Re: [PATCH 1/2] spi/rockchip: fix bug that case spi can't go as fast as slave request

[addy ke <addy.ke@rock-chips.com>]

On 2014/10/16 17:34, Mark Brown wrote:
> On Thu, Oct 16, 2014 at 05:16:02PM +0800, addy ke wrote:
>> On 2014/10/15 21:04, Mark Brown wrote:
>>> On Wed, Oct 15, 2014 at 07:25:49PM +0800, Addy Ke wrote:
> 
>>>> +	if (WARN_ON(rs->speed > MAX_SCLK_OUT))
>>>> +		rs->speed = MAX_SCLK_OUT;
> 
>>>> +	/* the minimum divsor is 2 */
>>>> +	if (rs->max_freq < 2 * rs->speed) {
>>>> +		clk_set_rate(rs->spiclk, 2 * rs->speed);
>>>> +		rs->max_freq = clk_get_rate(rs->spiclk);
>>>> +	}
> 
>>> I'll apply this but you should be checking the return code from
>>> clk_set_rate() here, please send a followup patch doing that.  It might
> 
>> If clk_set_rate return error, do I only put dev_warn here or return error value to spi core?
> 
> It'd be better to return an error if we need to set the rate and can't
> do it.
> 
>>> also be worth consdering just setting the rate unconditionally here, it
>>> seems like it should make things simpler.
> 
>> I think we need.
>> If we set the rate unconditionally here, clk_set_rate() will be executed in each spi transfer.
> 
> Is that really such a high cost?
> 
Not high cost, but  I think if the default spi_clk is enough, we do not need to set spi_clk again.

Maybe we can only set spi_clk as (2 * MAX_SCLK_OUT) in probe().