[PATCH 1/2] Adding platform level cpuidle driver for i.MX SoCs.

[shawn.guo@freescale.com (Shawn Guo)]

On Mon, Sep 12, 2011 at 12:11:30PM +0200, Sascha Hauer wrote:
> On Fri, Sep 09, 2011 at 10:33:40PM +0800, Shawn Guo wrote:
> > On Fri, Aug 26, 2011 at 09:27:04AM +0200, Sascha Hauer wrote:
> > > On Thu, Aug 25, 2011 at 09:33:14PM -0500, Robert Lee wrote:
> > > > Signed-off-by: Robert Lee <rob.lee@linaro.org>
> > > > ---
> > > > --- /dev/null
> > > > +++ b/arch/arm/plat-mxc/cpuidle.c
> > > > @@ -0,0 +1,112 @@
> > > > +/*
> > > > + * This file is licensed under the terms of the GNU General Public
> > > > + * License version 2.  This program is licensed "as is" without any
> > > > + * warranty of any kind, whether express or implied.
> > > > + */
> > > > +
> > > > +#include <linux/io.h>
> > > > +#include <linux/cpuidle.h>
> > > > +#include <asm/proc-fns.h>
> > > > +#include <mach/hardware.h>
> > > > +#include <mach/system.h>
> > > > +#include <mach/cpuidle.h>
> > > > +
> > > > +
> > > > +#ifndef MXC_OVERRIDE_DEFAULT_CPUIDLE_PARAMS
> > > 
> > > Either there is a possibility to overwrite the cpuidle parameters
> > > or there is none, but we don't need a define for this. I'm not
> > > convinced we need this possibility at all though.
> > > 
> > > > +
> > > > +#define MXC_X_MACRO(a, b, c) {c}
> > > > +static struct imx_cpuidle_params default_cpuidle_params[] = \
> > > > +	MXC_CPUIDLE_TABLE;
> > > > +#undef MXC_X_MACRO
> > > 
> > > Hell! This is one of the worst unnecessary preprocessor abuses I've ever
> > > seen. Do not show this in public again.
> > > 
> > > > +
> > > > +static struct imx_cpuidle_params *imx_cpuidle_params = default_cpuidle_params;
> > > > +
> > > > +#else
> > > > +
> > > > +static struct imx_cpuidle_params *imx_cpuidle_params;
> > > > +
> > > > +#endif
> > > > +
> > > > +
> > > > +
> > > > +/* in case you want to override the mach level params at the board level */
> > > > +void imx_cpuidle_board_params(struct imx_cpuidle_params *cpuidle_params)
> > > > +{
> > > > +	imx_cpuidle_params = cpuidle_params;
> > > > +}
> > > > +
> > > > +static int imx_enter_idle(struct cpuidle_device *dev,
> > > > +			       struct cpuidle_state *state)
> > > > +{
> > > > +	struct timeval before, after;
> > > > +	int idle_time, i;
> > > > +
> > > > +	/* We only need to pass an index to the mach level so here we
> > > > +	 * find the index of the name contained in the cpuidle_state
> > > > +	 * to pass. */
> > > > +	for (i = 0; i < MXC_NUM_CPUIDLE_STATES && i < CPUIDLE_STATE_MAX; i++)
> > > 
> > > A define for MXC_NUM_CPUIDLE_STATES looks wrong, it is not constant for
> > > different SoCs.
> > > 
> > > > +		if (state == &dev->states[i])
> > > > +			break;
> > > > +
> > > > +	local_irq_disable();
> > > > +	local_fiq_disable();
> > > > +
> > > > +	do_gettimeofday(&before);
> > > > +
> > > > +	imx_cpu_do_idle(i);
> > > 
> > > We are currently working on running a single image on as many SoCs we
> > > can. Take a step back and imagine what the linker says when it finds
> > > 6 different functions named imx_cpu_do_idle()
> > > 
> > > > +
> > > > +	do_gettimeofday(&after);
> > > > +
> > > > +	local_fiq_enable();
> > > > +	local_irq_enable();
> > > > +
> > > > +	idle_time = (after.tv_sec - before.tv_sec) * USEC_PER_SEC +
> > > > +			(after.tv_usec - before.tv_usec);
> > > > +
> > > > +	return idle_time;
> > > > +}
> > > > +
> > > > +static struct cpuidle_driver imx_cpuidle_driver = {
> > > > +	.name =         "imx_idle",
> > > > +	.owner =        THIS_MODULE,
> > > > +};
> > > > +
> > > > +static DEFINE_PER_CPU(struct cpuidle_device, imx_cpuidle_device);
> > > > +
> > > > +static int __init imx_cpuidle_init(void)
> > > > +{
> > > > +	struct cpuidle_device *device;
> > > > +	int i;
> > > > +
> > > > +	#define MXC_X_MACRO(a, b, c) #a
> > > > +	char *mxc_cpuidle_state_name[] = MXC_CPUIDLE_TABLE;
> > > > +	#undef MXC_X_MACRO
> > > > +
> > > > +	#define MXC_X_MACRO(a, b, c) b
> > > > +	char *mxc_cpuidle_state_desc[] = MXC_CPUIDLE_TABLE;
> > > > +	#undef MXC_X_MACRO
> > > > +
> > > > +	if (imx_cpuidle_params == NULL)
> > > > +		return -ENODEV;
> > > > +
> > > > +	cpuidle_register_driver(&imx_cpuidle_driver);
> > > > +
> > > > +	device = &per_cpu(imx_cpuidle_device, smp_processor_id());
> > > > +	device->state_count = MXC_NUM_CPUIDLE_STATES;
> > > > +
> > > > +	for (i = 0; i < MXC_NUM_CPUIDLE_STATES && i < CPUIDLE_STATE_MAX; i++) {
> > > > +		device->states[i].enter = imx_enter_idle;
> > > > +		device->states[i].exit_latency = imx_cpuidle_params[i].latency;
> > > > +		device->states[i].flags = CPUIDLE_FLAG_TIME_VALID;
> > > > +		strcpy(device->states[i].name, mxc_cpuidle_state_name[i]);
> > > > +		strcpy(device->states[i].desc, mxc_cpuidle_state_desc[i]);
> > > > +	}
> > > > +
> > > > +	if (cpuidle_register_device(device)) {
> > > > +		printk(KERN_ERR "imx_cpuidle_init: Failed registering\n");
> > > > +		return -ENODEV;
> > > > +	}
> > > > +	return 0;
> > > 
> > > This should really be a platform driver.
> > > 
> > I do not understand the reason behind this comment.  The cpuidle is
> > not really a platform device which typically has IO and IRQ such
> > hardware resources requiring a platform driver to talk to.
> > 
> > Looking at the following cpuidle drivers, only davinci implemented it
> > as a platform driver.
> > 
> >     arch/arm/mach-at91/cpuidle.c
> >     arch/arm/mach-davinci/cpuidle.c
> >     arch/arm/mach-exynos4/cpuidle.c
> >     arch/arm/mach-kirkwood/cpuidle.c
> >     arch/arm/mach-omap2/cpuidle34xx.c
> >     arch/arm/mach-shmobile/cpuidle.c
> > 
> > Also, I'm seeing imx cpufreq sitting on the mainline is not a platform
> > driver either.  I really did not think of any good reason for cpuidle
> > to be a platform driver necessarily.
> 
> The reason may not be really of technical nature. A platform driver only
> changes the thinking:
> 
> - A driver makes sure that the author knows that the hardware may or may
>   not be present.

The authors can know or have known that without creating a unnecessary
platform driver.

> - A driver makes sure that it may run on different types of hardware if
>   passed different platform specific data.

Without creating a platform driver, I'm sure there are some way to pass
platform specific data.  (How omap cpuidle and imx cpufreq did that?)  

> - A driver motivates the author to view the code as a single seperated
>   topic and to put all related code into the driver.
> 
A separated cpuidle.c file can also do the same.

> In the 'real' device driver world all points above are clear to driver
> authors, but when it comes to SoC internal stuff people tend to forget
> this.
> 
Again, it seems to me that we do not need to create an unnecessary
platform driver to get clear about all points above.

With cpuidle implemented as a platform driver, we will need to register
a platform device for it.  It's just silly to me, because we do not
have a cpuidle device at all.

-- 
Regards,
Shawn