2014-04-10 08:38:51

by Sonic Zhang

[permalink] [raw]
Subject: [PATCH 1/2] crypto: bfin_crc: ignore duplicated registration of the same algorithm

From: Sonic Zhang <[email protected]>

in case of multiple crc devices are probed.

Signed-off-by: Sonic Zhang <[email protected]>
---
drivers/crypto/bfin_crc.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/crypto/bfin_crc.c b/drivers/crypto/bfin_crc.c
index cea3e8c..5f5f427 100644
--- a/drivers/crypto/bfin_crc.c
+++ b/drivers/crypto/bfin_crc.c
@@ -672,7 +672,7 @@ static int bfin_crypto_crc_probe(struct platform_device *pdev)
platform_set_drvdata(pdev, crc);

ret = crypto_register_ahash(&algs);
- if (ret) {
+ if (ret && ret != -EEXIST) {
spin_lock(&crc_list.lock);
list_del(&crc->list);
spin_unlock(&crc_list.lock);
--
1.8.2.3


2014-04-10 08:38:45

by Sonic Zhang

[permalink] [raw]
Subject: [PATCH 2/2] crypto: bfin_crc: avoid get physical address of coherence memory by dma_map_single

From: Sonic Zhang <[email protected]>

- The 4-byte sg_mid_buf is located in the middle of the coherence memory
sg_cpu. Don't call dma_map_single to get its physical address. Get the its
base physical address from the physical address of sg_cpu instead.
- Should set up the dma descriptor data after the 4-byte sg_mid_buf is
filled in completely from next sg buffer.
- memory copy from sg buffer should be done via virtual address.
- Remove unused reference to blackfin header

Signed-off-by: Sonic Zhang <[email protected]>
---
drivers/crypto/bfin_crc.c | 43 +++++++++++++++++++++++--------------------
1 file changed, 23 insertions(+), 20 deletions(-)

diff --git a/drivers/crypto/bfin_crc.c b/drivers/crypto/bfin_crc.c
index 5f5f427..94a0792 100644
--- a/drivers/crypto/bfin_crc.c
+++ b/drivers/crypto/bfin_crc.c
@@ -29,7 +29,6 @@
#include <crypto/hash.h>
#include <crypto/internal/hash.h>

-#include <asm/blackfin.h>
#include <asm/dma.h>
#include <asm/portmux.h>
#include <asm/io.h>
@@ -62,6 +61,7 @@ struct bfin_crypto_crc {
struct dma_desc_array *sg_cpu; /* virt addr of sg dma descriptors */
dma_addr_t sg_dma; /* phy addr of sg dma descriptors */
u8 *sg_mid_buf;
+ dma_addr_t sg_mid_dma; /* phy addr of sg mid buffer */

struct tasklet_struct done_task;
struct crypto_queue queue; /* waiting requests */
@@ -196,7 +196,6 @@ static void bfin_crypto_crc_config_dma(struct bfin_crypto_crc *crc)
dma_map_sg(crc->dev, ctx->sg, ctx->sg_nents, DMA_TO_DEVICE);

for_each_sg(ctx->sg, sg, ctx->sg_nents, j) {
- dma_config = DMAFLOW_ARRAY | RESTART | NDSIZE_3 | DMAEN | PSIZE_32;
dma_addr = sg_dma_address(sg);
/* deduce extra bytes in last sg */
if (sg_is_last(sg))
@@ -209,12 +208,29 @@ static void bfin_crypto_crc_config_dma(struct bfin_crypto_crc *crc)
bytes in current sg buffer. Move addr of current
sg and deduce the length of current sg.
*/
- memcpy(crc->sg_mid_buf +((i-1) << 2) + mid_dma_count,
- (void *)dma_addr,
+ memcpy(crc->sg_mid_buf +(i << 2) + mid_dma_count,
+ sg_virt(sg),
CHKSUM_DIGEST_SIZE - mid_dma_count);
dma_addr += CHKSUM_DIGEST_SIZE - mid_dma_count;
dma_count -= CHKSUM_DIGEST_SIZE - mid_dma_count;
+
+ dma_config = DMAFLOW_ARRAY | RESTART | NDSIZE_3 |
+ DMAEN | PSIZE_32 | WDSIZE_32;
+
+ /* setup new dma descriptor for next middle dma */
+ crc->sg_cpu[i].start_addr = crc->sg_mid_dma + (i << 2);
+ crc->sg_cpu[i].cfg = dma_config;
+ crc->sg_cpu[i].x_count = 1;
+ crc->sg_cpu[i].x_modify = CHKSUM_DIGEST_SIZE;
+ dev_dbg(crc->dev, "%d: crc_dma: start_addr:0x%lx, "
+ "cfg:0x%lx, x_count:0x%lx, x_modify:0x%lx\n",
+ i, crc->sg_cpu[i].start_addr,
+ crc->sg_cpu[i].cfg, crc->sg_cpu[i].x_count,
+ crc->sg_cpu[i].x_modify);
+ i++;
}
+
+ dma_config = DMAFLOW_ARRAY | RESTART | NDSIZE_3 | DMAEN | PSIZE_32;
/* chop current sg dma len to multiple of 32 bits */
mid_dma_count = dma_count % 4;
dma_count &= ~0x3;
@@ -245,24 +261,9 @@ static void bfin_crypto_crc_config_dma(struct bfin_crypto_crc *crc)

if (mid_dma_count) {
/* copy extra bytes to next middle dma buffer */
- dma_config = DMAFLOW_ARRAY | RESTART | NDSIZE_3 |
- DMAEN | PSIZE_32 | WDSIZE_32;
memcpy(crc->sg_mid_buf + (i << 2),
- (void *)(dma_addr + (dma_count << 2)),
+ (u8*)sg_virt(sg) + (dma_count << 2),
mid_dma_count);
- /* setup new dma descriptor for next middle dma */
- crc->sg_cpu[i].start_addr = dma_map_single(crc->dev,
- crc->sg_mid_buf + (i << 2),
- CHKSUM_DIGEST_SIZE, DMA_TO_DEVICE);
- crc->sg_cpu[i].cfg = dma_config;
- crc->sg_cpu[i].x_count = 1;
- crc->sg_cpu[i].x_modify = CHKSUM_DIGEST_SIZE;
- dev_dbg(crc->dev, "%d: crc_dma: start_addr:0x%lx, "
- "cfg:0x%lx, x_count:0x%lx, x_modify:0x%lx\n",
- i, crc->sg_cpu[i].start_addr,
- crc->sg_cpu[i].cfg, crc->sg_cpu[i].x_count,
- crc->sg_cpu[i].x_modify);
- i++;
}
}

@@ -654,6 +655,8 @@ static int bfin_crypto_crc_probe(struct platform_device *pdev)
* 1 last + 1 next dma descriptors
*/
crc->sg_mid_buf = (u8 *)(crc->sg_cpu + ((CRC_MAX_DMA_DESC + 1) << 1));
+ crc->sg_mid_dma = crc->sg_dma + sizeof(struct dma_desc_array)
+ * ((CRC_MAX_DMA_DESC + 1) << 1);

writel(0, &crc->regs->control);
crc->poly = (u32)pdev->dev.platform_data;
--
1.8.2.3

2014-04-10 09:02:54

by Marek Vasut

[permalink] [raw]
Subject: Re: [PATCH 2/2] crypto: bfin_crc: avoid get physical address of coherence memory by dma_map_single

On Thursday, April 10, 2014 at 10:40:59 AM, Sonic Zhang wrote:
> From: Sonic Zhang <[email protected]>
>
> - The 4-byte sg_mid_buf is located in the middle of the coherence memory
> sg_cpu. Don't call dma_map_single to get its physical address. Get the its
> base physical address from the physical address of sg_cpu instead.
> - Should set up the dma descriptor data after the 4-byte sg_mid_buf is
> filled in completely from next sg buffer.
> - memory copy from sg buffer should be done via virtual address.
> - Remove unused reference to blackfin header
>
> Signed-off-by: Sonic Zhang <[email protected]>

I don't see any obvious problems,

Reviewed-by: Marek Vasut <[email protected]>

Best regards,
Marek Vasut

2014-04-10 09:02:52

by Marek Vasut

[permalink] [raw]
Subject: Re: [PATCH 1/2] crypto: bfin_crc: ignore duplicated registration of the same algorithm

On Thursday, April 10, 2014 at 10:40:58 AM, Sonic Zhang wrote:
> From: Sonic Zhang <[email protected]>
>
> in case of multiple crc devices are probed.
>
> Signed-off-by: Sonic Zhang <[email protected]>
> ---
> drivers/crypto/bfin_crc.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/crypto/bfin_crc.c b/drivers/crypto/bfin_crc.c
> index cea3e8c..5f5f427 100644
> --- a/drivers/crypto/bfin_crc.c
> +++ b/drivers/crypto/bfin_crc.c
> @@ -672,7 +672,7 @@ static int bfin_crypto_crc_probe(struct platform_device
> *pdev) platform_set_drvdata(pdev, crc);

Rant: Reading through the code a little, the platform_set_drvdata() should
happen before you add the new CRC device into the list (above this code).

> ret = crypto_register_ahash(&algs);
> - if (ret) {
> + if (ret && ret != -EEXIST) {

Uh oh, how is this supposed to work ? Does blackfin have multiple crc32
processing units ? Can you simply not check if the crc_list is not empty and
avoid calling crypto_register_ahash() for subsequent crc32 units at all ?

> spin_lock(&crc_list.lock);
> list_del(&crc->list);
> spin_unlock(&crc_list.lock);

Best regards,
Marek Vasut

2014-04-10 09:51:30

by Sonic Zhang

[permalink] [raw]
Subject: Re: [PATCH 1/2] crypto: bfin_crc: ignore duplicated registration of the same algorithm

Hi Marek,

On Thu, Apr 10, 2014 at 4:59 PM, Marek Vasut <[email protected]> wrote:
> On Thursday, April 10, 2014 at 10:40:58 AM, Sonic Zhang wrote:
>> From: Sonic Zhang <[email protected]>
>>
>> in case of multiple crc devices are probed.
>>
>> Signed-off-by: Sonic Zhang <[email protected]>
>> ---
>> drivers/crypto/bfin_crc.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/crypto/bfin_crc.c b/drivers/crypto/bfin_crc.c
>> index cea3e8c..5f5f427 100644
>> --- a/drivers/crypto/bfin_crc.c
>> +++ b/drivers/crypto/bfin_crc.c
>> @@ -672,7 +672,7 @@ static int bfin_crypto_crc_probe(struct platform_device
>> *pdev) platform_set_drvdata(pdev, crc);
>
> Rant: Reading through the code a little, the platform_set_drvdata() should
> happen before you add the new CRC device into the list (above this code).

OK

>
>> ret = crypto_register_ahash(&algs);
>> - if (ret) {
>> + if (ret && ret != -EEXIST) {
>
> Uh oh, how is this supposed to work ? Does blackfin have multiple crc32
> processing units ? Can you simply not check if the crc_list is not empty and
> avoid calling crypto_register_ahash() for subsequent crc32 units at all ?

Yes, Blackfin has 2 crc32 units CRC0 and CRC1.
I will have a try to see if it works.

Thanks

Sonic

2014-04-10 12:33:51

by Marek Vasut

[permalink] [raw]
Subject: Re: [PATCH 1/2] crypto: bfin_crc: ignore duplicated registration of the same algorithm

On Thursday, April 10, 2014 at 11:51:29 AM, Sonic Zhang wrote:
> Hi Marek,
>
> On Thu, Apr 10, 2014 at 4:59 PM, Marek Vasut <[email protected]> wrote:
> > On Thursday, April 10, 2014 at 10:40:58 AM, Sonic Zhang wrote:
> >> From: Sonic Zhang <[email protected]>
> >>
> >> in case of multiple crc devices are probed.
> >>
> >> Signed-off-by: Sonic Zhang <[email protected]>
> >> ---
> >>
> >> drivers/crypto/bfin_crc.c | 2 +-
> >> 1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/crypto/bfin_crc.c b/drivers/crypto/bfin_crc.c
> >> index cea3e8c..5f5f427 100644
> >> --- a/drivers/crypto/bfin_crc.c
> >> +++ b/drivers/crypto/bfin_crc.c
> >> @@ -672,7 +672,7 @@ static int bfin_crypto_crc_probe(struct
> >> platform_device *pdev) platform_set_drvdata(pdev, crc);
> >
> > Rant: Reading through the code a little, the platform_set_drvdata()
> > should happen before you add the new CRC device into the list (above
> > this code).
>
> OK
>
> >> ret = crypto_register_ahash(&algs);
> >>
> >> - if (ret) {
> >> + if (ret && ret != -EEXIST) {
> >
> > Uh oh, how is this supposed to work ? Does blackfin have multiple crc32
> > processing units ? Can you simply not check if the crc_list is not empty
> > and avoid calling crypto_register_ahash() for subsequent crc32 units at
> > all ?
>
> Yes, Blackfin has 2 crc32 units CRC0 and CRC1.
> I will have a try to see if it works.

Thank you !

Best regards,
Marek Vasut

2014-04-14 04:41:31

by Sonic Zhang

[permalink] [raw]
Subject: Re: [PATCH 2/2] crypto: bfin_crc: avoid get physical address of coherence memory by dma_map_single

Hi Herbert,

Could you please review this patch?

Thanks

Sonic Zhang


On Thu, Apr 10, 2014 at 5:02 PM, Marek Vasut <[email protected]> wrote:
> On Thursday, April 10, 2014 at 10:40:59 AM, Sonic Zhang wrote:
>> From: Sonic Zhang <[email protected]>
>>
>> - The 4-byte sg_mid_buf is located in the middle of the coherence memory
>> sg_cpu. Don't call dma_map_single to get its physical address. Get the its
>> base physical address from the physical address of sg_cpu instead.
>> - Should set up the dma descriptor data after the 4-byte sg_mid_buf is
>> filled in completely from next sg buffer.
>> - memory copy from sg buffer should be done via virtual address.
>> - Remove unused reference to blackfin header
>>
>> Signed-off-by: Sonic Zhang <[email protected]>
>
> I don't see any obvious problems,
>
> Reviewed-by: Marek Vasut <[email protected]>
>
> Best regards,
> Marek Vasut