Pull perf/core callchain fixes and improvements from Arnaldo Carvalho de Melo <acme@redhat.com:
User visible changes:
- Make --percent-limit apply to callchains also and fix some bugs
related to --percent-limit (Namhyung Kim)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull perf tooling changes from Arnaldo Carvalho de Melo:
New features:
- Port 'perf kvm stat' to PowerPC (Hemant Kumar)
Infrastructure changes:
- Use the 'feature-dump' target to do the feature checks just once and then
add code to reuse that in the tests/make makefile, speeding up the
'make -C tools/perf build-test' target (Wang Nan)
- Reduce the number of tests the 'build-test' target do to those that don't
pollute the source tree (Arnaldo Carvalho de Melo)
- Improve the output of the build tests a bit by aligning the name of the
tests, more can be done to filter out uninteresting info in the output
(Arnaldo Carvalho de Melo)
- Add perf_evlist pointer to *info_priv_size(), more prep work for
supporting the coresight architecture (Mathieu Poirier)
- Improve the 'perf test bp_signal' test (Wang Nan)
- Check environment before starting the BPF 'perf test', so that we can just
'Skip' older kernels instead of 'FAIL'ing them (Wang Nan)
- Fix cpumode of synthesized buildid event (Wang Nan)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
User visible changes:
- Rename the "colors.code" ~/.perfconfig variable to "colors.jump_arrows",
as it controls just the that UI element in the annotate browser (Taeung Song)
- Avoid trying to read ELF symtabs from device files, noticed while doing
memory profiling work (Jiri Olsa)
- Improve context detection when offering options in the hists browser,
i.e. some options don't make sense when the browser is not working with
a perf.data file ('perf top' mode), only in 'perf report' mode, like
scripting (Namhyung Kim)
Infrastructure changes:
- Elliminate duplication in the hists browser filter functions, getting the
common part into a function that receives callbacks for filtering by
DSO, thread, etc. (Namhyung Kim)
- Fix misleadingly indented assignment, found using
gcc6 -Wmisleading-indentation (Markus Trippelsdorf)
- Handle LLVM relocation oddities in libbpf, introducing a 'perf test' that
detects such problems and then fixing the problem, so that the test now
passes (Wang Nan)
- More improvements to the build infrastructure to allow reusing the
feature detection facilities (Wang Nan)
- Auto initialize the globals needed by cpu__max_{cpu,node}() routines
(Arnaldo Carvalho de Melo)
Documentation changes:
- Document the perf sysctls in Documentation/sysctl/kernel.txt (Ben Hutchings)
- Document a bunch more ~/.perfconfig knobs (Taeung Song)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
'perf probe' through debuginfo__find_probes() in util/probe-finder.c
checks for the functions' frame descriptions in either .eh_frame section
of an ELF or the .debug_frame.
The check is based on whether either one of these sections is present.
Depending on distro, toolchain defaults, architetcutre, build flags,
etc., CFI might be found in either .eh_frame and/or .debug_frame.
Sometimes, it may happen that, .eh_frame, even if present, may not be
complete and may miss some descriptions.
Therefore, to be sure, to find the CFI covering an address we will
always have to investigate both if available.
For e.g., in powerpc, this may happen:
$ gcc -g bin.c -o bin
$ objdump --dwarf ./bin
<1><145>: Abbrev Number: 7 (DW_TAG_subprogram)
<146> DW_AT_external : 1
<146> DW_AT_name : (indirect string, offset: 0x9e): main
<14a> DW_AT_decl_file : 1
<14b> DW_AT_decl_line : 39
<14c> DW_AT_prototyped : 1
<14c> DW_AT_type : <0x57>
<150> DW_AT_low_pc : 0x100007b8
If the .eh_frame and .debug_frame are checked for the same binary, we
will find that, .eh_frame (although present) doesn't contain a
description for "main" function.
But, .debug_frame has a description:
000000d8 00000024 00000000 FDE cie=00000000 pc=100007b8..10000838
DW_CFA_advance_loc: 16 to 100007c8
DW_CFA_def_cfa_offset: 144
DW_CFA_offset_extended_sf: r65 at cfa+16
...
Due to this (since, perf checks whether .eh_frame is present and goes on
searching for that address inside that frame), perf is unable to process
the probes:
# perf probe -x ./bin main
Failed to get call frame on 0x100007b8
Error: Failed to add events.
To avoid this issue, we need to check both the sections (.eh_frame and
.debug_frame), which is done in this patch.
Note that, we can always force everything into both .eh_frame and
.debug_frame by:
$ gcc bin.c -fasynchronous-unwind-tables -fno-dwarf2-cfi-asm -g -o bin
Signed-off-by: Hemant Kumar <hemant@linux.vnet.ibm.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Mark Wielaard <mjw@redhat.com>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1454426806-13974-1-git-send-email-hemant@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
intel_pt_process_auxtrace_info() creates a pt->unknown_thread thread
that eventually needs to be freed by the last thread__put() on it, when
its refcount hits zero, which may happen in
intel_pt_process_auxtrace_info() error handling path and triggers the
following segfault, which would happen as well at intel_pt_free, when
tools using this intel_pt codebase frees up resources:
# perf record -I -e intel_pt/tsc=1,noretcomp=1/u /bin/ls
0 a anaconda-ks.cfg bin perf.data perf.data.old perf-f23-bringup.todo
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.217 MB perf.data ]
#
# perf script -F event,comm,pid,tid,time,addr,ip,sym,dso,iregs
Samples for 'instructions:u' event do not have IREGS attribute set. Cannot print 'iregs' field.
intel_pt_synth_events: failed to synthesize 'instructions' event type
Segmentation fault (core dumped)
#
The problem is: there's a union in 'struct thread' combines a list_head
and a rb_node. The standard life cycle of a thread is: init rb_node in
the constructor, insert it into machine->threads rbtree using rb_node,
move it to machine->dead_threads using list_head, clean in the last
thread__put: list_del_init(&thread->node).
In the above command, it clean a thread before adding it into list,
causes the above segfault.
Since pt->unknown_thread will never live in an rbtree, initialize its
list node so that when list_del_init() is done on it we don't segfault.
After this patch:
# perf script -F event,comm,pid,tid,time,addr,ip,sym,dso,iregs
Samples for 'instructions:u' event do not have IREGS attribute set. Cannot print 'iregs' field.
intel_pt_synth_events: failed to synthesize 'instructions' event type
0x248 [0x88]: failed to process type: 70
#
Reported-by: Tong Zhang <ztong@vt.edu>
Reported-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Link: http://lkml.kernel.org/r/1454296865-19749-1-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull libnvdimm fixes from Dan Williams:
"1/ Fixes to the libnvdimm 'pfn' device that establishes a reserved
area for storing a struct page array.
2/ Fixes for dax operations on a raw block device to prevent pagecache
collisions with dax mappings.
3/ A fix for pfn_t usage in vm_insert_mixed that lead to a null
pointer de-reference.
These have received build success notification from the kbuild robot
across 153 configs and pass the latest ndctl tests"
* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
phys_to_pfn_t: use phys_addr_t
mm: fix pfn_t to page conversion in vm_insert_mixed
block: use DAX for partition table reads
block: revert runtime dax control of the raw block device
fs, block: force direct-I/O for dax-enabled block devices
devm_memremap_pages: fix vmem_altmap lifetime + alignment handling
libnvdimm, pfn: fix restoring memmap location
libnvdimm: fix mode determination for e820 devices
When there's only a single callchain, perf doesn't print its percentage
in front of the symbols. This is because it assumes that the percentage
is same as parents. But if a percent limit is applied, it's possible
that there are actually a couple of child nodes but only one of them is
shown. In this case it should display the percent to prevent
misunderstanding of its percentage is same as the parent's.
For example, let's see the following callchain.
$ perf report --no-children --percent-limit 0.01 --tui
...
- 0.06% sleep [kernel.vmlinux] [k] kmem_cache_alloc_trace
kmem_cache_alloc_trace
- perf_event_mmap
- 0.04% mmap_region
do_mmap_pgoff
- vm_mmap_pgoff
+ 0.02% sys_mmap_pgoff
+ 0.02% vm_mmap
+ 0.02% mprotect_fixup
Current code omits the percent if 'mmap_region' becomes the only node
when percent limit is set to 0.03%, its percent is not 0.06% but users
will assume it incorrectly.
Before:
$ perf report --no-children --percent-limit 0.03 --tui
...
0.06% sleep [kernel.vmlinux] [k] kmem_cache_alloc_trace
kmem_cache_alloc_trace
- perf_event_mmap
- mmap_region
do_mmap_pgoff
vm_mmap_pgoff
After:
$ perf report --no-children --percent-limit 0.03 --tui
...
0.06% sleep [kernel.vmlinux] [k] kmem_cache_alloc_trace
kmem_cache_alloc_trace
- perf_event_mmap
- 0.04% mmap_region
do_mmap_pgoff
vm_mmap_pgoff
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1453909257-26015-10-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When there's only a single callchain, perf doesn't print its percentage
in front of the symbols. This is because it assumes that the percentage
is same as parents. But if a percent limit is applied, it's possible
that there are actually a couple of child nodes but only one of them is
shown. In this case it should display the percent to prevent
misunderstanding of its percentage is same as the parent's.
For example, let's see the following callchain.
$ perf report -s comm --percent-limit 0.01 --stdio
...
9.95% swapper
|
|--7.57%--intel_idle
| cpuidle_enter_state
| cpuidle_enter
| call_cpuidle
| cpu_startup_entry
| |
| |--4.89%--start_secondary
| |
| --2.68%--rest_init
| start_kernel
| x86_64_start_reservations
| x86_64_start_kernel
|
|--0.15%--__schedule
| |
| |--0.13%--schedule
| | schedule_preempt_disable
| | cpu_startup_entry
| | |
| | |--0.09%--start_secondary
| | |
| | --0.04%--rest_init
| | start_kernel
| | x86_64_start_reservations
| | x86_64_start_kernel
| |
| --0.01%--schedule_preempt_disabled
| cpu_startup_entry
...
Current code omits the percent if 'intel_idle' becomes the only node
when percent limit is set to 0.5%, its percent is not 9.95% but users
will assume it incorrectly.
Before:
$ perf report --percent-limit 0.5 --stdio
...
9.95% swapper
|
---intel_idle
cpuidle_enter_state
cpuidle_enter
call_cpuidle
cpu_startup_entry
|
|--4.89%--start_secondary
|
--2.68%--rest_init
start_kernel
x86_64_start_reservations
x86_64_start_kernel
After:
$ perf report --percent-limit 0.5 --stdio
...
9.95% swapper
|
--7.57%--intel_idle
cpuidle_enter_state
cpuidle_enter
call_cpuidle
cpu_startup_entry
|
|--4.89%--start_secondary
|
--2.68%--rest_init
start_kernel
x86_64_start_reservations
x86_64_start_kernel
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1453909257-26015-7-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Fixes segmentation fault using, for instance:
(gdb) run record -I -e intel_pt/tsc=1,noretcomp=1/u /bin/ls
Starting program: /home/acme/bin/perf record -I -e intel_pt/tsc=1,noretcomp=1/u /bin/ls
Missing separate debuginfos, use: dnf debuginfo-install glibc-2.22-7.fc23.x86_64
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".
Program received signal SIGSEGV, Segmentation fault.
0 x00000000004b9ea5 in tracepoint_error (e=0x0, err=13, sys=0x19b1370 "sched", name=0x19a5d00 "sched_switch") at util/parse-events.c:410
(gdb) bt
#0 0x00000000004b9ea5 in tracepoint_error (e=0x0, err=13, sys=0x19b1370 "sched", name=0x19a5d00 "sched_switch") at util/parse-events.c:410
#1 0x00000000004b9fc5 in add_tracepoint (list=0x19a5d20, idx=0x7fffffffb8c0, sys_name=0x19b1370 "sched", evt_name=0x19a5d00 "sched_switch", err=0x0, head_config=0x0)
at util/parse-events.c:433
#2 0x00000000004ba334 in add_tracepoint_event (list=0x19a5d20, idx=0x7fffffffb8c0, sys_name=0x19b1370 "sched", evt_name=0x19a5d00 "sched_switch", err=0x0, head_config=0x0)
at util/parse-events.c:498
#3 0x00000000004bb699 in parse_events_add_tracepoint (list=0x19a5d20, idx=0x7fffffffb8c0, sys=0x19b1370 "sched", event=0x19a5d00 "sched_switch", err=0x0, head_config=0x0)
at util/parse-events.c:936
#4 0x00000000004f6eda in parse_events_parse (_data=0x7fffffffb8b0, scanner=0x19a49d0) at util/parse-events.y:391
#5 0x00000000004bc8e5 in parse_events__scanner (str=0x663ff2 "sched:sched_switch", data=0x7fffffffb8b0, start_token=258) at util/parse-events.c:1361
#6 0x00000000004bca57 in parse_events (evlist=0x19a5220, str=0x663ff2 "sched:sched_switch", err=0x0) at util/parse-events.c:1401
#7 0x0000000000518d5f in perf_evlist__can_select_event (evlist=0x19a3b90, str=0x663ff2 "sched:sched_switch") at util/record.c:253
#8 0x0000000000553c42 in intel_pt_track_switches (evlist=0x19a3b90) at arch/x86/util/intel-pt.c:364
#9 0x00000000005549d1 in intel_pt_recording_options (itr=0x19a2c40, evlist=0x19a3b90, opts=0x8edf68 <record+232>) at arch/x86/util/intel-pt.c:664
#10 0x000000000051e076 in auxtrace_record__options (itr=0x19a2c40, evlist=0x19a3b90, opts=0x8edf68 <record+232>) at util/auxtrace.c:539
#11 0x0000000000433368 in cmd_record (argc=1, argv=0x7fffffffde60, prefix=0x0) at builtin-record.c:1264
#12 0x000000000049bec2 in run_builtin (p=0x8fa2a8 <commands+168>, argc=5, argv=0x7fffffffde60) at perf.c:390
#13 0x000000000049c12a in handle_internal_command (argc=5, argv=0x7fffffffde60) at perf.c:451
#14 0x000000000049c278 in run_argv (argcp=0x7fffffffdcbc, argv=0x7fffffffdcb0) at perf.c:495
#15 0x000000000049c60a in main (argc=5, argv=0x7fffffffde60) at perf.c:618
(gdb)
Intel PT attempts to find the sched:sched_switch tracepoint but that seg
faults if tracefs is not readable, because the error reporting structure
is null, as errors are not reported when automatically adding
tracepoints. Fix by checking before using.
Committer note:
This doesn't take place in a kernel that supports
perf_event_attr.context_switch, that is the default way that will be
used for tracking context switches, only in older kernels, like 4.2, in
a machine with Intel PT (e.g. Broadwell) for non-priviledged users.
Further info from a similar patch by Wang:
The error is in tracepoint_error: it assumes the 'e' parameter is valid.
However, there are many situation a parse_event() can be called without
parse_events_error. See result of
$ grep 'parse_events(.*NULL)' ./tools/perf/ -r'
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Tong Zhang <ztong@vt.edu>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: stable@vger.kernel.org # v4.4+
Fixes: 196581717d ("perf tools: Enhance parsing events tracepoint error output")
Link: http://lkml.kernel.org/r/1453809921-24596-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull timer fixes from Thomas Gleixner:
"The timer departement delivers:
- a regression fix for the NTP code along with a proper selftest
- prevent a spurious timer interrupt in the NOHZ lowres code
- a fix for user space interfaces returning the remaining time on
architectures with CONFIG_TIME_LOW_RES=y
- a few patches to fix COMPILE_TEST fallout"
* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
tick/nohz: Set the correct expiry when switching to nohz/lowres mode
clocksource: Fix dependencies for archs w/o HAS_IOMEM
clocksource: Select CLKSRC_MMIO where needed
tick/sched: Hide unused oneshot timer code
kselftests: timers: Add adjtimex SETOFFSET validity tests
ntp: Fix ADJ_SETOFFSET being used w/ ADJ_NANO
itimers: Handle relative timers with CONFIG_TIME_LOW_RES proper
posix-timers: Handle relative timers with CONFIG_TIME_LOW_RES proper
timerfd: Handle relative timers with CONFIG_TIME_LOW_RES proper
hrtimer: Handle remaining time proper for TIME_LOW_RES
clockevents/tcb_clksrc: Prevent disabling an already disabled clock
Pull perf fixes from Thomas Gleixner:
"This is much bigger than typical fixes, but Peter found a category of
races that spurred more fixes and more debugging enhancements. Work
started before the merge window, but got finished only now.
Aside of that this contains the usual small fixes to perf and tools.
Nothing particular exciting"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (43 commits)
perf: Remove/simplify lockdep annotation
perf: Synchronously clean up child events
perf: Untangle 'owner' confusion
perf: Add flags argument to perf_remove_from_context()
perf: Clean up sync_child_event()
perf: Robustify event->owner usage and SMP ordering
perf: Fix STATE_EXIT usage
perf: Update locking order
perf: Remove __free_event()
perf/bpf: Convert perf_event_array to use struct file
perf: Fix NULL deref
perf/x86: De-obfuscate code
perf/x86: Fix uninitialized value usage
perf: Fix race in perf_event_exit_task_context()
perf: Fix orphan hole
perf stat: Do not clean event's private stats
perf hists: Fix HISTC_MEM_DCACHELINE width setting
perf annotate browser: Fix behaviour of Shift-Tab with nothing focussed
perf tests: Remove wrong semicolon in while loop in CQM test
perf: Synchronously free aux pages in case of allocation failure
...
A dma_addr_t is potentially smaller than a phys_addr_t on some archs.
Don't truncate the address when doing the pfn conversion.
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Matthew Wilcox <willy@linux.intel.com>
[willy: fix pfn_t_to_phys as well]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
$ make -C tools/perf build-test
make[1]: Entering directory `/home/acme/git/linux/tools/perf'
make_pure_O: cd . && make -f Makefile O=/tmp/tmp.mPx0Cmik3f DESTDIR=/tmp/tmp.U0SUmVbtJm
make_clean_all_O: cd . && make -f Makefile O=/tmp/tmp.Yl5UzhTU7T DESTDIR=/tmp/tmp.fop1E4jdER clean all
make_debug_O: cd . && make -f Makefile O=/tmp/tmp.pMn2ozBoXC DESTDIR=/tmp/tmp.azxhDp5sEp DEBUG=1
make_no_libperl_O: cd . && make -f Makefile O=/tmp/tmp.qJPiINMtA7 DESTDIR=/tmp/tmp.KNMrLeGDxZ NO_LIBPERL=1
<SNIP>
More needs to be done to make it more compact, i.e. elide the '-f Makefile',
remove that 'cd . &&', move the DESTDIR= and O= to the end, as they don't
convey that much information besides the fact that they are being set to some
random directory just for this build, move the meat, i.e. the meaningful
feature disabling bits to the start, etc.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-wir3w3o4f1nmbgcxgnx8cj9c@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Powerpc provides hcall events that also provides insights into guest
behaviour. Enhance perf kvm stat to record and analyze hcall events.
- To trace hcall events :
perf kvm stat record
- To show the results :
perf kvm stat report --event=hcall
The result shows the number of hypervisor calls from the guest grouped
by their respective reasons displayed with the frequency.
This patch makes use of two additional tracepoints
"kvm_hv:kvm_hcall_enter" and "kvm_hv:kvm_hcall_exit". To map the hcall
codes to their respective names, it needs a mapping. Such mapping is
added in this patch in book3s_hcalls.h.
# pgrep qemu
A sample output :
19378
60515
2 VMs running.
# perf kvm stat record -a
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 4.153 MB perf.data.guest (39624
samples) ]
# perf kvm stat report -p 60515 --event=hcall
Analyze events for all VMs, all VCPUs:
HCALL-EVENT Samples Samples% Time% MinTime MaxTime AvgTime
H_IPI 822 66.08% 88.10% 0.63us 11.38us 2.05us (+- 1.42%)
H_SEND_CRQ 144 11.58% 3.77% 0.41us 0.88us 0.50us (+- 1.47%)
H_VIO_SIGNAL 118 9.49% 2.86% 0.37us 0.83us 0.47us (+- 1.43%)
H_PUT_TERM_CHAR 76 6.11% 2.07% 0.37us 0.90us 0.52us (+- 2.43%)
H_GET_TERM_CHAR 74 5.95% 2.23% 0.37us 1.70us 0.58us (+- 4.77%)
H_RTAS 6 0.48% 0.85% 1.10us 9.25us 2.70us (+-48.57%)
H_PERFMON 4 0.32% 0.12% 0.41us 0.96us 0.59us (+-20.92%)
Total Samples:1244, Total events handled time:1916.69us.
Signed-off-by: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Alexander Yarygin <yarygin@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Scott Wood <scottwood@freescale.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/r/1453962787-15376-4-git-send-email-hemant@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf kvm can be used to analyze guest exit reasons. This support already
exists in x86. Hence, porting it to powerpc.
- To trace KVM events :
perf kvm stat record
If many guests are running, we can track for a specific guest by using
--pid as in : perf kvm stat record --pid <pid>
- To see the results :
perf kvm stat report
The result shows the number of exits (from the guest context to
host/hypervisor context) grouped by their respective exit reasons with
their frequency.
Since, different powerpc machines have different KVM tracepoints, this
patch discovers the available tracepoints dynamically and accordingly
looks for them. If any single tracepoint is not present, this support
won't be enabled for reporting. To record, this will fail if any of the
events we are looking to record isn't available. Right now, its only
supported on PowerPC Book3S_HV architectures.
To analyze the different exits, group them and present them (in a slight
descriptive way) to the user, we need a mapping between the "exit code"
(dumped in the kvm_guest_exit tracepoint data) and to its related
Interrupt vector description (exit reason). This patch adds this mapping
in book3s_hv_exits.h.
It records on two available KVM tracepoints for book3s_hv:
"kvm_hv:kvm_guest_exit" and "kvm_hv:kvm_guest_enter".
Here is a sample o/p:
# pgrep qemu
19378
60515
2 Guests are running on the host.
# perf kvm stat record -a
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 4.153 MB perf.data.guest (39624
samples) ]
# perf kvm stat report -p 60515
Analyze events for pid(s) 60515, all VCPUs:
VM-EXIT Samples Samples% Time% MinTime MaxTime Avg time
SYSCALL 9141 63.67% 7.49% 1.26us 5782.39us 9.87us (+- 6.46%)
H_DATA_STORAGE 4114 28.66% 5.07% 1.72us 4597.68us 14.84us (+-20.06%)
HV_DECREMENTER 418 2.91% 4.26% 0.70us 30002.22us 122.58us (+-70.29%)
EXTERNAL 392 2.73% 0.06% 0.64us 104.10us 1.94us (+-18.83%)
RETURN_TO_HOST 287 2.00% 83.11% 1.53us 124240.15us 3486.52us (+-16.81%)
H_INST_STORAGE 5 0.03% 0.00% 1.88us 3.73us 2.39us (+-14.20%)
Total Samples:14357, Total events handled time:1203918.42us.
Signed-off-by: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Alexander Yarygin <yarygin@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Scott Wood <scottwood@freescale.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/r/1453962787-15376-3-git-send-email-hemant@linux.vnet.ibm.com
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
There is a nasty confusion that, for kernel module, dso->kernel is not
necessary to be DSO_TYPE_KERNEL or DSO_TYPE_GUEST_KERNEL. These two
enums are for vmlinux. See thread [1]. We tried to fix this part but it
is costy.
Code machine__write_buildid_table() is another unfortunate function fall
into this trap that, when issuing buildid event for a kernel module,
cpumode it gives to the event is PERF_RECORD_MISC_USER, not
PERF_RECORD_MISC_KERNEL.
However, even with this bug, most of the time it doesn't causes real
problem. I find this issue when trying to use a perf before commit
3d39ac5386 ("perf machine: No need to have two DSOs lists") to parse a
perf.data generated by newest perf.
[1] https://lkml.org/lkml/2015/9/21/908
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1454089251-203152-1-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The 'tools/perf/test/make' makefile has in its default, 'all' target
builds that will pollute the source code directory, i.e. that will not
use O= variable.
The 'build-test' should be run as often as possible, preferrably after
each non strictly non-code commit, so speed it up by selecting just
the O= targets.
Furthermore it tests both the Makefile.perf file, that is normally
driven by the main Makefile, and the Makefile, reduce the time in half
by having just MK=Makefile, the most usual, tested by 'build-test'.
Please run:
make -C tools/perf -f tests/make
from time to time for testing also the in-place build tests.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-jrt9utscsiqkmjy3ccufostd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To prevent the feature check tests to run repeately, one time per
'tests/make' target/test, this patch utilizes the previously introduced
'feature-dump' make target and FEATURES_DUMP variable, making sure that
the feature checkers run only once when doing build-test for normal test
cases.
However, since standard users doesn't reuse features dump result, we'd
better give an option to check their behaviors. The above feature
should be used to make build-test faster only. Only utilize it for
build-test.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1454068269-235999-1-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
