[PATCH] firewire: fw-ohci: use dma_alloc_coherent for ar_buffer

[PATCH] firewire: fw-ohci: use dma_alloc_coherent for ar_buffer

[Jarod Wilson]

Currently, we do nothing to guarantee we have a consistent DMA buffer for
asynchronous receive packets. Rather than doing several sync's following a
dma_map_single() to get consistent buffers, just switch to using
dma_alloc_coherent().

Resolves constant buffer failures on my own x86_64 laptop w/4GB of RAM and
likely to fix a number of other failures witnessed on x86_64 systems with
4GB of RAM or more.

Signed-off-by: Jarod Wilson <jwilson@redhat.com>

---

 drivers/firewire/fw-ohci.c |   18 +++++-------------
 1 files changed, 5 insertions(+), 13 deletions(-)

diff --git a/drivers/firewire/fw-ohci.c b/drivers/firewire/fw-ohci.c
index 28ea78c..081a434 100644
--- a/drivers/firewire/fw-ohci.c
+++ b/drivers/firewire/fw-ohci.c
@@ -284,16 +284,10 @@ static int ar_context_add_page(struct ar_context *ctx)
 	dma_addr_t ab_bus;
 	size_t offset;
 
-	ab = (struct ar_buffer *) __get_free_page(GFP_ATOMIC);
+	ab = dma_alloc_coherent(dev, PAGE_SIZE, &ab_bus, GFP_ATOMIC);
 	if (ab == NULL)
 		return -ENOMEM;
 
-	ab_bus = dma_map_single(dev, ab, PAGE_SIZE, DMA_BIDIRECTIONAL);
-	if (dma_mapping_error(ab_bus)) {
-		free_page((unsigned long) ab);
-		return -ENOMEM;
-	}
-
 	memset(&ab->descriptor, 0, sizeof(ab->descriptor));
 	ab->descriptor.control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
 						    DESCRIPTOR_STATUS |
@@ -304,8 +298,6 @@ static int ar_context_add_page(struct ar_context *ctx)
 	ab->descriptor.res_count      = cpu_to_le16(PAGE_SIZE - offset);
 	ab->descriptor.branch_address = 0;
 
-	dma_sync_single_for_device(dev, ab_bus, PAGE_SIZE, DMA_BIDIRECTIONAL);
-
 	ctx->last_buffer->descriptor.branch_address = cpu_to_le32(ab_bus | 1);
 	ctx->last_buffer->next = ab;
 	ctx->last_buffer = ab;
@@ -409,6 +401,7 @@ static void ar_context_tasklet(unsigned long data)
 
 	if (d->res_count == 0) {
 		size_t size, rest, offset;
+		dma_addr_t buffer_bus;
 
 		/*
 		 * This descriptor is finished and we may have a
@@ -417,9 +410,7 @@ static void ar_context_tasklet(unsigned long data)
 		 */
 
 		offset = offsetof(struct ar_buffer, data);
-		dma_unmap_single(ohci->card.device,
-			le32_to_cpu(ab->descriptor.data_address) - offset,
-			PAGE_SIZE, DMA_BIDIRECTIONAL);
+		buffer_bus = le32_to_cpu(ab->descriptor.data_address) - offset;
 
 		buffer = ab;
 		ab = ab->next;
@@ -435,7 +426,8 @@ static void ar_context_tasklet(unsigned long data)
 		while (buffer < end)
 			buffer = handle_ar_packet(ctx, buffer);
 
-		free_page((unsigned long)buffer);
+		dma_free_coherent(ohci->card.device, PAGE_SIZE,
+				  buffer, buffer_bus);
 		ar_context_add_page(ctx);
 	} else {
 		buffer = ctx->pointer;

-- 
Jarod Wilson
jwilson@redhat.com

Re: [PATCH] firewire: fw-ohci: use dma_alloc_coherent for ar_buffer

[Stefan Richter]

Jarod Wilson wrote:
> Currently, we do nothing to guarantee we have a consistent DMA buffer for
> asynchronous receive packets. Rather than doing several sync's following a
> dma_map_single() to get consistent buffers, just switch to using
> dma_alloc_coherent().
> 
> Resolves constant buffer failures on my own x86_64 laptop w/4GB of RAM and
> likely to fix a number of other failures witnessed on x86_64 systems with
> 4GB of RAM or more.
> 
> Signed-off-by: Jarod Wilson <jwilson@redhat.com>

Looks good at first glance.  Alas my PCs aren't affected, so I can only 
test it for lack of regressions.

I'm tempted to push this to Linus later this week.  Unless e.g. Kristian 
has reservations.

> 
> ---
> 
>  drivers/firewire/fw-ohci.c |   18 +++++-------------
>  1 files changed, 5 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/firewire/fw-ohci.c b/drivers/firewire/fw-ohci.c
> index 28ea78c..081a434 100644
> --- a/drivers/firewire/fw-ohci.c
> +++ b/drivers/firewire/fw-ohci.c
> @@ -284,16 +284,10 @@ static int ar_context_add_page(struct ar_context *ctx)
>  	dma_addr_t ab_bus;
>  	size_t offset;
>  
> -	ab = (struct ar_buffer *) __get_free_page(GFP_ATOMIC);
> +	ab = dma_alloc_coherent(dev, PAGE_SIZE, &ab_bus, GFP_ATOMIC);
>  	if (ab == NULL)
>  		return -ENOMEM;
>  
> -	ab_bus = dma_map_single(dev, ab, PAGE_SIZE, DMA_BIDIRECTIONAL);
> -	if (dma_mapping_error(ab_bus)) {
> -		free_page((unsigned long) ab);
> -		return -ENOMEM;
> -	}
> -
>  	memset(&ab->descriptor, 0, sizeof(ab->descriptor));
>  	ab->descriptor.control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
>  						    DESCRIPTOR_STATUS |
> @@ -304,8 +298,6 @@ static int ar_context_add_page(struct ar_context *ctx)
>  	ab->descriptor.res_count      = cpu_to_le16(PAGE_SIZE - offset);
>  	ab->descriptor.branch_address = 0;
>  
> -	dma_sync_single_for_device(dev, ab_bus, PAGE_SIZE, DMA_BIDIRECTIONAL);
> -
>  	ctx->last_buffer->descriptor.branch_address = cpu_to_le32(ab_bus | 1);
>  	ctx->last_buffer->next = ab;
>  	ctx->last_buffer = ab;
> @@ -409,6 +401,7 @@ static void ar_context_tasklet(unsigned long data)
>  
>  	if (d->res_count == 0) {
>  		size_t size, rest, offset;
> +		dma_addr_t buffer_bus;
>  
>  		/*
>  		 * This descriptor is finished and we may have a
> @@ -417,9 +410,7 @@ static void ar_context_tasklet(unsigned long data)
>  		 */
>  
>  		offset = offsetof(struct ar_buffer, data);
> -		dma_unmap_single(ohci->card.device,
> -			le32_to_cpu(ab->descriptor.data_address) - offset,
> -			PAGE_SIZE, DMA_BIDIRECTIONAL);
> +		buffer_bus = le32_to_cpu(ab->descriptor.data_address) - offset;
>  
>  		buffer = ab;
>  		ab = ab->next;
> @@ -435,7 +426,8 @@ static void ar_context_tasklet(unsigned long data)
>  		while (buffer < end)
>  			buffer = handle_ar_packet(ctx, buffer);
>  
> -		free_page((unsigned long)buffer);
> +		dma_free_coherent(ohci->card.device, PAGE_SIZE,
> +				  buffer, buffer_bus);
>  		ar_context_add_page(ctx);
>  	} else {
>  		buffer = ctx->pointer;
> 


-- 
Stefan Richter
-=====-==--- --== -==-=
http://arcgraph.de/sr/

[PATCH update] firewire: fw-ohci: shut up false compiler warning on PPC32

[Stefan Richter]

Shut up "may be used uninitialised in this function" warnings due to
PPC32's implementation of dma_alloc_coherent().

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
---

updated to be enqueued after patch
"firewire: fw-ohci: use dma_alloc_coherent for ar_buffer"

 drivers/firewire/fw-ohci.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Index: linux/drivers/firewire/fw-ohci.c
===================================================================
--- linux.orig/drivers/firewire/fw-ohci.c
+++ linux/drivers/firewire/fw-ohci.c
@@ -281,7 +281,7 @@ static int ar_context_add_page(struct ar
 {
 	struct device *dev = ctx->ohci->card.device;
 	struct ar_buffer *ab;
-	dma_addr_t ab_bus;
+	dma_addr_t uninitialized_var(ab_bus);
 	size_t offset;
 
 	ab = dma_alloc_coherent(dev, PAGE_SIZE, &ab_bus, GFP_ATOMIC);
@@ -536,7 +536,7 @@ static int
 context_add_buffer(struct context *ctx)
 {
 	struct descriptor_buffer *desc;
-	dma_addr_t bus_addr;
+	dma_addr_t uninitialized_var(bus_addr);
 	int offset;
 
 	/*
@@ -1321,7 +1321,7 @@ ohci_set_config_rom(struct fw_card *card
 	unsigned long flags;
 	int retval = -EBUSY;
 	__be32 *next_config_rom;
-	dma_addr_t next_config_rom_bus;
+	dma_addr_t uninitialized_var(next_config_rom_bus);
 
 	ohci = fw_ohci(card);
 

-- 
Stefan Richter
-=====-==--- --== -===-
http://arcgraph.de/sr/

Re: [PATCH] firewire: fw-ohci: use dma_alloc_coherent for ar_buffer

[Stefan Richter]

I wrote:
> Jarod Wilson wrote:
>> Resolves constant buffer failures on my own x86_64 laptop w/4GB of RAM and
>> likely to fix a number of other failures witnessed on x86_64 systems with
>> 4GB of RAM or more.
>> 
>> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
> 
> Looks good at first glance.  Alas my PCs aren't affected, so I can only 
> test it for lack of regressions.
> 
> I'm tempted to push this to Linus later this week.

I tested it on a 2GB EM64T machine, on PPC32, and i686.  I used the
latter for stress-tests with SBP-2, DV, and IIDC traffic on two buses
with 3 devices each, all active at once.

I had one lock-up during the stress-tests which I suspect is the
handle_at_packet panic, http://bugzilla.kernel.org/show_bug.cgi?id=9617
which is unrelated to this patch.

I committed your patch to linux1394-2.6.git, whose shortlog and diff
from origin..for-linus looks now like this:

Jarod Wilson (2):
      firewire: fw-sbp2: set single-phase retry_limit
      firewire: fw-ohci: use dma_alloc_coherent for ar_buffer

Stefan Richter (9):
      firewire: endianess fix
      firewire: endianess annotations
      firewire: fw-ohci: PPC PMac platform code
      firewire: fw-ohci: Apple UniNorth 1st generation support
      firewire: warn on fatal condition in topology code
      firewire: update Kconfig help text
      firewire: fw-sbp2: fix for SYM13FW500 bridge (Datafab disk)
      ieee1394: sbp2: fix for SYM13FW500 bridge (Datafab disk)
      firewire: fw-ohci: shut up false compiler warning on PPC32

 drivers/firewire/Kconfig          |   50 +++++--------
 drivers/firewire/fw-ohci.c        |  108 +++++++++++++++++++++-------
 drivers/firewire/fw-sbp2.c        |   36 +++++++++-
 drivers/firewire/fw-topology.c    |    3 +-
 drivers/firewire/fw-transaction.c |    2 +-
 drivers/firewire/fw-transaction.h |    6 +-
 drivers/ieee1394/sbp2.c           |    5 ++
 7 files changed, 143 insertions(+), 67 deletions(-)

I'd like to submit this to Linus later today (Friday).  It all looks
like a bit too much for such a late -rc phase; but half of fw-ohci's
diffstat is actually just the added PMac code which has already proven
itself in ohci1394.

There are also some small new commits to linux1394-2.6.git master, whose
history is once again somewhat obfuscated by my pulling 'origin' and
'for-linus' into master, so that vice versa master can be cleanly pulled
into descendants of Linus' tree.

http://me.in-berlin.de/~s5r6/linux1394/updates/2.6.25-rc5/patches/series
gives a better picture of what is currently in the pipeline than git log
would give.
-- 
Stefan Richter
-=====-==--- --== -===-
http://arcgraph.de/sr/

Re: [PATCH update] firewire: fw-ohci: shut up false compiler warning on PPC32

[Jarod Wilson]

On Thursday 13 March 2008 07:27:49 pm Stefan Richter wrote:
> Shut up "may be used uninitialised in this function" warnings due to
> PPC32's implementation of dma_alloc_coherent().
>
> Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
> ---
>
> updated to be enqueued after patch
> "firewire: fw-ohci: use dma_alloc_coherent for ar_buffer"

Signed-off-by: Jarod Wilson <jwilson@redhat.com>

-- 
Jarod Wilson
jwilson@redhat.com