From 86e7972f690c1017fd086cdfe53d8524e68c661c Mon Sep 17 00:00:00 2001 From: Wang Nan Date: Mon, 28 Mar 2016 06:41:29 +0000 Subject: perf/ring_buffer: Introduce new ioctl options to pause and resume the ring-buffer Add new ioctl() to pause/resume ring-buffer output. In some situations we want to read from the ring-buffer only when we ensure nothing can write to the ring-buffer during reading. Without this patch we have to turn off all events attached to this ring-buffer to achieve this. This patch is a prerequisite to enable overwrite support for the perf ring-buffer support. Following commits will introduce new methods support reading from overwrite ring buffer. Before reading, caller must ensure the ring buffer is frozen, or the reading is unreliable. Signed-off-by: Wang Nan Signed-off-by: Peter Zijlstra (Intel) Cc: Cc: Alexander Shishkin Cc: Alexei Starovoitov Cc: Arnaldo Carvalho de Melo Cc: Brendan Gregg Cc: He Kuang Cc: Jiri Olsa Cc: Jiri Olsa Cc: Linus Torvalds Cc: Masami Hiramatsu Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Vince Weaver Cc: Zefan Li Link: http://lkml.kernel.org/r/1459147292-239310-2-git-send-email-wangnan0@huawei.com Signed-off-by: Ingo Molnar --- include/uapi/linux/perf_event.h | 1 + 1 file changed, 1 insertion(+) (limited to 'include') diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h index 1afe9623c1a7..a3c19034d5f8 100644 --- a/include/uapi/linux/perf_event.h +++ b/include/uapi/linux/perf_event.h @@ -401,6 +401,7 @@ struct perf_event_attr { #define PERF_EVENT_IOC_SET_FILTER _IOW('$', 6, char *) #define PERF_EVENT_IOC_ID _IOR('$', 7, __u64 *) #define PERF_EVENT_IOC_SET_BPF _IOW('$', 8, __u32) +#define PERF_EVENT_IOC_PAUSE_OUTPUT _IOW('$', 9, __u32) enum perf_event_ioc_flags { PERF_IOC_FLAG_GROUP = 1U << 0, -- cgit v1.2.3 From 1879445dfa7bbd6fe21b09c5cc72f4934798afed Mon Sep 17 00:00:00 2001 From: Wang Nan Date: Mon, 28 Mar 2016 06:41:30 +0000 Subject: perf/core: Set event's default ::overflow_handler() Set a default event->overflow_handler in perf_event_alloc() so don't need to check event->overflow_handler in __perf_event_overflow(). Following commits can give a different default overflow_handler. Initial idea comes from Peter: http://lkml.kernel.org/r/20130708121557.GA17211@twins.programming.kicks-ass.net Since the default value of event->overflow_handler is not NULL, existing 'if (!overflow_handler)' checks need to be changed. is_default_overflow_handler() is introduced for this. No extra performance overhead is introduced into the hot path because in the original code we still need to read this handler from memory. A conditional branch is avoided so actually we remove some instructions. Signed-off-by: Wang Nan Signed-off-by: Peter Zijlstra (Intel) Cc: Cc: Alexander Shishkin Cc: Alexei Starovoitov Cc: Arnaldo Carvalho de Melo Cc: Brendan Gregg Cc: He Kuang Cc: Jiri Olsa Cc: Jiri Olsa Cc: Linus Torvalds Cc: Masami Hiramatsu Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Vince Weaver Cc: Zefan Li Link: http://lkml.kernel.org/r/1459147292-239310-3-git-send-email-wangnan0@huawei.com Signed-off-by: Ingo Molnar --- arch/arm/kernel/hw_breakpoint.c | 4 ++-- arch/arm64/kernel/hw_breakpoint.c | 4 ++-- include/linux/perf_event.h | 6 ++++++ kernel/events/core.c | 14 ++++++++------ 4 files changed, 18 insertions(+), 10 deletions(-) (limited to 'include') diff --git a/arch/arm/kernel/hw_breakpoint.c b/arch/arm/kernel/hw_breakpoint.c index 6284779d64ee..b8df45883cf7 100644 --- a/arch/arm/kernel/hw_breakpoint.c +++ b/arch/arm/kernel/hw_breakpoint.c @@ -631,7 +631,7 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp) info->address &= ~alignment_mask; info->ctrl.len <<= offset; - if (!bp->overflow_handler) { + if (is_default_overflow_handler(bp)) { /* * Mismatch breakpoints are required for single-stepping * breakpoints. @@ -754,7 +754,7 @@ static void watchpoint_handler(unsigned long addr, unsigned int fsr, * mismatch breakpoint so we can single-step over the * watchpoint trigger. */ - if (!wp->overflow_handler) + if (is_default_overflow_handler(wp)) enable_single_step(wp, instruction_pointer(regs)); unlock: diff --git a/arch/arm64/kernel/hw_breakpoint.c b/arch/arm64/kernel/hw_breakpoint.c index b45c95d34b83..4ef5373f9a76 100644 --- a/arch/arm64/kernel/hw_breakpoint.c +++ b/arch/arm64/kernel/hw_breakpoint.c @@ -616,7 +616,7 @@ static int breakpoint_handler(unsigned long unused, unsigned int esr, perf_bp_event(bp, regs); /* Do we need to handle the stepping? */ - if (!bp->overflow_handler) + if (is_default_overflow_handler(bp)) step = 1; unlock: rcu_read_unlock(); @@ -712,7 +712,7 @@ static int watchpoint_handler(unsigned long addr, unsigned int esr, perf_bp_event(wp, regs); /* Do we need to handle the stepping? */ - if (!wp->overflow_handler) + if (is_default_overflow_handler(wp)) step = 1; unlock: diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 15588d4c581d..4065ca2d7149 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -838,6 +838,12 @@ extern void perf_event_output(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs); +static inline bool +is_default_overflow_handler(struct perf_event *event) +{ + return (event->overflow_handler == perf_event_output); +} + extern void perf_event_header__init_id(struct perf_event_header *header, struct perf_sample_data *data, diff --git a/kernel/events/core.c b/kernel/events/core.c index 51386e84293e..8c3b35f2a269 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -6628,10 +6628,7 @@ static int __perf_event_overflow(struct perf_event *event, irq_work_queue(&event->pending); } - if (event->overflow_handler) - event->overflow_handler(event, data, regs); - else - perf_event_output(event, data, regs); + event->overflow_handler(event, data, regs); if (*perf_event_fasync(event) && event->pending_kill) { event->pending_wakeup = 1; @@ -8152,8 +8149,13 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, context = parent_event->overflow_handler_context; } - event->overflow_handler = overflow_handler; - event->overflow_handler_context = context; + if (overflow_handler) { + event->overflow_handler = overflow_handler; + event->overflow_handler_context = context; + } else { + event->overflow_handler = perf_event_output; + event->overflow_handler_context = NULL; + } perf_event__state_init(event); -- cgit v1.2.3 From 9ecda41acb971ebd07c8fb35faf24005c0baea12 Mon Sep 17 00:00:00 2001 From: Wang Nan Date: Tue, 5 Apr 2016 14:11:18 +0000 Subject: perf/core: Add ::write_backward attribute to perf event This patch introduces 'write_backward' bit to perf_event_attr, which controls the direction of a ring buffer. After set, the corresponding ring buffer is written from end to beginning. This feature is design to support reading from overwritable ring buffer. Ring buffer can be created by mapping a perf event fd. Kernel puts event records into ring buffer, user tooling like perf fetch them from address returned by mmap(). To prevent racing between kernel and tooling, they communicate to each other through 'head' and 'tail' pointers. Kernel maintains 'head' pointer, points it to the next free area (tail of the last record). Tooling maintains 'tail' pointer, points it to the tail of last consumed record (record has already been fetched). Kernel determines the available space in a ring buffer using these two pointers to avoid overwrite unfetched records. By mapping without 'PROT_WRITE', an overwritable ring buffer is created. Different from normal ring buffer, tooling is unable to maintain 'tail' pointer because writing is forbidden. Therefore, for this type of ring buffers, kernel overwrite old records unconditionally, works like flight recorder. This feature would be useful if reading from overwritable ring buffer were as easy as reading from normal ring buffer. However, there's an obscure problem. The following figure demonstrates a full overwritable ring buffer. In this figure, the 'head' pointer points to the end of last record, and a long record 'E' is pending. For a normal ring buffer, a 'tail' pointer would have pointed to position (X), so kernel knows there's no more space in the ring buffer. However, for an overwritable ring buffer, kernel ignore the 'tail' pointer. (X) head . | . V +------+-------+----------+------+---+ |A....A|B.....B|C........C|D....D| | +------+-------+----------+------+---+ Record 'A' is overwritten by event 'E': head | V +--+---+-------+----------+------+---+ |.E|..A|B.....B|C........C|D....D|E..| +--+---+-------+----------+------+---+ Now tooling decides to read from this ring buffer. However, none of these two natural positions, 'head' and the start of this ring buffer, are pointing to the head of a record. Even the full ring buffer can be accessed by tooling, it is unable to find a position to start decoding. The first attempt tries to solve this problem AFAIK can be found from [1]. It makes kernel to maintain 'tail' pointer: updates it when ring buffer is half full. However, this approach introduces overhead to fast path. Test result shows a 1% overhead [2]. In addition, this method utilizes no more tham 50% records. Another attempt can be found from [3], which allows putting the size of an event at the end of each record. This approach allows tooling to find records in a backward manner from 'head' pointer by reading size of a record from its tail. However, because of alignment requirement, it needs 8 bytes to record the size of a record, which is a huge waste. Its performance is also not good, because more data need to be written. This approach also introduces some extra branch instructions to fast path. 'write_backward' is a better solution to this problem. Following figure demonstrates the state of the overwritable ring buffer when 'write_backward' is set before overwriting: head | V +---+------+----------+-------+------+ | |D....D|C........C|B.....B|A....A| +---+------+----------+-------+------+ and after overwriting: head | V +---+------+----------+-------+---+--+ |..E|D....D|C........C|B.....B|A..|E.| +---+------+----------+-------+---+--+ In each situation, 'head' points to the beginning of the newest record. From this record, tooling can iterate over the full ring buffer and fetch records one by one. The only limitation that needs to be considered is back-to-back reading. Due to the non-deterministic of user programs, it is impossible to ensure the ring buffer keeps stable during reading. Consider an extreme situation: tooling is scheduled out after reading record 'D', then a burst of events come, eat up the whole ring buffer (one or multiple rounds). When the tooling process comes back, reading after 'D' is incorrect now. To prevent this problem, we need to find a way to ensure the ring buffer is stable during reading. ioctl(PERF_EVENT_IOC_PAUSE_OUTPUT) is suggested because its overhead is lower than ioctl(PERF_EVENT_IOC_ENABLE). By carefully verifying 'header' pointer, reader can avoid pausing the ring-buffer. For example: /* A union of all possible events */ union perf_event event; p = head = perf_mmap__read_head(); while (true) { /* copy header of next event */ fetch(&event.header, p, sizeof(event.header)); /* read 'head' pointer */ head = perf_mmap__read_head(); /* check overwritten: is the header good? */ if (!verify(sizeof(event.header), p, head)) break; /* copy the whole event */ fetch(&event, p, event.header.size); /* read 'head' pointer again */ head = perf_mmap__read_head(); /* is the whole event good? */ if (!verify(event.header.size, p, head)) break; p += event.header.size; } However, the overhead is high because: a) In-place decoding is not safe. Copying-verifying-decoding is required. b) Fetching 'head' pointer requires additional synchronization. (From Alexei Starovoitov: Even when this trick works, pause is needed for more than stability of reading. When we collect the events into overwrite buffer we're waiting for some other trigger (like all cpu utilization spike or just one cpu running and all others are idle) and when it happens the buffer has valuable info from the past. At this point new events are no longer interesting and buffer should be paused, events read and unpaused until next trigger comes.) This patch utilizes event's default overflow_handler introduced previously. perf_event_output_backward() is created as the default overflow handler for backward ring buffers. To avoid extra overhead to fast path, original perf_event_output() becomes __perf_event_output() and marked '__always_inline'. In theory, there's no extra overhead introduced to fast path. Performance testing: Calling 3000000 times of 'close(-1)', use gettimeofday() to check duration. Use 'perf record -o /dev/null -e raw_syscalls:*' to capture system calls. In ns. Testing environment: CPU : Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz Kernel : v4.5.0 MEAN STDVAR BASE 800214.950 2853.083 PRE1 2253846.700 9997.014 PRE2 2257495.540 8516.293 POST 2250896.100 8933.921 Where 'BASE' is pure performance without capturing. 'PRE1' is test result of pure 'v4.5.0' kernel. 'PRE2' is test result before this patch. 'POST' is test result after this patch. See [4] for the detailed experimental setup. Considering the stdvar, this patch doesn't introduce performance overhead to the fast path. [1] http://lkml.iu.edu/hypermail/linux/kernel/1304.1/04584.html [2] http://lkml.iu.edu/hypermail/linux/kernel/1307.1/00535.html [3] http://lkml.iu.edu/hypermail/linux/kernel/1512.0/01265.html [4] http://lkml.kernel.org/g/56F89DCD.1040202@huawei.com Signed-off-by: Wang Nan Signed-off-by: Peter Zijlstra (Intel) Acked-by: Alexei Starovoitov Cc: Cc: Cc: Alexander Shishkin Cc: Arnaldo Carvalho de Melo Cc: Brendan Gregg Cc: He Kuang Cc: Jiri Olsa Cc: Jiri Olsa Cc: Linus Torvalds Cc: Masami Hiramatsu Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Vince Weaver Cc: Zefan Li Link: http://lkml.kernel.org/r/1459865478-53413-1-git-send-email-wangnan0@huawei.com [ Fixed the changelog some more. ] Signed-off-by: Ingo Molnar Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 28 +++++++++++++++++++++--- include/uapi/linux/perf_event.h | 3 ++- kernel/events/core.c | 48 ++++++++++++++++++++++++++++++++++++----- kernel/events/ring_buffer.c | 16 +++++++++++++- 4 files changed, 85 insertions(+), 10 deletions(-) (limited to 'include') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index b8b195fbe787..85749ae8cb5f 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -834,14 +834,24 @@ extern int perf_event_overflow(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs); +extern void perf_event_output_forward(struct perf_event *event, + struct perf_sample_data *data, + struct pt_regs *regs); +extern void perf_event_output_backward(struct perf_event *event, + struct perf_sample_data *data, + struct pt_regs *regs); extern void perf_event_output(struct perf_event *event, - struct perf_sample_data *data, - struct pt_regs *regs); + struct perf_sample_data *data, + struct pt_regs *regs); static inline bool is_default_overflow_handler(struct perf_event *event) { - return (event->overflow_handler == perf_event_output); + if (likely(event->overflow_handler == perf_event_output_forward)) + return true; + if (unlikely(event->overflow_handler == perf_event_output_backward)) + return true; + return false; } extern void @@ -1051,8 +1061,20 @@ static inline bool has_aux(struct perf_event *event) return event->pmu->setup_aux; } +static inline bool is_write_backward(struct perf_event *event) +{ + return !!event->attr.write_backward; +} + extern int perf_output_begin(struct perf_output_handle *handle, struct perf_event *event, unsigned int size); +extern int perf_output_begin_forward(struct perf_output_handle *handle, + struct perf_event *event, + unsigned int size); +extern int perf_output_begin_backward(struct perf_output_handle *handle, + struct perf_event *event, + unsigned int size); + extern void perf_output_end(struct perf_output_handle *handle); extern unsigned int perf_output_copy(struct perf_output_handle *handle, const void *buf, unsigned int len); diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h index a3c19034d5f8..43fc8d213472 100644 --- a/include/uapi/linux/perf_event.h +++ b/include/uapi/linux/perf_event.h @@ -340,7 +340,8 @@ struct perf_event_attr { comm_exec : 1, /* flag comm events that are due to an exec */ use_clockid : 1, /* use @clockid for time fields */ context_switch : 1, /* context switch data */ - __reserved_1 : 37; + write_backward : 1, /* Write ring buffer from end to beginning */ + __reserved_1 : 36; union { __u32 wakeup_events; /* wakeup every n events */ diff --git a/kernel/events/core.c b/kernel/events/core.c index 21ba024c9ed1..eabeb2aec00f 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -5694,9 +5694,13 @@ void perf_prepare_sample(struct perf_event_header *header, } } -void perf_event_output(struct perf_event *event, - struct perf_sample_data *data, - struct pt_regs *regs) +static void __always_inline +__perf_event_output(struct perf_event *event, + struct perf_sample_data *data, + struct pt_regs *regs, + int (*output_begin)(struct perf_output_handle *, + struct perf_event *, + unsigned int)) { struct perf_output_handle handle; struct perf_event_header header; @@ -5706,7 +5710,7 @@ void perf_event_output(struct perf_event *event, perf_prepare_sample(&header, data, event, regs); - if (perf_output_begin(&handle, event, header.size)) + if (output_begin(&handle, event, header.size)) goto exit; perf_output_sample(&handle, &header, data, event); @@ -5717,6 +5721,30 @@ exit: rcu_read_unlock(); } +void +perf_event_output_forward(struct perf_event *event, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + __perf_event_output(event, data, regs, perf_output_begin_forward); +} + +void +perf_event_output_backward(struct perf_event *event, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + __perf_event_output(event, data, regs, perf_output_begin_backward); +} + +void +perf_event_output(struct perf_event *event, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + __perf_event_output(event, data, regs, perf_output_begin); +} + /* * read event_id */ @@ -8153,8 +8181,11 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, if (overflow_handler) { event->overflow_handler = overflow_handler; event->overflow_handler_context = context; + } else if (is_write_backward(event)){ + event->overflow_handler = perf_event_output_backward; + event->overflow_handler_context = NULL; } else { - event->overflow_handler = perf_event_output; + event->overflow_handler = perf_event_output_forward; event->overflow_handler_context = NULL; } @@ -8388,6 +8419,13 @@ perf_event_set_output(struct perf_event *event, struct perf_event *output_event) if (output_event->clock != event->clock) goto out; + /* + * Either writing ring buffer from beginning or from end. + * Mixing is not allowed. + */ + if (is_write_backward(output_event) != is_write_backward(event)) + goto out; + /* * If both events generate aux data, they must be on the same PMU */ diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c index 60be55a64040..c49bab42dc57 100644 --- a/kernel/events/ring_buffer.c +++ b/kernel/events/ring_buffer.c @@ -230,10 +230,24 @@ out: return -ENOSPC; } +int perf_output_begin_forward(struct perf_output_handle *handle, + struct perf_event *event, unsigned int size) +{ + return __perf_output_begin(handle, event, size, false); +} + +int perf_output_begin_backward(struct perf_output_handle *handle, + struct perf_event *event, unsigned int size) +{ + return __perf_output_begin(handle, event, size, true); +} + int perf_output_begin(struct perf_output_handle *handle, struct perf_event *event, unsigned int size) { - return __perf_output_begin(handle, event, size, false); + + return __perf_output_begin(handle, event, size, + unlikely(is_write_backward(event))); } unsigned int perf_output_copy(struct perf_output_handle *handle, -- cgit v1.2.3 From c5dfd78eb79851e278b7973031b9ca363da87a7e Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 21 Apr 2016 12:28:50 -0300 Subject: perf core: Allow setting up max frame stack depth via sysctl The default remains 127, which is good for most cases, and not even hit most of the time, but then for some cases, as reported by Brendan, 1024+ deep frames are appearing on the radar for things like groovy, ruby. And in some workloads putting a _lower_ cap on this may make sense. One that is per event still needs to be put in place tho. The new file is: # cat /proc/sys/kernel/perf_event_max_stack 127 Chaging it: # echo 256 > /proc/sys/kernel/perf_event_max_stack # cat /proc/sys/kernel/perf_event_max_stack 256 But as soon as there is some event using callchains we get: # echo 512 > /proc/sys/kernel/perf_event_max_stack -bash: echo: write error: Device or resource busy # Because we only allocate the callchain percpu data structures when there is a user, which allows for changing the max easily, its just a matter of having no callchain users at that point. Reported-and-Tested-by: Brendan Gregg Reviewed-by: Frederic Weisbecker Acked-by: Alexei Starovoitov Acked-by: David Ahern Cc: Adrian Hunter Cc: Alexander Shishkin Cc: He Kuang Cc: Jiri Olsa Cc: Linus Torvalds Cc: Masami Hiramatsu Cc: Milian Wolff Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Vince Weaver Cc: Wang Nan Cc: Zefan Li Link: http://lkml.kernel.org/r/20160426002928.GB16708@kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- Documentation/sysctl/kernel.txt | 14 ++++++++++++++ arch/arm/kernel/perf_callchain.c | 2 +- arch/arm64/kernel/perf_callchain.c | 4 ++-- arch/metag/kernel/perf_callchain.c | 2 +- arch/mips/kernel/perf_event.c | 4 ++-- arch/powerpc/perf/callchain.c | 4 ++-- arch/sparc/kernel/perf_event.c | 6 +++--- arch/x86/events/core.c | 4 ++-- arch/xtensa/kernel/perf_event.c | 4 ++-- include/linux/perf_event.h | 8 ++++++-- kernel/bpf/stackmap.c | 8 ++++---- kernel/events/callchain.c | 35 +++++++++++++++++++++++++++++++++-- kernel/sysctl.c | 12 ++++++++++++ 13 files changed, 84 insertions(+), 23 deletions(-) (limited to 'include') diff --git a/Documentation/sysctl/kernel.txt b/Documentation/sysctl/kernel.txt index 57653a44b128..260cde08e92e 100644 --- a/Documentation/sysctl/kernel.txt +++ b/Documentation/sysctl/kernel.txt @@ -60,6 +60,7 @@ show up in /proc/sys/kernel: - panic_on_warn - perf_cpu_time_max_percent - perf_event_paranoid +- perf_event_max_stack - pid_max - powersave-nap [ PPC only ] - printk @@ -654,6 +655,19 @@ users (without CAP_SYS_ADMIN). The default value is 1. ============================================================== +perf_event_max_stack: + +Controls maximum number of stack frames to copy for (attr.sample_type & +PERF_SAMPLE_CALLCHAIN) configured events, for instance, when using +'perf record -g' or 'perf trace --call-graph fp'. + +This can only be done when no events are in use that have callchains +enabled, otherwise writing to this file will return -EBUSY. + +The default value is 127. + +============================================================== + pid_max: PID allocation wrap value. When the kernel's next PID value diff --git a/arch/arm/kernel/perf_callchain.c b/arch/arm/kernel/perf_callchain.c index 4e02ae5950ff..27563befa8a2 100644 --- a/arch/arm/kernel/perf_callchain.c +++ b/arch/arm/kernel/perf_callchain.c @@ -75,7 +75,7 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs) tail = (struct frame_tail __user *)regs->ARM_fp - 1; - while ((entry->nr < PERF_MAX_STACK_DEPTH) && + while ((entry->nr < sysctl_perf_event_max_stack) && tail && !((unsigned long)tail & 0x3)) tail = user_backtrace(tail, entry); } diff --git a/arch/arm64/kernel/perf_callchain.c b/arch/arm64/kernel/perf_callchain.c index ff4665462a02..32c3c6e70119 100644 --- a/arch/arm64/kernel/perf_callchain.c +++ b/arch/arm64/kernel/perf_callchain.c @@ -122,7 +122,7 @@ void perf_callchain_user(struct perf_callchain_entry *entry, tail = (struct frame_tail __user *)regs->regs[29]; - while (entry->nr < PERF_MAX_STACK_DEPTH && + while (entry->nr < sysctl_perf_event_max_stack && tail && !((unsigned long)tail & 0xf)) tail = user_backtrace(tail, entry); } else { @@ -132,7 +132,7 @@ void perf_callchain_user(struct perf_callchain_entry *entry, tail = (struct compat_frame_tail __user *)regs->compat_fp - 1; - while ((entry->nr < PERF_MAX_STACK_DEPTH) && + while ((entry->nr < sysctl_perf_event_max_stack) && tail && !((unsigned long)tail & 0x3)) tail = compat_user_backtrace(tail, entry); #endif diff --git a/arch/metag/kernel/perf_callchain.c b/arch/metag/kernel/perf_callchain.c index 315633461a94..252abc12a5a3 100644 --- a/arch/metag/kernel/perf_callchain.c +++ b/arch/metag/kernel/perf_callchain.c @@ -65,7 +65,7 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs) --frame; - while ((entry->nr < PERF_MAX_STACK_DEPTH) && frame) + while ((entry->nr < sysctl_perf_event_max_stack) && frame) frame = user_backtrace(frame, entry); } diff --git a/arch/mips/kernel/perf_event.c b/arch/mips/kernel/perf_event.c index c1cf9c6c3f77..5021c546ad07 100644 --- a/arch/mips/kernel/perf_event.c +++ b/arch/mips/kernel/perf_event.c @@ -35,7 +35,7 @@ static void save_raw_perf_callchain(struct perf_callchain_entry *entry, addr = *sp++; if (__kernel_text_address(addr)) { perf_callchain_store(entry, addr); - if (entry->nr >= PERF_MAX_STACK_DEPTH) + if (entry->nr >= sysctl_perf_event_max_stack) break; } } @@ -59,7 +59,7 @@ void perf_callchain_kernel(struct perf_callchain_entry *entry, } do { perf_callchain_store(entry, pc); - if (entry->nr >= PERF_MAX_STACK_DEPTH) + if (entry->nr >= sysctl_perf_event_max_stack) break; pc = unwind_stack(current, &sp, pc, &ra); } while (pc); diff --git a/arch/powerpc/perf/callchain.c b/arch/powerpc/perf/callchain.c index e04a6752b399..22d9015c1acc 100644 --- a/arch/powerpc/perf/callchain.c +++ b/arch/powerpc/perf/callchain.c @@ -247,7 +247,7 @@ static void perf_callchain_user_64(struct perf_callchain_entry *entry, sp = regs->gpr[1]; perf_callchain_store(entry, next_ip); - while (entry->nr < PERF_MAX_STACK_DEPTH) { + while (entry->nr < sysctl_perf_event_max_stack) { fp = (unsigned long __user *) sp; if (!valid_user_sp(sp, 1) || read_user_stack_64(fp, &next_sp)) return; @@ -453,7 +453,7 @@ static void perf_callchain_user_32(struct perf_callchain_entry *entry, sp = regs->gpr[1]; perf_callchain_store(entry, next_ip); - while (entry->nr < PERF_MAX_STACK_DEPTH) { + while (entry->nr < sysctl_perf_event_max_stack) { fp = (unsigned int __user *) (unsigned long) sp; if (!valid_user_sp(sp, 0) || read_user_stack_32(fp, &next_sp)) return; diff --git a/arch/sparc/kernel/perf_event.c b/arch/sparc/kernel/perf_event.c index 6596f66ce112..a4b8b5aed21c 100644 --- a/arch/sparc/kernel/perf_event.c +++ b/arch/sparc/kernel/perf_event.c @@ -1756,7 +1756,7 @@ void perf_callchain_kernel(struct perf_callchain_entry *entry, } } #endif - } while (entry->nr < PERF_MAX_STACK_DEPTH); + } while (entry->nr < sysctl_perf_event_max_stack); } static inline int @@ -1790,7 +1790,7 @@ static void perf_callchain_user_64(struct perf_callchain_entry *entry, pc = sf.callers_pc; ufp = (unsigned long)sf.fp + STACK_BIAS; perf_callchain_store(entry, pc); - } while (entry->nr < PERF_MAX_STACK_DEPTH); + } while (entry->nr < sysctl_perf_event_max_stack); } static void perf_callchain_user_32(struct perf_callchain_entry *entry, @@ -1822,7 +1822,7 @@ static void perf_callchain_user_32(struct perf_callchain_entry *entry, ufp = (unsigned long)sf.fp; } perf_callchain_store(entry, pc); - } while (entry->nr < PERF_MAX_STACK_DEPTH); + } while (entry->nr < sysctl_perf_event_max_stack); } void diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c index 041e442a3e28..41d93d0e972b 100644 --- a/arch/x86/events/core.c +++ b/arch/x86/events/core.c @@ -2277,7 +2277,7 @@ perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry) fp = compat_ptr(ss_base + regs->bp); pagefault_disable(); - while (entry->nr < PERF_MAX_STACK_DEPTH) { + while (entry->nr < sysctl_perf_event_max_stack) { unsigned long bytes; frame.next_frame = 0; frame.return_address = 0; @@ -2337,7 +2337,7 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs) return; pagefault_disable(); - while (entry->nr < PERF_MAX_STACK_DEPTH) { + while (entry->nr < sysctl_perf_event_max_stack) { unsigned long bytes; frame.next_frame = NULL; frame.return_address = 0; diff --git a/arch/xtensa/kernel/perf_event.c b/arch/xtensa/kernel/perf_event.c index 54f01188c29c..a6b00b3af429 100644 --- a/arch/xtensa/kernel/perf_event.c +++ b/arch/xtensa/kernel/perf_event.c @@ -332,14 +332,14 @@ static int callchain_trace(struct stackframe *frame, void *data) void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs) { - xtensa_backtrace_kernel(regs, PERF_MAX_STACK_DEPTH, + xtensa_backtrace_kernel(regs, sysctl_perf_event_max_stack, callchain_trace, NULL, entry); } void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs) { - xtensa_backtrace_user(regs, PERF_MAX_STACK_DEPTH, + xtensa_backtrace_user(regs, sysctl_perf_event_max_stack, callchain_trace, entry); } diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 85749ae8cb5f..a090700cccca 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -58,7 +58,7 @@ struct perf_guest_info_callbacks { struct perf_callchain_entry { __u64 nr; - __u64 ip[PERF_MAX_STACK_DEPTH]; + __u64 ip[0]; /* /proc/sys/kernel/perf_event_max_stack */ }; struct perf_raw_record { @@ -993,9 +993,11 @@ get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user, extern int get_callchain_buffers(void); extern void put_callchain_buffers(void); +extern int sysctl_perf_event_max_stack; + static inline int perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) { - if (entry->nr < PERF_MAX_STACK_DEPTH) { + if (entry->nr < sysctl_perf_event_max_stack) { entry->ip[entry->nr++] = ip; return 0; } else { @@ -1017,6 +1019,8 @@ extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, void __user *buffer, size_t *lenp, loff_t *ppos); +int perf_event_max_stack_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos); static inline bool perf_paranoid_tracepoint_raw(void) { diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c index 499d9e933f8e..f5a19548be12 100644 --- a/kernel/bpf/stackmap.c +++ b/kernel/bpf/stackmap.c @@ -66,7 +66,7 @@ static struct bpf_map *stack_map_alloc(union bpf_attr *attr) /* check sanity of attributes */ if (attr->max_entries == 0 || attr->key_size != 4 || value_size < 8 || value_size % 8 || - value_size / 8 > PERF_MAX_STACK_DEPTH) + value_size / 8 > sysctl_perf_event_max_stack) return ERR_PTR(-EINVAL); /* hash table size must be power of 2 */ @@ -124,8 +124,8 @@ static u64 bpf_get_stackid(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5) struct perf_callchain_entry *trace; struct stack_map_bucket *bucket, *new_bucket, *old_bucket; u32 max_depth = map->value_size / 8; - /* stack_map_alloc() checks that max_depth <= PERF_MAX_STACK_DEPTH */ - u32 init_nr = PERF_MAX_STACK_DEPTH - max_depth; + /* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */ + u32 init_nr = sysctl_perf_event_max_stack - max_depth; u32 skip = flags & BPF_F_SKIP_FIELD_MASK; u32 hash, id, trace_nr, trace_len; bool user = flags & BPF_F_USER_STACK; @@ -143,7 +143,7 @@ static u64 bpf_get_stackid(u64 r1, u64 r2, u64 flags, u64 r4, u64 r5) return -EFAULT; /* get_perf_callchain() guarantees that trace->nr >= init_nr - * and trace-nr <= PERF_MAX_STACK_DEPTH, so trace_nr <= max_depth + * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth */ trace_nr = trace->nr - init_nr; diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c index 343c22f5e867..b9325e7dcba1 100644 --- a/kernel/events/callchain.c +++ b/kernel/events/callchain.c @@ -18,6 +18,14 @@ struct callchain_cpus_entries { struct perf_callchain_entry *cpu_entries[0]; }; +int sysctl_perf_event_max_stack __read_mostly = PERF_MAX_STACK_DEPTH; + +static inline size_t perf_callchain_entry__sizeof(void) +{ + return (sizeof(struct perf_callchain_entry) + + sizeof(__u64) * sysctl_perf_event_max_stack); +} + static DEFINE_PER_CPU(int, callchain_recursion[PERF_NR_CONTEXTS]); static atomic_t nr_callchain_events; static DEFINE_MUTEX(callchain_mutex); @@ -73,7 +81,7 @@ static int alloc_callchain_buffers(void) if (!entries) return -ENOMEM; - size = sizeof(struct perf_callchain_entry) * PERF_NR_CONTEXTS; + size = perf_callchain_entry__sizeof() * PERF_NR_CONTEXTS; for_each_possible_cpu(cpu) { entries->cpu_entries[cpu] = kmalloc_node(size, GFP_KERNEL, @@ -147,7 +155,8 @@ static struct perf_callchain_entry *get_callchain_entry(int *rctx) cpu = smp_processor_id(); - return &entries->cpu_entries[cpu][*rctx]; + return (((void *)entries->cpu_entries[cpu]) + + (*rctx * perf_callchain_entry__sizeof())); } static void @@ -215,3 +224,25 @@ exit_put: return entry; } + +int perf_event_max_stack_handler(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, loff_t *ppos) +{ + int new_value = sysctl_perf_event_max_stack, ret; + struct ctl_table new_table = *table; + + new_table.data = &new_value; + ret = proc_dointvec_minmax(&new_table, write, buffer, lenp, ppos); + if (ret || !write) + return ret; + + mutex_lock(&callchain_mutex); + if (atomic_read(&nr_callchain_events)) + ret = -EBUSY; + else + sysctl_perf_event_max_stack = new_value; + + mutex_unlock(&callchain_mutex); + + return ret; +} diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 725587f10667..c8b318663525 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -130,6 +130,9 @@ static int one_thousand = 1000; #ifdef CONFIG_PRINTK static int ten_thousand = 10000; #endif +#ifdef CONFIG_PERF_EVENTS +static int six_hundred_forty_kb = 640 * 1024; +#endif /* this is needed for the proc_doulongvec_minmax of vm_dirty_bytes */ static unsigned long dirty_bytes_min = 2 * PAGE_SIZE; @@ -1144,6 +1147,15 @@ static struct ctl_table kern_table[] = { .extra1 = &zero, .extra2 = &one_hundred, }, + { + .procname = "perf_event_max_stack", + .data = NULL, /* filled in by handler */ + .maxlen = sizeof(sysctl_perf_event_max_stack), + .mode = 0644, + .proc_handler = perf_event_max_stack_handler, + .extra1 = &zero, + .extra2 = &six_hundred_forty_kb, + }, #endif #ifdef CONFIG_KMEMCHECK { -- cgit v1.2.3 From 375637bc524952f1122ea22caf5a8f1fecad8228 Mon Sep 17 00:00:00 2001 From: Alexander Shishkin Date: Wed, 27 Apr 2016 18:44:46 +0300 Subject: perf/core: Introduce address range filtering Many instruction tracing PMUs out there support address range-based filtering, which would, for example, generate trace data only for a given range of instruction addresses, which is useful for tracing individual functions, modules or libraries. Other PMUs may also utilize this functionality to allow filtering to or filtering out code at certain address ranges. This patch introduces the interface for userspace to specify these filters and for the PMU drivers to apply these filters to hardware configuration. The user interface is an ASCII string that is passed via an ioctl() and specifies (in the form of an ASCII string) address ranges within certain object files or within kernel. There is no special treatment for kernel modules yet, but it might be a worthy pursuit. The PMU driver interface basically adds two extra callbacks to the PMU driver structure, one of which validates the filter configuration proposed by the user against what the hardware is actually capable of doing and the other one translates hardware-independent filter configuration into something that can be programmed into the hardware. Signed-off-by: Alexander Shishkin Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: Mathieu Poirier Cc: Arnaldo Carvalho de Melo Cc: Arnaldo Carvalho de Melo Cc: Borislav Petkov Cc: Jiri Olsa Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Vince Weaver Cc: vince@deater.net Link: http://lkml.kernel.org/r/1461771888-10409-6-git-send-email-alexander.shishkin@linux.intel.com Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 98 +++++++ kernel/events/core.c | 623 +++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 705 insertions(+), 16 deletions(-) (limited to 'include') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index a090700cccca..c77e4a159fa2 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -151,6 +151,15 @@ struct hw_perf_event { */ struct task_struct *target; + /* + * PMU would store hardware filter configuration + * here. + */ + void *addr_filters; + + /* Last sync'ed generation of filters */ + unsigned long addr_filters_gen; + /* * hw_perf_event::state flags; used to track the PERF_EF_* state. */ @@ -240,6 +249,9 @@ struct pmu { int task_ctx_nr; int hrtimer_interval_ms; + /* number of address filters this PMU can do */ + unsigned int nr_addr_filters; + /* * Fully disable/enable this PMU, can be used to protect from the PMI * as well as for lazy/batch writing of the MSRs. @@ -392,12 +404,71 @@ struct pmu { */ void (*free_aux) (void *aux); /* optional */ + /* + * Validate address range filters: make sure the HW supports the + * requested configuration and number of filters; return 0 if the + * supplied filters are valid, -errno otherwise. + * + * Runs in the context of the ioctl()ing process and is not serialized + * with the rest of the PMU callbacks. + */ + int (*addr_filters_validate) (struct list_head *filters); + /* optional */ + + /* + * Synchronize address range filter configuration: + * translate hw-agnostic filters into hardware configuration in + * event::hw::addr_filters. + * + * Runs as a part of filter sync sequence that is done in ->start() + * callback by calling perf_event_addr_filters_sync(). + * + * May (and should) traverse event::addr_filters::list, for which its + * caller provides necessary serialization. + */ + void (*addr_filters_sync) (struct perf_event *event); + /* optional */ + /* * Filter events for PMU-specific reasons. */ int (*filter_match) (struct perf_event *event); /* optional */ }; +/** + * struct perf_addr_filter - address range filter definition + * @entry: event's filter list linkage + * @inode: object file's inode for file-based filters + * @offset: filter range offset + * @size: filter range size + * @range: 1: range, 0: address + * @filter: 1: filter/start, 0: stop + * + * This is a hardware-agnostic filter configuration as specified by the user. + */ +struct perf_addr_filter { + struct list_head entry; + struct inode *inode; + unsigned long offset; + unsigned long size; + unsigned int range : 1, + filter : 1; +}; + +/** + * struct perf_addr_filters_head - container for address range filters + * @list: list of filters for this event + * @lock: spinlock that serializes accesses to the @list and event's + * (and its children's) filter generations. + * + * A child event will use parent's @list (and therefore @lock), so they are + * bundled together; see perf_event_addr_filters(). + */ +struct perf_addr_filters_head { + struct list_head list; + raw_spinlock_t lock; +}; + /** * enum perf_event_active_state - the states of a event */ @@ -566,6 +637,12 @@ struct perf_event { atomic_t event_limit; + /* address range filters */ + struct perf_addr_filters_head addr_filters; + /* vma address array for file-based filders */ + unsigned long *addr_filters_offs; + unsigned long addr_filters_gen; + void (*destroy)(struct perf_event *); struct rcu_head rcu_head; @@ -1070,6 +1147,27 @@ static inline bool is_write_backward(struct perf_event *event) return !!event->attr.write_backward; } +static inline bool has_addr_filter(struct perf_event *event) +{ + return event->pmu->nr_addr_filters; +} + +/* + * An inherited event uses parent's filters + */ +static inline struct perf_addr_filters_head * +perf_event_addr_filters(struct perf_event *event) +{ + struct perf_addr_filters_head *ifh = &event->addr_filters; + + if (event->parent) + ifh = &event->parent->addr_filters; + + return ifh; +} + +extern void perf_event_addr_filters_sync(struct perf_event *event); + extern int perf_output_begin(struct perf_output_handle *handle, struct perf_event *event, unsigned int size); extern int perf_output_begin_forward(struct perf_output_handle *handle, diff --git a/kernel/events/core.c b/kernel/events/core.c index 2bb7c47f18e3..ffdc096a4f4e 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -44,6 +44,8 @@ #include #include #include +#include +#include #include "internal.h" @@ -2365,11 +2367,17 @@ void perf_event_enable(struct perf_event *event) } EXPORT_SYMBOL_GPL(perf_event_enable); +struct stop_event_data { + struct perf_event *event; + unsigned int restart; +}; + static int __perf_event_stop(void *info) { - struct perf_event *event = info; + struct stop_event_data *sd = info; + struct perf_event *event = sd->event; - /* for AUX events, our job is done if the event is already inactive */ + /* if it's already INACTIVE, do nothing */ if (READ_ONCE(event->state) != PERF_EVENT_STATE_ACTIVE) return 0; @@ -2385,9 +2393,86 @@ static int __perf_event_stop(void *info) event->pmu->stop(event, PERF_EF_UPDATE); + /* + * May race with the actual stop (through perf_pmu_output_stop()), + * but it is only used for events with AUX ring buffer, and such + * events will refuse to restart because of rb::aux_mmap_count==0, + * see comments in perf_aux_output_begin(). + * + * Since this is happening on a event-local CPU, no trace is lost + * while restarting. + */ + if (sd->restart) + event->pmu->start(event, PERF_EF_START); + return 0; } +static int perf_event_restart(struct perf_event *event) +{ + struct stop_event_data sd = { + .event = event, + .restart = 1, + }; + int ret = 0; + + do { + if (READ_ONCE(event->state) != PERF_EVENT_STATE_ACTIVE) + return 0; + + /* matches smp_wmb() in event_sched_in() */ + smp_rmb(); + + /* + * We only want to restart ACTIVE events, so if the event goes + * inactive here (event->oncpu==-1), there's nothing more to do; + * fall through with ret==-ENXIO. + */ + ret = cpu_function_call(READ_ONCE(event->oncpu), + __perf_event_stop, &sd); + } while (ret == -EAGAIN); + + return ret; +} + +/* + * In order to contain the amount of racy and tricky in the address filter + * configuration management, it is a two part process: + * + * (p1) when userspace mappings change as a result of (1) or (2) or (3) below, + * we update the addresses of corresponding vmas in + * event::addr_filters_offs array and bump the event::addr_filters_gen; + * (p2) when an event is scheduled in (pmu::add), it calls + * perf_event_addr_filters_sync() which calls pmu::addr_filters_sync() + * if the generation has changed since the previous call. + * + * If (p1) happens while the event is active, we restart it to force (p2). + * + * (1) perf_addr_filters_apply(): adjusting filters' offsets based on + * pre-existing mappings, called once when new filters arrive via SET_FILTER + * ioctl; + * (2) perf_addr_filters_adjust(): adjusting filters' offsets based on newly + * registered mapping, called for every new mmap(), with mm::mmap_sem down + * for reading; + * (3) perf_event_addr_filters_exec(): clearing filters' offsets in the process + * of exec. + */ +void perf_event_addr_filters_sync(struct perf_event *event) +{ + struct perf_addr_filters_head *ifh = perf_event_addr_filters(event); + + if (!has_addr_filter(event)) + return; + + raw_spin_lock(&ifh->lock); + if (event->addr_filters_gen != event->hw.addr_filters_gen) { + event->pmu->addr_filters_sync(event); + event->hw.addr_filters_gen = event->addr_filters_gen; + } + raw_spin_unlock(&ifh->lock); +} +EXPORT_SYMBOL_GPL(perf_event_addr_filters_sync); + static int _perf_event_refresh(struct perf_event *event, int refresh) { /* @@ -3237,16 +3322,6 @@ out: put_ctx(clone_ctx); } -void perf_event_exec(void) -{ - int ctxn; - - rcu_read_lock(); - for_each_task_context_nr(ctxn) - perf_event_enable_on_exec(ctxn); - rcu_read_unlock(); -} - struct perf_read_data { struct perf_event *event; bool group; @@ -3748,6 +3823,9 @@ static bool exclusive_event_installable(struct perf_event *event, return true; } +static void perf_addr_filters_splice(struct perf_event *event, + struct list_head *head); + static void _free_event(struct perf_event *event) { irq_work_sync(&event->pending); @@ -3775,6 +3853,8 @@ static void _free_event(struct perf_event *event) } perf_event_free_bpf_prog(event); + perf_addr_filters_splice(event, NULL); + kfree(event->addr_filters_offs); if (event->destroy) event->destroy(event); @@ -5846,6 +5926,57 @@ next: rcu_read_unlock(); } +/* + * Clear all file-based filters at exec, they'll have to be + * re-instated when/if these objects are mmapped again. + */ +static void perf_event_addr_filters_exec(struct perf_event *event, void *data) +{ + struct perf_addr_filters_head *ifh = perf_event_addr_filters(event); + struct perf_addr_filter *filter; + unsigned int restart = 0, count = 0; + unsigned long flags; + + if (!has_addr_filter(event)) + return; + + raw_spin_lock_irqsave(&ifh->lock, flags); + list_for_each_entry(filter, &ifh->list, entry) { + if (filter->inode) { + event->addr_filters_offs[count] = 0; + restart++; + } + + count++; + } + + if (restart) + event->addr_filters_gen++; + raw_spin_unlock_irqrestore(&ifh->lock, flags); + + if (restart) + perf_event_restart(event); +} + +void perf_event_exec(void) +{ + struct perf_event_context *ctx; + int ctxn; + + rcu_read_lock(); + for_each_task_context_nr(ctxn) { + ctx = current->perf_event_ctxp[ctxn]; + if (!ctx) + continue; + + perf_event_enable_on_exec(ctxn); + + perf_event_aux_ctx(ctx, perf_event_addr_filters_exec, NULL, + true); + } + rcu_read_unlock(); +} + struct remote_output { struct ring_buffer *rb; int err; @@ -5856,6 +5987,9 @@ static void __perf_event_output_stop(struct perf_event *event, void *data) struct perf_event *parent = event->parent; struct remote_output *ro = data; struct ring_buffer *rb = ro->rb; + struct stop_event_data sd = { + .event = event, + }; if (!has_aux(event)) return; @@ -5868,7 +6002,7 @@ static void __perf_event_output_stop(struct perf_event *event, void *data) * ring-buffer, but it will be the child that's actually using it: */ if (rcu_dereference(parent->rb) == rb) - ro->err = __perf_event_stop(event); + ro->err = __perf_event_stop(&sd); } static int __perf_pmu_output_stop(void *info) @@ -6329,6 +6463,87 @@ got_name: kfree(buf); } +/* + * Whether this @filter depends on a dynamic object which is not loaded + * yet or its load addresses are not known. + */ +static bool perf_addr_filter_needs_mmap(struct perf_addr_filter *filter) +{ + return filter->filter && filter->inode; +} + +/* + * Check whether inode and address range match filter criteria. + */ +static bool perf_addr_filter_match(struct perf_addr_filter *filter, + struct file *file, unsigned long offset, + unsigned long size) +{ + if (filter->inode != file->f_inode) + return false; + + if (filter->offset > offset + size) + return false; + + if (filter->offset + filter->size < offset) + return false; + + return true; +} + +static void __perf_addr_filters_adjust(struct perf_event *event, void *data) +{ + struct perf_addr_filters_head *ifh = perf_event_addr_filters(event); + struct vm_area_struct *vma = data; + unsigned long off = vma->vm_pgoff << PAGE_SHIFT, flags; + struct file *file = vma->vm_file; + struct perf_addr_filter *filter; + unsigned int restart = 0, count = 0; + + if (!has_addr_filter(event)) + return; + + if (!file) + return; + + raw_spin_lock_irqsave(&ifh->lock, flags); + list_for_each_entry(filter, &ifh->list, entry) { + if (perf_addr_filter_match(filter, file, off, + vma->vm_end - vma->vm_start)) { + event->addr_filters_offs[count] = vma->vm_start; + restart++; + } + + count++; + } + + if (restart) + event->addr_filters_gen++; + raw_spin_unlock_irqrestore(&ifh->lock, flags); + + if (restart) + perf_event_restart(event); +} + +/* + * Adjust all task's events' filters to the new vma + */ +static void perf_addr_filters_adjust(struct vm_area_struct *vma) +{ + struct perf_event_context *ctx; + int ctxn; + + rcu_read_lock(); + for_each_task_context_nr(ctxn) { + ctx = rcu_dereference(current->perf_event_ctxp[ctxn]); + if (!ctx) + continue; + + perf_event_aux_ctx(ctx, __perf_addr_filters_adjust, vma, true); + } + rcu_read_unlock(); +} + void perf_event_mmap(struct vm_area_struct *vma) { struct perf_mmap_event mmap_event; @@ -6360,6 +6575,7 @@ void perf_event_mmap(struct vm_area_struct *vma) /* .flags (attr_mmap2 only) */ }; + perf_addr_filters_adjust(vma); perf_event_mmap_event(&mmap_event); } @@ -7319,13 +7535,370 @@ void perf_bp_event(struct perf_event *bp, void *data) } #endif +/* + * Allocate a new address filter + */ +static struct perf_addr_filter * +perf_addr_filter_new(struct perf_event *event, struct list_head *filters) +{ + int node = cpu_to_node(event->cpu == -1 ? 0 : event->cpu); + struct perf_addr_filter *filter; + + filter = kzalloc_node(sizeof(*filter), GFP_KERNEL, node); + if (!filter) + return NULL; + + INIT_LIST_HEAD(&filter->entry); + list_add_tail(&filter->entry, filters); + + return filter; +} + +static void free_filters_list(struct list_head *filters) +{ + struct perf_addr_filter *filter, *iter; + + list_for_each_entry_safe(filter, iter, filters, entry) { + if (filter->inode) + iput(filter->inode); + list_del(&filter->entry); + kfree(filter); + } +} + +/* + * Free existing address filters and optionally install new ones + */ +static void perf_addr_filters_splice(struct perf_event *event, + struct list_head *head) +{ + unsigned long flags; + LIST_HEAD(list); + + if (!has_addr_filter(event)) + return; + + /* don't bother with children, they don't have their own filters */ + if (event->parent) + return; + + raw_spin_lock_irqsave(&event->addr_filters.lock, flags); + + list_splice_init(&event->addr_filters.list, &list); + if (head) + list_splice(head, &event->addr_filters.list); + + raw_spin_unlock_irqrestore(&event->addr_filters.lock, flags); + + free_filters_list(&list); +} + +/* + * Scan through mm's vmas and see if one of them matches the + * @filter; if so, adjust filter's address range. + * Called with mm::mmap_sem down for reading. + */ +static unsigned long perf_addr_filter_apply(struct perf_addr_filter *filter, + struct mm_struct *mm) +{ + struct vm_area_struct *vma; + + for (vma = mm->mmap; vma; vma = vma->vm_next) { + struct file *file = vma->vm_file; + unsigned long off = vma->vm_pgoff << PAGE_SHIFT; + unsigned long vma_size = vma->vm_end - vma->vm_start; + + if (!file) + continue; + + if (!perf_addr_filter_match(filter, file, off, vma_size)) + continue; + + return vma->vm_start; + } + + return 0; +} + +/* + * Update event's address range filters based on the + * task's existing mappings, if any. + */ +static void perf_event_addr_filters_apply(struct perf_event *event) +{ + struct perf_addr_filters_head *ifh = perf_event_addr_filters(event); + struct task_struct *task = READ_ONCE(event->ctx->task); + struct perf_addr_filter *filter; + struct mm_struct *mm = NULL; + unsigned int count = 0; + unsigned long flags; + + /* + * We may observe TASK_TOMBSTONE, which means that the event tear-down + * will stop on the parent's child_mutex that our caller is also holding + */ + if (task == TASK_TOMBSTONE) + return; + + mm = get_task_mm(event->ctx->task); + if (!mm) + goto restart; + + down_read(&mm->mmap_sem); + + raw_spin_lock_irqsave(&ifh->lock, flags); + list_for_each_entry(filter, &ifh->list, entry) { + event->addr_filters_offs[count] = 0; + + if (perf_addr_filter_needs_mmap(filter)) + event->addr_filters_offs[count] = + perf_addr_filter_apply(filter, mm); + + count++; + } + + event->addr_filters_gen++; + raw_spin_unlock_irqrestore(&ifh->lock, flags); + + up_read(&mm->mmap_sem); + + mmput(mm); + +restart: + perf_event_restart(event); +} + +/* + * Address range filtering: limiting the data to certain + * instruction address ranges. Filters are ioctl()ed to us from + * userspace as ascii strings. + * + * Filter string format: + * + * ACTION RANGE_SPEC + * where ACTION is one of the + * * "filter": limit the trace to this region + * * "start": start tracing from this address + * * "stop": stop tracing at this address/region; + * RANGE_SPEC is + * * for kernel addresses: [/] + * * for object files: [/]@ + * + * if is not specified, the range is treated as a single address. + */ +enum { + IF_ACT_FILTER, + IF_ACT_START, + IF_ACT_STOP, + IF_SRC_FILE, + IF_SRC_KERNEL, + IF_SRC_FILEADDR, + IF_SRC_KERNELADDR, +}; + +enum { + IF_STATE_ACTION = 0, + IF_STATE_SOURCE, + IF_STATE_END, +}; + +static const match_table_t if_tokens = { + { IF_ACT_FILTER, "filter" }, + { IF_ACT_START, "start" }, + { IF_ACT_STOP, "stop" }, + { IF_SRC_FILE, "%u/%u@%s" }, + { IF_SRC_KERNEL, "%u/%u" }, + { IF_SRC_FILEADDR, "%u@%s" }, + { IF_SRC_KERNELADDR, "%u" }, +}; + +/* + * Address filter string parser + */ +static int +perf_event_parse_addr_filter(struct perf_event *event, char *fstr, + struct list_head *filters) +{ + struct perf_addr_filter *filter = NULL; + char *start, *orig, *filename = NULL; + struct path path; + substring_t args[MAX_OPT_ARGS]; + int state = IF_STATE_ACTION, token; + unsigned int kernel = 0; + int ret = -EINVAL; + + orig = fstr = kstrdup(fstr, GFP_KERNEL); + if (!fstr) + return -ENOMEM; + + while ((start = strsep(&fstr, " ,\n")) != NULL) { + ret = -EINVAL; + + if (!*start) + continue; + + /* filter definition begins */ + if (state == IF_STATE_ACTION) { + filter = perf_addr_filter_new(event, filters); + if (!filter) + goto fail; + } + + token = match_token(start, if_tokens, args); + switch (token) { + case IF_ACT_FILTER: + case IF_ACT_START: + filter->filter = 1; + + case IF_ACT_STOP: + if (state != IF_STATE_ACTION) + goto fail; + + state = IF_STATE_SOURCE; + break; + + case IF_SRC_KERNELADDR: + case IF_SRC_KERNEL: + kernel = 1; + + case IF_SRC_FILEADDR: + case IF_SRC_FILE: + if (state != IF_STATE_SOURCE) + goto fail; + + if (token == IF_SRC_FILE || token == IF_SRC_KERNEL) + filter->range = 1; + + *args[0].to = 0; + ret = kstrtoul(args[0].from, 0, &filter->offset); + if (ret) + goto fail; + + if (filter->range) { + *args[1].to = 0; + ret = kstrtoul(args[1].from, 0, &filter->size); + if (ret) + goto fail; + } + + if (token == IF_SRC_FILE) { + filename = match_strdup(&args[2]); + if (!filename) { + ret = -ENOMEM; + goto fail; + } + } + + state = IF_STATE_END; + break; + + default: + goto fail; + } + + /* + * Filter definition is fully parsed, validate and install it. + * Make sure that it doesn't contradict itself or the event's + * attribute. + */ + if (state == IF_STATE_END) { + if (kernel && event->attr.exclude_kernel) + goto fail; + + if (!kernel) { + if (!filename) + goto fail; + + /* look up the path and grab its inode */ + ret = kern_path(filename, LOOKUP_FOLLOW, &path); + if (ret) + goto fail_free_name; + + filter->inode = igrab(d_inode(path.dentry)); + path_put(&path); + kfree(filename); + filename = NULL; + + ret = -EINVAL; + if (!filter->inode || + !S_ISREG(filter->inode->i_mode)) + /* free_filters_list() will iput() */ + goto fail; + } + + /* ready to consume more filters */ + state = IF_STATE_ACTION; + filter = NULL; + } + } + + if (state != IF_STATE_ACTION) + goto fail; + + kfree(orig); + + return 0; + +fail_free_name: + kfree(filename); +fail: + free_filters_list(filters); + kfree(orig); + + return ret; +} + +static int +perf_event_set_addr_filter(struct perf_event *event, char *filter_str) +{ + LIST_HEAD(filters); + int ret; + + /* + * Since this is called in perf_ioctl() path, we're already holding + * ctx::mutex. + */ + lockdep_assert_held(&event->ctx->mutex); + + if (WARN_ON_ONCE(event->parent)) + return -EINVAL; + + /* + * For now, we only support filtering in per-task events; doing so + * for CPU-wide events requires additional context switching trickery, + * since same object code will be mapped at different virtual + * addresses in different processes. + */ + if (!event->ctx->task) + return -EOPNOTSUPP; + + ret = perf_event_parse_addr_filter(event, filter_str, &filters); + if (ret) + return ret; + + ret = event->pmu->addr_filters_validate(&filters); + if (ret) { + free_filters_list(&filters); + return ret; + } + + /* remove existing filters, if any */ + perf_addr_filters_splice(event, &filters); + + /* install new filters */ + perf_event_for_each_child(event, perf_event_addr_filters_apply); + + return ret; +} + static int perf_event_set_filter(struct perf_event *event, void __user *arg) { char *filter_str; int ret = -EINVAL; - if (event->attr.type != PERF_TYPE_TRACEPOINT || - !IS_ENABLED(CONFIG_EVENT_TRACING)) + if ((event->attr.type != PERF_TYPE_TRACEPOINT || + !IS_ENABLED(CONFIG_EVENT_TRACING)) && + !has_addr_filter(event)) return -EINVAL; filter_str = strndup_user(arg, PAGE_SIZE); @@ -7336,6 +7909,8 @@ static int perf_event_set_filter(struct perf_event *event, void __user *arg) event->attr.type == PERF_TYPE_TRACEPOINT) ret = ftrace_profile_set_filter(event, event->attr.config, filter_str); + else if (has_addr_filter(event)) + ret = perf_event_set_addr_filter(event, filter_str); kfree(filter_str); return ret; @@ -8130,6 +8705,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, INIT_LIST_HEAD(&event->sibling_list); INIT_LIST_HEAD(&event->rb_entry); INIT_LIST_HEAD(&event->active_entry); + INIT_LIST_HEAD(&event->addr_filters.list); INIT_HLIST_NODE(&event->hlist_entry); @@ -8137,6 +8713,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, init_irq_work(&event->pending, perf_pending_event); mutex_init(&event->mmap_mutex); + raw_spin_lock_init(&event->addr_filters.lock); atomic_long_set(&event->refcount, 1); event->cpu = cpu; @@ -8221,11 +8798,22 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, if (err) goto err_pmu; + if (has_addr_filter(event)) { + event->addr_filters_offs = kcalloc(pmu->nr_addr_filters, + sizeof(unsigned long), + GFP_KERNEL); + if (!event->addr_filters_offs) + goto err_per_task; + + /* force hw sync on the address filters */ + event->addr_filters_gen = 1; + } + if (!event->parent) { if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) { err = get_callchain_buffers(); if (err) - goto err_per_task; + goto err_addr_filters; } } @@ -8234,6 +8822,9 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu, return event; +err_addr_filters: + kfree(event->addr_filters_offs); + err_per_task: exclusive_event_destroy(event); -- cgit v1.2.3 From 5101ef20f0ef1de79091a1fdb6b1a7f07565545a Mon Sep 17 00:00:00 2001 From: Mark Rutland Date: Tue, 26 Apr 2016 11:33:46 +0100 Subject: perf/arm: Special-case hetereogeneous CPUs Commit: 26657848502b7847 ("perf/core: Verify we have a single perf_hw_context PMU") forcefully prevents multiple PMUs from sharing perf_hw_context, as this generally doesn't make sense. It is a common bug for uncore PMUs to use perf_hw_context rather than perf_invalid_context, which this detects. However, systems exist with heterogeneous CPUs (and hence heterogeneous HW PMUs), for which sharing perf_hw_context is necessary, and possible in some limited cases. To make this work we have to perform some gymnastics, as we did in these commits: 66eb579e66ecfea5 ("perf: allow for PMU-specific event filtering") c904e32a69b7c779 ("arm: perf: filter unschedulable events") To allow those systems to work, we must allow PMUs for heterogeneous CPUs to share perf_hw_context, though we must still disallow sharing otherwise to detect the common misuse of perf_hw_context. This patch adds a new PERF_PMU_CAP_HETEROGENEOUS_CPUS for this, updates the core logic to account for this, and makes use of it in the arm_pmu code that is used for systems with heterogeneous CPUs. Comments are added to make the rationale clear and hopefully avoid accidental abuse. Signed-off-by: Mark Rutland Signed-off-by: Peter Zijlstra (Intel) Cc: Alexander Shishkin Cc: Arnaldo Carvalho de Melo Cc: Catalin Marinas Cc: Jiri Olsa Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Thomas Gleixner Cc: Vince Weaver Cc: Will Deacon Cc: linux-arm-kernel@lists.infradead.org Link: http://lkml.kernel.org/r/20160426103346.GA20836@leverpostej Signed-off-by: Ingo Molnar --- drivers/perf/arm_pmu.c | 8 ++++++++ include/linux/perf_event.h | 1 + kernel/events/core.c | 8 +++++++- 3 files changed, 16 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c index f70090897fdf..f2d01d4d9364 100644 --- a/drivers/perf/arm_pmu.c +++ b/drivers/perf/arm_pmu.c @@ -847,6 +847,14 @@ static int cpu_pmu_init(struct arm_pmu *cpu_pmu) if (!platform_get_irq(cpu_pmu->plat_device, 0)) cpu_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT; + /* + * This is a CPU PMU potentially in a heterogeneous configuration (e.g. + * big.LITTLE). This is not an uncore PMU, and we have taken ctx + * sharing into account (e.g. with our pmu::filter_match callback and + * pmu::event_init group validation). + */ + cpu_pmu->pmu.capabilities |= PERF_PMU_CAP_HETEROGENEOUS_CPUS; + return 0; out_unregister: diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index c77e4a159fa2..9e1c3ada91c4 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -225,6 +225,7 @@ struct perf_event; #define PERF_PMU_CAP_AUX_SW_DOUBLEBUF 0x08 #define PERF_PMU_CAP_EXCLUSIVE 0x10 #define PERF_PMU_CAP_ITRACE 0x20 +#define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40 /** * struct pmu - generic performance monitoring unit diff --git a/kernel/events/core.c b/kernel/events/core.c index 63be65437e9e..fc0290f25482 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -8443,7 +8443,13 @@ skip_type: if (pmu->task_ctx_nr == perf_hw_context) { static int hw_context_taken = 0; - if (WARN_ON_ONCE(hw_context_taken)) + /* + * Other than systems with heterogeneous CPUs, it never makes + * sense for two PMUs to share perf_hw_context. PMUs which are + * uncore must use perf_invalid_context. + */ + if (WARN_ON_ONCE(hw_context_taken && + !(pmu->capabilities & PERF_PMU_CAP_HETEROGENEOUS_CPUS))) pmu->task_ctx_nr = perf_invalid_context; hw_context_taken = 1; -- cgit v1.2.3