GVT-g KVM potential regression with 6.1.0-rc5

[He, Yu]

Hi GVT Folks,

I recently encountered a blank display issue when testing GVT-g on Intel iGFX. 
Below are my tests.

Kernels I used:
#1 37c9e6e4 -> Jason Gunthorpe's vfio iommufd compat based on 6.1.0-rc3
#2 77f7ccbc -> Jason Gunthorpe's vfio iommufd compat based on 6.1.0-rc3
                newer version compared with #1
#3 57f62422 -> Jason Gunthorpe's vfio iommufd compat (77f7ccbc) + 6.1.0-rc5 merge
#4 Jason Gunthorpe's vfio iommufd compat + 6.1.0-rc6 merge


Results I got:

- For #1&2 , display was normal after testing finished
- For #3, blank screen after 2D cairo running
- For #4, display was normal during/after testing, but met blank screen once in 24hrs after testing finished (display could recover when restarting gdm3 which not worked with rc5). For a double check, I re-tested the whole cases but the blank display issue didn't reproduce in another 15 hrs.

I checked with Jason Gunthorpe offline, the vfio diff between #2 and #3 is not likely to result in the blank display. So highly suspect it is a regression in gvt or i915.


Logs of VM when display was blank:

[   14.361137] [drm] Initialized i915 1.6.0 20201103 for 0000:00:03.0 on minor 0
[   14.388315] fbcon: i915drmfb (fb0) is primary device
[   24.595849] i915 0000:00:03.0: [drm] *ERROR* flip_done timed out
[   24.595856] i915 0000:00:03.0: [drm] *ERROR* [CRTC:51:pipe A] commit wait timed out
[   24.617269] Console: switching to colour frame buffer device 240x75
[   24.639108] i915 0000:00:03.0: [drm] fb0: i915drmfb frame buffer device
[   27.606771] rfkill: input handler disabled
[115345.696470] gdm3[650]: segfault at 0 ip 00007f3310708494 sp 00007ffcfc18e9a8 error 4 in libglib-2.0.so.0.6400.2[7f33106e4000+84000] likely on CPU 2 (core 2, socket 0)
[115345.696500] Code: f8 5b 5d c3 66 90 f3 0f 1e fa 48 83 ec 08 e8 a3 cf fd ff 85 c0 0f 94 c0 48 83 c4 08 0f b6 c0 c3 66 0f 1f 44 00 00 f3 0f 1e fa <0f> be 07 41 b8 05 15 00 00 84 c0 74 1d 0f 1f 80 00 00 00 00 44 89
[115345.696582] gsd-keyboard[1088]: segfault at 0 ip 00007fc70eed6494 sp 00007ffdc623c1b8 error 4 in libglib-2.0.so.0.6400.2[7fc70eeb2000+84000] likely on CPU 2 (core 2, socket 0)
[115345.696597] Code: f8 5b 5d c3 66 90 f3 0f 1e fa 48 83 ec 08 e8 a3 cf fd ff 85 c0 0f 94 c0 48 83 c4 08 0f b6 c0 c3 66 0f 1f 44 00 00 f3 0f 1e fa <0f> be 07 41 b8 05 15 00 00 84 c0 74 1d 0f 1f 80 00 00 00 00 44 89
[115345.698706] gdm-session-wor[661]: segfault at 0 ip 00007f9990c4e494 sp 00007ffc50aae428 error 4 in libglib-2.0.so.0.6400.2[7f9990c2a000+84000] likely on CPU 1 (core 1, socket 0)
[115345.698716] Code: f8 5b 5d c3 66 90 f3 0f 1e fa 48 83 ec 08 e8 a3 cf fd ff 85 c0 0f 94 c0 48 83 c4 08 0f b6 c0 c3 66 0f 1f 44 00 00 f3 0f 1e fa <0f> be 07 41 b8 05 15 00 00 84 c0 74 1d 0f 1f 80 00 00 00 00 44 89
[115346.878394] rfkill: input handler enabled
[115348.518884] rfkill: input handler disabled
[115500.612521] SGI XFS with ACLs, security attributes, realtime, quota, no debug enabled
[115500.621449] JFS: nTxBlock = 8192, nTxLock = 65536
[115500.632789] ntfs: driver 2.1.32 [Flags: R/O MODULE].
[115500.648481] QNX4 filesystem 0.2.3 registered.


Best regards,
He,Yu