Re: [PATCH] ASoC: meson: aiu-fifo: prevent sleeping when spinlock is held

[Valerio Setti <vsetti@baylibre.com>]

Hi Martin

On 11/13/26 12:05 AM, Martin Blumenstingl wrote:
> On Wed, Nov 11, 2026 at 11:09 AM Valerio Setti <vsetti@baylibre.com> wrote:
> [...]
>> @@ -37,8 +38,11 @@ snd_pcm_uframes_t aiu_fifo_pointer(struct snd_soc_component *component,
>>          struct aiu_fifo *fifo = snd_soc_dai_dma_data_get_playback(dai);
>>          struct snd_pcm_runtime *runtime = substream->runtime;
>>          unsigned int addr;
>> +       int ret;
>>
>> -       addr = snd_soc_component_read(component, fifo->mem_offset + AIU_MEM_RD);
>> +       ret = regmap_read(component->regmap, fifo->mem_offset + AIU_MEM_RD, &addr);
>> +       if (ret < 0)
>> +               return snd_soc_ret(component->dev, ret, "Failed to read FIFO pointer");
> As far as I understand:
> - snd_pcm_uframes_t is "unsigned long"
> - snd_soc_ret() returns a (signed) int
> 
> Doesn't that lead to a signed to unsigned conversion error (highest
> bit set -> large value)?

That's something that made me thinking a bit as well and the reason why 
I implemented it this way is the following.
'snd_soc_component_read' internally goes into 
'soc_component_read_no_lock' and then to 'regmap_read'. If the latter 
fails then 'return soc_component_ret_reg_rw()' is called and that calls 
'return snd_soc_ret()' internally.

So my goal was to keep the same return value in case of errors.

> axg-fifo simply discards regmap_read's return value.
> That seems odd at first, but knowing that its regmap is MMIO backed
> (same as for AIU) means there can't ever be any error.

That's a good point indeed. I will follow this design then.
Thanks for pointing this out.

-- 
Valerio