Closed markjdb closed 6 months ago
Hello, is there any interest in taking this PR? It's useful when CK is used with code instrumented by MSAN and is low-risk otherwise.
Hi @markjdb, sorry about that, I just forgot, it is now merged, thank you!
Hi @markjdb, sorry about that, I just forgot, it is now merged, thank you!
Thank you! Can I simply cherry-pick this into FreeBSD?
Yes, please do
@cognet I'm very sorry, I just realized that this commit doesn't exactly match the one I'm using in FreeBSD. It's not right. Could you please revert, and I'll submit the updated patch? I'm sorry again.
@markjdb done! Out of curiosity, what's wrong with this one?
Thanks. I failed to adjust the indexes of the operands in the actual assembly. Now that there's an output operand, %0 becomes %1, %1 becomes %2. I'm running the test suite on an arm64 board now just to confirm that I didn't miss something else.
Oh of course, I wonder how I missed that
I resubmitted the corrected and properly tested patch in #216.
This is required to enable KMSAN for aarch64 on FreeBSD. It will likely also be useful for userspace code which makes use of CK and is compiled with MSAN enabled.