Skip to contentSkip to conversation

perf script: Support filtering by hex address

PATCH In review Jin Yao 3 participants 9 messages Jan 25, 2021 – Jan 29, 2021 perf events

lore
Share & tools
lore

Jin Yao

author
#0 lore raw

PATCH 1/2 perf script: Support filtering by hex address

Perf-script supports '-S' or '--symbol' options to only list the trace records in given symbols. Symbol is typically a name or hex address. If it's hex address, it is the start address of one symbol.

While it would be useful if we can filter trace records by any hex address (not only the start address of symbol). So now we support filtering trace records by more conditions, such as:
- symbol name
- start address of symbol
- any hexadecimal address
- address range

The comparison order is defined as:

1. symbol name comparison
2. symbol start address comparison.
3. any hexadecimal address comparison.
4. address range comparison.

Let's see some examples:

root@kbl-ppc:~# ./perf script -S 0xffffffff9ca77308
            perf 18123 [000] 6142863.075104:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075107:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075108:         10   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075156:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075158:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075159:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [002] 6142863.075202:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [002] 6142863.075204:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [002] 6142863.075205:         16   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [003] 6142863.075250:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [003] 6142863.075252:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [003] 6142863.075253:         16   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])

Filter the traced records by hex address 0xffffffff9ca77308.

Easy to use, we support the hex address without '0x' prefix, e.g.
./perf script -S ffffffff9ca77308

It has the same effect.

We also support to filter trace records by a address range.

root@kbl-ppc:~# ./perf script -S ffffffff9ca77304 --addr-range 16
            perf 18123 [000] 6142863.075104:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075107:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075108:         10   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075109:        273   cycles:  ffffffff9ca7730a native_write_msr+0xa ([kernel.kallsyms])
            perf 18123 [001] 6142863.075156:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075158:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075159:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075160:        456   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])
            perf 18123 [002] 6142863.075202:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [002] 6142863.075204:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [002] 6142863.075205:         16   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [002] 6142863.075206:        436   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])
            perf 18123 [003] 6142863.075250:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [003] 6142863.075252:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [003] 6142863.075253:         16   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [003] 6142863.075254:        436   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])
            perf 18123 [004] 6142863.075299:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [004] 6142863.075301:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [004] 6142863.075302:         16   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [004] 6142863.075303:        431   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])
            perf 18123 [005] 6142863.075348:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [005] 6142863.075349:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [005] 6142863.075351:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [005] 6142863.075352:        454   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])
            perf 18123 [006] 6142863.075390:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [006] 6142863.075392:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [006] 6142863.075393:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [006] 6142863.075394:        459   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])
            perf 18123 [007] 6142863.075438:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [007] 6142863.075440:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [007] 6142863.075441:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [007] 6142863.075442:        466   cycles:  ffffffff9ca7730f native_write_msr+0xf ([kernel.kallsyms])

Filter the traced records from address range [0xffffffff9ca77304, 0xffffffff9ca77304 + 15].

---
 tools/perf/Documentation/perf-script.txt | 19 +++++++++++++
 tools/perf/builtin-script.c              |  2 ++
 tools/perf/util/event.c                  | 36 ++++++++++++++++++++++++
 tools/perf/util/symbol_conf.h            |  1 +
 4 files changed, 58 insertions(+)
tools/perf/Documentation/perf-script.txt +19 -0
index 44d37210fc8f..0ef2261baeb9 100644
@@ -422,9 +422,28 @@ include::itrace.txt[]
 	Only consider the listed symbols. Symbols are typically a name
 	but they may also be hexadecimal address.
 
+	The hexadecimal address may be the start address of a symbol or
+	any other address to filter the trace records.
+
 	For example, to select the symbol noploop or the address 0x4007a0:
 	perf script --symbols=noploop,0x4007a0
 
+	Support filtering trace records by symbol name, start address of
+	symbol, any hexadecimal address and address range.
+
+	The comparison order is:
+	1. symbol name comparison
+	2. symbol start address comparison.
+	3. any hexadecimal address comparison.
+	4. address range comparison (see --addr-range).
+
+--addr-range::
+	Use with -S or --symbols to list traced records within address range.
+
+	For example, to list the traced records within the address range
+	[0x4007a0, 0x0x4007a9]:
+	perf script -S 0x4007a0 --addr-range 10
+
 --call-trace::
 	Show call stream for intel_pt traces. The CPUs are interleaved, but
 	can be filtered with -C.
tools/perf/builtin-script.c +2 -0
index edacfa98d073..e0feda33dbb9 100644
@@ -3525,6 +3525,8 @@ int cmd_script(int argc, const char **argv)
 		    "system-wide collection from all CPUs"),
 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 		   "only consider these symbols"),
+	OPT_INTEGER(0, "addr-range", &symbol_conf.addr_range,
+		    "Use with -S to list traced records within address range"),
 	OPT_CALLBACK_OPTARG(0, "insn-trace", &itrace_synth_opts, NULL, NULL,
 			"Decode instructions from itrace", parse_insn_trace),
 	OPT_CALLBACK_OPTARG(0, "xed", NULL, NULL, NULL,
tools/perf/util/event.c +36 -0
index fbe8578e4c47..525b859cb445 100644
@@ -645,6 +645,22 @@ struct symbol *thread__find_symbol_fb(struct thread *thread, u8 cpumode,
 	return al->sym;
 }
 
+static bool check_address_range(struct strlist *sym_list, int addr_range,
+				struct addr_location *al)
+{
+	struct str_node *pos;
+	char *endptr;
+	u64 addr, al_addr = al->map->unmap_ip(al->map, al->addr);
+
+	strlist__for_each_entry(pos, sym_list) {
+		addr = strtoull(pos->s, &endptr, 16);
+		if (al_addr >= addr && al_addr < addr + addr_range)
+			return true;
+	}
+
+	return false;
+}
+
 /*
  * Callers need to drop the reference to al->thread, obtained in
  * machine__findnew_thread()
@@ -709,6 +725,26 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
 			ret = strlist__has_entry(symbol_conf.sym_list,
 						al_addr_str);
 		}
+		if (!ret && al->map) {
+			snprintf(al_addr_str, sz, "0x%"PRIx64,
+				al->map->unmap_ip(al->map, al->addr));
+			ret = strlist__has_entry(symbol_conf.sym_list,
+						 al_addr_str);
+			if (!ret) {
+				/* Check for hex without "0x" prefix */
+				snprintf(al_addr_str, sz, "%"PRIx64,
+					al->map->unmap_ip(al->map, al->addr));
+				ret = strlist__has_entry(symbol_conf.sym_list,
+							 al_addr_str);
+			}
+
+			if (!ret && symbol_conf.addr_range) {
+				ret = check_address_range(symbol_conf.sym_list,
+							  symbol_conf.addr_range,
+							  al);
+			}
+		}
+
 		if (!ret)
 			al->filtered |= (1 << HIST_FILTER__SYMBOL);
 	}
tools/perf/util/symbol_conf.h +1 -0
index b18f9c8dbb75..07e247cdb43c 100644
@@ -75,6 +75,7 @@ struct symbol_conf {
 	int		res_sample;
 	int		pad_output_len_dso;
 	int		group_sort_idx;
+	int		addr_range;
 };
 
 extern struct symbol_conf symbol_conf;
Signature · 1 line
2.17.1
author
#1 lore raw

PATCH 2/2 perf script: Support dso filter

Other perf tool builtins have already supported dso filter.

For example,
perf report --dso, which only considers symbols in these dsos.

Now dso filter is supported for perf-script.

root@kbl-ppc:~# ./perf script --dso "[kernel.kallsyms]"
            perf 18123 [000] 6142863.075104:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075107:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075108:         10   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075109:        273   cycles:  ffffffff9ca7730a native_write_msr+0xa ([kernel.kallsyms])
            perf 18123 [000] 6142863.075110:       7684   cycles:  ffffffff9ca3c9c0 native_sched_clock+0x50 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075112:     213017   cycles:  ffffffff9d765a92 syscall_exit_to_user_mode+0x32 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075156:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075158:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075159:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
---
 tools/perf/Documentation/perf-script.txt | 3 +++
 tools/perf/builtin-script.c              | 2 ++
 2 files changed, 5 insertions(+)
tools/perf/Documentation/perf-script.txt +3 -0
index 0ef2261baeb9..6e32b2f9b828 100644
@@ -444,6 +444,9 @@ include::itrace.txt[]
 	[0x4007a0, 0x0x4007a9]:
 	perf script -S 0x4007a0 --addr-range 10
 
+--dsos=::
+	Only consider symbols in these dsos.
+
 --call-trace::
 	Show call stream for intel_pt traces. The CPUs are interleaved, but
 	can be filtered with -C.
tools/perf/builtin-script.c +2 -0
index e0feda33dbb9..f8d3d8582ea2 100644
@@ -3523,6 +3523,8 @@ int cmd_script(int argc, const char **argv)
 		     parse_output_fields),
 	OPT_BOOLEAN('a', "all-cpus", &system_wide,
 		    "system-wide collection from all CPUs"),
+	OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
+		   "only consider symbols in these dsos"),
 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 		   "only consider these symbols"),
 	OPT_INTEGER(0, "addr-range", &symbol_conf.addr_range,
Signature · 1 line
2.17.1
Acked-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] perf script: Support dso filter

Quoted 18 lines · Jin Yao

Other perf tool builtins have already supported dso filter.

For example,
perf report --dso, which only considers symbols in these dsos.

Now dso filter is supported for perf-script.

root@kbl-ppc:~# ./perf script --dso "[kernel.kallsyms]"
            perf 18123 [000] 6142863.075104:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075107:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075108:         10   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075109:        273   cycles:  ffffffff9ca7730a native_write_msr+0xa ([kernel.kallsyms])
            perf 18123 [000] 6142863.075110:       7684   cycles:  ffffffff9ca3c9c0 native_sched_clock+0x50 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075112:     213017   cycles:  ffffffff9d765a92 syscall_exit_to_user_mode+0x32 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075156:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075158:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075159:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
Quoted 1 line

thanks,
jirka

Quoted 35 lines
---
 tools/perf/Documentation/perf-script.txt | 3 +++
 tools/perf/builtin-script.c              | 2 ++
 2 files changed, 5 insertions(+)
tools/perf/Documentation/perf-script.txt +3 -0
index 0ef2261baeb9..6e32b2f9b828 100644
@@ -444,6 +444,9 @@ include::itrace.txt[]
 	[0x4007a0, 0x0x4007a9]:
 	perf script -S 0x4007a0 --addr-range 10
 
+--dsos=::
+	Only consider symbols in these dsos.
+
 --call-trace::
 	Show call stream for intel_pt traces. The CPUs are interleaved, but
 	can be filtered with -C.
tools/perf/builtin-script.c +2 -0
index e0feda33dbb9..f8d3d8582ea2 100644
@@ -3523,6 +3523,8 @@ int cmd_script(int argc, const char **argv)
 		     parse_output_fields),
 	OPT_BOOLEAN('a', "all-cpus", &system_wide,
 		    "system-wide collection from all CPUs"),
+	OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
+		   "only consider symbols in these dsos"),
 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 		   "only consider these symbols"),
 	OPT_INTEGER(0, "addr-range", &symbol_conf.addr_range,
Signature · 1 line
2.17.1
#3 lore raw

Re: [PATCH 2/2] perf script: Support dso filter

In reply to Jin Yao

Em Mon, Jan 25, 2021 at 07:27:50AM +0800, Jin Yao escreveu:

Quoted 7 lines

Other perf tool builtins have already supported dso filter.

For example,
perf report --dso, which only considers symbols in these dsos.

Now dso filter is supported for perf-script.

Quoted 10 lines
root@kbl-ppc:~# ./perf script --dso "[kernel.kallsyms]"
            perf 18123 [000] 6142863.075104:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075107:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075108:         10   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075109:        273   cycles:  ffffffff9ca7730a native_write_msr+0xa ([kernel.kallsyms])
            perf 18123 [000] 6142863.075110:       7684   cycles:  ffffffff9ca3c9c0 native_sched_clock+0x50 ([kernel.kallsyms])
            perf 18123 [000] 6142863.075112:     213017   cycles:  ffffffff9d765a92 syscall_exit_to_user_mode+0x32 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075156:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075158:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
            perf 18123 [001] 6142863.075159:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])

Applied, and in testing it I noticed this fix was also needed, added just before it, please ack :-)

- Arnaldo

commit d4b6078bd58b799a8ea2a57842193504c867bee5

Author: Arnaldo Carvalho de Melo <[email protected]>
Date:   Thu Jan 28 09:52:47 2021 -0300
    perf tools: Fix DSO filtering when not finding a map for a sampled address
    
    When we lookup an address and don't find a map we should filter that
    sample if the user specified a list of --dso entries to filter on, fix
    it.
    
    Before:
    
      $ perf script
                 sleep 274800  2843.556162:          1 cycles:u:  ffffffffbb26bff4 [unknown] ([unknown])
                 sleep 274800  2843.556168:          1 cycles:u:  ffffffffbb2b047d [unknown] ([unknown])
                 sleep 274800  2843.556171:          1 cycles:u:  ffffffffbb2706b2 [unknown] ([unknown])
                 sleep 274800  2843.556174:          6 cycles:u:  ffffffffbb2b0267 [unknown] ([unknown])
                 sleep 274800  2843.556176:         59 cycles:u:  ffffffffbb2b03b1 [unknown] ([unknown])
                 sleep 274800  2843.556180:        691 cycles:u:  ffffffffbb26bff4 [unknown] ([unknown])
                 sleep 274800  2843.556189:       9160 cycles:u:      7fa9550eeaa3 __GI___tunables_init+0xf3 (/usr/lib64/ld-2.32.so)
                 sleep 274800  2843.556312:      86937 cycles:u:      7fa9550e157b _dl_lookup_symbol_x+0x4b (/usr/lib64/ld-2.32.so)
      $
    
    So we have some samples we somehow didn't find in a map for, if we now
    do:
    
      $ perf report --stdio --dso /usr/lib64/ld-2.32.so
      # dso: /usr/lib64/ld-2.32.so
      #
      # Total Lost Samples: 0
      #
      # Samples: 8  of event 'cycles:u'
      # Event count (approx.): 96856
      #
      # Overhead  Command  Symbol
      # ........  .......  ........................
      #
          89.76%  sleep    [.] _dl_lookup_symbol_x
           9.46%  sleep    [.] __GI___tunables_init
           0.71%  sleep    [k] 0xffffffffbb26bff4
           0.06%  sleep    [k] 0xffffffffbb2b03b1
           0.01%  sleep    [k] 0xffffffffbb2b0267
           0.00%  sleep    [k] 0xffffffffbb2706b2
           0.00%  sleep    [k] 0xffffffffbb2b047d
      $
    
    After this patch we get the right output with just entries for the DSOs
    specified in --dso:
    
      $ perf report --stdio --dso /usr/lib64/ld-2.32.so
      # dso: /usr/lib64/ld-2.32.so
      #
      # Total Lost Samples: 0
      #
      # Samples: 8  of event 'cycles:u'
      # Event count (approx.): 96856
      #
      # Overhead  Command  Symbol
      # ........  .......  ........................
      #
          89.76%  sleep    [.] _dl_lookup_symbol_x
           9.46%  sleep    [.] __GI___tunables_init
      $
    
      # To display the perf.data header info, please use --header/--header-only options.
      #
    
    Cc: Alexander Shishkin <[email protected]>
    Cc: Andi Kleen <[email protected]>
    Cc: Ingo Molnar <[email protected]>
    Cc: Jin Yao <[email protected]>
    Cc: Jiri Olsa <[email protected]>
    Cc: Kan Liang <[email protected]>
    Cc: Namhyung Kim <[email protected]>
    Cc: Peter Zijlstra <[email protected]>
    Fixes: 96415e4d3f5fdf9c ("perf symbols: Avoid unnecessary symbol loading when dso list is specified")
    Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
tools/perf/util/event.c +2 -0
index fbe8578e4c475773..208b6c141d98bb5d 100644
@@ -692,6 +692,8 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
 		}
 
 		al->sym = map__find_symbol(al->map, al->addr);
+	} else if (symbol_conf.dso_list) {
+		al->filtered |= (1 << HIST_FILTER__DSO);
 	}
 
 	if (symbol_conf.sym_list) {
author Acked-by
#4 lore raw

Re: [PATCH 2/2] perf script: Support dso filter

Hi Arnaldo,

Quoted 108 lines · Arnaldo Carvalho de Melo

Em Mon, Jan 25, 2021 at 07:27:50AM +0800, Jin Yao escreveu:

Quoted 7 lines

Other perf tool builtins have already supported dso filter.

For example,
perf report --dso, which only considers symbols in these dsos.

Now dso filter is supported for perf-script.

Quoted 10 lines
root@kbl-ppc:~# ./perf script --dso "[kernel.kallsyms]"
             perf 18123 [000] 6142863.075104:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
             perf 18123 [000] 6142863.075107:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
             perf 18123 [000] 6142863.075108:         10   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
             perf 18123 [000] 6142863.075109:        273   cycles:  ffffffff9ca7730a native_write_msr+0xa ([kernel.kallsyms])
             perf 18123 [000] 6142863.075110:       7684   cycles:  ffffffff9ca3c9c0 native_sched_clock+0x50 ([kernel.kallsyms])
             perf 18123 [000] 6142863.075112:     213017   cycles:  ffffffff9d765a92 syscall_exit_to_user_mode+0x32 ([kernel.kallsyms])
             perf 18123 [001] 6142863.075156:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
             perf 18123 [001] 6142863.075158:          1   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])
             perf 18123 [001] 6142863.075159:         17   cycles:  ffffffff9ca77308 native_write_msr+0x8 ([kernel.kallsyms])

Applied, and in testing it I noticed this fix was also needed, added just before it, please ack :-)

- Arnaldo

commit d4b6078bd58b799a8ea2a57842193504c867bee5

Author: Arnaldo Carvalho de Melo <[email protected]>
Date:   Thu Jan 28 09:52:47 2021 -0300
     perf tools: Fix DSO filtering when not finding a map for a sampled address
     
     When we lookup an address and don't find a map we should filter that
     sample if the user specified a list of --dso entries to filter on, fix
     it.
     
     Before:
     
       $ perf script
                  sleep 274800  2843.556162:          1 cycles:u:  ffffffffbb26bff4 [unknown] ([unknown])
                  sleep 274800  2843.556168:          1 cycles:u:  ffffffffbb2b047d [unknown] ([unknown])
                  sleep 274800  2843.556171:          1 cycles:u:  ffffffffbb2706b2 [unknown] ([unknown])
                  sleep 274800  2843.556174:          6 cycles:u:  ffffffffbb2b0267 [unknown] ([unknown])
                  sleep 274800  2843.556176:         59 cycles:u:  ffffffffbb2b03b1 [unknown] ([unknown])
                  sleep 274800  2843.556180:        691 cycles:u:  ffffffffbb26bff4 [unknown] ([unknown])
                  sleep 274800  2843.556189:       9160 cycles:u:      7fa9550eeaa3 __GI___tunables_init+0xf3 (/usr/lib64/ld-2.32.so)
                  sleep 274800  2843.556312:      86937 cycles:u:      7fa9550e157b _dl_lookup_symbol_x+0x4b (/usr/lib64/ld-2.32.so)
       $
     
     So we have some samples we somehow didn't find in a map for, if we now
     do:
     
       $ perf report --stdio --dso /usr/lib64/ld-2.32.so
       # dso: /usr/lib64/ld-2.32.so
       #
       # Total Lost Samples: 0
       #
       # Samples: 8  of event 'cycles:u'
       # Event count (approx.): 96856
       #
       # Overhead  Command  Symbol
       # ........  .......  ........................
       #
           89.76%  sleep    [.] _dl_lookup_symbol_x
            9.46%  sleep    [.] __GI___tunables_init
            0.71%  sleep    [k] 0xffffffffbb26bff4
            0.06%  sleep    [k] 0xffffffffbb2b03b1
            0.01%  sleep    [k] 0xffffffffbb2b0267
            0.00%  sleep    [k] 0xffffffffbb2706b2
            0.00%  sleep    [k] 0xffffffffbb2b047d
       $
     
     After this patch we get the right output with just entries for the DSOs
     specified in --dso:
     
       $ perf report --stdio --dso /usr/lib64/ld-2.32.so
       # dso: /usr/lib64/ld-2.32.so
       #
       # Total Lost Samples: 0
       #
       # Samples: 8  of event 'cycles:u'
       # Event count (approx.): 96856
       #
       # Overhead  Command  Symbol
       # ........  .......  ........................
       #
           89.76%  sleep    [.] _dl_lookup_symbol_x
            9.46%  sleep    [.] __GI___tunables_init
       $
     
       # To display the perf.data header info, please use --header/--header-only options.
       #
     
     Cc: Alexander Shishkin <[email protected]>
     Cc: Andi Kleen <[email protected]>
     Cc: Ingo Molnar <[email protected]>
     Cc: Jin Yao <[email protected]>
     Cc: Jiri Olsa <[email protected]>
     Cc: Kan Liang <[email protected]>
     Cc: Namhyung Kim <[email protected]>
     Cc: Peter Zijlstra <[email protected]>
     Fixes: 96415e4d3f5fdf9c ("perf symbols: Avoid unnecessary symbol loading when dso list is specified")
     Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
tools/perf/util/event.c +0 -0
index fbe8578e4c475773..208b6c141d98bb5d 100644

tools/perf/util/event.c int machine__resolve(struct machine *machine, struct addr_location *al, · line 699

Quoted 8 lines
tools/perf/util/event.c +2 -0
@@ -692,6 +692,8 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
  		}
 
  		al->sym = map__find_symbol(al->map, al->addr);
+	} else if (symbol_conf.dso_list) {
+		al->filtered |= (1 << HIST_FILTER__DSO);
  	}
 
  	if (symbol_conf.sym_list) {

This fix guarantees the "[unknown]" dso is filtered when we enable "--dso" option (whatever the dso list in --dso).

Thanks
Jin Yao

#5 lore raw
In reply to Jin Yao

On Mon, Jan 25, 2021 at 07:27:49AM +0800, Jin Yao wrote:

SNIP

Quoted 67 lines
tools/perf/Documentation/perf-script.txt +16 -0
 	perf script --symbols=noploop,0x4007a0
 
+	Support filtering trace records by symbol name, start address of
+	symbol, any hexadecimal address and address range.
+
+	The comparison order is:
+	1. symbol name comparison
+	2. symbol start address comparison.
+	3. any hexadecimal address comparison.
+	4. address range comparison (see --addr-range).
+
+--addr-range::
+	Use with -S or --symbols to list traced records within address range.
+
+	For example, to list the traced records within the address range
+	[0x4007a0, 0x0x4007a9]:
+	perf script -S 0x4007a0 --addr-range 10
+
 --call-trace::
 	Show call stream for intel_pt traces. The CPUs are interleaved, but
 	can be filtered with -C.
tools/perf/builtin-script.c +2 -0
index edacfa98d073..e0feda33dbb9 100644
@@ -3525,6 +3525,8 @@ int cmd_script(int argc, const char **argv)
 		    "system-wide collection from all CPUs"),
 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 		   "only consider these symbols"),
+	OPT_INTEGER(0, "addr-range", &symbol_conf.addr_range,
+		    "Use with -S to list traced records within address range"),
 	OPT_CALLBACK_OPTARG(0, "insn-trace", &itrace_synth_opts, NULL, NULL,
 			"Decode instructions from itrace", parse_insn_trace),
 	OPT_CALLBACK_OPTARG(0, "xed", NULL, NULL, NULL,
tools/perf/util/event.c +18 -0
index fbe8578e4c47..525b859cb445 100644
@@ -645,6 +645,22 @@ struct symbol *thread__find_symbol_fb(struct thread *thread, u8 cpumode,
 	return al->sym;
 }
 
+static bool check_address_range(struct strlist *sym_list, int addr_range,
+				struct addr_location *al)
+{
+	struct str_node *pos;
+	char *endptr;
+	u64 addr, al_addr = al->map->unmap_ip(al->map, al->addr);
+
+	strlist__for_each_entry(pos, sym_list) {
+		addr = strtoull(pos->s, &endptr, 16);
+		if (al_addr >= addr && al_addr < addr + addr_range)
+			return true;
+	}
+
+	return false;
+}
+
 /*
  * Callers need to drop the reference to al->thread, obtained in
  * machine__findnew_thread()
@@ -709,6 +725,26 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
 			ret = strlist__has_entry(symbol_conf.sym_list,
 						al_addr_str);
 		}
+		if (!ret && al->map) {
+			snprintf(al_addr_str, sz, "0x%"PRIx64,

tools/perf/util/event.c int machine__resolve(struct machine *machine, struct addr_location *al, · line 739

Quoted 10 lines
tools/perf/util/event.c +10 -0
+				al->map->unmap_ip(al->map, al->addr));
+			ret = strlist__has_entry(symbol_conf.sym_list,
+						 al_addr_str);
+			if (!ret) {
+				/* Check for hex without "0x" prefix */
+				snprintf(al_addr_str, sz, "%"PRIx64,
+					al->map->unmap_ip(al->map, al->addr));
+				ret = strlist__has_entry(symbol_conf.sym_list,
+							 al_addr_str);
+			}

that seems tricky.. what if user specify more leading zeros, I think it'd be better to search intlist instead

we could move all 'address' entries from sym_list to new intlist (in symbol__init) and use it for this search

jirka

author
#6 lore raw

Hi Jiri,

Quoted 19 lines · Jiri Olsa
Quoted 6 lines
tools/perf/util/event.c +2 -0
@@ -709,6 +725,26 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
  			ret = strlist__has_entry(symbol_conf.sym_list,
  						al_addr_str);
  		}
+		if (!ret && al->map) {
+			snprintf(al_addr_str, sz, "0x%"PRIx64,

tools/perf/util/event.c int machine__resolve(struct machine *machine, struct addr_location *al, · line 739

Quoted 10 lines
tools/perf/util/event.c +10 -0
+				al->map->unmap_ip(al->map, al->addr));
+			ret = strlist__has_entry(symbol_conf.sym_list,
+						 al_addr_str);
+			if (!ret) {
+				/* Check for hex without "0x" prefix */
+				snprintf(al_addr_str, sz, "%"PRIx64,
+					al->map->unmap_ip(al->map, al->addr));
+				ret = strlist__has_entry(symbol_conf.sym_list,
+							 al_addr_str);
+			}

that seems tricky.. what if user specify more leading zeros, I think it'd be better to search intlist instead

Quoted 5 lines

we could move all 'address' entries from sym_list to new intlist (in symbol__init) and use it for this search

jirka

intlist now only supports 'int' type.

I'm considering to use 'long' to replace original 'int' in struct int_node.

struct int_node {
	struct rb_node rb_node;
-	int i;
+	long i;
	void *priv;
};

On 32 bits platform, sizeof(long) is 4, and on 64 bits platform, sizeof(long) is 8. So that should be OK for storing the values such as pid/tid and address.

I'm not very clear why currently it uses 'int' for i in struct int_node, maybe something I don't understand correctly. Please correct me if my understanding is wrong.

Thanks
Jin Yao

#7 lore raw

Em Thu, Jan 28, 2021 at 11:31:26AM +0800, Jin, Yao escreveu:

Quoted 33 lines · Jiri Olsa
Quoted 18 lines
Quoted 5 lines
tools/perf/util/event.c +2 -0
@@ -709,6 +725,26 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
  			ret = strlist__has_entry(symbol_conf.sym_list, al_addr_str);
  		}
+		if (!ret && al->map) {
+			snprintf(al_addr_str, sz, "0x%"PRIx64,

tools/perf/util/event.c int machine__resolve(struct machine *machine, struct addr_location *al, · line 738

Quoted 10 lines
tools/perf/util/event.c +10 -0
+				al->map->unmap_ip(al->map, al->addr));
+			ret = strlist__has_entry(symbol_conf.sym_list,
+						 al_addr_str);
+			if (!ret) {
+				/* Check for hex without "0x" prefix */
+				snprintf(al_addr_str, sz, "%"PRIx64,
+					al->map->unmap_ip(al->map, al->addr));
+				ret = strlist__has_entry(symbol_conf.sym_list,
+							 al_addr_str);
+			}

that seems tricky.. what if user specify more leading zeros, I think it'd be better to search intlist instead

Quoted 2 lines

we could move all 'address' entries from sym_list to new intlist (in symbol__init) and use it for this search

intlist now only supports 'int' type.

I'm considering to use 'long' to replace original 'int' in struct int_node.

struct int_node {
	struct rb_node rb_node;
(quoted diff) +1 -1
-	int i;
+	long i;
	void *priv;
};
Quoted 7 lines

On 32 bits platform, sizeof(long) is 4, and on 64 bits platform, sizeof(long) is 8. So that should be OK for storing the values such as pid/tid and address.

I'm not very clear why currently it uses 'int' for i in struct int_node, maybe something I don't understand correctly. Please correct me if my understanding is wrong.

I think its ok to make it a long, it won't even enlarge the struct because:

[acme@quaco pahole]$ pahole -C int_node ~/bin/perf

struct int_node {
	struct rb_node             rb_node __attribute__((__aligned__(8))); /*     0    24 */
	int                        i;                                       /*    24     4 */

	/* XXX 4 bytes hole, try to pack */

	void *                     priv;                                    /*    32     8 */

	/* size: 40, cachelines: 1, members: 3 */
	/* sum members: 36, holes: 1, sum holes: 4 */
	/* forced alignments: 1 */
	/* last cacheline: 40 bytes */
} __attribute__((__aligned__(8)));

[acme@quaco pahole]$

:-)

- Arnaldo

author
#8 lore raw
Quoted 59 lines · Arnaldo Carvalho de Melo

Em Thu, Jan 28, 2021 at 11:31:26AM +0800, Jin, Yao escreveu:

Quoted 33 lines · Jiri Olsa
Quoted 18 lines
Quoted 5 lines
tools/perf/util/event.c +2 -0
@@ -709,6 +725,26 @@ int machine__resolve(struct machine *machine, struct addr_location *al,
   			ret = strlist__has_entry(symbol_conf.sym_list, al_addr_str);
   		}
+		if (!ret && al->map) {
+			snprintf(al_addr_str, sz, "0x%"PRIx64,

tools/perf/util/event.c int machine__resolve(struct machine *machine, struct addr_location *al, · line 738

Quoted 10 lines
tools/perf/util/event.c +10 -0
+				al->map->unmap_ip(al->map, al->addr));
+			ret = strlist__has_entry(symbol_conf.sym_list,
+						 al_addr_str);
+			if (!ret) {
+				/* Check for hex without "0x" prefix */
+				snprintf(al_addr_str, sz, "%"PRIx64,
+					al->map->unmap_ip(al->map, al->addr));
+				ret = strlist__has_entry(symbol_conf.sym_list,
+							 al_addr_str);
+			}

that seems tricky.. what if user specify more leading zeros, I think it'd be better to search intlist instead

Quoted 2 lines

we could move all 'address' entries from sym_list to new intlist (in symbol__init) and use it for this search

intlist now only supports 'int' type.

I'm considering to use 'long' to replace original 'int' in struct int_node.

struct int_node {
	struct rb_node rb_node;
(quoted diff) +1 -1
-	int i;
+	long i;
	void *priv;
};
Quoted 7 lines

On 32 bits platform, sizeof(long) is 4, and on 64 bits platform, sizeof(long) is 8. So that should be OK for storing the values such as pid/tid and address.

I'm not very clear why currently it uses 'int' for i in struct int_node, maybe something I don't understand correctly. Please correct me if my understanding is wrong.

I think its ok to make it a long, it won't even enlarge the struct because:

[acme@quaco pahole]$ pahole -C int_node ~/bin/perf

struct int_node {
	struct rb_node             rb_node __attribute__((__aligned__(8))); /*     0    24 */
	int                        i;                                       /*    24     4 */

	/* XXX 4 bytes hole, try to pack */

	void *                     priv;                                    /*    32     8 */

	/* size: 40, cachelines: 1, members: 3 */
	/* sum members: 36, holes: 1, sum holes: 4 */
	/* forced alignments: 1 */
	/* last cacheline: 40 bytes */
} __attribute__((__aligned__(8)));
Quoted 6 lines

[acme@quaco pahole]$

:-)

- Arnaldo

Thanks Arnaldo!

OK, let me try to improve the intlist.

Thanks
Jin Yao