Re: [PATCH V5 1/3] mmc: mmci: add hardware busy timeout feature

From: Ludovic BARRE
Date: Wed Sep 04 2019 - 11:05:02 EST


hi Ulf

On 8/26/19 1:39 PM, Ulf Hansson wrote:
On Tue, 13 Aug 2019 at 12:00, Ludovic Barre <ludovic.Barre@xxxxxx> wrote:

From: Ludovic Barre <ludovic.barre@xxxxxx>

In some variants, the data timer starts and decrements
when the DPSM enters in Wait_R or Busy state
(while data transfer or MMC_RSP_BUSY), and generates a
data timeout error if the counter reach 0.

I don't quite follow here, sorry. Can you please try to elaborate on
the use case(s) more exactly?

For example, what happens when a data transfer has just finished (for
example when MCI_DATAEND has been received) and we are going to send a
CMD12 to stop it? In this case the CMD12 has the MMC_RSP_BUSY flag
set.


example with cmd25 (write multi block)
mmci_request
- mmci_start_data
set MMCIDATATIMER, MMCIDATALENGTH, MMCIMASK0
- mmci_start_command:
set MMCIARGUMENT, MMCICOMMAND (cmd25)

mmci_irq:
- irq MCI_CMDRESPEND
- irq MCI_DATAEND
- send cmd12 => mmci_start_command(host->stop_abort or data->stop)
these cmds have flag rsp_busy and no data associate
host->cmd = cmd (host->stop_abort or data->stop) for next irq

mmci_irq:
- irq MCI_CMDRESPEND
- irq BUSYD0END
- mmci_request_end

Another example is the CMD5, which has no data with it.


-Define max_busy_timeout (in ms) according to clock.
-Set data timer register if the command has rsp_busy flag.
If busy_timeout is not defined by framework, the busy
length after Data Burst is defined as 1 second
(refer: 4.6.2.2 Write of sd specification part1 v6-0).

One second is not sufficient for all operations, like ERASE for
example. However, I understand that you want to pick some value, as a
safety. I guess that's fine.

I am thinking that if the command has the MMC_RSP_BUSY flag set, the
core should really provide a busy timeout for it. That said, maybe the
host driver should splat a WARN in case there is not busy timeout
specified.

Today, I just see a busy_timeout not defined on write request.
On erase request, the timeout is defined in function mmc_do_erase.
In core, there are several paths to done a write request, and I not
be sure to fix all. For safety, I preferred fix with the max value of
write request.


-Add MCI_DATATIMEOUT error management in mmci_cmd_irq.

Signed-off-by: Ludovic Barre <ludovic.barre@xxxxxx>
---
drivers/mmc/host/mmci.c | 37 ++++++++++++++++++++++++++++++++-----
drivers/mmc/host/mmci.h | 3 +++
2 files changed, 35 insertions(+), 5 deletions(-)

diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
index c37e70dbe250..c50586540765 100644
--- a/drivers/mmc/host/mmci.c
+++ b/drivers/mmc/host/mmci.c
@@ -1075,6 +1075,7 @@ static void
mmci_start_command(struct mmci_host *host, struct mmc_command *cmd, u32 c)
{
void __iomem *base = host->base;
+ unsigned long long clks = 0;

dev_dbg(mmc_dev(host->mmc), "op %02x arg %08x flags %08x\n",
cmd->opcode, cmd->arg, cmd->flags);
@@ -1097,6 +1098,19 @@ mmci_start_command(struct mmci_host *host, struct mmc_command *cmd, u32 c)
else
c |= host->variant->cmdreg_srsp;
}
+
+ if (host->variant->busy_timeout && !host->mrq->data) {

Suppose this is a CMD12 command, having the MMC_RSP_BUSY flag set. The
command would then be sent to stop the transmission and then
host->mrq->data would also be set.

Sorry, it's a mistake introduce by v5.
I would keep the clear of datatimer when is not needed
(no data & no rsp busy, see below). But on cmd23 (set_block_count) with datactrl_first variant property the datatimer should be protected.

To simplify and fix the code, I will remove the clear of datatimer
when there is no data & no rsp busy.

- if (host->variant->busy_timeout && !host->mrq->data) {
+ if (host->variant->busy_timeout && !cmd->flags & MMC_RSP_BUSY) {
+ ....
+ writel_relaxed(clks, host->base + MMCIDATATIMER);
+ }


If I recall earlier what you stated about the new sdmmc variant, the
CMD12 is needed to exit the DPSM. Hence don't you need to re-program a
new value for the MMCIDATATIMER register for this scenario?

+ if (cmd->flags & MMC_RSP_BUSY) {
+ if (!cmd->busy_timeout)
+ cmd->busy_timeout = 1000;
+
+ clks = (unsigned long long)cmd->busy_timeout;
+ clks *= host->cclk;

Any problems with putting the above on one line?

No, it was just to not exceed 80 characters.


+ do_div(clks, MSEC_PER_SEC);
+ }
+ writel_relaxed(clks, host->base + MMCIDATATIMER);

This is writing zero to MMCIDATATIMER in case the MMC_RSP_BUSY isn't
set, is that on purpose?

It was to clear the datatimer when the command has
no data & no rsp_busy. This allowed to look if the datatimer
was used and not correctly set with the right value (with datatimeout).

Like said above, I will remove this and set datatimer only
on rsp_busy flag.


+ }
+
if (/*interrupt*/0)
c |= MCI_CPSM_INTERRUPT;

@@ -1203,6 +1217,7 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
{
void __iomem *base = host->base;
bool sbc, busy_resp;
+ u32 err_msk;

if (!cmd)
return;
@@ -1215,8 +1230,12 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
* handling. Note that we tag on any latent IRQs postponed
* due to waiting for busy status.
*/
- if (!((status|host->busy_status) &
- (MCI_CMDCRCFAIL|MCI_CMDTIMEOUT|MCI_CMDSENT|MCI_CMDRESPEND)))
+ err_msk = MCI_CMDCRCFAIL | MCI_CMDTIMEOUT;

You might as well move the initial assignment of err_msk to the its
declaration above.


OK, thx

+ if (host->variant->busy_timeout && busy_resp)
+ err_msk |= MCI_DATATIMEOUT;
+
+ if (!((status | host->busy_status) &
+ (err_msk | MCI_CMDSENT | MCI_CMDRESPEND)))
return;

/* Handle busy detection on DAT0 if the variant supports it. */
@@ -1235,8 +1254,7 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
* while, to allow it to be set, but tests indicates that it
* isn't needed.
*/
- if (!host->busy_status &&
- !(status & (MCI_CMDCRCFAIL|MCI_CMDTIMEOUT)) &&
+ if (!host->busy_status && !(status & err_msk) &&
(readl(base + MMCISTATUS) & host->variant->busy_detect_flag)) {

writel(readl(base + MMCIMASK0) |
@@ -1290,6 +1308,9 @@ mmci_cmd_irq(struct mmci_host *host, struct mmc_command *cmd,
cmd->error = -ETIMEDOUT;
} else if (status & MCI_CMDCRCFAIL && cmd->flags & MMC_RSP_CRC) {
cmd->error = -EILSEQ;
+ } else if (host->variant->busy_timeout && busy_resp &&
+ status & MCI_DATATIMEOUT) {
+ cmd->error = -ETIMEDOUT;
} else {
cmd->resp[0] = readl(base + MMCIRESPONSE0);
cmd->resp[1] = readl(base + MMCIRESPONSE1);
@@ -1948,6 +1969,8 @@ static int mmci_probe(struct amba_device *dev,
* Enable busy detection.
*/
if (variant->busy_detect) {
+ u32 max_busy_timeout = 0;
+
mmci_ops.card_busy = mmci_card_busy;
/*
* Not all variants have a flag to enable busy detection
@@ -1957,7 +1980,11 @@ static int mmci_probe(struct amba_device *dev,
mmci_write_datactrlreg(host,
host->variant->busy_dpsm_flag);
mmc->caps |= MMC_CAP_WAIT_WHILE_BUSY;
- mmc->max_busy_timeout = 0;
+
+ if (variant->busy_timeout)
+ max_busy_timeout = ~0UL / (mmc->f_max / MSEC_PER_SEC);

It looks like the max busy timeout is depending on the current picked
clock rate, right?

In such case, perhaps it's better to update mmc->max_busy_timeout as
part of the ->set_ios() callback, as it's from there the clock rate
gets updated. Or what do you think?

yes, it's possible


+
+ mmc->max_busy_timeout = max_busy_timeout;
}

/* Prepare a CMD12 - needed to clear the DPSM on some variants. */
diff --git a/drivers/mmc/host/mmci.h b/drivers/mmc/host/mmci.h
index 833236ecb31e..d8b7f6774e8f 100644
--- a/drivers/mmc/host/mmci.h
+++ b/drivers/mmc/host/mmci.h
@@ -287,6 +287,8 @@ struct mmci_host;
* @signal_direction: input/out direction of bus signals can be indicated
* @pwrreg_clkgate: MMCIPOWER register must be used to gate the clock
* @busy_detect: true if the variant supports busy detection on DAT0.
+ * @busy_timeout: true if the variant starts data timer when the DPSM
+ * enter in Wait_R or Busy state.
* @busy_dpsm_flag: bitmask enabling busy detection in the DPSM
* @busy_detect_flag: bitmask identifying the bit in the MMCISTATUS register
* indicating that the card is busy
@@ -333,6 +335,7 @@ struct variant_data {
u8 signal_direction:1;
u8 pwrreg_clkgate:1;
u8 busy_detect:1;
+ u8 busy_timeout:1;
u32 busy_dpsm_flag;
u32 busy_detect_flag;
u32 busy_detect_mask;
--
2.17.1


Kind regards
Uffe