https://bugs.kde.org/show_bug.cgi?id=492422
--- Comment #10 from Michael Catanzaro <mcatanz...@gnome.org> --- (In reply to Mark Wielaard from comment #8) > Could you test this updated patch to see if it handles file descriptor > tracking for all three ioctls now? Yes. There is a massive improvement in the results I see. There are only a few fd reuse errors remaining, and they all look like this: ==14900== File descriptor 17: /memfd:mutter-shared (deleted) is already closed ==14900== at 0x60F9F9C: close (close.c:27) ==14900== by 0x433901CD: wsi_destroy_image (wsi_common.c:789) ==14900== by 0x4339B192: wsi_wl_swapchain_images_free (wsi_common_wayland.c:2238) ==14900== by 0x4339B1F5: wsi_wl_swapchain_destroy (wsi_common_wayland.c:2306) ==14900== by 0x50DF509: gdk_vulkan_context_dispose (gdkvulkancontext.c:370) ==14900== by 0x49E9E7A: g_object_unref (gobject.c:4415) ==14900== by 0x51A5F35: gsk_gpu_renderer_unrealize (gskgpurenderer.c:247) ==14900== by 0x51409FC: gsk_vulkan_renderer_unrealize (gskvulkanrenderer.c:142) ==14900== by 0x5102635: gsk_renderer_unrealize (gskrenderer.c:380) ==14900== by 0x4DD12A7: gtk_window_unrealize (gtkwindow.c:4407) ==14900== by 0x4B2AEB3: gtk_application_window_real_unrealize (gtkapplicationwindow.c:528) ==14900== by 0x49DEE79: g_cclosure_marshal_VOID__VOID (gmarshal.c:117) I suspect we're just missing one last ioctl that creates fds. I will investigate. And then there are several fd leaks reported. Some of them are surely real, but some look like we're perhaps missing a DRM ioctl that closes fds? Probably best for me to investigate and create a separate bug report if valgrind turns out to be to blame. -- You are receiving this mail because: You are watching all bug changes.
