Since we have the reliable way to detect the availability of i915 audio component, the HD-audio driver may wait for the binding safely.
This patch puts a simple completion call in the binding callback and let snd_hdac_i915_init() waiting until the i915 component gets bound or it notifies. The wait for binding has a timeout of 60 seconds, in case where the i915 module isn't present on the media by some reason. Such a situation is rather the system error, but it's still safe to get a timeout and indicate user about that. The explicit call of request_module() is still kept in this patch in order to avoid regression even for a system without the implicit driver loading. But now it's called in the condition only when no ops is found, so it's slightly optimized. The pin-down of i915 module is still kept, too, as the dynamic unbinding is still problematic. It may be racy and lead to the refcount unbalance. Signed-off-by: Takashi Iwai <[email protected]> --- sound/hda/hdac_i915.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) diff --git a/sound/hda/hdac_i915.c b/sound/hda/hdac_i915.c index 812777f9af1f..6455d239a928 100644 --- a/sound/hda/hdac_i915.c +++ b/sound/hda/hdac_i915.c @@ -22,6 +22,7 @@ #include <sound/hda_i915.h> static struct i915_audio_component *hdac_acomp; +static DECLARE_COMPLETION(bind_comp); /** * snd_hdac_set_codec_wakeup - Enable / disable HDMI/DP codec wakeup @@ -248,10 +249,12 @@ static int hdac_component_master_bind(struct device *dev) goto out_unbind; } + complete_all(&bind_comp); return 0; out_unbind: component_unbind_all(dev, acomp); + complete_all(&bind_comp); return ret; } @@ -339,6 +342,7 @@ int snd_hdac_i915_init(struct hdac_bus *bus) return -ENOMEM; bus->audio_component = acomp; hdac_acomp = acomp; + init_completion(&bind_comp); component_match_add(dev, &match, hdac_component_master_match, bus); ret = component_master_add_with_match(dev, &hdac_component_master_ops, @@ -346,12 +350,10 @@ int snd_hdac_i915_init(struct hdac_bus *bus) if (ret < 0) goto out_err; - /* - * Atm, we don't support deferring the component binding, so make sure - * i915 is loaded and that the binding successfully completes. - */ - request_module("i915"); + if (!acomp->ops) + request_module("i915"); + wait_for_completion_timeout(&bind_comp, msecs_to_jiffies(60 * 1000)); if (!acomp->ops || acomp->ops->disabled) { ret = -ENODEV; goto out_master_del; -- 2.8.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
