diff mbox series

dmaengine: ti: omap-dma: Improved memcpy polling support

Message ID 20190618132416.26874-1-peter.ujfalusi@ti.com (mailing list archive)
State New, archived
Headers show
Series dmaengine: ti: omap-dma: Improved memcpy polling support | expand

Commit Message

Peter Ujfalusi June 18, 2019, 1:24 p.m. UTC
When a DMA client driver does not set the DMA_PREP_INTERRUPT because it
does not want to use interrupts for DMA completion or because it can not
rely on DMA interrupts due to executing the memcpy when interrupts are
disabled it will poll the status of the transfer.

If the interrupts are enabled then the cookie will be set completed in the
interrupt handler so only check in HW completion when the polling is really
needed.

Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
---
Hi,

This patch fine-tunes the omap-dma polled memcpy support to be inline with how
the EDMA driver is handling it.

The polled completion can be tested by applying:
https://patchwork.kernel.org/patch/10966499/

and run the dmatest with polled = 1 on boards where sDMA is used.

Or boot up any dra7 family device with display enabled. The workaround for DMM
errata i878 uses polled DMA memcpy.

Regards,
Peter

 drivers/dma/ti/omap-dma.c | 37 ++++++++++++++++++++++++-------------
 1 file changed, 24 insertions(+), 13 deletions(-)

Comments

Vinod Koul July 5, 2019, 6:23 a.m. UTC | #1
On 18-06-19, 16:24, Peter Ujfalusi wrote:
> When a DMA client driver does not set the DMA_PREP_INTERRUPT because it
> does not want to use interrupts for DMA completion or because it can not
> rely on DMA interrupts due to executing the memcpy when interrupts are
> disabled it will poll the status of the transfer.
> 
> If the interrupts are enabled then the cookie will be set completed in the
> interrupt handler so only check in HW completion when the polling is really
> needed.
> 
> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
> ---
> Hi,
> 
> This patch fine-tunes the omap-dma polled memcpy support to be inline with how
> the EDMA driver is handling it.
> 
> The polled completion can be tested by applying:
> https://patchwork.kernel.org/patch/10966499/
> 
> and run the dmatest with polled = 1 on boards where sDMA is used.
> 
> Or boot up any dra7 family device with display enabled. The workaround for DMM
> errata i878 uses polled DMA memcpy.
> 
> Regards,
> Peter
> 
>  drivers/dma/ti/omap-dma.c | 37 ++++++++++++++++++++++++-------------
>  1 file changed, 24 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/dma/ti/omap-dma.c b/drivers/dma/ti/omap-dma.c
> index 5ba7d8485026..75d8f7e13c8d 100644
> --- a/drivers/dma/ti/omap-dma.c
> +++ b/drivers/dma/ti/omap-dma.c
> @@ -94,6 +94,7 @@ struct omap_desc {
>  	bool using_ll;
>  	enum dma_transfer_direction dir;
>  	dma_addr_t dev_addr;
> +	bool polled;
>  
>  	int32_t fi;		/* for OMAP_DMA_SYNC_PACKET / double indexing */
>  	int16_t ei;		/* for double indexing */
> @@ -834,20 +835,10 @@ static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
>  
>  	ret = dma_cookie_status(chan, cookie, txstate);
>  
> -	if (!c->paused && c->running) {
> -		uint32_t ccr = omap_dma_chan_read(c, CCR);
> -		/*
> -		 * The channel is no longer active, set the return value
> -		 * accordingly
> -		 */
> -		if (!(ccr & CCR_ENABLE))
> -			ret = DMA_COMPLETE;
> -	}
> -
> +	spin_lock_irqsave(&c->vc.lock, flags);
>  	if (ret == DMA_COMPLETE || !txstate)
> -		return ret;
> +		goto out;
>  
> -	spin_lock_irqsave(&c->vc.lock, flags);
>  	vd = vchan_find_desc(&c->vc, cookie);
>  	if (vd) {
>  		txstate->residue = omap_dma_desc_size(to_omap_dma_desc(&vd->tx));
> @@ -868,6 +859,23 @@ static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
>  	}
>  	if (ret == DMA_IN_PROGRESS && c->paused)
>  		ret = DMA_PAUSED;
> +
> +out:
> +	if (ret == DMA_IN_PROGRESS && c->running && c->desc &&
> +	    c->desc->polled && c->desc->vd.tx.cookie == cookie) {

heh, that makes quite a read!

checking DMA_IN_PROGRESS should not make sense, as we should bail out at
the start if it is completed

I think other can be optimzed to make it a better read!

> +		uint32_t ccr = omap_dma_chan_read(c, CCR);
> +		/*
> +		 * The channel is no longer active, set the return value
> +		 * accordingly
> +		 */
> +		if (!(ccr & CCR_ENABLE)) {
> +			struct omap_desc *d = c->desc;
> +			ret = DMA_COMPLETE;
> +			omap_dma_start_desc(c);
> +			vchan_cookie_complete(&d->vd);
> +		}
> +	}
> +
>  	spin_unlock_irqrestore(&c->vc.lock, flags);
>  
>  	return ret;
> @@ -1233,7 +1241,10 @@ static struct dma_async_tx_descriptor *omap_dma_prep_dma_memcpy(
>  	d->ccr = c->ccr;
>  	d->ccr |= CCR_DST_AMODE_POSTINC | CCR_SRC_AMODE_POSTINC;
>  
> -	d->cicr = CICR_DROP_IE | CICR_FRAME_IE;
> +	if (tx_flags & DMA_PREP_INTERRUPT)
> +		d->cicr |= CICR_FRAME_IE;
> +	else
> +		d->polled = true;
>  
>  	d->csdp = data_type;
>  
> -- 
> Peter
> 
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
Peter Ujfalusi July 12, 2019, 9:10 p.m. UTC | #2
On 5.7.2019 9.23, Vinod Koul wrote:
> On 18-06-19, 16:24, Peter Ujfalusi wrote:
>> When a DMA client driver does not set the DMA_PREP_INTERRUPT because it
>> does not want to use interrupts for DMA completion or because it can not
>> rely on DMA interrupts due to executing the memcpy when interrupts are
>> disabled it will poll the status of the transfer.
>>
>> If the interrupts are enabled then the cookie will be set completed in the
>> interrupt handler so only check in HW completion when the polling is really
>> needed.
>>
>> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
>> ---
>> Hi,
>>
>> This patch fine-tunes the omap-dma polled memcpy support to be inline with how
>> the EDMA driver is handling it.
>>
>> The polled completion can be tested by applying:
>> https://patchwork.kernel.org/patch/10966499/
>>
>> and run the dmatest with polled = 1 on boards where sDMA is used.
>>
>> Or boot up any dra7 family device with display enabled. The workaround for DMM
>> errata i878 uses polled DMA memcpy.
>>
>> Regards,
>> Peter
>>
>>  drivers/dma/ti/omap-dma.c | 37 ++++++++++++++++++++++++-------------
>>  1 file changed, 24 insertions(+), 13 deletions(-)
>>
>> diff --git a/drivers/dma/ti/omap-dma.c b/drivers/dma/ti/omap-dma.c
>> index 5ba7d8485026..75d8f7e13c8d 100644
>> --- a/drivers/dma/ti/omap-dma.c
>> +++ b/drivers/dma/ti/omap-dma.c
>> @@ -94,6 +94,7 @@ struct omap_desc {
>>  	bool using_ll;
>>  	enum dma_transfer_direction dir;
>>  	dma_addr_t dev_addr;
>> +	bool polled;
>>  
>>  	int32_t fi;		/* for OMAP_DMA_SYNC_PACKET / double indexing */
>>  	int16_t ei;		/* for double indexing */
>> @@ -834,20 +835,10 @@ static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
>>  
>>  	ret = dma_cookie_status(chan, cookie, txstate);
>>  
>> -	if (!c->paused && c->running) {
>> -		uint32_t ccr = omap_dma_chan_read(c, CCR);
>> -		/*
>> -		 * The channel is no longer active, set the return value
>> -		 * accordingly
>> -		 */
>> -		if (!(ccr & CCR_ENABLE))
>> -			ret = DMA_COMPLETE;
>> -	}
>> -
>> +	spin_lock_irqsave(&c->vc.lock, flags);
>>  	if (ret == DMA_COMPLETE || !txstate)
>> -		return ret;
>> +		goto out;
>>  
>> -	spin_lock_irqsave(&c->vc.lock, flags);
>>  	vd = vchan_find_desc(&c->vc, cookie);
>>  	if (vd) {
>>  		txstate->residue = omap_dma_desc_size(to_omap_dma_desc(&vd->tx));
>> @@ -868,6 +859,23 @@ static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
>>  	}
>>  	if (ret == DMA_IN_PROGRESS && c->paused)
>>  		ret = DMA_PAUSED;
>> +
>> +out:
>> +	if (ret == DMA_IN_PROGRESS && c->running && c->desc &&
>> +	    c->desc->polled && c->desc->vd.tx.cookie == cookie) {
> 
> heh, that makes quite a read!
> 
> checking DMA_IN_PROGRESS should not make sense, as we should bail out at
> the start if it is completed
> 
> I think other can be optimzed to make it a better read!

True, a simple re-ordering should make it easier to read or to split
them up as two if() checks. I'll try to figure out something to simplify it.

> 
>> +		uint32_t ccr = omap_dma_chan_read(c, CCR);
>> +		/*
>> +		 * The channel is no longer active, set the return value
>> +		 * accordingly
>> +		 */
>> +		if (!(ccr & CCR_ENABLE)) {
>> +			struct omap_desc *d = c->desc;
>> +			ret = DMA_COMPLETE;
>> +			omap_dma_start_desc(c);
>> +			vchan_cookie_complete(&d->vd);
>> +		}
>> +	}
>> +
>>  	spin_unlock_irqrestore(&c->vc.lock, flags);
>>  
>>  	return ret;
>> @@ -1233,7 +1241,10 @@ static struct dma_async_tx_descriptor *omap_dma_prep_dma_memcpy(
>>  	d->ccr = c->ccr;
>>  	d->ccr |= CCR_DST_AMODE_POSTINC | CCR_SRC_AMODE_POSTINC;
>>  
>> -	d->cicr = CICR_DROP_IE | CICR_FRAME_IE;
>> +	if (tx_flags & DMA_PREP_INTERRUPT)
>> +		d->cicr |= CICR_FRAME_IE;
>> +	else
>> +		d->polled = true;
>>  
>>  	d->csdp = data_type;
>>  
>> -- 
>> Peter
>>
>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
>> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>
diff mbox series

Patch

diff --git a/drivers/dma/ti/omap-dma.c b/drivers/dma/ti/omap-dma.c
index 5ba7d8485026..75d8f7e13c8d 100644
--- a/drivers/dma/ti/omap-dma.c
+++ b/drivers/dma/ti/omap-dma.c
@@ -94,6 +94,7 @@  struct omap_desc {
 	bool using_ll;
 	enum dma_transfer_direction dir;
 	dma_addr_t dev_addr;
+	bool polled;
 
 	int32_t fi;		/* for OMAP_DMA_SYNC_PACKET / double indexing */
 	int16_t ei;		/* for double indexing */
@@ -834,20 +835,10 @@  static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
 
 	ret = dma_cookie_status(chan, cookie, txstate);
 
-	if (!c->paused && c->running) {
-		uint32_t ccr = omap_dma_chan_read(c, CCR);
-		/*
-		 * The channel is no longer active, set the return value
-		 * accordingly
-		 */
-		if (!(ccr & CCR_ENABLE))
-			ret = DMA_COMPLETE;
-	}
-
+	spin_lock_irqsave(&c->vc.lock, flags);
 	if (ret == DMA_COMPLETE || !txstate)
-		return ret;
+		goto out;
 
-	spin_lock_irqsave(&c->vc.lock, flags);
 	vd = vchan_find_desc(&c->vc, cookie);
 	if (vd) {
 		txstate->residue = omap_dma_desc_size(to_omap_dma_desc(&vd->tx));
@@ -868,6 +859,23 @@  static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
 	}
 	if (ret == DMA_IN_PROGRESS && c->paused)
 		ret = DMA_PAUSED;
+
+out:
+	if (ret == DMA_IN_PROGRESS && c->running && c->desc &&
+	    c->desc->polled && c->desc->vd.tx.cookie == cookie) {
+		uint32_t ccr = omap_dma_chan_read(c, CCR);
+		/*
+		 * The channel is no longer active, set the return value
+		 * accordingly
+		 */
+		if (!(ccr & CCR_ENABLE)) {
+			struct omap_desc *d = c->desc;
+			ret = DMA_COMPLETE;
+			omap_dma_start_desc(c);
+			vchan_cookie_complete(&d->vd);
+		}
+	}
+
 	spin_unlock_irqrestore(&c->vc.lock, flags);
 
 	return ret;
@@ -1233,7 +1241,10 @@  static struct dma_async_tx_descriptor *omap_dma_prep_dma_memcpy(
 	d->ccr = c->ccr;
 	d->ccr |= CCR_DST_AMODE_POSTINC | CCR_SRC_AMODE_POSTINC;
 
-	d->cicr = CICR_DROP_IE | CICR_FRAME_IE;
+	if (tx_flags & DMA_PREP_INTERRUPT)
+		d->cicr |= CICR_FRAME_IE;
+	else
+		d->polled = true;
 
 	d->csdp = data_type;