If the initialization is not finished, then filling input data to
the FIFO may fail. So it is better to add initialization finishing
check in the runtime resume for suspend & resume case.
And consider the case of three instances working in parallel,
increase the retry times to 50 for more initialization time.
Signed-off-by: Shengjiu Wang <[email protected]>
---
sound/soc/fsl/fsl_asrc.c | 19 ++++++++++++++++++-
1 file changed, 18 insertions(+), 1 deletion(-)
diff --git a/sound/soc/fsl/fsl_asrc.c b/sound/soc/fsl/fsl_asrc.c
index aa5edf32d988..bae263b90ac1 100644
--- a/sound/soc/fsl/fsl_asrc.c
+++ b/sound/soc/fsl/fsl_asrc.c
@@ -20,6 +20,7 @@
#define IDEAL_RATIO_DECIMAL_DEPTH 26
#define DIVIDER_NUM 64
+#define INIT_TRY_NUM 50
#define pair_err(fmt, ...) \
dev_err(&asrc->pdev->dev, "Pair %c: " fmt, 'A' + index, ##__VA_ARGS__)
@@ -579,7 +580,7 @@ static void fsl_asrc_start_pair(struct fsl_asrc_pair *pair)
{
struct fsl_asrc *asrc = pair->asrc;
enum asrc_pair_index index = pair->index;
- int reg, retry = 10, i;
+ int reg, retry = INIT_TRY_NUM, i;
/* Enable the current pair */
regmap_update_bits(asrc->regmap, REG_ASRCTR,
@@ -592,6 +593,10 @@ static void fsl_asrc_start_pair(struct fsl_asrc_pair *pair)
reg &= ASRCFG_INIRQi_MASK(index);
} while (!reg && --retry);
+ /* FIXME: Doesn't treat initialization timeout as error */
+ if (!retry)
+ dev_warn(&asrc->pdev->dev, "initialization isn't finished\n");
+
/* Make the input fifo to ASRC STALL level */
regmap_read(asrc->regmap, REG_ASRCNCR, ®);
for (i = 0; i < pair->channels * 4; i++)
@@ -1257,6 +1262,7 @@ static int fsl_asrc_runtime_resume(struct device *dev)
{
struct fsl_asrc *asrc = dev_get_drvdata(dev);
struct fsl_asrc_priv *asrc_priv = asrc->private;
+ int reg, retry = INIT_TRY_NUM;
int i, ret;
u32 asrctr;
@@ -1295,6 +1301,17 @@ static int fsl_asrc_runtime_resume(struct device *dev)
regmap_update_bits(asrc->regmap, REG_ASRCTR,
ASRCTR_ASRCEi_ALL_MASK, asrctr);
+ /* Wait for status of initialization for every enabled pairs */
+ do {
+ udelay(5);
+ regmap_read(asrc->regmap, REG_ASRCFG, ®);
+ reg = (reg >> ASRCFG_INIRQi_SHIFT(0)) & 0x7;
+ } while ((reg != ((asrctr >> ASRCTR_ASRCEi_SHIFT(0)) & 0x7)) && --retry);
+
+ /* FIXME: Doesn't treat initialization timeout as error */
+ if (!retry)
+ dev_warn(dev, "initialization isn't finished\n");
+
return 0;
disable_asrck_clk:
--
2.34.1
On Mon, Sep 5, 2022 at 3:47 AM Shengjiu Wang <[email protected]> wrote:
> @@ -1295,6 +1301,17 @@ static int fsl_asrc_runtime_resume(struct device *dev)
> regmap_update_bits(asrc->regmap, REG_ASRCTR,
> ASRCTR_ASRCEi_ALL_MASK, asrctr);
>
> + /* Wait for status of initialization for every enabled pairs */
> + do {
> + udelay(5);
> + regmap_read(asrc->regmap, REG_ASRCFG, ®);
> + reg = (reg >> ASRCFG_INIRQi_SHIFT(0)) & 0x7;
> + } while ((reg != ((asrctr >> ASRCTR_ASRCEi_SHIFT(0)) & 0x7)) && --retry);
> +
> + /* FIXME: Doesn't treat initialization timeout as error */
> + if (!retry)
> + dev_warn(dev, "initialization isn't finished\n");
Any reason why not just dev_err?
On Mon, Sep 5, 2022 at 6:54 PM Shengjiu Wang <[email protected]> wrote:
>> > + /* Wait for status of initialization for every enabled pairs */
>> > + do {
>> > + udelay(5);
>> > + regmap_read(asrc->regmap, REG_ASRCFG, ®);
>> > + reg = (reg >> ASRCFG_INIRQi_SHIFT(0)) & 0x7;
>> > + } while ((reg != ((asrctr >> ASRCTR_ASRCEi_SHIFT(0)) & 0x7)) && --retry);
>> > +
>> > + /* FIXME: Doesn't treat initialization timeout as error */
>> > + if (!retry)
>> > + dev_warn(dev, "initialization isn't finished\n");
>>
>> Any reason why not just dev_err?
>
> Just hesitate to use dev_err. if use dev_err, then should return an error.
> May one of the pairs is finished, it still can continue.
Makes sense. In that case, why "FIXME" :)
On Tue, Sep 6, 2022 at 3:50 AM Shengjiu Wang <[email protected]> wrote:
>> >> > + /* Wait for status of initialization for every enabled pairs */
>> >> > + do {
>> >> > + udelay(5);
>> >> > + regmap_read(asrc->regmap, REG_ASRCFG, ®);
>> >> > + reg = (reg >> ASRCFG_INIRQi_SHIFT(0)) & 0x7;
>> >> > + } while ((reg != ((asrctr >> ASRCTR_ASRCEi_SHIFT(0)) & 0x7)) && --retry);
>> >> > +
>> >> > + /* FIXME: Doesn't treat initialization timeout as error */
>> >> > + if (!retry)
>> >> > + dev_warn(dev, "initialization isn't finished\n");
>> >>
>> >> Any reason why not just dev_err?
>> >
>> > Just hesitate to use dev_err. if use dev_err, then should return an error.
>> > May one of the pairs is finished, it still can continue.
>>
>> Makes sense. In that case, why "FIXME" :)
> Just want to have a record/note here, need to care about this warning.
"FIXME" feels like something is wrong and literally means that it is
waiting for a fix/solution. In your case, it's not waiting for a fix
at all, but just an annotation? So, shouldn't it be just "Note:"?