This patch adds platform dependency into the test case 15
(perf_event_attr). It is based on a suggestion from Jiri Olsa.
Add a new optional attribute named 'arch' in the [config] section of the
test case file. It is a comma separated list of architecture names this
test can be executed on. For example:
arch = x86_64,alpha,ppc
If this attribute is missing the test is executed on any platform. This
does not break existing behavior.
The values listed for this attribute should be identical to uname -m
output.
If the list starts with an exclamation mark (!) the comparison is
inverted, for example for
arch = !s390x,ppc
the test is not executed on s390x or ppc platforms. The exclamation
mark must be at the beginnning of the list.
Here is an example debug output:
[root@s35lp76]# fgrep arch tests/attr/test-stat-C2
arch = x86_64,alpha,ppc
[root@s35lp76]# PERF_TEST_ATTR=/tmp /usr/bin/python2 ./tests/attr.py \
-d ./tests/attr/ -p ./perf -vvvvv -t test-stat-C1
provides the following output:
running './tests/attr//test-stat-C1'
test limitation 'x86_64,alpha,ppc' <--- new
loading expected events
Event event:base-stat
fd = 1
group_fd = -1
.....
Here is the output when a test is skipped:
[root@s35lp76]# fgrep arch tests/attr/test-stat-C1
arch = !s390x
[root@s35lp76]# PERF_TEST_ATTR=/tmp /usr/bin/python2 ./tests/attr.py \
-d ./tests/attr/ -p ./perf -vvvvv -t test-stat-C1
provides the following output:
test limitation '!s390x' <--- new
skipped [s390x] './tests/attr//test-stat-C1' <--- new
The test is skipped with return code 0.
Suggested-and-Acked-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: linux-s390@vger.kernel.org
Link: http://lkml.kernel.org/r/20170622073625.86762-1-tmricht@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Michael reported the segfault when kernel.kptr_restrict=2 is set.
$ perf record ls
...
perf: Segmentation fault
Obtained 16 stack frames.
./perf(dump_stack+0x2d) [0x5068df]
./perf(sighandler_dump_stack+0x2d) [0x5069bf]
./perf() [0x43e47b]
/lib64/libc.so.6(+0x3594f) [0x7f762004794f]
/lib64/libc.so.6(strlen+0x26) [0x7f762009ef86]
/lib64/libc.so.6(__strdup+0xd) [0x7f762009ecbd]
./perf(maps__set_kallsyms_ref_reloc_sym+0x4d) [0x51590f]
./perf(machine__create_kernel_maps+0x136) [0x50a7de]
./perf(perf_session__create_kernel_maps+0x2c) [0x510a81]
./perf(perf_session__new+0x13d) [0x510e23]
./perf() [0x43fd61]
./perf(cmd_record+0x704) [0x441823]
./perf() [0x4bc1a0]
./perf() [0x4bc40d]
./perf() [0x4bc55f]
./perf(main+0x2d5) [0x4bc939]
Segmentation fault (core dumped)
The reason is that with kernel.kptr_restrict=2, we don't get
the symbol from machine__get_running_kernel_start, which we
want to use in maps__set_kallsyms_ref_reloc_sym and we crash.
Check the symbol name value before calling
maps__set_kallsyms_ref_reloc_sym() and succeed without ref_reloc_sym
being set. It's safe because we check its existence before we use it.
Reported-by: Michael Petlan <mpetlan@redhat.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20170626095153.553-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In commit 613f050d68 ("perf probe: Fix to probe on gcc generated
functions in modules"), the offset from symbol is, incorrectly, added
to the trace point address. This leads to incorrect probe trace points
for inlined functions and when using relative line number on symbols.
Prior this patch:
$ perf probe -m nf_nat -D in_range
p:probe/in_range nf_nat:in_range.isra.9+0
$ perf probe -m i40e -D i40e_clean_rx_irq
p:probe/i40e_clean_rx_irq i40e:i40e_napi_poll+2212
$ perf probe -m i40e -D i40e_clean_rx_irq:16
p:probe/i40e_clean_rx_irq i40e:i40e_lan_xmit_frame+626
After:
$ perf probe -m nf_nat -D in_range
p:probe/in_range nf_nat:in_range.isra.9+0
$ perf probe -m i40e -D i40e_clean_rx_irq
p:probe/i40e_clean_rx_irq i40e:i40e_napi_poll+1106
$ perf probe -m i40e -D i40e_clean_rx_irq:16
p:probe/i40e_clean_rx_irq i40e:i40e_napi_poll+2665
Committer testing:
Using 'pfunct', a tool found in the 'dwarves' package [1], one can ask what are
the functions that while not being explicitely marked as inline, were inlined
by the compiler:
# pfunct --cc_inlined /lib/modules/4.12.0-rc4+/kernel/drivers/net/ethernet/intel/e1000e/e1000e.ko | head
__ew32
e1000_regdump
e1000e_dump_ps_pages
e1000_desc_unused
e1000e_systim_to_hwtstamp
e1000e_rx_hwtstamp
e1000e_update_rdt_wa
e1000e_update_tdt_wa
e1000_put_txbuf
e1000_consume_page
Then ask 'perf probe' to produce the kprobe_tracer probe definitions for two of
them:
# perf probe -m e1000e -D e1000e_rx_hwtstamp
p:probe/e1000e_rx_hwtstamp e1000e:e1000_receive_skb+74
# perf probe -m e1000e -D e1000_consume_page
p:probe/e1000_consume_page e1000e:e1000_clean_jumbo_rx_irq+876
p:probe/e1000_consume_page_1 e1000e:e1000_clean_jumbo_rx_irq+1506
p:probe/e1000_consume_page_2 e1000e:e1000_clean_rx_irq_ps+1074
Now lets concentrate on the 'e1000_consume_page' one, that was inlined twice in
e1000_clean_jumbo_rx_irq(), lets see what readelf says about the DWARF tags for
that function:
$ readelf -wi /lib/modules/4.12.0-rc4+/kernel/drivers/net/ethernet/intel/e1000e/e1000e.ko
<SNIP>
<1><13e27b>: Abbrev Number: 121 (DW_TAG_subprogram)
<13e27c> DW_AT_name : (indirect string, offset: 0xa8945): e1000_clean_jumbo_rx_irq
<13e287> DW_AT_low_pc : 0x17a30
<3><13e6ef>: Abbrev Number: 119 (DW_TAG_inlined_subroutine)
<13e6f0> DW_AT_abstract_origin: <0x13ed2c>
<13e6f4> DW_AT_low_pc : 0x17be6
<SNIP>
<1><13ed2c>: Abbrev Number: 142 (DW_TAG_subprogram)
<13ed2e> DW_AT_name : (indirect string, offset: 0xa54c3): e1000_consume_page
So, the first time in e1000_clean_jumbo_rx_irq() where e1000_consume_page() is
inlined is at PC 0x17be6, which subtracted from e1000_clean_jumbo_rx_irq()'s
address, gives us the offset we should use in the probe definition:
0x17be6 - 0x17a30 = 438
but above we have 876, which is twice as much.
Lets see the second inline expansion of e1000_consume_page() in
e1000_clean_jumbo_rx_irq():
<3><13e86e>: Abbrev Number: 119 (DW_TAG_inlined_subroutine)
<13e86f> DW_AT_abstract_origin: <0x13ed2c>
<13e873> DW_AT_low_pc : 0x17d21
0x17d21 - 0x17a30 = 753
So we where adding it at twice the offset from the containing function as we
should.
And then after this patch:
# perf probe -m e1000e -D e1000e_rx_hwtstamp
p:probe/e1000e_rx_hwtstamp e1000e:e1000_receive_skb+37
# perf probe -m e1000e -D e1000_consume_page
p:probe/e1000_consume_page e1000e:e1000_clean_jumbo_rx_irq+438
p:probe/e1000_consume_page_1 e1000e:e1000_clean_jumbo_rx_irq+753
p:probe/e1000_consume_page_2 e1000e:e1000_clean_jumbo_rx_irq+1353
#
Which matches the two first expansions and shows that because we were
doubling the offset it would spill over the next function:
readelf -sw /lib/modules/4.12.0-rc4+/kernel/drivers/net/ethernet/intel/e1000e/e1000e.ko
673: 0000000000017a30 1626 FUNC LOCAL DEFAULT 2 e1000_clean_jumbo_rx_irq
674: 0000000000018090 2013 FUNC LOCAL DEFAULT 2 e1000_clean_rx_irq_ps
This is the 3rd inline expansion of e1000_consume_page() in
e1000_clean_jumbo_rx_irq():
<3><13ec77>: Abbrev Number: 119 (DW_TAG_inlined_subroutine)
<13ec78> DW_AT_abstract_origin: <0x13ed2c>
<13ec7c> DW_AT_low_pc : 0x17f79
0x17f79 - 0x17a30 = 1353
So:
0x17a30 + 2 * 1353 = 0x184c2
And:
0x184c2 - 0x18090 = 1074
Which explains the bogus third expansion for e1000_consume_page() to end up at:
p:probe/e1000_consume_page_2 e1000e:e1000_clean_rx_irq_ps+1074
All fixed now :-)
[1] https://git.kernel.org/pub/scm/devel/pahole/pahole.git/
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: stable@vger.kernel.org
Fixes: 613f050d68 ("perf probe: Fix to probe on gcc generated functions in modules")
Link: http://lkml.kernel.org/r/20170621164134.5701-1-bjorn.topel@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The kernel now supports the disabling of branch tracing, however the
decoder assumes branch tracing is always enabled. Pass through a parameter
to indicate whether branch tracing is enabled and use it to avoid cases
when the decoder is expecting branch packets. There are 2 such cases.
First, FUP packets which can bind to an IP even when there is no branch
tracing. Secondly, the decoder will try to use branch packets to find an IP
to start decoding or to recover from errors.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Link: http://lkml.kernel.org/r/1495786658-18063-11-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Intel PT uses IP compression based on the last IP. For decoding
purposes, 'last IP' is considered to be reset to zero whenever there is
a synchronization packet (PSB). The decoder wasn't doing that, and was
treating the zero value to mean that there was no last IP, whereas
compression can be done against the zero value. Fix by setting last_ip
to zero when a PSB is received and keep track of have_last_ip.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1495786658-18063-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Porting PPC to libdw only needs an architecture-specific hook to move
the register state from perf to libdw.
The ARM and x86 architectures already use libdw, and it is useful to
have as much common code for the unwinder as possible. Mark Wielaard
has contributed a frame-based unwinder to libdw, so that unwinding works
even for binaries that do not have CFI information. In addition,
libunwind is always preferred to libdw by the build machinery so this
cannot introduce regressions on machines that have both libunwind and
libdw installed.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Milian Wolff <milian.wolff@kdab.com>
Acked-by: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/r/1496312681-20133-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Implementing a new --smi-cost mode in perf stat to measure SMI cost.
During the measurement, the /sys/device/cpu/freeze_on_smi will be set.
The measurement can be done with one counter (unhalted core cycles), and
two free running MSR counters (IA32_APERF and SMI_COUNT).
In practice, the percentages of SMI core cycles should be more useful
than absolute value. So the output will be the percentage of SMI core
cycles and SMI#. metric_only will be set by default.
SMI cycles% = (aperf - unhalted core cycles) / aperf
Here is an example output.
Performance counter stats for 'sudo echo ':
SMI cycles% SMI#
0.1% 1
0.010858678 seconds time elapsed
Users who wants to get the actual value can apply additional
--no-metric-only.
Signed-off-by: Kan Liang <Kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Elliott <elliott@hpe.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1495825538-5230-3-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
While trying to reduce util.[ch] I noticed that fetch_kernel_version()
and fetch_ubuntu_kernel_version() do lots of operations only to check if
they are needed, i.e. it checks if the pointer where to return the
kernel version is NULL only after obtaining the kernel version from
/proc/version_signature or by parsing the results from uname().
Do it earlier not to confuse people reading this code in the future :-)
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-i94qwyekk4tzbu0b9ce1r1mz@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The idea here is to make AutoFDO easier in cloud environment with ASLR.
It's easiest to show how this is useful by example. I built a small test
akin to "while(1) { do_nothing(); }" where the do_nothing function is
loaded from a dso:
$ cat burncpu.cpp
#include <dlfcn.h>
int main() {
void* handle = dlopen("./dso.so", RTLD_LAZY);
if (!handle) return -1;
typedef void (*fp)();
fp do_nothing = (fp) dlsym(handle, "do_nothing");
while(1) {
do_nothing();
}
}
$ cat dso.cpp
extern "C" void do_nothing() {}
$ cat build.sh
#!/bin/bash
g++ -shared dso.cpp -o dso.so
g++ burncpu.cpp -o burncpu -ldl
I sampled the execution of this program with perf record -b.
Using the existing "brstack,dso", we get absolute addresses that are
affected by ASLR, and could be different on different hosts. The address
does not uniquely identify a branch/target in the binary:
$ perf script -F brstack,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
0x7f967139b6aa(/tmp/burncpu/dso.so)/0x4006b1(/tmp/burncpu/exe)/P/-/-/0
Using the existing "brstacksym,dso" is a little better, because the
symbol plus offset and dso name *does* uniquely identify a branch/target
in the binary. Ultimately, however, AutoFDO wants a simple offset into
the binary, so we'd have to undo all the work perf did to symbolize in
the first place:
$ perf script -F brstacksym,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
do_nothing+0x5(/tmp/burncpu/dso.so)/main+0x44(/tmp/burncpu/exe)/P/-/-/0
With the new "brstackoff,dso" we get what we need: a simple offset into a
specific dso/binary that uniquely identifies a branch/target:
$ perf script -F brstackoff,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
0x6aa(/tmp/burncpu/dso.so)/0x4006b1(/tmp/burncpu/exe)/P/-/-/0
Signed-off-by: Mark Santaniello <marksan@fb.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170619163825.2012979-2-marksan@fb.com
[ Updated documentation about 'brstackoff' using text from above ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Perf script can report the dso for "addr" and "ip" fields.
This adds the same support for the "brstack" and "brstacksym" fields.
This can be helpful for AutoFDO: we can ignore LBR entries unless the
source and target address are both in the target module we are about to
build.
I built a small test akin to "while(1) { do_nothing(); }" where the
do_nothing function is loaded from a dso:
$ cat burncpu.cpp
#include <dlfcn.h>
int main() {
void* handle = dlopen("./dso.so", RTLD_LAZY);
if (!handle) return -1;
typedef void (*fp)();
fp do_nothing = (fp) dlsym(handle, "do_nothing");
while(1) {
do_nothing();
}
}
$ cat dso.cpp
extern "C" void do_nothing() {}
$ cat build.sh
#!/bin/bash
g++ -shared dso.cpp -o dso.so
g++ burncpu.cpp -o burncpu -ldl
I sampled the execution with perf record -b. Using the new perf script
functionality I can easily find cases where there was a transition from one
dso to another:
$ perf record -a -b -- sleep 5
[ perf record: Woken up 55 times to write data ]
[ perf record: Captured and wrote 18.815 MB perf.data (43593 samples) ]
$ perf script -F brstack,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
0x7f967139b6aa(/tmp/burncpu/dso.so)/0x4006b1(/tmp/burncpu/exe)/P/-/-/0
$ perf script -F brstacksym,dso | sed 's/\/0 /\/0\n/g' | grep burncpu | grep dso.so | head -n 1
do_nothing+0x5(/tmp/burncpu/dso.so)/main+0x44(/tmp/burncpu/exe)/P/-/-/0
Signed-off-by: Mark Santaniello <marksan@fb.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170619163825.2012979-1-marksan@fb.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The 'if' keyword is a define that expands to complex code when
CONFIG_PROFILE_ALL_BRANCHES is selected, which causes a 'perf test LLVM'
failure like:
$ ./perf test LLVM
35: LLVM search and compile :
35.1: Basic BPF llvm compile : Ok
35.2: kbuild searching : Ok
35.3: Compile source for BPF prologue generation: FAILED!
35.4: Compile source for BPF relocation : Skip
The only affected test case is bpf-script-test-prologue.c
because it uses kernel headers and has 'if' inside.
This patch undefines 'if' to make it passes perf test.
More detailed analysis from a message in this thread, also by Wang:
The problem is caused by following relocation information:
$ readelf -a ./llvmsubtest3
...
[ 5] _ftrace_branch PROGBITS 0000000000000000 00000260
00000000000000a0 0000000000000000 WA 0 0 4
...
Relocation section '.relfunc=null_lseek file->f_mode offset orig' at
offset 0x490 contains 4 entries:
Offset Info Type Sym. Value Sym. Name
000000000038 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
0000000000b0 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
000000000128 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
0000000001c0 000b00000001 unrecognized: 1 0000000000000000 _ftrace_branch
Relocation section '.rel_ftrace_branch' at offset 0x4d0 contains 8 entries:
Offset Info Type Sym. Value Sym. Name
000000000000 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000008 000100000001 unrecognized: 1 0000000000000015 .L.str
000000000028 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000030 000100000001 unrecognized: 1 0000000000000015 .L.str
000000000050 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000058 000100000001 unrecognized: 1 0000000000000015 .L.str
000000000078 000200000001 unrecognized: 1 0000000000000000 .L__func__.bpf_func__n
000000000080 000100000001 unrecognized: 1 0000000000000015 .L.str
...
So I think the failure is because you enabled CONFIG_PROFILE_ALL_BRANCHES.
I can reproduce your buggy result by selecting
CONFIG_PROFILE_ALL_BRANCHES in my kbuild:
$ ./perf test LLVM
35: LLVM search and compile :
35.1: Basic BPF llvm compile : Ok
35.2: kbuild searching : Ok
35.3: Compile source for BPF prologue generation: FAILED!
35.4: Compile source for BPF relocation : Skip
Simply undef CONFIG_PROFILE_ALL_BRANCHES in clang opts not working
because it is introduced by "#include <uapi/linux/fs.h>", which override
cmdline options. So I think the best way is to undefine 'if' inside BPF
script.
Reported-and-Tested-by: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Zefan Li <lizefan@huawei.com>
Link: http://lkml.kernel.org/r/20170620183203.2517-1-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
