On 16/07/2018 13:58, Adam Ford wrote:
> On Thu, Jul 12, 2018 at 5:39 PM David Rivshin <drivshin@xxxxxxxxx> wrote:
>>
>> On Wed, 20 Jun 2018 11:53:21 -0400
>> David Rivshin <drivshin@xxxxxxxxx> wrote:
>>
>>> On Wed, 20 Jun 2018 09:11:19 +0200
>>> Neil Armstrong <narmstrong@xxxxxxxxxxxx> wrote:
>>>
>>>> Hi,
>>>>
>>>> On 20/06/2018 08:30, Tero Kristo wrote:
>>>>> On 19/06/18 21:53, David Rivshin wrote:
>>>>>> I think I've run across a regression in 4.16 that resulted from the TI clock
>>>>>> infrastructure changes. A bisect pointed to [1] as the first breaking commit.
>>>>>>
>>>>>> Our use-case is a pwm-dmtimer-omap using timer7, with timer7_fck parented
>>>>>> by clkdiv32k_ick (aka timer_32k_ck aka l4_per_cm:clk:0138:0). Starting with
>>>>>> 4.16 that parenting fails in omap_dm_timer_set_source(), called from
>>>>>> pwm_omap_dmtimer_probe, at this check:
>>>>>> if (clk_hw_get_num_parents(__clk_get_hw(timer->fclk)) < 2)
>>>>>> return 0;
>>>>>> because timer->fclk is "l4_per_cm:clk:0068:0", whose only possible parent
>>>>>> is "timer7_fck".
>>>>>>
>>>>>> It seemed to me that timer->fclk should instead be "timer7_fck" (which has
>>>>>> the desired set of possible parents). I tried following the lead set in [2]
>>>>>> and added "clocks" and "clock-names" properties on timer7. However, that
>>>>>> resulted in omap_dm_timer_of_set_source(), called by omap_dm_timer_prepare(),
>>>>>> failing when it tries to parent "timer7_fck" to itself.
>>>>>>
>>>>>> The following change in omap_dm_timer_prepare() does seem to work:
>>>>>> - timer->fclk = clk_get(&timer->pdev->dev, "fck");
>>>>>> + timer->fclk = clk_get_parent(clk_get(&timer->pdev->dev, "fck"));
>>>>>
>>>>> It seems omap_dm_timer_of_set_source() is actually completely broken and isn't doing anything useful. Neil, what was this function supposed to be doing, seeing you invented the initial patch for that (commit 31a7448f4fa8a528040e3df593e9781f55218183)?
>>>>
>>>> The original change was supposed to set the timer source clock by using the clock defined in the DT, but
>>>> it was made to work when the code was in arch/arm and the fck was not found from DT.
>>>>
>>>>
>>>> The omap_dm_timer_of_set_source() should be changed to get a specified clock name like "src":
>>>> parent = clk_get(&timer->pdev->dev, "src");
>>>> if (IS_ERR(parent))
>>>> return -ENODEV;
>>>>
>>>> And the fck clock should be added to all timer nodes.
>>>>
>>>> For example ed839c3664cb308144fe70a3a95813ab59a03dfb or 521f9e5e3f871cdac3ecd2a583d14d0472278053 only adds fck to timer1/2, the 7 timers should have the fck clock.
>>>>
>>>> @David: chan you test the omap_dm_timer_of_set_source() change and in DT update the timer7 node with :
>>>> clocks = <&timer7_fck>, <&clkdiv32k_ck>;
>>>> clock-names = "fck", "src";
>>>
>>> That didn't work as-is because clkdiv32k_ck isn't a direct parent of
>>> timer7_fck anymore. However, if I change it to
>>> <&l4_per_clkctrl AM3_CLKDIV32K_CLKCTRL 0>
>>> then all seems well.
>>>
>>> Since pwm-dmtimer-omap sets the source clock explicitly anyways
>>> (based on the DT), using any of the parents of timer7_fck as
>>> the timer7 "src" clock works in my case.
>>>
>>> Also, if I leave off the "src" clock entirely, then omap_dm_timer_prepare()
>>> defaults it to using that same clock:
>>> rc = omap_dm_timer_of_set_source(timer);
>>> if (rc == -ENODEV)
>>> return omap_dm_timer_set_source(timer, OMAP_TIMER_SRC_32_KHZ);
>>> and that also works for me.
>>>
>>> Thanks,
>>> David Rivshin
>>>
>>
>> Hi Neil,
>>
>> Are you planning on submitting a patch with this fix? Or would you prefer
>> I spun one up? One concern I have is that while I use an am335x and feel
>> fairly confident on those DT changes, I'm not so confident about what
>> changes would be needed for other SoC's devicetrees.
>
> I use the dmtimer to run a PWM on the backlight of a device I have on
> both an OMAP35xx and OMAP36/37. If someone wants to CC me on the
> patch, I can test it my boards.
Hi,
I can send a patch, but I'll need you test to make sure it still works.
Neil
>
> adam
>>
>>
>>>>
>>>> Thanks,
>>>> Neil
>>>>
>>>>>
>>>>> -Tero
>>>>>
>>>>>> but I'm very uncertain whether that's the right path.
>>>>>> Should omap_dm_timer know that the clock returned is actually a child of the
>>>>>> clock it's interested in? Or should clk_get(<timerN>,"fck") be returning the
>>>>>> timerN_fck in the first place? Or is the problem something else entirely?
>>>>>>
>>>>>> As I'm not sure how these pieces are all intended to fit together, any
>>>>>> pointers would be appreciated.
>>>>>>
>>>>>> Thanks,
>>>>>> David Rivshin
>>>>>>
>>>>>> [1] df54bfc5502a ("clk: ti: am33xx: add clkctrl clock data")
>>>>>> https://www.spinics.net/lists/linux-omap/msg139798.html
>>>>>> [2] 521f9e5e3f87 ("ARM: dts: am33xx: add fck under timers1/2")
>>>>>> https://www.spinics.net/lists/linux-omap/msg140605.html
>>>>>>
>>>>>
>>>>> --
>>>>> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
>>>>
>>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-omap" in
>> the body of a message to majordomo@xxxxxxxxxxxxxxx
>> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
