i#5256 sigprocmask: enable linux.sigaction on MacOS

[abhinav92003]

Fixes and enables the linux.sigaction test on MacOS.

On MacOS, sigprocmask has different behaviour than Linux. It
does not fail for an unreadable or unwriteable old sigset. Also,
it does not need a size arg.

Fixes terminate_via_kill on MacOS. We need to add a constant
for Unix syscalls on Mac. Otherwise it results in a hang due to
the debug infinite loop at global_do_syscall_syscall when we
attempt to SIGTERM the process in linux.sigaction.

Fixes is_sys_kill detection on Mac. Things seem to work without
this, but the latest Mac documentation shows that the method
we use for non-MacOS (si_code <= 0) would work on Mac too:
https://www.unix.com/man-page/osx/3HEAD/siginfo.h

Fixes sigaction mask comparison in the linux.sigaction test.

Fixes a corner case where a bad old sigset fails the syscall with
EFAULT, but still succeeds in setting the new sigset.

Also adds a native execution test for linux.sigaction. This is to
verify kernel's behaviour in the CI.

Adds a -no_intercept_all_signals test for linux.sigaction.This is to
verify write faults in the path where DR has to manually modify
the old sigset returned by the kernel in sigprocmask.

Issue: #5257, #5254
Fixes: #5256

[abhinav92003]

I'm seeing a hang in the linux.sigaction test on MacOS: https://gist.github.com/abhinav92003/3e59ff8fa71872d398890ff37293dd2f. It works fine till the end, but hangs at the SIGTERM. It also hangs when I remove everything but the kill(SIGTERM... from the test.

The test ends up succeeding after 90s but that's just because of the timeout.

It works fine when the test is run with -no_code_api -no_intercept_all_signals.

[abhinav92003]

The problem seems to be that the is_sys_kill call at

dynamorio/core/unix/signal.c

Line 6343 in 686f1a1

is_sys_kill(dcontext, pc, (byte *)sc->SC_XSP, &frame->info)) {

does not return true on Mac.

On Linux, the SIGTERM is recognised as a user-generated signal at

dynamorio/core/unix/signal.c

Line 4851 in 686f1a1

if (info->si_code <= 0)

. So it returns true early.
The second condition in is_sys_kill also returns false because is_at_do_syscall is false, but the syscall is a SYS_kill.

[abhinav92003]

I made a fix to is_sys_kill to allow detecting user-sent signals on Mac, but then I realised that it doesn't make a difference because the if-block executes even if is_sys_kill returns false.

Following the bug more, I see that this may be the loop it's hitting:

dynamorio/core/arch/x86/x86.asm

Line 842 in 686f1a1

jmp GLOBAL_REF(debug_infinite_loop)

This could be because SYS_kill is failing.

[abhinav92003]

run arm tests