lkml.org 
[lkml]   [2021]   [Jul]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] perf probe: Fix add event failed when 32-bit perf running in 64-bit kernel
From
Date
Hello Hiramatsu,

On 2021/7/15 12:58, Masami Hiramatsu wrote:
> On Thu, 15 Jul 2021 09:20:01 +0800
> Yang Jihong <yangjihong1@huawei.com> wrote:
>
>> Hello Hiramatsu,
>>
>> On 2021/7/14 16:35, Masami Hiramatsu wrote:
>>> Hi Yang,
>>>
>>> On Wed, 14 Jul 2021 14:54:32 +0800
>>> Yang Jihong <yangjihong1@huawei.com> wrote:
>>>
>>>> The "address" member of "struct probe_trace_point" uses long data type.
>>>> If kernel is 64-bit and perf program is 32-bit, size of "address" variable is
>>>> 32 bits. As a result, upper 32 bits of address read from kernel are truncated,
>>>> An error occurs during address comparison in kprobe_warn_out_range function.
>>>
>>> Good catch!
>>> I didn't imagine that such a use case. But that is important because perf
>>> probe can be used for cross-arch probe definition too.
>>>
>>>>
>>>> Before:
>>>>
>>>> # perf probe -a schedule
>>>> schedule is out of .text, skip it.
>>>> Error: Failed to add events.
>>>>
>>>> Solution:
>>>> Change data type of "address" variable to u64 and change corresponding
>>>> address printing and value assignment.
>>>
>>> OK, as far as I can see, the other parts of the perf also uses u64 for
>>> "address" storing variables. (e.g. symbols, maps etc.)
>>>
>>>>
>>>> After:
>>>>
>>>> # perf.new.new probe -a schedule
>>>> Added new event:
>>>> probe:schedule (on schedule)
>>>>
>>>> You can now use it in all perf tools, such as:
>>>>
>>>> perf record -e probe:schedule -aR sleep 1
>>>>
>>>> # perf probe -l
>>>> probe:schedule (on schedule)
>>>
>>> I think you missed one thing here.
>>> Usually, this shows the filename and line number of schedule().
>> Yes, I tried the following diff and now it can show the filename (as is
>> function entry, relative line number is 0), thank you very much :)
>>
>> The test result in my environment is as follows:
>>
>> # perf probe -a schedule
>> Added new event:
>> probe:schedule (on schedule)
>>
>> You can now use it in all perf tools, such as:
>>
>> perf record -e probe:schedule -aR sleep 1
>>
>> # perf probe -l schedule -k vmlinux.debug
>> probe:schedule (on schedule@kernel/sched/core.c)
>>
>
> OK, good!
>
>>
>> Can I put the following diff together and submit a v2 patch?
>
> Yes, please include this fix in your patch, because it is what your patch
> has to do :)
>
OK,I have submitted the v2 patch:
https://lore.kernel.org/patchwork/patch/1460154/

Thanks very much for your patient review.:)

Jihong
> Thank you!
>
>>>
>>> Could you try below diff?
>>>
>>> diff --git a/tools/perf/util/dwarf-aux.c b/tools/perf/util/dwarf-aux.c
>>> index 7d2ba8419b0c..609ca1671501 100644
>>> --- a/tools/perf/util/dwarf-aux.c
>>> +++ b/tools/perf/util/dwarf-aux.c
>>> @@ -113,14 +113,14 @@ static Dwarf_Line *cu_getsrc_die(Dwarf_Die *cu_die, Dwarf_Addr addr)
>>> *
>>> * Find a line number and file name for @addr in @cu_die.
>>> */
>>> -int cu_find_lineinfo(Dwarf_Die *cu_die, unsigned long addr,
>>> - const char **fname, int *lineno)
>>> +int cu_find_lineinfo(Dwarf_Die *cu_die, Dwarf_Addr addr,
>>> + const char **fname, int *lineno)
>>> {
>>> Dwarf_Line *line;
>>> Dwarf_Die die_mem;
>>> Dwarf_Addr faddr;
>>>
>>> - if (die_find_realfunc(cu_die, (Dwarf_Addr)addr, &die_mem)
>>> + if (die_find_realfunc(cu_die, addr, &die_mem)
>>> && die_entrypc(&die_mem, &faddr) == 0 &&
>>> faddr == addr) {
>>> *fname = dwarf_decl_file(&die_mem);
>>> @@ -128,7 +128,7 @@ int cu_find_lineinfo(Dwarf_Die *cu_die, unsigned long addr,
>>> goto out;
>>> }
>>>
>>> - line = cu_getsrc_die(cu_die, (Dwarf_Addr)addr);
>>> + line = cu_getsrc_die(cu_die, addr);
>>> if (line && dwarf_lineno(line, lineno) == 0) {
>>> *fname = dwarf_linesrc(line, NULL, NULL);
>>> if (!*fname)
>>> diff --git a/tools/perf/util/dwarf-aux.h b/tools/perf/util/dwarf-aux.h
>>> index cb99646843a9..7ee0fa19b5c4 100644
>>> --- a/tools/perf/util/dwarf-aux.h
>>> +++ b/tools/perf/util/dwarf-aux.h
>>> @@ -19,7 +19,7 @@ const char *cu_find_realpath(Dwarf_Die *cu_die, const char *fname);
>>> const char *cu_get_comp_dir(Dwarf_Die *cu_die);
>>>
>>> /* Get a line number and file name for given address */
>>> -int cu_find_lineinfo(Dwarf_Die *cudie, unsigned long addr,
>>> +int cu_find_lineinfo(Dwarf_Die *cudie, Dwarf_Addr addr,
>>> const char **fname, int *lineno);
>>>
>>> /* Walk on functions at given address */
>>> diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
>>> index 8ba01bbdf05c..50d861a80f57 100644
>>> --- a/tools/perf/util/probe-finder.c
>>> +++ b/tools/perf/util/probe-finder.c
>>> @@ -1727,7 +1727,7 @@ int debuginfo__find_probe_point(struct debuginfo *dbg, u64 addr,
>>> }
>>>
>>> /* Find a corresponding line (filename and lineno) */
>>> - cu_find_lineinfo(&cudie, addr, &fname, &lineno);
>>> + cu_find_lineinfo(&cudie, (Dwarf_Addr)addr, &fname, &lineno);
>>> /* Don't care whether it failed or not */
>>>
>>> /* Find a corresponding function (name, baseline and baseaddr) */
>>> @@ -1828,8 +1828,7 @@ static int line_range_add_line(const char *src, unsigned int lineno,
>>> }
>>>
>>> static int line_range_walk_cb(const char *fname, int lineno,
>>> - Dwarf_Addr addr __maybe_unused,
>>> - void *data)
>>> + Dwarf_Addr addr, void *data)
>>> {
>>> struct line_finder *lf = data;
>>> const char *__fname;
>>>
>>>
>
>

\
 
 \ /
  Last update: 2021-07-15 08:48    [W:0.046 / U:0.372 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site