> -------- Original Message --------
> Subject: Re: [iGVT-g] problems with Tina Zhang's dmabuf patches
> Local Time: September 1, 2017 4:11 AM
> UTC Time: September 1, 2017 1:11 AM
> From: zhiyuan.lv(a)intel.com
> To: Ethidium Bromide <ebrombugs(a)protonmail.com>
> igvt-g(a)lists.01.org <igvt-g(a)lists.01.org>
> On Thu, Aug 31, 2017 at 01:28:36PM -0400, Ethidium Bromide wrote:
>> I"m using GVT with Tina Zhang"s dmabuf patches on gvt-staging and
topic/dmabuf. I can still VNC into my Windows VM (and can confirm that the Intel drivers
do in fact load properly, run benchmarks etc.) but the SDL display shuts off about halfway
into the boot process (when the Intel graphics drivers are loaded, I"m assuming) with
the error message "primary create buffer failed using ioctl:-19". Here"s a
pastebin of the entire output from qemu: https://paste.debian.net/983916/
(it stays like
that for a while, then starts repeating that last error message very rapidly - over 10
times per second probably)
>> How would I go about debugging this?
> Thanks for trying this!
> Most likely it is because the framebuffer on the virtual display port
> not available. We have a known issue on the display part. Currently we
> give VM a virtual monitor on specific display port. When that port is
> not available on the host machine, there will be some problem.
> We have a workaround for that issue. As the formal fix, we will give VM
> a pure virtual VBT table. If you ran into the same issue, below patch
> might help:
> If there are still problems, Tina could provide more inputs next week,
> who is on vacation this week. Thanks!
>> iGVT-g mailing list
Original message was formatted incorrectly and isn't showing up in the mailing list
archives, have to double post, sorry.
That patch doesn't work for me, here's the dmesg output I get with drm.debug=3
when I try to create a vgpu with the patch applied:
[ 1252.302526] iommu: Adding device 3065ccfd-89cc-46f0-9cf4-1ab7c74b7607 to group 16
[ 1252.302532] vfio_mdev 3065ccfd-89cc-46f0-9cf4-1ab7c74b7607: MDEV: group_id = 16
[ 1252.302741] [drm:intel_gvt_create_vgpu] gvt: core: handle 0 low 128 MB high 512 MB
[ 1252.304939] [drm:intel_vgpu_alloc_resource] gvt: core: vgpu1: alloc low GM start
1444000 size 8000000
[ 1252.304945] [drm:intel_vgpu_alloc_resource] gvt: core: vgpu1: alloc high GM start
dffa9000 size 20000000
[ 1252.304975] [drm:populate_pvinfo_page] gvt: core: Populate PVINFO PAGE for vGPU 1
[ 1252.304980] [drm:populate_pvinfo_page] gvt: core: aperture base [GMADR] 0x1444000 size
[ 1252.304984] [drm:populate_pvinfo_page] gvt: core: hidden base [GMADR] 0xdffa9000
[ 1252.304988] [drm:populate_pvinfo_page] gvt: core: fence size 4
[ 1252.308529] [drm:intel_vgpu_init_gtt] gvt: mm: vgpu1 create scratch_pt: type 12
[ 1252.308532] [drm:intel_vgpu_init_gtt] gvt: mm: vgpu1 create scratch_pt: type 13
[ 1252.308537] [drm:intel_vgpu_init_gtt] gvt: mm: vgpu1 create scratch_pt: type 14
[ 1252.308541] [drm:intel_vgpu_init_gtt] gvt: mm: vgpu1 create scratch_pt: type 15
[ 1252.308915] [drm:intel_vgpu_create] gvt: failed to create intel vgpu: -22
[ 1252.309027] iommu: Removing device 3065ccfd-89cc-46f0-9cf4-1ab7c74b7607 from group 16
[ 1252.309031] vfio_mdev 3065ccfd-89cc-46f0-9cf4-1ab7c74b7607: MDEV: detaching iommu
Looks like the patched display.c is giving an -EINVAL, any ideas?
You mean the function "intel_vgpu_init_display()"? If that is the case,
the port B/C/D of the display are all disabled in video bios. Are you
using a laptop for the experiment?
What the patch tries to do is to align the virtual port used in VM with
the physical port information in VBT. Currently the patch did not handle
the case if the machine only enabled eDP (port A).
As a formal fix, we are cooking patch to provide a pure virtual OpRegin
to VM, which is not related to host BIOS setting. Then all will be fine.
iGVT-g mailing list