843364 - Late write from mozilla_sampler_shutdown

Status: RESOLVED WONTFIX

(Core :: Gecko Profiler, defect)

Description

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

We are getting stacks like these in Telemetry:

mozilla::ValidWriteAssert(bool) (in XUL)
(anonymous namespace)::AbortOnBadWrite(int, void const*, unsigned long) (in XUL)
_ZN12_GLOBAL__N_116wrap_writev_tempILZNS_11writev_dataEEEEliPK5ioveci (in XUL)
std::__basic_file<char>::xsputn_2(char const*, long, char const*, long) (in libstdc++.6.dylib)
std::basic_filebuf<char, std::char_traits<char> >::xsputn(char const*, long) (in libstdc++.6.dylib)
std::basic_ostream<char, std::char_traits<char> >& std::__ostream_insert<char, std::char_traits<char> >(std::basic_ostream<char, std::char_traits<char> >&, char const*, long) (in libstdc++.6.dylib)
HashTableOutput(nsACString_internal const&, PropertyValue*, void*) (in XUL)
nsBaseHashtable<nsCStringHashKey, PropertyValue*, PropertyValue*>::s_EnumReadStub(PLDHashTable*, PLDHashEntryHdr*, unsigned int, void*) (in XUL)
PL_DHashTableEnumerate (in XUL)
TemplatePropertyValue<JSCustomObject*>::SendToStream(std::ostream&) (in XUL)
TemplatePropertyValue<JSCustomArray*>::SendToStream(std::ostream&) (in XUL)
HashTableOutput(nsACString_internal const&, PropertyValue*, void*) (in XUL)
nsBaseHashtable<nsCStringHashKey, PropertyValue*, PropertyValue*>::s_EnumReadStub(PLDHashTable*, PLDHashEntryHdr*, unsigned int, void*) (in XUL)
PL_DHashTableEnumerate (in XUL)
TemplatePropertyValue<JSCustomObject*>::SendToStream(std::ostream&) (in XUL)
TemplatePropertyValue<JSCustomArray*>::SendToStream(std::ostream&) (in XUL)
HashTableOutput(nsACString_internal const&, PropertyValue*, void*) (in XUL)
nsBaseHashtable<nsCStringHashKey, PropertyValue*, PropertyValue*>::s_EnumReadStub(PLDHashTable*, PLDHashEntryHdr*, unsigned int, void*) (in XUL)
PL_DHashTableEnumerate (in XUL)
TemplatePropertyValue<JSCustomObject*>::SendToStream(std::ostream&) (in XUL)
TemplatePropertyValue<JSCustomArray*>::SendToStream(std::ostream&) (in XUL)
HashTableOutput(nsACString_internal const&, PropertyValue*, void*) (in XUL)
nsBaseHashtable<nsCStringHashKey, PropertyValue*, PropertyValue*>::s_EnumReadStub(PLDHashTable*, PLDHashEntryHdr*, unsigned int, void*) (in XUL)
PL_DHashTableEnumerate (in XUL)
JSCustomObjectBuilder::Serialize(JSCustomObject*, std::ostream&) (in XUL)
mozilla_sampler_shutdown() (in XUL)
XREMain::XRE_main(int, char**, nsXREAppData const*) (in XUL)
XRE_main (in XUL)
main (in firefox)

Comment 1

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

It looks like the code responding to MOZ_PROFILER_SHUTDOWN.  The options I see are

* Move it earlier
* Mark the fd as debug
* Disable write poisoning in mozilla_sampler_shutdown if MOZ_PROFILER_SHUTDOWN is set.

It is hard for me to decide since I don't know what is being written.

Comment 2

[Benoit Girard (:BenWa)]

We will want to mark it as a debug FD and when we exit 0 we will want to run this code before exit(0) as well.

Comment 3

[Benoit Girard (:BenWa)]

Attachment: patch

Comment 4

[Benoit Girard (:BenWa)]

Comment on attachment 716267 [details] [diff] [review]
patch

Opps

Comment 5

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

(In reply to Benoit Girard (:BenWa) from comment #2)
> We will want to mark it as a debug FD and when we exit 0 we will want to run
> this code before exit(0) as well.

Remember that when exit(0) will be called is a runtime setup. So for example:

MOZ_SHUTDOWN_CHECKS=nothing firefox  # will call exit(0) early.
MOZ_SHUTDOWN_CHECKS=debug firefox # will not call exit

Should they do these writes at different times? If so you do need to mark it as debug fd, but if the objective is to be able to profile "normal" firefox runs the write should always go before the place where we call exit(0), no?

Comment 6

[Benoit Girard (:BenWa)]

Sadly we're not working on exit(0) anymore :(