Skip to contentSkip to conversation

dma: various minor clean ups for slave drivers

PATCH In review Andy Shevchenko 6 participants 24 messages May 27, 2013 – Jul 15, 2013

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

cover dma: various minor clean ups for slave drivers

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

Andy Shevchenko (12):
imx-sdma: remove useless variable
mxs-dma: remove useless variable
edma: no need to assign residue to 0 explicitly
ep93xx_dma: remove useless use of lock
fsldma: remove useless use of lock
mmp_pdma: remove useless use of lock
mpc512x_dma: remove useless use of lock
pch_dma: remove useless use of lock
tegra20-apb-dma: remove useless use of lock
ipu_idmac: re-use dma_cookie_status()
mmp_tdma: set cookies as well when asked for tx status
txx9dmac: return DMA_SUCCESS immediately from device_tx_status()

 drivers/dma/edma.c            |  2 --
 drivers/dma/ep93xx_dma.c      | 10 +---------
 drivers/dma/fsldma.c          | 10 +---------
 drivers/dma/imx-sdma.c        |  9 +++------
 drivers/dma/ipu/ipu_idmac.c   |  5 +----
 drivers/dma/mmp_pdma.c        | 10 +---------
 drivers/dma/mmp_tdma.c        |  3 ++-
 drivers/dma/mpc512x_dma.c     | 10 +---------
 drivers/dma/mxs-dma.c         |  4 +---
 drivers/dma/pch_dma.c         |  9 +--------
 drivers/dma/tegra20-apb-dma.c |  8 +++-----
 drivers/dma/txx9dmac.c        | 13 ++++++-------
 12 files changed, 21 insertions(+), 72 deletions(-)
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#1 lore raw

PATCH 1/12 imx-sdma: remove useless variable

last_used variable is applied only once, so, let's substitute it by its value.

---
 drivers/dma/imx-sdma.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
drivers/dma/imx-sdma.c +3 -6
index 092867b..285e10c 100644
@@ -1119,15 +1119,12 @@ static int sdma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
 }
 
 static enum dma_status sdma_tx_status(struct dma_chan *chan,
-					    dma_cookie_t cookie,
-					    struct dma_tx_state *txstate)
+				      dma_cookie_t cookie,
+				      struct dma_tx_state *txstate)
 {
 	struct sdma_channel *sdmac = to_sdma_chan(chan);
-	dma_cookie_t last_used;
 
-	last_used = chan->cookie;
-
-	dma_set_tx_state(txstate, chan->completed_cookie, last_used,
+	dma_set_tx_state(txstate, chan->completed_cookie, chan->cookie,
 			sdmac->chn_count - sdmac->chn_real_count);
 
 	return sdmac->status;
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#2 lore raw

PATCH 2/12 mxs-dma: remove useless variable

In reply to Andy Shevchenko

last_used variable is applied only once, so, let's substitute it by its value.

---
 drivers/dma/mxs-dma.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/dma/mxs-dma.c +1 -3
index b48a79c..b3a6379 100644
@@ -622,10 +622,8 @@ static enum dma_status mxs_dma_tx_status(struct dma_chan *chan,
 			dma_cookie_t cookie, struct dma_tx_state *txstate)
 {
 	struct mxs_dma_chan *mxs_chan = to_mxs_dma_chan(chan);
-	dma_cookie_t last_used;
 
-	last_used = chan->cookie;
-	dma_set_tx_state(txstate, chan->completed_cookie, last_used, 0);
+	dma_set_tx_state(txstate, chan->completed_cookie, chan->cookie, 0);
 
 	return mxs_chan->status;
 }
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#3 lore raw

PATCH 3/12 edma: no need to assign residue to 0 explicitly

In reply to Andy Shevchenko

Residue value is assigned to 0 by dma_cookie_status().

---
 drivers/dma/edma.c | 2 --
 1 file changed, 2 deletions(-)
drivers/dma/edma.c +0 -2
index cd7e328..8e11499 100644
@@ -502,8 +502,6 @@ static enum dma_status edma_tx_status(struct dma_chan *chan,
 	} else if (echan->edesc && echan->edesc->vdesc.tx.cookie == cookie) {
 		struct edma_desc *edesc = echan->edesc;
 		txstate->residue = edma_desc_size(edesc);
-	} else {
-		txstate->residue = 0;
 	}
 	spin_unlock_irqrestore(&echan->vchan.lock, flags);
 
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#4 lore raw

PATCH 4/12 ep93xx_dma: remove useless use of lock

In reply to Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

---
 drivers/dma/ep93xx_dma.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
drivers/dma/ep93xx_dma.c +1 -9
index f2bf8c0..591cd8c 100644
@@ -1313,15 +1313,7 @@ static enum dma_status ep93xx_dma_tx_status(struct dma_chan *chan,
 					    dma_cookie_t cookie,
 					    struct dma_tx_state *state)
 {
-	struct ep93xx_dma_chan *edmac = to_ep93xx_dma_chan(chan);
-	enum dma_status ret;
-	unsigned long flags;
-
-	spin_lock_irqsave(&edmac->lock, flags);
-	ret = dma_cookie_status(chan, cookie, state);
-	spin_unlock_irqrestore(&edmac->lock, flags);
-
-	return ret;
+	return dma_cookie_status(chan, cookie, state);
 }
 
 /**
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#5 lore raw

PATCH 5/12 fsldma: remove useless use of lock

In reply to Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

---
 drivers/dma/fsldma.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
drivers/dma/fsldma.c +1 -9
index 49e8fbd..b3f3e90 100644
@@ -979,15 +979,7 @@ static enum dma_status fsl_tx_status(struct dma_chan *dchan,
 					dma_cookie_t cookie,
 					struct dma_tx_state *txstate)
 {
-	struct fsldma_chan *chan = to_fsl_chan(dchan);
-	enum dma_status ret;
-	unsigned long flags;
-
-	spin_lock_irqsave(&chan->desc_lock, flags);
-	ret = dma_cookie_status(dchan, cookie, txstate);
-	spin_unlock_irqrestore(&chan->desc_lock, flags);
-
-	return ret;
+	return dma_cookie_status(dchan, cookie, txstate);
 }
 
 /*----------------------------------------------------------------------------*/
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#6 lore raw

PATCH 6/12 mmp_pdma: remove useless use of lock

In reply to Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

---
 drivers/dma/mmp_pdma.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
drivers/dma/mmp_pdma.c +1 -9
index c26699f..ad4b0a7 100644
@@ -632,15 +632,7 @@ static int mmp_pdma_control(struct dma_chan *dchan, enum dma_ctrl_cmd cmd,
 static enum dma_status mmp_pdma_tx_status(struct dma_chan *dchan,
 			dma_cookie_t cookie, struct dma_tx_state *txstate)
 {
-	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
-	enum dma_status ret;
-	unsigned long flags;
-
-	spin_lock_irqsave(&chan->desc_lock, flags);
-	ret = dma_cookie_status(dchan, cookie, txstate);
-	spin_unlock_irqrestore(&chan->desc_lock, flags);
-
-	return ret;
+	return dma_cookie_status(dchan, cookie, txstate);
 }
 
 /**
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#7 lore raw

PATCH 7/12 mpc512x_dma: remove useless use of lock

In reply to Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

---
 drivers/dma/mpc512x_dma.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
drivers/dma/mpc512x_dma.c +1 -9
index 2d95673..2fe4353 100644
@@ -556,15 +556,7 @@ static enum dma_status
 mpc_dma_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
 	       struct dma_tx_state *txstate)
 {
-	struct mpc_dma_chan *mchan = dma_chan_to_mpc_dma_chan(chan);
-	enum dma_status ret;
-	unsigned long flags;
-
-	spin_lock_irqsave(&mchan->lock, flags);
-	ret = dma_cookie_status(chan, cookie, txstate);
-	spin_unlock_irqrestore(&mchan->lock, flags);
-
-	return ret;
+	return dma_cookie_status(chan, cookie, txstate);
 }
 
 /* Prepare descriptor for memory to memory copy */
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#8 lore raw

PATCH 8/12 pch_dma: remove useless use of lock

In reply to Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

---
 drivers/dma/pch_dma.c | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)
drivers/dma/pch_dma.c +1 -8
index ce3dc3e..237f1d5 100644
@@ -564,14 +564,7 @@ static void pd_free_chan_resources(struct dma_chan *chan)
 static enum dma_status pd_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
 				    struct dma_tx_state *txstate)
 {
-	struct pch_dma_chan *pd_chan = to_pd_chan(chan);
-	enum dma_status ret;
-
-	spin_lock_irq(&pd_chan->lock);
-	ret = dma_cookie_status(chan, cookie, txstate);
-	spin_unlock_irq(&pd_chan->lock);
-
-	return ret;
+	return dma_cookie_status(chan, cookie, txstate);
 }
 
 static void pd_issue_pending(struct dma_chan *chan)
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#9 lore raw

PATCH 9/12 tegra20-apb-dma: remove useless use of lock

In reply to Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

---
 drivers/dma/tegra20-apb-dma.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
drivers/dma/tegra20-apb-dma.c +3 -5
index 33f59ec..019ccfa 100644
@@ -767,13 +767,11 @@ static enum dma_status tegra_dma_tx_status(struct dma_chan *dc,
 	unsigned long flags;
 	unsigned int residual;
 
-	spin_lock_irqsave(&tdc->lock, flags);
-
 	ret = dma_cookie_status(dc, cookie, txstate);
-	if (ret == DMA_SUCCESS) {
-		spin_unlock_irqrestore(&tdc->lock, flags);
+	if (ret == DMA_SUCCESS)
 		return ret;
-	}
+
+	spin_lock_irqsave(&tdc->lock, flags);
 
 	/* Check on wait_ack desc status */
 	list_for_each_entry(dma_desc, &tdc->free_dma_desc, node) {
Signature · 1 line
1.8.2.rc0.22.gb3600c3
maintainer
#10 lore raw

Re: [PATCH 09/12] tegra20-apb-dma: remove useless use of lock

Quoted 2 lines · Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

I think we need lock here:
From isr handler, we call dma_cookie_complete() which is in spin-locked. This function updates tx->chan->completed_cookie = tx->cookie; In tegra_dma_tx_status(), we check for dma_cookie_status() which access the chan->completed_cookie; and it decides status based on this

As the access of chan->completed_cookie are from different context, we need this locking.

But did not get why it is documented as locking is not require if shared variable is getting changed/access from different context simultaneously.

#11 lore raw

Re: [PATCH 09/12] tegra20-apb-dma: remove useless use of lock

Quoted 4 lines · Laxman Dewangan
Quoted 3 lines · Andy Shevchenko

Accordingly to dma_cookie_status() description locking is not required.

Quoted 8 lines

I think we need lock here:
From isr handler, we call dma_cookie_complete() which is in spin-locked. This function updates tx->chan->completed_cookie = tx->cookie; In tegra_dma_tx_status(), we check for dma_cookie_status() which access the chan->completed_cookie; and it decides status based on this

As the access of chan->completed_cookie are from different context, we need this locking.

You need to have a consistent data in the cookies. This is guaranteed by memory barrier if I got it correctly.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#12 lore raw

PATCH 10/12 ipu_idmac: re-use dma_cookie_status()

In reply to Andy Shevchenko

It's better to use generic dma_cookie_status() that allows user to get standard possible return codes independently of the DMAC driver in charge.

---
 drivers/dma/ipu/ipu_idmac.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/dma/ipu/ipu_idmac.c +1 -4
index d39c2cd..608d4a2 100644
@@ -1593,10 +1593,7 @@ static void idmac_free_chan_resources(struct dma_chan *chan)
 static enum dma_status idmac_tx_status(struct dma_chan *chan,
 		       dma_cookie_t cookie, struct dma_tx_state *txstate)
 {
-	dma_set_tx_state(txstate, chan->completed_cookie, chan->cookie, 0);
-	if (cookie != chan->cookie)
-		return DMA_ERROR;
-	return DMA_SUCCESS;
+	return dma_cookie_status(chan, cookie, txstate);
 }
 
 static int __init ipu_idmac_init(struct ipu *ipu)
Signature · 1 line
1.8.2.rc0.22.gb3600c3
Acked-by on v1 10/12
#13 lore raw

Re: [PATCH 10/12] ipu_idmac: re-use dma_cookie_status()

Quoted 5 lines · Andy Shevchenko

It's better to use generic dma_cookie_status() that allows user to get standard possible return codes independently of the DMAC driver in charge.

Quoted 23 lines
---
 drivers/dma/ipu/ipu_idmac.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/dma/ipu/ipu_idmac.c +1 -4
index d39c2cd..608d4a2 100644
@@ -1593,10 +1593,7 @@ static void idmac_free_chan_resources(struct dma_chan *chan)
 static enum dma_status idmac_tx_status(struct dma_chan *chan,
 		       dma_cookie_t cookie, struct dma_tx_state *txstate)
 {
-	dma_set_tx_state(txstate, chan->completed_cookie, chan->cookie, 0);
-	if (cookie != chan->cookie)
-		return DMA_ERROR;
-	return DMA_SUCCESS;
+	return dma_cookie_status(chan, cookie, txstate);
 }
 
 static int __init ipu_idmac_init(struct ipu *ipu)
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#14 lore raw

PATCH 11/12 mmp_tdma: set cookies as well when asked for tx status

In reply to Andy Shevchenko

dma_set_residue() sets only residue value, so user can't rely on the returned values of cookies. That patch standardize the behaviour.

---
 drivers/dma/mmp_tdma.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/dma/mmp_tdma.c +2 -1
index 43d5a6c..4472915 100644
@@ -456,7 +456,8 @@ static enum dma_status mmp_tdma_tx_status(struct dma_chan *chan,
 {
 	struct mmp_tdma_chan *tdmac = to_mmp_tdma_chan(chan);
 
-	dma_set_residue(txstate, tdmac->buf_len - tdmac->pos);
+	dma_set_tx_state(txstate, chan->completed_cookie, chan->cookie,
+			 tdmac->buf_len - tdmac->pos);
 
 	return tdmac->status;
 }
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#15 lore raw

PATCH 12/12 txx9dmac: return DMA_SUCCESS immediately from device_tx_status()

In reply to Andy Shevchenko

There is no point to go throught the rest of the function if first call to dma_cookie_status() returned DMA_SUCCESS.

---
 drivers/dma/txx9dmac.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
drivers/dma/txx9dmac.c +6 -7
index a59fb48..59357db 100644
@@ -962,15 +962,14 @@ txx9dmac_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
 	enum dma_status ret;
 
 	ret = dma_cookie_status(chan, cookie, txstate);
-	if (ret != DMA_SUCCESS) {
-		spin_lock_bh(&dc->lock);
-		txx9dmac_scan_descriptors(dc);
-		spin_unlock_bh(&dc->lock);
+	if (ret == DMA_SUCCESS)
+		return DMA_SUCCESS;
 
-		ret = dma_cookie_status(chan, cookie, txstate);
-	}
+	spin_lock_bh(&dc->lock);
+	txx9dmac_scan_descriptors(dc);
+	spin_unlock_bh(&dc->lock);
 
-	return ret;
+	return dma_cookie_status(chan, cookie, txstate);
 }
 
 static void txx9dmac_chain_dynamic(struct txx9dmac_chan *dc,
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#16 lore raw
In reply to Andy Shevchenko
Quoted 2 lines · Andy Shevchenko

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

The series looks fine. I am going to wait a day more and apply, pls speak up if you disagree and ack if you agree

--
~Vinod

Quoted 32 lines

Andy Shevchenko (12):
imx-sdma: remove useless variable
mxs-dma: remove useless variable
edma: no need to assign residue to 0 explicitly
ep93xx_dma: remove useless use of lock
fsldma: remove useless use of lock
mmp_pdma: remove useless use of lock
mpc512x_dma: remove useless use of lock
pch_dma: remove useless use of lock
tegra20-apb-dma: remove useless use of lock
ipu_idmac: re-use dma_cookie_status()
mmp_tdma: set cookies as well when asked for tx status
txx9dmac: return DMA_SUCCESS immediately from device_tx_status()

 drivers/dma/edma.c            |  2 --
 drivers/dma/ep93xx_dma.c      | 10 +---------
 drivers/dma/fsldma.c          | 10 +---------
 drivers/dma/imx-sdma.c        |  9 +++------
 drivers/dma/ipu/ipu_idmac.c   |  5 +----
 drivers/dma/mmp_pdma.c        | 10 +---------
 drivers/dma/mmp_tdma.c        |  3 ++-
 drivers/dma/mpc512x_dma.c     | 10 +---------
 drivers/dma/mxs-dma.c         |  4 +---
 drivers/dma/pch_dma.c         |  9 +--------
 drivers/dma/tegra20-apb-dma.c |  8 +++-----
 drivers/dma/txx9dmac.c        | 13 ++++++-------
 12 files changed, 21 insertions(+), 72 deletions(-)
Signature · 1 line
1.8.2.rc0.22.gb3600c3
#17 lore raw
Quoted 5 lines · Vinod Koul
Quoted 2 lines · Andy Shevchenko

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

The series looks fine. I am going to wait a day more and apply, pls speak up if you disagree and ack if you agree

I'm not in hurry with it. Please, take your time and do whatever it requires. Thank you!

--
With Best Regards,
Andy Shevchenko

#18 lore raw
Quoted 2 lines · Andy Shevchenko
Quoted 2 lines
Quoted 2 lines

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

Applied thanks

--
~Vinod

#19 lore raw
Quoted 5 lines · Vinod Koul
Quoted 2 lines · Andy Shevchenko
Quoted 2 lines
Quoted 2 lines

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

Applied thanks

Thank you. You were faster than me, I was just about to send rebased version.

Signature · 2 lines
Andy Shevchenko <andriy.shevchenko-VuQAYsv1563Yd54FQh9/[email protected]>
Intel Finland Oy
#20 lore raw
Quoted 9 lines · Andy Shevchenko
Quoted 5 lines · Vinod Koul
Quoted 2 lines · Andy Shevchenko
Quoted 2 lines
Quoted 2 lines

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

Applied thanks

Thank you. You were faster than me, I was just about to send rebased version.

:)

I suspected changes are trivial and it should apply or with slight modfications. it did apply cleanly so no reason to delay applying on shiny new -rc1

For folks in this part of world, monday morning Linus drops the rc1 usually, so get staright to it!

--
~Vinod

#21 lore raw
In reply to Vinod Koul
Quoted 5 lines · Vinod Koul
Quoted 2 lines · Andy Shevchenko

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

The series looks fine. I am going to wait a day more and apply, pls speak up if you disagree and ack if you agree

Looks ok to me. Reminds we can probably take this one step further and provide a generic implementation for the common case. It's just a bit inconsistent though that some engines will poll the completion handler (try to advance the state of the last completed cookie) whereas others just assume things will be completed asynchronously.

--
Dan

#22 lore raw
Quoted 4 lines · Dan Williams
Quoted 3 lines · Vinod Koul
Quoted 2 lines · Andy Shevchenko

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

Quoted 8 lines
Quoted 2 lines

The series looks fine. I am going to wait a day more and apply, pls speak up if you disagree and ack if you agree

Looks ok to me. Reminds we can probably take this one step further and provide a generic implementation for the common case. It's just a bit inconsistent though that some engines will poll the completion handler (try to advance the state of the last completed cookie) whereas others just assume things will be completed asynchronously.

agree with that. These are the inconsistencies some of which are based on hardware and some are user iterpretations...

--
~Vinod

#23 lore raw
In reply to Vinod Koul
Quoted 5 lines · Vinod Koul
Quoted 2 lines · Andy Shevchenko

Here is a set of small independent patches that clean up or fix minor things across DMA slave drivers.

The series looks fine. I am going to wait a day more and apply, pls speak up if you disagree and ack if you agree

Kindly remind about this series.

Quoted 36 lines

--
~Vinod

Quoted 32 lines

Andy Shevchenko (12):
imx-sdma: remove useless variable
mxs-dma: remove useless variable
edma: no need to assign residue to 0 explicitly
ep93xx_dma: remove useless use of lock
fsldma: remove useless use of lock
mmp_pdma: remove useless use of lock
mpc512x_dma: remove useless use of lock
pch_dma: remove useless use of lock
tegra20-apb-dma: remove useless use of lock
ipu_idmac: re-use dma_cookie_status()
mmp_tdma: set cookies as well when asked for tx status
txx9dmac: return DMA_SUCCESS immediately from device_tx_status()

 drivers/dma/edma.c            |  2 --
 drivers/dma/ep93xx_dma.c      | 10 +---------
 drivers/dma/fsldma.c          | 10 +---------
 drivers/dma/imx-sdma.c        |  9 +++------
 drivers/dma/ipu/ipu_idmac.c   |  5 +----
 drivers/dma/mmp_pdma.c        | 10 +---------
 drivers/dma/mmp_tdma.c        |  3 ++-
 drivers/dma/mpc512x_dma.c     | 10 +---------
 drivers/dma/mxs-dma.c         |  4 +---
 drivers/dma/pch_dma.c         |  9 +--------
 drivers/dma/tegra20-apb-dma.c |  8 +++-----
 drivers/dma/txx9dmac.c        | 13 ++++++-------
 12 files changed, 21 insertions(+), 72 deletions(-)
Signature · 1 line
1.8.2.rc0.22.gb3600c3
Signature · 2 lines
Andy Shevchenko <andriy.shevchenko-VuQAYsv1563Yd54FQh9/[email protected]>
Intel Finland Oy