Crash caused by exception thown from DMABufTextureBinder::ensure_egl_image()

[Alan Griffiths]

The most reliable way I have to reproduce is:

1. Run (Mir-on-X) $ miral-shell --window-manager tiling
2. Run a script that iterates over mir_demo_client_* launching all our demo clients.

The following exception causes a segfault in report_exception() - because by then the library has been unloaded:

Thread 3 "Mir/Comp" hit Catchpoint 1 (exception thrown), 0x00007ffff6ee58bd in __cxa_throw ()
   from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
(gdb) bt
#0 0x00007ffff6ee58bd in __cxa_throw () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#1 0x00007ffff18275be in throw_exception (e=...) at /usr/include/boost/throw_exception.hpp:69
#2 boost::exception_detail::throw_exception_<std::system_error>(std::system_error const&, char const*, char const*, int) [clone .constprop.114] (x=...,
    current_function=0x7ffff182b2e0 <(anonymous namespace)::DMABufTextureBinder::ensure_egl_image()::__PRETTY_FUNCTION__> "virtual void {anonymous}::DMABufTextureBinder::ensure_egl_image()", file=<optimised out>, line=<optimised out>)
    at /usr/include/boost/throw_exception.hpp:86
#3 0x00007ffff1822967 in (anonymous namespace)::DMABufTextureBinder::ensure_egl_image() [clone .lto_priv.250] (
    this=<optimised out>)
    at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/platforms/mesa/server/common/buffer_allocator.cpp:147
#4 0x00007ffff182308e in (anonymous namespace)::EGLImageBufferTextureBinder::gl_bind_to_texture() [clone .lto_priv.247] (
    this=0x7fffe8509c50) at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/platforms/mesa/server/common/buffer_allocator.cpp:71
#5 mir::graphics::mesa::GBMBuffer::gl_bind_to_texture() (this=<optimised out>)
    at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/platforms/mesa/server/common/gbm_buffer.cpp:166
#6 0x00007ffff65e402f in mir::gl::RecentlyUsedCache::load(mir::graphics::Renderable const&) (this=<optimised out>,
    renderable=...) at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/gl/recently_used_cache.cpp:45
#7 0x00007ffff65fd2f9 in mir::renderer::gl::Renderer::draw(mir::graphics::Renderable const&, mir::renderer::gl::Renderer::Program const&) const (this=this@entry=0x7fffe80008c0, renderable=..., prog=...)
    at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/renderers/gl/renderer.cpp:247
#8 0x00007ffff65fd590 in mir::renderer::gl::Renderer::render(std::vector<std::shared_ptr<mir::graphics::Renderable>, std::allocator<std::shared_ptr<mir::graphics::Renderable> > > const&) const (this=0x7fffe80008c0, renderables=...)
    at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/renderers/gl/renderer.cpp:205
#9 0x00007ffff65a30be in composite (this=this@entry=0x7fffe84c76c0,
    scene_elements=scene_elements@entry=<unknown type in /usr/lib/debug/.build-id/01/9340776985304e24a5e24d7cdcce3fc030bb9f.debug, CU 0x216701, DIE 0x23f84d>)
    at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/server/compositor/default_display_buffer_compositor.cpp:83
#10 0x00007ffff65a1076 in mir::compositor::CompositingFunctor::operator()() (this=0x7c4570)
    at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/server/compositor/multi_threaded_compositor.cpp:141
#11 0x00007ffff6601087 in std::function<void ()>::operator()() const () at /usr/include/c++/5/functional:2267
#12 execute (this=0x7fffef7afdf0) at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/server/thread/basic_thread_pool.cpp:40
#13 operator() (this=0x7c4660) at /build/mir-pkdHET/mir-0.21.0+16.04.20160330/src/server/thread/basic_thread_pool.cpp:91
#14 __invoke () at /usr/include/c++/5/functional:201
#15 operator() () at /usr/include/c++/5/functional:428
#16 _M_invoke () at /usr/include/c++/5/functional:1531
#17 operator() () at /usr/include/c++/5/functional:1520
#18 _M_run (this=<optimised out>) at /usr/include/c++/5/thread:115
#19 0x00007ffff6f10c70 in ?? () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#20 0x00007ffff62e46fa in start_thread (arg=0x7fffef7b0700) at pthread_create.c:333
#21 0x00007ffff697fb5d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109

[Alan Griffiths (alan-griffiths)]

The exception comes from:

            auto ret = drmPrimeHandleToFD(drm_fd, gem_handle, DRM_CLOEXEC, &raw_fd);
            prime_fd = mir::Fd{raw_fd};
            if (ret)
            {
                std::string const msg("Failed to get PRIME fd from gbm bo");
                BOOST_THROW_EXCEPTION(
                    std::system_error(errno, std::system_category(), "Failed to get PRIME fd from gbm bo"));
            }

Is this something we should be able to handle? How and where?

[Alan Griffiths (alan-griffiths)]

Running on lp:mir trunk I get somewhat better behaviour. Instead of a segfault with no information I get:

ERROR: /home/alan/display_server/mir/src/platforms/mesa/server/buffer_allocator.cpp(150): Throw in function virtual void {anonymous}::DMABufTextureBinder::ensure_egl_image()
Dynamic exception type: boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<std::system_error> >
std::exception::what: Failed to get PRIME fd from gbm bo: No such file or directory

Segmentation fault (core dumped)

[Michał Sawicz (saviq)]

Syncing task from Mir.