569270 - e10s: Assertion on child process shutdown in xpcshell

Status: RESOLVED FIXED

(Core :: IPC, defect)

Description

[Josh Matthews [:jdm]]

(gdb) bt
#0  0x00532416 in __kernel_vsyscall ()
#1  0x005ced96 in nanosleep () from /lib/libc.so.6
#2  0x005cebb1 in sleep () from /lib/libc.so.6
#3  0x00ec2289 in ah_crap_handler (signum=6) at /home/t_mattjo/src/firefox/mobilebase/toolkit/xre/nsSigHandlers.cpp:132
#4  0x00ec22e2 in child_ah_crap_handler (signum=6) at /home/t_mattjo/src/firefox/mobilebase/toolkit/xre/nsSigHandlers.cpp:145
#5  <signal handler called>
#6  0x00532416 in __kernel_vsyscall ()
#7  0x0055da81 in raise () from /lib/libc.so.6
#8  0x0055f34a in abort () from /lib/libc.so.6
#9  0x0047ef84 in mozalloc_abort (msg=0xbfee47cc "###!!! ASSERTION: Content Protocol is NULL!: 'cpc', file /home/t_mattjo/src/firefox/mobilebase/modules/libpref/src/nsPrefBranch.cpp, line 833")
    at /home/t_mattjo/src/firefox/mobilebase/memory/mozalloc/mozalloc_abort.cpp:73
#10 0x021abad6 in Abort (aMsg=0xbfee47cc "###!!! ASSERTION: Content Protocol is NULL!: 'cpc', file /home/t_mattjo/src/firefox/mobilebase/modules/libpref/src/nsPrefBranch.cpp, line 833")
    at /home/t_mattjo/src/firefox/mobilebase/xpcom/base/nsDebugImpl.cpp:371
#11 0x021aba99 in NS_DebugBreak_P (aSeverity=1, aStr=0x2547705 "Content Protocol is NULL!", aExpr=0x2547701 "cpc", aFile=0x2547610 "/home/t_mattjo/src/firefox/mobilebase/modules/libpref/src/nsPrefBranch.cpp", aLine=833)
    at /home/t_mattjo/src/firefox/mobilebase/xpcom/base/nsDebugImpl.cpp:358
#12 0x01039980 in nsPrefBranch::RemoveObserver (this=0xb51ec9c0, aDomain=0x24547df "network.tcp.sendbuffer", aObserver=0xb7522010) at /home/t_mattjo/src/firefox/mobilebase/modules/libpref/src/nsPrefBranch.cpp:833
#13 0x0103ee3c in nsPrefService::RemoveObserver (this=0xb7511a00, aDomain=0x24547df "network.tcp.sendbuffer", aObserver=0xb7522010) at /home/t_mattjo/src/firefox/mobilebase/modules/libpref/src/nsPrefService.h:61
#14 0x00f17177 in nsSocketTransportService::Shutdown (this=0xb7522000) at /home/t_mattjo/src/firefox/mobilebase/netwerk/base/src/nsSocketTransportService2.cpp:475
#15 0x00eed3fd in nsIOService::SetOffline (this=0xb752ad80, offline=1) at /home/t_mattjo/src/firefox/mobilebase/netwerk/base/src/nsIOService.cpp:723
#16 0x00eedeee in nsIOService::Observe (this=0xb752ad80, subject=0xb755f104, topic=0x28126d0 "xpcom-shutdown", data=0x0) at /home/t_mattjo/src/firefox/mobilebase/netwerk/base/src/nsIOService.cpp:933
#17 0x0214efa7 in nsObserverList::NotifyObservers (this=0xb7555dc0, aSubject=0xb755f104, aTopic=0x28126d0 "xpcom-shutdown", someData=0x0) at /home/t_mattjo/src/firefox/mobilebase/xpcom/ds/nsObserverList.cpp:130
#18 0x02150945 in nsObserverService::NotifyObservers (this=0xb7511a30, aSubject=0xb755f104, aTopic=0x28126d0 "xpcom-shutdown", someData=0x0) at /home/t_mattjo/src/firefox/mobilebase/xpcom/ds/nsObserverService.cpp:182
#19 0x02142bda in mozilla::ShutdownXPCOM (servMgr=0x0) at /home/t_mattjo/src/firefox/mobilebase/xpcom/build/nsXPComInit.cpp:795
#20 0x021429df in NS_ShutdownXPCOM_P (servMgr=0x0) at /home/t_mattjo/src/firefox/mobilebase/xpcom/build/nsXPComInit.cpp:758
#21 0x00ec2aae in XRE_TermEmbedding () at /home/t_mattjo/src/firefox/mobilebase/toolkit/xre/nsEmbedFunctions.cpp:242
#22 0x0202d5c3 in mozilla::ipc::ScopedXREEmbed::Stop (this=0xb75341b8) at /home/t_mattjo/src/firefox/mobilebase/ipc/glue/ScopedXREEmbed.cpp:100
#23 0x01ff5c2a in mozilla::dom::ContentProcessProcess::CleanUp (this=0xb7534020) at /home/t_mattjo/src/firefox/mobilebase/dom/ipc/ContentProcessProcess.cpp:63
#24 0x00ec2ffb in XRE_InitChildProcess (aArgc=1, aArgv=0xbfee5b74, aProcess=GeckoProcessType_Content) at /home/t_mattjo/src/firefox/mobilebase/toolkit/xre/nsEmbedFunctions.cpp:451
#25 0x08049050 in main (argc=3, argv=0xbfee5b74) at /home/t_mattjo/src/firefox/mobilebase/ipc/app/MozillaRuntimeMain.cpp:87

Comment 1

[Frederic Plourde (:fred23)]

So if I understand this trace correctly, some consumer in the content process is trying to remove pref observers (nsPrefBranch::RemoveObserver) AFTER it's been cleaned-up.

I guess in that case, nsPrefBranch::RemoveObserver could just silently return without asserting because we already removed every prefObservers from the content process at child destruction.

Dan, what do you think?

Comment 2

[[e-mail]]

That sounds fine, with a comment explaining why. The parent doesn't care at all, right?

Comment 3

[Frederic Plourde (:fred23)]

right, the parent doesn't care. it's just forwarding every pref (listening for "") to the child and doesn't care about actual observer structs. I'll create a patch and submit for review.

Comment 4

[Frederic Plourde (:fred23)]

Attachment: patch

patch, according to discussions

Comment 5

[Josh Matthews [:jdm]]

>+      rv = cpc->RemoveRemotePrefObserver(nsDependentCString(aDomain), 
>+                                                            mPrefRoot, 
>+                                                            aObserver);

Line up mPrefRoot and aObserver with nsDependentCString, please.

Comment 6

[Frederic Plourde (:fred23)]

Attachment: patch

Comment 7

[Frederic Plourde (:fred23)]

Attachment: patch v.2

(In reply to comment #5)
> >+      rv = cpc->RemoveRemotePrefObserver(nsDependentCString(aDomain), 
> >+                                                            mPrefRoot, 
> >+                                                            aObserver);
> 
> Line up mPrefRoot and aObserver with nsDependentCString, please.

jeez, you're right ! thanks.

Comment 8

[[e-mail]]

Comment on attachment 449640 [details] [diff] [review]
patch v.2

Looks good. r=dwitte

Comment 9

[Frederic Plourde (:fred23)]

pushed to e10s
http://hg.mozilla.org/projects/electrolysis/rev/c11f32e403db