- 07 Jul, 2011 1 commit
-
-
Steven Rostedt authored
I triggered a triple fault with gcc 4.5.1 because it did not honor the inline annotation to arch_local_save_flags() function and that function was added to the pool of functions traced by the function tracer. When preempt_schedule() called arch_local_save_flags() (called by irqs_disabled()), it was traced, but the first thing the function tracer does is disable preemption. When it enables preemption, the NEED_RESCHED flag will not have been cleared and the preemption check will trigger the call to preempt_schedule() again. Although the dynamic function tracer crashed immediately, the static version of the function tracer (CONFIG_DYNAMIC_FTRACE is not set) actually was able to show where the problem was. swapper-1 3.N.. 103885us : arch_local_save_flags <-preempt_schedule swapper-1 3.N.. 103886us : arch_local_save_flags <-preempt_schedule swapper-1 3.N.. 103886us : arch_local_save_flags <-preempt_schedule swapper-1 3.N.. 103887us : arch_local_save_flags <-preempt_schedule swapper-1 3.N.. 103887us : arch_local_save_flags <-preempt_schedule swapper-1 3.N.. 103888us : arch_local_save_flags <-preempt_schedule swapper-1 3.N.. 103888us : arch_local_save_flags <-preempt_schedule It went on for a while before it triple faulted with a corrupted stack. The arch_local_save_flags and arch_local_irq_* functions should not be traced. Even though they are marked as inline, gcc may still make them a function and enable tracing of them. The simple solution is to just mark them as notrace. I had to add the <linux/types.h> for this file to include the notrace tag. Signed-off-by:
Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/20110702033852.733414762@goodmis.orgSigned-off-by:
Ingo Molnar <mingo@elte.hu>
-
- 05 Jul, 2011 2 commits
-
-
Ingo Molnar authored
Merge branch 'tip/perf/core-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-2.6-trace into perf/core
-
Anton Blanchard authored
Add an option to perf report/annotate/script to specify which CPUs to operate on. This enables us to take a single system wide profile and analyse each CPU (or group of CPUs) in isolation. This was useful when profiling a multiprocess workload where the bottleneck was on one CPU but this was hidden in the overall profile. Per process and per thread breakdowns didn't help because multiple processes were running on each CPU and no single process consumed an entire CPU. The patch converts the list of CPUs returned by cpu_map__new into a bitmap for fast lookup. I wanted to use -C to be consistent with perf top/record/stat, but unfortunately perf report already uses -C <comms>. v2: Incorporate suggestions from David Ahern: - Added -c to perf script - Check that SAMPLE_CPU is set when -c is used - Update documentation v3: Create perf_session__cpu_bitmap() Signed-off-by:
Anton Blanchard <anton@samba.org> Acked-by:
David Ahern <dsahern@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Link: http://lkml.kernel.org/r/20110704215750.11647eb9@krytenSigned-off-by:
-
- 04 Jul, 2011 1 commit
-
-
- 03 Jul, 2011 1 commit
-
-
Ingo Molnar authored
Merge branch 'perf/stacktrace' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing into perf/core
-
- 02 Jul, 2011 6 commits
-
-
Frederic Weisbecker authored
rbp is used in SAVE_ARGS_IRQ to save the old stack pointer in order to restore it later in ret_from_intr. It is convenient because we save its value in the irq regs and it's easily restored using the leave instruction. However this is a kind of abuse of the frame pointer which role is to help unwinding the kernel by chaining frames together, each node following the return address to the previous frame. But although we are breaking the frame by changing the stack pointer, there is no preceding return address before the new frame. Hence using the frame pointer to link the two stacks breaks the stack unwinders that find a random value instead of a return address here. There is no workaround that can work in every case. We are using the fixup_bp_irq_link() function to dereference that abused frame pointer in the case of non nesting interrupt (which means stack changed). But that doesn't fix the case of interrupts that don't change the stack (but we still have the unconditional frame link), which is the case of hardirq interrupting softirq. We have no way to detect this transition so the frame irq link is considered as a real frame pointer and the return address is dereferenced but it is still a spurious one. There are two possible results of this: either the spurious return address, a random stack value, luckily belongs to the kernel text and then the unwinding can continue and we just have a weird entry in the stack trace. Or it doesn't belong to the kernel text and unwinding stops there. This is the reason why stacktraces (including perf callchains) on irqs that interrupted softirqs don't work very well. To solve this, we don't save the old stack pointer on rbp anymore but we save it to a scratch register that we push on the new stack and that we pop back later on irq return. This preserves the whole frame chain without spurious return addresses in the middle and drops the need for the horrid fixup_bp_irq_link() workaround. And finally irqs that interrupt softirq are sanely unwinded. Before: 99.81% perf [kernel.kallsyms] [k] perf_pending_event | --- perf_pending_event irq_work_run smp_irq_work_interrupt irq_work_interrupt | |--41.60%-- __read | | | |--99.90%-- create_worker | | bench_sched_messaging | | cmd_bench | | run_builtin | | main | | __libc_start_main | --0.10%-- [...] After: 1.64% swapper [kernel.kallsyms] [k] perf_pending_event | --- perf_pending_event irq_work_run smp_irq_work_interrupt irq_work_interrupt | |--95.00%-- arch_irq_work_raise | irq_work_queue | __perf_event_overflow | perf_swevent_overflow | perf_swevent_event | perf_tp_event | perf_trace_softirq | __do_softirq | call_softirq | do_softirq | irq_exit | | | |--73.68%-- smp_apic_timer_interrupt | | apic_timer_interrupt | | | | | |--96.43%-- amd_e400_idle | | | cpu_idle | | | start_secondary Signed-off-by:Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jan Beulich <JBeulich@novell.com>
-
Frederic Weisbecker authored
The unwinder backlink in interrupt entry is very useless. It's actually not part of the stack frame chain and thus is never used. Signed-off-by: -
Frederic Weisbecker authored
Just for clarity in the code. Have a first block that handles the frame pointer and a separate one that handles pt_regs pointer and its use. Signed-off-by: -
Frederic Weisbecker authored
The save_regs function that saves the regs on low level irq entry is complicated because of the fact it changes its stack in the middle and also because it manipulates data allocated in the caller frame and accesses there are directly calculated from callee rsp value with the return address in the middle of the way. This complicates the static stack offsets calculation and require more dynamic ones. It also needs a save/restore of the function's return address. To simplify and optimize this, turn save_regs() into a macro. Signed-off-by: -
Frederic Weisbecker authored
When regs are passed to dump_stack(), we fetch the frame pointer from the regs but the stack pointer is taken from the current frame. Thus the frame and stack pointers may not come from the same context. For example this can result in the unwinder to think the context is in irq, due to the current value of the stack, but the frame pointer coming from the regs points to a frame from another place. It then tries to fix up the irq link but ends up dereferencing a random frame pointer that doesn't belong to the irq stack: [ 9131.706906] ------------[ cut here ]------------ [ 9131.707003] WARNING: at arch/x86/kernel/dumpstack_64.c:129 dump_trace+0x2aa/0x330() [ 9131.707003] Hardware name: AMD690VM-FMH [ 9131.707003] Perf: bad frame pointer = 0000000000000005 in callchain [ 9131.707003] Modules linked in: [ 9131.707003] Pid: 1050, comm: perf Not tainted 3.0.0-rc3+ #181 [ 9131.707003] Call Trace: [ 9131.707003] <IRQ> [<ffffffff8104bd4a>] warn_slowpath_common+0x7a/0xb0 [ 9131.707003] [<ffffffff8104be21>] warn_slowpath_fmt+0x41/0x50 [ 9131.707003] [<ffffffff8178b873>] ? bad_to_user+0x6d/0x10be [ 9131.707003] [<ffffffff8100c2da>] dump_trace+0x2aa/0x330 [ 9131.707003] [<ffffffff810107d3>] ? native_sched_clock+0x13/0x50 [ 9131.707003] [<ffffffff8101b164>] perf_callchain_kernel+0x54/0x70 [ 9131.707003] [<ffffffff810d391f>] perf_prepare_sample+0x19f/0x2a0 [ 9131.707003] [<ffffffff810d546c>] __perf_event_overflow+0x16c/0x290 [ 9131.707003] [<ffffffff810d5430>] ? __perf_event_overflow+0x130/0x290 [ 9131.707003] [<ffffffff810107d3>] ? native_sched_clock+0x13/0x50 [ 9131.707003] [<ffffffff8100fbb9>] ? sched_clock+0x9/0x10 [ 9131.707003] [<ffffffff810752e5>] ? T.375+0x15/0x90 [ 9131.707003] [<ffffffff81084da4>] ? trace_hardirqs_on_caller+0x64/0x180 [ 9131.707003] [<ffffffff810817bd>] ? trace_hardirqs_off+0xd/0x10 [ 9131.707003] [<ffffffff810d5764>] perf_event_overflow+0x14/0x20 [ 9131.707003] [<ffffffff810d588c>] perf_swevent_hrtimer+0x11c/0x130 [ 9131.707003] [<ffffffff817821a1>] ? error_exit+0x51/0xb0 [ 9131.707003] [<ffffffff81072e93>] __run_hrtimer+0x83/0x1e0 [ 9131.707003] [<ffffffff810d5770>] ? perf_event_overflow+0x20/0x20 [ 9131.707003] [<ffffffff81073256>] hrtimer_interrupt+0x106/0x250 [ 9131.707003] [<ffffffff812a3bfd>] ? trace_hardirqs_off_thunk+0x3a/0x3c [ 9131.707003] [<ffffffff81024833>] smp_apic_timer_interrupt+0x53/0x90 [ 9131.707003] [<ffffffff81789053>] apic_timer_interrupt+0x13/0x20 [ 9131.707003] <EOI> [<ffffffff817821a1>] ? error_exit+0x51/0xb0 [ 9131.707003] [<ffffffff8178219c>] ? error_exit+0x4c/0xb0 [ 9131.707003] ---[ end trace b2560d4876709347 ]--- Fix this by simply taking the stack pointer from regs->sp when regs are provided. Signed-off-by: -
Frederic Weisbecker authored
In order to prepare for fetching the stack pointer from the regs when possible in dump_trace() instead of taking the local one, save the current stack pointer in perf live regs saving. Signed-off-by:
-
- 01 Jul, 2011 22 commits
-
-
Peter Zijlstra authored
The patch a8b0ca17 ("perf: Remove the nmi parameter from the swevent and overflow interface") missed a spot in the ppc hw_breakpoint code, fix this up so things compile again. Reported-by:
Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/n/tip-09pfip95g88s70iwkxu6nnbt@git.kernel.orgSigned-off-by:
-
Zhengyu He authored
Previously, when you want perf-stat to output the statistics in csv mode, no information of the noise will be printed out. For example right now we output this --repeat information: ./perf stat -r3 -x, sleep 1 1.164789,task-clock 8,context-switches 0,CPU-migrations 219,page-faults 3337800,cycles With this patch, the output will be appended with an additional entry for the noise value: ./perf stat -r3 -x, sleep 1 1.164789,task-clock,3.75% 8,context-switches,75.00% 0,CPU-migrations,100.00% 219,page-faults,0.00% 3337800,cycles,3.36% Signed-off-by:
Zhengyu He <zhengyuh@google.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Stephane Eranian <eranian@google.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Frederic Weisbecker <fweisbec@gmail.com> Link: http://lkml.kernel.org/r/1308861942-4945-1-git-send-email-zhengyuh@google.comSigned-off-by:
-
Ingo Molnar authored
Merge branch 'perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing into perf/core
-
Avi Kivity authored
KVM needs one-shot samples, since a PMC programmed to -X will fire after X events and then again after 2^40 events (i.e. variable period). Signed-off-by:
Avi Kivity <avi@redhat.com> Signed-off-by:
-
Avi Kivity authored
The v1 PMU does not have any fixed counters. Using the v2 constraints, which do have fixed counters, causes an additional choice to be present in the weight calculation, but not when actually scheduling the event, leading to an event being not scheduled at all. Signed-off-by:
-
Avi Kivity authored
The perf_event overflow handler does not receive any caller-derived argument, so many callers need to resort to looking up the perf_event in their local data structure. This is ugly and doesn't scale if a single callback services many perf_events. Fix by adding a context parameter to perf_event_create_kernel_counter() (and derived hardware breakpoints APIs) and storing it in the perf_event. The field can be accessed from the callback as event->overflow_handler_context. All callers are updated. Signed-off-by:
-
Peter Zijlstra authored
Add a NODE level to the generic cache events which is used to measure local vs remote memory accesses. Like all other cache events, an ACCESS is HIT+MISS, if there is no way to distinguish between reads and writes do reads only etc.. The below needs filling out for !x86 (which I filled out with unsupported events). I'm fairly sure ARM can leave it like that since it doesn't strike me as an architecture that even has NUMA support. SH might have something since it does appear to have some NUMA bits. Sparc64, PowerPC and MIPS certainly want a good look there since they clearly are NUMA capable. Signed-off-by:
-
Peter Zijlstra authored
Since the OFFCORE registers are fully symmetric, try the other one when the specified one is already in use. Signed-off-by:
-
Stephane Eranian authored
This patch adds Intel Sandy Bridge offcore_response support by providing the low-level constraint table for those events. On Sandy Bridge, there are two offcore_response events. Each uses its own dedictated extra register. But those registers are NOT shared between sibling CPUs when HT is on unlike Nehalem/Westmere. They are always private to each CPU. But they still need to be controlled within an event group. All events within an event group must use the same value for the extra MSR. That's not controlled by the second patch in this series. Furthermore on Sandy Bridge, the offcore_response events have NO counter constraints contrary to what the official documentation indicates, so drop the events from the contraint table. Signed-off-by:
Stephane Eranian <eranian@google.com> Signed-off-by:
-
Stephane Eranian authored
The validate_group() function needs to validate events with extra shared regs. Within an event group, only events with the same value for the extra reg can co-exist. This was not checked by validate_group() because it was missing the shared_regs logic. This patch changes the allocation of the fake cpuc used for validation to also point to a fake shared_regs structure such that group events be properly testing. It modifies __intel_shared_reg_get_constraints() to use spin_lock_irqsave() to avoid lockdep issues. Signed-off-by:
-
Stephane Eranian authored
This patch improves the code managing the extra shared registers used for offcore_response events on Intel Nehalem/Westmere. The idea is to use static allocation instead of dynamic allocation. This simplifies greatly the get and put constraint routines for those events. The patch also renames per_core to shared_regs because the same data structure gets used whether or not HT is on. When HT is off, those events still need to coordination because they use a extra MSR that has to be shared within an event group. Signed-off-by:
-
Peter Zijlstra authored
Since only samples call perf_output_sample() its much saner (and more correct) to put the sample logic in there than in the perf_output_begin()/perf_output_end() pair. Saves a useless argument, reduces conditionals and shrinks struct perf_output_handle, win! Signed-off-by:
-
Peter Zijlstra authored
The nmi parameter indicated if we could do wakeups from the current context, if not, we would set some state and self-IPI and let the resulting interrupt do the wakeup. For the various event classes: - hardware: nmi=0; PMI is in fact an NMI or we run irq_work_run from the PMI-tail (ARM etc.) - tracepoint: nmi=0; since tracepoint could be from NMI context. - software: nmi=[0,1]; some, like the schedule thing cannot perform wakeups, and hence need 0. As one can see, there is very little nmi=1 usage, and the down-side of not using it is that on some platforms some software events can have a jiffy delay in wakeup (when arch_irq_work_raise isn't implemented). The up-side however is that we can remove the nmi parameter and save a bunch of conditionals in fast paths. Signed-off-by: -
Cyrill Gorcunov authored
Due to restriction and specifics of Netburst PMU we need a separated event for NMI watchdog. In particular every Netburst event consumes not just a counter and a config register, but also an additional ESCR register. Since ESCR registers are grouped upon counters (i.e. if ESCR is occupied for some event there is no room for another event to enter until its released) we need to pick up the "least" used ESCR (or the most available one) for nmi-watchdog purposes -- so MSR_P4_CRU_ESCR2/3 was chosen. With this patch nmi-watchdog and perf top should be able to run simultaneously. Signed-off-by:
Cyrill Gorcunov <gorcunov@openvz.org> CC: Lin Ming <ming.m.lin@intel.com> CC: Arnaldo Carvalho de Melo <acme@redhat.com> CC: Frederic Weisbecker <fweisbec@gmail.com> Tested-and-reviewed-by:
Don Zickus <dzickus@redhat.com> Tested-and-reviewed-by:
-
Eric B Munson authored
The event tracing infrastructure exposes two timers which should be updated each time the value of the counter is updated. Currently, these counters are only updated when userspace calls read() on the fd associated with an event. This means that counters which are read via the mmap'd page exclusively never have their timers updated. This patch adds ensures that the timers are updated each time the values in the mmap'd page are updated. Signed-off-by:
Eric B Munson <emunson@mgebm.net> Signed-off-by:
-
Eric B Munson authored
Take the timer calculation from perf_output_read and move it to a helper function for any place that needs timer values but cannot take the ctx->lock. Signed-off-by:
-
Eric B Munson authored
Signed-off-by:
-
Richard Kennedy authored
Reorder perf_event_context to remove 8 bytes of 64 bit alignment padding shrinking its size to 192 bytes, allowing it to fit into a smaller slab and use one fewer cache lines. Signed-off-by:
Richard Kennedy <richard@rsk.demon.co.uk> Signed-off-by:
-
Vince Weaver authored
Since 2.6.36 (specifically commit d57e34fd ("perf: Simplify the ring-buffer logic: make perf_buffer_alloc() do everything needed"), the perf_buffer_init_code() has been mis-setting the buffer watermark if perf_event_attr.wakeup_events has a non-zero value. This is because perf_event_attr.wakeup_events is a union with perf_event_attr.wakeup_watermark. This commit re-enables the check for perf_event_attr.watermark being set before continuing with setting a non-default watermark. This bug is most noticable when you are trying to use PERF_IOC_REFRESH with a value larger than one and perf_event_attr.wakeup_events is set to one. In this case the buffer watermark will be set to 1 and you will get extraneous POLL_IN overflows rather than POLL_HUP as expected. [ avoid using attr.wakeup_events when attr.watermark is set ] Signed-off-by:
Vince Weaver <vweaver1@eecs.utk.edu> Signed-off-by:
-
Peter Zijlstra authored
Commit e360adbe ("irq_work: Add generic hardirq context callbacks") fouled up the Alpha bit, not properly naming the arch specific function that raises the 'self-IPI'. Signed-off-by:
-
Peter Zijlstra authored
Commit e360adbe ("irq_work: Add generic hardirq context callbacks") fouled up the ppc bit, not properly naming the arch specific function that raises the 'self-IPI'. Cc: Huang Ying <ying.huang@intel.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Anton Blanchard <anton@samba.org> Cc: Eric B Munson <emunson@mgebm.net> Cc: stable@kernel.org # 37+ Signed-off-by:
-
Ingo Molnar authored
Merge reason: Pick up the latest fixes. Signed-off-by:
-
- 29 Jun, 2011 6 commits
-
-
Frederic Weisbecker authored
We don't need to display the parent field if the parent sorting machinery is only used for parent filtering (as in "-p foo"). However if parent filtering is used in combination with explicit parent sorting ( -s parent), we want to display it. Result with: perf report -p kernel_thread -s parent Before: # Overhead Parent symbol # ........ ............. # 0.07% | --- ioread8 ata_sff_check_status ata_sff_tf_load ata_sff_qc_issue ata_bmdma_qc_issue ata_qc_issue ata_scsi_translate ata_scsi_queuecmd scsi_dispatch_cmd scsi_request_fn __blk_run_queue __make_request generic_make_request submit_bio submit_bh journal_submit_commit_record jbd2_journal_commit_transaction kjournald2 kthread kernel_thread_helpe After: # Overhead Parent symbol # ........ ............. # 0.07% kernel_thread_helper | --- ioread8 ata_sff_check_status ata_sff_tf_load ata_sff_qc_issue ata_bmdma_qc_issue ata_qc_issue ata_scsi_translate ata_scsi_queuecmd scsi_dispatch_cmd scsi_request_fn __blk_run_queue __make_request generic_make_request submit_bio submit_bh journal_submit_commit_record jbd2_journal_commit_transaction kjournald2 kthread kernel_thread_helper Signed-off-by: -
Frederic Weisbecker authored
So that the parent sort dimension can be registered twice: once if we add it as an explicit sort dimension (-s parent) and twice if we request a parent filter (-p foo). We'll have only one parent sort dimension in the end but this allows to override the default parent filter with we gave in "-p" option. The goal of this is to prepare to allow the use of "-s parent" and "-p foo" at the same time, ie: sort by filtered parent. Signed-off-by: -
Frederic Weisbecker authored
As for newt ui, don't display entries that have been marked as ignored. The practical current effect of this is to make parent filtering really working. Before, entries that were ignored were given a null parent but were still displayed. This resulted in some weird effects: # Overhead Command Shared Object Symbol # ........ ........... ................. ............ # ^A | --- __lock_acquire | |--95.97%-- lock_acquire | | | |--30.75%-- _raw_spin_lock Discard these from the stdio display. Signed-off-by: -
Frederic Weisbecker authored
These are probably some old leftovers. Signed-off-by: -
Frederic Weisbecker authored
These don't need to be globally visible. Signed-off-by: -
Sam Liao authored
Add "caller/callee" option to support inverted butterfly report, in the inverted report (with caller option), the call graph start from the callee's ancestor. Users can use such view to catch system's performance bottleneck from a sysprof like view. Using this option with specified sort order like pid gives us high level view of call graph statistics. Also add "-G" alias for inverted call graph. Signed-off-by:
Sam Liao <phyomh@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Stephane Eranian <eranian@google.com> Cc: David Ahern <dsahern@gmail.com> Signed-off-by:
-
- 28 Jun, 2011 1 commit
-
-
Linus Torvalds authored
-
