Daniel Micay
Daniel Micay
It's not redundant since it might be all people read. Going to leave it as is.
Perhaps we should remove it as supported until it works properly, if ever.
Not going to be changed by us in that case.
We're working around this by going back to a 4k stack guard temporarily. We can add detection of the Hermes library and downgrade security for apps using it in the...
This is the stack overflow detection code that's buggy: https://github.com/facebook/hermes/blob/0ff6c2ab52062bc6ace2a0675b5367dcca5e0a05/lib/Support/OSCompatPosix.cpp#L638 https://github.com/facebook/hermes/blob/0ff6c2ab52062bc6ace2a0675b5367dcca5e0a05/include/hermes/Support/StackOverflowGuard.h#L94 It doesn't take into account the guard size in the calculations, which isn't always 4k. It works with a...
The calculation wrongly thinks the stack has overflowed and Hermes aborts, wrongly thinking there's an overflow. Do not know the exact details of how it goes wrong, just that it...
You can explicitly set the guard size to 64k when creating the threads and it should fail in the same way.
It needs to be done on Android. Bionic handles things differently than glibc.
That's different from what I'm describing since Bionic adds space for the stack guard so there should be just as much space available with a larger stack guard and it...
@tmikov That's not what I was reporting here. I suggest not working around how Android reports the guard page as part of `pthread_attr_getstack` because that can be changed. It seems...