Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <95116fcd-a374-d0c7-47f3-10921325e331@intel.com>
Date:   Fri, 29 Jul 2022 13:17:48 +0300
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Firefox/91.0 Thunderbird/91.11.0
Subject: Re: [PATCH] mmc: block: Dont report successful writes with errors
Content-Language: en-US
To:     Christian Loehle <CLoehle@hyperstone.com>,
        "ulf.hansson@linaro.org" <ulf.hansson@linaro.org>,
        Avri Altman <Avri.Altman@wdc.com>,
        "andriy.shevchenko@linux.intel.com" 
        <andriy.shevchenko@linux.intel.com>,
        Linux MMC List <linux-mmc@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
References: <ca06b94aa48a484d965744e64e17a4ef@hyperstone.com>
 <992cc198-6c7c-977a-1af8-56665e939cc9@intel.com>
 <a66fe29560f545e18b91af57f4d0dddb@hyperstone.com>
From:   Adrian Hunter <adrian.hunter@intel.com>
Organization: Intel Finland Oy, Registered Address: PL 281, 00181 Helsinki,
 Business Identity Code: 0357606 - 4, Domiciled in Helsinki
In-Reply-To: <a66fe29560f545e18b91af57f4d0dddb@hyperstone.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 23/07/22 18:08, Christian Loehle wrote:
>> On 19/07/22 18:34, Christian Loehle wrote:
>>> Be as conservative about successful write reporting to the block layer 
>>> for SPI as with normal SD and MMC.
>>> That means on any errors bytes_xfered is ignored and the whole write 
>>> must be repeated.
>>>
>>> Signed-off-by: Christian Loehle <cloehle@hyperstone.com>
>>> ---
>>>  drivers/mmc/core/block.c | 6 +++++-
>>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/mmc/core/block.c b/drivers/mmc/core/block.c index 
>>> f4a1281658db..63d1c05582a9 100644
>>> --- a/drivers/mmc/core/block.c
>>> +++ b/drivers/mmc/core/block.c
>>> @@ -1765,8 +1765,12 @@ static bool mmc_blk_status_error(struct request *req, u32 status)
>>>  	struct mmc_queue *mq = req->q->queuedata;
>>>  	u32 stop_err_bits;
>>>  
>>> +	/*
>>> +	 * Either write timed out during busy and data->error is set
>>> +	 * or we actually received a valid R2 and check for error bits.
>>> +	 */
>>>  	if (mmc_host_is_spi(mq->card->host))
>>> -		return false;
>>> +		return brq->data.error || !!status;
>>
>> This function is for checking status, so brq->data.error does not belong here.  Also it would be more readable to use a define e.g.
>>
>> 		return status & SPI_R2_ERRORS;
>>
>> I think clearing bytes_xfered for SPI brq->data.error should be a separate patch, and you would need to explain a bit more for that case too.
> 
> I understand that, but there is no way of checking status in SPI mode.
> The behavior of mmc block is only changed in a minor way here anyway, that is, checking for status is done one more time than before.
> If brq->data.error is set directly after a write e.g. then bytes_xfered is already 0.

The expectation is that the driver sets bytes_xfered correctly,
based controller errors.  The driver is not expected to check for
status errors, hence in that case the bytes_xfered is set to 0 by
error recovery.

> My intention was mostly to improve on the flow of the recovery and get the mmc_is_host_spi out of there for the most part with future patches.
> After all it feels weird to do a single step read retry before ensuring a fix_state, and I ran into that quite often.
> Unfortunately, I now realized that fix_state cannot properly be implemented within the spec or even real-world card's behavior and I won't be taking this further.
> The best attempt I came up with is doing a loop of CMD12 and CMD13 in SPI and if CMD12 was ILLEGAL and CMD12 has no bits set, state is fixed.
> But CMD12 is only defined for multiple block transfers in SPI and cards treat it differently on e.g. CMD17 transfers.
> Instead I would just do a soft reset for SPI and retry and maybe increase the read timeout of 100ms which larger SD cards can fail sometimes.
> Anyway since SPI initialization is quite fast, especially for soft resets there is likely no recovery to beat that performance-wise.
> I will send an RFC for the soft reset in the coming days.

That sounds like it would be a good improvement to have.

> If not I would at least add the !mmc_is_host_spi condition for calling mmc_blk_status_error to make it a bit more clear that this function does do what is intended for SPI cards.

I am not sure what you mean.  Isn't it OK to check CMD13 response for SPI?