[Plat-9721] Fix race condition in KSMachHeaders

[kstenerud]

Goal

This PR fixes some race conditions in KSMachHeaders

Changeset

• The image header linked list now uses atomics.
• The list head is now a dummy entry to simplify the algorithm. The real head is accessed via bsg_mach_headers_get_images().
• The initial headers list is now created at startup rather than lazily (it was being called at startup every time anyway, so this complication is unnecessary).
• Reduced the exposed API space, and clearly marked those intended for unit tests.

Testing

Updated unit tests to capture more edge cases.

[github-actions]

Bugsnag.framework binary size decreased by 368 bytes from 706,696 to 706,328 🎉

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  [ = ]       0  +2.6%    +368    [__LINKEDIT]
  +0.7%    +177  +0.7%    +177    [__TEXT]
  [ = ]       0  +0.5%     +48    __DATA,__bss
  +0.3%     +24  -0.2%     -24    [__DATA]
  +0.9%      +8  +0.9%      +8    __DATA,__data
  -0.3%      -8  -0.3%      -8    __TEXT,__unwind_info
  -0.1%     -25  -0.1%     -25    __TEXT,__cstring
  -1.0%     -32  -1.0%     -32    __DATA,__const
  -0.1%    -112  -0.1%    -112    Symbol Table
  -0.1%    -144  -0.1%    -144    __TEXT,__text
  -0.2%    -256  -0.2%    -256    String Table
  -0.1%    -368  [ = ]       0    TOTAL

Generated by 🚫 Danger

[lemnik]

I think we're missing a CAS operation here to make sure the load/store don't race?

Something like:

Suggested change

	BSG_Mach_Header_Info *oldTail = atomic_load(&g_images_tail);
	atomic_store(&g_images_tail, newImage);
	atomic_store(&oldTail->next, newImage);
	BSG_Mach_Header_Info *oldTail = atomic_load(&g_images_tail);
	for(;;) {
	if(atomic_compare_exchange_strong(&g_images_tail, &oldTail, newImage)) {
	atomic_store(&oldTail->next, newImage);
	break;
	}
	}

[lemnik]

LGTM