Re: [PATCH 1/5] ASoC: sh: rz-ssi: Drop calling rz_ssi_pio_recv() recursively

From: Lad, Prabhakar
Date: Mon Jan 10 2022 - 11:03:46 EST


Hi Cezary,

Thank you for the review.

On Mon, Jan 10, 2022 at 3:48 PM Cezary Rojewski
<cezary.rojewski@xxxxxxxxx> wrote:
>
> On 2022-01-10 10:47 AM, Lad Prabhakar wrote:
> > Instead of recursively calling rz_ssi_pio_recv() use a while loop
> > to read the samples from RX fifo.
>
> Recursion and loops are means for doing something repeatedly. Could you
> specify _why_ such change was made i.e. the conversion from one method
> into the other? I bet the code is not being changed for the sake of
> changing it, the reason is simply missing in the commit message.
>
I had feedback from Pavel "recursion is unwelcome in kernel due to
limited stack use." which I did agree with as a result I have come up
with this patch. Also to add this driver will later be used on Renesas
RZ/A2 SoC's which runs with limited memory.

> Please note that refactoring below function into while-loop has a side
> effect: everything had to be indented by additional tab. Generally,
> readability increases if function is shaped 'linearly'.
>
I do agree, my initial patch just added a jump back to the start of
the function if there are more samples, but Biju suggested to use a
while loop instead.

> > This also fixes an issue where the return value of rz_ssi_pio_recv()
> > was ignored when called recursively.
> >
> > Fixes: 03e786bd4341 ("ASoC: sh: Add RZ/G2L SSIF-2 driver")
> > Reported-by: Pavel Machek <pavel@xxxxxxx>
> > Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > Reviewed-by: Biju Das <biju.das.jz@xxxxxxxxxxxxxx>
> > ---
> > sound/soc/sh/rz-ssi.c | 68 ++++++++++++++++++++++---------------------
> > 1 file changed, 35 insertions(+), 33 deletions(-)
> >
> > diff --git a/sound/soc/sh/rz-ssi.c b/sound/soc/sh/rz-ssi.c
> > index fa0cc08f70ec..37466f65c2b0 100644
> > --- a/sound/soc/sh/rz-ssi.c
> > +++ b/sound/soc/sh/rz-ssi.c
> > @@ -411,54 +411,56 @@ static int rz_ssi_pio_recv(struct rz_ssi_priv *ssi, struct rz_ssi_stream *strm)
> > {
> > struct snd_pcm_substream *substream = strm->substream;
> > struct snd_pcm_runtime *runtime;
> > + bool done = false;
> > u16 *buf;
> > int fifo_samples;
> > int frames_left;
> > - int samples = 0;
> > + int samples;
> > int i;
> >
> > if (!rz_ssi_stream_is_valid(ssi, strm))
> > return -EINVAL;
> >
> > runtime = substream->runtime;
> > - /* frames left in this period */
> > - frames_left = runtime->period_size - (strm->buffer_pos %
> > - runtime->period_size);
> > - if (frames_left == 0)
> > - frames_left = runtime->period_size;
> >
> > - /* Samples in RX FIFO */
> > - fifo_samples = (rz_ssi_reg_readl(ssi, SSIFSR) >>
> > - SSIFSR_RDC_SHIFT) & SSIFSR_RDC_MASK;
> > -
> > - /* Only read full frames at a time */
> > - while (frames_left && (fifo_samples >= runtime->channels)) {
> > - samples += runtime->channels;
> > - fifo_samples -= runtime->channels;
> > - frames_left--;
> > - }
> > + while (!done) {
>
> I wonder if converting this into do-while isn't a better option. Maybe
> I'm missing something but 'done' flag seems to be changed only as an
> outcome of the last if-statement (last step) in this entire procedure.
> Perhaps condition from said if-statement could also be moved into
> 'while' portion of do-while loop.
>
Agreed.

> > + /* frames left in this period */
> > + frames_left = runtime->period_size -
> > + (strm->buffer_pos % runtime->period_size);
> > + if (!frames_left)
> > + frames_left = runtime->period_size;
> > +
> > + /* Samples in RX FIFO */
> > + fifo_samples = (rz_ssi_reg_readl(ssi, SSIFSR) >>
> > + SSIFSR_RDC_SHIFT) & SSIFSR_RDC_MASK;
> > +
> > + /* Only read full frames at a time */
> > + samples = 0;
> > + while (frames_left && (fifo_samples >= runtime->channels)) {
> > + samples += runtime->channels;
> > + fifo_samples -= runtime->channels;
> > + frames_left--;
> > + }
> >
> > - /* not enough samples yet */
> > - if (samples == 0)
> > - return 0;
> > + /* not enough samples yet */
> > + if (!samples)
> > + break;
> >
> > - /* calculate new buffer index */
> > - buf = (u16 *)(runtime->dma_area);
> > - buf += strm->buffer_pos * runtime->channels;
> > + /* calculate new buffer index */
> > + buf = (u16 *)(runtime->dma_area);
>
> Is the second pair of brackets needed?
>
Nope can be dropped.

Cheers,
Prabhakar