[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1664532445.32klbp0ljq.naveen@linux.ibm.com>
Date: Fri, 30 Sep 2022 15:39:53 +0530
From: "Naveen N. Rao" <naveen.n.rao@...ux.vnet.ibm.com>
To: Thadeu Lima de Souza Cascardo <cascardo@...onical.com>,
Srikar Dronamraju <srikar@...ux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@...nel.org>, Josh Poimboeuf <jpoimboe@...hat.com>,
LKML <linux-kernel@...r.kernel.org>,
linuxppc-dev <linuxppc-dev@...ts.ozlabs.org>,
Peter Zijlstra <peterz@...radead.org>
Subject: Re: [PATCH] scripts/faddr2line: Fix regression in name resolution on
ppc64le
Thadeu Lima de Souza Cascardo wrote:
> On Tue, Sep 27, 2022 at 01:22:11PM +0530, Srikar Dronamraju wrote:
>> Commit 1d1a0e7c5100 ("scripts/faddr2line: Fix overlapping text section failures")
>> can cause scripts/faddr2line to fail on ppc64le machines on few
>> distributions, while working on other distributions. The failure can be
>> attributed to difference in readelf output on various distributions.
>>
>> $ ./scripts/faddr2line vmlinux find_busiest_group+0x00
>> no match for find_busiest_group+0x00
>>
>> Expected output was:
>> $ ./scripts/faddr2line vmlinux find_busiest_group+0x00
>> find_busiest_group+0x00/0x3d0:
>> find_busiest_group at kernel/sched/fair.c:9595
>>
>> On ppc64le, readelf adds localentry tag before the symbol name on few
>> distributions and adds the localentry tag after the symbol name on few
>> other distributions. This problem has been discussed in the reference
>> URL given below. This problem can be overcome by filtering out
>> localentry tags in readelf output. Similar fixes are already present in
>> kernel by way of commits:
>>
>> 1fd6cee127e2 ("libbpf: Fix VERSIONED_SYM_COUNT number parsing")
>> aa915931ac3e ("libbpf: Fix readelf output parsing for Fedora")
>>
>> Fixes: 1d1a0e7c5100 ("scripts/faddr2line: Fix overlapping text section failures")
>> Reference: https://lore.kernel.org/bpf/20191211160133.GB4580@calabresa/
>> Cc: "Naveen N. Rao" <naveen.n.rao@...ux.vnet.ibm.com>
>> Cc: Jiri Olsa <jolsa@...nel.org>
>> Cc: Thadeu Lima de Souza Cascardo <cascardo@...onical.com>
>
> Reviewed-by: Thadeu Lima de Souza Cascardo <cascardo@...onical.com>
>
> The other instances of readelf --wide on faddr2line use --section-headers and
> should not required the same mangling.
There's one usage of readelf with --file-header in faddr2line which also
doesn't need this. The extra information being printed is from st_other
and is very specific to the symbol table.
Acked-by: Naveen N. Rao <naveen.n.rao@...ux.vnet.ibm.com>
- Naveen
Powered by blists - more mailing lists