[PATCH v3 0/2] dma: pl330: improve status checking

[PATCH v3 0/2] dma: pl330: improve status checking

[Robert Baldyga]

Hi,

This series allows to check DMA transfer residue (number of bytes left
to send/receive) by modifying pl330_tx_status() function, when struct
dma_tx_state is filled with needed data. It also introduces DMA_PAUSE
feature, which allows to halt DMA transfer before termination and read
residue without risk of data loss.

This features are needed for proper implementation of DMA transfers,
particulary for serial drivers when transfer sizes are unknown and
requests on DMA channels are terminated before transfer completion
very ofter (it's becouse we terminate then in timeout interrupt to
avoid latency which is usually undesirable).

Best regards,
Robert Baldyga

Changelog:

v3:
- remove double pm_runtime_put()

v2: https://lkml.org/lkml/2014/12/5/94
- add pm_runtime support
- make it working for multi-segment transfers
- some cleanups

v1: https://lkml.org/lkml/2014/11/25/554

Robert Baldyga (2):
  dma: pl330: improve pl330_tx_status() function
  dma: pl330: add DMA_PAUSE feature

 drivers/dma/pl330.c | 87 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 85 insertions(+), 2 deletions(-)

-- 
1.9.1

[PATCH v3 1/2] dma: pl330: improve pl330_tx_status() function

[Robert Baldyga]

This patch adds possibility to read residue of DMA transfer. It's useful
when we want to know how many bytes have been transferred before we
terminate channel. It can take place, for example, on timeout interrupt.

Signed-off-by: Lukasz Czerwinski <l.czerwinski@samsung.com>
Signed-off-by: Robert Baldyga <r.baldyga@samsung.com>
---
 drivers/dma/pl330.c | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 66 insertions(+), 2 deletions(-)

diff --git a/drivers/dma/pl330.c b/drivers/dma/pl330.c
index bdf40b5..2f4d561 100644
--- a/drivers/dma/pl330.c
+++ b/drivers/dma/pl330.c
@@ -504,6 +504,9 @@ struct dma_pl330_desc {
 
 	enum desc_status status;
 
+	int bytes_requested;
+	bool last;
+
 	/* The channel which currently holds this desc */
 	struct dma_pl330_chan *pchan;
 
@@ -2182,11 +2185,68 @@ static void pl330_free_chan_resources(struct dma_chan *chan)
 	pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
 }
 
+int pl330_get_current_xferred_count(struct dma_pl330_chan *pch,
+		struct dma_pl330_desc *desc)
+{
+	struct pl330_thread *thrd = pch->thread;
+	struct pl330_dmac *pl330 = pch->dmac;
+	void __iomem *regs = thrd->dmac->base;
+	u32 val, addr;
+
+	pm_runtime_get_sync(pl330->ddma.dev);
+	val = addr = 0;
+	if (desc->rqcfg.src_inc) {
+		val = readl(regs + SA(thrd->id));
+		addr = desc->px.src_addr;
+	} else {
+		val = readl(regs + DA(thrd->id));
+		addr = desc->px.dst_addr;
+	}
+	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
+	pm_runtime_put_autosuspend(pl330->ddma.dev);
+	return val - addr;
+}
+
 static enum dma_status
 pl330_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
 		 struct dma_tx_state *txstate)
 {
-	return dma_cookie_status(chan, cookie, txstate);
+	enum dma_status ret;
+	unsigned long flags;
+	struct dma_pl330_desc *desc, *running = NULL;
+	struct dma_pl330_chan *pch = to_pchan(chan);
+	unsigned int transferred, residual = 0;
+
+	spin_lock_irqsave(&pch->lock, flags);
+
+	if (pch->thread->req_running != -1)
+		running = pch->thread->req[pch->thread->req_running].desc;
+
+	/* Check in pending list */
+	list_for_each_entry(desc, &pch->work_list, node) {
+		if (desc->status == DONE)
+			transferred = desc->bytes_requested;
+		else if (running && desc == running)
+			transferred =
+				pl330_get_current_xferred_count(pch, desc);
+		else
+			transferred = 0;
+		residual += desc->bytes_requested - transferred;
+		if (desc->txd.cookie == cookie) {
+			dma_set_residue(txstate, residual);
+			ret = desc->status;
+			spin_unlock_irqrestore(&pch->lock, flags);
+			return ret;
+		}
+		if (desc->last)
+			residual = 0;
+	}
+	spin_unlock_irqrestore(&pch->lock, flags);
+
+	ret = dma_cookie_status(chan, cookie, txstate);
+	dma_set_residue(txstate, 0);
+
+	return ret;
 }
 
 static void pl330_issue_pending(struct dma_chan *chan)
@@ -2231,12 +2291,14 @@ static dma_cookie_t pl330_tx_submit(struct dma_async_tx_descriptor *tx)
 			desc->txd.callback = last->txd.callback;
 			desc->txd.callback_param = last->txd.callback_param;
 		}
+		last->last = false;
 
 		dma_cookie_assign(&desc->txd);
 
 		list_move_tail(&desc->node, &pch->submitted_list);
 	}
 
+	last->last = true;
 	cookie = dma_cookie_assign(&last->txd);
 	list_add_tail(&last->node, &pch->submitted_list);
 	spin_unlock_irqrestore(&pch->lock, flags);
@@ -2459,6 +2521,7 @@ static struct dma_async_tx_descriptor *pl330_prep_dma_cyclic(
 		desc->rqtype = direction;
 		desc->rqcfg.brst_size = pch->burst_sz;
 		desc->rqcfg.brst_len = 1;
+		desc->bytes_requested = period_len;
 		fill_px(&desc->px, dst, src, period_len);
 
 		if (!first)
@@ -2601,6 +2664,7 @@ pl330_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		desc->rqcfg.brst_size = pch->burst_sz;
 		desc->rqcfg.brst_len = 1;
 		desc->rqtype = direction;
+		desc->bytes_requested = sg_dma_len(sg);
 	}
 
 	/* Return the last desc in the chain */
@@ -2631,7 +2695,7 @@ static int pl330_dma_device_slave_caps(struct dma_chan *dchan,
 	caps->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
 	caps->cmd_pause = false;
 	caps->cmd_terminate = true;
-	caps->residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
+	caps->residue_granularity = DMA_RESIDUE_GRANULARITY_SEGMENT;
 
 	return 0;
 }
-- 
1.9.1

Re: [PATCH v3 1/2] dma: pl330: improve pl330_tx_status() function

[Vinod Koul]

On Wed, Dec 10, 2014 at 11:55:17AM +0100, Robert Baldyga wrote:
> This patch adds possibility to read residue of DMA transfer. It's useful
> when we want to know how many bytes have been transferred before we
> terminate channel. It can take place, for example, on timeout interrupt.
> 
> Signed-off-by: Lukasz Czerwinski <l.czerwinski@samsung.com>
> Signed-off-by: Robert Baldyga <r.baldyga@samsung.com>
> ---
>  drivers/dma/pl330.c | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 66 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/dma/pl330.c b/drivers/dma/pl330.c
> index bdf40b5..2f4d561 100644
> --- a/drivers/dma/pl330.c
> +++ b/drivers/dma/pl330.c
> @@ -504,6 +504,9 @@ struct dma_pl330_desc {
>  
>  	enum desc_status status;
>  
> +	int bytes_requested;
> +	bool last;
> +
>  	/* The channel which currently holds this desc */
>  	struct dma_pl330_chan *pchan;
>  
> @@ -2182,11 +2185,68 @@ static void pl330_free_chan_resources(struct dma_chan *chan)
>  	pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
>  }
>  
> +int pl330_get_current_xferred_count(struct dma_pl330_chan *pch,
> +		struct dma_pl330_desc *desc)
> +{
> +	struct pl330_thread *thrd = pch->thread;
> +	struct pl330_dmac *pl330 = pch->dmac;
> +	void __iomem *regs = thrd->dmac->base;
> +	u32 val, addr;
> +
> +	pm_runtime_get_sync(pl330->ddma.dev);
> +	val = addr = 0;
> +	if (desc->rqcfg.src_inc) {
> +		val = readl(regs + SA(thrd->id));
> +		addr = desc->px.src_addr;
> +	} else {
> +		val = readl(regs + DA(thrd->id));
> +		addr = desc->px.dst_addr;
> +	}
> +	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
> +	pm_runtime_put_autosuspend(pl330->ddma.dev);
> +	return val - addr;
> +}
> +
>  static enum dma_status
>  pl330_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
>  		 struct dma_tx_state *txstate)
>  {
> -	return dma_cookie_status(chan, cookie, txstate);
> +	enum dma_status ret;
> +	unsigned long flags;
> +	struct dma_pl330_desc *desc, *running = NULL;
> +	struct dma_pl330_chan *pch = to_pchan(chan);
> +	unsigned int transferred, residual = 0;
> +
> +	spin_lock_irqsave(&pch->lock, flags);
You want to check the dma_cookie_status here first and then based on status
go into residue calcaultion, that too only when the txstate is NON null.

-- 
~Vinod

> +
> +	if (pch->thread->req_running != -1)
> +		running = pch->thread->req[pch->thread->req_running].desc;
> +
> +	/* Check in pending list */
> +	list_for_each_entry(desc, &pch->work_list, node) {
> +		if (desc->status == DONE)
> +			transferred = desc->bytes_requested;
> +		else if (running && desc == running)
> +			transferred =
> +				pl330_get_current_xferred_count(pch, desc);
> +		else
> +			transferred = 0;
> +		residual += desc->bytes_requested - transferred;
> +		if (desc->txd.cookie == cookie) {
> +			dma_set_residue(txstate, residual);
> +			ret = desc->status;
> +			spin_unlock_irqrestore(&pch->lock, flags);
> +			return ret;
> +		}
> +		if (desc->last)
> +			residual = 0;
> +	}
> +	spin_unlock_irqrestore(&pch->lock, flags);
> +
> +	ret = dma_cookie_status(chan, cookie, txstate);
> +	dma_set_residue(txstate, 0);
> +
> +	return ret;
>  }
>  
>  static void pl330_issue_pending(struct dma_chan *chan)
> @@ -2231,12 +2291,14 @@ static dma_cookie_t pl330_tx_submit(struct dma_async_tx_descriptor *tx)
>  			desc->txd.callback = last->txd.callback;
>  			desc->txd.callback_param = last->txd.callback_param;
>  		}
> +		last->last = false;
>  
>  		dma_cookie_assign(&desc->txd);
>  
>  		list_move_tail(&desc->node, &pch->submitted_list);
>  	}
>  
> +	last->last = true;
>  	cookie = dma_cookie_assign(&last->txd);
>  	list_add_tail(&last->node, &pch->submitted_list);
>  	spin_unlock_irqrestore(&pch->lock, flags);
> @@ -2459,6 +2521,7 @@ static struct dma_async_tx_descriptor *pl330_prep_dma_cyclic(
>  		desc->rqtype = direction;
>  		desc->rqcfg.brst_size = pch->burst_sz;
>  		desc->rqcfg.brst_len = 1;
> +		desc->bytes_requested = period_len;
>  		fill_px(&desc->px, dst, src, period_len);
>  
>  		if (!first)
> @@ -2601,6 +2664,7 @@ pl330_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
>  		desc->rqcfg.brst_size = pch->burst_sz;
>  		desc->rqcfg.brst_len = 1;
>  		desc->rqtype = direction;
> +		desc->bytes_requested = sg_dma_len(sg);
>  	}
>  
>  	/* Return the last desc in the chain */
> @@ -2631,7 +2695,7 @@ static int pl330_dma_device_slave_caps(struct dma_chan *dchan,
>  	caps->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
>  	caps->cmd_pause = false;
>  	caps->cmd_terminate = true;
> -	caps->residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
> +	caps->residue_granularity = DMA_RESIDUE_GRANULARITY_SEGMENT;
>  
>  	return 0;
>  }
> -- 
> 1.9.1
> 

-- 

Re: [PATCH v3 1/2] dma: pl330: improve pl330_tx_status() function

[Robert Baldyga]

On 02/11/2015 01:23 AM, Vinod Koul wrote:
> On Wed, Dec 10, 2014 at 11:55:17AM +0100, Robert Baldyga wrote:
>> This patch adds possibility to read residue of DMA transfer. It's useful
>> when we want to know how many bytes have been transferred before we
>> terminate channel. It can take place, for example, on timeout interrupt.
>>
>> Signed-off-by: Lukasz Czerwinski <l.czerwinski@samsung.com>
>> Signed-off-by: Robert Baldyga <r.baldyga@samsung.com>
>> ---
>>  drivers/dma/pl330.c | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++--
>>  1 file changed, 66 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/dma/pl330.c b/drivers/dma/pl330.c
>> index bdf40b5..2f4d561 100644
>> --- a/drivers/dma/pl330.c
>> +++ b/drivers/dma/pl330.c
>> @@ -504,6 +504,9 @@ struct dma_pl330_desc {
>>  
>>  	enum desc_status status;
>>  
>> +	int bytes_requested;
>> +	bool last;
>> +
>>  	/* The channel which currently holds this desc */
>>  	struct dma_pl330_chan *pchan;
>>  
>> @@ -2182,11 +2185,68 @@ static void pl330_free_chan_resources(struct dma_chan *chan)
>>  	pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
>>  }
>>  
>> +int pl330_get_current_xferred_count(struct dma_pl330_chan *pch,
>> +		struct dma_pl330_desc *desc)
>> +{
>> +	struct pl330_thread *thrd = pch->thread;
>> +	struct pl330_dmac *pl330 = pch->dmac;
>> +	void __iomem *regs = thrd->dmac->base;
>> +	u32 val, addr;
>> +
>> +	pm_runtime_get_sync(pl330->ddma.dev);
>> +	val = addr = 0;
>> +	if (desc->rqcfg.src_inc) {
>> +		val = readl(regs + SA(thrd->id));
>> +		addr = desc->px.src_addr;
>> +	} else {
>> +		val = readl(regs + DA(thrd->id));
>> +		addr = desc->px.dst_addr;
>> +	}
>> +	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
>> +	pm_runtime_put_autosuspend(pl330->ddma.dev);
>> +	return val - addr;
>> +}
>> +
>>  static enum dma_status
>>  pl330_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
>>  		 struct dma_tx_state *txstate)
>>  {
>> -	return dma_cookie_status(chan, cookie, txstate);
>> +	enum dma_status ret;
>> +	unsigned long flags;
>> +	struct dma_pl330_desc *desc, *running = NULL;
>> +	struct dma_pl330_chan *pch = to_pchan(chan);
>> +	unsigned int transferred, residual = 0;
>> +
>> +	spin_lock_irqsave(&pch->lock, flags);
> You want to check the dma_cookie_status here first and then based on status
> go into residue calcaultion, that too only when the txstate is NON null.
> 

Ok, I will send v4.

Thanks,
Robert Baldyga

[PATCH v3 2/2] dma: pl330: add DMA_PAUSE feature

[Robert Baldyga]

DMA_PAUSE command is used for halting DMA transfer on chosen channel.
It can be useful when we want to safely read residue before terminating
all requests on channel. Otherwise there can be situation when some data
is transferred before channel termination but after reading residue,
which obviously results with data loss. To avoid this situation we can
pause channel, read residue and then terminate all requests.
This scenario is common, for example, in serial port drivers.

Signed-off-by: Robert Baldyga <r.baldyga@samsung.com>
---
 drivers/dma/pl330.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/drivers/dma/pl330.c b/drivers/dma/pl330.c
index 2f4d561..ead4369 100644
--- a/drivers/dma/pl330.c
+++ b/drivers/dma/pl330.c
@@ -2137,6 +2137,25 @@ static int pl330_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd, unsigned
 		pm_runtime_mark_last_busy(pl330->ddma.dev);
 		pm_runtime_put_autosuspend(pl330->ddma.dev);
 		break;
+	case DMA_PAUSE:
+		/*
+		 * We don't support DMA_RESUME command because of hardware
+		 * limitations, so after pausing the channel we cannot restore
+		 * it to active state. We have to terminate channel and setup
+		 * DMA transfer again. This pause feature was implemented to
+		 * allow safely read residue before channel termination.
+		 */
+		pm_runtime_get_sync(pl330->ddma.dev);
+		spin_lock_irqsave(&pch->lock, flags);
+
+		spin_lock(&pl330->lock);
+		_stop(pch->thread);
+		spin_unlock(&pl330->lock);
+
+		spin_unlock_irqrestore(&pch->lock, flags);
+		pm_runtime_mark_last_busy(pl330->ddma.dev);
+		pm_runtime_put_autosuspend(pl330->ddma.dev);
+		break;
 	case DMA_SLAVE_CONFIG:
 		slave_config = (struct dma_slave_config *)arg;
 
-- 
1.9.1

Re: [PATCH v3 2/2] dma: pl330: add DMA_PAUSE feature

[Vinod Koul]

On Wed, Dec 10, 2014 at 11:55:18AM +0100, Robert Baldyga wrote:
> DMA_PAUSE command is used for halting DMA transfer on chosen channel.
> It can be useful when we want to safely read residue before terminating
> all requests on channel. Otherwise there can be situation when some data
> is transferred before channel termination but after reading residue,
> which obviously results with data loss. To avoid this situation we can
> pause channel, read residue and then terminate all requests.
> This scenario is common, for example, in serial port drivers.
And where is the resume here? Also this needs rebase

-- 
~Vinod

> 
> Signed-off-by: Robert Baldyga <r.baldyga@samsung.com>
> ---
>  drivers/dma/pl330.c | 19 +++++++++++++++++++
>  1 file changed, 19 insertions(+)
> 
> diff --git a/drivers/dma/pl330.c b/drivers/dma/pl330.c
> index 2f4d561..ead4369 100644
> --- a/drivers/dma/pl330.c
> +++ b/drivers/dma/pl330.c
> @@ -2137,6 +2137,25 @@ static int pl330_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd, unsigned
>  		pm_runtime_mark_last_busy(pl330->ddma.dev);
>  		pm_runtime_put_autosuspend(pl330->ddma.dev);
>  		break;
> +	case DMA_PAUSE:
> +		/*
> +		 * We don't support DMA_RESUME command because of hardware
> +		 * limitations, so after pausing the channel we cannot restore
> +		 * it to active state. We have to terminate channel and setup
> +		 * DMA transfer again. This pause feature was implemented to
> +		 * allow safely read residue before channel termination.
> +		 */
> +		pm_runtime_get_sync(pl330->ddma.dev);
> +		spin_lock_irqsave(&pch->lock, flags);
> +
> +		spin_lock(&pl330->lock);
> +		_stop(pch->thread);
> +		spin_unlock(&pl330->lock);
> +
> +		spin_unlock_irqrestore(&pch->lock, flags);
> +		pm_runtime_mark_last_busy(pl330->ddma.dev);
> +		pm_runtime_put_autosuspend(pl330->ddma.dev);
> +		break;
>  	case DMA_SLAVE_CONFIG:
>  		slave_config = (struct dma_slave_config *)arg;
>  
> -- 
> 1.9.1
> 

-- 

Re: [PATCH v3 2/2] dma: pl330: add DMA_PAUSE feature

[Robert Baldyga]

On 02/11/2015 01:24 AM, Vinod Koul wrote:
> On Wed, Dec 10, 2014 at 11:55:18AM +0100, Robert Baldyga wrote:
>> DMA_PAUSE command is used for halting DMA transfer on chosen channel.
>> It can be useful when we want to safely read residue before terminating
>> all requests on channel. Otherwise there can be situation when some data
>> is transferred before channel termination but after reading residue,
>> which obviously results with data loss. To avoid this situation we can
>> pause channel, read residue and then terminate all requests.
>> This scenario is common, for example, in serial port drivers.
> And where is the resume here? Also this needs rebase
> 

We can't have resume on this hardware. It's in comment. DMA_PAUSE
feature is added only to freeze channel state before its termination to
allow safe residue read.

BTW I was almost sure that you have applied these patches to your tree
about two months ago. In mainline kernel there is already samsung serial
driver using DMA, which wouldn't work without these changes. Hence it
would be great to have my patches applied ASAP. I will send v4 today.

Thanks,
Robert Baldyga