perf probe: Check address correctness by map instead of _etext
authorMasami Hiramatsu <mhiramat@kernel.org>
Thu, 23 Apr 2020 11:01:13 +0000 (20:01 +0900)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Thu, 28 May 2020 13:03:24 +0000 (10:03 -0300)
Since commit 03db8b583d1c ("perf tools: Fix
maps__find_symbol_by_name()") introduced map address range check in
maps__find_symbol_by_name(), we can not get "_etext" from kernel map
because _etext is placed on the edge of the kernel .text section (=
kernel map in perf.)

To fix this issue, this checks the address correctness by map address
range information (map->start and map->end) instead of using _etext
address.

This can cause an error if the target inlined function is embedded in
both __init function and normal function.

For exaample, request_resource() is a normal function but also embedded
in __init reserve_setup(). In this case, the probe point in
reserve_setup() must be skipped.

However, without this fix, it failes to setup all probe points:

  # ./perf probe -v request_resource
  probe-definition(0): request_resource
  symbol:request_resource file:(null) line:0 offset:0 return:0 lazy:(null)
  0 arguments
  Looking at the vmlinux_path (8 entries long)
  Using /usr/lib/debug/lib/modules/5.5.17-200.fc31.x86_64/vmlinux for symbols
  Open Debuginfo file: /usr/lib/debug/lib/modules/5.5.17-200.fc31.x86_64/vmlinux
  Try to find probe point from debuginfo.
  Matched function: request_resource [15e29ad]
  found inline addr: 0xffffffff82fbf892
  Probe point found: reserve_setup+204
  found inline addr: 0xffffffff810e9790
  Probe point found: request_resource+0
  Found 2 probe_trace_events.
  Opening /sys/kernel/debug/tracing//kprobe_events write=1
  Opening /sys/kernel/debug/tracing//README write=0
  Writing event: p:probe/request_resource _text+33290386
  Failed to write event: Invalid argument
    Error: Failed to add events. Reason: Invalid argument (Code: -22)
  #

With this fix,

  # ./perf probe request_resource
  reserve_setup is out of .text, skip it.
  Added new events:
    (null):(null)        (on request_resource)
    probe:request_resource (on request_resource)

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

   perf record -e probe:request_resource -aR sleep 1

  #

Fixes: 03db8b583d1c ("perf tools: Fix maps__find_symbol_by_name()")
Reported-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/158763967332.30755.4922496724365529088.stgit@devnote2
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/probe-event.c

index 63d936f..a08f373 100644 (file)
@@ -236,21 +236,22 @@ static void clear_probe_trace_events(struct probe_trace_event *tevs, int ntevs)
 static bool kprobe_blacklist__listed(unsigned long address);
 static bool kprobe_warn_out_range(const char *symbol, unsigned long address)
 {
-       u64 etext_addr = 0;
-       int ret;
-
-       /* Get the address of _etext for checking non-probable text symbol */
-       ret = kernel_get_symbol_address_by_name("_etext", &etext_addr,
-                                               false, false);
+       struct map *map;
+       bool ret = false;
 
-       if (ret == 0 && etext_addr < address)
-               pr_warning("%s is out of .text, skip it.\n", symbol);
-       else if (kprobe_blacklist__listed(address))
+       map = kernel_get_module_map(NULL);
+       if (map) {
+               ret = address <= map->start || map->end < address;
+               if (ret)
+                       pr_warning("%s is out of .text, skip it.\n", symbol);
+               map__put(map);
+       }
+       if (!ret && kprobe_blacklist__listed(address)) {
                pr_warning("%s is blacklisted function, skip it.\n", symbol);
-       else
-               return false;
+               ret = true;
+       }
 
-       return true;
+       return ret;
 }
 
 /*