Re: [RFC 2/9] opp-modifier: Add opp-modifier-reg driver

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




On Mon, Mar 17, 2014 at 9:30 AM, Nishanth Menon <nm@xxxxxx> wrote:
> On 03/14/2014 04:00 PM, Rob Herring wrote:
>> On Fri, Mar 14, 2014 at 2:25 PM, Dave Gerlach <d-gerlach@xxxxxx> wrote:
>>> Driver to read from a register and depending on either set bits or
>>> a specific known selectively enable or disable OPPs based on DT node.
>>>
>>> Can support opp-modifier-reg-bit where single bits within the register
>>> determine the availability of an OPP or opp-modifier-reg-val where a
>>> certain value inside the register or a portion of it determine what the
>>> maximum allowed OPP is.
>>>
>>> The driver expects a device that has already has its OPPs loaded
>>> and then will disable the OPPs not matching the criteria specified in
>>> the opp-modifier table.
>>>
>>> Signed-off-by: Dave Gerlach <d-gerlach@xxxxxx>
>>> ---
>>>  .../devicetree/bindings/power/opp-modifier.txt     | 111 +++++++++
>>>  drivers/power/opp/Makefile                         |   1 +
>>>  drivers/power/opp/opp-modifier-reg.c               | 259 +++++++++++++++++++++
>>>  3 files changed, 371 insertions(+)
>>>  create mode 100644 Documentation/devicetree/bindings/power/opp-modifier.txt
>>>  create mode 100644 drivers/power/opp/opp-modifier-reg.c
>>>
>>> diff --git a/Documentation/devicetree/bindings/power/opp-modifier.txt b/Documentation/devicetree/bindings/power/opp-modifier.txt
>>> new file mode 100644
>>> index 0000000..af8a2e9
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/power/opp-modifier.txt
>>> @@ -0,0 +1,111 @@
>>> +* OPP-Modifier - opp modifier to selectively enable operating points
>>> +
>>> +Many SoCs that have selectively modifiable OPPs can specify
>>> +all available OPPs in their operating-points listing and then define
>>> +opp_modifiers to enable or disable the OPPs that are actually available
>>> +on the specific hardware.
>>> +
>>> +* OPP Modifier Provider
>>
>> Uggg. Please stop designing around the current OPP binding which has
>> the problem that the OPP table is not extensible to add more data.
>> Define a new OPP binding that solves these problems. This is at least
> Generically, there are three different issues with current OPP bindings:
> a) ability to enable disable certain OPPs depending on SoC OTP/Efuse
> settings.

More generically: ...depending on variety of factors.

> b) ability to reuse OPPs defined for one device node for another (cpu1
> to reuse OPP definitions of cpu0)
> c) ability to add additional information per OPP. we can argue this is
> a superset of (a), but really, the problems are different.

It is all additional data per OPP. Additional different information is
of course for different problems. That doesn't mean we need different
solutions.

> Previous proposals include making each OPP as a phandle, but there
> does not seem much traction in that direction either. - proposal here
> has nothing to do with (b) or (c).

They may have nothing to do with each other, but they all have to do
with the OPP binding. If we're going to change/extend the binding,
then all issues need to be taken into account.

>> the 3rd OPP related binding addition I've seen recently. But I
>> wouldn't spend a lot of effort on a new OPP binding just to add the
>> functionality you are adding here because I don't like the whole
>> concept in general. This might be a common way to determine valid OPPs
>> on TI chips, but I think it is too low level and I don't want to see
>
> Not just TI chips, but iMX, now, Marvell, Xilinx as well. potentially
> more as well. doing OTP/Efuse based decision on which OPPs are valid
> on a chip is not a TI specific thing. This was the reason for us to
> try to define something generic enough to be reused by more SoCs than
> just TI.

Agreed, but I'm not convinced how different SOCs determine valid OPPs
is common enough. Certainly how to mark an entry disabled is common
though.

>> bindings for every different possible way. Just add platform code to
>> do the OPP setup you need.
> Errr.. adding platform code means the hardware description goes back
> to kernel. is'nt that giving up on device tree binding for describing
> hardware?

We're always going to have some platform code. I'm not saying you have
to in this case. I'm saying either come up with an OPP binding
addressing all these issues or live with the existing one and fix it
up in the kernel or bootloader.

>> Frankly, I prefer the bootloader/firmware fixup the OPP table approach
>> mentioned in the cpufreq-cpu0 thread. Somewhat less desirable, but the
>> kernel could do the fixups as well (via of_update_property).
>
> a) Trying to move the hardware definition away from device tree seems
> to me a major step backwards.
> b) Allowing for definitions in platform code is a step backwards again
> for a generic solution that works for more than 1 vendor.
> c) moving the logic away to bootloader when it can easily be done in
> kernel again is adding burden to bootloader for data it does need to
> handle.

The burden has to be somewhere. Maintaining a binding forever in the
kernel is a burden as well if it is poorly designed.

Valid OPPs are not going to just be random. There's probably on a few
combinations and they'll be based on part# or speed grade or something
(which in turn defines the efuses in your case). While a dev board may
have random parts on it, an actual product would not. I could argue
that your DTB just needs to be correct to begin with for a given
part/design. Obviously, managing minor differences in a DTB like this
can be a pain. This is why firmware or bootloaders do adjustments to
the DTB at runtime and it is quite common.

> OPP is a hardware behavior, which OPPs are enabled are described in
> hardware on certain SoCs. the current proposal is to provide a generic
> solution for those devices that allow for dynamic definition of OPPs
> based on SoC efuse definition.

What if the decision is not based on a single register bit? Perhaps
efuses are not directly memory mapped. Maybe it is based on Si
revision. Or you need to limit frequency because a certain board can't
supply adequate current. You call this generic, but it is not. It
doesn't even solve the part that is generic which is marking some OPPs
disabled.

Rob
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]
  Powered by Linux