Date:   Wed, 5 Aug 2020 10:29:16 -0700
From:   Sowjanya Komatineni <skomatineni@...dia.com>
To:     Dmitry Osipenko <digetx@...il.com>,
        Thierry Reding <thierry.reding@...il.com>
CC:     <jonathanh@...dia.com>, <frankc@...dia.com>, <hverkuil@...all.nl>,
        <sakari.ailus@....fi>, <robh+dt@...nel.org>,
        <helen.koike@...labora.com>, <gregkh@...uxfoundation.org>,
        <linux-media@...r.kernel.org>, <devicetree@...r.kernel.org>,
        <linux-tegra@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v8 08/10] gpu: host1x: mipi: Keep MIPI clock enabled till
 calibration is done


On 8/5/20 10:04 AM, Sowjanya Komatineni wrote:
>
> On 8/5/20 9:57 AM, Dmitry Osipenko wrote:
>> 05.08.2020 19:50, Sowjanya Komatineni пишет:
>>> On 8/5/20 9:47 AM, Dmitry Osipenko wrote:
>>>> 05.08.2020 19:33, Sowjanya Komatineni пишет:
>>>>> On 8/5/20 7:19 AM, Dmitry Osipenko wrote:
>>>>>> 05.08.2020 17:05, Dmitry Osipenko пишет:
>>>>>>> 05.08.2020 16:46, Thierry Reding пишет:
>>>>>>>> On Mon, Aug 03, 2020 at 08:42:24AM -0700, Sowjanya Komatineni 
>>>>>>>> wrote:
>>>>>>>>> With the split of MIPI calibration into tegra_mipi_calibrate() 
>>>>>>>>> and
>>>>>>>>> tegra_mipi_wait(), MIPI clock is not kept enabled till the
>>>>>>>>> calibration
>>>>>>>>> is done.
>>>>>>>>>
>>>>>>>>> So, this patch skips disabling MIPI clock after triggering 
>>>>>>>>> start of
>>>>>>>>> calibration and disables it only after waiting for done status 
>>>>>>>>> from
>>>>>>>>> the calibration logic.
>>>>>>>>>
>>>>>>>>> This patch renames tegra_mipi_calibrate() as
>>>>>>>>> tegra_mipi_start_calibration()
>>>>>>>>> and tegra_mipi_wait() as tegra_mipi_finish_calibration() to be
>>>>>>>>> inline
>>>>>>>>> with their usage.
>>>>>>>>>
>>>>>>>>> As MIPI clock is left enabled and in case of any failures with 
>>>>>>>>> CSI
>>>>>>>>> input
>>>>>>>>> streaming tegra_mipi_finish_calibration() will not get invoked.
>>>>>>>>> So added new API tegra_mipi_cancel_calibration() which disables
>>>>>>>>> MIPI clock
>>>>>>>>> and consumer drivers can call this in such cases.
>>>>>>>>>
>>>>>>>>> Reviewed-by: Dmitry Osipenko <digetx@...il.com>
>>>>>>>>> Signed-off-by: Sowjanya Komatineni <skomatineni@...dia.com>
>>>>>>>>> ---
>>>>>>>>>     drivers/gpu/drm/tegra/dsi.c |  4 ++--
>>>>>>>>>     drivers/gpu/host1x/mipi.c   | 19 ++++++++++---------
>>>>>>>>>     include/linux/host1x.h      |  5 +++--
>>>>>>>>>     3 files changed, 15 insertions(+), 13 deletions(-)
>>>>>>>>>
>>>>>>>>> diff --git a/drivers/gpu/drm/tegra/dsi.c
>>>>>>>>> b/drivers/gpu/drm/tegra/dsi.c
>>>>>>>>> index 3820e8d..a7864e9 100644
>>>>>>>>> --- a/drivers/gpu/drm/tegra/dsi.c
>>>>>>>>> +++ b/drivers/gpu/drm/tegra/dsi.c
>>>>>>>>> @@ -694,11 +694,11 @@ static int tegra_dsi_pad_calibrate(struct
>>>>>>>>> tegra_dsi *dsi)
>>>>>>>>>             DSI_PAD_PREEMP_PD(0x03) | DSI_PAD_PREEMP_PU(0x3);
>>>>>>>>>         tegra_dsi_writel(dsi, value, DSI_PAD_CONTROL_3);
>>>>>>>>>     -    err = tegra_mipi_calibrate(dsi->mipi);
>>>>>>>>> +    err = tegra_mipi_start_calibration(dsi->mipi);
>>>>>>>>>         if (err < 0)
>>>>>>>>>             return err;
>>>>>>>>>     -    return tegra_mipi_wait(dsi->mipi);
>>>>>>>>> +    return tegra_mipi_finish_calibration(dsi->mipi);
>>>>>>>>>     }
>>>>>>>>>       static void tegra_dsi_set_timeout(struct tegra_dsi *dsi,
>>>>>>>>> unsigned long bclk,
>>>>>>>>> diff --git a/drivers/gpu/host1x/mipi.c 
>>>>>>>>> b/drivers/gpu/host1x/mipi.c
>>>>>>>>> index e606464..b15ab6e 100644
>>>>>>>>> --- a/drivers/gpu/host1x/mipi.c
>>>>>>>>> +++ b/drivers/gpu/host1x/mipi.c
>>>>>>>>> @@ -293,17 +293,19 @@ int tegra_mipi_disable(struct
>>>>>>>>> tegra_mipi_device *dev)
>>>>>>>>>     }
>>>>>>>>>     EXPORT_SYMBOL(tegra_mipi_disable);
>>>>>>>>>     -int tegra_mipi_wait(struct tegra_mipi_device *device)
>>>>>>>>> +void tegra_mipi_cancel_calibration(struct tegra_mipi_device
>>>>>>>>> *device)
>>>>>>>>> +{
>>>>>>>>> +    clk_disable(device->mipi->clk);
>>>>>>>> Do we need to do anything with the MIPI_CAL_CTRL and 
>>>>>>>> MIPI_CAL_STATUS
>>>>>>>> registers here? We don't clear the START bit in the former when 
>>>>>>>> the
>>>>>>>> calibration has successfully finished, but I suspect that's 
>>>>>>>> because
>>>>>>>> the bit is self-clearing. But I wonder if we still need to 
>>>>>>>> clear it
>>>>>>>> upon cancellation to make sure the calibration does indeed stop.
>>>>>>> Apparently there is no way to explicitly stop calibration other
>>>>>>> than to
>>>>>>> reset MIPI calibration block, but Sowjanya says this is 
>>>>>>> unnecessary.
>>>>>>>
>>>>>>> Perhaps having a fixed delay before disabling clock could be 
>>>>>>> enough to
>>>>>>> ensure that calibration is stopped before the clock is disabled?
>>>>>>>
>>>>>> Actually, there is a MIPI_CAL_ACTIVE bit in the status register. 
>>>>>> Maybe
>>>>>> it needs to be polled until it's unset?
>>>>> Confirmed with HW design team during this patch update.
>>>>>
>>>>> SW does not need to clear START bit and only write 1 takes effect to
>>>>> that bit.
>>>>>
>>>>> Also, no need to have delay or do any other register settings 
>>>>> unclear as
>>>>> its FSM and there's nothing to get stuck.
>>>>>
>>>>> Also it goes thru small finite set of codes and by the time sensor
>>>>> programming happens for enabling streaming FSM will finish its
>>>>> calibration sequence much early and it will only wait for pads LP-11.
>>>>>
>>>>> So, during cancel we only need disable MIPI clock.
>>>>>
>>>> But there is no guarantee that cancel_calibration() couldn't be 
>>>> invoked
>>>> in the middle of the calibration process, hence FSM could freeze in an
>>>> intermediate state if it's running on the disabled MIPI clock, this
>>>> doesn't sound good.
>>> Actual calibration logic uses UART_FST_CAL clock which is always 
>>> enabled
>> What enables the UART_FST_CAL clock? I don't see this clock used 
>> anywhere.
>
> UART_FST_MIPI_CAL is shared with uart and is always enabled all the time.
>
> I don't see mipi driver handling this and I think that's because this 
> clock is enabled all the time as its used for UART as well. Probably 
> thierry can comment on this clock.
>
> Also regarding cancel calibration, as FSM goes thru only finite 
> sequence codes by the time csi stream and sensor stream happens which 
> is where we check for calibration to complete for sure calibration 
> will be finished and calibration logic will only wait for pads to be 
> in LP-11 to apply results. So nothing special needed during cancel 
> except to turn clock off to balance its usage count.
>
>
UART_FST_MIPI_CAL is the clock used for calibration logic which is FSM 
that goes thru sequence codes and when done waits for pads to be in 
LP-11 to apply results.

MIPI_CLK is controller gate clock which is also need to be kept enabled 
as incase if it sees LP-11 it updates registers so its recommended to 
have this clock enabled.

We can cancel_calibration() in CSI only when csi/sensor stream on fails 
and in which case there will be no LP-11 so we can unconditionally 
disable MIPI_CLK.

Powered by blists - more mailing lists