[PATCH libdrm] tests/util: Add support for meson module

[Emil Velikov]

On 20 January 2017 at 16:17, Thierry Reding <treding at nvidia.com> wrote:
> On Fri, Jan 20, 2017 at 02:13:00PM +0000, Emil Velikov wrote:
>> On 19 January 2017 at 09:19, Thierry Reding <treding at nvidia.com> wrote:
>> > On Wed, Jan 18, 2017 at 02:59:21PM +0100, Neil Armstrong wrote:
>> >> Add support for Amlogic Meson DRM driver merged for Linux 4.10.
>> >>
>> >> Signed-off-by: Neil Armstrong <narmstrong at baylibre.com>
>> >> ---
>> >>  tests/util/kms.c | 1 +
>> >>  1 file changed, 1 insertion(+)
>> >
>> > Applied, thanks.
>> >
>> Was going to say "NACK use Thierry's helpers" but you've beat me to it.
>> As a Tl;DR: we _really_ want to stop using drmOpen* for anything that's KMS.
>
> I think it's still useful to have this helper to iterate over all
> supported driver because it enables easy testing with just a simple
> modetest.
>
> That said, for (almost) as long as I remember I've been using the -M
> option to prevent modetest from iterating over the list, which can take
> fairly long if you've got DRM_DEBUG messages enabled.
>
> I'm not sure I understand exactly what you're suggesting by "use
> Thierry's helpers". modetest and other tests use util_open() internally
> now. That in turn uses drmOpen(), though we're of course open to change
> that. Are you suggesting we somehow use drmDevice to locate existing
> devices?
>
Yes using drmDevice is what I had in mind. The only thing that was
stopping me from doing that is the lack of platform devices support.
With that in we can start purging _everything_ that uses drmOpen*.

> We could implement some heuristic that finds the first device with a
> primary node, but what if we want to support the -M option? There is
> currently no way of getting the driver from drmDevice. Maybe that's
> something we should add anyway.
>
The -M (kernel module name as given by drmGetVersion) is another
interesting topic. Feel free to skip to the Tl;Dr below.

If doing the ioctl implicitly via drmOpen or drmDevice wakes up the
device, keeping in mind that:
 - it can take some time, and
 - you don't always need the info

A couple of elaborate workarounds include:
 - use only on demand - add DRM_DEVICE_GET_MODULE_NAME or alike to the
drmDevice2 API
Needs a big "this can be VERY slow" warning in the documentation/man
pages... first we need actual man pages for libdrm ;-)
 - deprecate the whole thing and use the compat strings - how do we
handle the PCI devices ?
 - use the kernel module name (foo.ko) - kind of making it an ABI, but
iirc we've already have users which depend on it
Neither PCI devices nor platform ones are consistent in their naming -
foo.ko vs the value returned by the IOCTL.

We might be able to combine the latter two ... need to double-check.

Tl;Dr: For the moment I'd leave it to the user to call drmGetVersion.

Thanks
Emil