Re: [PATCH v2 2/4] ALSA: hda: Rework snd_hdac_stream_reset() to use macros

[Amadeusz Sławiński]

On 10/5/2022 4:26 PM, Jon Hunter wrote:
> On 05/10/2022 15:07, Takashi Iwai wrote:
> > On Wed, 05 Oct 2022 15:52:01 +0200,
> > Jon Hunter wrote:
> > > On 05/10/2022 13:29, Takashi Iwai wrote:
> > >
> > > ...
> > >
> > > > > HDA playback is failing on -next for various Tegra boards. Bisect is
> > > > > point to this commit and reverting it fixes the problem. I was a bit
> > > > > puzzled why this change is causing a problem, but looking closer there
> > > > > is a difference between the previous code that was calling
> > > > > snd_hdac_stream_readb() and the new code that is calling
> > > > > snd_hdac_stream_readb_poll(). The function snd_hdac_stream_readb()
> > > > > calls snd_hdac_aligned_mmio() is see if the device has an aligned MMIO
> > > > > which Tegra does and then would call snd_hdac_aligned_read(). However,
> > > > > now the code always call readb() and this is breaking Tegra.
> > > > >
> > > > > So it is either necessary to update snd_hdac_stream_readb_poll() to
> > > > > handle this or revert this change.
> > > >
> > > > Does the patch below work?
> > > >
> > > >
> > > > thanks,
> > > >
> > > > Takashi
> > > >
> > > > -- 8< --
> > > > --- a/include/sound/hdaudio.h
> > > > +++ b/include/sound/hdaudio.h
> > > > @@ -592,8 +592,8 @@ int snd_hdac_get_stream_stripe_ctl(struct 
> > > > hdac_bus *bus,
> > > >    #define snd_hdac_stream_readb(dev, reg) \
> > > >        snd_hdac_reg_readb((dev)->bus, (dev)->sd_addr + AZX_REG_ ## reg)
> > > >    #define snd_hdac_stream_readb_poll(dev, reg, val, cond, delay_us, 
> > > > timeout_us) \
> > > > -    readb_poll_timeout((dev)->sd_addr + AZX_REG_ ## reg, val, cond, \
> > > > -               delay_us, timeout_us)
> > > > +    read_poll_timeout(snd_hdac_reg_readb, val, cond, delay_us, 
> > > > timeout_us,\
> > > > +              false, (dev)->bus, (dev)->sd_addr + AZX_REG_ ## reg)
> > > >    #define snd_hdac_stream_readl_poll(dev, reg, val, cond, delay_us, 
> > > > timeout_us) \
> > > >        readl_poll_timeout((dev)->sd_addr + AZX_REG_ ## reg, val, 
> > > > cond, \
> > > >                   delay_us, timeout_us)
> > >
> > >
> > > Amazingly it does not work. I would have thought that would, but it
> > > does not. I am a bit puzzled by that?
> >
> > Interesting, it must be a subtle difference.
> > What about passing true?  It seems that the original code has the
> > udelay(3) before the loop.
>
>
> I wondered the same and tried that, but still not working.
>
> Jon

Well in worse case we can revert the patch in question, but I would like 
to get it working...

Maybe also try to raise timeout to 1000, as what original code called 
timeout, was actually number of retries? So 300 * udelay(3) which is 
more or less 900us, so we can round it up for test?

I mean, something like:

--- a/sound/hda/hdac_stream.c
+++ b/sound/hda/hdac_stream.c
@@ -176,7 +176,7 @@ void snd_hdac_stream_reset(struct hdac_stream *azx_dev)
        snd_hdac_stream_updateb(azx_dev, SD_CTL, 0, SD_CTL_STREAM_RESET);

        /* wait for hardware to report that the stream entered reset */
-       snd_hdac_stream_readb_poll(azx_dev, SD_CTL, val, (val & 
SD_CTL_STREAM_RESET), 3, 300);
+       snd_hdac_stream_readb_poll(azx_dev, SD_CTL, val, (val & 
SD_CTL_STREAM_RESET), 3, 1000);

        if (azx_dev->bus->dma_stop_delay && dma_run_state)
                udelay(azx_dev->bus->dma_stop_delay);
@@ -184,7 +184,7 @@ void snd_hdac_stream_reset(struct hdac_stream *azx_dev)
        snd_hdac_stream_updateb(azx_dev, SD_CTL, SD_CTL_STREAM_RESET, 0);

        /* wait for hardware to report that the stream is out of reset */
-       snd_hdac_stream_readb_poll(azx_dev, SD_CTL, val, !(val & 
SD_CTL_STREAM_RESET), 3, 300);
+       snd_hdac_stream_readb_poll(azx_dev, SD_CTL, val, !(val & 
SD_CTL_STREAM_RESET), 3, 1000);

        /* reset first position - may not be synced with hw at this time */
        if (azx_dev->posbuf)


in addition to Takashi suggestion?