linux-kernel - Re: [PATCH 5.10.y 0/2] Fixed perf abort when taken branch stack sampling enabled

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <2024111029-gorged-humiliate-f0bb@gregkh>
Date: Sun, 10 Nov 2024 06:12:08 +0100
From: Greg KH <gregkh@...uxfoundation.org>
To: Shuai Xue <xueshuai@...ux.alibaba.com>
Cc: stable@...r.kernel.org, linux-kernel@...r.kernel.org,
	linux-perf-users@...r.kernel.org, acme@...nel.org,
	adrian.hunter@...el.com, alexander.shishkin@...ux.intel.com,
	irogers@...gle.com, mark.rutland@....com, namhyung@...nel.org,
	peterz@...radead.org, acme@...hat.com, kprateek.nayak@....com,
	ravi.bangoria@....com, sandipan.das@....com,
	anshuman.khandual@....com, german.gomez@....com,
	james.clark@....com, terrelln@...com, seanjc@...gle.com,
	changbin.du@...wei.com, liuwenyu7@...wei.com,
	yangjihong1@...wei.com, mhiramat@...nel.org, ojeda@...nel.org,
	song@...nel.org, leo.yan@...aro.org, kjain@...ux.ibm.com,
	ak@...ux.intel.com, kan.liang@...ux.intel.com,
	atrajeev@...ux.vnet.ibm.com, siyanteng@...ngson.cn,
	liam.howlett@...cle.com, pbonzini@...hat.com, jolsa@...nel.org
Subject: Re: [PATCH 5.10.y 0/2] Fixed perf abort when taken branch stack
 sampling enabled

On Mon, Nov 04, 2024 at 07:27:34PM +0800, Shuai Xue wrote:
> On x86 platform, kernel v5.10.228, perf-report command aborts due to "free():
> invalid pointer" when perf-record command is run with taken branch stack
> sampling enabled. This regression can be reproduced with the following steps:
> 
> 	- sudo perf record -b
> 	- sudo perf report
> 
> The root cause is that bi[i].to.ms.maps does not always point to thread->maps,
> which is a buffer dynamically allocated by maps_new(). Instead, it may point to
> &machine->kmaps, while kmaps is not a pointer but a variable. The original
> upstream commit c1149037f65b ("perf hist: Add missing puts to
> hist__account_cycles") worked well because machine->kmaps had been refactored to
> a pointer by the previous commit 1a97cee604dc ("perf maps: Use a pointer for
> kmaps").
> 
> The memory leak issue, which the reverted patch intended to fix, has been solved
> by commit cf96b8e45a9b ("perf session: Add missing evlist__delete when deleting
> a session"). The root cause is that the evlist is not being deleted on exit in
> perf-report, perf-script, and perf-data. Consequently, the reference count of
> the thread increased by thread__get() in hist_entry__init() is not decremented
> in hist_entry__delete(). As a result, thread->maps is not properly freed.
> 
> To this end,
> 
> - PATCH 1/2 reverts commit a83fc293acd5c5050a4828eced4a71d2b2fffdd3 to fix the
>   abort regression.
> - PATCH 2/2 backports cf96b8e45a9b ("perf session: Add missing evlist__delete
>   when deleting a session") to fix memory leak issue.
> 
> Riccardo Mancini (1):
>   perf session: Add missing evlist__delete when deleting a session
> 
> Shuai Xue (1):
>   Revert "perf hist: Add missing puts to hist__account_cycles"
> 
>  tools/perf/util/hist.c    | 10 +++-------
>  tools/perf/util/session.c |  5 ++++-
>  2 files changed, 7 insertions(+), 8 deletions(-)

perf actually works and builds on this kernel tree?  That's news to me,
but hey, I'll take these now as obviously someone is still trying to run
it.

But why not just use the latest version of perf instead?

thanks,

greg k-h