Re: [PATCH v3 2/2] perf: report/annotate: fix segfault problem.

[Wang Nan <wangnan0@huawei.com>]

On 2015/4/7 23:13, Arnaldo Carvalho de Melo wrote:
> Em Tue, Apr 07, 2015 at 08:22:46AM +0000, Wang Nan escreveu:
>> perf report and perf annotate are easy to trigger segfault if trace data
>> contain kernel module information like this:
>>
>>  # perf report -D -i ./perf.data
>>  ...
>>  0 0 0x188 [0x50]: PERF_RECORD_MMAP -1/0: [0xffffffbff1018000(0xf068000) @ 0]: x [test_module]
>>  ...
>>
>>  # perf report -i ./perf.data --objdump=/path/to/objdump --kallsyms=/path/to/kallsyms
>>
>>  perf: Segmentation fault
>>  -------- backtrace --------
>>  /path/to/perf[0x503478]
>>  /lib64/libc.so.6(+0x3545f)[0x7fb201f3745f]
>>  /path/to/perf[0x499b56]
> 
> So, with this patch applied I am seeing this:
> 
> [root@ssdandy ~]# perf sched record -a
> Kmod (modules.order) and cpumode (0) inconsistent
> Kmod (modules.builtin) and cpumode (0) inconsistent
> Kmod (modules.dep) and cpumode (0) inconsistent
> Kmod (modules.dep.bin) and cpumode (0) inconsistent
> Kmod (modules.alias) and cpumode (0) inconsistent
> Kmod (modules.alias.bin) and cpumode (0) inconsistent
> Kmod (modules.softdep) and cpumode (0) inconsistent
> Kmod (modules.symbols) and cpumode (0) inconsistent
> Kmod (modules.symbols.bin) and cpumode (0) inconsistent
> Kmod (modules.builtin.bin) and cpumode (0) inconsistent
> Kmod (modules.devname) and cpumode (0) inconsistent
> ^C[ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 0.465 MB perf.data (1595 samples) ]
> 
> [root@ssdandy ~]#o
> 
> Can you please check this?
> 
> Why do we get these warning for these files?
> 

This is because in map_groups__set_modules_path_dir() we enumerate over each file
in /lib/modules/`uname -a`/ and use kmod_path__parse_name() to parse their names.
In this v3 patch, I update __kmod_path__parse() to check consistency between m.kmod and
cpumode. It will warn if one passes a kernel module name with user
space cpumode, or passes kernel space cpumode with a non-kmodule name.

I'll post a patch v4 to avoid such warning if passed cpumode is not explicitly set to
user or kernel.

Difference from v3:

diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index fb1ed7d..875f7c9 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -215,7 +215,7 @@ int __kmod_path__parse(struct kmod_path *m, const char *path,
        const char *name = strrchr(path, '/');
        const char *ext  = strrchr(path, '.');
        bool is_simple_name = false;
-       bool cpu_mode_kernel, is_kernel = false;
+       bool cpu_mode_kernel, cpu_mode_unknown = false, is_kernel = false;

        /* treat PERF_RECORD_MISC_CPUMODE_UNKNOWN as kernel. */
        switch (cpumode & PERF_RECORD_MISC_CPUMODE_MASK) {
@@ -224,6 +224,9 @@ int __kmod_path__parse(struct kmod_path *m, const char *path,
        case PERF_RECORD_MISC_GUEST_USER:
                cpu_mode_kernel = false;
                break;
+       case PERF_RECORD_MISC_CPUMODE_UNKNOWN:
+               cpu_mode_unknown = true;
+               /* fall through */
        default:
                cpu_mode_kernel = true;
        }
@@ -288,10 +291,12 @@ int __kmod_path__parse(struct kmod_path *m, const char *path,
                }
        }
 out:
-       if ((m->kmod && !cpu_mode_kernel) ||
-                       (cpu_mode_kernel && !m->kmod && !is_kernel))
-               pr_warning("Kmod (%s) and cpumode (%d) inconsistent\n",
-                               path, cpumode);
+       if (!cpu_mode_unknown) {
+               if ((m->kmod && !cpu_mode_kernel) ||
+                               (cpu_mode_kernel && !m->kmod && !is_kernel))
+                       pr_warning("Kmod (%s) and cpumode (%d) inconsistent\n",
+                                       path, cpumode);
+       }

        return 0;
 }

> 
> Applied the other patch:
> 
> -> 28 T 04/07 Wang Nan (8,8K) [PATCH v3 1/2] perf: kmaps: enforce usage of kmaps to protect futher bugs.
> 
> 
> - Arnaldo
>