Re: [PATCH v3 1/3] mmc: mmci: Support any block sizes for ux500v2 and qcom variant

From: Srinivas Kandagatla
Date: Wed Sep 10 2014 - 05:07:40 EST

Hi Ulf,

On 10/09/14 08:58, Ulf Hansson wrote:
On 22 August 2014 06:54, Srinivas Kandagatla
<srinivas.kandagatla@xxxxxxxxxx> wrote:
From: Ulf Hansson <ulf.hansson@xxxxxxxxxx>

For the ux500v2 variant of the PL18x block, any block sizes are
supported. This will make it possible to decrease data overhead
for SDIO transfers.

This patch is based on Ulf Hansson patch
http://www.spinics.net/lists/linux-mmc/msg12160.html

Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@xxxxxxxxxx>
enabled this support on qcom variant.

Signed-off-by: Ulf Hansson <ulf.hansson@xxxxxxxxxx>
Signed-off-by: Linus Walleij <linus.walleij@xxxxxxxxxx>

I am not sure how to handle this patch.

It will as you say in the cover letter for this patchset, improve
situations for the ath6kl driver when it's issuing 12 bytes and 24
bytes reads and solve those issues.

On the other hand, as stated earlier mmci_pio_write need to be fixed
to have full support for any block size. That applies to the qcom
variant as well.
looking at current mmci_pio_write, I see it can support any block sizes as it is. Unless Am missing something obvious.

block size aligned to 4 is taken care in the code and is straight forward.
block sizes not aligned to 4 are also partly taken care in pio_write and partly by programming blksz in datactrl register. However Am not sure if it was safe to handle the buffer pointer out of its boundary.

For ux500, I am sure this won't cause any regressions since the cw1200
isn't probed. Also, I am not sure the "any block size" support is even
enabled for that driver.

How about, that we add a comment in the pio_write function describing
that we need to fix it for "SDIO any block size" support? And leave
that as a future improvement?

Is below patch any good?
With the below patch It should be possible to address the case where buffer passed the length is handled safely. If you ok with the approach I can send a patch as RFC.


----cut here----
diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
index 264c947..8480c02 100644
--- a/drivers/mmc/host/mmci.c
+++ b/drivers/mmc/host/mmci.c
@@ -1134,7 +1134,20 @@ static int mmci_pio_write(struct mmci_host *host, char *buffer, unsigned int rem
* byte become a 32bit write, 7 bytes will be two
* 32bit writes etc.
*/
- iowrite32_rep(base + MMCIFIFO, ptr, (count + 3) >> 2);
+ if (unlikely(count & 0x3)) {
+ unsigned char buf[4] = {0, };
+
+ if (count < 4) {
+ memcpy(buf, ptr, count);
+ iowrite32_rep(base + MMCIFIFO, buf, 1);
+ } else {
+ iowrite32_rep(base + MMCIFIFO, ptr, count >> 2);
+ memcpy(buf, ptr + (count & ~0x3), count & 0x3);
+ iowrite32_rep(base + MMCIFIFO, buf, 1);
+ }
+ } else {
+ iowrite32_rep(base + MMCIFIFO, ptr, count >> 2);
+ }

ptr += count;
remain -= count;

----cut here----

thanks,
srini

Kind regards
Uffe

---
drivers/mmc/host/mmci.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
index c11cb05..533ad2b 100644
--- a/drivers/mmc/host/mmci.c
+++ b/drivers/mmc/host/mmci.c
@@ -77,6 +77,7 @@ static unsigned int fmax = 515633;
* @qcom_fifo: enables qcom specific fifo pio read logic.
* @reversed_irq_handling: handle data irq before cmd irq.
* @qcom_dml: enables qcom specific dma glue for dma transfers.
+ * @any_blksize: true if block any sizes are supported
*/
struct variant_data {
unsigned int clkreg;
@@ -102,6 +103,7 @@ struct variant_data {
bool qcom_fifo;
bool reversed_irq_handling;
bool qcom_dml;
+ bool any_blksize;
};

static struct variant_data variant_arm = {
@@ -194,6 +196,7 @@ static struct variant_data variant_ux500v2 = {
.pwrreg_clkgate = true,
.busy_detect = true,
.pwrreg_nopower = true,
+ .any_blksize = true,
};

static struct variant_data variant_qcom = {
@@ -212,6 +215,7 @@ static struct variant_data variant_qcom = {
.explicit_mclk_control = true,
.qcom_fifo = true,
.qcom_dml = true,
+ .any_blksize = true,
};

static int mmci_card_busy(struct mmc_host *mmc)
@@ -239,10 +243,11 @@ static int mmci_card_busy(struct mmc_host *mmc)
static int mmci_validate_data(struct mmci_host *host,
struct mmc_data *data)
{
+ struct variant_data *variant = host->variant;
+
if (!data)
return 0;
-
- if (!is_power_of_2(data->blksz)) {
+ if (!is_power_of_2(data->blksz) && !variant->any_blksize) {
dev_err(mmc_dev(host->mmc),
"unsupported block size (%d bytes)\n", data->blksz);
return -EINVAL;
@@ -796,7 +801,6 @@ static void mmci_start_data(struct mmci_host *host, struct mmc_data *data)
writel(host->size, base + MMCIDATALENGTH);

blksz_bits = ffs(data->blksz) - 1;
- BUG_ON(1 << blksz_bits != data->blksz);

if (variant->blksz_datactrl16)
datactrl = MCI_DPSM_ENABLE | (data->blksz << 16);
--
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/