On Tue, 13 Sep 2022 14:04:40 +0200
Janusz Krzysztofik <[email protected]> wrote:
> Starting from commit 0edabd8a4a08 ("core_hotunplug: fix audio unbind
> logic") audio modules are unloaded unconditionally before graphics driver
> unbind. This is required for i915 to be unbound cleanly from a graphics
> device. However, we have no indications that non-Intel platforms require
> the same.
Actually, there was a much more generic solution merged, but that depends
on a patch upstream to lib/module not applied. See:
https://lore.kernel.org/all/[email protected]/
So, we ended reverting the changes on this changeset:
03706cd84099 ("lib/igt_kmod: remove the conditional audio removal code")
In any case, other drivers may have similar issues, if the audio driver
binds into the video driver. So, I would place any chipset-specific logic
for the current logic inside igt_audio_driver_unload(), as this will make it
more generic, as any other audio driver unbind needs will be self-contained
inside the library, and, if some day Kernel lib/modules.c gets fixed to
properly report bind (as proposed on my patchset), the library will get
the hack to have the snd-hda-intel name hardcoded on it, and the code
will work with any GPU/snd audio driver combination.
> Since we care for other platforms using this test and contributing to it,
> not developing their own, keep i915 workarounds limited to i915 devices.
>
> Signed-off-by: Janusz Krzysztofik <[email protected]>
> ---
> tests/core_hotunplug.c | 58 +++++++++++++++++++++++-------------------
> 1 file changed, 32 insertions(+), 26 deletions(-)
>
> diff --git a/tests/core_hotunplug.c b/tests/core_hotunplug.c
> index c3756889c2..29cb26ab12 100644
> --- a/tests/core_hotunplug.c
> +++ b/tests/core_hotunplug.c
> @@ -55,6 +55,7 @@ struct hotunplug {
> bool need_healthcheck;
> bool has_intel_perf;
> char *snd_driver;
> + int chipset;
> };
>
> /* Helpers */
> @@ -140,19 +141,20 @@ static void prepare(struct hotunplug *priv)
> static void driver_unbind(struct hotunplug *priv, const char *prefix,
> int timeout)
> {
> - /*
> - * FIXME: on some devices, the audio driver (snd_hda_intel)
> - * binds into the i915 driver. On such hardware, kernel warnings
> - * and errors may happen if i915 is unbind/removed before removing
> - * first the audio driver.
> - * So, add a logic that unloads the audio driver before trying to
> - * unbind i915 driver, reloading it when binding again.
> - */
> - if (igt_audio_driver_unload(&priv->snd_driver)) {
> - igt_skip("Audio driver %s in use, skipping test\n",
> - priv->snd_driver);
> - } else if (priv->snd_driver) {
> - igt_info("Unloaded audio driver %s\n", priv->snd_driver);
> + if (priv->chipset == DRIVER_INTEL) {
> + /*
> + * FIXME: on some devices, the audio driver (snd_hda_intel)
> + * binds into the i915 driver. On such hardware, kernel warnings
> + * and errors may happen if i915 is unbind/removed before
> removing
> + * first the audio driver.
> + * So, add a logic that unloads the audio driver before trying
> to
> + * unbind i915 driver, reloading it when binding again.
> + */
> + if (igt_audio_driver_unload(&priv->snd_driver))
> + igt_skip("Audio driver %s in use, skipping test\n",
> + priv->snd_driver);
> + else if (priv->snd_driver)
> + igt_info("Unloaded audio driver %s\n",
> priv->snd_driver);
> }
>
> local_debug("%sunbinding the driver from the device\n", prefix);
> @@ -198,19 +200,20 @@ static void device_unplug(struct hotunplug *priv, const
> char *prefix,
> {
> igt_require(priv->fd.sysfs_dev == -1);
>
> - /*
> - * FIXME: on some devices, the audio driver (snd_hda_intel)
> - * binds into the i915 driver. On such hardware, kernel warnings
> - * and errors may happen if i915 is unbind/removed before removing
> - * first the audio driver.
> - * So, add a logic that unloads the audio driver before trying to
> - * unbind i915 driver, reloading it when binding again.
> - */
> - if (igt_audio_driver_unload(&priv->snd_driver)) {
> - igt_skip("Audio driver %s in use, skipping test\n",
> - priv->snd_driver);
> - } else if (priv->snd_driver) {
> - igt_info("Unloaded audio driver %s\n", priv->snd_driver);
> + if (priv->chipset == DRIVER_INTEL) {
> + /*
> + * FIXME: on some devices, the audio driver (snd_hda_intel)
> + * binds into the i915 driver. On such hardware, kernel warnings
> + * and errors may happen if i915 is unbind/removed before
> removing
> + * first the audio driver.
> + * So, add a logic that unloads the audio driver before trying
> to
> + * unbind i915 driver, reloading it when binding again.
> + */
> + if (igt_audio_driver_unload(&priv->snd_driver))
> + igt_skip("Audio driver %s in use, skipping test\n",
> + priv->snd_driver);
> + else if (priv->snd_driver)
> + igt_info("Unloaded audio driver %s\n",
> priv->snd_driver);
> }
>
> priv->fd.sysfs_dev = openat(priv->fd.sysfs_bus, priv->dev_bus_addr,
> @@ -625,6 +628,7 @@ igt_main
> .need_healthcheck = true,
> .has_intel_perf = false,
> .snd_driver = NULL,
> + .chipset = DRIVER_ANY,
> };
>
> igt_fixture {
> @@ -634,6 +638,8 @@ igt_main
> igt_skip_on_f(fd_drm < 0, "No known DRM device found\n");
>
> if (is_i915_device(fd_drm)) {
> + priv.chipset = DRIVER_INTEL;
> +
> gem_quiescent_gpu(fd_drm);
> igt_require_gem(fd_drm);
>
