Date: Tue, 15 Mar 2022 12:55:38 -0600 From: Gavin Howard <gavin.d.howard@...il.com> To: Rich Felker <dalias@...c.org> Cc: musl@...ts.openwall.com Subject: Re: Possible PEBKAC or Bug in musl Semaphores and/or Mutexes > I might suggest a programming questions forum like Stack Overflow as a > more productive place to pursue this than the musl mailing list, since > this is pretty clearly not an actionable bug report (which would need > a complete test case you *can* publish that reproduces the purported > bug) and almost surely not related to any bug in musl. I am pretty sure that this is the right place and that it is a bug in musl. I spent the time to be able to make the code in question public and to create an easy reproducer for you so that you will have an actionable bug report. The code is at https://git.yzena.com/Yzena/Yc . Make sure you have CMake and Clang installed (I use Clang to test bootstrap on other platforms). Also be sure you have musl-gcc in /usr/local/musl/bin. Once you do, just run the following: ``` git clone https://git.yzena.com/Yzena/Yc.git cd Yc ./tools/rwlock_repro.sh -m ``` The rwlock_repro.sh script with the -m option will initialize the repo, build the build system using the musl-gcc, and then repeatedly run the build system on itself until some error happens. With this setup, I have been able to get it to show that more than one thread is allowed into the write lock critical section simultaneously, and this usually happens within two minutes with an average of about a minute. If it does happen, you should see a message like this: ``` Panic: More than one thread in the critical section Source: /home/gavin/Yc2/src/rig/build.c:555 Function: rig_searchPath() ``` This happens because there is a global variable that is incremented by threads when they enter that critical section and decremented when they leave, and that is the only place the global is used. (It's just for testing this.) Yet, sometimes, threads will find that it is greater than 1, which means that more than one thread is in that critical section. Even if there are bugs in my code, I'm pretty sure that is not supposed to happen with musl's read/write locks. Then again, I could be wrong; if I am wrong about what is supposed to happen with musl's rwlocks, please let me know. I tested this with musl 1.2.2 and the latest master as of last night. And this is on an AMD Ryzen Threadripper 1900X x86_64, if that helps. Gavin Howard
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.