Re: [PATCH i-g-t] lib/drmtest: Remove VKMS from Exclusion list

["José Expósito" <jose.exposito89@gmail.com>]

Hi Mark,

On Thu, Oct 16, 2025 at 10:51:07AM -0400, Mark Yacoub wrote:
> On Thu, Oct 16, 2025 at 7:00 AM José Expósito <jose.exposito89@gmail.com> wrote:
> >
> > Hi everyone,
> >
> > Thanks for CCing me Louis.
> >
> > On Thu, Oct 16, 2025 at 10:40:32AM +0200, Louis Chauvet wrote:
> > >
> > >
> > > Le 15/10/2025 à 20:16, Kamil Konieczny a écrit :
> > > > Hi Mark,
> > > > On 2025-10-09 at 12:53:24 -0400, Mark Yacoub wrote:
> > > > > Hi Kamil,
> > > > >
> > > > > Thanks for your feedback. I'd like to elaborate on my use case to provide
> > > > > more context about the problem I'm trying to solve.
> > > > >
> > > > > I am building IGT for Android, and the same build system is deployed on
> > > > > both physical hardware and virtual machines. On the hardware, the
> > > > > appropriate hardware driver is correctly detected, so there are no issues
> > > > > there. However, on the VM, VKMS is running, but it's currently excluded
> > > > > from `DRIVER_ANY`, which prevents the tests from running against it. I need
> > > > > the tests to execute on VKMS in this scenario.
> > > > >
> > > > > Using `IGT_FORCE_DRIVER` would require me to fork the build processes for
> > > > > hardware and VM, which I would prefer to avoid. My goal is to maintain a
> > > >
> > > > You do not need to fork it nor recompile, this is env var for
> > > > CI run with tests, it uses the same igt binaries only makes them
> > > > using a forced driver when opening card. Now it do not work as
> > > > no user before needed to force it into vgem or VKMS.
> > > >
> > > > In theory you could define it like: export IGT_FORCE_DRIVER=vkms
> > > > and then all kms tests and core_ ones should honor it.
> > >
> > > +CC: José
> > >
> > > Hi,
> > >
> > > I had the same issue this week, but IGT_FORCE_DRIVER is not enough: all the
> > > tests using DRIVER_ANY are skipped.
> > >
> > > # export IGT_DEVICE=drm:/dev/dri/card0
> > > # export IGT_FORCE_DRIVER=vkms
> > > # export IGT_LOG_LEVEL=debug
> > > # /usr/libexec/igt-gpu-tools/kms_plane
> > > (kms_plane:125) igt_core-DEBUG: Notice: using IGT_DEVICE env:
> > > (kms_plane:125) igt_core-DEBUG: [drm:/dev/dri/card0]
> > > [IGT] kms_plane: executing
> > > IGT-Version: 2.2-g6d82d231ced6 (x86_64) (Linux:
> > > 6.17.0-rc6-01217-g1ddd2876eeb9 x86_64)
> > > Using IGT_SRANDOM=1760602572 for randomisation
> > > (kms_plane:125) drmtest-DEBUG: Looking for devices to open using filter 0:
> > > drm:/dev/dri/card0
> > > (kms_plane:125) drmtest-DEBUG: Filter matched /dev/dri/card0 |
> > > (kms_plane:125) igt_kmod-DEBUG: Module vkms already inserted
> > > Test requirement not met in function drm_open_driver, file
> > > ../igt/lib/drmtest.c:735:
> > > Test requirement: !(fd<0)
> > > No known gpu found for chipset flags 0x4294965755 (any)
> > > Last errno: 2, No such file or directory
> > > Subtest pixel-format: SKIP (0.000s)
> > > Subtest pixel-format-source-clamping: SKIP (0.000s)
> > > Subtest plane-position-covered: SKIP (0.000s)
> > > Subtest plane-position-hole: SKIP (0.000s)
> > > Subtest plane-position-hole-dpms: SKIP (0.000s)
> > > Subtest plane-panning-top-left: SKIP (0.000s)
> > > Subtest plane-panning-bottom-right: SKIP (0.000s)
> > > Subtest plane-panning-bottom-right-suspend: SKIP (0.000s)
> > > Subtest planar-pixel-format-settings: SKIP (0.000s)
> > > [IGT] kms_plane: exiting, ret=77
> > > (kms_plane:125) igt_core-DEBUG: Exiting with status code 77
> > >
> > > José suggested this fix, it seems to properly use vkms if IGT_FORCE_DRIVER
> > > is set:
> > >
> > > diff --git a/lib/drmtest.c b/lib/drmtest.c
> > > index a1fb0b5ba..f630bc157 100644
> > > --- a/lib/drmtest.c
> > > +++ b/lib/drmtest.c
> > > @@ -350,7 +350,7 @@ int __drm_open_device(const char *name, unsigned int
> > > chipset)
> > >
> > >         modulename_to_chipset(dev_name, &chip);
> > >
> > > -       if ((chipset & chip) == chip) {
> > > +       if ((chipset & chip) == chip || (forced && strcmp(forced, dev_name)
> > > == 0)) {
> > >                 log_opened_device_path(name);
> > >                 return fd;
> > >         }
> > >
> > > @Mark, is it enough for you?
> > > @José, do you want to send the proper patch?
> >
> > Sorry I couldn't send a patch earlier, I have been pretty busy this week.
> >
> > The reason why VKMS is excluded from DRIVERS_ANY is that, when a test
> > selects any driver ("drm_open_driver(DRIVER_ANY)") hardware drivers are
> > prioritized over virtual drivers (vgem, vkms or virtio) to avoid them
> > being accidentally selected due to device-load ordering.
> Hey, so I'm a bit confused here. I have a VM with VKMS, and another
> hardware DUT with intel, so there is never an accidental use of the
> wrong driver.
> Do you have hardware that might carry both drivers?

I have never run into this issue, but looking at the comment in drmtest.h
this has been an issue in the past.

There is a patch by Kamil addressing this. I have tested it and it worked
for me:
https://lore.kernel.org/igt-dev/20251016160544.122657-1-kamil.konieczny@linux.intel.com/T/#t

Jose

> >
> > There is a nice comment in drmtest.h explaining it with more detail. So, in
> > my opinion, VKMS is correctly excluded from DRIVERS_ANY.
> >
> > However, I'd expect that setting the IGT_FORCE_DRIVER environment variable to
> > "vkms" would actually use VKMS for testing.
> > I have refined a bit the initial workaround I discuss with  Louis 2 days ago
> > and I think this could be an acceptable solution:
> >
> > diff --git a/lib/drmtest.c b/lib/drmtest.c
> > index a1fb0b5ba..8f295d436 100644
> > --- a/lib/drmtest.c
> > +++ b/lib/drmtest.c
> > @@ -350,6 +350,14 @@ int __drm_open_device(const char *name, unsigned int chipset)
> >
> >         modulename_to_chipset(dev_name, &chip);
> >
> > +       /*
> > +        * Virtual drivers like vgem or vkms are excluded from DRIVER_ANY and
> > +        * therefore it is not possible to use them in tests using the
> > +        * IGT_FORCE_DRIVER environment variable. Include them when forced.
> > +        */
> > +       if (forced && chipset == DRIVER_ANY)
> > +               chipset |= chip;
> > +
> >         if ((chipset & chip) == chip) {
> >                 log_opened_device_path(name);
> >                 return fd;
> >
> > I don't know if making that check even more precise would be a better solution.
> > For example, calling a "is_virtual_driver()" helper:
> >
> >   if (forced && chipset == DRIVER_ANY && is_virtual_driver(chip))
> >           chipset |= chip;
> >
> > I can send a patch with that fix, but I'd appreciate as much feedback as possible
> > as this code is used by every tests and I don't have enough experience with IGT
> > to make sure it doesn't introduce a bug.
> >
> > Best wishes,
> > Jose
> >
> > >
> > > Hav a nice day,
> > > Louis Chauvet
> > >
> > > > Regards,
> > > > Kamil
> > > >
> > > > > unified build system and allow IGT to run on whatever driver is present. Is
> > > > > there a way to achieve this?
> > > > >
> > > > > I understand that the current code prioritizes the hardware driver, which
> > > > > is a desirable behavior. I'm trying to understand the potential issues that
> > > > > might arise if VKMS were not excluded in other contexts.
> > > > >
> > > > > Thank you!
> > > > > Mark
> > > > >
> > > > >
> > > > > On Thu, Oct 9, 2025 at 12:46 PM Kamil Konieczny <
> > > > > kamil.konieczny@linux.intel.com> wrote:
> > > > >
> > > > > > Hi Mark,
> > > > > > On 2025-10-08 at 12:02:09 -0400, Mark Yacoub wrote:
> > > > > > > [Why]
> > > > > > > VKMS (Virtual VKMS) is typically running on a virtual machine where it's
> > > > > > > expected to be without a contention with a hardware driver.
> > > > > > > When it's there, we would like to run IGT on it.
> > > > > > >
> > > > > > > [How]
> > > > > > > Delete VKMS from DRIVER_ANY's exclusion list to allow IGT to be picked
> > > > > > > up.
> > > > > > >
> > > > > > > Tested: Ran kms_bw on Android's VM with VKMS enabled.
> > > > > > >
> > > > > > > Change-Id: Ib13ad9261b8c2617370c02d0ffb5aa878bf3d8f8
> > > > > > > Signed-off-by: Mark Yacoub <markyacoub@google.com>
> > > > > > > ---
> > > > > > >   lib/drmtest.h | 4 +---
> > > > > > >   1 file changed, 1 insertion(+), 3 deletions(-)
> > > > > > >
> > > > > > > diff --git a/lib/drmtest.h b/lib/drmtest.h
> > > > > > > index 8416aa491..ab991da57 100644
> > > > > > > --- a/lib/drmtest.h
> > > > > > > +++ b/lib/drmtest.h
> > > > > > > @@ -63,10 +63,8 @@ int __get_drm_device_name(int fd, char *name, int
> > > > > > name_size);
> > > > > > >    * on a system with them as well as a supported driver, you can end up
> > > > > > >    * with a near-100% skip rate if you don't explicitly specify the
> > > > > > device,
> > > > > > >    * depending on device-load ordering.
> > > > > > > - *
> > > > > > > - * Exclude VKMS to prefer hardware drivers.
> > > > > > >    */
> > > > > > > -#define DRIVER_ANY   ~(DRIVER_VGEM | DRIVER_VKMS | DRIVER_VIRTIO)
> > > > > > > +#define DRIVER_ANY   ~(DRIVER_VGEM | DRIVER_VIRTIO)
> > > > > >
> > > > > > Nak, this is bad as there could be vkms module also on host.
> > > > > >
> > > > > > Let me work on improving IGT_FORCE_DRIVER.
> > > > > >
> > > > > > Regards,
> > > > > > Kamil
> > > > > >
> > > > > > >
> > > > > > >   /*
> > > > > > >    * Compile friendly enum for i915/xe.
> > > > > > > --
> > > > > > > 2.51.0.740.g6adb054d12-goog
> > > > > > >
> > > > > >
> > >
> > > --
> > > --
> > > Louis Chauvet, Bootlin
> > > Embedded Linux and Kernel engineering
> > > https://bootlin.com
> > >