DMARC-Filter: OpenDMARC Filter v1.4.2 delorie.com 54U8m7Kp1008685
DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 7DFE7385840D
Date: Fri, 30 May 2025 17:47:30 +0900
To: cygwin@cygwin.com
Subject: Re: Crash or hang if SIGSEGV+SIGALRM are nested
Message-Id: <20250530174730.7475a800c9f189b21ee0646e@nifty.ne.jp>
In-Reply-To: <32cf9316-3e97-7fab-f975-851c3dfc25b9@t-online.de>
References: <c4335d1d-5394-2a41-f06b-feb5c9fb1398@t-online.de>
 <20250528215707.ddb72fb28122c3ed07da8c5b@nifty.ne.jp>
 <20250529102846.8144f502c3f17decbe6700c1@nifty.ne.jp>
 <76940263-49e8-1b4a-17f2-11f5545014ab@t-online.de>
 <20250530015410.e8c4808c7947682b3309fbe7@nifty.ne.jp>
 <32cf9316-3e97-7fab-f975-851c3dfc25b9@t-online.de>
Mime-Version: 1.0
Precedence: list
From: Takashi Yano via Cygwin <cygwin@cygwin.com>
Reply-To: Takashi Yano <takashi.yano@nifty.ne.jp>
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Errors-To: cygwin-bounces~archive-cygwin=delorie.com@cygwin.com
Sender: "Cygwin" <cygwin-bounces~archive-cygwin=delorie.com@cygwin.com>

On Thu, 29 May 2025 20:01:31 +0200
Christian Franke wrote:
> Takashi Yano via Cygwin wrote:
> > On Thu, 29 May 2025 17:32:19 +0200
> > Christian Franke wrote:
> > ...
> >> I still don't fully understand why a SIGSEGV triggered by an instruction
> >> could interrupt a SIGALRM handler.
> >> https://sourceware.org/pipermail/cygwin/2025-May/258145.html
> >>
> >> I guess such behavior is valid from the POSIX point of view, but it is
> >> at least unexpected. If the SIGALRM handler is already running, it
> >> should have interrupted the thread such that the instruction triggering
> >> the segfault is not executed until the SIGALRM handler returns.
> > I try to explain what is happing using the figure below.
> >
> >     <<<<<< Main thread >>>>>>                       < Signal Thread >
> >                                              SIGNAL
> >    main()   handler1() handler2()             QUEUE    wait_sig()
> >      |          .          .                    |          |
> >      |          .          .       ALRM         |          |
> >      +----------------------------------------->|   ALRM   |
> >      |          .          .       SEGV         +--------->|
> >      X----------------------------------------->|          |
> >      |          .       arm ALRM                |          |
> >      +----------+ <----------------------------------------+
> >                 |          .                    |   SEGV   |
> >                 |          .                    +--------->|
> >                 |          .                    |          |
> >                 |          .      arm SEGV      |          |
> >                 +----------+ <-----------------------------+
> >                 .          |                    |          |
> >           longjmp()        |                    |          |
> >      +---------------------+                    |          |
> >      |          .          .                    |          |
> >      |          .          .                    |          |
> >
> > The point is the exception handler does not arm SIGSEGV handler
> > directly. It just pushes the SIGSEGV into the signal queue.
> > The signal thread reads the queue and process it asynchronously,
> > and arms the handler().
> >
> 
> I see. Thanks for nice explanation!
> 
> I created an issue at stress-ng upstream because at least the --mprotect 
> test is affected:
> https://github.com/ColinIanKing/stress-ng/issues/529

Thanks!

BTW, do you think this is the right thing?
https://github.com/RISCV-Tools/stress-ng/commit/bf24393357cde137bb8e7b38f917f565946199fd

I think the patch should set SIGSEGV mask on SIGALRM.

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>

-- 
Problem reports:      https://cygwin.com/problems.html
FAQ:                  https://cygwin.com/faq/
Documentation:        https://cygwin.com/docs.html
Unsubscribe info:     https://cygwin.com/ml/#unsubscribe-simple