PCSX2 Linux Development

[gregory]

Not exactly.

A sigsegv is a memory signal (often (wrongly) call segmentation fault). PCSX2 uses a virtual memory that is not fully allocated. Allocation will only be done once you access it. This access will generate an expected sigsegv. I didn't find any nice way to separate the good sigsegv from the bad. Normally the bad will send an abort signal, so disabled gdb sigsegv handler will work.

[rama]

Every time I hear about this I wonder if there isn't some saner way to do this..

[gregory]

Well no. SIGSEGV is really the signal that handle bad memory access.

However I'm not sure Jake's reservation system is really critical on linux. 32 bits linux gets 3GB (technically you can choose it in the kernel) of VMem (could be 4GB with pae). And 64 bits linux gets 4GB of VMem. Versus the only 2GB of standards 32bit windows.
Note: I'm not sure it is used only for reservation.

[krysto]

Thanks for being patient with me... I am not as experienced with low level programming!

I was able to narrow it down to my std::mutex implementation of GSCritSec.

[krysto]

How recent was the asan option added? I just got from upstream this morning.

[gregory]

well a couple of hours ago => https://github.com/PCSX2/pcsx2/commit/c6...0398044c19

It has been in my backlog since many months/years.

[krysto]

If you could be so kind as to walk me through the debugging process...

where do I drop ASAN_OPTIONS=allow_user_segv_handler=1?

or is

Code:

ASAN_OPTIONS=allow_user_segv_handler=1 ./bin/pcsx2-dbg

all one line in the terminal?

[Urisma]

yes, all in one line in the terminal. If you want to run that while inside gdb I believe you'd do the following:

Code:

$gdb pcsx2-dbg
#handle SIGSEGV noprint
#set environment ASAN_OPTIONS allow_user_segv_handler=1
#run

[krysto]

hmm... I have no binary named pcsx2-gdb. Should I be getting one after building with the --asan option?

[Urisma]

pcsx2-dbg, not pcsx2-gdb