Re: [PATCH v2] host/rootfs: Don't wait for /dev/dri/card0

[Alyssa Ross <hi@alyssa.is>]

[-- Attachment #1: Type: text/plain, Size: 2389 bytes --]

Demi Marie Obenour <demiobenour@gmail.com> writes:

> On 6/24/26 04:55, Alyssa Ross wrote:
>> Alyssa Ross <hi@alyssa.is> writes:
>> 
>>> Demi Marie Obenour <demiobenour@gmail.com> writes:
>>>
>>>> Weston can find a card itself.  If the card is not ready, it will exit,
>>>> but s6 will automatically restart it.
>>>>
>>>> Signed-off-by: Demi Marie Obenour <demiobenour@gmail.com>
>>>
>>> I don't think this commit message really captures it.  As long as the
>>> card is not hotplugged, it _will_ be ready, as far as I know, because it
>>> will become ready when DRM is loaded, before userspace starts.  This
>>> makes it sound like Weston not finding a card and being restarted by s6
>>> would be routine behaviour, whereas we'd actually not expect to ever see
>>> it, except in the very unusual case of somebody having no cards at boot
>>> and later hotplugging one.  (Even then it's not ideal to need to restart
>>> in a loop, but that case can wait for COSMIC.)
>>>
>>> I suggest the following alternative, which also removes some other
>>> obsolete instances of the same pattern.
>>>
>>> From 71216f0aa36049cbd3c7b7562e6e7ff225a1c810 Mon Sep 17 00:00:00 2001
>>> From: Alyssa Ross <hi@alyssa.is>
>>> Date: Tue, 26 May 2026 16:05:21 +0200
>>> Subject: [PATCH] Stop waiting for /dev/dri/card0
>>>
>>> On the host, which does not run in a well-controlled VM environment,
>>> there might not be a card0 (sometimes you just get card1).  This
>>> causes hangs.  It hasn't been necessary to wait for non-hotplugged DRI
>>> devices to exist since the Nixpkgs kernel set CONFIG_DRM=y.  (If
>> 
>> Alas this is not true upon further investigation.  I saw
>> release/checks/wayland fail with this applied, because while DRM was
>> loaded pre-userspace, DRM_BOCHS still wasn't.  s6 would indeed just
>> restart Weston until it works, but I don't really want a restart loop to
>> be a routine thing…
>
> Is it okay to just let the restart loop happen until Spectrum switches
> to COSMIC?  An alternative would be to use a udev rule and a FIFO,
> but that's extra work that will be thrown away later.

I think it's a small amount of work, and we can reuse some of what we
did before with mdevd, so I'll just do it.  More is needed than just
switching to COSMIC: COSMIC needs to be modified so that it won't refuse
to start up if there are no cards.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 227 bytes --]