Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
User visible changes:
- Add ability to specify to select which registers to record,
to reduce the size of perf.data files, and also allow printing
the registers in 'perf script': (Stephane Eranian)
# perf record --intr-regs=AX,SP usleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.016 MB perf.data (8 samples) ]
# perf script -F ip,sym,iregs | tail -5
ffffffff8105f42a native_write_msr_safe AX:0xf SP:0xffff8802629c3c00
ffffffff8105f42a native_write_msr_safe AX:0xf SP:0xffff8802629c3c00
ffffffff81761ac0 _raw_spin_lock AX:0xffff8801bfcf8020 SP:0xffff8802629c3ce8
ffffffff81202bf8 __vma_adjust_trans_huge AX:0x7ffc75200000 SP:0xffff8802629c3b30
ffffffff8122b089 dput AX:0x101 SP:0xffff8802629c3c78
#
Infrastructure changes:
- Open event on evsel cpus and threads. (Kan Liang)
- Add new bpf API to get name from a BPF object. (Wang Nan)
Build fixes:
- Fix build on powerpc broken by pt/bts. (Adrian Hunter)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull perf updates from Ingo Molnar:
"Main perf kernel side changes:
- uprobes updates/fixes. (Oleg Nesterov)
- Add PERF_RECORD_SWITCH to indicate context switches and use it in
tooling. (Adrian Hunter)
- Support BPF programs attached to uprobes and first steps for BPF
tooling support. (Wang Nan)
- x86 generic x86 MSR-to-perf PMU driver. (Andy Lutomirski)
- x86 Intel PT, LBR and BTS updates. (Alexander Shishkin)
- x86 Intel Skylake support. (Andi Kleen)
- x86 Intel Knights Landing (KNL) RAPL support. (Dasaratharaman
Chandramouli)
- x86 Intel Broadwell-DE uncore support. (Kan Liang)
- x86 hw breakpoints robustization (Andy Lutomirski)
Main perf tooling side changes:
- Support Intel PT in several tools, enabling the use of the
processor trace feature introduced in Intel Broadwell processors:
(Adrian Hunter)
# dmesg | grep Performance
# [0.188477] Performance Events: PEBS fmt2+, 16-deep LBR, Broadwell events, full-width counters, Intel PMU driver.
# perf record -e intel_pt//u -a sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.216 MB perf.data ]
# perf script # then navigate in the tool output to some area, like this one:
184 1030 dl_main (/usr/lib64/ld-2.17.so) => 7f21ba661440 dl_main (/usr/lib64/ld-2.17.so)
185 1457 dl_main (/usr/lib64/ld-2.17.so) => 7f21ba669f10 _dl_new_object (/usr/lib64/ld-2.17.so)
186 9f37 _dl_new_object (/usr/lib64/ld-2.17.so) => 7f21ba677b90 strlen (/usr/lib64/ld-2.17.so)
187 7ba3 strlen (/usr/lib64/ld-2.17.so) => 7f21ba677c75 strlen (/usr/lib64/ld-2.17.so)
188 7c78 strlen (/usr/lib64/ld-2.17.so) => 7f21ba669f3c _dl_new_object (/usr/lib64/ld-2.17.so)
189 9f8a _dl_new_object (/usr/lib64/ld-2.17.so) => 7f21ba65fab0 calloc@plt (/usr/lib64/ld-2.17.so)
190 fab0 calloc@plt (/usr/lib64/ld-2.17.so) => 7f21ba675e70 calloc (/usr/lib64/ld-2.17.so)
191 5e87 calloc (/usr/lib64/ld-2.17.so) => 7f21ba65fa90 malloc@plt (/usr/lib64/ld-2.17.so)
192 fa90 malloc@plt (/usr/lib64/ld-2.17.so) => 7f21ba675e60 malloc (/usr/lib64/ld-2.17.so)
193 5e68 malloc (/usr/lib64/ld-2.17.so) => 7f21ba65fa80 __libc_memalign@plt (/usr/lib64/ld-2.17.so)
194 fa80 __libc_memalign@plt (/usr/lib64/ld-2.17.so) => 7f21ba675d50 __libc_memalign (/usr/lib64/ld-2.17.so)
195 5d63 __libc_memalign (/usr/lib64/ld-2.17.so) => 7f21ba675e20 __libc_memalign (/usr/lib64/ld-2.17.so)
196 5e40 __libc_memalign (/usr/lib64/ld-2.17.so) => 7f21ba675d73 __libc_memalign (/usr/lib64/ld-2.17.so)
197 5d97 __libc_memalign (/usr/lib64/ld-2.17.so) => 7f21ba675e18 __libc_memalign (/usr/lib64/ld-2.17.so)
198 5e1e __libc_memalign (/usr/lib64/ld-2.17.so) => 7f21ba675df9 __libc_memalign (/usr/lib64/ld-2.17.so)
199 5e10 __libc_memalign (/usr/lib64/ld-2.17.so) => 7f21ba669f8f _dl_new_object (/usr/lib64/ld-2.17.so)
200 9fc2 _dl_new_object (/usr/lib64/ld-2.17.so) => 7f21ba678e70 memcpy (/usr/lib64/ld-2.17.so)
201 8e8c memcpy (/usr/lib64/ld-2.17.so) => 7f21ba678ea0 memcpy (/usr/lib64/ld-2.17.so)
- Add support for using several Intel PT features (CYC, MTC packets),
the relevant documentation was updated in:
tools/perf/Documentation/intel-pt.txt
briefly describing those packets, its purposes, how to configure
them in the event config terms and relevant external documentation
for further reading. (Adrian Hunter)
- Introduce support for probing at an absolute address, for user and
kernel 'perf probe's, useful when one have the symbol maps on a
developer machine but not on an embedded system. (Wang Nan)
- Add Intel BTS support, with a call-graph script to show it and PT
in use in a GUI using 'perf script' python scripting with
postgresql and Qt. (Adrian Hunter)
- Allow selecting the type of callchains per event, including
disabling callchains in all but one entry in an event list, to save
space, and also to ask for the callchains collected in one event to
be used in other events. (Kan Liang)
- Beautify more syscall arguments in 'perf trace': (Arnaldo Carvalho
de Melo)
* A bunch more translate file/pathnames from pointers to strings.
* Convert numbers to strings for the 'keyctl' syscall 'option'
arg.
* Add missing 'clockid' entries.
- Introduce 'srcfile' sort key: (Andi Kleen)
# perf record -F 10000 usleep 1
# perf report --stdio --dsos '[kernel.vmlinux]' -s srcfile
<SNIP>
# Overhead Source File
26.49% copy_page_64.S
5.49% signal.c
0.51% msr.h
#
It can be combined with other fields, for instance, experiment with
'-s srcfile,symbol'.
There are some oddities in some distros and with some specific
DSOs, being investigated, so your mileage may vary.
- Support per-event 'freq' term: (Namhyung Kim)
$ perf record -e 'cpu/instructions,freq=1234/',cycles -c 1000 sleep 1
$ perf evlist -F
cpu/instructions,freq=1234/: sample_freq=1234
cycles: sample_period=1000
$
- Deref sys_enter pointer args with contents from probe:vfs_getname,
showing pathnames instead of pointers in many syscalls in 'perf
trace'. (Arnaldo Carvalho de Melo)
- Stop collecting /proc/kallsyms in perf.data files, saving about
4.5MB on a typical x86-64 system, use the the symbol resolution
routines used in all the other tools (report, top, etc) now that we
can ask libtraceevent to use perf's symbol resolution code.
(Arnaldo Carvalho de Melo)
- Allow filtering out of perf's PID via 'perf record --exclude-perf'.
(Wang Nan)
- 'perf trace' now supports syscall groups, like strace, i.e:
$ trace -e file touch file
Will expand 'file' into multiple, file related, syscalls. More
work needed to add extra groups for other syscall groups, and also
to complement what was added for the 'file' group, included as a
proof of concept. (Arnaldo Carvalho de Melo)
- Add lock_pi stresser to 'perf bench futex', to test the kernel code
related to FUTEX_(UN)LOCK_PI. (Davidlohr Bueso)
- Let user have timestamps with per-thread recording in 'perf record'
(Adrian Hunter)
- ... and tons of other changes, see the shortlog and the Git log for
details"
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (240 commits)
perf evlist: Add backpointer for perf_env to evlist
perf tools: Rename perf_session_env to perf_env
perf tools: Do not change lib/api/fs/debugfs directly
perf tools: Add tracing_path and remove unneeded functions
perf buildid: Introduce sysfs/filename__sprintf_build_id
perf evsel: Add a backpointer to the evlist a evsel is in
perf trace: Add header with copyright and background info
perf scripts python: Add new compaction-times script
perf stat: Get correct cpu id for print_aggr
tools lib traceeveent: Allow for negative numbers in print format
perf script: Add --[no-]-demangle/--[no-]-demangle-kernel
tracing/uprobes: Do not print '0x (null)' when offset is 0
perf probe: Support probing at absolute address
perf probe: Fix error reported when offset without function
perf probe: Fix list result when address is zero
perf probe: Fix list result when symbol can't be found
tools build: Allow duplicate objects in the object list
perf tools: Remove export.h from MANIFEST
perf probe: Prevent segfault when reading probe point with absolute address
perf tools: Update Intel PT documentation
...
It was reported that "%-8s" does not parse well when used in the printk
format. The '-' is what is throwing it off. Allow that to be included.
Reporter note:
Example before:
transhuge-stres-10730 [004] 5897.713989: mm_compaction_finished: node=0
zone=>-<8s order=-2119871790 ret=
Example after:
transhuge-stres-4235 [000] 453.149280: mm_compaction_finished: node=0
zone=ffffffff81815d7a order=9 ret=
(I will send patches to fix the string handling in the tracepoints so
it's on par with in-kernel printing via trace_pipe:)
transhuge-stres-10921 [007] ...1 6307.140205: mm_compaction_finished: node=0
zone=Normal order=9 ret=partial
Reported-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Tested-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20150827094601.46518bcc@gandalf.local.home
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Running the following perf-stat command on an arm64 system produces the
following result...
[root@aarch64 ~]# perf stat -e kmem:mm_page_alloc -a sleep 1
Warning: [kmem:mm_page_alloc] function sizeof not defined
Warning: Error: expected type 4 but read 0
Segmentation fault
[root@aarch64 ~]#
The second warning was a result of the first warning not stopping
processing after it detected the issue.
That is, code that found the issue reported the first problem, but
because it did not exit out of the functions smoothly, it caused the
other warning to appear and not only that, it later caused the SIGSEGV.
Signed-off-by: Dean Nelson <dnelson@redhat.com>
Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20150820151632.13927.13791.email-sent-by-dnelson@teal
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch utilizes previous introduced bpf_load_program to load
programs in the ELF file into kernel. Result is stored in 'fd' field in
'struct bpf_program'.
During loading, it allocs a log buffer and free it before return. Note
that that buffer is not passed to bpf_load_program() if the first
loading try is successful. Doesn't use a statically allocated log buffer
to avoid potention multi-thread problem.
Instructions collected during opening is cleared after loading.
load_program() is created for loading a 'struct bpf_insn' array into
kernel, bpf_program__load() calls it. By this design we have a function
loads instructions into kernel. It will be used by further patches,
which creates different instances from a program and load them into
kernel.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-20-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If an eBPF program accesses a map, LLVM generates a load instruction
which loads an absolute address into a register, like this:
ld_64 r1, <MCOperand Expr:(mymap)>
...
call 2
That ld_64 instruction will be recorded in relocation section.
To enable the usage of that map, relocation must be done by replacing
the immediate value by real map file descriptor so it can be found by
eBPF map functions.
This patch to the relocation work based on information collected by
patches:
'bpf tools: Collect symbol table from SHT_SYMTAB section',
'bpf tools: Collect relocation sections from SHT_REL sections'
and
'bpf tools: Record map accessing instructions for each program'.
For each instruction which needs relocation, it inject corresponding
file descriptor to imm field. As a part of protocol, src_reg is set to
BPF_PSEUDO_MAP_FD to notify kernel this is a map loading instruction.
This is the final part of map relocation patch. The principle of map
relocation is described in commit message of 'bpf tools: Collect symbol
table from SHT_SYMTAB section'.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-18-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch records the indices of instructions which are needed to be
relocated. That information is saved in the 'reloc_desc' field in
'struct bpf_program'. In the loading phase (this patch takes effect in
the opening phase), the collected instructions will be replaced by map
loading instructions.
Since we are going to close the ELF file and clear all data at the end
of the 'opening' phase, the ELF information will no longer be valid in
the 'loading' phase. We have to locate the instructions before maps are
loaded, instead of directly modifying the instruction.
'struct bpf_map_def' is introduced in this patch to let us know how many
maps are defined in the object.
This is the third part of map relocation. The principle of map relocation
is described in commit message of 'bpf tools: Collect symbol table from
SHT_SYMTAB section'.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-15-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch collects symbols section. This section is useful when linking
BPF maps.
What 'bpf_map_xxx()' functions actually require are map's file
descriptors (and the internal verifier converts fds into pointers to
'struct bpf_map'), which we don't know when compiling. Therefore, we
should make compiler generate a 'ldr_64 r1, <imm>' instruction, and
fill the 'imm' field with the actual file descriptor when loading in
libbpf.
BPF programs should be written in this way:
struct bpf_map_def SEC("maps") my_map = {
.type = BPF_MAP_TYPE_HASH,
.key_size = sizeof(unsigned long),
.value_size = sizeof(unsigned long),
.max_entries = 1000000,
};
SEC("my_func=sys_write")
int my_func(void *ctx)
{
...
bpf_map_update_elem(&my_map, &key, &value, BPF_ANY);
...
}
Compiler should convert '&my_map' into a 'ldr_64, r1, <imm>'
instruction, where imm should be the address of 'my_map'. According to
the address, libbpf knows which map it actually referenced, and then
fills the imm field with the 'fd' of that map created by it.
However, since we never really 'link' the object file, the imm field is
only a record in relocation section. Therefore libbpf should do the
relocation:
1. In relocation section (type == SHT_REL), positions of each such
'ldr_64' instruction are recorded with a reference of an entry in
symbol table (SHT_SYMTAB);
2. From records in symbol table we can find the indics of map
variables.
Libbpf first record SHT_SYMTAB and positions of each instruction which
required bu such operation. Then create file descriptor. Finally, after
map creation complete, replace the imm field.
This is the first patch of BPF map related stuff. It records SHT_SYMTAB
into object's efile field for further use.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-12-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch defines basic interface of libbpf. 'struct bpf_object' will
be the handler of each object file. Its internal structure is hide to
user. eBPF object files are compiled by LLVM as ELF format. In this
patch, libelf is used to open those files, read EHDR and do basic
validation according to e_type and e_machine.
All elf related staffs are grouped together and reside in efile field of
'struct bpf_object'. bpf_object__elf_finish() is introduced to clear it.
After all eBPF programs in an object file are loaded, related ELF
information is useless. Close the object file and free those memory.
The zfree() and zclose() functions are introduced to ensure setting NULL
pointers and negative file descriptors after resources are released.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-6-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
