1426466 - When linked with LLD, Firefox crashes on startup with "Assertion failure: !aOther.IsNull()" at TimeStamp.h:571

Status: RESOLVED WORKSFORME

(Firefox Build System :: General, defect)

Description

[Botond Ballo [:botond]]

STR:
  1. Build a debug version of Firefox with:
     
       ac_add_options --enable-linker=lld
       ac_add_options --disable-elf-hack

  2. Run the resulting build.

Actual results:

  Firefox crashes on startup with the following assertion failure:

Program received signal SIGSEGV, Segmentation fault.
0x00000000002362dd in mozilla::TimeStamp::operator> (this=0x7fffffffbaf0, aOther=...) at /home/botond/dev/mozilla/central/objdir-desktop-clang/dist/include/mozilla/TimeStamp.h:571
571         MOZ_ASSERT(!aOther.IsNull(), "Cannot compute with aOther null value");
(gdb) i s
#0  0x00000000002362dd in mozilla::TimeStamp::operator>(mozilla::TimeStamp const&) const (this=0x7fffffffbaf0, aOther=...)
    at /home/botond/dev/mozilla/central/objdir-desktop-clang/dist/include/mozilla/TimeStamp.h:571
#1  0x0000000000235f74 in mozilla::TimeStamp::ProcessCreation(bool*) (aIsInconsistent=0x0) at /home/botond/dev/mozilla/central/mozglue/misc/TimeStamp.cpp:73
#2  0x00007fffe98df760 in CorePS::CorePS() (this=0x7ffff6b48a00) at /home/botond/dev/mozilla/central/tools/profiler/core/platform.cpp:213
#3  0x00007fffe98d7ec9 in CorePS::Create(mozilla::BaseAutoLock<PSMutex> const&) (aLock=...) at /home/botond/dev/mozilla/central/tools/profiler/core/platform.cpp:235
#4  0x00007fffe98c12fa in profiler_init(void*) (aStackTop=0x7fffffffbf28) at /home/botond/dev/mozilla/central/tools/profiler/core/platform.cpp:2298
#5  0x00007fffe9de0c19 in mozilla::AutoProfilerInit::AutoProfilerInit(mozilla::detail::GuardObjectNotifier&&) (this=0x7fffffffbf28, _notifier=<unknown type in /home/botond/dev/mozilla/central/objdir-desktop-clang/dist/bin/libxul.so, CU 0x11e5eaf7, DIE 0x11e8ffea>) at /home/botond/dev/mozilla/central/objdir-desktop-clang/dist/include/GeckoProfiler.h:524
#6  0x00007fffe9ddcfd1 in XREMain::XRE_main(int, char**, mozilla::BootstrapConfig const&) (this=0x7fffffffbf78, argc=4, argv=0x7fffffffd308, aConfig=...)
    at /home/botond/dev/mozilla/central/toolkit/xre/nsAppRunner.cpp:4758
#7  0x00007fffe9dde1ec in XRE_main(int, char**, mozilla::BootstrapConfig const&) (argc=4, argv=0x7fffffffd308, aConfig=...)
    at /home/botond/dev/mozilla/central/toolkit/xre/nsAppRunner.cpp:4963
#8  0x00007fffe9df3fb7 in mozilla::BootstrapImpl::XRE_main(int, char**, mozilla::BootstrapConfig const&) (this=0x7ffff6b506a0, argc=4, argv=0x7fffffffd308, aConfig=...)
    at /home/botond/dev/mozilla/central/toolkit/xre/Bootstrap.cpp:49
#9  0x000000000021672f in do_main(int, char**, char**) (argc=4, argv=0x7fffffffd308, envp=0x7fffffffd330) at /home/botond/dev/mozilla/central/browser/app/nsBrowserApp.cpp:231
#10 0x00000000002162fe in main(int, char**, char**) (argc=4, argv=0x7fffffffd308, envp=0x7fffffffd330) at /home/botond/dev/mozilla/central/browser/app/nsBrowserApp.cpp:304

This suggests sInitOnce.mFirstTimeStamp has not been initialized by the time TimeStamp::ProcessCreation() is called. 

sInitOnce.mFirstTimeStamp is initialized by the constructor of sInitOnce, which is a global object, which should have been called by the time we get into main().

Comment 1

[(Away)]

What version of lld are you using?

Comment 2

[Botond Ballo [:botond]]

5.0.1 (and likewise for clang).

Comment 3

[Nicholas Nethercote [inactive]]

botond, is this still a problem? Bug 1473436 made lld the default on Linux...

Comment 4

[Botond Ballo [:botond]]

Hmm, I just tried pulling m-c and building (without --enable-linker=lld in my .mozconfig), and the build system used ld.gold to link libxul. I do have LLD installed.

Comment 5

[Botond Ballo [:botond]]

I'd debug the python code that chooses the linker by dropping an |import pdb; pdb.set_trace()| in it, but that doesn't seem to work any more:

  ImportError: Importing modules is forbidden

How do you debug the build system these days?

Comment 6

[Botond Ballo [:botond]]

I tried |@imports('pdb')| above the function, but that just caused it to hang.

Comment 7

[Mike Hommey [:glandium]]

check config.log in your objdir.

Comment 8

[Botond Ballo [:botond]]

Thanks. The problem was that I was building with GCC. I guess GCC doesn't support LLD. (Yet?)

Comment 9

[Mike Hommey [:glandium]]

it does if you have a ld.lld binary in your PATH.

Comment 10

[Botond Ballo [:botond]]

I do have an ld.lld binary in my PATH. The config.log said something like "option -fuse-ld=lld not recognized, did you mean -fuse-ld=bfd", so maybe my GCC version (6.2) is too old to support it.

Comment 11

[Botond Ballo [:botond]]

Anyways, having successfuly built with clang + LLD, I no longer experience the mentioned startup crash. Closing as WORKSFORME.

Comment 12

[Brian Birtles (:birtles)]

I started hitting this today after installing lld and running ./mach bootstrap.

I am on Ubuntu 17.10, clang --version is:

  clang version 4.0.1-6 (tags/RELEASE_401/final)
  Target: x86_64-pc-linux-gnu
  Thread model: posix
  InstalledDir: /usr/bin

I tried installing clang-5.0 and lld-5.0, setting `export CC=clang-5.0 / CXX = clang++-5.0` in mozconfig and clobber building but it failed to compile with errors such as:

  ...
  6:15.58 /usr/lib/llvm-5.0/lib/clang/5.0.0/include/avx512bwintrin.h:509:10: error: invalid conversion between vector type '__m512i' (vector of 8 'long long' values) and integer type 'int' of different size
  6:15.58   return (__m512i)__builtin_ia32_packssdw512((__v16si)__A, (__v16si)__B);
  ...

I've reverted my mozconfig and uninstalled lld for now.

Comment 13

[Brian Birtles (:birtles)]

I guess I didn't actually uninstall lld since now I get:

  /usr/lib/llvm-4.0/bin/ld.lld: error: /home/birtles/src2/dom/media/gmp-plugin-openh264/gmp-fake-openh264.cpp:425: undefined symbol 'operator new(unsigned long)'

I added:

  ac_add_options --enable-linker=gold

to my mozconfig and now I can finally get a working build again.