Re: [PATCH] perf probe: Fix module probing with shortname

[Arnaldo Carvalho de Melo <acme@redhat.com>]

Em Tue, Sep 22, 2015 at 03:34:32AM +0000, Wang Nan escreveu:
> After commit 3d39ac538629e4f00a6e1c38d46346f1b8e69505 ("perf machine:
> No need to have two DSOs lists"), perf probe with module short name doesn't
> work again. For example:
> 
>  # lsmod | grep e1000e
>  e1000e                233472  0
> 
>  # cat /proc/modules | grep e1000e
>  e1000e 233472 0 - Live 0xffffffffa0073000
> 
>  # cat /proc/kallsyms | grep '\<e1000e_up\>'
>  ffffffffa0093860 t e1000e_up[e1000e]
> 
>  # perf probe -v -m e1000e --add e1000e_up
>  probe-definition(0): e1000e_up
>  symbol:e1000e_up file:(null) line:0 offset:0 return:0 lazy:(null)
>  0 arguments
>  Failed to find module e1000e.
>  Could not open debuginfo. Try to use symbols.
>  Looking at the vmlinux_path (7 entries long)
>  Using /lib/modules/4.2.0-rc7+/build/vmlinux for symbols
>  e1000e_up is out of .text, skip it.
>    Error: Failed to add events. Reason: No such file or directory (Code: -2)
> 
> This is caused by a misunderstood of dso->kernel in kernel_get_module_dso()
> that, for kernel module, dso->kernel is DSO_TYPE_USER. dso->kernel is DSO_TYPE_KERNEL
> iff dso is vmlinux.

Kernel modules having DSO_TYPE_USER seems to be the bug, no? I'll try to
check that...

- Arnaldo

> 
> This patch fix 'perf probe -m' with an ad-hoc way.
> 
> After this patch:
> 
>  # perf probe -v -m e1000e --add e1000e_up
>  probe-definition(0): e1000e_up
>  symbol:e1000e_up file:(null) line:0 offset:0 return:0 lazy:(null)
>  0 arguments
>  Open Debuginfo file: /lib/modules/4.2.0-rc7+/kernel/drivers/net/ethernet/intel/e1000e/e1000e.ko
>  Try to find probe point from debuginfo.
>  Matched function: e1000e_up
>  Probe point found: e1000e_up+0
>  Found 1 probe_trace_events.
>  Opening /sys/kernel/debug/tracing//kprobe_events write=1
>  Writing event: p:probe/e1000e_up e1000e:e1000e_up+0
>  Added new event:
>    probe:e1000e_up      (on e1000e_up in e1000e)
> 
>  You can now use it in all perf tools, such as:
> 
>  	perf record -e probe:e1000e_up -aR sleep 1
> 
>  # perf probe -l
>  Failed to find debug information for address ffffffffa0093860
>    probe:e1000e_up      (on e1000e_up in e1000e)
> 
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
> Cc: Namhyung Kim <namhyung@kernel.org>
> Cc: Jiri Olsa <jolsa@redhat.com>
> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> ---
> 
> I think there may be other places where dso->kernel is misused.
> machine__process_kernel_mmap_event() may be one of them. If I understand
> correctly, 'dso->kernel && is_kernel_module(dso->long_name)' should always
> false theoretically. However, I don't have enough time to check whether that
> code really cause problem.
> 
> ---
>  tools/perf/util/probe-event.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
> index 2b78e8f..c7d6d3d 100644
> --- a/tools/perf/util/probe-event.c
> +++ b/tools/perf/util/probe-event.c
> @@ -270,7 +270,7 @@ static int kernel_get_module_dso(const char *module, struct dso **pdso)
>  
>  	if (module) {
>  		list_for_each_entry(dso, &host_machine->dsos.head, node) {
> -			if (!dso->kernel)
> +			if (dso->kernel)
>  				continue;
>  			if (strncmp(dso->short_name + 1, module,
>  				    dso->short_name_len - 2) == 0)
> -- 
> 1.8.3.4