linux_demon: PTRACE_EVENT_STOPs can sometimes be reported even if a thread has already hit a trap (e.g. a trap net or breakpoint trap). this would sometimes cause spurious SIGTRAPs after issuing a PTRACE_INTERRUPT - so we need to flag halted threads to not treat that as a 'real' trap. furthermore, even though that trap was hit, it does not guarantee that the debugger would next see a SIGTRAP. that would cause spurious SIGILLs, because we'd resume threads *after* the trap, despite not seeing the trap, so they'd just see a malformed instruction (the bytes subsequent the initial trap). so in those cases, we need to roll such threads back, so that they re-execute the trap, and so that we see a regular SIGTRAP, and handle the trap correctly.

This commit is contained in:
Ryan Fleury committed 2026-09-29 12:28:05 -06:00
1 parent 2933143b08
commit 0b91cab00a
3 files changed
+75 -47

No files matched your search

+7 -6
View File
@@ -7,23 +7,24 @@
// initial alpha:
// [x] .so loading/debugging - missing debug info
// [x] missing disasm
// [ ] recycling semaphore files - or alt strat?
// [ ] control
// [ ] multithreaded stepping - trap hits reported on other threads.
// [ ] stale memory - things evaluating to stale values that are definitely not zero
// [ ] support
// [ ] .eh_frame unwinding without .eh_frame_hdr
// [ ] asan
// [ ] stale memory - things evaluating to stale values that are definitely not zero
// [ ] rendering
// [ ] bitmaps / geometry
// [ ] bitmaps / geometry
//
// post initial alpha:
// [ ] windowing
// [ ] wayland
// [ ] support
// [ ] .eh_frame unwinding without .eh_frame_hdr
//
// [x] clipboard
// [x] jai exes - busted call stacks / unwind info / paths?
// [x] statically linked EXEs - busted unwinding
// [x] statically linked EXEs, not parsing maps file correctly
// [x] asan
// [x] multithreaded stepping - trap hits reported on other threads.
////////////////////////////////
//~ rjf: evaluation system pass TODO notes