linux-stable.git/tools/perf/util/map.c, branch v5.5

perf maps: Rename 'mg' variables to 'maps'

2019-11-26T14:07:46+00:00

Continuing the merge of 'struct maps' with 'struct map_groups'.

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-z8d14wrw393a0fbvmnk1bqd9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf thread: Rename thread->mg to thread->maps

2019-11-26T14:07:46+00:00

One more step on the merge of 'struct maps' with 'struct map_groups'.

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-69vcr8pubpym90skxhmbwhiw@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf maps: Merge 'struct maps' with 'struct map_groups'

2019-11-26T14:07:46+00:00

And pick the shortest name: 'struct maps'.

The split existed because we used to have two groups of maps, one for
functions and one for variables, but that only complicated things,
sometimes we needed to figure out what was at some address and then had
to first try it on the functions group and if that failed, fall back to
the variables one.

That split is long gone, so for quite a while we had only one struct
maps per struct map_groups, simplify things by combining those structs.

First patch is the minimum needed to merge both, follow up patches will
rename 'thread->mg' to 'thread->maps', etc.

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-hom6639ro7020o708trhxh59@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf map: Remove unused functions

2019-11-26T14:07:46+00:00

At some point those stopped being used, prune them.

Cc: Adrian Hunter 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-p2k98mj3ff2uk1z95sbl5r6e@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf script: Move map__fprintf_srccode() to near its only user

2019-11-26T14:07:45+00:00

No need to have it elsewhere.

Cc: Adrian Hunter 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-8cw846pudpxo0xdkvi9qnvrh@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf dso: Move dso_id from 'struct map' to 'struct dso'

2019-11-19T22:12:26+00:00

And take it into account when looking up DSOs when we have the dso_id
fields obtained from somewhere, like from PERF_RECORD_MMAP2 records.

Instances of struct map pointing to the same DSO pathname but with
anything in dso_id different are in fact different DSOs, so better have
different 'struct dso' instances to reflect that. At some point we may
want to get copies of the contents of the different objects if we want
to do correct annotation or other analysis.

With this we get 'struct map' 24 bytes leaner:

  $ pahole -C map ~/bin/perf
  struct map {
  	union {
  		struct rb_node     rb_node __attribute__((__aligned__(8))); /*     0    24 */
  		struct list_head   node;                 /*     0    16 */
  	} __attribute__((__aligned__(8)));               /*     0    24 */
  	u64                        start;                /*    24     8 */
  	u64                        end;                  /*    32     8 */
  	_Bool                      erange_warned:1;      /*    40: 0  1 */
  	_Bool                      priv:1;               /*    40: 1  1 */

  	/* XXX 6 bits hole, try to pack */
  	/* XXX 3 bytes hole, try to pack */

  	u32                        prot;                 /*    44     4 */
  	u64                        pgoff;                /*    48     8 */
  	u64                        reloc;                /*    56     8 */
  	/* --- cacheline 1 boundary (64 bytes) --- */
  	u64                        (*map_ip)(struct map *, u64); /*    64     8 */
  	u64                        (*unmap_ip)(struct map *, u64); /*    72     8 */
  	struct dso *               dso;                  /*    80     8 */
  	refcount_t                 refcnt;               /*    88     4 */
  	u32                        flags;                /*    92     4 */

  	/* size: 96, cachelines: 2, members: 13 */
  	/* sum members: 92, holes: 1, sum holes: 3 */
  	/* sum bitfield members: 2 bits, bit holes: 1, sum bit holes: 6 bits */
  	/* forced alignments: 1 */
  	/* last cacheline: 32 bytes */
  } __attribute__((__aligned__(8)));
  $

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-g4hxxmraplo7wfjmk384mfsb@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf map: Pass a dso_id to map__new()

2019-11-19T18:09:26+00:00

Instead of the 4 fields, a step in the direction of moving this to
struct dso.

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-gp5s1xgxacurmih5d1l94ymy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf map: Move maj/min/ino/ino_generation to separate struct

2019-11-19T18:09:26+00:00

And this patch highlights where these fields are being used: in the sort
order where it uses it to compare maps and classify samples taking into
account not just the DSO, but those DSO id fields.

I think these should be used to differentiate DSOs with the same name
but different 'struct dso_id' fields, i.e. these fields should move to
'struct dso' and then be used as part of the key when doing lookups for
DSOs, in addition to the DSO name.

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-8v5isitqy0dup47nnwkpc80f@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf map_groups: Auto sort maps by name, if needed

2019-11-18T16:01:58+00:00

There are still lots of lookups by name, even if just when loading
vmlinux, till that code is studied to figure out if its possible to do
away with those map lookup by names, provide a way to sort it using
libc's qsort/bsearch.

Doing it at the first lookup defers the sorting a bit, and as the code
stands now, is never done for user maps, just for the kernel ones.

  # perf probe -l
  # perf probe -x ~/bin/perf -L __map_groups__find_by_name
  <__map_groups__find_by_name@/home/acme/git/perf/tools/perf/util/symbol.c:0>
        0  static struct map *__map_groups__find_by_name(struct map_groups *mg, const char *name)
        1  {
                  struct map **mapp;

        4         if (mg->maps_by_name == NULL &&
        5             map__groups__sort_by_name_from_rbtree(mg))
        6                 return NULL;

        8         mapp = bsearch(name, mg->maps_by_name, mg->nr_maps, sizeof(*mapp), map__strcmp_name);
        9         if (mapp)
       10                 return *mapp;
       11         return NULL;
       12  }

           struct map *map_groups__find_by_name(struct map_groups *mg, const char *name)
           {

  # perf probe -x ~/bin/perf 'found=__map_groups__find_by_name:10 name:string'
  Added new event:
    probe_perf:found     (on __map_groups__find_by_name:10 in /home/acme/bin/perf with name:string)

  You can now use it in all perf tools, such as:

  	perf record -e probe_perf:found -aR sleep 1

  #
  # perf probe -x ~/bin/perf -L map_groups__find_by_name
  
        0  struct map *map_groups__find_by_name(struct map_groups *mg, const char *name)
        1  {
        2         struct maps *maps = &mg->maps;
                  struct map *map;

        5         down_read(&maps->lock);

        7         if (mg->last_search_by_name && strcmp(mg->last_search_by_name->dso->short_name, name) == 0) {
        8                 map = mg->last_search_by_name;
        9                 goto out_unlock;
                  }
                  /*
                   * If we have mg->maps_by_name, then the name isn't in the rbtree,
                   * as mg->maps_by_name mirrors the rbtree when lookups by name are
                   * made.
                   */
       16         map = __map_groups__find_by_name(mg, name);
       17         if (map || mg->maps_by_name != NULL)
       18                 goto out_unlock;

                  /* Fallback to traversing the rbtree... */
       21         maps__for_each_entry(maps, map)
       22                 if (strcmp(map->dso->short_name, name) == 0) {
       23                         mg->last_search_by_name = map;
       24                         goto out_unlock;
                          }

       27         map = NULL;

           out_unlock:
       30         up_read(&maps->lock);
       31         return map;
       32  }

           int dso__load_vmlinux(struct dso *dso, struct map *map,
                                const char *vmlinux, bool vmlinux_allocated)

  # perf probe -x ~/bin/perf 'fallback=map_groups__find_by_name:21 name:string'
  Added new events:
    probe_perf:fallback  (on map_groups__find_by_name:21 in /home/acme/bin/perf with name:string)
    probe_perf:fallback_1 (on map_groups__find_by_name:21 in /home/acme/bin/perf with name:string)

  You can now use it in all perf tools, such as:

  	perf record -e probe_perf:fallback_1 -aR sleep 1

  #
  # perf probe -l
    probe_perf:fallback  (on map_groups__find_by_name:21@util/symbol.c in /home/acme/bin/perf with name_string)
    probe_perf:fallback_1 (on map_groups__find_by_name:21@util/symbol.c in /home/acme/bin/perf with name_string)
    probe_perf:found     (on __map_groups__find_by_name:10@util/symbol.c in /home/acme/bin/perf with name_string)
  #
  # perf stat -e probe_perf:*

Now run 'perf top' in another term and then, after a while, stop 'perf stat':

Furthermore, if we ask for interval printing, we can see that that is done just
at the start of the workload:

  # perf stat -I1000 -e probe_perf:*
  #           time             counts unit events
       1.000319513                  0      probe_perf:found
       1.000319513                  0      probe_perf:fallback_1
       1.000319513                  0      probe_perf:fallback
       2.001868092             23,251      probe_perf:found
       2.001868092                  0      probe_perf:fallback_1
       2.001868092                  0      probe_perf:fallback
       3.002901597                  0      probe_perf:found
       3.002901597                  0      probe_perf:fallback_1
       3.002901597                  0      probe_perf:fallback
       4.003358591                  0      probe_perf:found
       4.003358591                  0      probe_perf:fallback_1
       4.003358591                  0      probe_perf:fallback
  ^C
  #

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-c5lmbyr14x448rcfii7y6t3k@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf map_groups: Add a front end cache for map lookups by name

2019-11-18T14:21:32+00:00

Lets see if it helps:

First look at the probeable lines for the function that does lookups by
name in a map_groups struct:

  # perf probe -x ~/bin/perf -L map_groups__find_by_name
  
        0  struct map *map_groups__find_by_name(struct map_groups *mg, const char *name)
        1  {
        2         struct maps *maps = &mg->maps;
                  struct map *map;

        5         down_read(&maps->lock);

        7         if (mg->last_search_by_name && strcmp(mg->last_search_by_name->dso->short_name, name) == 0) {
        8                 map = mg->last_search_by_name;
        9                 goto out_unlock;
                  }

       12         maps__for_each_entry(maps, map)
       13                 if (strcmp(map->dso->short_name, name) == 0) {
       14                         mg->last_search_by_name = map;
       15                         goto out_unlock;
                          }

       18         map = NULL;

           out_unlock:
       21         up_read(&maps->lock);
       22         return map;
       23  }

           int dso__load_vmlinux(struct dso *dso, struct map *map,
                                const char *vmlinux, bool vmlinux_allocated)

  #

Now add a probe to the place where we reuse the last search:

  # perf probe -x ~/bin/perf map_groups__find_by_name:8
  Added new event:
    probe_perf:map_groups__find_by_name (on map_groups__find_by_name:8 in /home/acme/bin/perf)

  You can now use it in all perf tools, such as:

  	perf record -e probe_perf:map_groups__find_by_name -aR sleep 1

  #

Now lets do a system wide 'perf stat' counting those events:

  # perf stat -e probe_perf:*

Leave it running and lets do a 'perf top', then, after a while, stop the
'perf stat':

  # perf stat -e probe_perf:*
  ^C
   Performance counter stats for 'system wide':

               3,603      probe_perf:map_groups__find_by_name

        44.565253139 seconds time elapsed
  #

yeah, good to have.

Cc: Adrian Hunter 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Link: https://lkml.kernel.org/n/tip-tcz37g3nxv3tvxw3q90vga3p@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo