Hi Hans,
Problem was that lock was held while calling dmaengine APIs, which
causes double lock issue when dma callback was called under context of
dma_async_issue_pending() (see dcmi_start_dma() & below comments).
Rest of changes are around spin_lock() changed to spin_lock_irq() for
safer lock management.
On 02/26/2018 02:56 PM, Hans Verkuil wrote:
> On 02/22/2018 10:49 AM, Hugues Fruchet wrote:
>> Fix lock scheme leading to spurious freeze.
>
> Can you elaborate a bit more? It's hard to review since you don't
> describe what was wrong and why this fixes the problem.
>
> Regards,
>
> Hans
>
>>
>> Signed-off-by: Hugues Fruchet <hugues.fruc...@st.com>
>> ---
>> version 2:
>> - dcmi_buf_queue() refactor to avoid to have "else" after "return"
>> (warning detected by checkpatch.pl --strict -f stm32-dcmi.c)
>>
>> drivers/media/platform/stm32/stm32-dcmi.c | 57
>> +++++++++++++------------------
>> 1 file changed, 24 insertions(+), 33 deletions(-)
>>
>> diff --git a/drivers/media/platform/stm32/stm32-dcmi.c
>> b/drivers/media/platform/stm32/stm32-dcmi.c
>> index 2fd8bed..5de18ad 100644
>> --- a/drivers/media/platform/stm32/stm32-dcmi.c
>> +++ b/drivers/media/platform/stm32/stm32-dcmi.c
>> @@ -197,7 +197,7 @@ static void dcmi_dma_callback(void *param)
>> struct dma_tx_state state;
>> enum dma_status status;
>>
>> - spin_lock(&dcmi->irqlock);
>> + spin_lock_irq(&dcmi->irqlock);
>>
>> /* Check DMA status */
>> status = dmaengine_tx_status(chan, dcmi->dma_cookie, &state);
>> @@ -239,7 +239,7 @@ static void dcmi_dma_callback(void *param)
>> dcmi->errors_count++;
>> dcmi->active = NULL;
>>
>> - spin_unlock(&dcmi->irqlock);
>> + spin_unlock_irq(&dcmi->irqlock);
>> return;
>> }
>>
>> @@ -248,13 +248,11 @@ static void dcmi_dma_callback(void *param)
>>
>> list_del_init(&dcmi->active->list);
>>
>> - if (dcmi_start_capture(dcmi)) {
>> + spin_unlock_irq(&dcmi->irqlock);
Lock is released here before calling dcmi_start_capture() which calls
dma_async_issue_pending()
>> + if (dcmi_start_capture(dcmi))
>> dev_err(dcmi->dev, "%s: Cannot restart capture
>> on DMA complete\n",
>> __func__);
>> -
>> - spin_unlock(&dcmi->irqlock);
>> - return;
>> - }
>> + return;
>> }
>>
>> break;
>> @@ -263,7 +261,7 @@ static void dcmi_dma_callback(void *param)
>> break;
>> }
>>
>> - spin_unlock(&dcmi->irqlock);
>> + spin_unlock_irq(&dcmi->irqlock);
>> }
>>
>> static int dcmi_start_dma(struct stm32_dcmi *dcmi,
>> @@ -360,7 +358,7 @@ static irqreturn_t dcmi_irq_thread(int irq, void *arg)
>> {
>> struct stm32_dcmi *dcmi = arg;
>>
>> - spin_lock(&dcmi->irqlock);
>> + spin_lock_irq(&dcmi->irqlock);
>>
>> /* Stop capture is required */
>> if (dcmi->state == STOPPING) {
>> @@ -370,7 +368,7 @@ static irqreturn_t dcmi_irq_thread(int irq, void *arg)
>>
>> complete(&dcmi->complete);
>>
>> - spin_unlock(&dcmi->irqlock);
>> + spin_unlock_irq(&dcmi->irqlock);
>> return IRQ_HANDLED;
>> }
>>
>> @@ -383,35 +381,34 @@ static irqreturn_t dcmi_irq_thread(int irq, void *arg)
>> __func__);
>>
>> dcmi->errors_count++;
>> - dmaengine_terminate_all(dcmi->dma_chan);
>> -
>> dev_dbg(dcmi->dev, "Restarting capture after DCMI error\n");
>>
>> - if (dcmi_start_capture(dcmi)) {
>> + spin_unlock_irq(&dcmi->irqlock);
Lock is released here before calling dma APIs (terminate then
async_issue_pending)
>> + dmaengine_terminate_all(dcmi->dma_chan);
>> +
>> + if (dcmi_start_capture(dcmi))
>> dev_err(dcmi->dev, "%s: Cannot restart capture on
>> overflow or error\n",
>> __func__);
>> -
>> - spin_unlock(&dcmi->irqlock);
>> - return IRQ_HANDLED;
>> - }
>> + return IRQ_HANDLED;
>> }
>>
>> - spin_unlock(&dcmi->irqlock);
>> + spin_unlock_irq(&dcmi->irqlock);
>> return IRQ_HANDLED;
>> }
>>
>> static irqreturn_t dcmi_irq_callback(int irq, void *arg)
>> {
>> struct stm32_dcmi *dcmi = arg;
>> + unsigned long flags;
>>
>> - spin_lock(&dcmi->irqlock);
>> + spin_lock_irqsave(&dcmi->irqlock, flags);
>>
>> dcmi->misr = reg_read(dcmi->regs, DCMI_MIS);
>>
>> /* Clear interrupt */
>> reg_set(dcmi->regs, DCMI_ICR, IT_FRAME | IT_OVR | IT_ERR);
>>
>> - spin_unlock(&dcmi->irqlock);
>> + spin_unlock_irqrestore(&dcmi->irqlock, flags);
>>
>> return IRQ_WAKE_THREAD;
>> }
>> @@ -490,9 +487,8 @@ static void dcmi_buf_queue(struct vb2_buffer *vb)
>> struct stm32_dcmi *dcmi = vb2_get_drv_priv(vb->vb2_queue);
>> struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
>> struct dcmi_buf *buf = container_of(vbuf, struct dcmi_buf, vb);
>> - unsigned long flags = 0;
>>
>> - spin_lock_irqsave(&dcmi->irqlock, flags);
>> + spin_lock_irq(&dcmi->irqlock);
>>
>> if ((dcmi->state == RUNNING) && (!dcmi->active)) {
>> dcmi->active = buf;
>> @@ -500,19 +496,15 @@ static void dcmi_buf_queue(struct vb2_buffer *vb)
>> dev_dbg(dcmi->dev, "Starting capture on buffer[%d] queued\n",
>> buf->vb.vb2_buf.index);
>>
>> - if (dcmi_start_capture(dcmi)) {
>> + spin_unlock_irq(&dcmi->irqlock);
Lock is released here before calling dma APIs (terminate then
async_issue_pending)
>> + if (dcmi_start_capture(dcmi))
>> dev_err(dcmi->dev, "%s: Cannot restart capture on
>> overflow or error\n",
>> __func__);
>> -
>> - spin_unlock_irqrestore(&dcmi->irqlock, flags);
>> - return;
>> - }
>> } else {
>> /* Enqueue to video buffers list */
>> list_add_tail(&buf->list, &dcmi->buffers);
>> + spin_unlock_irq(&dcmi->irqlock);
>> }
>> -
>> - spin_unlock_irqrestore(&dcmi->irqlock, flags);
>> }
>>
>> static int dcmi_start_streaming(struct vb2_queue *vq, unsigned int count)
>> @@ -598,20 +590,17 @@ static int dcmi_start_streaming(struct vb2_queue *vq,
>> unsigned int count)
>>
>> dev_dbg(dcmi->dev, "Start streaming, starting capture\n");
>>
>> + spin_unlock_irq(&dcmi->irqlock);
>> ret = dcmi_start_capture(dcmi);
>> if (ret) {
>> dev_err(dcmi->dev, "%s: Start streaming failed, cannot start
>> capture\n",
>> __func__);
>> -
>> - spin_unlock_irq(&dcmi->irqlock);
>> goto err_subdev_streamoff;
>> }
>>
>> /* Enable interruptions */
>> reg_set(dcmi->regs, DCMI_IER, IT_FRAME | IT_OVR | IT_ERR);
>>
>> - spin_unlock_irq(&dcmi->irqlock);
>> -
>> return 0;
>>
>> err_subdev_streamoff:
>> @@ -654,7 +643,9 @@ static void dcmi_stop_streaming(struct vb2_queue *vq)
>> dev_err(dcmi->dev, "%s: Failed to stop streaming, subdev
>> streamoff error (%d)\n",
>> __func__, ret);
>>
>> + spin_lock_irq(&dcmi->irqlock);
>> dcmi->state = STOPPING;
>> + spin_unlock_irq(&dcmi->irqlock);
>>
>> timeout = wait_for_completion_interruptible_timeout(&dcmi->complete,
>> time_ms);
>>
>
Best regards,
Hugues.
