ZJIT: Fix SP alignment on JIT-to-JIT calls for x86_64 #13900
Merged
+82
−45
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
XrXr
approved these changes
Jul 15, 2025
There was a problem hiding this comment.
It always seemed weird to me that it's aligned right before CALL runs, not after, since that forces the callee to align the stack if it wants it, which kind of defeats the whole point. But ay, I guess they assume everyone wants to at least push rbp in the olden days and this is some sort of carry-over.
Comment on lines
-514
to
-517
| // On x86_64, maintain 16-byte stack alignment | ||
| if cfg!(target_arch = "x86_64") { | ||
| asm.cpush(SP); | ||
| } |
There was a problem hiding this comment.
Nice, this looked suspect to me, so I'm glad we're fixing it.
st0012
approved these changes
Jul 16, 2025
tekknolagi
approved these changes
Jul 16, 2025
Co-authored-by: Alan Wu <XrXr@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR fixes the alignment of the native stack pointer on JIT-to-JIT calls for x86_64. It fixes the crash on the 30k_ifelse benchmark.
What was happening
When making a call on x86_64, SP should be 16-byte aligned. But the call instruction itself bumps it by 8 bytes, so when you're at the beginning of the function, it's misaligned.
gen_entry_prologue()was pushing 32 bytes, so the JIT trampoline carried the misaligned state to the call into the JIT function.When the call instruction makes another 8-byte bump to enter the JIT function, it's aligned, so
aligned_stack_bytes()andalloc_regs()pushing 16-byte-aligned bytes made sense. However, when it makes a JIT-to-JIT call, the call instruction makes it misaligned on the callee JIT function, so C functions called in it had misaligned SP.Moreover, when a JIT-to-JIT side-exits, the caller frame did not restore the frame size bumped for spilled basic block arguments. So leaving JIT-to-JIT calls with many basic block arguments could lead to wrong/misaligned SP as well.
What happens now
This PR changes
gen_entry_prologue()to push 24 bytes instead. Combined with the 8 byte pushed by the call instruction, the JIT function is called with SP aligned properly. The JIT function then needs to fix the SP misaligned by the call instruction, soaligned_stack_bytes()now pushes an odd number of slots.This PR also changes JIT-to-JIT calls to use a small side exit that only restores SP when the call returns Qundef.