index 4d0dfa0d998e..d25e74cc1a50 100644 --- a/arch/x86/include/asm/msr-index.h +++ b/arch/x86/include/asm/msr-index.h @@ -47,6 +47,9 @@ #define MSR_IA32_MCG_STATUS 0x0000017a #define MSR_IA32_MCG_CTL 0x0000017b +#define MSR_OFFCORE_RSP_0 0x000001a6 +#define MSR_OFFCORE_RSP_1 0x000001a7 + #define MSR_IA32_PEBS_ENABLE 0x000003f1 #define MSR_IA32_DS_AREA 0x00000600 #define MSR_IA32_PERF_CAPABILITIES 0x00000345 diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index ea03c725e465..ec6a6db07332 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -93,6 +93,8 @@ struct amd_nb { struct event_constraint event_constraints[X86_PMC_IDX_MAX]; }; +struct intel_percore; + #define MAX_LBR_ENTRIES 16 struct cpu_hw_events { @@ -127,6 +129,13 @@ struct cpu_hw_events { struct perf_branch_stack lbr_stack; struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES]; + /* + * Intel percore register state. + * Coordinate shared resources between HT threads. + */ + int percore_used; /* Used by this CPU? */ + struct intel_percore *per_core; + /* * AMD specific bits */ @@ -177,6 +186,28 @@ struct cpu_hw_events { #define for_each_event_constraint(e, c) \ for ((e) = (c); (e)->weight; (e)++) +/* + * Extra registers for specific events. + * Some events need large masks and require external MSRs. + * Define a mapping to these extra registers. + */ +struct extra_reg { + unsigned int event; + unsigned int msr; + u64 config_mask; + u64 valid_mask; +}; + +#define EVENT_EXTRA_REG(e, ms, m, vm) { \ + .event = (e), \ + .msr = (ms), \ + .config_mask = (m), \ + .valid_mask = (vm), \ + } +#define INTEL_EVENT_EXTRA_REG(event, msr, vm) \ + EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm) +#define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0) + union perf_capabilities { struct { u64 lbr_format : 6; @@ -221,6 +252,7 @@ struct x86_pmu { void (*put_event_constraints)(struct cpu_hw_events *cpuc, struct perf_event *event); struct event_constraint *event_constraints; + struct event_constraint *percore_constraints; void (*quirks)(void); int perfctr_second_write; @@ -249,6 +281,11 @@ struct x86_pmu { */ unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */ int lbr_nr; /* hardware stack size */ + + /* + * Extra registers for events + */ + struct extra_reg *extra_regs; }; static struct x86_pmu x86_pmu __read_mostly; @@ -341,6 +378,31 @@ static inline unsigned int x86_pmu_event_addr(int index) return x86_pmu.perfctr + x86_pmu_addr_offset(index); } +/* + * Find and validate any extra registers to set up. + */ +static int x86_pmu_extra_regs(u64 config, struct perf_event *event) +{ + struct extra_reg *er; + + event->hw.extra_reg = 0; + event->hw.extra_config = 0; + + if (!x86_pmu.extra_regs) + return 0; + + for (er = x86_pmu.extra_regs; er->msr; er++) { + if (er->event != (config & er->config_mask)) + continue; + if (event->attr.config1 & ~er->valid_mask) + return -EINVAL; + event->hw.extra_reg = er->msr; + event->hw.extra_config = event->attr.config1; + break; + } + return 0; +} + static atomic_t active_events; static DEFINE_MUTEX(pmc_reserve_mutex); @@ -665,6 +727,8 @@ static void x86_pmu_disable(struct pmu *pmu) static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc, u64 enable_mask) { + if (hwc->extra_reg) + wrmsrl(hwc->extra_reg, hwc->extra_config); wrmsrl(hwc->config_base, hwc->config | enable_mask); } diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c index c3ce053ecb46..13cb6cf013f6 100644 --- a/arch/x86/kernel/cpu/perf_event_intel.c +++ b/arch/x86/kernel/cpu/perf_event_intel.c @@ -1,5 +1,27 @@ #ifdef CONFIG_CPU_SUP_INTEL +#define MAX_EXTRA_REGS 2 + +/* + * Per register state. + */ +struct er_account { + int ref; /* reference count */ + unsigned int extra_reg; /* extra MSR number */ + u64 extra_config; /* extra MSR config */ +}; + +/* + * Per core state + * This used to coordinate shared registers for HT threads. + */ +struct intel_percore { + raw_spinlock_t lock; /* protect structure */ + struct er_account regs[MAX_EXTRA_REGS]; + int refcnt; /* number of threads */ + unsigned core_id; +}; + /* * Intel PerfMon, used on Core and later. */ @@ -64,6 +86,18 @@ static struct event_constraint intel_nehalem_event_constraints[] = EVENT_CONSTRAINT_END }; +static struct extra_reg intel_nehalem_extra_regs[] = +{ + INTEL_EVENT_EXTRA_REG(0xb7, MSR_OFFCORE_RSP_0, 0xffff), + EVENT_EXTRA_END +}; + +static struct event_constraint intel_nehalem_percore_constraints[] = +{ + INTEL_EVENT_CONSTRAINT(0xb7, 0), + EVENT_CONSTRAINT_END +}; + static struct event_constraint intel_westmere_event_constraints[] = { FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */ @@ -89,6 +123,20 @@ static struct event_constraint intel_snb_event_constraints[] = EVENT_CONSTRAINT_END }; +static struct extra_reg intel_westmere_extra_regs[] = +{ + INTEL_EVENT_EXTRA_REG(0xb7, MSR_OFFCORE_RSP_0, 0xffff), + INTEL_EVENT_EXTRA_REG(0xbb, MSR_OFFCORE_RSP_1, 0xffff), + EVENT_EXTRA_END +}; + +static struct event_constraint intel_westmere_percore_constraints[] = +{ + INTEL_EVENT_CONSTRAINT(0xb7, 0), + INTEL_EVENT_CONSTRAINT(0xbb, 0), + EVENT_CONSTRAINT_END +}; + static struct event_constraint intel_gen_event_constraints[] = { FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */ @@ -906,6 +954,67 @@ intel_bts_constraints(struct perf_event *event) return NULL; } +static struct event_constraint * +intel_percore_constraints(struct cpu_hw_events *cpuc, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + unsigned int e = hwc->config & ARCH_PERFMON_EVENTSEL_EVENT; + struct event_constraint *c; + struct intel_percore *pc; + struct er_account *era; + int i; + int free_slot; + int found; + + if (!x86_pmu.percore_constraints || hwc->extra_alloc) + return NULL; + + for (c = x86_pmu.percore_constraints; c->cmask; c++) { + if (e != c->code) + continue; + + /* + * Allocate resource per core. + */ + pc = cpuc->per_core; + if (!pc) + break; + c = &emptyconstraint; + raw_spin_lock(&pc->lock); + free_slot = -1; + found = 0; + for (i = 0; i < MAX_EXTRA_REGS; i++) { + era = &pc->regs[i]; + if (era->ref > 0 && hwc->extra_reg == era->extra_reg) { + /* Allow sharing same config */ + if (hwc->extra_config == era->extra_config) { + era->ref++; + cpuc->percore_used = 1; + hwc->extra_alloc = 1; + c = NULL; + } + /* else conflict */ + found = 1; + break; + } else if (era->ref == 0 && free_slot == -1) + free_slot = i; + } + if (!found && free_slot != -1) { + era = &pc->regs[free_slot]; + era->ref = 1; + era->extra_reg = hwc->extra_reg; + era->extra_config = hwc->extra_config; + cpuc->percore_used = 1; + hwc->extra_alloc = 1; + c = NULL; + } + raw_spin_unlock(&pc->lock); + return c; + } + + return NULL; +} + static struct event_constraint * intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event) { @@ -919,9 +1028,51 @@ intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event if (c) return c; + c = intel_percore_constraints(cpuc, event); + if (c) + return c; + return x86_get_event_constraints(cpuc, event); } +static void intel_put_event_constraints(struct cpu_hw_events *cpuc, + struct perf_event *event) +{ + struct extra_reg *er; + struct intel_percore *pc; + struct er_account *era; + struct hw_perf_event *hwc = &event->hw; + int i, allref; + + if (!cpuc->percore_used) + return; + + for (er = x86_pmu.extra_regs; er->msr; er++) { + if (er->event != (hwc->config & er->config_mask)) + continue; + + pc = cpuc->per_core; + raw_spin_lock(&pc->lock); + for (i = 0; i < MAX_EXTRA_REGS; i++) { + era = &pc->regs[i]; + if (era->ref > 0 && + era->extra_config == hwc->extra_config && + era->extra_reg == er->msr) { + era->ref--; + hwc->extra_alloc = 0; + break; + } + } + allref = 0; + for (i = 0; i < MAX_EXTRA_REGS; i++) + allref += pc->regs[i].ref; + if (allref == 0) + cpuc->percore_used = 0; + raw_spin_unlock(&pc->lock); + break; + } +} + static int intel_pmu_hw_config(struct perf_event *event) { int ret = x86_pmu_hw_config(event); @@ -993,11 +1144,43 @@ static __initconst const struct x86_pmu core_pmu = { */ .max_period = (1ULL << 31) - 1, .get_event_constraints = intel_get_event_constraints, + .put_event_constraints = intel_put_event_constraints, .event_constraints = intel_core_event_constraints, }; +static int intel_pmu_cpu_prepare(int cpu) +{ + struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu); + + cpuc->per_core = kzalloc_node(sizeof(struct intel_percore), + GFP_KERNEL, cpu_to_node(cpu)); + if (!cpuc->per_core) + return NOTIFY_BAD; + + raw_spin_lock_init(&cpuc->per_core->lock); + cpuc->per_core->core_id = -1; + return NOTIFY_OK; +} + static void intel_pmu_cpu_starting(int cpu) { + struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu); + int core_id = topology_core_id(cpu); + int i; + + for_each_cpu(i, topology_thread_cpumask(cpu)) { + struct intel_percore *pc = per_cpu(cpu_hw_events, i).per_core; + + if (pc && pc->core_id == core_id) { + kfree(cpuc->per_core); + cpuc->per_core = pc; + break; + } + } + + cpuc->per_core->core_id = core_id; + cpuc->per_core->refcnt++; + init_debug_store_on_cpu(cpu); /* * Deal with CPUs that don't clear their LBRs on power-up. @@ -1007,6 +1190,15 @@ static void intel_pmu_cpu_starting(int cpu) static void intel_pmu_cpu_dying(int cpu) { + struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu); + struct intel_percore *pc = cpuc->per_core; + + if (pc) { + if (pc->core_id == -1 || --pc->refcnt == 0) + kfree(pc); + cpuc->per_core = NULL; + } + fini_debug_store_on_cpu(cpu); } @@ -1031,7 +1223,9 @@ static __initconst const struct x86_pmu intel_pmu = { */ .max_period = (1ULL << 31) - 1, .get_event_constraints = intel_get_event_constraints, + .put_event_constraints = intel_put_event_constraints, + .cpu_prepare = intel_pmu_cpu_prepare, .cpu_starting = intel_pmu_cpu_starting, .cpu_dying = intel_pmu_cpu_dying, }; @@ -1151,7 +1345,9 @@ static __init int intel_pmu_init(void) x86_pmu.event_constraints = intel_nehalem_event_constraints; x86_pmu.pebs_constraints = intel_nehalem_pebs_event_constraints; + x86_pmu.percore_constraints = intel_nehalem_percore_constraints; x86_pmu.enable_all = intel_pmu_nhm_enable_all; + x86_pmu.extra_regs = intel_nehalem_extra_regs; pr_cont("Nehalem events, "); break; @@ -1174,8 +1370,10 @@ static __init int intel_pmu_init(void) intel_pmu_lbr_init_nhm(); x86_pmu.event_constraints = intel_westmere_event_constraints; + x86_pmu.percore_constraints = intel_westmere_percore_constraints; x86_pmu.enable_all = intel_pmu_nhm_enable_all; x86_pmu.pebs_constraints = intel_westmere_pebs_event_constraints; + x86_pmu.extra_regs = intel_westmere_extra_regs; pr_cont("Westmere events, "); break; diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 8ceb5a6fd9c9..614615b8d42b 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -225,8 +225,14 @@ struct perf_event_attr { }; __u32 bp_type; - __u64 bp_addr; - __u64 bp_len; + union { + __u64 bp_addr; + __u64 config1; /* extension of config */ + }; + union { + __u64 bp_len; + __u64 config2; /* extension of config1 */ + }; }; /* @@ -541,6 +547,9 @@ struct hw_perf_event { unsigned long event_base; int idx; int last_cpu; + unsigned int extra_reg; + u64 extra_config; + int extra_alloc; }; struct { /* software */ struct hrtimer hrtimer; -- cgit v1.2.3 From 750912fa366312e9c5bc83eab352898a26750401 Mon Sep 17 00:00:00 2001 From: David Sharp Date: Wed, 8 Dec 2010 13:46:47 -0800 Subject: tracing: Add an 'overwrite' trace_option. Add an "overwrite" trace_option for ftrace to control whether the buffer should be overwritten on overflow or not. The default remains to overwrite old events when the buffer is full. This patch adds the option to instead discard newest events when the buffer is full. This is useful to get a snapshot of traces just after enabling traces. Dropping the current event is also a simpler code path. Signed-off-by: David Sharp LKML-Reference: <1291844807-15481-1-git-send-email-dhsharp@google.com> Signed-off-by: Steven Rostedt --- Documentation/trace/ftrace.txt | 5 +++++ include/linux/ring_buffer.h | 2 ++ kernel/trace/ring_buffer.c | 11 +++++++++++ kernel/trace/trace.c | 17 +++++++++++------ kernel/trace/trace.h | 1 + 5 files changed, 30 insertions(+), 6 deletions(-) (limited to 'include/linux') diff --git a/Documentation/trace/ftrace.txt b/Documentation/trace/ftrace.txt index 67f1cc473257..1ebc24cf9a55 100644 --- a/Documentation/trace/ftrace.txt +++ b/Documentation/trace/ftrace.txt @@ -454,6 +454,11 @@ x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] latencies, as described in "Latency trace format". + overwrite - This controls what happens when the trace buffer is + full. If "1" (default), the oldest events are + discarded and overwritten. If "0", then the newest + events are discarded. + ftrace_enabled -------------- diff --git a/include/linux/ring_buffer.h b/include/linux/ring_buffer.h index 8d3a2486544d..ab38ac80b0f9 100644 --- a/include/linux/ring_buffer.h +++ b/include/linux/ring_buffer.h @@ -100,6 +100,8 @@ void ring_buffer_free(struct ring_buffer *buffer); int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size); +void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val); + struct ring_buffer_event *ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length); int ring_buffer_unlock_commit(struct ring_buffer *buffer, diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index bd1c35a4fbcc..269db80a961e 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1429,6 +1429,17 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) } EXPORT_SYMBOL_GPL(ring_buffer_resize); +void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) +{ + mutex_lock(&buffer->mutex); + if (val) + buffer->flags |= RB_FL_OVERWRITE; + else + buffer->flags &= ~RB_FL_OVERWRITE; + mutex_unlock(&buffer->mutex); +} +EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); + static inline void * __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) { diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 8dc8da6733f9..85e3ee1e474e 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -41,8 +41,6 @@ #include "trace.h" #include "trace_output.h" -#define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE) - /* * On boot up, the ring buffer is set to the minimum size, so that * we do not waste memory on systems that are not using tracing. @@ -340,7 +338,7 @@ static DECLARE_WAIT_QUEUE_HEAD(trace_wait); /* trace_flags holds trace_options default values */ unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME | - TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD; + TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE; static int trace_stop_count; static DEFINE_SPINLOCK(tracing_start_lock); @@ -425,6 +423,7 @@ static const char *trace_options[] = { "sleep-time", "graph-time", "record-cmd", + "overwrite", NULL }; @@ -2529,6 +2528,9 @@ static void set_tracer_flags(unsigned int mask, int enabled) if (mask == TRACE_ITER_RECORD_CMD) trace_event_enable_cmd_record(enabled); + + if (mask == TRACE_ITER_OVERWRITE) + ring_buffer_change_overwrite(global_trace.buffer, enabled); } static ssize_t @@ -4555,9 +4557,11 @@ void ftrace_dump(enum ftrace_dump_mode oops_dump_mode) __init static int tracer_alloc_buffers(void) { int ring_buf_size; + enum ring_buffer_flags rb_flags; int i; int ret = -ENOMEM; + if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL)) goto out; @@ -4570,12 +4574,13 @@ __init static int tracer_alloc_buffers(void) else ring_buf_size = 1; + rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0; + cpumask_copy(tracing_buffer_mask, cpu_possible_mask); cpumask_copy(tracing_cpumask, cpu_all_mask); /* TODO: make the number of buffers hot pluggable with CPUS */ - global_trace.buffer = ring_buffer_alloc(ring_buf_size, - TRACE_BUFFER_FLAGS); + global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags); if (!global_trace.buffer) { printk(KERN_ERR "tracer: failed to allocate ring buffer!\n"); WARN_ON(1); @@ -4585,7 +4590,7 @@ __init static int tracer_alloc_buffers(void) #ifdef CONFIG_TRACER_MAX_TRACE - max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS); + max_tr.buffer = ring_buffer_alloc(1, rb_flags); if (!max_tr.buffer) { printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n"); WARN_ON(1); diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index 856e73cc1d3f..951d0b7e7062 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -606,6 +606,7 @@ enum trace_iterator_flags { TRACE_ITER_SLEEP_TIME = 0x40000, TRACE_ITER_GRAPH_TIME = 0x80000, TRACE_ITER_RECORD_CMD = 0x100000, + TRACE_ITER_OVERWRITE = 0x200000, }; /* -- cgit v1.2.3 From e6e1e2593592a8f6f6380496655d8c6f67431266 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Wed, 9 Mar 2011 10:41:56 -0500 Subject: tracing: Remove lock_depth from event entry The lock_depth field in the event headers was added as a temporary data point for help in removing the BKL. Now that the BKL is pretty much been removed, we can remove this field. This in turn changes the header from 12 bytes to 8 bytes, removing the 4 byte buffer that gcc would insert if the first field in the data load was 8 bytes in size. Signed-off-by: Steven Rostedt --- include/linux/ftrace_event.h | 1 - kernel/trace/trace.c | 8 +++----- kernel/trace/trace_events.c | 1 - kernel/trace/trace_output.c | 10 ++-------- 4 files changed, 5 insertions(+), 15 deletions(-) (limited to 'include/linux') diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index 1a99e7939c2b..22b32af1b5ec 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -37,7 +37,6 @@ struct trace_entry { unsigned char flags; unsigned char preempt_count; int pid; - int lock_depth; }; #define FTRACE_MAX_EVENT \ diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 85e3ee1e474e..fd6e1b906b3c 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -1101,7 +1101,6 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, entry->preempt_count = pc & 0xff; entry->pid = (tsk) ? tsk->pid : 0; - entry->lock_depth = (tsk) ? tsk->lock_depth : 0; entry->flags = #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) | @@ -1748,10 +1747,9 @@ static void print_lat_help_header(struct seq_file *m) seq_puts(m, "# | / _----=> need-resched \n"); seq_puts(m, "# || / _---=> hardirq/softirq \n"); seq_puts(m, "# ||| / _--=> preempt-depth \n"); - seq_puts(m, "# |||| /_--=> lock-depth \n"); - seq_puts(m, "# |||||/ delay \n"); - seq_puts(m, "# cmd pid |||||| time | caller \n"); - seq_puts(m, "# \\ / |||||| \\ | / \n"); + seq_puts(m, "# |||| / delay \n"); + seq_puts(m, "# cmd pid ||||| time | caller \n"); + seq_puts(m, "# \\ / ||||| \\ | / \n"); } static void print_func_help_header(struct seq_file *m) diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c index 5f499e0438a4..e1d579b19834 100644 --- a/kernel/trace/trace_events.c +++ b/kernel/trace/trace_events.c @@ -116,7 +116,6 @@ static int trace_define_common_fields(void) __common_field(unsigned char, flags); __common_field(unsigned char, preempt_count); __common_field(int, pid); - __common_field(int, lock_depth); return ret; } diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 02272baa2206..151f32e5f2c6 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -529,7 +529,7 @@ seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags) * @entry: The trace entry field from the ring buffer * * Prints the generic fields of irqs off, in hard or softirq, preempt - * count and lock depth. + * count. */ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry) { @@ -554,13 +554,7 @@ int trace_print_lat_fmt(struct trace_seq *s, struct trace_entry *entry) else ret = trace_seq_putc(s, '.'); - if (!ret) - return 0; - - if (entry->lock_depth < 0) - return trace_seq_putc(s, '.'); - - return trace_seq_printf(s, "%d", entry->lock_depth); + return ret; } static int -- cgit v1.2.3