Skip to contentSkip to conversation

dma: mv_xor: spinlock related improvements

PATCH New Thomas Petazzoni 5 participants 9 messages Dec 27, 2013 – Mar 7, 2014 dmaengine

lore
Share & tools
lore
#0 lore raw

cover dma: mv_xor: spinlock related improvements

Vinod, Dan,

Here are three patches that make improvements to the spinlock handling of the mv_xor driver. The first patch fixes a lock/release of the channel spinlock in mv_xor_status(), and the next two patches make followup code improvements/refactoring.

Ezequiel, it would be good to have your review on these patches, since I know you've spent some time lately doing mv_xor stuff.

Thanks!

Thomas

Thomas Petazzoni (3):
dma: mv_xor: take channel spinlock in mv_xor_status()
dma: mv_xor: use __mv_xor_slot_cleanup() in
mv_xor_free_chan_resources()
dma: mv_xor: rename __mv_xor_slot_cleanup() to mv_xor_slot_cleanup()

 drivers/dma/mv_xor.c | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)
Signature · 1 line
1.8.3.2
#1 lore raw

PATCH 1/3 dma: mv_xor: take channel spinlock in mv_xor_status()

The mv_xor_status() function accesses the mv_xor_chan structure, but was not taking the corresponding spinlock. This patch fixes this problem.

---
 drivers/dma/mv_xor.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/dma/mv_xor.c +8 -2
index 53fb0c8..526ab27 100644
@@ -701,14 +701,20 @@ static enum dma_status mv_xor_status(struct dma_chan *chan,
 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
 	enum dma_status ret;
 
+	spin_lock_bh(&mv_chan->lock);
+
 	ret = dma_cookie_status(chan, cookie, txstate);
 	if (ret == DMA_COMPLETE) {
 		mv_xor_clean_completed_slots(mv_chan);
+		spin_unlock_bh(&mv_chan->lock);
 		return ret;
 	}
-	mv_xor_slot_cleanup(mv_chan);
+	__mv_xor_slot_cleanup(mv_chan);
 
-	return dma_cookie_status(chan, cookie, txstate);
+	ret = dma_cookie_status(chan, cookie, txstate);
+	spin_unlock_bh(&mv_chan->lock);
+
+	return ret;
 }
 
 static void mv_dump_xor_regs(struct mv_xor_chan *chan)
Signature · 1 line
1.8.3.2
#2 lore raw

PATCH 1/3 dma: mv_xor: take channel spinlock in mv_xor_status()

Quoted 14 lines · Thomas Petazzoni

The mv_xor_status() function accesses the mv_xor_chan structure, but was not taking the corresponding spinlock. This patch fixes this problem.

---
 drivers/dma/mv_xor.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/dma/mv_xor.c +0 -0
index 53fb0c8..526ab27 100644

drivers/dma/mv_xor.c static enum dma_status mv_xor_status(struct dma_chan *chan, · line 705

Quoted 5 lines
drivers/dma/mv_xor.c +2 -0
@@ -701,14 +701,20 @@ static enum dma_status mv_xor_status(struct dma_chan *chan,
 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
 	enum dma_status ret;
 
+	spin_lock_bh(&mv_chan->lock);
+

I don't think it's a proper place for this spinlock since dma_cookie_status is independent to that.

drivers/dma/mv_xor.c static enum dma_status mv_xor_status(struct dma_chan *chan, · line 707

Quoted 2 lines
drivers/dma/mv_xor.c +0 -0
 	ret = dma_cookie_status(chan, cookie, txstate);
 	if (ret == DMA_COMPLETE) {

If you are caring about mv_chan access, better to take it here...

drivers/dma/mv_xor.c static enum dma_status mv_xor_status(struct dma_chan *chan, · line 711

Quoted 4 lines
drivers/dma/mv_xor.c +1 -0
 		mv_xor_clean_completed_slots(mv_chan);
+		spin_unlock_bh(&mv_chan->lock);
 		return ret;
 	}

...and here.

Quoted 11 lines
drivers/dma/mv_xor.c +5 -2
-	mv_xor_slot_cleanup(mv_chan);
+	__mv_xor_slot_cleanup(mv_chan);
 
-	return dma_cookie_status(chan, cookie, txstate);
+	ret = dma_cookie_status(chan, cookie, txstate);
+	spin_unlock_bh(&mv_chan->lock);
+
+	return ret;
 }
 
 static void mv_dump_xor_regs(struct mv_xor_chan *chan)
Signature · 12 lines
Andy Shevchenko <[email protected]>
Intel Finland Oy
---------------------------------------------------------------------
Intel Finland Oy
Registered Address: PL 281, 00181 Helsinki 
Business Identity Code: 0357606 - 4 
Domiciled in Helsinki 

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.
#3 lore raw

PATCH 1/3 dma: mv_xor: take channel spinlock in mv_xor_status()

In reply to Thomas Petazzoni
Quoted 14 lines · Thomas Petazzoni

The mv_xor_status() function accesses the mv_xor_chan structure, but was not taking the corresponding spinlock. This patch fixes this problem.

---
 drivers/dma/mv_xor.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/dma/mv_xor.c +0 -0
index 53fb0c8..526ab27 100644

drivers/dma/mv_xor.c static enum dma_status mv_xor_status(struct dma_chan *chan, · line 708

Quoted 8 lines
drivers/dma/mv_xor.c +2 -0
@@ -701,14 +701,20 @@ static enum dma_status mv_xor_status(struct dma_chan *chan,
        struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
        enum dma_status ret;
 
+       spin_lock_bh(&mv_chan->lock);
+
        ret = dma_cookie_status(chan, cookie, txstate);
        if (ret == DMA_COMPLETE) {
                mv_xor_clean_completed_slots(mv_chan);

I think you can just delete this call to mv_xor_clean_completed_slots(). The fact that the descriptors are complete means that __mv_xor_slot_cleanup ran, and if that is the case there should be nothing to cleanup.

--
Dan

#4 lore raw

PATCH 1/3 dma: mv_xor: take channel spinlock in mv_xor_status()

Hello Dan,

Thomas is a bit busy right now, so I'm picking this where he left it.

Quoted 2 lines · Dan Williams

On Fri, Dec 27, 2013 at 3:38 AM, Thomas Petazzoni
<[email protected]> wrote:

[..]

Quoted 4 lines
Quoted 4 lines
drivers/dma/mv_xor.c +0 -0
@@ -701,14 +701,20 @@ static enum dma_status mv_xor_status(struct dma_chan *chan,
        struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
        enum dma_status ret;
 
Quoted 11 lines

drivers/dma/mv_xor.c static enum dma_status mv_xor_status(struct dma_chan *chan, · line 708

Quoted 5 lines
drivers/dma/mv_xor.c +2 -0
+       spin_lock_bh(&mv_chan->lock);
+
        ret = dma_cookie_status(chan, cookie, txstate);
        if (ret == DMA_COMPLETE) {
                mv_xor_clean_completed_slots(mv_chan);

I think you can just delete this call to mv_xor_clean_completed_slots(). The fact that the descriptors are complete means that __mv_xor_slot_cleanup ran, and if that is the case there should be nothing to cleanup.

In that case, we don't need to take the lock anywhere in this function:

1. It's not needed for dma_cookie_status() (as per Andy's comment).
2. It's not needed for mv_xor_slot_cleanup(), which takes the lock.

Signature · 3 lines
Ezequiel Garc?a, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com
#5 lore raw

PATCH 2/3 dma: mv_xor: use __mv_xor_slot_cleanup() in mv_xor_free_chan_resources()

In reply to Thomas Petazzoni

The mv_xor_free_chan_resources() was doing:

1/ Call mv_xor_slot_cleanup, which take the channel spinlock, does
its work, and releases the channel spinlock
2/ Take the channel spinlock
3/ Do its work
4/ Release the channel spinlock

This commit changes the function to use the __mv_xor_slot_cleanup() variant of mv_xor_slot_cleanup(), which assumes the caller already holds the channel spinlock. The sequence becomes:

1/ Take the channel spinlock
2/ Call __mv_xor_slot_cleanup
3/ Do the rest of the work
4/ Release the channel spinlock

---
 drivers/dma/mv_xor.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/dma/mv_xor.c +3 -2
index 526ab27..3d7d36b 100644
@@ -658,9 +658,10 @@ static void mv_xor_free_chan_resources(struct dma_chan *chan)
 	struct mv_xor_desc_slot *iter, *_iter;
 	int in_use_descs = 0;
 
-	mv_xor_slot_cleanup(mv_chan);
-
 	spin_lock_bh(&mv_chan->lock);
+
+	__mv_xor_slot_cleanup(mv_chan);
+
 	list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
 					chain_node) {
 		in_use_descs++;
Signature · 1 line
1.8.3.2
#6 lore raw

PATCH 3/3 dma: mv_xor: rename __mv_xor_slot_cleanup() to mv_xor_slot_cleanup()

In reply to Thomas Petazzoni

Now that mv_xor_slot_cleanup() has only one remaining caller: the mv_xor_tasklet(), we:

1/ get rid of mv_xor_slot_cleanup(), and directly take/release the
spinlock in mv_xor_tasklet()

2/ rename __mv_xor_slot_cleanup() to mv_xor_slot_cleanup()

We take this opportunity to add a comment that makes it clear that the channel spinlock should be held before calling mv_xor_slot_cleanup().

---
 drivers/dma/mv_xor.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)
drivers/dma/mv_xor.c +6 -11
index 3d7d36b..611795e 100644
@@ -312,7 +312,8 @@ mv_xor_clean_slot(struct mv_xor_desc_slot *desc,
 	return 0;
 }
 
-static void __mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
+/* This function must be called with the mv_xor_chan spinlock held */
+static void mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
 {
 	struct mv_xor_desc_slot *iter, *_iter;
 	dma_cookie_t cookie = 0;
@@ -368,18 +369,12 @@ static void __mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
 		mv_chan->dmachan.completed_cookie = cookie;
 }
 
-static void
-mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
-{
-	spin_lock_bh(&mv_chan->lock);
-	__mv_xor_slot_cleanup(mv_chan);
-	spin_unlock_bh(&mv_chan->lock);
-}
-
 static void mv_xor_tasklet(unsigned long data)
 {
 	struct mv_xor_chan *chan = (struct mv_xor_chan *) data;
+	spin_lock_bh(&chan->lock);
 	mv_xor_slot_cleanup(chan);
+	spin_unlock_bh(&chan->lock);
 }
 
 static struct mv_xor_desc_slot *
@@ -660,7 +655,7 @@ static void mv_xor_free_chan_resources(struct dma_chan *chan)
 
 	spin_lock_bh(&mv_chan->lock);
 
-	__mv_xor_slot_cleanup(mv_chan);
+	mv_xor_slot_cleanup(mv_chan);
 
 	list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
 					chain_node) {
@@ -710,7 +705,7 @@ static enum dma_status mv_xor_status(struct dma_chan *chan,
 		spin_unlock_bh(&mv_chan->lock);
 		return ret;
 	}
-	__mv_xor_slot_cleanup(mv_chan);
+	mv_xor_slot_cleanup(mv_chan);
 
 	ret = dma_cookie_status(chan, cookie, txstate);
 	spin_unlock_bh(&mv_chan->lock);
Signature · 1 line
1.8.3.2
#7 lore raw

cover dma: mv_xor: spinlock related improvements

In reply to Thomas Petazzoni
Quoted 10 lines · Thomas Petazzoni

Vinod, Dan,

Here are three patches that make improvements to the spinlock handling of the mv_xor driver. The first patch fixes a lock/release of the channel spinlock in mv_xor_status(), and the next two patches make followup code improvements/refactoring.

Ezequiel, it would be good to have your review on these patches, since I know you've spent some time lately doing mv_xor stuff.

Spinlocks aren't the easiest thing to get right, and this driver is not particularly simple. Anyway, to the best of my knowledge:

Signature · 3 lines
Ezequiel Garc?a, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com
#8 lore raw

cover dma: mv_xor: spinlock related improvements

In reply to Thomas Petazzoni
Quoted 14 lines · Thomas Petazzoni

Vinod, Dan,

Here are three patches that make improvements to the spinlock handling of the mv_xor driver. The first patch fixes a lock/release of the channel spinlock in mv_xor_status(), and the next two patches make followup code improvements/refactoring.

Ezequiel, it would be good to have your review on these patches, since I know you've spent some time lately doing mv_xor stuff.

Thanks!

Thomas

Quoted 8 lines

Thomas Petazzoni (3):
dma: mv_xor: take channel spinlock in mv_xor_status()
dma: mv_xor: use __mv_xor_slot_cleanup() in
mv_xor_free_chan_resources()
dma: mv_xor: rename __mv_xor_slot_cleanup() to mv_xor_slot_cleanup()

 drivers/dma/mv_xor.c | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

for the whole series:

thx,

Jason.