Re: [PATCH v3 07/11] mmc: sdhci: Program a relatively accurate SW timeout value

[Kishon Vijay Abraham I]

Hi Adrian,

On Monday 19 March 2018 03:49 PM, Kishon Vijay Abraham I wrote:
> Hi Adrian,
> 
> On Monday 19 March 2018 03:30 PM, Adrian Hunter wrote:
>> On 19/03/18 11:20, Kishon Vijay Abraham I wrote:
>>> Hi Adrian,
>>>
>>> On Friday 16 March 2018 07:51 PM, Adrian Hunter wrote:
>>>> On 16/03/18 08:29, Kishon Vijay Abraham I wrote:
>>>>> Hi,
>>>>>
>>>>> On Thursday 15 March 2018 06:43 PM, Adrian Hunter wrote:
>>>>>> On 07/03/18 15:20, Kishon Vijay Abraham I wrote:
>>>>>>> sdhci has a 10 second timeout to catch devices that stop responding.
>>>>>>> Instead of programming 10 second arbitrary value, calculate the total time
>>>>>>> it would take for the entire transfer to happen and program the timeout
>>>>>>> value accordingly.
>>>>>>>
>>>>>>> Signed-off-by: Kishon Vijay Abraham I <kishon@xxxxxx>
>>>>>>> ---
>>>>>>>  drivers/mmc/host/sdhci.c | 47 ++++++++++++++++++++++++++++++++++++++++-------
>>>>>>>  drivers/mmc/host/sdhci.h | 10 ++++++++++
>>>>>>>  2 files changed, 50 insertions(+), 7 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
>>>>>>> index 1dd117cbeb6e..baab67bfa39b 100644
>>>>>>> --- a/drivers/mmc/host/sdhci.c
>>>>>>> +++ b/drivers/mmc/host/sdhci.c
>>>>>>> @@ -709,6 +709,36 @@ static u32 sdhci_sdma_address(struct sdhci_host *host)
>>>>>>>  		return sg_dma_address(host->data->sg);
>>>>>>>  }
>>>>>>>  
>>>>>>> +static void sdhci_calc_sw_timeout(struct sdhci_host *host,
>>>>>>> +				  struct mmc_command *cmd,
>>>>>>> +				  unsigned int target_timeout)
>>>>>>> +{
>>>>>>> +	struct mmc_data *data = cmd->data;
>>>>>>> +	struct mmc_host *mmc = host->mmc;
>>>>>>> +	u64 transfer_time;
>>>>>>> +	struct mmc_ios *ios = &mmc->ios;
>>>>>>> +	unsigned char bus_width = 1 << ios->bus_width;
>>>>>>> +	unsigned int blksz;
>>>>>>> +	unsigned int freq;
>>>>>>> +
>>>>>>> +	if (data) {
>>>>>>> +		blksz = data->blksz;
>>>>>>> +		freq = host->mmc->actual_clock ? : host->clock;
>>>>>>> +		transfer_time = (u64)blksz * NSEC_PER_SEC * (8 / bus_width);
>>>>>>> +		do_div(transfer_time, freq);
>>>>>>> +		/* multiply by '2' to account for any unknowns */
>>>>>>> +		transfer_time = transfer_time * 2;
>>>>>>> +		/* calculate timeout for the entire data */
>>>>>>> +		host->data_timeout = (data->blocks * ((target_timeout *
>>>>>>> +						       NSEC_PER_USEC) +
>>>>>>> +						       transfer_time));
>>>>>>
>>>>>> (target_timeout * NSEC_PER_USEC) might be 32-bit and therefore overflow
>>>>>> for timeouts greater than about 4 seconds.
>>>>>>
>>>>>>> +	} else {
>>>>>>> +		host->data_timeout = (u64)target_timeout * NSEC_PER_USEC;
>>>>>>> +	}
>>>>>>> +
>>>>>>> +	host->data_timeout += MMC_CMD_TRANSFER_TIME;
>>>>>>
>>>>>> Need to allow for target_timeout == 0 so:
>>>>>>
>>>>>> 	if (host->data_timeout)
>>>>>> 		host->data_timeout += MMC_CMD_TRANSFER_TIME;
>>>>>>
>>>>>>> +}
>>>>>>> +
>>>>>>>  static u8 sdhci_calc_timeout(struct sdhci_host *host, struct mmc_command *cmd)
>>>>>>>  {
>>>>>>>  	u8 count;
>>>>>>> @@ -766,6 +796,7 @@ static u8 sdhci_calc_timeout(struct sdhci_host *host, struct mmc_command *cmd)
>>>>>>>  		if (count >= 0xF)
>>>>>>>  			break;
>>>>>>>  	}
>>>>>>> +	sdhci_calc_sw_timeout(host, cmd, target_timeout);
>>>>>>
>>>>>> If you make the changes I suggest for patch 6, then this would
>>>>>> move sdhci_calc_sw_timeout() into sdhci_set_timeout().
>>>>>>
>>>>>> I suggest you factor out the target_timeout calculation e.g.
>>>>>>
>>>>>> static unsigned int sdhci_target_timeout(struct sdhci_host *host,
>>>>>> 					 struct mmc_command *cmd,
>>>>>> 					 struct mmc_data *data)
>>>>>> {
>>>>>> 	unsigned int target_timeout;
>>>>>>
>>>>>> 	/* timeout in us */
>>>>>> 	if (!data)
>>>>>> 		target_timeout = cmd->busy_timeout * 1000;
>>>>>> 	else {
>>>>>> 		target_timeout = DIV_ROUND_UP(data->timeout_ns, 1000);
>>>>>> 		if (host->clock && data->timeout_clks) {
>>>>>> 			unsigned long long val;
>>>>>>
>>>>>> 			/*
>>>>>> 			 * data->timeout_clks is in units of clock cycles.
>>>>>> 			 * host->clock is in Hz.  target_timeout is in us.
>>>>>> 			 * Hence, us = 1000000 * cycles / Hz.  Round up.
>>>>>> 			 */
>>>>>> 			val = 1000000ULL * data->timeout_clks;
>>>>>> 			if (do_div(val, host->clock))
>>>>>> 				target_timeout++;
>>>>>> 			target_timeout += val;
>>>>>> 		}
>>>>>> 	}
>>>>>>
>>>>>> 	return target_timeout;
>>>>>> }
>>>>>>
>>>>>> And call it from sdhci_calc_sw_timeout()
>>>>>>
>>>>>>>  
>>>>>>>  	return count;
>>>>>>>  }
>>>>>>> @@ -1175,13 +1206,6 @@ void sdhci_send_command(struct sdhci_host *host, struct mmc_command *cmd)
>>>>>>>  		mdelay(1);
>>>>>>>  	}
>>>>>>>  
>>>>>>> -	timeout = jiffies;
>>>>>>> -	if (!cmd->data && cmd->busy_timeout > 9000)
>>>>>>> -		timeout += DIV_ROUND_UP(cmd->busy_timeout, 1000) * HZ + HZ;
>>>>>>> -	else
>>>>>>> -		timeout += 10 * HZ;
>>>>>>> -	sdhci_mod_timer(host, cmd->mrq, timeout);
>>>>>>> -
>>>>>>>  	host->cmd = cmd;
>>>>>>>  	if (sdhci_data_line_cmd(cmd)) {
>>>>>>>  		WARN_ON(host->data_cmd);
>>>>>>> @@ -1221,6 +1245,15 @@ void sdhci_send_command(struct sdhci_host *host, struct mmc_command *cmd)
>>>>>>>  	    cmd->opcode == MMC_SEND_TUNING_BLOCK_HS200)
>>>>>>>  		flags |= SDHCI_CMD_DATA;
>>>>>>>  
>>>>>>> +	timeout = jiffies;
>>>>>>> +	if (host->data_timeout > 0) {
>>>>>>
>>>>>> This can be just:
>>>>>>
>>>>>> 	if (host->data_timeout) {
>>>>>>
>>>>>>> +		timeout += nsecs_to_jiffies(host->data_timeout);
>>>>>>> +		host->data_timeout = 0;
>>>>>>
>>>>>> It would be better to initialize host->data_timeout = 0 at the top of
>>>>>> sdhci_prepare_data().
>>>>>>
>>>>>> Also still need:
>>>>>>
>>>>>> 	else if (!cmd->data && cmd->busy_timeout > 9000) {
>>>>>> 		timeout += DIV_ROUND_UP(cmd->busy_timeout, 1000) * HZ + HZ;
>>>>>
>>>>> sdhci_calc_sw_timeout should have calculated the timeout for this case too no?
>>>>
>>>> Yes, but I was thinking you would only calculate when it was needed.
>>>
>>> I feel since we would have anyways calculated data_timeout, we should use that
>>> instead unless you see a problem with that.
>>
>> I would prefer not to calculate data_timeout when a hardware timeout is
>> being used.
>>
> 
> That differs from what I had thought. This patch tries to program a relatively
> accurate SW timeout value (for data_timer) irrespective of whether hardware
> timeout is used or not. This only tries to change the 10 Sec SW timeout value
> programmed for all data transfer commands.

IMHO since we calculate the worst case timeout value we should be using that
for all cases where we are able to calculate the timeout value so that we don't
give a too high or too low timeout value. Let me know If this sounds okay to you.

Thanks
Kishon