SOF_HDA: need to clean-up exit sequences #299

Activity

1. plbossart

   assigned

   keyonjie

   on Nov 17, 2018
2. mengdonglin

   added

   enhancementNew feature or requestNew feature or request

   on Nov 22, 2018
3. jocelyn-li

   assigned

   bardliao

   and unassigned

   keyonjie

   on Nov 26, 2018
4. mengdonglin

   added

   bugSomething isn't workingSomething isn't working

   P1Blocker bugs or important featuresBlocker bugs or important features

   and removed

   bugSomething isn't workingSomething isn't working

   on Nov 26, 2018
5. bardliao commented on Nov 27, 2018

   

   bardliao

   on Nov 27, 2018

   Collaborator

   More actions

   @plbossart It seems we didn't use hdac stream control APIs. I think that's why we don't use snd_hdac_ext_stop_streams(). And hda_dsp_stream_free() do things similar to snd_hdac_stream_free_all() + snd_hdac_bus_free_stream_pages()

   snd_hdac_bus_free_stream_pages()
                   bus->io_ops->dma_free_pages(bus, &s->bdl);
                   bus->io_ops->dma_free_pages(bus, &bus->rb);
                   bus->io_ops->dma_free_pages(bus, &bus->posbuf);
   

   snd_hdac_stream_free_all()
                   snd_hdac_ext_stream_decouple(bus, stream, false);
                   list_del(&s->list);
                   kfree(stream);
   

   hda_dsp_stream_free()
                   snd_dma_free_pages(&bus->posbuf);
                   snd_dma_free_pages(&bus->rb);
                   snd_dma_free_pages(&s->bdl);
   		list_del(&s->list);
   		devm_kfree(sdev->dev, stream);
   

   The only thing we missed here is snd_hdac_ext_stream_decouple(). But I don't know if we need to do it or it is done in somewhere else.

   void snd_hdac_ext_stream_decouple(struct hdac_bus *bus,
                                   struct hdac_ext_stream *stream, bool decouple)
   {
           struct hdac_stream *hstream = &stream->hstream;
           u32 val;
           int mask = AZX_PPCTL_PROCEN(hstream->index);
   
           spin_lock_irq(&bus->reg_lock);
           val = readw(bus->ppcap + AZX_REG_PP_PPCTL) & mask;
   
           if (decouple && !val)
                   snd_hdac_updatel(bus->ppcap, AZX_REG_PP_PPCTL, mask, mask);
           else if (!decouple && val)
                   snd_hdac_updatel(bus->ppcap, AZX_REG_PP_PPCTL, mask, 0);
   
           stream->decoupled = decouple;
           spin_unlock_irq(&bus->reg_lock);
   }
   

   snd_hdac_ext_bus_exit() and snd_hdac_link_free_all() are add in #311
6. plbossart commented on Nov 27, 2018

   

   plbossart

   on Nov 27, 2018

   MemberAuthor

   More actions

   snd_hdac_ext_stream_decouple() is used in hda-dai.c

   linux/sound/soc/sof/intel/hda-dai.c

   Line 138 in f585396

   snd_hdac_ext_stream_decouple(bus, stream, false);

   
   I have absolutely no idea why it's only called on TRIGGER_SUSPEND, makes no sense to me.
7. bardliao commented on Dec 3, 2018

   

   bardliao

   on Dec 3, 2018

   Collaborator

   More actions

   @plbossart I just checked snd_hdac_ext_stream_decouple(). I believe that we need to call snd_hdac_ext_stream_decouple(false) in hda_dsp_stream_free(). I didn't get SNDRV_PCM_TRIGGER_SUSPEND cmd even when the system is suspended. And hdac_ext_link_stream_assign() will call snd_hdac_ext_stream_decouple(true) when system boots up.
   I will send a PR for it tomorrow.
8. plbossart

   closed this as completedin #361on Jan 16, 2019
9. plbossart

   added a commit that references this issue on Nov 29, 2022

   tracing: Fix wild-memory-access in register_synth_event()

   ...

   1b5f1c3
10. oder-chiou

    added a commit that references this issue on May 5, 2026

    mm: fix kernel BUG when userfaultfd_move encounters swapcache

    ...

    c50f8e6