On Tue, 06 Oct 2020 13:30:40 +0200, Kai Vehmanen wrote:
Hi,
this simple bugfix started to feel a bit like getting stuck in quicksand, so I'm looking for some early input via this RFC series.
Basicly hdac_i915.c should not use global state to track communication with i915 driver. But how to get handle of "hdac_bus*? I considered a few options:
- add hdac_bus as a member of drm_audio_component.h
-> seems wrong as this is really an audio side implementation)
- embed copy of drm_audio_component to 'struct hdac_bus', so I could use container_of() on the device handle to get to the bus
-> wasted space to keep a copy at hdac_bus level (note: snd-hda-codec-hdmi do this by embedding a copy of ops to "struct hdmi_spec")
- add another devres entry to store the hdac_bus directly in acomp_init and a new helper function to query it
I now implemented option 3 in this RFC series as it seemed cleanest and most local to hdac_component.c, where the problem stems from. It's still somewhat messy, and I'm wondering if I'm overlooking some obvious alternative. We could dig this deeper into i915 specific code, but OTOH, hdac_bus is an argument snd_hdac_acomp_init(), so it's common for all.
Kai Vehmanen (2): ALSA: hda - keep track of HDA core bus instance in acomp ALSA: hda/i915 - fix list corruption with concurrent probes
Another option would be to move the completion into the common acomp helper from i915-specific one. That is,
- Add bind_complete field into struct drm_audio_component, initialize it at snd_hdac_acomp_init()
- Call complete_all(&acomp->bind_complete) at the end of hdac_component_master_bind()
- Remove / replace i915's own completion with the hdac's one. The i915_init_ops can be dropped.
thanks,
Takashi