Re: [PATCHv1 1/1] omap3: pm: Delink opp layer and cpufreq

[Nishanth Menon <menon.nishanth@gmail.com>]

On 06/01/2010 03:01 PM, Premi, Sanjeev wrote:
>
>> -----Original Message-----
>> From: Nishanth Menon [mailto:menon.nishanth@gmail.com]
>> Sent: Monday, May 31, 2010 10:59 PM
>> To: Premi, Sanjeev
>> Cc: linux-omap@vger.kernel.org
>> Subject: Re: [PATCHv1 1/1] omap3: pm: Delink opp layer and cpufreq
>>
>> On 05/31/2010 03:39 PM, Sanjeev Premi wrote:
>>> The OPP layer was contained in the CONFIG_CPU_FREQ.
>>> This patch removes this containment relation.
>>>
>>> Signed-off-by: Sanjeev Premi<premi@ti.com>
>>> ---
>>>    arch/arm/mach-omap2/Makefile          |    6 +-
>>>    arch/arm/mach-omap2/board-omap3evm.c  |    2 +-
>
> [snip]--[snip]
>
>> you sure this is the only board file having "omap3-opp.h" ?
>> anyway.. the
>> need for board files to use opp_init is gone with my patch
>> http://marc.info/?l=linux-omap&m=127507237109393&w=2
>> so I wont harp on it, I would rather post a cleanup patch for
>> all board
>> files once the patch is in..(or mebbe kevin could drop the patch that
>> adds opp_init_table to board files ;) )..
>>
>
> [sp] You didn't reead the 0/1 of the patch series, where I have clearly
> mentioned that I will make changes to the other board specific files
> once there rest of the changes are well discussed. There may be, possibly,
> more changes in the board specific files and we can review them in the
> context of this file and then same can be repeated for other board files.
ok

>
>>>    arch/arm/mach-omap2/cpufreq34xx.c     |  164
>> --------------------------------
>>>    arch/arm/mach-omap2/omap3-opp.h       |   20 ----
>>>    arch/arm/mach-omap2/opp34xx_data.c    |  166
>> +++++++++++++++++++++++++++++++++
>>>    arch/arm/mach-omap2/pm34xx.c          |    1 -
>>>    arch/arm/plat-omap/Makefile           |    7 +-
>>>    arch/arm/plat-omap/cpu-omap.c         |   47 +++++++++
>>>    arch/arm/plat-omap/include/plat/opp.h |   82 +---------------
>>>    arch/arm/plat-omap/opp.c              |   46 ---------
>>>    10 files changed, 225 insertions(+), 316 deletions(-)
>>>    delete mode 100644 arch/arm/mach-omap2/cpufreq34xx.c
>>>    delete mode 100644 arch/arm/mach-omap2/omap3-opp.h
>>>    create mode 100644 arch/arm/mach-omap2/opp34xx_data.c
>>>
>
> [sp]
>> finding it difficult to align with this change, you introduce
>> omap3_pm_init_opp_table later into plat/opp.h which defeats generic
>> nature of opp.h - as it was supposed to be used for other
>> omaps as well..
>
> In that case the function omap3_pm_init_opp_table() can be made
> generic by renaming to omap_pm_init_opp_table and can be implemented
> for each omap family.
Do you intend to handle multiomap case by calling
each omap[1234]_pm_init_opp_table() if cpu_is_omap34xx() etc? you will 
still need a custom omap family specific init_opp_table - that is what 
this header provides.

>
> If opp table has to be implementyed for each family then why have
> different funtion with family specific prefixes?
opp table contents will be different for each family and they should all 
build and co-exist in a single uImage.

>
> Also, what this headerf ile is/was doing? only defining the
> function to return -EINVAL when CONFIG_CPU_FREQ is not selected;
> which is not required. For OPP layer to be used this table needs
> to be populated. Now, there is only one place this function is
> used, so why do/should be need a header for the same.

To allow the the external function that triggers it to be able to use 
it.. :)

>
> [snip]--[snip]
>
>> +obj-y				+= opp.o
>> +obj-$(CONFIG_TWL4030_POWER)	+= opp_twl_tps.o
> NAK. you just need TWL4030_CORE not power here. any reason to retain
> power? it has no dependency on power..
>
> [sp] Isn't this purpose of this opp to TWL linkage is to define
> the voltages in terms the PMIC connected; and later make sure
> that correct voltages are set via the PMIC? This is very much related
no it does not. these are just translation functions, as long as _CORE 
exists, it means that the system uses twl and we should be good to go.

> to POWER. We could also do it on CORE; but I don't see this as a
> big issue.
ok.

>
> But TWL4030 has more feature beyond PMIC but this is not the case
> with other simpler PMICs and I wanted to use CONFIG option that
> can be easier for someone else make the port easy to spot as a necessary
> change.

i am aligned with the change, except that I believe you should not be 
using POWER as the prefix for the config build dependency.

>
> [snip]--[snip]
>
>>> +	freq_table[i].index = i;
>>> +	freq_table[i].frequency = CPUFREQ_TABLE_END;
>>> +
>>> +	*table =&freq_table[0];
>>> +}
>>> +#endif
>>> +
>> errrr.... why? it used to be here and was moved to opp.c - see
>> http://git.kernel.org/?p=linux/kernel/git/khilman/linux-omap-p
> m.git;a=commit;h=9a6b00f70e9f4bce30ad4f8fab41a24bd3706dbd
>> you are essentially reverting that patch!
>
> [sp] May be I am reverting the patch, but I don't see this function
>       being used anywhere else. Most of the other cpufreq related
>       initialization is happening at this place.
>
>       Only the function omap_cpu_init() calls this function and it
>       is in the same file.
>
>       It also helps in need of an additional header; which seem
>       to make "de-linking" more complex - in terms of #ifdefs.

the idea was to have a common function for ALL omaps to create the table 
and reuse it where ever needed, if you look beyond omap3 into omap1,2 
and 4, the ability to do this is invaluable. does it matter if a 
function exists in the library even if not used?

>
> [snip]--[snip]
>
>>
>>> +/**
>>> + * omap3_pm_init_opp_table() - Initialize the OPP table
>> for OMAP3 devices.
>>> + *
>>> + * Initializes the OPP table for the current OMAP3 device.
>>> + */
>>> +int __init omap3_pm_init_opp_table(void);
>> NAK. opp. is meant to be used by omap2, OMAP4 etc..
>> when you removed from omap3-opp.h, it kinda needed you to
>> have it here,
>> which breaks the generic nature of this header.
>
> [sp] See my comment earlier. The function for init-ing the OPP
>       table can be made generic. Then (after rename) this is a
>       generic function itself.
>
>       Rather than making sweelping changes, I am only delinking
>       the OPP layer and CPU freq. These changes can be done
>       separately.
replied above.

>
>>
>>>
>>> -#endif		/* CONFIG_CPU_FREQ */
>>>    #endif		/* __ASM_ARM_OMAP_OPP_H */
>>> diff --git a/arch/arm/plat-omap/opp.c b/arch/arm/plat-omap/opp.c
>>> index 13da451..3ed3ec1 100644
>>> --- a/arch/arm/plat-omap/opp.c
>>> +++ b/arch/arm/plat-omap/opp.c
>>> @@ -351,49 +351,3 @@ int opp_disable(struct omap_opp *opp)
>>>    	opp->enabled = false;
>>>    	return 0;
>>>    }
>>> -
>>> -/* XXX document */
>>> -void opp_init_cpufreq_table(enum opp_t opp_type,
>>> -			    struct cpufreq_frequency_table **table)
>>> -{
>>> -	int i = 0, j;
>>> -	int opp_num;
>>> -	struct cpufreq_frequency_table *freq_table;
>>> -	struct omap_opp *opp;
>>> -
>>> -	if (opp_type>= OPP_TYPES_MAX) {
>>> -		pr_warning("%s: failed to initialize frequency"
>>> -				"table\n", __func__);
>>> -		return;
>>> -	}
>>> -
>>> -	opp_num = opp_get_opp_count(opp_type);
>>> -	if (opp_num<   0) {
>>> -		pr_err("%s: no opp table?\n", __func__);
>>> -		return;
>>> -	}
>>> -
>>> -	freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) *
>>> -			     (opp_num + 1), GFP_ATOMIC);
>>> -	if (!freq_table) {
>>> -		pr_warning("%s: failed to allocate frequency"
>>> -				"table\n", __func__);
>>> -		return;
>>> -	}
>>> -
>>> -	opp = _opp_list[opp_type];
>>> -	opp += opp_num;
>>> -	for (j = opp_num; j>= 0; j--) {
>>> -		if (opp->enabled) {
>>> -			freq_table[i].index = i;
>>> -			freq_table[i].frequency = opp->rate / 1000;
>>> -			i++;
>>> -		}
>>> -		opp--;
>>> -	}
>>> -
>>> -	freq_table[i].index = i;
>>> -	freq_table[i].frequency = CPUFREQ_TABLE_END;
>>> -
>>> -	*table =&freq_table[0];
>>> -}
>> not sure why you removed this..
>>
>
> [sp] It isn't removed but simply moved to the only file in the code where it
>       is being used... along with rest of the code related to CPU_FREQ.
>
> The way I see, the OPP layer has been mixed with CPUFREQ usage in the
> cureent code; but if you look at OPP layer as as "real" layer then the
> CPUFREQ implementation should be the "client" to this later and use its
> services - not get 'mingled' into the layer itself. If you go by this
> reasoning, this init function belong outside the OPP layer.
I have explained on top, further, the only set of dependencies i see:
a) naming of the the files
b) build and #ifdef CPUFREQ dependencies
these are changes I liked from your patch.

Regards,
Nishanth Menon