[PATCH v3] ASoC: qcom: Fix for DMA interrupt clear reg overwriting

[Marek Szyprowski]

Hi,

On 21.06.2021 14:52, Srinivas Kandagatla wrote:
>
> On 17/06/2021 20:35, Marek Szyprowski wrote:
>>
>> On 09.06.2021 09:23, Srinivasa Rao Mandadapu wrote:
>>> The DMA interrupt clear register overwritten during
>>> simultaneous playback and capture in lpass platform
>>> interrupt handler. It's causing playback or capture stuck
>>> in similtaneous plaback on speaker and capture on dmic test.
>>> Update appropriate reg fields of corresponding channel instead
>>> of entire register write.
>>>
>>> Fixes: commit c5c8635a04711 ("ASoC: qcom: Add LPASS platform driver")
>>>
>>> Signed-off-by: Srinivasa Rao Mandadapu <srivasam at codeaurora.org>
>>
> Can you please try this patch and let us know if this fixes the issue
>
> ------------------------->cut<-------------------------------
> Author: Srinivas Kandagatla <srinivas.kandagatla at linaro.org>
> Date:   Mon Jun 21 12:38:43 2021 +0100
>
>     ASoC: qcom: lpass-cpu: mark IRQ_CLEAR register as volatile and 
> readable
>
>     Currently IRQ_CLEAR register is marked as write-only, however using
>     regmap_update_bits on this register will have some side effects.
>     so mark IRQ_CLEAR register appropriately as readable and volatile.
>
>     Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla at linaro.org>

This fixes the issue observed on DragonBoard410c. Feel free to add:

Reported-by: Marek Szyprowski <m.szyprowski at samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski at samsung.com>
> diff --git a/sound/soc/qcom/lpass-cpu.c b/sound/soc/qcom/lpass-cpu.c
> index 0b9cbf2ce505..8998697cd1e1 100644
> --- a/sound/soc/qcom/lpass-cpu.c
> +++ b/sound/soc/qcom/lpass-cpu.c
> @@ -525,6 +525,8 @@ static bool lpass_cpu_regmap_readable(struct 
> device *dev, unsigned int reg)
>                         return true;
>
>         for (i = 0; i < v->irq_ports; ++i) {
> +               if (reg == LPAIF_IRQCLEAR_REG(v, i))
> +                       return true;
>                 if (reg == LPAIF_IRQEN_REG(v, i))
>                         return true;
>                 if (reg == LPAIF_IRQSTAT_REG(v, i))
> @@ -566,9 +568,12 @@ static bool lpass_cpu_regmap_volatile(struct 
> device *dev, unsigned int reg)
>         struct lpass_variant *v = drvdata->variant;
>         int i;
>
> -       for (i = 0; i < v->irq_ports; ++i)
> +       for (i = 0; i < v->irq_ports; ++i) {
> +               if (reg == LPAIF_IRQCLEAR_REG(v, i))
> +                       return true;
>                 if (reg == LPAIF_IRQSTAT_REG(v, i))
>                         return true;
> +       }
>
>         for (i = 0; i < v->rdma_channels; ++i)
>                 if (reg == LPAIF_RDMACURR_REG(v, i))
>
> ------------------------->cut<-------------------------------
>
> --srini
>
>> This patch landed recently in linux-next as commit da0363f7bfd3 ("ASoC:
>> qcom: Fix for DMA interrupt clear reg overwriting"). It breaks ALSA
>> playback on DragonBoard 410c (arch/arm64/boot/dts/qcom/apq8016-sbc.dts).
>> After applying this patch, running 'speaker-test -l1' never finishes.
>> There is no error nor kernel warning message. Before that commit, the
>> playback worked fine on that board.
>>
>>> ---
>>> Changes since v2:
>>>     -- Removed redundant variables.
>>> Changes since v1:
>>>     -- Subject lines changed.
>>>    sound/soc/qcom/lpass-platform.c | 12 ++++++------
>>>    1 file changed, 6 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/sound/soc/qcom/lpass-platform.c 
>>> b/sound/soc/qcom/lpass-platform.c
>>> index 0df9481ea4c6..f9df76d37858 100644
>>> --- a/sound/soc/qcom/lpass-platform.c
>>> +++ b/sound/soc/qcom/lpass-platform.c
>>> @@ -526,7 +526,7 @@ static int lpass_platform_pcmops_trigger(struct 
>>> snd_soc_component *component,
>>>                return -EINVAL;
>>>            }
>>>    -        ret = regmap_write(map, reg_irqclr, val_irqclr);
>>> +        ret = regmap_update_bits(map, reg_irqclr, val_irqclr, 
>>> val_irqclr);
>>>            if (ret) {
>>>                dev_err(soc_runtime->dev, "error writing to irqclear 
>>> reg: %d\n", ret);
>>>                return ret;
>>> @@ -650,10 +650,11 @@ static irqreturn_t lpass_dma_interrupt_handler(
>>>        struct lpass_variant *v = drvdata->variant;
>>>        irqreturn_t ret = IRQ_NONE;
>>>        int rv;
>>> -    unsigned int reg = 0, val = 0;
>>> +    unsigned int reg, val, mask;
>>>        struct regmap *map;
>>>        unsigned int dai_id = cpu_dai->driver->id;
>>>    +    mask = LPAIF_IRQ_ALL(chan);
>>>        switch (dai_id) {
>>>        case LPASS_DP_RX:
>>>            map = drvdata->hdmiif_map;
>>> @@ -676,8 +677,7 @@ static irqreturn_t lpass_dma_interrupt_handler(
>>>        return -EINVAL;
>>>        }
>>>        if (interrupts & LPAIF_IRQ_PER(chan)) {
>>> -
>>> -        rv = regmap_write(map, reg, LPAIF_IRQ_PER(chan) | val);
>>> +        rv = regmap_update_bits(map, reg, mask, 
>>> (LPAIF_IRQ_PER(chan) | val));
>>>            if (rv) {
>>>                dev_err(soc_runtime->dev,
>>>                    "error writing to irqclear reg: %d\n", rv);
>>> @@ -688,7 +688,7 @@ static irqreturn_t lpass_dma_interrupt_handler(
>>>        }
>>>           if (interrupts & LPAIF_IRQ_XRUN(chan)) {
>>> -        rv = regmap_write(map, reg, LPAIF_IRQ_XRUN(chan) | val);
>>> +        rv = regmap_update_bits(map, reg, mask, 
>>> (LPAIF_IRQ_XRUN(chan) | val));
>>>            if (rv) {
>>>                dev_err(soc_runtime->dev,
>>>                    "error writing to irqclear reg: %d\n", rv);
>>> @@ -700,7 +700,7 @@ static irqreturn_t lpass_dma_interrupt_handler(
>>>        }
>>>           if (interrupts & LPAIF_IRQ_ERR(chan)) {
>>> -        rv = regmap_write(map, reg, LPAIF_IRQ_ERR(chan) | val);
>>> +        rv = regmap_update_bits(map, reg, mask, 
>>> (LPAIF_IRQ_ERR(chan) | val));
>>>            if (rv) {
>>>                dev_err(soc_runtime->dev,
>>>                    "error writing to irqclear reg: %d\n", rv);
>>
>
Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland