1757217 - https://crash-stats.mozilla.org/report/index/9be477f2-0692-45b1-8303-62a8e0220214 not properly unwound

Status: RESOLVED FIXED

(Socorro :: Symbols, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

Among the problems are the inclusion of _tailMerge_d3dcompiler_47.dll with very large offsets.

Comment 1

[Andrew McCreight [:mccr8]]

This tailmerge thing shows up fairly often.

This search indicates there are almost 40,000 of crashes with tailMerge_d3dcompiler in the proto stack in the last week.

Comment 2

[Gabriele Svelto [:gsvelto]]

What's happening with tailMerge_d3dcompiler is that it's the last symbol inside libxul. The stack walker probably finds a pointer into libxul's mapping that's past the last symbol we have, and we emit it because there's nothing else to go on. The actual issue here is that we're missing symbols and unwinding information for KERNELBASE.dll. Once we get those this should unwind properly. Once bug 1757890 gets fixed we should resume scraping Microsoft libraries and these problems should go away.

Comment 3

[Gabriele Svelto [:gsvelto]]

I reprocessed the crash and no symbols are missing, marking this as fixed.

Comment 4

[Gabriele Svelto [:gsvelto]]

Clearing the NI? as the stack-walking issue was caused by the missing symbols. With the symbols available the stack unwinds correctly.

Comment 5

[Jeff Muizelaar [:jrmuizel]]

Should https://crash-stats.mozilla.org/report/index/9be477f2-0692-45b1-8303-62a8e0220214 now have the correct stack? It doesn't seem to.

Comment 6

[Gabriele Svelto [:gsvelto]]

It appears we're getting lost in the JIT'd code after frame 5 and then again after frame 10. I'll dig a bit further into it but it looks like a different problem so maybe it's better to open a different bug for it. I want to first dig in the crash with other tools though to be sure. Leaving the NI? until I do that.

Comment 7

[Gabriele Svelto [:gsvelto]]

I've checked what kind of output we got with the old stack walker and it's the same as with the new one so I suspect there might be a problem with the unwinding directives within libxul. I'll file a separate bug to tackle that.