Re: [PATCH] mmc: sdhci: don't limit discard timeout by data line timeout

[Vladimir Zapolskiy <vz@mleia.com>]

On 11/27/13 10:21, Adrian Hunter wrote:
> On 26/11/13 18:33, Vladimir Zapolskiy wrote:
>> On 11/26/13 11:04, Adrian Hunter wrote:
>>> On 22/11/13 17:21, Vladimir Zapolskiy wrote:
>>>> On 22.11.2013 16:04, Adrian Hunter wrote:
>>>>> On 22/11/13 15:50, Vladimir Zapolskiy wrote:
>>>>>> On 22.11.2013 14:04, Adrian Hunter wrote:
>>>>>>> On 22/11/13 14:24, Vladimir Zapolskiy wrote:
>>>>>>>> On 22.11.2013 12:38, Adrian Hunter wrote:
>>>>>>>>> On 21/11/13 17:07, Vladimir Zapolskiy wrote:
>>>>>>>>>> JEDEC specification defines quite high erase timeout value for 300ms
>>>>>>>>>> multiplied by erase group number, and SD Host Controller specification
>>>>>>>>>> data line timeout may be much less, e.g. 2^13 / 52MHz ~ 160us.
>>>>>>>>>>
>>>>>>>>>>>        From block layer and MMC perfromance perspective it is desirable
>>>>>>>>>>> that
>>>>>>>>>> millions of erase groups are discarded at once, so there is no much
>>>>>>>>>> sense to limit maximum erase timeout by data line timeout, if a
>>>>>>>>>> controller handles correctly erase operation without indication of
>>>>>>>>>> data line timeout.
>>>>>>>>>
>>>>>>>>> Would you explain that some more.  Do you mean that:
>>>>>>>>>         a) it does not have a timeout
>>>>>>>>
>>>>>>>> JEDEC defines a timeout on erase/trim operations, also in
>>>>>>>> drivers/mmc/core/core.c
>>>>>>>> there is a reasonable enough 10 minutes limitation for discard
>>>>>>>> operations.
>>>>>>>>
>>>>>>>>>         b) it has a timeout which is less than the timeout specified
>>>>>>>>> by the
>>>>>>>>> standard but the operation nevertheless completes
>>>>>>>>
>>>>>>>> SDHC data line timeout is enormously less than erase group timeout, and
>>>>>>>> trivial testing shows that those two timeouts are independent, probably
>>>>>>>> except some particular cases of controllers not known before commits
>>>>>>>> 58d1246db3 and e056a1b5b.
>>>>>>>>
>>>>>>>> According to the currently implemented logic, mmc_do_erase() commonly is
>>>>>>>> instructed to discard 1-2 erase groups at maximum, however it tends
>>>>>>>> to be
>>>>>>>> capable to successfully discard millions of erase groups at once
>>>>>>>> ignoring
>>>>>>>> that SDHC data line timeout limitation.
>>>>>>>>
>>>>>>>
>>>>>>> You seem to be trying to say that the SDHCI spec. says that the host
>>>>>>> controller does not timeout erase operations or uses a different timeout
>>>>>>> than the one programmed in the "Timeout Control Register".  Where is
>>>>>>> that is
>>>>>>> the SDHCI spec?
>>>>>>
>>>>>> According to the spec a host controller timeouts erase operations like any
>>>>>> other R1B command.
>>>>>>
>>>>>> So in your opinion, should there be SDHCI_QUIRK_BROKEN_TIMEOUT_VAL instead
>>>>>> of the new quirk?
>>>>>
>>>>> I don't understand how SDHCI_QUIRK_BROKEN_TIMEOUT_VAL would help.  It just
>>>>> sets the timeout to maximum but max_discard_to is the maximum timeout.
>>>>
>>>> Here I meant to do something like:
>>>>
>>>>       if (host->quirks&   SDHCI_QUIRK_BROKEN_TIMEOUT_VAL)
>>>>           mmc->max_discard_to = 0;
>>>>
>>>> Again I'm not sure that this applies well to all
>>>> SDHCI_QUIRK_BROKEN_TIMEOUT_VAL
>>>> controllers, therefore a new quirk might be better.
>>>>
>>>>> As I understand it you don't want to limit the discard size, either because
>>>>> your controller does not timeout, or because you are happy that the maximum
>>>>> timeout is enough for your users and their use-cases.
>>>>>
>>>>> If that is the case then the original patch just needs the quirk the other
>>>>> way around. i.e.
>>>>>
>>>>>        if (host->quirks2&    SDHCI_QUIRK2_NO_DISCARD_LIMIT)
>>>>>            mmc->max_discard_to = 0;
>>>>>        else
>>>>>            mmc->max_discard_to = (1<<    27) / host->timeout_clk;
>>>>
>>>> This suits me fine, thanks for review, and I'll resend a change based on
>>>> this.
>>>>
>>>> Also I'd like to pay your attention to (1<<   27) / host->timeout_clk part of
>>>> calculation, following the spec it might be better to account the actual
>>>> value of Data Timeout Counter, otherwise a controller may get unintentional
>>>> Data Timeout Error pretty soon. Please correct me, if I'm mistaken here.
>>>
>>> Not sure what you mean.  max_discard_to is the maximum timeout (in
>>> milliseconds) that the host controller supports.  The intent is to limit
>>> erase operations to ones that have a timeout that is less than or equal to
>>> that.
>>
>> That's clear. But it's not obvious why do you prefer (1<<  27) numerator
>> instead
>> of secure (1<<  13) or (1<<  (13 + sdhci_readl(host, SDHCI_TIMEOUT_CONTROL))).
>
> The maximum value of "Data Timeout Counter Value" in "Timeout Control
> Register" is 14 and the maximum timeout is therefore (1<<  27).

So, from this perspective I assume this is a potential theoretical maximum
timeout for a controller, which may be 16384 times more than the maximum
guaranteed timeout before getting a DAT timeout. Why is the theoretical 
maximum
supposed to be used in calculations of a guaranteed discard operation 
timeout
instead of promised DAT timeout by a controller?

>>
>>> Currently, the limit gets applied by the block layer before the mmc layer is
>>> involved so there is no possibility to take the actual timeout into account.
>>>    However if you have erase_group_def set, then it won't make any difference
>>> i.e. the limit will be the same.
>>>
>>>>
>>>>>>
>>>>>>>>>>
>>>>>>>>>> Potentially the change may break some of the SDHCs on discard of mmc,
>>>>>>>>>> and for backward compatibility a new quirk is introduced, which is NOT
>>>>>>>>>> set by default.
>>>>>>>>>
>>>>>>>>> It sounds to me that what you want to do is not standard so the quirk
>>>>>>>>> should
>>>>>>>>> be the other way around.
>>>>>>>>
>>>>>>>> Please take a look at commits 58d1246db3 and e056a1b5b, I'd be glad, if
>>>>>>>> you
>>>>>>>> could elaborate to which "some host controllers" the quirk in my
>>>>>>>> definition
>>>>>>>> applies, I believe all other host controllers present at that time in
>>>>>>>> drivers/mmc/host/* are capable to discard without introduced limitation.
>>>>>>>
>>>>>>> "some host controllers" == SDHCI i.e. to all of the ones you are applying
>>>>>>> the change.
>>>>>>>
>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Vladimir Zapolskiy<vladimir_zapolskiy@mentor.com>
>>>>>>>>>> Reported-by: Ed Sutter<ed.sutter@alcatel-lucent.com>
>>>>>>>>>> Cc: Chris Ball<cjb@laptop.org>
>>>>>>>>>> Cc: Adrian Hunter<adrian.hunter@intel.com>
>>>>>>>>>> ---
>>>>>>>>>>       drivers/mmc/host/sdhci.c  |    5 ++++-
>>>>>>>>>>       include/linux/mmc/sdhci.h |    1 +
>>>>>>>>>>       2 files changed, 5 insertions(+), 1 deletion(-)
>>>>>>>>>>
>>>>>>>>>> diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
>>>>>>>>>> index bd8a098..b1fdddb 100644
>>>>>>>>>> --- a/drivers/mmc/host/sdhci.c
>>>>>>>>>> +++ b/drivers/mmc/host/sdhci.c
>>>>>>>>>> @@ -2930,7 +2930,10 @@ int sdhci_add_host(struct sdhci_host *host)
>>>>>>>>>>           if (host->quirks&      SDHCI_QUIRK_DATA_TIMEOUT_USES_SDCLK)
>>>>>>>>>>               host->timeout_clk = mmc->f_max / 1000;
>>>>>>>>>>
>>>>>>>>>> -    mmc->max_discard_to = (1<<      27) / host->timeout_clk;
>>>>>>>>>> +    if (host->quirks2&      SDHCI_QUIRK2_DATA_TIMEOUT_ON_DISCARD)
>>>>>>>>>> +        mmc->max_discard_to = (1<<      27) / host->timeout_clk;
>>>>>>>>>> +    else
>>>>>>>>>> +        mmc->max_discard_to = 0;
>>>>>>>>>>
>>>>>>>>>>           mmc->caps |= MMC_CAP_SDIO_IRQ | MMC_CAP_ERASE |
>>>>>>>>>> MMC_CAP_CMD23;
>>>>>>>>>>
>>>>>>>>>> diff --git a/include/linux/mmc/sdhci.h b/include/linux/mmc/sdhci.h
>>>>>>>>>> index 3e781b8..e7f6bd2 100644
>>>>>>>>>> --- a/include/linux/mmc/sdhci.h
>>>>>>>>>> +++ b/include/linux/mmc/sdhci.h
>>>>>>>>>> @@ -98,6 +98,7 @@ struct sdhci_host {
>>>>>>>>>>       #define SDHCI_QUIRK2_CARD_ON_NEEDS_BUS_ON        (1<<4)
>>>>>>>>>>       /* Controller has a non-standard host control register */
>>>>>>>>>>       #define SDHCI_QUIRK2_BROKEN_HOST_CONTROL        (1<<5)
>>>>>>>>>> +#define SDHCI_QUIRK2_DATA_TIMEOUT_ON_DISCARD        (1<<6)
>>>>>>>>>>
>>>>>>>>>>           int irq;        /* Device IRQ */
>>>>>>>>>>           void __iomem *ioaddr;    /* Mapped address */
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>
>>>>
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html