Re: [PATCH RESEND] dma: cppi41: only allocate descriptor memory once

2013-12-04 Thread Daniel Mack
On 12/04/2013 12:51 PM, Steve Cotton wrote:
> On Wed, Dec 04, 2013 at 11:21 +0100, Daniel Mack wrote:
>> While at it, remove the intermediate variable mem_decs (I guess it was
>> only there to make the code comply to the 80-chars CodingSytle rule).
> 
> purge_descs() still has a (now unused) mem_decs intermediate variable.

You're right. Thanks for spotting it. I'll send a v2.


Daniel

--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html


Re: [PATCH RESEND] dma: cppi41: only allocate descriptor memory once

2013-12-04 Thread Steve Cotton
On Wed, Dec 04, 2013 at 11:21 +0100, Daniel Mack wrote:
> While at it, remove the intermediate variable mem_decs (I guess it was
> only there to make the code comply to the 80-chars CodingSytle rule).

purge_descs() still has a (now unused) mem_decs intermediate variable.

Regards,
Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[PATCH RESEND] dma: cppi41: only allocate descriptor memory once

2013-12-04 Thread Daniel Mack
cdd->cd and cdd->descs_phys are allocated DESCS_AREAS times from
init_descs() and freed as often from purge_descs(). This doesn't cause
any trouble as DESCS_AREAS is ensured to be 1, but it looks wrong to
do the (de)allocation inside the loop.

While at it, remove the intermediate variable mem_decs (I guess it was
only there to make the code comply to the 80-chars CodingSytle rule).

Signed-off-by: Daniel Mack 
---

Hi Vinod,
Hi Sebastian,

I know we discussed that patch before, but AFAIR, there was no
conclusion. Could we have another look please?

As mentioned, I still think the current code looks wrong, even though it
doesn't harm because DESCS_AREAS is always 1, as ensured by 
BUILD_BUG_ON(DESCS_AREAS != 1).

Still, calling dma_{alloc,free}_coherent inside the loop without the
loop variable anywhere in the assigned pointer really smells like
a memory leak or double-free, respectively, doesn't it? :)


Many thanks,
Daniel

 drivers/dma/cppi41.c | 25 ++---
 1 file changed, 10 insertions(+), 15 deletions(-)

diff --git a/drivers/dma/cppi41.c b/drivers/dma/cppi41.c
index c29dacf..896252f 100644
--- a/drivers/dma/cppi41.c
+++ b/drivers/dma/cppi41.c
@@ -719,13 +719,12 @@ static void purge_descs(struct device *dev, struct 
cppi41_dd *cdd)
mem_decs = ALLOC_DECS_NUM * sizeof(struct cppi41_desc);
 
for (i = 0; i < DESCS_AREAS; i++) {
-
cppi_writel(0, cdd->qmgr_mem + QMGR_MEMBASE(i));
cppi_writel(0, cdd->qmgr_mem + QMGR_MEMCTRL(i));
-
-   dma_free_coherent(dev, mem_decs, cdd->cd,
-   cdd->descs_phys);
}
+
+   dma_free_coherent(dev, ALLOC_DECS_NUM * sizeof(struct cppi41_desc),
+ cdd->cd, cdd->descs_phys);
 }
 
 static void disable_sched(struct cppi41_dd *cdd)
@@ -747,8 +746,7 @@ static void deinit_cppi41(struct device *dev, struct 
cppi41_dd *cdd)
 
 static int init_descs(struct device *dev, struct cppi41_dd *cdd)
 {
-   unsigned int desc_size;
-   unsigned int mem_decs;
+   unsigned int desc_size = sizeof(struct cppi41_desc);
int i;
u32 reg;
u32 idx;
@@ -757,28 +755,25 @@ static int init_descs(struct device *dev, struct 
cppi41_dd *cdd)
(sizeof(struct cppi41_desc) - 1));
BUILD_BUG_ON(sizeof(struct cppi41_desc) < 32);
BUILD_BUG_ON(ALLOC_DECS_NUM < 32);
+   BUILD_BUG_ON(DESCS_AREAS != 1);
 
-   desc_size = sizeof(struct cppi41_desc);
-   mem_decs = ALLOC_DECS_NUM * desc_size;
+   cdd->cd = dma_alloc_coherent(dev, ALLOC_DECS_NUM * desc_size,
+&cdd->descs_phys, GFP_KERNEL);
+   if (!cdd->cd)
+   return -ENOMEM;
 
idx = 0;
for (i = 0; i < DESCS_AREAS; i++) {
-
reg = idx << QMGR_MEMCTRL_IDX_SH;
reg |= (ilog2(desc_size) - 5) << QMGR_MEMCTRL_DESC_SH;
reg |= ilog2(ALLOC_DECS_NUM) - 5;
 
-   BUILD_BUG_ON(DESCS_AREAS != 1);
-   cdd->cd = dma_alloc_coherent(dev, mem_decs,
-   &cdd->descs_phys, GFP_KERNEL);
-   if (!cdd->cd)
-   return -ENOMEM;
-
cppi_writel(cdd->descs_phys, cdd->qmgr_mem + QMGR_MEMBASE(i));
cppi_writel(reg, cdd->qmgr_mem + QMGR_MEMCTRL(i));
 
idx += ALLOC_DECS_NUM;
}
+
return 0;
 }
 
-- 
1.8.4.2

--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html