Problem and motivation: Once a breakpoint perf event (PERF_TYPE_BREAKPOINT)
is created, there is no flexibility to change the breakpoint type
(bp_type), breakpoint address (bp_addr), or breakpoint length (bp_len). The
only option is to close the perf event and configure a new breakpoint
event. This inflexibility has a significant performance overhead. For
example, sampling-based, lightweight performance profilers (and also
concurrency bug detection tools), monitor different addresses for a short
duration using PERF_TYPE_BREAKPOINT and change the address (bp_addr) to
another address or change the kind of breakpoint (bp_type) from "write" to
a "read" or vice-versa or change the length (bp_len) of the address being
monitored. The cost of these modifications is prohibitive since it involves
unmapping the circular buffer associated with the perf event, closing the
perf event, opening another perf event and mmaping another circular buffer.
Solution: The new ioctl flag for perf events,
PERF_EVENT_IOC_MODIFY_ATTRIBUTES, introduced in this patch takes a pointer
to a struct perf_event_attr as an argument to update an old breakpoint
event with new address, type, and size. This facility allows retaining a
previous mmaped perf events ring buffer and avoids having to close and
reopen another perf event.
This patch supports only changing PERF_TYPE_BREAKPOINT event type; future
implementations can extend this feature. The patch replicates some of its
functionality of modify_user_hw_breakpoint() in
kernel/events/hw_breakpoint.c. modify_user_hw_breakpoint cannot be called
directly since perf_event_ctx_lock() is already held in _perf_ioctl().
Evidence: Experiments show that the baseline (not able to modify an already
created breakpoint) costs an order of magnitude (~10x) more than the
suggested optimization (having the ability to dynamically modifying a
configured breakpoint via ioctl). When the breakpoints typically do not
trap, the speedup due to the suggested optimization is ~10x; even when the
breakpoints always trap, the speedup is ~4x due to the suggested
optimization.
Testing: tests posted at
https://github.com/linux-contrib/perf_event_modify_bp demonstrate the
performance significance of this patch. Tests also check the functional
correctness of the patch.
Signed-off-by: Milind Chabbi <chabbi.milind@gmail.com>
[ Using modify_user_hw_breakpoint_check function. ]
[ Reformated PERF_EVENT_IOC_*, so the values are all in one column. ]
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oleg Nesterov <onestero@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180312134548.31532-8-jolsa@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Mark reported his arm64 perf fuzzer runs sometimes splat like:
armv8pmu_read_counter+0x1e8/0x2d8
armpmu_event_update+0x8c/0x188
armpmu_read+0xc/0x18
perf_output_read+0x550/0x11e8
perf_event_read_event+0x1d0/0x248
perf_event_exit_task+0x468/0xbb8
do_exit+0x690/0x1310
do_group_exit+0xd0/0x2b0
get_signal+0x2e8/0x17a8
do_signal+0x144/0x4f8
do_notify_resume+0x148/0x1e8
work_pending+0x8/0x14
which asserts that we only call pmu::read() on ACTIVE events.
The above callchain does:
perf_event_exit_task()
perf_event_exit_task_context()
task_ctx_sched_out() // INACTIVE
perf_event_exit_event()
perf_event_set_state(EXIT) // EXIT
sync_child_event()
perf_event_read_event()
perf_output_read()
perf_output_read_group()
leader->pmu->read()
Which results in doing a pmu::read() on an !ACTIVE event.
I _think_ this is 'new' since we added attr.inherit_stat, which added
the perf_event_read_event() to the exit path, without that
perf_event_read_output() would only trigger from samples and for
@event to trigger a sample, it's leader _must_ be ACTIVE too.
Still, adding this check makes it consistent with the @sub case for
the siblings.
Reported-and-Tested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
- Support to display the IPC/Cycle in 'annotate' TUI, for systems
where this info can be obtained, like Intel's >= Skylake (Jin Yao)
- Support wildcards on PMU name in dynamic PMU events (Agustin Vega-Frias)
- Display pmu name when printing unmerged events in stat (Agustin Vega-Frias)
- Auto-merge PMU events created by prefix or glob match (Agustin Vega-Frias)
- Fix s390 'call' operations target function annotation (Thomas Richter)
- Handle s390 PC relative load and store instruction in the augmented
'annotate', code, used so far in the TUI modes of 'perf report' and
'perf annotate' (Thomas Richter)
- Provide libtraceevent with a kernel symbol resolver, so that
symbols in tracepoint fields can be resolved when showing them in
tools such as 'perf report' (Wang YanQing)
- Refactor the cgroups code to look more like other code in tools/perf,
using cgroup__{put,get} for refcount operations instead of its
open-coded equivalent, breaking larger functions, etc (Arnaldo Carvalho de Melo)
- Implement support for the -G/--cgroup target in 'perf trace', allowing
strace like tracing (plus other events, backtraces, etc) for cgroups
(Arnaldo Carvalho de Melo)
- Update thread shortname in 'perf sched map' when the thread's COMM
changes (Changbin Du)
- refcount 'struct mem_info', for better sharing it over several
users, avoid duplicating structs and fixing crashes related to
use after free (Jiri Olsa)
- Display perf.data version, offsets in 'perf report --header' (Jiri Olsa)
- Record the machine's memory topology information in a perf.data
feature section, to be used by tools such as 'perf c2c' (Jiri Olsa)
- Fix output of forced groups in the header for 'perf report' --stdio
and --tui (Jiri Olsa)
- Better support llvm, clang, cxx make tests in the build process (Jiri Olsa)
- Streamline the 'struct perf_mmap' methods, storing some info in the
struct instead of passing it via various methods, shortening its
signatures (Kan Liang)
- Update the quipper perf.data parser library site information (Stephane Eranian)
- Correct perf's man pages title markers for asciidoctor (Takashi Iwai)
- Intel PT fixes and refactorings paving the way for implementing
support for AUX area sampling (Adrian Hunter)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There is a bug when reading event->count with large PEBS enabled.
Here is an example:
# ./read_count
0x71f0
0x122c0
0x1000000001c54
0x100000001257d
0x200000000bdc5
In fixed period mode, the auto-reload mechanism could be enabled for
PEBS events, but the calculation of event->count does not take the
auto-reload values into account.
Anyone who reads event->count will get the wrong result, e.g x86_pmu_read().
This bug was introduced with the auto-reload mechanism enabled since
commit:
851559e35f ("perf/x86/intel: Use the PEBS auto reload mechanism when possible")
Introduce intel_pmu_save_and_restart_reload() to calculate the
event->count only for auto-reload.
Since the counter increments a negative counter value and overflows on
the sign switch, giving the interval:
[-period, 0]
the difference between two consequtive reads is:
A) value2 - value1;
when no overflows have happened in between,
B) (0 - value1) + (value2 - (-period));
when one overflow happened in between,
C) (0 - value1) + (n - 1) * (period) + (value2 - (-period));
when @n overflows happened in between.
Here A) is the obvious difference, B) is the extension to the discrete
interval, where the first term is to the top of the interval and the
second term is from the bottom of the next interval and C) the extension
to multiple intervals, where the middle term is the whole intervals
covered.
The equation for all cases is:
value2 - value1 + n * period
Previously the event->count is updated right before the sample output.
But for case A, there is no PEBS record ready. It needs to be specially
handled.
Remove the auto-reload code from x86_perf_event_set_period() since
we'll not longer call that function in this case.
Based-on-code-from: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Fixes: 851559e35f ("perf/x86/intel: Use the PEBS auto reload mechanism when possible")
Link: http://lkml.kernel.org/r/1518474035-21006-2-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Large fixed period values could be truncated on Broadwell, for example:
perf record -e cycles -c 10000000000
Here the fixed period is 0x2540BE400, but the period which finally applied is
0x540BE400 - which is wrong.
The reason is that x86_pmu::limit_period() uses an u32 parameter, so the
high 32 bits of 'period' get truncated.
This bug was introduced in:
commit 294fe0f52a ("perf/x86/intel: Add INST_RETIRED.ALL workarounds")
It's safe to use u64 instead of u32:
- Although the 'left' is s64, the value of 'left' must be positive when
calling limit_period().
- bdw_limit_period() only modifies the lowest 6 bits, it doesn't touch
the higher 32 bits.
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Fixes: 294fe0f52a ("perf/x86/intel: Add INST_RETIRED.ALL workarounds")
Link: http://lkml.kernel.org/r/1519926894-3520-1-git-send-email-kan.liang@linux.intel.com
[ Rewrote unacceptably bad changelog. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
S390 has several load and store instructions with target operand
addressing relative to the program counter, for example lrl, lgrl, strl,
stgrl.
These instructions are handled similar to x86. Objdump output displays
those instructions as:
9595c: c4 2d 00 09 9c 54 lgrl %r7,1c8540 <mp_+0x60>
This output is parsed (like on x86) and perf annotate shows those lines
as:
lgrl %r7,mp_+0x60
This patch handles the s390 specific instruction parsing for PC relative
load and store instructions.
Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20180308120913.14802-1-tmricht@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
So that beautifiers wanting to resolve kernel function addresses to
names can do its work, and when we use "perf report" for output of "perf
kmem record", we will get kernel symbol output.
This patch affect the output of "perf report" for the record data
generated by "perf kmem record" looks like below:
Before patch:
0.01% call_site=ffffffff814e5828 ptr=0x99bb000 bytes_req=3616 bytes_alloc=4096 gfp_flags=GFP_ATOMIC
0.01% call_site=ffffffff81370b87 ptr=0x428a3060 bytes_req=32 bytes_alloc=32 gfp_flags=GFP_KERNEL|GFP_ZERO
After patch:
0.01% (aa_alloc_task_context+0x27) call_site=ffffffff81370b87 ptr=0x428a3060 bytes_req=32 bytes_alloc=32 gfp_flags=GFP_KERNEL|GFP_ZERO
0.01% (__tty_buffer_request_room+0x88) call_site=ffffffff814e5828 ptr=0x99bb000 bytes_req=3616 bytes_alloc=4096 gfp_flags=GFP_ATOMIC
Signed-off-by: Wang YanQing <udknight@gmail.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180308032850.GA12383@udknight-ThinkPad-E550
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adding MEM_TOPOLOGY feature to perf data file,
that will carry physical memory map and its
node assignments.
The format of data in MEM_TOPOLOGY is as follows:
0 - version | for future changes
8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
16 - count | number of nodes
For each node we store map of physical indexes for
each node:
32 - node id | node index
40 - size | size of bitmap
48 - bitmap | bitmap of memory indexes that belongs to node
| /sys/devices/system/node/node<NODE>/memory<INDEX>
The MEM_TOPOLOGY could be displayed with following
report command:
$ perf report --header-only -I
...
# memory nodes (nr 1, block size 0x8000000):
# 0 [7G]: 0-23,32-69
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-8-jolsa@kernel.org
[ Rename 'index' to 'idx', as this breaks the build in rhel5, 6 and other systems where this is used by glibc headers ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It's passed along several hists entries in --hierarchy mode, so it's
better we keep track of it.
The current fail I see is that it gets removed in hierarchy --mem-mode
mode, where it's shared in the different hierarchies, but removed from
the template hist entry, so the report crashes.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180307155020.32613-6-jolsa@kernel.org
[ Rename mem_info__aloc() to mem_info__new(), to fix the typo and use the convention for constructors ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
'perf annotate' displays function call assembler instructions with a
right arrow. Hitting enter on this line/instruction causes the browser
to disassemble this target function and show it on the screen. On s390
this results in an error message 'The called function was not found.'
The function call assembly line parsing does not handle the s390 bras
and brasl instructions. Function call__parse expects the target as first
operand:
callq e9140 <__fxstat>
S390 has a register number as first operand:
brasl %r14,41d60 <abort>
Therefore the target addresses on s390 are always zero which is an
invalid address.
Introduce a s390 specific call parsing function which skips the first
operand on s390.
Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20180307134325.96106-1-tmricht@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Intel PT code already has some preparation for AUX area sampling mode.
However the implementation has changed from the first proposal and one
of the side-effects is that it will not be impossible to support snapshot
mode and sampling mode at the same time.
Although there are no plans to support it, let validation (not yet
implemented) control whether it is allowed rather than low-level
functions.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-9-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
intel_pt_get_trace() fixes overlaps between the current buffer and the
previous buffer ('old_buffer').
However the previous buffer might not have had usable data (no PSB) so
the comparison must be made against the previous buffer that had usable
data.
Tidy that by keeping a pointer for that purpose in struct intel_pt_queue.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1520431349-30689-8-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
timestamp_insn_cnt is used to estimate the timestamp based on the number of
instructions since the last known timestamp.
If the estimate is not accurate enough decoding might not be correctly
synchronized with side-band events causing more trace errors.
However there are always timestamps following an overflow, so the
estimate is not needed and can indeed result in more errors.
Suppress the estimate by setting timestamp_insn_cnt to zero.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1520431349-30689-5-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
