Crash on OwnerLastRenderTimePtr update

Development Rendering
, ,
1

I am currently looking at a crash dump that so far has been reported only once. Looking at the code and the disassembly for the dump, it looks like the code is passing a nullptr check on OwnerLastRenderTimePtr, and then immediately crashing because that pointer is now null:

[Image Removed]

[Image Removed]

This leads me to assume that another concurrent thread happened to bash that value at exactly the right time.

Searching existing threads lead me to this one from last June: [Dangling pointer + mem corruption related to [Content removed] It seems plausible that our crash is related to the same issue described there. Has there been any progress on an official fix for that issue that we could try pulling into our build? Or is the PostRename() workaround in that thread still the best way to handle that issue?

[Attachment Removed]

2

Steps to Reproduce
It popped up once randomly during a playthrough on 2/4, has not been reported again since.

[Attachment Removed]

3

Hello,

We do have a fix coming in 5.8 in this area in CL#48956492 Make LastRenderTime / LastRenderTimeOnScreen relaxed atomic to fix TSan issue, but I’m not 100% certain if this is going to address the issue you’re seeing. I’m passing this along to my colleague who is more familiar with the related issue linked above.

[Attachment Removed]

4

CL#4895649 would not fix that issue. Since we never reset that ptr to nullptr, I’d assume the ptr to the FPrimitiveInfoData is stale.

[Attachment Removed]