Am 09.11.2017 um 11:04 schrieb Sascha Hauer:
> On Wed, Nov 08, 2017 at 04:33:08PM +0100, Oleksij Rempel wrote:
>> this part of code is executed at least 4 times in eMMC probe sequence.
>> Optimizing it is reducing 20-30 msec of boot time.
>>
>> Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
>> ---
>>  drivers/mci/imx-esdhc.c | 14 ++++++++++----
>>  1 file changed, 10 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/mci/imx-esdhc.c b/drivers/mci/imx-esdhc.c
>> index 141d715c90..5c2a3a36e3 100644
>> --- a/drivers/mci/imx-esdhc.c
>> +++ b/drivers/mci/imx-esdhc.c
>> @@ -95,6 +95,7 @@ struct fsl_esdhc_host {
>>  	struct device_d		*dev;
>>  	struct clk		*clk;
>>  	const struct esdhc_soc_data *socdata;
>> +	u32			last_clock;
>>  };
>>  
>>  #define to_fsl_esdhc(mci)	container_of(mci, struct fsl_esdhc_host, mci)
>> @@ -408,7 +409,7 @@ esdhc_send_cmd(struct mci_host *mci, struct mci_cmd *cmd, struct mci_data *data)
>>  
>>  static void set_sysctl(struct mci_host *mci, u32 clock)
>>  {
>> -	int div, pre_div;
>> +	int div, pre_div, i;
>>  	struct fsl_esdhc_host *host = to_fsl_esdhc(mci);
>>  	void __iomem *regs = host->regs;
>>  	int sdhc_clk = clk_get_rate(host->clk);
>> @@ -453,13 +454,17 @@ static void set_sysctl(struct mci_host *mci, u32 clock)
>>  	esdhc_clrsetbits32(regs + SDHCI_CLOCK_CONTROL__TIMEOUT_CONTROL__SOFTWARE_RESET,
>>  			SYSCTL_CLOCK_MASK, clk);
>>  
>> -	wait_on_timeout(10 * MSECOND,
>> -			!(esdhc_read32(regs + SDHCI_PRESENT_STATE) & PRSSTAT_SDSTB));
>> +	for (i = 0; i < 1000; i++) {
>> +		if (esdhc_read32(regs + SDHCI_PRESENT_STATE) & PRSSTAT_SDSTB)
>> +			break;
>> +		udelay(10);
>> +	}
> 
> I hope this hunk doesn't make anything better. If yes, we have a
> problem.

this is the part which need most of the time. Some times it will pass
immediately or will take exactly 10ms.

>>  
>>  	clk = SYSCTL_PEREN | SYSCTL_CKEN;
>>  
>>  	esdhc_setbits32(regs + SDHCI_CLOCK_CONTROL__TIMEOUT_CONTROL__SOFTWARE_RESET,
>>  			clk);
>> +	host->last_clock = clock;
> 
> Instead of storing the desired clock I would rather store the clock we
> actually make from the desired clock. This way we could optimze for
> multiple input frequencies that result in the same actual frequency.
> 
> Also please rename the variable from last_clock to cur_clock.

ok.

-- 
Regards,
Oleksij