Fix 32 bit ARM crashing bugs#555
Conversation
|
"Approximate understanding" is what I have as well... |
| } | ||
|
|
||
| return 0; | ||
| } |
There was a problem hiding this comment.
This looks like it was borrowed from the allocptr_displacement() function.
Right I really struggled with understanding the mechanism here, my only understanding is that it's a very specific memory/tagging race with exception handling. If it's hard to make sense I suggest we wait a couple weeks while my boards run on this branch a bit. That doesn't add much on top of the years these bugs are there. If they still work then, merging perhaps still would make sense as in my experience arm32 runtime is effectively broken. |
|
OK, we'll wait while you run this code for a while. I agree that the arm32 runtime is buggy, and given that your patch seems to help, I'm inclined to merge even if I don't completely understand it. The whole pc_luser_xp() scheme is extremely complicated. I'm becoming convinced that we'd be better off if we polled for gc/interrupts instead. |
2 participants
This is to address #520 and #428 crashing bugs on 32 bit ARM. The patch was generated after some time with Codex and only approximate understanding on my part so discretion is advised. Nonetheless it appears to have stopped crashes on my test board both with our codebase and lparallel as referred in #520. Both were reproducing robustly before the change.
What it's supposed to do: