i#5383: Fix macOS arm64 test build/run #7171
Conversation
- Disable selfmod tests due to build failures on M3 - Fix macOS build for mangle_pauth test (test still fails for some reason) - Mark dump_ucontext as NYI on macOS - Fixes NULL sigcontext_t in os_forge_exception - Adds -vm_size 500M to client tests to avoid reachablity assert
|
only a64 failure is raw-zlib #5635, doesn't seem to be my fault |
ndrewh
changed the title
| @@ -6040,7 +6047,9 @@ if (NOT ARM) # FIXME i#1551: fix bugs on ARM | |||
| endif () | |||
| endif (NOT ARM) | |||
|
|
|||
| if (X86 OR AARCH64) # FIXME i#1551: port asm to ARM | |||
| # FIXME i#1551: port asm to ARM | |||
| # FIXME i#xxx: selfmod tests fail to build on macOS + ARM64 | |||
There was a problem hiding this comment.
Change i#xxx with i#5383 (this falls within that issue).
| /* Since SIGCXT_FROM_UCXT just accesses the uc->uc_mcontext ptr field on | ||
| * macOS, sc will be NULL below if we do not initialize uc_mcontext first | ||
| */ | ||
| uc->IF_X64_ELSE(uc_mcontext64, uc_mcontext) = (void *)&frame->mc; |
There was a problem hiding this comment.
This seems weird.
frame is zeroed-out anyway, isn't this just changing sc to be a "different" NULL?
Also, shouldn't sc being NULL break more than MACOS?
There was a problem hiding this comment.
frame is zeroed-out anyway,
frame is indeed zeroed out; which means frame->uc.uc_mcontext is NULL before L7717.
isn't this just changing sc to be a "different" NULL?
L7717 seems to set frame->uc.uc_mcontext (which was NULL before) to &frame->mc; the address of mc within the frame is not NULL.
I'm not familiar with these Mac structs. I wonder if it is simply a requirement for frame->uc.uc_mcontext to point to frame->mc.
There was a problem hiding this comment.
@ndrewh The comment at L7714 justifies that uc->IF_X64_ELSE(uc_mcontext64, uc_mcontext) must be set to something, but it doesn't explain why &frame->mc is the correct intended value. Can you add that reasoning to the comment?
| # On macOS + arm64 tests will randomly fail depending on | ||
| # whether mmap returns a region satisfying 32-bit reachability constraints. | ||
| # Reducing the vm_size appears to eliminate this issue. | ||
| set(dr_test_ops -vm_size 500M ${dr_test_ops}) |
There was a problem hiding this comment.
Would be useful to file a new issue with details of failures you saw.
This fixes several build and runtime issues under macOS on ARM64 on macOS 14.4.1 (Sonoma). - `FEATURE_PAUTH` is now detected on macOS using a sysctl (capability MSRs cannot be read in userspace on M3 at least). - In the preload library and os.c, `_init` needs `__attribute__((constructor))` to be run by dyld. - `PLATFORM_SUPPORTS_SCATTER_GATHER` is broken, disabled for now on a64 macOS. - Fixed sigcontext/mcontext conversion which assumes old `simd` size and was failing on an assert. Known issues that remain: - VM allocation frequently but nondeterministically fails reachability constraints. `-vm_size 500M` seems to fix this. #7171 adds this flag to tests on macOS+ARM64. Another option is to make this the default on macOS+ARM64. - Sometimes some bookkeeping asserts in `vmh_exit` are reached upon process exit. The end result is that the `bbcount` tool works on a simple toy program on macOS, in debug mode, without hitting any asserts: ``` bin64/drrun -debug -vm_size 500M -c api/bin/libbbcount.dylib -- ./main ``` Issue #5383
This makes tests build and run properly on macOS (14.4) arm64 (M3). Many tests do not pass, but you can at least run them now.