Skip to contentSkip to conversation

dmatest: properly handle duplicate DMA channels

PATCH Superseded Timur Tabi 4 participants 8 messages Sep 18, 2008 – Sep 20, 2008 dmaengine

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Timur Tabi

author
#0 lore raw

PATCH 1/1 dmatest: properly handle duplicate DMA channels

Update the the dmatest driver so that it handles duplicate DMA channels properly.

When a DMA client is notified of an available DMA channel, it must check if it has already allocated resources for that channel. If so, it should return DMA_DUP. This can happen, for example, if a DMA driver calls dma_async_device_register() more than once.

---
 drivers/dma/dmatest.c |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)
drivers/dma/dmatest.c +6 -1
index a08d197..2689d90 100644
@@ -321,10 +321,15 @@ static void dmatest_cleanup_channel(struct dmatest_chan *dtc)
 
 static enum dma_state_client dmatest_add_channel(struct dma_chan *chan)
 {
-	struct dmatest_chan	*dtc;
+	struct dmatest_chan	*dtc, *_dtc;
 	struct dmatest_thread	*thread;
 	unsigned int		i;
 
+	/* Have we already been told about this channel? */
+	list_for_each_entry_safe(dtc, _dtc, &dmatest_channels, node)
+		if (dtc->chan == chan)
+			return DMA_DUP;
+
 	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_ATOMIC);
 	if (!dtc) {
 		pr_warning("dmatest: No memory for %s\n", chan->dev.bus_id);
Signature · 1 line
1.5.5
Acked-by on v1 1/1
#1 lore raw
Quoted 9 lines · Timur Tabi

Update the the dmatest driver so that it handles duplicate DMA channels properly.

When a DMA client is notified of an available DMA channel, it must check if it has already allocated resources for that channel. If so, it should return DMA_DUP. This can happen, for example, if a DMA driver calls dma_async_device_register() more than once.

...though I don't really see why you have to use
list_for_entry_safe()...

Thanks a lot!

Haavard

author
#2 lore raw
Quoted 2 lines · Haavard Skinnemoen

...though I don't really see why you have to use
list_for_entry_safe()...

You use it in dmatest_remove_channel(), so I figured I'd do the same thing.

Signature · 2 lines
Timur Tabi
Linux kernel developer at Freescale
#3 lore raw
Quoted 6 lines · Timur Tabi
Quoted 2 lines · Haavard Skinnemoen

...though I don't really see why you have to use
list_for_entry_safe()...

You use it in dmatest_remove_channel(), so I figured I'd do the same thing.

dmatest_remove_channel() may call list_del() on the current entry, so it has to. As long as you're not altering the list, the _safe() variant shouldn't be necessary.

Haavard

#4 lore raw
In reply to Timur Tabi
Quoted 9 lines · On Thu, 18 Sep 2008 10:21:19 -0500 Timur Tabi <[email protected]> wrote:
drivers/dma/dmatest.c +1 -1
@@ -321,10 +321,15 @@ static void dmatest_cleanup_channel(struct dmatest_chan *dtc)
 
 static enum dma_state_client dmatest_add_channel(struct dma_chan *chan)
 {
-	struct dmatest_chan	*dtc;
+	struct dmatest_chan	*dtc, *_dtc;
 	struct dmatest_thread	*thread;

drivers/dma/dmatest.c static void dmatest_cleanup_channel(struct dmatest_chan *dtc) · line 335

Quoted 10 lines
drivers/dma/dmatest.c +5 -0
 	unsigned int		i;
 
+	/* Have we already been told about this channel? */
+	list_for_each_entry_safe(dtc, _dtc, &dmatest_channels, node)
+		if (dtc->chan == chan)
+			return DMA_DUP;
+
 	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_ATOMIC);
 	if (!dtc) {
 		pr_warning("dmatest: No memory for %s\n", chan->dev.bus_id);

hm. A few lines after that GFP_ATOMIC the driver does a GFP_KERNEL allocation.

One of them is incorrect. The interface is undocumented (natch), but I assume that GFP_KERNEL is the one to use here.

drivers/dma/dmatest.c renamed from drivers/dma/dmatest.c~a +1 -1
@@ -330,7 +330,7 @@ static enum dma_state_client dmatest_add
 		if (dtc->chan == chan)
 			return DMA_DUP;
 
-	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_ATOMIC);
+	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL);
 	if (!dtc) {
 		pr_warning("dmatest: No memory for %s\n", chan->dev.bus_id);
 		return DMA_NAK;

_

#5 lore raw
Quoted 2 lines · Andrew Morton

One of them is incorrect. The interface is undocumented (natch), but I assume that GFP_KERNEL is the one to use here.

You're right -- there's no reason to use GFP_ATOMIC there. I messed around with various kinds of locks before I realized that it was already protected by a mutex.

Haavard

#6 lore raw
In reply to Andrew Morton
Quoted 26 lines · Andrew Morton
Quoted 9 lines · On Thu, 18 Sep 2008 10:21:19 -0500 Timur Tabi <[email protected]> wrote:
drivers/dma/dmatest.c +1 -1
@@ -321,10 +321,15 @@ static void dmatest_cleanup_channel(struct dmatest_chan *dtc)
 
 static enum dma_state_client dmatest_add_channel(struct dma_chan *chan)
 {
-     struct dmatest_chan     *dtc;
+     struct dmatest_chan     *dtc, *_dtc;
      struct dmatest_thread   *thread;

drivers/dma/dmatest.c static void dmatest_cleanup_channel(struct dmatest_chan *dtc) · line 335

Quoted 10 lines
drivers/dma/dmatest.c +5 -0
      unsigned int            i;
 
+     /* Have we already been told about this channel? */
+     list_for_each_entry_safe(dtc, _dtc, &dmatest_channels, node)
+             if (dtc->chan == chan)
+                     return DMA_DUP;
+
      dtc = kmalloc(sizeof(struct dmatest_chan), GFP_ATOMIC);
      if (!dtc) {
              pr_warning("dmatest: No memory for %s\n", chan->dev.bus_id);

hm. A few lines after that GFP_ATOMIC the driver does a GFP_KERNEL allocation.

Quoted 3 lines

One of them is incorrect. The interface is undocumented (natch), but I assume that GFP_KERNEL is the one to use here.

The interface is documented, although not the locking (natch), at include/linux/dmaengine.h:229. When I fix this up is there a canonical location to document a callback interface rather than at the callback's typedef?

Thanks,
Dan

#7 lore raw
Quoted 3 lines · On Sat, 20 Sep 2008 14:40:40 -0700 "Dan Williams" <[email protected]>…

When I fix this up is there a
canonical location to document a callback interface rather than at the callback's typedef?

Personally I like to see at at the definition site, so that's within the struct which contains the function pointer. eg:

struct address_space_operations {
	int (*writepage)(struct page *page, struct writeback_control *wbc);
	int (*readpage)(struct file *, struct page *);
	void (*sync_page)(struct page *);

	/* Write back some dirty pages from this mapping. */
	int (*writepages)(struct address_space *, struct writeback_control *);

	/* Set a page dirty.  Return true if this dirtied it */
	int (*set_page_dirty)(struct page *page);

(sadly incomplete, but we tried)