There use to be ring buffer integrity checks after updating the
size of the ring buffer. But now that the ring buffer can modify
the size while the system is running, the integrity checks were
removed, as they require the ring buffer to be disabed to perform
the check.
Move the integrity check to the reading of the ring buffer via the
iterator reads (the "trace" file). As reading via an iterator requires
disabling the ring buffer, it is a perfect place to have it.
If the ring buffer happens to be disabled when updating the size,
we still perform the integrity check.
Cc: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
This patch adds the capability to remove pages from a ring buffer
without destroying any existing data in it.
This is done by removing the pages after the tail page. This makes sure
that first all the empty pages in the ring buffer are removed. If the
head page is one in the list of pages to be removed, then the page after
the removed ones is made the head page. This removes the oldest data
from the ring buffer and keeps the latest data around to be read.
To do this in a non-racey manner, tracing is stopped for a very short
time while the pages to be removed are identified and unlinked from the
ring buffer. The pages are freed after the tracing is restarted to
minimize the time needed to stop tracing.
The context in which the pages from the per-cpu ring buffer are removed
runs on the respective CPU. This minimizes the events not traced to only
NMI trace contexts.
Link: http://lkml.kernel.org/r/1336096792-25373-1-git-send-email-vnagarnaik@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
On gcc 4.5 the function tracing_mark_write() would give a warning
of page2 being uninitialized. This is due to a bug in gcc because
the logic prevents page2 from being used uninitialized, and
gcc 4.6+ does not complain (correctly).
Instead of adding a "unitialized" around page2, which could show
a bug later on, I combined page1 and page2 into an array map_pages[].
This binds the two and the two are modified according to nr_pages
(what gcc 4.5 seems to ignore). This no longer gives a warning with
gcc 4.5 nor with gcc 4.6.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Fixing i386 allnoconfig built errors:
arch/x86/built-in.o: In function `amd_pmu_hw_config':
perf_event_amd.c:(.text+0xc3e1): undefined reference to `get_ibs_caps'
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Arjan & Linus Annotation Edition
- Fix indirect calls beautifier, reported by Linus.
- Use the objdump comments to nuke specificities about how access to a well
know variable is encoded, suggested by Linus.
- Show the number of places that jump to a target, requested by Arjan.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Fixes and improvements for perf/core:
- perf_target: abstraction for --uid, --pid, --tid, --cpu, --all-cpus handling,
eliminating code duplicated in the tools, having constraints that apply to
all of them, from Namhyung Kim
- Fixes for handling fallback to cpu-clock on PPC, from David Ahern
- Fix for processing events with unknown size, from Jiri Olsa
- Compilation fix on 32-bit, from Jiri Olsa
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The 'perf top' command falls back to cpu-clock if the H/W cycles event
is not supported, but the event name is not updated leading to a
misleading header:
PerfTop: 8 irqs/sec kernel:75.0% exact: 0.0% [1000Hz cycles], ...
Update the event name when the event type is changed so that the
header displays correctly:
PerfTop: 794 irqs/sec kernel:100.0% exact: 0.0% [1000Hz cpu-clock], ...
Signed-off-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1336495789-58420-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf stat on PPC currently fails to run:
$ perf stat -- sleep 1
Error: open_counter returned with 6 (No such device or address). /bin/dmesg may provide additional information.
Fatal: Not all events could be opened.
The problem is that until 2.6.37 (behavior changed with commit b0a873e)
perf on PPC returns ENXIO when hw_perf_event_init() fails. With this
patch we get the expected behavior:
$ perf stat -v -- sleep 1
cycles event is not supported by the kernel.
stalled-cycles-frontend event is not supported by the kernel.
stalled-cycles-backend event is not supported by the kernel.
instructions event is not supported by the kernel.
branches event is not supported by the kernel.
branch-misses event is not supported by the kernel.
...
Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336490956-57145-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf-record on PPC is not falling back to cpu-clock:
$ perf record -ag -fo /tmp/perf.data -- sleep 1
Error: sys_perf_event_open() syscall returned with 6 (No such device or address). /bin/dmesg may provide additional information.
Fatal: No CONFIG_PERF_EVENTS=y kernel support configured?
The problem is that until 2.6.37 (behavior changed with commit b0a873e)
perf on PPC returns ENXIO when hw_perf_event_init() fails. With this
patch we get the expected behavior:
$ perf record -ag -fo /tmp/perf.data -v -- sleep 1
Old kernel, cannot exclude guest or host samples.
The cycles event is not supported, trying to fall back to cpu-clock-ticks
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.151 MB /tmp/perf.data (~6592 samples) ]
Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336490937-57106-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When disabling ibs there might be the case where hardware continuously
generates interrupts. This is described in erratum #420 (Instruction-
Based Sampling Engine May Generate Interrupt that Cannot Be Cleared).
To avoid this we must clear the counter mask first and then clear the
enable bit. This patch implements this.
See Revision Guide for AMD Family 10h Processors, Publication #41322.
Note: We now keep track of the last read ibs config value which is
then used to disable ibs. To update the config value we pass now a
pointer to the functions reading it.
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-11-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are cases where the remaining period is smaller than the minimal
possible value. In this case the counter is restarted with the minimal
period. This is of no use as the interrupt handler will trigger
immediately again and most likely hits itself. This biases the
results.
So, if the remaining period is within the min range, we better do not
restart the counter and instead trigger the overflow.
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-9-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch adds support for precise event sampling with IBS. There are
two counting modes to count either cycles or micro-ops. If the
corresponding performance counter events (hw events) are setup with
the precise flag set, the request is redirected to the ibs pmu:
perf record -a -e cpu-cycles:p ... # use ibs op counting cycle count
perf record -a -e r076:p ... # same as -e cpu-cycles:p
perf record -a -e r0C1:p ... # use ibs op counting micro-ops
Each ibs sample contains a linear address that points to the
instruction that was causing the sample to trigger. With ibs we have
skid 0. Thus, ibs supports precise levels 1 and 2. Samples are marked
with the PERF_EFLAGS_EXACT flag set. In rare cases the rip is invalid
when IBS was not able to record the rip correctly. Then the
PERF_EFLAGS_EXACT flag is cleared and the rip is taken from pt_regs.
V2:
* don't drop samples in precise level 2 if rip is invalid, instead
support the PERF_EFLAGS_EXACT flag
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20120502103309.GP18810@erda.amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Fixing profiling at a fixed frequency, in this case the freq value and
sample period was setup incorrectly. Since sampling periods are
adjusted we also allow periods that have lower 4 bits set.
Another fix is the setup of the hw counter: If we modify
hwc->sample_period, we also need to update hwc->last_period and
hwc->period_left.
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-5-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We always need to pass the last sample period to
perf_sample_data_init(), otherwise the event distribution will be
wrong. Thus, modifiyng the function interface with the required period
as argument. So basically a pattern like this:
perf_sample_data_init(&data, ~0ULL);
data.period = event->hw.last_period;
will now be like that:
perf_sample_data_init(&data, ~0ULL, event->hw.last_period);
Avoids unininitialized data.period and simplifies code.
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-3-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The ftrace_disable_cpu() and ftrace_enable_cpu() functions were
needed back before the ring buffer was lockless. Now that the
ring buffer is lockless (and has been for some time), these functions
serve no purpose, and unnecessarily slow down operations of the tracer.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
It's appropriate to use __seq_open_private interface to open
some of trace seq files, because it covers all steps we are
duplicating in tracing code - zallocating the iterator and
setting it as seq_file's private.
Using this for following files:
trace
available_filter_functions
enabled_functions
Link: http://lkml.kernel.org/r/1335342219-2782-5-git-send-email-jolsa@redhat.com
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
[
Fixed warnings for:
kernel/trace/trace.c: In function '__tracing_open':
kernel/trace/trace.c:2418:11: warning: unused variable 'ret' [-Wunused-variable]
kernel/trace/trace.c:2417:19: warning: unused variable 'm' [-Wunused-variable]
]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Perf annotate browser improvements:
- Get back the line separating the overheads from the disassembly, requested by
Peter Zijlstra, Linus agreed now that it is a solid line and more column real
state was harvested. Also it has the jump->arrow lines separated from it by
the address/jump target column.
- Don't change asm line color when toggling source code view. Requested by
Peter Zijlstra.
Current snapshot:
avtab_search_node
│ push %rbp
│ mov %rsp,%rbp
│ → callq mcount
│ movzwl 0x6(%rsi),%edx
│ and $0x7fff,%dx
│ test %rdi,%rdi
│ ↓ jne 20
0.42 │17:┌─→xor %eax,%eax
│19:│ leaveq
0.42 │ │← retq
│ │ nopl 0x0(%rax,%rax,1)
│20:│ mov (%rdi),%rax
0.08 │ │ test %rax,%rax
│ └──je 17
│ movzwl (%rsi),%ecx
│ movzwl 0x2(%rsi),%r9d
│ movzwl 0x4(%rsi),%r8d
│ movzwl %cx,%esi
│ movzwl %r9w,%r10d
│ shl $0x9,%esi
│ lea (%rsi,%r10,4),%esi
│ lea (%r8,%rsi,1),%esi
│ and 0x10(%rdi),%si
│ movzwl %si,%esi
│ mov (%rax,%rsi,8),%rax
1.01 │ test %rax,%rax
│ ↑ je 19
│ nopw 0x0(%rax,%rax,1)
3.19 │60: cmp %cx,(%rax)
│ ↓ jne 7e
0.08 │ cmp %r9w,0x2(%rax)
│ ↓ jne 7e
│ cmp %r8w,0x4(%rax)
│ ↓ jne 79
│ test %dx,0x6(%rax)
│ ↑ jne 19
│79: cmp %r8w,0x4(%rax)
83.45 │7e: ↑ ja 17
3.36 │ mov 0x10(%rax),%rax
7.98 │ test %rax,%rax
│ ↑ jne 60
│ leaveq
│ ← retq
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Currently, 'perf record -- sleep 1' creates a cpu map for all online
cpus since it turns out calling cpu_map__new(NULL). Fix it.
Also it is guaranteed that cpu_list is NULL if PID/TID is given by
calling perf_target__validate(), so we can make the conditional bit
simpler.
This also fixes perf test 7 (Validate) failure on my 6 core machine:
$ cat /sys/devices/system/cpu/online
0-11
$ ./perf test -v 7
7: Validate PERF_RECORD_* events & perf_sample fields:
--- start ---
perf_evlist__mmap: Operation not permitted
---- end ----
Validate PERF_RECORD_* events & perf_sample fields: FAILED!
Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-3-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
