Re: [PATCH] mmc: pxamci: prepare and unprepare the clocks

[Robert Jarzmik <robert.jarzmik@free.fr>]

Ulf Hansson <ulf.hansson@linaro.org> writes:

> I understand your concern, but I don't see why there should be any
> major difference in clock management code (clk tree wise), due to this
> patch. It worked before, so likely it will work now!?

It will ony work *differently*, it will change the clock management. It won't
break, but again it's *not* the purpose of the patch.  The patch is aimed at
removing a warning.

As for the clock management, it will the change the behaviour :
Let's see the current clock management :
 - pxamci_probe()
   => clock is disabled
   mmc_add_host()
     mmc_start_host()
       mmc_power_up() (as pxamci is unaware of caps2)
       Here the comment of the function is (drivers/mmc/core/core.c:1534):
       "First, we enable power to the card without the clock running"
       => this won't be true if the clock is enabled in pxamci_probe()
       mmc_set_ios(host, host->ios.clock=host->f_init)
         pxamci_set_ios()
           clk_enable()
             => here the clock is enabled, enable_count=1
       mmc_host_clk_release()
         pxamci_set_ios()
             => here the clock is disabled, enable_count=0

Let's see the your proposal clock management :
 - pxamci_probe()
   => clock is enabled
   mmc_add_host()
     mmc_start_host()
       mmc_power_up() (as pxamci is unaware of caps2)
       mmc_set_ios(host, host->ios.clock=host->f_init)
         pxamci_set_ios()
           clk_enable()
             => here the clock is enabled twice, enable_count=2
       mmc_host_clk_release()
         pxamci_set_ios()
             => here the clock *remains enabled*, enable_count=1

- time passes with clock enabled

So I think there is a difference, unless my understand of the MMC core stack is wrong.

Cheers.

-- 
Robert