rbtree infinite loop/crash

[zmike]

As I pasted to @spencer-lunarg, running this this piglit test with VVL causes an extended loop inside VVL on shutdown which eventually results in a crash:

ZINK_DEBUG=vvl MESA_LOADER_DRIVER_OVERRIDE=zink bin/object-namespace-pollution glBitmap framebuffer -auto -fbo

[spencer-lunarg]

(notes)

If I turn off everything but Core Checks ( VK_LAYER_DISABLES=VK_VALIDATION_FEATURE_DISABLE_SHADERS_EXT,VK_VALIDATION_FEATURE_DISABLE_THREAD_SAFETY_EXT,VK_VALIDATION_FEATURE_DISABLE_OBJECT_LIFETIMES_EXT,VK_VALIDATION_FEATURE_DISABLE_UNIQUE_HANDLES_EXT,VK_VALIDATION_FEATURE_DISABLE_API_PARAMETERS_EXT) it will still crash

If I turn off core checks (VK_LAYER_DISABLES=VK_VALIDATION_FEATURE_DISABLE_CORE_CHECKS_EXT) it works as expected... so seems to be isolated to just core checks

I noticed from API dump it goes

vkBindBufferMemory()
vkBindBufferMemory()
PIGLIT: {"result": "pass" }
vkCreateImageView()
vkGetPhysicalDeviceImageFormatProperties2KHR()
vkCreateImage()
vkGetImageMemoryRequirements2()
vkAllocateMemory()
vkBindImageMemory()
vkCmdPipelineBarrier2() <---- memory corruption noticed here

[spencer-lunarg]

seems all static const std::map are being corrupted, even in things in like GetAccessMask2CommonMap(), not sure yet when that corruption starts

[spencer-lunarg]

so Piglet is calling exit(0)

#0 __run_exit_handlers stdlib/exit.c:108
#1 __GI_exit stdlib/exit.c:138
#2 piglit_report_result piglit/tests/util/piglit-util.c:245-

but it is then handling it piglit-framework-gl.c via the destory() but our std::map are being destroyed as a library under us during that exit(0) call