Re: [RESEND RFC v2] mmc: Change the max discard sectors and erase response if mmc host supports busy signalling

From: Baolin Wang
Date: Wed Jul 20 2016 - 04:33:17 EST

On 19 July 2016 at 19:57, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> On 13 June 2016 at 10:54, Baolin Wang <baolin.wang@xxxxxxxxxx> wrote:
>> When mmc host HW supports busy signalling (using R1B as response), We
>> shouldn't use 'host->max_busy_timeout' as the limitation when deciding
>> the max discard sectors that we tell the generic BLOCK layer about.
>> Instead, we should pick one preferred erase size as the max discard
>> sectors.
>>
>> If the host controller supports busy signalling and the timeout for
>> the erase operation does not exceed the max_busy_timeout, we should
>> use R1B response. Or we need to prevent the host from doing hw busy
>> detection, which is done by converting to a R1 response instead.
>>
>> Changes since v1:
>> - Remove the 'MMC_CAP_WAIT_WHILE_BUSY' flag checking when deciding
>> the max discard sectors.
>>
>> Signed-off-by: Baolin Wang <baolin.wang@xxxxxxxxxx>
>> ---
>> drivers/mmc/core/core.c | 47 +++++++++++++++++++++++++++++++++++++++--------
>> 1 file changed, 39 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
>> index 8b4dfd4..edd43b1 100644
>> --- a/drivers/mmc/core/core.c
>> +++ b/drivers/mmc/core/core.c
>> @@ -2060,7 +2060,7 @@ static int mmc_do_erase(struct mmc_card *card, unsigned int from,
>> unsigned int to, unsigned int arg)
>> {
>> struct mmc_command cmd = {0};
>> - unsigned int qty = 0;
>> + unsigned int qty = 0, busy_timeout = 0;
>> unsigned long timeout;
>> int err;
>>
>> @@ -2128,8 +2128,23 @@ static int mmc_do_erase(struct mmc_card *card, unsigned int from,
>> memset(&cmd, 0, sizeof(struct mmc_command));
>> cmd.opcode = MMC_ERASE;
>> cmd.arg = arg;
>> - cmd.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC;
>> - cmd.busy_timeout = mmc_erase_timeout(card, arg, qty);
>> + busy_timeout = mmc_erase_timeout(card, arg, qty);
>> + /*
>> + * If the host controller supports busy signalling and the timeout for
>> + * the erase operation does not exceed the max_busy_timeout, we should
>> + * use R1B response. Or we need to prevent the host from doing hw busy
>> + * detection, which is done by converting to a R1 response instead.
>> + */
>> + if ((card->host->max_busy_timeout &&
>> + busy_timeout > card->host->max_busy_timeout) ||
>> + !(card->host->caps & MMC_CAP_WAIT_WHILE_BUSY)) {
>
> sdhci uses the max_busy_timeout, but doesn't always use MMC_CAP_WAIT_WHILE_BUSY.
> I have probably asked Adrian about this before, but right now I can't
> recall why this is the case.
>
> Anyway, I don't think we need to check MMC_CAP_WAIT_WHILE_BUSY at this point.

Make sense.

>
>> + cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC;
>> + cmd.busy_timeout = 0;
>> + } else {
>> + cmd.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC;
>> + cmd.busy_timeout = busy_timeout;
>> + }
>> +
>> err = mmc_wait_for_cmd(card->host, &cmd, 0);
>> if (err) {
>> pr_err("mmc_erase: erase error %d, status %#x\n",
>
> You also need to fix the loop for polling with CMD13 to get the card
> status. In case of when R1B+MMC_CAP_WAIT_WHILE_BUSY is used, the
> polling shall be avoided.

Make sense.

>
> Moreover, the polling loop doesn't care about the earlier used
> calculated erase timeout, but instead defaults to the
> MMC_CORE_TIMEOUT_MS, this is wrong.

Then we should use the earlier calculated erase timeout as the polling
time, right?

>
> Finally, perhaps as an improvement step; I think we shall make use of
> the host's ->card_busy() callback if implemented and when
> R1B+MMC_CAP_WAIT_WHILE_BUSY isn't used. This in favour of issuing
> CMD13 in the loop.
>
> You can see the sequence in __mmc_switch(), we should apply this to
> the erase case as well. Perhaps we can even split up some of the code
> in __mmc_switch() to allow it to re-used for the erase case as well!?

Make sense.

>
>> @@ -2321,23 +2336,39 @@ static unsigned int mmc_do_calc_max_discard(struct mmc_card *card,
>> unsigned int arg)
>> {
>> struct mmc_host *host = card->host;
>> - unsigned int max_discard, x, y, qty = 0, max_qty, timeout;
>> + unsigned int max_discard, x, y, qty = 0, max_qty, min_qty, timeout;
>> unsigned int last_timeout = 0;
>>
>> - if (card->erase_shift)
>> + if (card->erase_shift) {
>> max_qty = UINT_MAX >> card->erase_shift;
>> - else if (mmc_card_sd(card))
>> + min_qty = card->pref_erase >> card->erase_shift;
>> + } else if (mmc_card_sd(card)) {
>> max_qty = UINT_MAX;
>> - else
>> + min_qty = card->pref_erase;
>> + } else {
>> max_qty = UINT_MAX / card->erase_size;
>> + min_qty = card->pref_erase / card->erase_size;
>> + }
>>
>> /* Find the largest qty with an OK timeout */
>
> This comment needs to be updated.

OK.

>
>> do {
>> y = 0;
>> for (x = 1; x && x <= max_qty && max_qty - x >= qty; x <<= 1) {
>> timeout = mmc_erase_timeout(card, arg, qty + x);
>> - if (timeout > host->max_busy_timeout)
>> + /*
>> + * We should not only use 'host->max_busy_timeout' as
>> + * the limitation when deciding the max discard sectors.
>> + * We should set a balance value to improve the erase
>> + * speed, and it can not get too long timeout at the
>> + * same time.
>
> I am not really sure I understand this comment. Could you try to
> elaborate on what value that will be picked when the max_busy_timeout
> isn't just as the limiter.

It means we can get the value at least is 'min_qty' no matter what
size of max_busy_timeout. But if the max_busy_timeout is large enough
for 'min_qty' size, then we can continue to increase the max discard
sectors.

>
>> + *
>> + * Here we set 'card->pref_erase' as the minimal discard
>> + * sectors when deciding the max discard sectors.
>> + */
>> + if (qty + x > min_qty &&
>> + timeout > host->max_busy_timeout)
>> break;
>> +
>> if (timeout < last_timeout)
>> break;
>> last_timeout = timeout;
>> --
>> 1.7.9.5
>>
>
> Again, sorry for the delay in response!

It's okay:). Thanks for your comments.

--
Baolin.wang
Best Regards