[U-Boot] [PATCH v2] DW SPI: Get clock value from Device Tree

[Marek Vasut]

On 10/30/2017 12:36 PM, Jagan Teki wrote:
> On Mon, Oct 30, 2017 at 4:24 PM, Marek Vasut <marex at denx.de> wrote:
>> On 10/30/2017 07:04 AM, Jagan Teki wrote:
>>> On Sat, Oct 28, 2017 at 5:09 PM, Marek Vasut <marex at denx.de> wrote:
>>>> On 10/27/2017 03:54 PM, Eugeniy Paltsev wrote:
>>>>> On Tue, 2017-10-24 at 15:22 +0530, Jagan Teki wrote:
>>>>>> On Tue, Oct 24, 2017 at 11:38 AM, Marek Vasut <marex at denx.de> wrote:
>>>>>>> On 10/23/2017 01:43 PM, Eugeniy Paltsev wrote:
>>>>>>>> On Thu, 2017-10-19 at 13:20 -0500, Dinh Nguyen wrote:
>>>>>>>>>
>>>>>>>>> On 10/19/2017 10:51 AM, Marek Vasut wrote:
>>>>>>>>>> On 10/19/2017 05:36 PM, Eugeniy Paltsev wrote:
>>>>>>>>>>> On Tue, 2017-10-17 at 20:32 +0530, Jagan Teki wrote:
>>>>>>>>>>>> On Tue, Oct 17, 2017 at 8:27 PM, Alexey Brodkin
>>>>>>>>>>>> <Alexey.Brodkin at synopsys.com> wrote:
>>>>>>>>>>>>> Hi Jagan,
>>>>>>>>>>>>>
>>>>>>>>>>>>>> -----Original Message-----
>>>>>>>>>>>>>> From: Eugeniy Paltsev [mailto:paltsev at synopsys.com]
>>>>>>>>>>>>>> Sent: Tuesday, October 17, 2017 4:33 PM
>>>>>>>>>>>>>> To: jagannadh.teki at gmail.com
>>>>>>>>>>>>>> Cc: u-boot at lists.denx.de; uboot-snps-arc at synopsys.com
>>>>>>>>>>>>>> Subject: [uboot-snps-arc] Re: [PATCH v2] DW SPI: Get clock value from Device Tree
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> How hard it is to make others to use clock manager? do you have any list?
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> clock_manager.h is an old (and non-generic) way to deal with different clocks.
>>>>>>>>>>>>>> For example in SOCFPGA_GEN5 and SOCFPGA_ARRIA10 clock_manager.h provides
>>>>>>>>>>>>>> cm_get_spi_controller_clk_hz function to deal with spi controller clock.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> But today we have another, linux-like alternative: to bind clocks via device tree
>>>>>>>>>>>>>> and manipulate with clocks via generic functions provided by clk.h
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> In this patch I added option to get clock via device tree using standard bindings
>>>>>>>>>>>>>> and restrict clock_manager.h functions usage only to targets which still use it,
>>>>>>>>>>>>>> so new targets can simply bind clock via device tree and they do not need to
>>>>>>>>>>>>>> implement/define something in clock_manager.h
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> So we don't need to make others to use clock manager :)
>>>>>>>>>>>>>
>>>>>>>>>>>>> Maybe it worth trying the other way around and think about switching SOCFPGA platforms to
>>>>>>>>>>>>> generic clk framework?
>>>>>>>>>>>>
>>>>>>>>>>>> Yes, ie what exactly I thought of, thanks!
>>>>>>>>>>>
>>>>>>>>>>> I checked cm_get_spi_controller_clk_hz implementation in SOCFPGA_GEN5 and
>>>>>>>>>>> SOCFPGA_ARRIA10: we can't simply replace it with "fixed-clock" driver as it
>>>>>>>>>>> manipulate with real hardware.
>>>>>>>>>>> The only way to do it is to replace SOCFPGA* clock manager functions by real
>>>>>>>>>>> clock driver.
>>>>>>>>>>>
>>>>>>>>>>> And given I don't have mentioned hardware so I barely can help with
>>>>>>>>>>> those improvements on SOCFPGA. That said if there're no short-term plans to
>>>>>>>>>>> switch SOCFPGA to clk framework maybe we'll be OK with my workaround with #ifdefs?
>>>>>>>>>>
>>>>>>>>>> Wait for Dinh's reply ...
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Honestly, I don't have too much time to work on this right now. So I
>>>>>>>>> really don't when it can get done. But it'll go on my to-do list.
>>>>>>>>>
>>>>>>>>> Dinh
>>>>>>>>
>>>>>>>> Yep, thanks for your comments.
>>>>>>>>
>>>>>>>> So, Jagan,
>>>>>>>> given Dinh's reply, could you please apply this patch?
>>>>>>>
>>>>>>> I'd really hate it to start seeing soc-specific ifdefs in drivers,
>>>>>>> that's IMO not acceptable. A __weak override might be a temporary
>>>>>>> solution I'd be willing to live with though.
>>>>>>
>>>>>> I would rather like to see some check on clock manager itself whether
>>>>>> CONFIG_IS_ENABLED(CLK) is using or not? this can tends not to use
>>>>>> __weak as well soc #ifdefs in driver.
>>>>>>
>>>>>
>>>>> Actually I don't understand what do you mean.
>>>>> Even if I add any #ifdefs to the clock_manager.h I still need to wrap clock_manager.h
>>>>> include with #ifdefs as clock_manager.h is defined only for two
>>>>> targets - SOCFPGA_GEN5 and SOCFPGA_ARRIA10.
>>>>>
>>>>> #if defined(CONFIG_TARGET_SOCFPGA_GEN5) || defined(CONFIG_TARGET_SOCFPGA_ARRIA10)
>>>>> #include <asm/arch/clock_manager.h>
>>>>> #endif
>>>>>
>>>>> And I think it is better to add this #ifdef in driver than create empty
>>>>> clock_manager.h file for every new target which uses this driver.
>>>
>>> Check CLK in clock-manager like below, of-course cm_get_l4_noc_hz used
>>> in other IP's as well. we need to carefully adding these check and if
>>> require add CLK for remaining drivers as well. the reason for adding
>>> checks in clock-manager is it may be removable code if all use CLK.
>>
>> I do not understand what you're trying to tell me, but the patch below
>> makes no sense. What I would like to see is a weak function in the
>> driver and that function be overriden in the socfpga clock manager.
> 
> I'm trying to avoid __weak or #ifdef on driver instead make changes in
> clock manager which will remove in future(once all moved with CLK)

Well, if you do the proposed change, socfpga will break. The __weak is
the least possible evil here and I can live with that for now.

-- 
Best regards,
Marek Vasut