vt(4): Add support to "downgrade" from eg. vt_fb to vt_vga

[dumbbell]

Summary

The main purpose is to be able to unload a KMS driver. It's at least useful to me while working on the DRM stack, because it saves me a reboot :)

The patch introduces a new vt_driver callback, vd_reset_t, which is called after the old driver is stopped (old = fbd) and before the new driver is initialized (new = vt_vga). The callback is really implemented in drm_fb_helper.c where vga_pci_repost() is called (a new public function in vga_pci) to restore the VGA state.

Currently, the DRM code in HEAD seems to turn off the screen during shutdown, and re-POSTing the card isn't enough to turn it back on. The problem doesn't occur with my DRM 3.8 development branch (the console comes back with vt_vga and everything works).

Test Plan

kldload radeonkms
kldunload radeonkms

The console should still work

This can't be tested with i915kms which doesn't support unloading.

[emaste]

Adding jhb as a subscriber to look at vga_pci.c

sys/dev/drm2/radeon/radeon_fb.c

294

Why do we need vga_pci_repost both here and in vt_kms_reset?

[dumbbell]

sys/dev/drm2/radeon/radeon_fb.c

294

This is a mistake I don't have locally anymore. I still haven't published a revised patch (which is not finished anyway).

However, I only changed what calls what to address concerns expressed outside of Phabricator. The vga_pci_repost() function is unchanged and it needs review, because I'm not sure of what I do :)

[emaste]

What's next here?

[dumbbell]

I need to publish a newer version of the patch, but didn't have the time to do it so far. Once xserver 1.14 is committed, I will work on a few vt(4) changes, like D1004 and this one.

[dumbbell]

Each part should be properly isolated now. Only vt_vga nows about repost.

[dumbbell]

sys/dev/drm2/radeon/radeon_fb.c

295

Note that the device_delete_child() is moved to a central function in drm_fb_helper.c in my DRM update branch. This avoids to do it in each driver. This function is missing in our current DRM and I didn't want to import new code for now.

[jhb]

sys/dev/vt/vt.h

130

I would perhaps call these 'vd_prev_*' instead of 'vd_old_*'

[dumbbell]

Rename vd_old_* to vd_prev_*

[dumbbell]

Do you have any comment on this?

I'd like to commit it to HEAD to ease the work on i915.

[• hselasky]

Hi,

I need this for dev/usb/video/udl.c which unloads when the device detaches. Can you please commit it?

--HPS

[dumbbell]

Regenerated the patch after some changes in vt(4)

If there is no objection, I'm going to commit this patch as I need it to
work on DRM.

[dumbbell]

Closed by commit rS279487 (authored by @dumbbell).