tools lib traceevent,  perf tools: Rename traceevent_plugin_* APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "traceevent_". This changes
APIs: traceevent_plugin_list_options, traceevent_plugin_free_options_list,
traceevent_plugin_add_options, traceevent_plugin_remove_options,
traceevent_print_plugins

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180702.089951638@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Rename pevent_function* APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_function_handler, pevent_func_handler,
pevent_func_arg_type, PEVENT_FUNC_ARG_VOID, PEVENT_FUNC_ARG_INT,
PEVENT_FUNC_ARG_LONG, PEVENT_FUNC_ARG_STRING, PEVENT_FUNC_ARG_PTRS,
PEVENT_FUNC_ARG_MAX_TYPES

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180701.935881193@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, tools lib lockdep: Rename 'enum pevent_errno' to 'enum tep_errno'

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
'enum pevent_errno' to 'enum tep_errno'.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180701.770475059@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename 'enum pevent_flag' to 'enum tep_flag'

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
pevent_get_page_size API and enum pevent_flag to enum tep_flag

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180701.623942406@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename traceevent_* APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "traceevent_". This
changes APIs: traceevent_host_bigendian, traceevent_load_plugins and
traceevent_unload_plugins

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180701.484691639@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent_set_* APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_set_file_bigendian, pevent_set_flag,
pevent_set_function_resolver, pevent_set_host_bigendian,
pevent_set_long_size, pevent_set_page_size and pevent_get_long_size

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180701.256265951@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent_register_* APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_register_comm, pevent_register_print_string

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.948980691@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent_read_number_* APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_read_number, pevent_read_number_field

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.804271434@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent print APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_print_field, pevent_print_fields, pevent_print_funcs,
pevent_print_printk

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.654453763@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent parse APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_parse_event, pevent_parse_format, pevent_parse_header_page

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.469749700@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent find APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_find_any_field, pevent_find_common_field,
pevent_find_event, pevent_find_field

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.316995920@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent alloc / free APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
APIs: pevent_alloc, pevent_free, pevent_event_info and pevent_func_resolver_t

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.152609945@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename pevent plugin related APIs

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
the pevent plugin related API.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180700.005287044@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename 'struct pevent_record' to 'struct tep_record'

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
the 'struct pevent_record' to 'struct tep_record'.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180659.866021298@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename struct pevent to struct tep_handle

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. That prefix will be "tep_" and not "pevent_". This changes
the struct pevent to struct tep_handle.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yordan Karadzhov (VMware) <y.karadz@gmail.com>
Cc: linux-trace-devel@vger.kernel.org
Link: http://lkml.kernel.org/r/20180808180659.706175783@goodmis.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe powerpc: Fix trace event post-processing

In some cases, a symbol may have multiple aliases. Attempting to add an
entry probe for such symbols results in a probe being added at an
incorrect location while it fails altogether for return probes. This is
only applicable for binaries with debug information.

During the arch-dependent post-processing, the offset from the start of
the symbol at which the probe is to be attached is determined and added
to the start address of the symbol to get the probe's location.  In case
there are multiple aliases, this offset gets added multiple times for
each alias of the symbol and we end up with an incorrect probe location.

This can be verified on a powerpc64le system as shown below.

  $ nm /lib/modules/$(uname -r)/build/vmlinux | grep "sys_open$"
  ...
  c000000000414290 T __se_sys_open
  c000000000414290 T sys_open

  $ objdump -d /lib/modules/$(uname -r)/build/vmlinux | grep -A 10 "<__se_sys_open>:"

  c000000000414290 <__se_sys_open>:
  c000000000414290:       19 01 4c 3c     addis   r2,r12,281
  c000000000414294:       70 c4 42 38     addi    r2,r2,-15248
  c000000000414298:       a6 02 08 7c     mflr    r0
  c00000000041429c:       e8 ff a1 fb     std     r29,-24(r1)
  c0000000004142a0:       f0 ff c1 fb     std     r30,-16(r1)
  c0000000004142a4:       f8 ff e1 fb     std     r31,-8(r1)
  c0000000004142a8:       10 00 01 f8     std     r0,16(r1)
  c0000000004142ac:       c1 ff 21 f8     stdu    r1,-64(r1)
  c0000000004142b0:       78 23 9f 7c     mr      r31,r4
  c0000000004142b4:       78 1b 7e 7c     mr      r30,r3

  For both the entry probe and the return probe, the probe location
  should be _text+4276888 (0xc000000000414298). Since another alias
  exists for 'sys_open', the post-processing code will end up adding
  the offset (8 for powerpc64le) twice and perf will attempt to add
  the probe at _text+4276896 (0xc0000000004142a0) instead.

Before:

  # perf probe -v -a sys_open

  probe-definition(0): sys_open
  symbol:sys_open file:(null) line:0 offset:0 return:0 lazy:(null)
  0 arguments
  Looking at the vmlinux_path (8 entries long)
  Using /lib/modules/4.18.0-rc8+/build/vmlinux for symbols
  Open Debuginfo file: /lib/modules/4.18.0-rc8+/build/vmlinux
  Try to find probe point from debuginfo.
  Symbol sys_open address found : c000000000414290
  Matched function: __se_sys_open [2ad03a0]
  Probe point found: __se_sys_open+0
  Found 1 probe_trace_events.
  Opening /sys/kernel/debug/tracing/kprobe_events write=1
  Writing event: p:probe/sys_open _text+4276896
  Added new event:
    probe:sys_open       (on sys_open)
  ...

  # perf probe -v -a sys_open%return $retval

  probe-definition(0): sys_open%return
  symbol:sys_open file:(null) line:0 offset:0 return:1 lazy:(null)
  0 arguments
  Looking at the vmlinux_path (8 entries long)
  Using /lib/modules/4.18.0-rc8+/build/vmlinux for symbols
  Open Debuginfo file: /lib/modules/4.18.0-rc8+/build/vmlinux
  Try to find probe point from debuginfo.
  Symbol sys_open address found : c000000000414290
  Matched function: __se_sys_open [2ad03a0]
  Probe point found: __se_sys_open+0
  Found 1 probe_trace_events.
  Opening /sys/kernel/debug/tracing/README write=0
  Opening /sys/kernel/debug/tracing/kprobe_events write=1
  Parsing probe_events: p:probe/sys_open _text+4276896
  Group:probe Event:sys_open probe:p
  Writing event: r:probe/sys_open__return _text+4276896
  Failed to write event: Invalid argument
    Error: Failed to add events. Reason: Invalid argument (Code: -22)

After:

  # perf probe -v -a sys_open

  probe-definition(0): sys_open
  symbol:sys_open file:(null) line:0 offset:0 return:0 lazy:(null)
  0 arguments
  Looking at the vmlinux_path (8 entries long)
  Using /lib/modules/4.18.0-rc8+/build/vmlinux for symbols
  Open Debuginfo file: /lib/modules/4.18.0-rc8+/build/vmlinux
  Try to find probe point from debuginfo.
  Symbol sys_open address found : c000000000414290
  Matched function: __se_sys_open [2ad03a0]
  Probe point found: __se_sys_open+0
  Found 1 probe_trace_events.
  Opening /sys/kernel/debug/tracing/kprobe_events write=1
  Writing event: p:probe/sys_open _text+4276888
  Added new event:
    probe:sys_open       (on sys_open)
  ...

  # perf probe -v -a sys_open%return $retval

  probe-definition(0): sys_open%return
  symbol:sys_open file:(null) line:0 offset:0 return:1 lazy:(null)
  0 arguments
  Looking at the vmlinux_path (8 entries long)
  Using /lib/modules/4.18.0-rc8+/build/vmlinux for symbols
  Open Debuginfo file: /lib/modules/4.18.0-rc8+/build/vmlinux
  Try to find probe point from debuginfo.
  Symbol sys_open address found : c000000000414290
  Matched function: __se_sys_open [2ad03a0]
  Probe point found: __se_sys_open+0
  Found 1 probe_trace_events.
  Opening /sys/kernel/debug/tracing/README write=0
  Opening /sys/kernel/debug/tracing/kprobe_events write=1
  Parsing probe_events: p:probe/sys_open _text+4276888
  Group:probe Event:sys_open probe:p
  Writing event: r:probe/sys_open__return _text+4276888
  Added new event:
    probe:sys_open__return (on sys_open%return)
  ...

Reported-by: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Fixes: 99e608b5954c ("perf probe ppc64le: Fix probe location when using DWARF")
Link: http://lkml.kernel.org/r/20180809161929.35058-1-sandipan@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf map: Optimize maps__fixup_overlappings()

This function splits and removes overlapping areas.

Maps in tree are ordered by start address thus we could find first
overlap and stop if next map does not overlap.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/153365189407.435244.7234821822450484712.stgit@buzz
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf map: Synthesize maps only for thread group leader

Threads share map_groups, all map events are merged into it.

Thus we could send mmaps only for thread group leader.  Otherwise it
took ages to attach and record something from processes with many vmas
and threads.

Thread group leader could be already dead, but it seems perf cannot
handle this case anyway.

Testing dummy:

  #include <stdio.h>
  #include <stdlib.h>
  #include <sys/mman.h>
  #include <pthread.h>
  #include <unistd.h>

  void *thread(void *arg) {
          pause();
  }

  int main(int argc, char **argv) {
        int threads = 10000;
        int vmas = 50000;
        pthread_t th;
        for (int i = 0; i < threads; i++)
                pthread_create(&th, NULL, thread, NULL);
        for (int i = 0; i < vmas; i++)
                mmap(NULL, 4096, (i & 1) ? PROT_READ : PROT_WRITE,
                     MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0);
        sleep(60);
        return 0;
  }

Comment by Jiri Olsa:

We actualy synthesize the group leader (if we found one) for the thread
even if it's not present in the thread_map, so the process maps are
always in data.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/153363294102.396323.6277944760215058174.stgit@buzz
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Wire up the augmented syscalls with the syscalls:sys_enter_FOO beautifier

We just check that the evsel is the one we associated with the
bpf-output event associated with the "__augmented_syscalls__" eBPF map,
to show that the formatting is done properly:

  # perf trace -e perf/tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
     0.000 (         ): __augmented_syscalls__:dfd: CWD, filename: 0x43e06da8, flags: CLOEXEC
     0.006 (         ): syscalls:sys_enter_openat:dfd: CWD, filename: 0x43e06da8, flags: CLOEXEC
     0.007 ( 0.004 ms): cat/11486 openat(dfd: CWD, filename: 0x43e06da8, flags: CLOEXEC                 ) = 3
     0.029 (         ): __augmented_syscalls__:dfd: CWD, filename: 0x4400ece0, flags: CLOEXEC
     0.030 (         ): syscalls:sys_enter_openat:dfd: CWD, filename: 0x4400ece0, flags: CLOEXEC
     0.031 ( 0.004 ms): cat/11486 openat(dfd: CWD, filename: 0x4400ece0, flags: CLOEXEC                 ) = 3
     0.249 (         ): __augmented_syscalls__:dfd: CWD, filename: 0xc3700d6
     0.250 (         ): syscalls:sys_enter_openat:dfd: CWD, filename: 0xc3700d6
     0.252 ( 0.003 ms): cat/11486 openat(dfd: CWD, filename: 0xc3700d6                                  ) = 3
  #

Now we just need to get the full blown enter/exit handlers to check if the
evsel being processed is the augmented_syscalls one to go pick the pointer
payloads from the end of the payload.

We also need to state somehow what is the layout for multi pointer arg syscalls.

Also handy would be to have a BTF file with the struct definitions used in
syscalls, compact, generated at kernel built time and available for use in eBPF
programs.

Till we get there we can go on doing some manual coupling of the most relevant
syscalls with some hand built beautifiers.

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: https://lkml.kernel.org/n/tip-r6ba5izrml82nwfmwcp7jpkm@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Setup the augmented syscalls bpf-output event fields

The payload that is put in place by the eBPF script attached to
syscalls:sys_enter_openat (and other syscalls with pointers, in the
future) can be consumed by the existing sys_enter beautifiers if
evsel->priv is setup with a struct syscall_tp with struct tp_fields for
the 'syscall_id' and 'args' fields expected by the beautifiers, this
patch does just that.

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: https://lkml.kernel.org/n/tip-xfjyog8oveg2fjys9r1yy1es@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Make bpf__setup_output_event() return the bpf-output event

We're calling it to setup that event, and we'll need it later to decide
if the bpf-output event we're handling is the one setup for a specific
purpose, return it using ERR_PTR, etc.

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: https://lkml.kernel.org/n/tip-zhachv7il2n1lopt9aonwhu7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Handle "bpf-output" events associated with "__augmented_syscalls__" BPF map

Add an example BPF script that writes syscalls:sys_enter_openat raw
tracepoint payloads augmented with the first 64 bytes of the "filename"
syscall pointer arg.

Then catch it and print it just like with things written to the
"__bpf_stdout__" map associated with a PERF_COUNT_SW_BPF_OUTPUT software
event, by just letting the default tracepoint handler in 'perf trace',
trace__event_handler(), to use bpf_output__fprintf(trace, sample), just
like it does with all other PERF_COUNT_SW_BPF_OUTPUT events, i.e. just
do a dump on the payload, so that we can check if what is being printed
has at least the first 64 bytes of the "filename" arg:

The augmented_syscalls.c eBPF script:

  # cat tools/perf/examples/bpf/augmented_syscalls.c
  // SPDX-License-Identifier: GPL-2.0

  #include <stdio.h>

  struct bpf_map SEC("maps") __augmented_syscalls__ = {
       .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
       .key_size = sizeof(int),
       .value_size = sizeof(u32),
       .max_entries = __NR_CPUS__,
  };

  struct syscall_enter_openat_args {
unsigned long long common_tp_fields;
long    syscall_nr;
long    dfd;
char    *filename_ptr;
long    flags;
long    mode;
  };

  struct augmented_enter_openat_args {
struct syscall_enter_openat_args args;
char  filename[64];
  };

  int syscall_enter(openat)(struct syscall_enter_openat_args *args)
  {
struct augmented_enter_openat_args augmented_args;

probe_read(&augmented_args.args, sizeof(augmented_args.args), args);
probe_read_str(&augmented_args.filename, sizeof(augmented_args.filename), args->filename_ptr);
perf_event_output(args, &__augmented_syscalls__, BPF_F_CURRENT_CPU,
  &augmented_args, sizeof(augmented_args));
return 1;
  }

  license(GPL);
  #

So it will just prepare a raw_syscalls:sys_enter payload for the
"openat" syscall.

This will eventually be done for all syscalls with pointer args,
globally or just when the user asks, using some spec, which args of
which syscalls it wants "expanded" this way, we'll probably start with
just all the syscalls that have char * pointers with familiar names, the
ones we already handle with the probe:vfs_getname kprobe if it is in
place hooking the kernel getname_flags() function used to copy from user
the paths.

Running it we get:

  # perf trace -e perf/tools/perf/examples/bpf/augmented_syscalls.c,openat cat /etc/passwd > /dev/null
     0.000 (         ): __augmented_syscalls__:X?.C......................`\..................../etc/ld.so.cache..#......,....ao.k...............k......1.".........
     0.006 (         ): syscalls:sys_enter_openat:dfd: CWD, filename: 0x5c600da8, flags: CLOEXEC
     0.008 ( 0.005 ms): cat/31292 openat(dfd: CWD, filename: 0x5c600da8, flags: CLOEXEC                 ) = 3
     0.036 (         ): __augmented_syscalls__:X?.C.......................\..................../lib64/libc.so.6......... .\....#........?.......=.C..../.".........
     0.037 (         ): syscalls:sys_enter_openat:dfd: CWD, filename: 0x5c808ce0, flags: CLOEXEC
     0.039 ( 0.007 ms): cat/31292 openat(dfd: CWD, filename: 0x5c808ce0, flags: CLOEXEC                 ) = 3
     0.323 (         ): __augmented_syscalls__:X?.C.....................P....................../etc/passwd......>.C....@................>.C.....,....ao.>.C........
     0.325 (         ): syscalls:sys_enter_openat:dfd: CWD, filename: 0xe8be50d6
     0.327 ( 0.004 ms): cat/31292 openat(dfd: CWD, filename: 0xe8be50d6                                 ) = 3
  #

We need to go on optimizing this to avoid seding trash or zeroes in the
pointer content payload, using the return from bpf_probe_read_str(), but
to keep things simple at this stage and make incremental progress, lets
leave it at that for now.

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: https://lkml.kernel.org/n/tip-g360n1zbj6bkbk6q0qo11c28@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add wrappers to BPF_FUNC_probe_read(_str) functions

Will be used shortly in the augmented syscalls work together with a
PERF_COUNT_SW_BPF_OUTPUT software event to insert syscalls + pointer
contents in the perf ring buffer, to be consumed by 'perf trace'
beautifiers.

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: https://lkml.kernel.org/n/tip-ajlkpz4cd688ulx1u30htkj3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add bpf__setup_output_event() strerror() counterpart

That is just bpf__strerror_setup_stdout() renamed to the more general
"setup_output_event" method, keep the existing stdout() as a wrapper.

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: https://lkml.kernel.org/n/tip-nwnveo428qn0b48axj50vkc7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Generalize bpf__setup_stdout()

We will use it to set up other bpf-output events, for instance to
generate augmented syscall entry tracepoints with pointer contents.

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: https://lkml.kernel.org/n/tip-4r7kw0nsyi4vyz6xm1tzx6a3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Make bpf__for_each_stdout_map() generic

By passing a 'name' arg, that will eventually be used to setup more
"bpf-output" events, e.g. to create a event where to create raw_syscalls
like events that in addition to the syscall arguments will also copy the
pointer contents being passed from/to userspace.

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: https://lkml.kernel.org/n/tip-talrnxps9p3qozk3aeh91fgv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add bpf/stdio.h wrapper to bpf_perf_event_output function

That, together with the map __bpf_output__ that is already handled by
'perf trace' to print that event's contents as strings provides a
debugging facility, to show it in use, print a simple string everytime
the syscalls:sys_enter_openat() syscall tracepoint is hit:

  # cat tools/perf/examples/bpf/hello.c
  #include <stdio.h>

  int syscall_enter(openat)(void *args)
  {
  puts("Hello, world\n");
  return 0;
  }

  license(GPL);
  #
  # perf trace -e openat,tools/perf/examples/bpf/hello.c cat /etc/passwd > /dev/null
     0.016 (         ): __bpf_stdout__:Hello, world
     0.018 ( 0.010 ms): cat/9079 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
     0.057 (         ): __bpf_stdout__:Hello, world
     0.059 ( 0.011 ms): cat/9079 openat(dfd: CWD, filename: /lib64/libc.so.6, flags: CLOEXEC) = 3
     0.417 (         ): __bpf_stdout__:Hello, world
     0.419 ( 0.009 ms): cat/9079 openat(dfd: CWD, filename: /etc/passwd) = 3
  #

This is part of an ongoing experimentation on making eBPF scripts as
consumed by perf to be as concise as possible and using familiar
concepts such as stdio.h functions, that end up just wrapping the
existing BPF functions, trying to hide as much boilerplate as possible
while using just conventions and C preprocessor tricks.

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: https://lkml.kernel.org/n/tip-4tiaqlx5crf0fwpe7a6j84x7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add struct bpf_map struct

A helper structure used by eBPF C program to describe map attributes to
elf_bpf loader, to be used initially by the special __bpf_stdout__ map
used to print strings into the perf ring buffer in BPF scripts, e.g.:

Using the upcoming stdio.h and puts() macros to use the __bpf_stdout__
map to add strings to the ring buffer:

  # cat tools/perf/examples/bpf/hello.c
  #include <stdio.h>

  int syscall_enter(openat)(void *args)
  {
  puts("Hello, world\n");
  return 0;
  }

  license(GPL);
  #
  # cat ~/.perfconfig
  [llvm]
dump-obj = true
  # perf trace -e openat,tools/perf/examples/bpf/hello.c/call-graph=dwarf/ cat /etc/passwd > /dev/null
  LLVM: dumping tools/perf/examples/bpf/hello.o
     0.016 (         ): __bpf_stdout__:Hello, world
     0.018 ( 0.010 ms): cat/9079 openat(dfd: CWD, filename: /etc/ld.so.cache, flags: CLOEXEC           ) = 3
     0.057 (         ): __bpf_stdout__:Hello, world
     0.059 ( 0.011 ms): cat/9079 openat(dfd: CWD, filename: /lib64/libc.so.6, flags: CLOEXEC           ) = 3
     0.417 (         ): __bpf_stdout__:Hello, world
     0.419 ( 0.009 ms): cat/9079 openat(dfd: CWD, filename: /etc/passwd                                ) = 3
  #
  # file tools/perf/examples/bpf/hello.o
  tools/perf/examples/bpf/hello.o: ELF 64-bit LSB relocatable, *unknown arch 0xf7* version 1 (SYSV), not stripped
   # readelf -SW tools/perf/examples/bpf/hello.o
  There are 10 section headers, starting at offset 0x208:

  Section Headers:
    [Nr] Name              Type            Address          Off    Size   ES Flg Lk Inf Al
    [ 0]                   NULL            0000000000000000 000000 000000 00      0   0  0
    [ 1] .strtab           STRTAB          0000000000000000 000188 00007f 00      0   0  1
    [ 2] .text             PROGBITS        0000000000000000 000040 000000 00  AX  0   0  4
    [ 3] syscalls:sys_enter_openat PROGBITS        0000000000000000 000040 000088 00  AX  0   0  8
    [ 4] .relsyscalls:sys_enter_openat REL             0000000000000000 000178 000010 10      9   3  8
    [ 5] maps              PROGBITS        0000000000000000 0000c8 00001c 00  WA  0   0  4
    [ 6] .rodata.str1.1    PROGBITS        0000000000000000 0000e4 00000e 01 AMS  0   0  1
    [ 7] license           PROGBITS        0000000000000000 0000f2 000004 00  WA  0   0  1
    [ 8] version           PROGBITS        0000000000000000 0000f8 000004 00  WA  0   0  4
    [ 9] .symtab           SYMTAB          0000000000000000 000100 000078 18      1   1  8
  Key to Flags:
    W (write), A (alloc), X (execute), M (merge), S (strings), I (info),
    L (link order), O (extra OS processing required), G (group), T (TLS),
    C (compressed), x (unknown), o (OS specific), E (exclude),
    p (processor specific)
    # readelf -s tools/perf/examples/bpf/hello.o

  Symbol table '.symtab' contains 5 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT    5 __bpf_stdout__
     2: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT    7 _license
     3: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT    8 _version
     4: 0000000000000000     0 NOTYPE  GLOBAL DEFAULT    3 syscall_enter_openat
  #

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: https://lkml.kernel.org/n/tip-81fg60om2ifnatsybzwmiga3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add --percent-type option

Set annotation percent type from following choices:

  global-period, local-period, global-hits, local-hits

With following report option setup the percent type will be passed to
annotation browser:

  $ perf report --percent-type period-local

The local/global keywords set if the percentage is computed in the scope
of the function (local) or the whole data (global).  The period/hits
keywords set the base the percentage is computed on - the samples period
or the number of samples (hits).

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-21-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Add --percent-type option

Add --percent-type option to set annotation percent type from following
choices:

  global-period, local-period, global-hits, local-hits

Examples:

  $ perf annotate --percent-type period-local --stdio | head -1
   Percent         |      Source code ... es, percent: local period)
  $ perf annotate --percent-type hits-local --stdio | head -1
   Percent         |      Source code ... es, percent: local hits)
  $ perf annotate --percent-type hits-global --stdio | head -1
   Percent         |      Source code ... es, percent: global hits)
  $ perf annotate --percent-type period-global --stdio | head -1
   Percent         |      Source code ... es, percent: global period)

The local/global keywords set if the percentage is computed in the scope
of the function (local) or the whole data (global).

The period/hits keywords set the base the percentage is computed on -
the samples period or the number of samples (hits).

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-20-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Display percent type in stdio output

In following patches we will allow to switch percent type even for stdio
annotation outputs. Adding the percent type value into the annotation
outputs title.

  $ perf annotate --stdio
   Percent         |      Sou ... instructions:u } (2805 samples, percent: local period)
  --------------------------- ... ------------------------------------------------------
  ...

  $ perf annotate --stdio2
  Samples: 2K of events 'anon ...  count (approx.): 156525487, [percent: local period]
  safe_write.c() /usr/bin/yes
  Percent
  ...

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-19-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Make local period the default percent type

Currently we display the percentages in annotation output based on
number of samples hits. Switching it to period based percentage by
default, because it corresponds more to the time spent on the line.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-18-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Add support to toggle percent type

Add new key bindings to toggle percent type/base in annotation UI browser:

 'p' to switch between local and global percent type
 'b' to switch between hits and perdio percent base

Add the following help messages to the UI browser '?' window:

  ...
  p             Toggle percent type [local/global]
  b             Toggle percent base [period/hits]
  ...

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-17-jolsa@kernel.org
[ Moved percent_type to be the last arg to sym_title(), its an arg to what is being formmated (buf, size) ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Pass browser percent_type in annotate_browser__calc_percent()

Pass browser percent_type in annotate_browser__calc_percent().

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-16-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Pass 'struct annotation_options' to map_symbol__annotation_dump()

Pass 'struct annotation_options' to map_symbol__annotation_dump(), to
carry on and pass the percent_type value.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-15-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Pass struct annotation_options to symbol__calc_lines()

Pass struct annotation_options to symbol__calc_lines(), to carry on and
pass the percent_type value.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-14-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Add percent_type to struct annotation_options

It will be used to carry user selection of percent type for annotation
output.

Passing the percent_type to the annotation_line__print function as the
first step and making it default to current percentage type
(PERCENT_HITS_LOCAL) value.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-13-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Add PERCENT_PERIOD_GLOBAL percent value

Adding and computing global period percent value for annotation line.
Storing it in struct annotation_data percent array under new
PERCENT_PERIOD_GLOBAL index.

At the moment it's not displayed, it's coming in following patches.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-12-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Add PERCENT_PERIOD_LOCAL percent value

Adding and computing local period percent value for annotation line.
Storing it in struct annotation_data percent array under new
PERCENT_PERIOD_LOCAL index.

At the moment it's not displayed, it's coming in following patches.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-11-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Add PERCENT_HITS_GLOBAL percent value

Adding and computing global hits percent value for annotation line.
Storing it in struct annotation_data percent array under new
PERCENT_HITS_GLOBAL index.

At the moment it's not displayed, it's coming in following patches.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-10-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Switch struct annotation_data::percent to array

So we can hold multiple percent values for annotation line.

The first member of this array is current local hits percent value
(PERCENT_HITS_LOCAL index), so no functional change is expected.

Adding annotation_data__percent function to return requested percent
value from struct annotation_data.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-9-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Loop group events directly in annotation__calc_percent()

We need to bring in 'struct hists' object and for that we need 'struct
perf_evsel' object in the scope.

Switching the group data loop with the evsel group loop.  It does the
same thing, but it brings evsel object, that we can use later get the
'struct hists' object.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-8-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Rename hist to sym_hist in annotation__calc_percent

We will need to bring in 'struct hists' variable in this scope, so it's
better we do this rename first.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-7-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Rename local sample variables to data

Based on previous rename, changing also the local variable names to fit
properly.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-6-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Rename struct annotation_line::samples* to data*

The name 'samples*' is little confusing because we have nested 'struct
sym_hist_entry' under annotation_line struct, which holds 'nr_samples'
as well.

Also the holding struct name is 'annotation_data' so the 'data' name
fits better.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-5-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Get rid of annotation__scnprintf_samples_period()

We have more current function tto get the title for annotation,
which is hists__scnprintf_title. They both have same output as
far as the annotation's header line goes.

They differ in counting of the nr_samples, hists__scnprintf_title
provides more accurate number based on the setup of the
symbol_conf.filter_relative variable.

Plus it also displays any uid/thread/dso/socket filters/zooms
if there are set any, which annotation__scnprintf_samples_period
does not.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-4-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Make annotation_line__max_percent static

There's no outside user of it.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20180804130521.11408-3-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Make symbol__annotate_fprintf2() local

There's no outside user of it.

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>
Cc: Stephane Eranian <eranian@google.com>
Link: https://lkml.kernel.org/r/20180804130521.11408-2-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add 'syscall_enter' probe helper for syscall enter tracepoints

Allowing one to hook into the syscalls:sys_enter_NAME tracepoints,
an example is provided that hooks into the 'openat' syscall.

Using it with the probe:vfs_getname probe into getname_flags to get the
filename args as it is copied from userspace:

  # perf probe -l
  probe:vfs_getname    (on getname_flags:73@acme/git/linux/fs/namei.c with pathname)
  # perf trace -e probe:*getname,tools/perf/examples/bpf/sys_enter_openat.c cat /etc/passwd > /dev/null
     0.000 probe:vfs_getname:(ffffffffbd2a8983) pathname="/etc/ld.so.preload"
     0.022 syscalls:sys_enter_openat:dfd: CWD, filename: 0xafbe8da8, flags: CLOEXEC
     0.027 probe:vfs_getname:(ffffffffbd2a8983) pathname="/etc/ld.so.cache"
     0.054 syscalls:sys_enter_openat:dfd: CWD, filename: 0xafdf0ce0, flags: CLOEXEC
     0.057 probe:vfs_getname:(ffffffffbd2a8983) pathname="/lib64/libc.so.6"
     0.316 probe:vfs_getname:(ffffffffbd2a8983) pathname="/usr/lib/locale/locale-archive"
     0.375 syscalls:sys_enter_openat:dfd: CWD, filename: 0xe2b2b0b4
     0.379 probe:vfs_getname:(ffffffffbd2a8983) pathname="/etc/passwd"
  #

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: https://lkml.kernel.org/n/tip-2po9jcqv1qgj0koxlg8kkg30@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Drop unneeded bitmap_zero() calls

bitmap_zero() is called after bitmap_alloc() in perf code. But
bitmap_alloc() internally uses calloc() which guarantees that allocated
area is zeroed. So following bitmap_zero is unneeded. Drop it.

This happened because of confusing name for bitmap allocator. It
should has name bitmap_zalloc instead of bitmap_alloc.

This series:

  https://lkml.org/lkml/2018/6/18/841

introduces a new API for bitmap allocations in kernel, and functions
there are named correctly. Following patch propogates the API to tools,
and fixes naming issue.

Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andriy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kate Stewart <kstewart@linuxfoundation.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Mike Snitzer <snitzer@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Link: http://lkml.kernel.org/r/20180623073502.16321-1-ynorov@caviumnetworks.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf vendor events arm64: Enable JSON events for eMAG

This patch adds the Ampere Computing eMAG file.  This platform follows
the ARMv8 recommended IMPLEMENTATION DEFINED events, where applicable.

Signed-off-by: Sean V Kelley <seanvk.dev@oregontracks.org>
Reviewed-by: John Garry <john.garry@huawei.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ganapatrao Kulkarni <ganapatrao.kulkarni@cavium.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: William Cohen <wcohen@redhat.com>
Cc: linux-arm-kernel@lists.infradead.org
LPU-Reference: 20180803041811.17065-1-seanvk.dev@oregontracks.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add GUI report support for s390 auxiliary trace

Add support for s390 auxiliary trace support.

Use 'perf record -e rbd000 -- ls' to create the perf.data file.

Use 'perf report' to display the auxiliary trace data.

Output before:

  [root@s35lp76 perf]# ./perf report --stdio
  0x128 [0x10]: failed to process type: 70
  Error:
  failed to process sample
  [root@s35lp76 perf]#

Output after:

  [root@s35lp76 perf]# ./perf report --stdio

      18.21%    18.21%  ls     [kernel.kallsyms]       [k] ftrace_likely_update
       9.52%     9.52%  ls     [kernel.kallsyms]       [k] lock_acquire
       9.38%     9.38%  ls     [kernel.kallsyms]       [k] lock_release
       3.45%     3.45%  ls     [kernel.kallsyms]       [k] lock_acquired
       2.88%     2.88%  ls     [kernel.kallsyms]       [k] link_path_walk
       2.63%     2.63%  ls     [kernel.kallsyms]       [k] __d_lookup
       2.38%     2.38%  ls     [kernel.kallsyms]       [k] __d_lookup_rcu
       2.04%     2.04%  ls     [kernel.kallsyms]       [k] ___might_sleep
       1.83%     1.83%  ls     [kernel.kallsyms]       [k] debug_lockdep_rcu_enabled
       1.44%     1.44%  ls     [kernel.kallsyms]       [k] dput
     ....

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20180802074622.13641-4-tmricht@linux.ibm.com
[ Use PRI[xd]64 to fix the build on debian:experimental-x-mips (gcc 8.1.0) and others ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add raw report support for s390 auxiliary trace

Add support for s390 auxiliary trace support.

Use 'perf record -e rbd000' to create the perf.data file.  The event
also has the symbolic name SF_CYCLES_BASIC_DIAG, using 'perf record -e
SF_CYCLES_BASIC_DIAG' is equivalent.

Use 'perf report -D' to display the auxiliary trace data.

Output before:

 0 0 0x25a66 [0x30]: PERF_RECORD_AUXTRACE size: 0x40000
                 offset: 0  ref: 0  idx: 4  tid: -1  cpu: 4
     Nothing else

Output after:

 0 0 0x25a66 [0x30]: PERF_RECORD_AUXTRACE size: 0x40000
                  offset: 0  ref: 0  idx: 4  tid: -1  cpu: 4
 .
 . ... s390 AUX data: size 262144 bytes
    [00000000] Basic   Def:0001 Inst:0000 TW   AS:3 ASN:0xffff IA:0x0000000000c2f1bc
CL:1 HPP:0x8000000000000000 GPP:000000000000000000
    [0x000020] Diag    Def:8005
    [0x0000bf] Basic   Def:0001 Inst:0000 TW   AS:3 ASN:0xffff IA:0x0000000000c2f1bc
CL:1 HPP:0x8000000000000000 GPP:000000000000000000
    [0x0000df] Diag    Def:8005
    [0x00017e] Basic   Def:0001 Inst:0000 TW   AS:3 ASN:0xffff IA:0x0000000000c2f1bc
CL:1 HPP:0x8000000000000000 GPP:000000000000000000
    ....
    [0x000fc0] Trailer F T bsdes:32 dsdes:159 Overflow:0 Time:0xd4ab59a8450fa108
C:1 TOD:0xd4ab4ec98ceb3832 1:0x8000000000000000 2:0xd4ab4ec98ceb3832

This output is shown for every sampled data block. The
output contains the

 - basic-sampling data entry

 - diagnostic-sampling data entry

 - trailer entry

The basic sampling entry and diagnostic sampling entry sizes can be
extracted using the trailer entries in the SDB.  On older hardware these
values (bsdes and dsdes in the trailer entry) are reserved and zero.
Older hardware use hard coded values based on the s390 machine type.

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Link: http://lkml.kernel.org/r/20180802074622.13641-3-tmricht@linux.ibm.com
Link: http://lkml.kernel.org/r/eda2632e-7919-5ffd-5f68-821e77d216fa@linux.ibm.com
[ Merged a fix for a 'tipe puned' problem reported by Michael Ellerman see last Link tag. ]
[ Removed __packed from two structs, they're already naturally packed and having that. ]
[ attribute breaks the build in gcc 8.1.1 mips, 4.4.7 x86_64, 7.1.1 ARCompact ISA, etc) ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Support for perf report -D for s390

Add initial support for s390 auxiliary traces using the CPU-Measurement
Sampling Facility.

Support and ignore PERF_REPORT_AUXTRACE_INFO records in the perf data
file. Later patches will show the contents of the auxiliary traces.

Setup the auxtrace queues and data structures for s390.  A raw dump of
the perf.data file now does not show an error when an auxtrace event is
encountered.

Output before:

  [root@s35lp76 perf]# ./perf report -D -i perf.data.auxtrace
  0x128 [0x10]: failed to process type: 70
  Error:
  failed to process sample

  0x128 [0x10]: event: 70
  .
  . ... raw event: size 16 bytes
  .  0000:  00 00 00 46 00 00 00 10 00 00 00 00 00 00 00 00  ...F............

  0x128 [0x10]: PERF_RECORD_AUXTRACE_INFO type: 0
  [root@s35lp76 perf]#

Output after:

   # ./perf report -D -i perf.data.auxtrace |fgrep PERF_RECORD_AUXTRACE
  0 0 0x128 [0x10]: PERF_RECORD_AUXTRACE_INFO type: 5
  0 0 0x25a66 [0x30]: PERF_RECORD_AUXTRACE size: 0x40000
   offset: 0  ref: 0  idx: 4  tid: -1  cpu: 4
  ....

Additional notes about the underlying hardware and software
implementation, provided by Hendrik Brueckner (see Link: below).

=============================================================================

The CPU-Measurement Facility (CPU-MF) provides a set of functions to obtain
performance information on the mainframe.  Basically, it was introduced
with System z10 years ago for the z/Architecture, that means, 64-bit.
For Linux, there are two facilities of interest, counter facility and sampling
facility.  The counter facility provides hardware counters for instructions,
cycles, crypto-activities, and many more.

The sampling facility is a hardware sampler that when started will write
samples at a particular interval into a sampling buffer.  At some point,
for example, if a sample block is full, it generates an interrupt to collect
samples (while the sampler continues to run).

Few years ago, I started to provide the a perf PMU to use the counter
and sampling facilities.  Recently, the device driver was updated to also
"export" the sampling buffer into the AUX area.  Thomas now completed the
related perf work to interpret and process these AUX data.

If people are more interested in the sampling facility, they can have a
look into:

- The Load-Program-Parameter and the CPU-Measurement Facilities, SA23-2260-05
  http://www-01.ibm.com/support/docview.wss?uid=isg26fcd1cc32246f4c8852574ce0044734a

and to learn how-to use it for Linux on Z, have look at chapter 54,
"Using the CPU-measurement facilities" in the:

- Device Drivers, Features, and Commands, SC33-8411-34
  http://public.dhe.ibm.com/software/dw/linux390/docu/l416dd34.pdf

=============================================================================

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Link: http://lkml.kernel.org/r/20180803100758.GA28475@linux.ibm.com
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Link: http://lkml.kernel.org/r/20180802074622.13641-2-tmricht@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Use perf_evsel__sc_tp_{uint,ptr} for "id"/"args" handling syscalls:* events

Now it looks just about the same as for the trace__sys_{enter,exit}.

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: https://lkml.kernel.org/n/tip-y59may7zx1eccnp4m3qm4u0b@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Setup struct syscall_tp for syscalls:sys_{enter,exit}_NAME events

Mapping "__syscall_nr" to "id" and setting up "args" from the offset of
"__syscall_nr" + sizeof(u64), as the payload for syscalls:* is the same
as for raw_syscalls:*, just the fields have different names.

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: Steven Rostedt <rostedt@goodmis.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-ogeenrpviwcpwl3oy1l55f3m@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Allow setting up a syscall_tp struct without a format_field

To avoid having to ask libtraceevent to find a field by name when
handling each tracepoint event, we setup a struct syscall_tp with
a tp_field struct having an extractor function + the offset for the
"id", "args" and "ret" raw_syscalls:sys_{enter,exit} tracepoints.

Now that we want to do the same with syscalls:sys_{entry,exit}_NAME
individual syscall tracepoints, where we have "id" as "__syscall_nr" and
"args" as the actual series of per syscall parameters, we need more
flexibility from the routines that set up these pre-looked up syscall
tracepoint arg fields.

The next cset will use it.

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: Steven Rostedt <rostedt@goodmis.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-v59q5e0jrlzkpl9a1c7t81ni@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Rename some syscall_tp methods to raw_syscall

Because raw_syscalls have the field for the syscall number as 'id' while
the syscalls:sys_{enter,exit}_NAME have it as __syscall_nr...

Since we want to support both for being able to enable just a
syscalls:sys_{enter,exit}_name instead of asking for
raw_syscalls:sys_{enter,exit} plus filters, make the method names for
each kind of tracepoint more explicit.

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: Steven Rostedt <rostedt@goodmis.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-4rixbfzco6tsry0w9ghx3ktb@git.kernel.org
Signef-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Use beautifiers on syscalls:sys_enter_ handlers

We were using the beautifiers only when processing the
raw_syscalls:sys_enter events, but we can as well use them for the
syscalls:sys_enter_NAME events, as the layout is the same.

Some more tweaking is needed as we're processing them straight away,
i.e. there is no buffering in the sys_enter_NAME event to wait for
things like vfs_getname to provide pointer contents and then flushing
at sys_exit_NAME, so we need to state in the syscall_arg that this
is unbuffered, just print the pointer values, beautifying just
non-pointer syscall args.

This just shows an alternative way of processing tracepoints, that we
will end up using when creating "tracepoint" payloads that already copy
pointer contents (or chunks of it, i.e. not the whole filename, but just
the end of it, not all the bf for a read/write, but just the start,
etc), directly in the kernel using eBPF.

E.g.:

  # perf trace -e syscalls:*enter*sleep,*sleep sleep 1
     0.303 (         ): syscalls:sys_enter_nanosleep:rqtp: 0x7ffc93d5ecc0
     0.305 (1000.229 ms): sleep/8746 nanosleep(rqtp: 0x7ffc93d5ecc0) = 0
  # perf trace -e syscalls:*_*sleep,*sleep sleep 1
     0.288 (         ): syscalls:sys_enter_nanosleep:rqtp: 0x7ffecde87e40
     0.289 (         ): sleep/8748 nanosleep(rqtp: 0x7ffecde87e40) ...
  1000.479 (         ): syscalls:sys_exit_nanosleep:0x0
     0.289 (1000.208 ms): sleep/8748  ... [continued]: nanosleep()) = 0
  #

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: Steven Rostedt <rostedt@goodmis.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-jehyd2zwhw00z3p7v7mg9632@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Associate vfs_getname()'ed pathname with fd returned from 'openat'

When the vfs_getname() wannabe tracepoint is in place:

  # perf probe -l
    probe:vfs_getname    (on getname_flags:73@acme/git/linux/fs/namei.c with pathname)
  #

'perf trace' will use it to get the pathname when it is copied from
userspace to the kernel, right after syscalls:sys_enter_open, copied
in the 'probe:vfs_getname', stash it somewhere and then, at
syscalls:sys_exit_open time, if the 'open' return is not -1, i.e. a
successfull open syscall, associate that pathname to this return, i.e.
the fd.

We were not doing this for the 'openat' syscall, which would cause 'perf
trace' to fallback to using /proc to get the fd, change it so that we
use what we got from probe:vfs_getname, reducing the 'openat'
beautification process cost, ditching the syscalls performed to read
procfs state and avoiding some possible races in the process.

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: https://lkml.kernel.org/n/tip-xnp44ao3bkb6ejeczxfnjwsh@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-core-for-mingo-4.19-20180801' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

perf trace: (Arnaldo Carvalho de Melo)

- Do not require --no-syscalls to suppress strace like output, i.e.

     # perf trace -e sched:*switch

  will show just sched:sched_switch events, not strace-like formatted
  syscall events, use --syscalls to get the previous behaviour.

  If instead:

     # perf trace

  is used, i.e. no events specified, then --syscalls is implied and
  system wide strace like formatting will be applied to all syscalls.

  The behaviour when just a syscall subset is used with '-e' is unchanged:

     # perf trace -e *sleep,sched:*switch

  will work as before: just the 'nanosleep' syscall will be strace-like
  formatted plus the sched:sched_switch tracepoint event, system wide.

- Allow string table generators to use a default header dir, allowing
  use of them without parameters to see the table it generates on
  stdout, e.g.:

    $ tools/perf/trace/beauty/kvm_ioctl.sh
    static const char *kvm_ioctl_cmds[] = {
        [0x00] = "GET_API_VERSION",
        [0x01] = "CREATE_VM",
        [0x02] = "GET_MSR_INDEX_LIST",
        [0x03] = "CHECK_EXTENSION",
<BIG SNIP>
        [0xe0] = "CREATE_DEVICE",
        [0xe1] = "SET_DEVICE_ATTR",
        [0xe2] = "GET_DEVICE_ATTR",
        [0xe3] = "HAS_DEVICE_ATTR",
    };
    $

  See 'ls tools/perf/trace/beauty/*.sh' to see the available string
  table generators.

- Add a generator for IPPROTO_ socket's protocol constants.

perf record: (Kan Liang)

- Fix error out while applying initial delay and using LBR, due to
  the use of a PERF_TYPE_SOFTWARE/PERF_COUNT_SW_DUMMY event to track
  PERF_RECORD_MMAP events while waiting for the initial delay. Such
  events fail when configured asking PERF_SAMPLE_BRANCH_STACK in
  perf_event_attr.sample_type.

perf c2c: (Jiri Olsa)

- Fix report crash for empty browser, when processing a perf.data file
  without events of interest, either because not asked for in
  'perf record' or because the workload didn't triggered such events.

perf list: (Michael Petlan)

- Align metric group description format with PMU event description.

perf tests: (Sandipan Das)

- Fix indexing when invoking subtests, which caused BPF tests to
  get results for the next test in the list, with the last one
  reporting a failure.

eBPF:

- Fix installation directory for header files included from eBPF proggies,
  avoiding clashing with relative paths used to build other software projects
  such as glibc. (Thomas Richter)

- Show better message when failing to load an object. (Arnaldo Carvalho de Melo)

General: (Christophe Leroy)

- Allow overriding MAX_NR_CPUS at compile time, to make the tooling
  usable in systems with less memory, in time this has to be changed
  to properly allocate based on _NPROCESSORS_ONLN.

Architecture specific:

- Update arm64's ThunderX2 implementation defined pmu core events (Ganapatrao Kulkarni)

- Fix complex event name parsing in 'perf test' for PowerPC, where the 'umask' event
  modifier isn't present. (Sandipan Das)

CoreSight ARM hardware tracing: (Leo Yan)

- Fix start tracing packet handling.

- Support dummy address value for CS_ETM_TRACE_ON packet.

- Generate branch sample when receiving a CS_ETM_TRACE_ON packet.

- Generate branch sample for CS_ETM_TRACE_ON packet.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge branch 'perf/urgent' into perf/core, to pick up fixes

Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf trace: Do not require --no-syscalls to suppress strace like output

So far the --syscalls option was the default, requiring explicit
--no-syscalls when wanting to process just some other event, invert that
and assume it only when no other event was specified, allowing its
explicit enablement when wanting to see all syscalls together with some
other event:

E.g:

The existing default is maintained for a single workload:

  # perf trace sleep 1
<SNIP>
     0.264 ( 0.003 ms): sleep/12762 mmap(len: 113045344, prot: READ, flags: PRIVATE, fd: 3) = 0x7f62cbf04000
     0.271 ( 0.001 ms): sleep/12762 close(fd: 3) = 0
     0.295 (1000.130 ms): sleep/12762 nanosleep(rqtp: 0x7ffd15194fd0) = 0
  1000.469 ( 0.006 ms): sleep/12762 close(fd: 1) = 0
  1000.480 ( 0.004 ms): sleep/12762 close(fd: 2) = 0
  1000.502 (         ): sleep/12762 exit_group()
  #

For a pid:

  # pidof ssh
  7826 3961 3226 2628 2493
  # perf trace -p 3961
         ? (         ):  ... [continued]: select()) = 1
     0.023 ( 0.005 ms): clock_gettime(which_clock: BOOTTIME, tp: 0x7ffcc8fce870               ) = 0
     0.036 ( 0.009 ms): read(fd: 5</dev/pts/7>, buf: 0x7ffcc8fca7b0, count: 16384             ) = 3
     0.060 ( 0.004 ms): getpid(                                                               ) = 3961 (ssh)
     0.079 ( 0.004 ms): clock_gettime(which_clock: BOOTTIME, tp: 0x7ffcc8fce8e0               ) = 0
     0.088 ( 0.003 ms): clock_gettime(which_clock: BOOTTIME, tp: 0x7ffcc8fce7c0               ) = 0
<SNIP>

For system wide, threads, cgroups, user, etc when no event is specified,
the existing behaviour is maintained, i.e. --syscalls is selected.

When some event is specified, then --no-syscalls doesn't need to be
specified:

  # perf trace -e tcp:tcp_probe ssh localhost
     0.000 tcp:tcp_probe:src=[::1]:22 dest=[::1]:39074 mark=0 length=53 snd_nxt=0xb67ce8f7 snd_una=0xb67ce8f7 snd_cwnd=10 ssthresh=2147483647 snd_wnd=43776 srtt=18 rcv_wnd=43690
     0.010 tcp:tcp_probe:src=[::1]:39074 dest=[::1]:22 mark=0 length=32 snd_nxt=0xa8f9ef38 snd_una=0xa8f9ef23 snd_cwnd=10 ssthresh=2147483647 snd_wnd=43690 srtt=31 rcv_wnd=43776
     4.525 tcp:tcp_probe:src=[::1]:22 dest=[::1]:39074 mark=0 length=1240 snd_nxt=0xb67ce90c snd_una=0xb67ce90c snd_cwnd=10 ssthresh=2147483647 snd_wnd=43776 srtt=18 rcv_wnd=43776
     7.242 tcp:tcp_probe:src=[::1]:22 dest=[::1]:39074 mark=0 length=80 snd_nxt=0xb67ced44 snd_una=0xb67ce90c snd_cwnd=10 ssthresh=2147483647 snd_wnd=43776 srtt=18 rcv_wnd=174720
  The authenticity of host 'localhost (::1)' can't be established.
  ECDSA key fingerprint is SHA256:TKZS58923458203490asekfjaklskljmkjfgPMBfHzY.
  ECDSA key fingerprint is MD5:d8:29:54:40:71:fa:b8:44:89:52:64:8a:35:42:d0:e8.
  Are you sure you want to continue connecting (yes/no)?
^C
  #

To get the previous behaviour just use --syscalls and get all syscalls formatted
strace like + the specified extra events:

  # trace -e sched:*switch --syscalls sleep 1
  <SNIP>
     0.160 ( 0.003 ms): sleep/12877 mprotect(start: 0x7fdfe2361000, len: 4096, prot: READ) = 0
     0.164 ( 0.009 ms): sleep/12877 munmap(addr: 0x7fdfe2345000, len: 113155) = 0
     0.211 ( 0.001 ms): sleep/12877 brk() = 0x55d3ce68e000
     0.212 ( 0.002 ms): sleep/12877 brk(brk: 0x55d3ce6af000) = 0x55d3ce6af000
     0.215 ( 0.001 ms): sleep/12877 brk() = 0x55d3ce6af000
     0.219 ( 0.004 ms): sleep/12877 open(filename: 0xe1f07c00, flags: CLOEXEC) = 3
     0.225 ( 0.001 ms): sleep/12877 fstat(fd: 3, statbuf: 0x7fdfe2138aa0) = 0
     0.227 ( 0.003 ms): sleep/12877 mmap(len: 113045344, prot: READ, flags: PRIVATE, fd: 3) = 0x7fdfdb1b8000
     0.234 ( 0.001 ms): sleep/12877 close(fd: 3) = 0
     0.257 (         ): sleep/12877 nanosleep(rqtp: 0x7fffb36b6020) ...
     0.260 (         ): sched:sched_switch:prev_comm=sleep prev_pid=12877 prev_prio=120 prev_state=D ==> next_comm=swapper/3 next_pid=0 next_prio=120
     0.257 (1000.134 ms): sleep/12877  ... [continued]: nanosleep()) = 0
  1000.428 ( 0.006 ms): sleep/12877 close(fd: 1) = 0
  1000.440 ( 0.004 ms): sleep/12877 close(fd: 2) = 0
  1000.461 (         ): sleep/12877 exit_group()
  #

When specifiying just some syscalls, the behaviour doesn't change, i.e.:

  # trace -e nanosleep -e sched:*switch sleep 1
     0.000 (         ): sleep/14974 nanosleep(rqtp: 0x7ffc344ba9c0                                        ) ...
     0.007 (         ): sched:sched_switch:prev_comm=sleep prev_pid=14974 prev_prio=120 prev_state=D ==> next_comm=swapper/2 next_pid=0 next_prio=120
     0.000 (1000.139 ms): sleep/14974  ... [continued]: nanosleep()) = 0
  #

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: https://lkml.kernel.org/n/tip-om2fulll97ytnxv40ler8jkf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Include uapi/linux/bpf.h from the 'perf trace' script's bpf.h

The next example scripts need the definition for the BPF functions, i.e.
things like BPF_FUNC_probe_read, and in time will require lots of other
definitions found in uapi/linux/bpf.h, so include it from the bpf.h file
included from the eBPF scripts build with clang via '-e bpf_script.c'
like in this example:

  $ tail -8 tools/perf/examples/bpf/5sec.c
  #include <bpf.h>

  int probe(hrtimer_nanosleep, rqtp->tv_sec)(void *ctx, int err, long sec)
  {
return sec == 5;
  }

  license(GPL);
  $

That 'bpf.h' include in the 5sec.c eBPF example will come from a set of
header files crafted for building eBPF objects, that in a end-user
system will come from:

  /usr/lib/perf/include/bpf/bpf.h

And will include <uapi/linux/bpf.h> either from the place where the
kernel was built, or from a kernel-devel rpm package like:

  -working-directory /lib/modules/4.17.9-100.fc27.x86_64/build

That is set up by tools/perf/util/llvm-utils.c, and can be overriden
by setting the 'kbuild-dir' variable in the "llvm" ~/.perfconfig file,
like:

  # cat ~/.perfconfig
  [llvm]
       kbuild-dir = /home/foo/git/build/linux

This usually doesn't need any change, just documenting here my findings
while working with this code.

In the future we may want to instead just use what is in
/usr/include/linux/bpf.h, that comes from the UAPI provided from the
kernel sources, for now, to avoid getting the kernel's non-UAPI
"linux/bpf.h" file, that will cause clang to fail and is not what we
want anyway (no BPF function definitions, etc), do it explicitely by
asking for "uapi/linux/bpf.h".

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: https://lkml.kernel.org/n/tip-zd8zeyhr2sappevojdem9xxt@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Allow overriding MAX_NR_CPUS at compile time

After update of kernel, the perf tool doesn't run anymore on my 32MB RAM
powerpc board, but still runs on a 128MB RAM board:

  ~# strace perf
  execve("/usr/sbin/perf", ["perf"], [/* 12 vars */]) = -1 ENOMEM (Cannot allocate memory)
  --- SIGSEGV {si_signo=SIGSEGV, si_code=SI_KERNEL, si_addr=0} ---
  +++ killed by SIGSEGV +++
  Segmentation fault

objdump -x shows that .bss section has a huge size of 24Mbytes:

 27 .bss          016baca8  101cebb8  101cebb8  001cd988  2**3

With especially the following objects having quite big size:

  10205f80 l     O .bss 00140000     runtime_cycles_stats
  10345f80 l     O .bss 00140000     runtime_stalled_cycles_front_stats
  10485f80 l     O .bss 00140000     runtime_stalled_cycles_back_stats
  105c5f80 l     O .bss 00140000     runtime_branches_stats
  10705f80 l     O .bss 00140000     runtime_cacherefs_stats
  10845f80 l     O .bss 00140000     runtime_l1_dcache_stats
  10985f80 l     O .bss 00140000     runtime_l1_icache_stats
  10ac5f80 l     O .bss 00140000     runtime_ll_cache_stats
  10c05f80 l     O .bss 00140000     runtime_itlb_cache_stats
  10d45f80 l     O .bss 00140000     runtime_dtlb_cache_stats
  10e85f80 l     O .bss 00140000     runtime_cycles_in_tx_stats
  10fc5f80 l     O .bss 00140000     runtime_transaction_stats
  11105f80 l     O .bss 00140000     runtime_elision_stats
  11245f80 l     O .bss 00140000     runtime_topdown_total_slots
  11385f80 l     O .bss 00140000     runtime_topdown_slots_retired
  114c5f80 l     O .bss 00140000     runtime_topdown_slots_issued
  11605f80 l     O .bss 00140000     runtime_topdown_fetch_bubbles
  11745f80 l     O .bss 00140000     runtime_topdown_recovery_bubbles

This is due to commit 4d255766d28b1 ("perf: Bump max number of cpus
to 1024"), because many tables are sized with MAX_NR_CPUS

This patch gives the opportunity to redefine MAX_NR_CPUS via

  $ make EXTRA_CFLAGS=-DMAX_NR_CPUS=1

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/r/20170922112043.8349468C57@po15668-vm-win7.idsi0.si.c-s.fr
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Show better message when failing to load an object

Before:

  libbpf: license of tools/perf/examples/bpf/etcsnoop.c is GPL
  libbpf: section(6) version, size 4, link 0, flags 3, type=1
  libbpf: kernel version of tools/perf/examples/bpf/etcsnoop.c is 41200
  libbpf: section(7) .symtab, size 120, link 1, flags 0, type=2
  bpf: config program 'syscalls:sys_enter_openat'
  libbpf: load bpf program failed: Operation not permitted
  libbpf: failed to load program 'syscalls:sys_enter_openat'
  libbpf: failed to load object 'tools/perf/examples/bpf/etcsnoop.c'
  bpf: load objects failed

After: (just the last line changes)

  bpf: load objects failed: err=-4009: (Incorrect kernel version)

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: https://lkml.kernel.org/n/tip-wi44iid0yjfht3lcvplc75fm@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf list: Unify metric group description format with PMU event description

PMU event descriptions use 7 spaces + '[' or 8 spaces as indentation.
Metric groups used a tab + '['. This patch unifies it to the way PMU
event descriptions are indented.

BEFORE:

  $ perf list
  [...]
  Metric Groups:

  DSB:
    DSB_Coverage
  [Fraction of Uops delivered by the DSB (aka Decoded Icache; or Uop Cache)]
  [...]

AFTER:

  $ perf list
  [...]
  Metric Groups:

  DSB:
    DSB_Coverage
         [Fraction of Uops delivered by the DSB (aka Decoded Icache; or Uop Cache)]
  [...]

Signed-off-by: Michael Petlan <mpetlan@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Kim Phillips <kim.phillips@arm.com>
LPU-Reference: 771439042.22924766.1532986504631.JavaMail.zimbra@redhat.com
Link: https://lkml.kernel.org/n/tip-mlo850517m6u1rbjndvd1bwr@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf vendor events arm64: Update ThunderX2 implementation defined pmu core events

Signed-off-by: Ganapatrao Kulkarni <ganapatrao.kulkarni@cavium.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ganapatrao Kulkarni <gklkml16@gmail.com>
Cc: Jan Glauber <jan.glauber@cavium.com>
Cc: Jayachandran C <jnair@caviumnetworks.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <robert.richter@cavium.com>
Cc: Vadim Lomovtsev <vadim.lomovtsev@cavium.com>
Cc: Will Deacon <will.deacon@arm.com>
Link: http://lkml.kernel.org/r/20180731100251.23575-1-ganapatrao.kulkarni@cavium.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Generate branch sample for CS_ETM_TRACE_ON packet

CS_ETM_TRACE_ON packet itself can give the info that there have a
discontinuity in the trace, this patch is to add branch sample for
CS_ETM_TRACE_ON packet if it is inserted in the middle of CS_ETM_RANGE
packets; as result we can have hint for the trace discontinuity.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Walker <robert.walker@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1531295145-596-7-git-send-email-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Generate branch sample when receiving a CS_ETM_TRACE_ON packet

If one CS_ETM_TRACE_ON packet is inserted, we miss to generate branch
sample for the previous CS_ETM_RANGE packet.

This patch is to generate branch sample when receiving a CS_ETM_TRACE_ON
packet, so this can save complete info for the previous CS_ETM_RANGE
packet just before CS_ETM_TRACE_ON packet.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Walker <robert.walker@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1531295145-596-6-git-send-email-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Support dummy address value for CS_ETM_TRACE_ON packet

For CS_ETM_TRACE_ON packet, its fields 'packet->start_addr' and
'packet->end_addr' equal to 0xdeadbeefdeadbeefUL which are emitted in
the decoder layer as dummy value, but the dummy value is pointless for
branch sample when we use 'perf script' command to check program flow.

This patch is a preparation to support CS_ETM_TRACE_ON packet for branch
sample, it converts the dummy address value to zero for more readable;
this is accomplished by cs_etm__last_executed_instr() and
cs_etm__first_executed_instr().  The later one is a new function
introduced by this patch.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Walker <robert.walker@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1531295145-596-5-git-send-email-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Fix start tracing packet handling

Usually the start tracing packet is a CS_ETM_TRACE_ON packet, this
packet is passed to cs_etm__flush();  cs_etm__flush() will check the
condition 'prev_packet->sample_type == CS_ETM_RANGE' but 'prev_packet'
is allocated by zalloc() so 'prev_packet->sample_type' is zero in
initialization and this condition is false.  So cs_etm__flush() will
directly bail out without handling the start tracing packet.

This patch is to introduce a new sample type CS_ETM_EMPTY, which is used
to indicate the packet is an empty packet.  cs_etm__flush() will swap
packets when it finds the previous packet is empty, so this can record
the start tracing packet into 'etmq->prev_packet'.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Walker <robert.walker@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1531295145-596-4-git-send-email-leo.yan@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Fix installation directory for eBPF

The perf tool build and install is controlled via a Makefile. The
'install' rule creates directories and copies files. Among them are
header files installed in /usr/lib/include/perf/bpf/.

However all listed examples are installing its header files in

  /usr/lib/<tool-name>/...[/include]/header.h

and not in

  /usr/lib/include/<tool-name>/.../header.h.

Background information:

Building the Fedora 28 glibc RPM on s390x and s390 fails on s390 (gcc
-m31) as gcc is not able to find header-files like stdbool.h.

In the glibc.spec file, you can see that glibc is configured with
"--with-headers". In this case, first -nostdinc is added to the CFLAGS
and then further include paths are added via -isystem.  One of those
paths should contain header files like stdbool.h.

In order to get this path, gcc is invoked with:

- on Fedora 28 (with 4.18 kernel):

  $ gcc -print-file-name=include
  /usr/lib/gcc/s390x-redhat-linux/8/include
  $ gcc -m31 -print-file-name=include
  /usr/lib/gcc/s390x-redhat-linux/8/../../../../lib/include
  => If perf is installed, this is: /usr/lib/include
  On my machine this directory is only containing the directory "perf".
  If perf is not installed gcc returns: /usr/lib/gcc/s390x-redhat-linux/8/include

- on Ubuntu 18.04 (with 4.15 kernel):

  $ gcc  -print-file-name=include
  /usr/lib/gcc/s390x-linux-gnu/7/include
  $ gcc -m31 -print-file-name=include
  /usr/lib/gcc/s390x-linux-gnu/7/include
  => gcc returns the correct path even if perf is installed.

In each case, the introduction of the subdirectory /usr/lib/include
leads to the regression that one can not build the glibc RPM for s390
anymore as gcc can not find headers like stdbool.h.

To remedy this install bpf.h to /usr/lib/perf/include/bpf/bpf.h

Output before using the command 'perf test -Fv 40':

  echo '...[bpf-program-source]...' | /usr/bin/clang ... \
   -I/root/lib/include/perf/bpf ...
                               ^^^^^^^^^^^^
...
  [root@p23lp27 perf]# perf test -F 40
  40: BPF filter                                            :
  40.1: Basic BPF filtering                                 : Ok
  40.2: BPF pinning                                         : Ok
  40.3: BPF prologue generation                             : Ok
  40.4: BPF relocation checker                              : Ok
  [root@p23lp27 perf]#

Output after using command 'perf test -Fv 40':

  echo '...[bpf-program-source]...' | /usr/bin/clang ... \
 -I/root/lib/perf/include/bpf ...
                             ^^^^^^^^^^^^
...
  [root@p23lp27 perf]# perf test -F 40
  40: BPF filter                                            :
  40.1: Basic BPF filtering                                 : Ok
  40.2: BPF pinning                                         : Ok
  40.3: BPF prologue generation                             : Ok
  40.4: BPF relocation checker                              : Ok
  [root@p23lp27 perf]#

Committer testing:

While the above 'perf test -F 40' (or 'perf test bpf') will allow us
to see that the correct path is now added via -I, to actually test this
we better try to use a bpf script that includes files in the changed
directory.

We have the files that now reside in /root/lib/perf/examples/bpf/ to do
just that:

  # tail -8 /root/lib/perf/examples/bpf/5sec.c
  #include <bpf.h>

  int probe(hrtimer_nanosleep, rqtp->tv_sec)(void *ctx, int err, long sec)
  {
  return sec == 5;
  }

  license(GPL);
  # perf trace -e *sleep -e /root/lib/perf/examples/bpf/5sec.c sleep 4
       0.333 (4000.086 ms): sleep/9248 nanosleep(rqtp: 0x7ffc155f3300) = 0
  # perf trace -e *sleep -e /root/lib/perf/examples/bpf/5sec.c sleep 5
       0.287 (         ): sleep/9659 nanosleep(rqtp: 0x7ffeafe38200) ...
       0.290 (         ): perf_bpf_probe:hrtimer_nanosleep:(ffffffff9911efe0) tv_sec=5
       0.287 (5000.059 ms): sleep/9659  ... [continued]: nanosleep()) = 0
  # perf trace -e *sleep -e /root/lib/perf/examples/bpf/5sec.c sleep 6
       0.247 (5999.951 ms): sleep/10068 nanosleep(rqtp: 0x7fff2086d900) = 0
  # perf trace -e *sleep -e /root/lib/perf/examples/bpf/5sec.c sleep 5.987
       0.293 (         ): sleep/10489 nanosleep(rqtp: 0x7ffdd4fc10e0) ...
       0.296 (         ): perf_bpf_probe:hrtimer_nanosleep:(ffffffff9911efe0) tv_sec=5
       0.293 (5986.912 ms): sleep/10489  ... [continued]: nanosleep()) = 0
  #

Suggested-by: Stefan Liebler <stli@linux.ibm.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.ibm.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Fixes: 1b16fffa389d ("perf llvm-utils: Add bpf include path to clang command line")
Link: http://lkml.kernel.org/r/20180731073254.91090-1-tmricht@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf c2c report: Fix crash for empty browser

'perf c2c' scans read/write accesses and tries to find false sharing
cases, so when the events it wants were not asked for or ended up not
taking place, we get no histograms.

So do not try to display entry details if there's not any. Currently
this ends up in crash:

  $ perf c2c report # then press 'd'
  perf: Segmentation fault
  $

Committer testing:

Before:

Record a perf.data file without events of interest to 'perf c2c report',
then call it and press 'd':

  # perf record sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.001 MB perf.data (6 samples) ]
  # perf c2c report
  perf: Segmentation fault
  -------- backtrace --------
  perf[0x5b1d2a]
  /lib64/libc.so.6(+0x346df)[0x7fcb566e36df]
  perf[0x46fcae]
  perf[0x4a9f1e]
  perf[0x4aa220]
  perf(main+0x301)[0x42c561]
  /lib64/libc.so.6(__libc_start_main+0xe9)[0x7fcb566cff29]
  perf(_start+0x29)[0x42c999]
  #

After the patch the segfault doesn't take place, a follow up patch to
tell the user why nothing changes when 'd' is pressed would be good.

Reported-by: rodia@autistici.org
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: Don Zickus <dzickus@redhat.com>
Cc: Joe Mario <jmario@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: f1c5fd4d0bb9 ("perf c2c report: Add TUI cacheline browser")
Link: http://lkml.kernel.org/r/20180724062008.26126-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Fix indexing when invoking subtests

Recently, the subtest numbering was changed to start from 1.  While it
is fine for displaying results, this should not be the case when the
subtests are actually invoked.

Typically, the subtests are stored in zero-indexed arrays and invoked
based on the index passed to the main test function.  Since the index
now starts from 1, the second subtest in the array (index 1) gets
invoked instead of the first (index 0).  This applies to all of the
following subtests but for the last one, the subtest always fails
because it does not meet the boundary condition of the subtest index
being lesser than the number of subtests.

This can be observed on powerpc64 and x86_64 systems running Fedora 28
as shown below.

Before:

  # perf test "builtin clang support"
  55: builtin clang support                                 :
  55.1: builtin clang compile C source to IR                : Ok
  55.2: builtin clang compile C source to ELF object        : FAILED!

  # perf test "LLVM search and compile"
  38: LLVM search and compile                               :
  38.1: Basic BPF llvm compile                              : Ok
  38.2: kbuild searching                                    : Ok
  38.3: Compile source for BPF prologue generation          : Ok
  38.4: Compile source for BPF relocation                   : FAILED!

  # perf test "BPF filter"
  40: BPF filter                                            :
  40.1: Basic BPF filtering                                 : Ok
  40.2: BPF pinning                                         : Ok
  40.3: BPF prologue generation                             : Ok
  40.4: BPF relocation checker                              : FAILED!

After:

  # perf test "builtin clang support"
  55: builtin clang support                                 :
  55.1: builtin clang compile C source to IR                : Ok
  55.2: builtin clang compile C source to ELF object        : Ok

  # perf test "LLVM search and compile"
  38: LLVM search and compile                               :
  38.1: Basic BPF llvm compile                              : Ok
  38.2: kbuild searching                                    : Ok
  38.3: Compile source for BPF prologue generation          : Ok
  38.4: Compile source for BPF relocation                   : Ok

  # perf test "BPF filter"
  40: BPF filter                                            :
  40.1: Basic BPF filtering                                 : Ok
  40.2: BPF pinning                                         : Ok
  40.3: BPF prologue generation                             : Ok
  40.4: BPF relocation checker                              : Ok

Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Fixes: 9ef0112442bd ("perf test: Fix subtest number when showing results")
Link: http://lkml.kernel.org/r/20180726171733.33208-1-sandipan@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Beautify the AF_INET & AF_INET6 'socket' syscall 'protocol' args

For instance:

  $ trace -e socket* ssh sandy
     0.000 ( 0.031 ms): ssh/19919 socket(family: LOCAL, type: STREAM|CLOEXEC|NONBLOCK                   ) = 3
     0.052 ( 0.015 ms): ssh/19919 socket(family: LOCAL, type: STREAM|CLOEXEC|NONBLOCK                   ) = 3
     1.568 ( 0.020 ms): ssh/19919 socket(family: LOCAL, type: STREAM|CLOEXEC|NONBLOCK                   ) = 3
     1.603 ( 0.012 ms): ssh/19919 socket(family: LOCAL, type: STREAM|CLOEXEC|NONBLOCK                   ) = 3
     1.699 ( 0.014 ms): ssh/19919 socket(family: LOCAL, type: STREAM|CLOEXEC|NONBLOCK                   ) = 3
     1.724 ( 0.012 ms): ssh/19919 socket(family: LOCAL, type: STREAM|CLOEXEC|NONBLOCK                   ) = 3
     1.804 ( 0.020 ms): ssh/19919 socket(family: INET, type: STREAM, protocol: TCP                      ) = 3
    17.549 ( 0.098 ms): ssh/19919 socket(family: LOCAL, type: STREAM                                    ) = 4
  acme@sandy's password:

Just like with other syscall args, the common bits are supressed so that
the output is more compact, i.e. we use "TCP" instead of "IPPROTO_TCP",
but we can make this show the original constant names if we like it by
using some command line knob or ~/.perfconfig "[trace]" section
variable.

Also needed is to make perf's event parser accept things like:

  $ perf trace -e socket*/protocol=TCP/

By using both the tracefs event 'format' files and these tables built
from the kernel sources.

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: https://lkml.kernel.org/n/tip-l39jz1vnyda0b6jsufuc8bz7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace beauty: Add beautifiers for 'socket''s 'protocol' arg

It'll be wired to 'perf trace' in the next cset.

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: https://lkml.kernel.org/n/tip-2i9vkvm1ik8yu4hgjmxhsyjv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace beauty: Do not print NULL strarray entries

We may have string tables where not all slots have values, in those
cases its better to print the numeric value, for instance:

In the table below we would show "protocol: (null)" for

      socket_ipproto[3]

Where it would be better to show "protocol: 3".

      $ tools/perf/trace/beauty/socket_ipproto.sh
      static const char *socket_ipproto[] = {
            [0] = "IP",
            [103] = "PIM",
            [108] = "COMP",
            [12] = "PUP",
            [132] = "SCTP",
            [136] = "UDPLITE",
            [137] = "MPLS",
            [17] = "UDP",
            [1] = "ICMP",
            [22] = "IDP",
            [255] = "RAW",
            [29] = "TP",
            [2] = "IGMP",
            [33] = "DCCP",
            [41] = "IPV6",
            [46] = "RSVP",
            [47] = "GRE",
            [4] = "IPIP",
            [50] = "ESP",
            [51] = "AH",
            [6] = "TCP",
            [8] = "EGP",
            [92] = "MTP",
            [94] = "BEETPH",
            [98] = "ENCAP",
      };
      $

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: https://lkml.kernel.org/n/tip-7djfak94eb3b9ltr79cpn3ti@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf beauty: Add a generator for IPPROTO_ socket's protocol constants

It'll use tools/include copy of linux/in.h to generate a table to be
used by tools, initially by the 'socket' and 'socketpair' beautifiers in
'perf trace', but that could also be used to translate from a string
constant to the integer value to be used in a eBPF or tracefs tracepoint
filter.

When used without any args it produces:

  $ tools/perf/trace/beauty/socket_ipproto.sh
  static const char *socket_ipproto[] = {
[0] = "IP",
[103] = "PIM",
[108] = "COMP",
[12] = "PUP",
[132] = "SCTP",
[136] = "UDPLITE",
[137] = "MPLS",
[17] = "UDP",
[1] = "ICMP",
[22] = "IDP",
[255] = "RAW",
[29] = "TP",
[2] = "IGMP",
[33] = "DCCP",
[41] = "IPV6",
[46] = "RSVP",
[47] = "GRE",
[4] = "IPIP",
[50] = "ESP",
[51] = "AH",
[6] = "TCP",
[8] = "EGP",
[92] = "MTP",
[94] = "BEETPH",
[98] = "ENCAP",
  };
  $

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: https://lkml.kernel.org/n/tip-v9rafqh3qn6b9kp9vfvj9f8s@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools include uapi: Grab a copy of linux/in.h

We'll use it to create tables for the 'protocol' argument to the
socket syscall when the 'family' arg is one of AF_INET or AF_INET6.

Add it to check_headers.sh so that when a new protocol gets added we get
a notification during the build process.

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: https://lkml.kernel.org/n/tip-2amnveu1ns4emjn70xuavpje@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Fix complex event name parsing

The 'umask' event parameter is unsupported on some architectures like
powerpc64.

This can be observed on a powerpc64le system running Fedora 27 as shown
below.

  # perf test "Parse event definition strings" -v
   6: Parse event definition strings                        :
  --- start ---
  test child forked, pid 45915
  ...
  running test 3 'cpu/name='COMPLEX_CYCLES_NAME:orig=cycles,desc=chip-clock-ticks',period=0x1,event=0x2,umask=0x3/ukp'Invalid event/parameter 'umask'
  Invalid event/parameter 'umask'
  failed to parse event 'cpu/name='COMPLEX_CYCLES_NAME:orig=cycles,desc=chip-clock-ticks',period=0x1,event=0x2,umask=0x3/ukp', err 1, str 'unknown term'
  event syntax error: '..,event=0x2,umask=0x3/ukp'
                                    \___ unknown term

  valid terms: event,mark,pmc,cache_sel,pmcxsel,unit,thresh_stop,thresh_start,combine,thresh_sel,thresh_cmp,sample_mode,config,config1,config2,name,period,freq,branch_type,time,call-graph,stack-size,no-inherit,inherit,max-stack,no-overwrite,overwrite,driver-config

  mem_access -> cpu/event=0x10401e0/
  running test 0 'config=10,config1,config2=3,umask=1'
  test child finished with 1
  ---- end ----
  Parse event definition strings: FAILED!

Committer testing:

After applying the patch these test passes and in verbose mode we get:

  # perf test -v "event definition"
   6: Parse event definition strings:
  --- start ---
  test child forked, pid 11061
  running test 0 'syscalls:sys_enter_openat'Using CPUID GenuineIntel-6-9E
  <SNIP>
  running test 53 'cycles/name='COMPLEX_CYCLES_NAME:orig=cycles,desc=chip-clock-ticks'/Duk'
  running test 0 'cpu/config=10,config1,config2=3,period=1000/u'
  running test 1 'cpu/config=1,name=krava/u,cpu/config=2/u'
  running test 2 'cpu/config=1,call-graph=fp,time,period=100000/,cpu/config=2,call-graph=no,time=0,period=2000/'
  running test 3 'cpu/name='COMPLEX_CYCLES_NAME:orig=cycles,desc=chip-clock-ticks',period=0x1,event=0x2/ukp'
  <SNIP>
  test child finished with 0
  ---- end ----
  Parse event definition strings: Ok
  #

Suggested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Fixes: 06dc5bf21f3f ("perf tests: Check that complex event name is parsed correctly")
Link: http://lkml.kernel.org/r/20180726105502.31670-1-sandipan@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Fix error out while applying initial delay and LBR

'perf record' will error out if both --delay and LBR are applied.

For example:

  # perf record -D 1000 -a -e cycles -j any -- sleep 2
  Error:
  dummy:HG: PMU Hardware doesn't support sampling/overflow-interrupts.
  Try 'perf stat'
  #

A dummy event is added implicitly for initial delay, which has the same
configurations as real sampling events. The dummy event is a software
event. If LBR is configured, perf must error out.

The dummy event will only be used to track PERF_RECORD_MMAP while perf
waits for the initial delay to enable the real events. The BRANCH_STACK
bit can be safely cleared for the dummy event.

After applying the patch:

  # perf record -D 1000 -a -e cycles -j any -- sleep 2
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 1.054 MB perf.data (828 samples) ]
  #

Reported-by: Sunil K Pandey <sunil.k.pandey@intel.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1531145722-16404-1-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace beauty: Default header_dir to cwd to work without parms

Useful when checking the effects of header synchs for the files it uses
as a input to generate string tables, in retrospect this is how it
should've been done from day 1, not requiring the header_dir to be set
on the Makefile, will change everything later, so that the only parm,
common to all generators will be $(srctree) and $(beauty_outdir).

So, to see what it generates, just call it without any parameters:

  $ tools/perf/trace/beauty/vhost_virtio_ioctl.sh
  static const char *vhost_virtio_ioctl_cmds[] = {
[0x00] = "SET_FEATURES",
[0x01] = "SET_OWNER",
[0x02] = "RESET_OWNER",
[0x03] = "SET_MEM_TABLE",
[0x04] = "SET_LOG_BASE",
[0x07] = "SET_LOG_FD",
[0x10] = "SET_VRING_NUM",
[0x11] = "SET_VRING_ADDR",
[0x12] = "SET_VRING_BASE",
[0x13] = "SET_VRING_ENDIAN",
[0x14] = "GET_VRING_ENDIAN",
[0x20] = "SET_VRING_KICK",
[0x21] = "SET_VRING_CALL",
[0x22] = "SET_VRING_ERR",
[0x23] = "SET_VRING_BUSYLOOP_TIMEOUT",
[0x24] = "GET_VRING_BUSYLOOP_TIMEOUT",
[0x30] = "NET_SET_BACKEND",
[0x40] = "SCSI_SET_ENDPOINT",
[0x41] = "SCSI_CLEAR_ENDPOINT",
[0x42] = "SCSI_GET_ABI_VERSION",
[0x43] = "SCSI_SET_EVENTS_MISSED",
[0x44] = "SCSI_GET_EVENTS_MISSED",
[0x60] = "VSOCK_SET_GUEST_CID",
[0x61] = "VSOCK_SET_RUNNING",
  };
  static const char *vhost_virtio_ioctl_read_cmds[] = {
[0x00] = "GET_FEATURES",
[0x12] = "GET_VRING_BASE",
  };
  $

Or:

  $ tools/perf/trace/beauty/sndrv_pcm_ioctl.sh
  static const char *sndrv_pcm_ioctl_cmds[] = {
[0x00] = "PVERSION",
[0x01] = "INFO",
[0x02] = "TSTAMP",
[0x03] = "TTSTAMP",
[0x04] = "USER_PVERSION",
[0x10] = "HW_REFINE",
[0x11] = "HW_PARAMS",
[0x12] = "HW_FREE",
[0x13] = "SW_PARAMS",
[0x20] = "STATUS",
[0x21] = "DELAY",
[0x22] = "HWSYNC",
[0x23] = "SYNC_PTR",
[0x24] = "STATUS_EXT",
[0x32] = "CHANNEL_INFO",
[0x40] = "PREPARE",
[0x41] = "RESET",
[0x42] = "START",
[0x43] = "DROP",
[0x44] = "DRAIN",
[0x45] = "PAUSE",
[0x46] = "REWIND",
[0x47] = "RESUME",
[0x48] = "XRUN",
[0x49] = "FORWARD",
[0x50] = "WRITEI_FRAMES",
[0x51] = "READI_FRAMES",
[0x52] = "WRITEN_FRAMES",
[0x53] = "READN_FRAMES",
[0x60] = "LINK",
[0x61] = "UNLINK",
  };
  $

Etc.

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: https://lkml.kernel.org/n/tip-90am4vm8hh1osms894dp2otr@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge remote-tracking branch 'tip/perf/urgent' into perf/core

To pick up fixes.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-urgent-for-mingo-4.18-20180730' of git://git./linux/kernel/git/acme/linux into perf/urgent

Pull perf/urgent fixes from Arnaldo Carvalho de Melo:

- Update the tools copy of several files, including perf_event.h,
  powerpc's asm/unistd.h (new io_pgetevents syscall), bpf.h and
  x86's memcpy_64.s (used in 'perf bench mem'), silencing the
  respective warnings during the perf tools build.

- Fix the build on the alpine:edge distro.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Fix hardcoded index of Broadwell extra PCI devices

Masayoshi Mizuma reported that a warning message is shown while a CPU is
hot-removed on Broadwell servers:

  WARNING: CPU: 126 PID: 6 at arch/x86/events/intel/uncore.c:988
  uncore_pci_remove+0x10b/0x150
  Call Trace:
   pci_device_remove+0x42/0xd0
   device_release_driver_internal+0x148/0x220
   pci_stop_bus_device+0x76/0xa0
   pci_stop_root_bus+0x44/0x60
   acpi_pci_root_remove+0x1f/0x80
   acpi_bus_trim+0x57/0x90
   acpi_bus_trim+0x2e/0x90
   acpi_device_hotplug+0x2bc/0x4b0
   acpi_hotplug_work_fn+0x1a/0x30
   process_one_work+0x174/0x3a0
   worker_thread+0x4c/0x3d0
   kthread+0xf8/0x130

This bug was introduced by:

  commit 15a3e845b01c ("perf/x86/intel/uncore: Fix SBOX support for Broadwell CPUs")

The index of "QPI Port 2 filter" was hardcode to 2, but this conflicts with the
index of "PCU.3" which is "HSWEP_PCI_PCU_3", which equals to 2 as well.

To fix the conflict, the hardcoded index needs to be cleaned up:

 - introduce a new enumerator "BDX_PCI_QPI_PORT2_FILTER" for "QPI Port 2
   filter" on Broadwell,
 - increase UNCORE_EXTRA_PCI_DEV_MAX by one,
 - clean up the hardcoded index.

Debugged-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Suggested-by: Ingo Molnar <mingo@kernel.org>
Reported-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Tested-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
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: msys.mizuma@gmail.com
Cc: stable@vger.kernel.org
Fixes: 15a3e845b01c ("perf/x86/intel/uncore: Fix SBOX support for Broadwell CPUs")
Link: http://lkml.kernel.org/r/1532953688-15008-1-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "Misc fixes:

   - a build race fix

   - a Xen entry fix

   - a TSC_DEADLINE quirk future-proofing fix"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/boot: Fix if_changed build flip/flop bug
  x86/entry/64: Remove %ebx handling from error_entry/exit
  x86/apic: Future-proof the TSC_DEADLINE quirk for SKX

Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Ingo Molnar:
 "Misc fixes:

   - a deadline scheduler related bug fix which triggered a kernel
     warning

   - an RT_RUNTIME_SHARE fix

   - a stop_machine preemption fix

   - a potential NULL dereference fix in sched_domain_debug_one()"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/rt: Restore rt_runtime after disabling RT_RUNTIME_SHARE
  sched/deadline: Update rq_clock of later_rq when pushing a task
  stop_machine: Disable preemption after queueing stopper threads
  sched/topology: Check variable group before dereferencing it

Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "Misc fixes:

   - AMD IBS data corruptor fix (uncovered by UBSAN)

   - an Intel PEBS entry unwind error fix

   - a HW-tracing crash fix

   - a MAINTAINERS update"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/core: Fix crash when using HW tracing kernel filters
  perf/x86/intel: Fix unwind errors from PEBS entries (mk-II)
  MAINTAINERS: Add Naveen N. Rao as kprobes co-maintainer
  perf/x86/amd/ibs: Don't access non-started event

Merge branch 'locking-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull locking fixes from Ingo Molnar:
 "A paravirt UP-patching fix, and an I2C MUX driver lockdep warning fix"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/pvqspinlock/x86: Use LOCK_PREFIX in __pv_queued_spin_unlock() assembly code
  i2c/mux, locking/core: Annotate the nested rt_mutex usage
  locking/rtmutex: Allow specifying a subclass for nested locking

Merge branch 'efi-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull EFI fix from Ingo Molnar:
 "An UEFI variables fix for SEV guests"

* 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/efi: Access EFI MMIO data as unencrypted when SEV is active

perf tools: Fix the build on the alpine:edge distro

The UAPI file byteorder/little_endian.h uses the __always_inline define
without including the header where it is defined, linux/stddef.h, this
ends up working in all the other distros because that file gets included
seemingly by luck from one of the files included from little_endian.h.

But not on Alpine:edge, that fails for all files where perf_event.h is
included but linux/stddef.h isn't include before that.

Adding the missing linux/stddef.h file where it breaks on Alpine:edge to
fix that, in all other distros, that is just a very small header anyway.

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: https://lkml.kernel.org/n/tip-9r1pifftxvuxms8l7ir73p5l@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools arch: Update arch/x86/lib/memcpy_64.S copy used in 'perf bench mem memcpy'

To cope with the changes in:

  12c89130a56a ("x86/asm/memcpy_mcsafe: Add write-protection-fault handling")
  60622d68227d ("x86/asm/memcpy_mcsafe: Return bytes remaining")
  bd131544aa7e ("x86/asm/memcpy_mcsafe: Add labels for __memcpy_mcsafe() write fault handling")
  da7bc9c57eb0 ("x86/asm/memcpy_mcsafe: Remove loop unrolling")

This needed introducing a file with a copy of the mcsafe_handle_tail()
function, that is used in the new memcpy_64.S file, as well as a dummy
mcsafe_test.h header.

Testing it:

  $ nm ~/bin/perf | grep mcsafe
  0000000000484130 T mcsafe_handle_tail
  0000000000484300 T __memcpy_mcsafe
  $
  $ perf bench mem memcpy
  # Running 'mem/memcpy' benchmark:
  # function 'default' (Default memcpy() provided by glibc)
  # Copying 1MB bytes ...

      44.389205 GB/sec
  # function 'x86-64-unrolled' (unrolled memcpy() in arch/x86/lib/memcpy_64.S)
  # Copying 1MB bytes ...

      22.710756 GB/sec
  # function 'x86-64-movsq' (movsq-based memcpy() in arch/x86/lib/memcpy_64.S)
  # Copying 1MB bytes ...

      42.459239 GB/sec
  # function 'x86-64-movsb' (movsb-based memcpy() in arch/x86/lib/memcpy_64.S)
  # Copying 1MB bytes ...

      42.459239 GB/sec
  $

This silences this perf tools build warning:

  Warning: Kernel ABI header at 'tools/arch/x86/lib/memcpy_64.S' differs from latest version at 'arch/x86/lib/memcpy_64.S'

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Mika Penttilä <mika.penttila@nextfour.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-igdpciheradk3gb3qqal52d0@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools headers uapi: Refresh linux/bpf.h copy

To get the changes in:

  4c79579b44b1 ("bpf: Change bpf_fib_lookup to return lookup status")

That do not entail changes in tools/perf/ use of it, elliminating the
following perf build warning:

  Warning: Kernel ABI header at 'tools/include/uapi/linux/bpf.h' differs from latest version at 'include/uapi/linux/bpf.h'

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
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: https://lkml.kernel.org/n/tip-yei494y6b3mn6bjzz9g0ws12@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools headers powerpc: Update asm/unistd.h copy to pick new

The new 'io_pgetevents' syscall was wired up in PowerPC in the following
cset:

  b2f82565f2ca ("powerpc: Wire up io_pgetevents")

Update tools/arch/powerpc/ copy of the asm/unistd.h file so that 'perf
trace' on PowerPC gets it in its syscall table.

This elliminated the following perf build warning:

  Warning: Kernel ABI header at 'tools/arch/powerpc/include/uapi/asm/unistd.h' differs from latest version at 'arch/powerpc/include/uapi/asm/unistd.h'

Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Breno Leitao <leitao@debian.org>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Thomas Richter <tmricht@linux.vnet.ibm.com>
Link: https://lkml.kernel.org/n/tip-9uvu7tz4ud3bxxfyxwryuz47@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools headers uapi: Update tools's copy of linux/perf_event.h

To get the changes in:

  6cbc304f2f36 ("perf/x86/intel: Fix unwind errors from PEBS entries (mk-II)")

That do not imply any changes in the tooling side, the (ab)use of
sample_type is entirely done in kernel space, nothing for userspace to
witness here.

This cures the following warning during perf's build:

  Warning: Kernel ABI header at 'tools/include/uapi/linux/perf_event.h' differs from latest version at 'include/uapi/linux/perf_event.h'

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Prashant Bhole <bhole_prashant_q7@lab.ntt.co.jp>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-o64mjoy35s9gd1gitunw1zg4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Linux 4.18-rc7

Merge tag 'ext4_for_linus_stable' of git://git./linux/kernel/git/tytso/ext4

Pull ext4 fixes from Ted Ts'o:
 "Some miscellaneous ext4 fixes for 4.18; one fix is for a regression
  introduced in 4.18-rc4.

  Sorry for the late-breaking pull. I was originally going to wait for
  the next merge window, but Eric Whitney found a regression introduced
  in 4.18-rc4, so I decided to push out the regression plus the other
  fixes now. (The other commits have been baking in linux-next since
  early July)"

* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: fix check to prevent initializing reserved inodes
  ext4: check for allocation block validity with block group locked
  ext4: fix inline data updates with checksums enabled
  ext4: clear mmp sequence number when remounting read-only
  ext4: fix false negatives *and* false positives in ext4_check_descriptors()

squashfs: be more careful about metadata corruption

Anatoly Trosinenko reports that a corrupted squashfs image can cause a
kernel oops.  It turns out that squashfs can end up being confused about
negative fragment lengths.

The regular squashfs_read_data() does check for negative lengths, but
squashfs_read_metadata() did not, and the fragment size code just
blindly trusted the on-disk value.  Fix both the fragment parsing and
the metadata reading code.

Reported-by: Anatoly Trosinenko <anatoly.trosinenko@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Phillip Lougher <phillip@squashfs.org.uk>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext4: fix check to prevent initializing reserved inodes

Commit 8844618d8aa7: "ext4: only look at the bg_flags field if it is
valid" will complain if block group zero does not have the
EXT4_BG_INODE_ZEROED flag set.  Unfortunately, this is not correct,
since a freshly created file system has this flag cleared.  It gets
almost immediately after the file system is mounted read-write --- but
the following somewhat unlikely sequence will end up triggering a
false positive report of a corrupted file system:

   mkfs.ext4 /dev/vdc
   mount -o ro /dev/vdc /vdc
   mount -o remount,rw /dev/vdc

Instead, when initializing the inode table for block group zero, test
to make sure that itable_unused count is not too large, since that is
the case that will result in some or all of the reserved inodes
getting cleared.

This fixes the failures reported by Eric Whiteney when running
generic/230 and generic/231 in the the nojournal test case.

Fixes: 8844618d8aa7 ("ext4: only look at the bg_flags field if it is valid")
Reported-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>