| Summary: | On X11, plasma crashes in ScreenPool::handleScreenRemoved after waking monitor | ||
|---|---|---|---|
| Product: | [Plasma] plasmashell | Reporter: | sp <sp139> |
| Component: | generic-crash | Assignee: | Plasma Bugs List <plasma-bugs> |
| Status: | RESOLVED FIXED | ||
| Severity: | crash | CC: | dougshaw77, itsbugfood, kde, meven29, nate, nicolas.fella, notmart, okitain, postix, salamanetwork, tkaratzo |
| Priority: | NOR | Keywords: | drkonqi |
| Version: | 6.0.3 | ||
| Target Milestone: | 1.0 | ||
| Platform: | unspecified | ||
| OS: | Linux | ||
| See Also: | https://bugs.kde.org/show_bug.cgi?id=444386 | ||
| Latest Commit: | https://invent.kde.org/plasma/plasma-workspace/-/commit/27e16b504b96a717691abe5cef63e9fdc3ee0670 | Version Fixed In: | 6.2.0 |
| Sentry Crash Report: | https://crash-reports.kde.org/organizations/kde/issues/3211/ | ||
| Attachments: | New crash information added by DrKonqi | ||
|
Description
sp
2024-03-28 22:49:08 UTC
Created attachment 167906 [details]
New crash information added by DrKonqi
DrKonqi auto-attaching complete backtrace.
#6 __pthread_kill_implementation (no_tid=0, signo=6, threadid=125444339747520) at ./nptl/pthread_kill.c:44 #7 __pthread_kill_internal (signo=6, threadid=125444339747520) at ./nptl/pthread_kill.c:78 #8 __GI___pthread_kill (threadid=125444339747520, signo=signo@entry=6) at ./nptl/pthread_kill.c:89 #9 0x000072174d442476 in __GI_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26 #10 0x000072174d4287f3 in __GI_abort () at ./stdlib/abort.c:79 #11 0x000072174e0db597 in qAbort () at ./src/corelib/global/qglobal.cpp:161 #12 0x000072174e0d67f5 in qt_message_fatal<QString&> (message=..., context=...) at ./src/corelib/global/qlogging.cpp:2003 #13 qt_message(QtMsgType, const QMessageLogContext &, const char *, typedef __va_list_tag __va_list_tag *) (msgType=msgType@entry=QtFatalMsg, context=..., msg=<optimized out>, ap=ap@entry=0x7fffefc3fcd0) at ./src/corelib/global/qlogging.cpp:378 #14 0x000072174e0dbfc3 in QMessageLogger::fatal (this=<optimized out>, msg=<optimized out>) at ./src/corelib/global/qlogging.cpp:901 #15 0x000072174e0a0387 in qt_assert_x (where=where@entry=0x5d49db706d88 "void ScreenPool::handleScreenRemoved(QScreen*)", what=<optimized out>, file=file@entry=0x5d49db70700f "./shell/screenpool.cpp", line=line@entry=265) at ./src/corelib/global/qassert.cpp:77 #16 0x00005d49db64fa10 in ScreenPool::handleScreenRemoved (this=0x5d49dd19b4d0, screen=<optimized out>) at ./shell/screenpool.cpp:265 #17 0x000072174e03c3ce in QtPrivate::QSlotObjectBase::call (a=<optimized out>, r=<optimized out>, this=<optimized out>, this=<optimized out>, r=<optimized out>, a=<optimized out>) at ./src/corelib/kernel/qobjectdefs_impl.h:433 #18 doActivate<false> (sender=0x7fffefc407e0, signal_index=10, argv=0x7fffefc40000) at ./src/corelib/kernel/qobject.cpp:4039 #19 0x000072174e4f2646 in QGuiApplication::screenRemoved (this=<optimized out>, _t1=<optimized out>) at ./obj-x86_64-linux-gnu/src/gui/Gui_autogen/include/moc_qguiapplication.cpp:560 #20 0x000072174e54cc02 in QWindowSystemInterface::handleScreenRemoved (platformScreen=0x5d49def4d6f0) at ./src/corelib/kernel/qcoreapplication.h:96 #21 0x00007217489b155c in QXcbConnection::initializeScreensFromMonitor (this=this@entry=0x5d49dcdbab80, it=it@entry=0x7fffefc40240, xcbScreenNumber=xcbScreenNumber@entry=0, primaryScreen=primaryScreen@entry=0x7fffefc40220, initialized=initialized@entry=true) at ./src/plugins/platforms/xcb/qxcbconnection_screens.cpp:567 #22 0x00007217489b1b58 in QXcbConnection::initializeScreens (this=0x5d49dcdbab80, initialized=true) at ./src/plugins/platforms/xcb/qxcbconnection_screens.cpp:317 #23 0x00007217489a6ae2 in QXcbConnection::handleXcbEvent (this=0x5d49dcdbab80, event=0x5d49dd8c0da0) at ./src/plugins/platforms/xcb/qxcbconnection.cpp:590 #24 0x00007217489ad0ee in QXcbConnection::processXcbEvents (this=0x5d49dcdbab80, flags=...) at ./src/plugins/platforms/xcb/qxcbconnection.cpp:1087 #25 0x00007217489c4a87 in xcbSourceDispatch (source=<optimized out>) at ./src/plugins/platforms/xcb/qxcbeventdispatcher.cpp:55 #26 0x000072174c8e1d3b in g_main_context_dispatch () from /lib/x86_64-linux-gnu/libglib-2.0.so.0 #27 0x000072174c937258 in ?? () from /lib/x86_64-linux-gnu/libglib-2.0.so.0 #28 0x000072174c8df3e3 in g_main_context_iteration () from /lib/x86_64-linux-gnu/libglib-2.0.so.0 #29 0x000072174de55c30 in QEventDispatcherGlib::processEvents (this=0x5d49dce6f520, flags=...) at ./src/corelib/kernel/qeventdispatcher_glib.cpp:393 #30 0x000072174e07fceb in QEventLoop::exec (this=0x7fffefc40700, flags=...) at ./src/corelib/global/qflags.h:34 #31 0x000072174e07a36c in QCoreApplication::exec () at ./src/corelib/global/qflags.h:74 #32 0x000072174e4e9a20 in QGuiApplication::exec () at ./src/gui/kernel/qguiapplication.cpp:1925 #33 0x00007217501ee469 in QApplication::exec () at ./src/widgets/kernel/qapplication.cpp:2574 #34 0x00005d49db652bb4 in main (argc=<optimized out>, argv=<optimized out>) at ./shell/main.cpp:214 Video driver is amdgpu. More or less 9 times out 10, when monitor goes to powersave/sleep, when waking it up, plasmashell crashes. So it almost always crashes. "Crash report" lets us send "automatic" crash report. So I've been sending them also. Thanks, that's indeed helpful to do. https://invent.kde.org/plasma/plasma-workspace/-/merge_requests/4238 *might* fix that, at least fixes it for the way i could reproduce (disabling internal monitor and enabling external in one go from the kcm) needs to see if it will make sentry reports go down Git commit dfd160bdf6f660c10b712a664c86a2a7339e8e72 by Marco Martin. Committed on 15/04/2024 at 12:36. Pushed by mart into branch 'master'. Fix redundantoutputs when the enabled screens change very quickly on X11, when a single transaction from KScreen disables screenA and enabled screenB, a new output order arrivers before the old output gets removed and in that moment the new screen will be considered redundant to che screen that is about to be removed, causing an invalid internal state Related: bug 456947 M +4 -3 shell/screenpool.cpp https://invent.kde.org/plasma/plasma-workspace/-/commit/dfd160bdf6f660c10b712a664c86a2a7339e8e72 A possibly relevant merge request was started @ https://invent.kde.org/plasma/plasma-workspace/-/merge_requests/4239 Git commit da35849d5f4913221ba362d3961fd29526f364d5 by Marco Martin. Committed on 15/04/2024 at 12:37. Pushed by mart into branch 'Plasma/6.0'. Fix redundantoutputs when the enabled screens change very quickly on X11, when a single transaction from KScreen disables screenA and enabled screenB, a new output order arrivers before the old output gets removed and in that moment the new screen will be considered redundant to che screen that is about to be removed, causing an invalid internal state Related: bug 456947 (cherry picked from commit dfd160bdf6f660c10b712a664c86a2a7339e8e72) c6ff79fd Fix redundantoutputs when the enabled screens change very quickly M +4 -3 shell/screenpool.cpp https://invent.kde.org/plasma/plasma-workspace/-/commit/da35849d5f4913221ba362d3961fd29526f364d5 That commit made it into Plasma 6.0.4, which has been released for a few weeks now. We think this may have resolved the issue fully; closing the bug report. However please do feel free to re-open if if you can still reproduce the crash with 6.0.4 or later. If you do, make sure it's the exact same crash, with the same backtrace. Thanks! We still have a lot of crashes in KDE Neon with Plasma 6.1.2: ``` KCrash: Application 'plasmashell' crashing... crashRecursionCounter = 2 ASSERT failure in void ScreenPool::handleScreenRemoved(QScreen*): "ScreenPool ( 0x62adc81a8310 ) Internal state: Screen Order: QList(QScreen(0x7285680235f0, name="DP-0")) "Fake" screens: QSet() Redundant screens covered by other ones: QHash() All screens, ordered by size: QList(QScreen(0x7285680235f0, name="DP-0")) All screen that QGuiApplication knows: QList(QScreen(0x7285680235f0, name="DP-0")) Current screen: QScreen(0x62adc78cd610, name=":0.0") ", file ./shell/screenpool.cpp, line 261 ``` https://crash-reports.kde.org/organizations/kde/issues/3211/?project=9&query=is%3Aunresolved&referrer=issue-stream&stream_index=1 *** Bug 490426 has been marked as a duplicate of this bug. *** See also older bug reports that have QScreen in the backtrace: Bug 454057. This one doesn't, though both end up in ScreenPool::handleScreenRemoved. *** Bug 490993 has been marked as a duplicate of this bug. *** *** Bug 486863 has been marked as a duplicate of this bug. *** A possibly relevant merge request was started @ https://invent.kde.org/plasma/plasma-workspace/-/merge_requests/4642 *** Bug 492273 has been marked as a duplicate of this bug. *** Git commit 7975ecf9631b01650f24c1f98582907e1bc3ddb6 by Méven Car. Committed on 31/08/2024 at 07:30. Pushed by meven into branch 'master'. ScreenPool: add fake screens initially So that fake screens at the gui starts are in m_fakeScreens. M +49 -0 shell/autotests/screenpooltest.cpp M +5 -0 shell/screenpool.cpp https://invent.kde.org/plasma/plasma-workspace/-/commit/7975ecf9631b01650f24c1f98582907e1bc3ddb6 Git commit ac473d8e358ac8a0f25a6236673b8aeccc3822df by Méven Car. Committed on 08/09/2024 at 09:16. Pushed by meven into branch 'Plasma/6.1'. ScreenPool: add fake screens initially So that fake screens at the gui starts are in m_fakeScreens. (cherry picked from commit 7975ecf9631b01650f24c1f98582907e1bc3ddb6) M +49 -0 shell/autotests/screenpooltest.cpp M +5 -0 shell/screenpool.cpp https://invent.kde.org/plasma/plasma-workspace/-/commit/ac473d8e358ac8a0f25a6236673b8aeccc3822df https://crash-reports.kde.org/organizations/kde/issues/3211/?project=9&query=is%3Aunresolved+issue.priority%3A%5Bhigh%2C+medium%5D&referrer=issue-stream&statsPeriod=14d&stream_index=2 Not fixed: ``` default ASSERT failure in void ScreenPool::handleScreenRemoved(QScreen*): "ScreenPool ( 0x643342b05c00 ) Internal state: Screen Order: QList(QScreen(0x64334695ae60, name="HDMI-A-0")) "Fake" screens: QSet() Redundant screens covered by other ones: QHash() All screens, ordered by size: QList(QScreen(0x64334695ae60, name="HDMI-A-0")) All screen that QGuiApplication knows: QList(QScreen(0x64334695ae60, name="HDMI-A-0")) Current screen: QScreen(0x64334273faa0, name=":0.0") ", file ./shell/screenpool.cpp, line 266 ``` Fake screen can still be not added to `m_fakeScreens`. A possibly relevant merge request was started @ https://invent.kde.org/plasma/plasma-workspace/-/merge_requests/4704 *** Bug 454057 has been marked as a duplicate of this bug. *** *** Bug 461605 has been marked as a duplicate of this bug. *** *** Bug 463396 has been marked as a duplicate of this bug. *** A possibly relevant merge request was started @ https://invent.kde.org/plasma/plasma-workspace/-/merge_requests/4761 Git commit b833966ad2e56714eddc999420dec24f7fa3e0cd by Marco Martin. Committed on 30/09/2024 at 09:32. Pushed by mart into branch 'master'. outputorderwatcher: watch for when the last output is removed watch for XCB_RANDR_CONNECTION_DISCONNECTED to immediately remove from the screen order, to avoid the data inconsistency when the last screen gets removed When the ast screen is removed, its qscreen becomes name ":0.0" as the fake screen, but nothing happens really, screenpool doesn't notice (and looking at the assert_x there are, that was expected" then the screen gets connected again, a new screen gets conencted, the old 0.0 one gets disconnected, but the screen order stuff doesn't say anything as it's still the same connector name as before so screenpool finds itself with an empty screenorder M +26 -0 libkworkspace/outputorderwatcher.cpp M +0 -5 shell/screenpool.cpp https://invent.kde.org/plasma/plasma-workspace/-/commit/b833966ad2e56714eddc999420dec24f7fa3e0cd Git commit 27e16b504b96a717691abe5cef63e9fdc3ee0670 by Marco Martin. Committed on 30/09/2024 at 09:37. Pushed by mart into branch 'Plasma/6.2'. outputorderwatcher: watch for when the last output is removed watch for XCB_RANDR_CONNECTION_DISCONNECTED to immediately remove from the screen order, to avoid the data inconsistency when the last screen gets removed When the ast screen is removed, its qscreen becomes name ":0.0" as the fake screen, but nothing happens really, screenpool doesn't notice (and looking at the assert_x there are, that was expected" then the screen gets connected again, a new screen gets conencted, the old 0.0 one gets disconnected, but the screen order stuff doesn't say anything as it's still the same connector name as before so screenpool finds itself with an empty screenorder (cherry picked from commit b833966ad2e56714eddc999420dec24f7fa3e0cd) 16c98c28 outputorderwatcher: watch for when the last output is removed 9aec2c7a add comments on what's going on b1911f16 more comments Co-authored-by: Marco Martin <notmart@gmail.com> M +26 -0 libkworkspace/outputorderwatcher.cpp M +0 -5 shell/screenpool.cpp https://invent.kde.org/plasma/plasma-workspace/-/commit/27e16b504b96a717691abe5cef63e9fdc3ee0670 |