Heap allocation functions - are they really thread-safe as documented? about ps2sdk HOT 2 OPEN

I guess this depends on another descussion about how threads work on the IOP:

• threads can only be interrupted by interrupts
• interrupt handlers cannot call these functions

So we only need to worry about thread safety if another thread starts as a result from an interrupt while inside AllocSysMemory() or FreeSysMemory(). And there's been some discussion about whether this is possible or not.

Can an interrupt return to a higher priority thread? If so then these functions need to be thread safe, or called with interrupts disabled.

Why don't we make a simple test .irx, with 2 threads:

• thread1: low priority, runs forever in a loop, calling AllocSysMemory() and FreeSysMemory()
• thread2: high priority, triggered periodically with an interrupt, calling AllocSysMemory() and FreeSysMemory() once, then going back to sleep

If interrupts cannot return to a higher priority thread, then the IOP will be stuck inside thread1 (but I don't think this is the case).

If interrupts do return to the higher priority thread, then calling any thread-unsafe function will cause problems. Probably a memory leak, or crash.

from ps2sdk.

By the official documentation, it is possible for a higher-priority thread to be run instead of the previous running thread, if the higher-priority thread is changed into READY state by the interrupt handler. That would cause the previous running thread to be changed to READY state instead and the higher-priority thread will be changed from READY to RUN, when scheduling is resumed.

Even if it was not allowed, I was thinking that the AllocSysMemory function could be made to (accidentally) allocate the same region of memory twice, if the thread calling AllocSysMemory is suspended with iSuspendThread(), called by an interrupt handler. Another thread that runs could be allowed to execute AllocSysMemory(), which would then allow two threads to execute AllocSysMemory with the same (stale) state.

I am a little burned out now. But if nobody does any tests in a few days, I might do them.
Your test idea is great. But perhaps I'll do an additional check on whether both threads can somehow get the same memory pointer from AllocSysMemory, just in case the design does not allow a crash to occur. I can check for the amount of free memory too, I believe.

Thanks!

from ps2sdk.