Bug 102277

Not aware of anything wrong with window switching since we fixed the last bug in Xorg circa xorg-server-1.18.3. Start with attaching your Xorg, but it sounds like we will need to trawl through --enable-debug=full to see what is actually being requested.

Also, if you can a short video would be useful. From your description, it sounds like there is a stale tearfree backbuffer (which may be lost damage, i.e. we didn't copy across the changes made before switching), but I am not aware of Ratpoison so I don't know how that operates.

Created attachment 133594 [details]
Xorg.0.log with Intel --enable-debug=full, gzipped

Xorg.0.log with Intel --enable-debug=full, gzipped

While I was recording the server crashed so I'll attach what dmesg I have too. Hopefully you can play this ogv format recordmydesktop created, I was able to play it with mpv. The video shows me trying to tab out but I couldn't manage. I didn't spam it since I was afraid of another crash, but sometimes it works.

Created attachment 133596 [details]
video of tabbing attempt

Created attachment 133597 [details]
remaining dmesg

(In reply to Tod Jackson from comment #4)
> While I was recording the server crashed so I'll attach what dmesg I have
> too. Hopefully you can play this ogv format recordmydesktop created, I was
> able to play it with mpv. The video shows me trying to tab out but I
> couldn't manage. I didn't spam it since I was afraid of another crash, but
> sometimes it works.

The crash will be an assertion, useful to know, but it may be one of the dubious sanity checks (there are quite a few consistency checks that forget they are acting on shared buffers for example). I did mean to go back and soften those asserts that are handled anyway (I made them explode to make it easier to debug).

[  +0.000004] [drm:drm_atomic_set_fb_for_plane] Set [FB:51] for plane state ffff88000cf4d3c0
[  +0.000003] [drm:drm_atomic_check_only] checking ffff88002dbcd000
[  +0.000011] [drm:intel_plane_atomic_calc_changes] [CRTC:26:pipe A] has [PLANE:25:cursor A] with fb 51
[  +0.000004] [drm:intel_plane_atomic_calc_changes] [PLANE:25:cursor A] visible 1 -> 1, off 0, on 0, ms 0
[  +0.000009] [drm:drm_atomic_commit] commiting ffff88002dbcd000
[  +0.000012] [drm:drm_vblank_enable] enabling vblank on crtc 0, ret: 0
[  +0.000007] [drm:drm_calc_vbltimestamp_from_scanoutpos] crtc 0 : v 0x5 p(0,0)@ 138379.223796 -> 138379.223796 [e 0 us, 0 rep]
[  +0.000005] [drm:drm_update_vblank_count] updating vblank count on crtc 0: current=8281713, diff=0, hw=8282350 hw_last=8282350
[  +0.000005] [drm:vblank_disable_fn] disabling vblank on crtc 0
[  +0.000007] [drm:drm_calc_vbltimestamp_from_scanoutpos] crtc 0 : v 0x5 p(0,0)@ 138379.223813 -> 138379.223813 [e 0 us, 0 rep]
[  +0.000004] [drm:drm_update_vblank_count] updating vblank count on crtc 0: current=8281713, diff=0, hw=8282350 hw_last=8282350
[  +0.000010] [drm:drm_atomic_state_default_clear] Clearing atomic state ffff88002dbcd000
[  +0.000004] [drm:drm_mode_object_unreference] OBJ ID: 58 (2)
[  +0.000003] [drm:drm_mode_object_unreference] OBJ ID: 60 (2)
[  +0.000005] [drm:drm_mode_object_unreference] OBJ ID: 51 (4)
[  +0.000003] [drm:drm_atomic_state_free] Freeing atomic state ffff88002dbcd000
[  +0.000004] [drm:drm_mode_object_unreference] OBJ ID: 51 (3)
[  +0.295171] dota2[6076]: segfault at 0 ip 00007f51a9dad2c6 sp 00007ffc566cc430 error 4 in libclient.so[7f51a74be000+4aa8000]
[  +6.303233] [drm:drm_ioctl] pid=6163, dev=0xe200, auth=1, DRM_IOCTL_MODE_SETCRTC
[  +0.000006] [drm:drm_mode_setcrtc] [CRTC:26:pipe A]
[  +0.000004] [drm:drm_mode_setcrtc] [CONNECTOR:36:LVDS-1]
[  +0.000003] [drm:drm_atomic_state_init] Allocated atomic state ffff880084124800

I think libclient.so is from SDL 2. I just hours ago reported an mpv bug with that same library segfaulting. So something's broken on my system. That being said, I had this issue a long time ago on different distros. Still, interesting...

After spending all day yesterday playing with this, I don't think it's an Intel bug. On other drivers in environments lacking moderm WMs the SDL in Dota has various problems, such as raising the window by itself.