Hi Laurent,
On 18/06/2020 03:36, Laurent Pinchart wrote:
> Hi Kieran,
>
> On Wed, Jun 17, 2020 at 11:16:45AM +0100, Kieran Bingham wrote:
>> On 17/06/2020 11:05, Laurent Pinchart wrote:
>>> On Wed, Jun 17, 2020 at 10:56:29AM +0100, Kieran Bingham wrote:
>>>> On 15/06/2020 00:59, Laurent Pinchart wrote:
>>>>> The external pixel rate is retrieved when starting the camerarx and only
>>>>> used then. There's no need to store it in the cal_camerarx structure, it
>>>>> can be returned by cal_camerarx_get_external_info() and explicitly
>>>>> passed to cal_camerarx_config().
>>>>>
>>>>> While at it, rename cal_camerarx_get_external_info() to
>>>>> cal_camerarx_get_external_rate() to better reflect the function's
>>>>> purpose.
>>>>>
>>>>> Signed-off-by: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
>>>>> ---
>>>>> drivers/media/platform/ti-vpe/cal.c | 24 ++++++++++++------------
>>>>> 1 file changed, 12 insertions(+), 12 deletions(-)
>>>>>
>>>>> diff --git a/drivers/media/platform/ti-vpe/cal.c b/drivers/media/platform/ti-vpe/cal.c
>>>>> index 8326db0e4197..a11457909134 100644
>>>>> --- a/drivers/media/platform/ti-vpe/cal.c
>>>>> +++ b/drivers/media/platform/ti-vpe/cal.c
>>>>> @@ -272,7 +272,6 @@ struct cal_camerarx {
>>>>> struct v4l2_fwnode_endpoint endpoint;
>>>>> struct device_node *sensor_node;
>>>>> struct v4l2_subdev *sensor;
>>>>> - unsigned int external_rate;
>>>>
>>>> Here, external_rate is 32 bit,
>>>>
>>>>> };
>>>>>
>>>>> struct cal_dev {
>>>>> @@ -481,9 +480,10 @@ static void cal_quickdump_regs(struct cal_dev *cal)
>>>>> * ------------------------------------------------------------------
>>>>> */
>>>>>
>>>>> -static int cal_camerarx_get_external_info(struct cal_camerarx *phy)
>>>>> +static s64 cal_camerarx_get_external_rate(struct cal_camerarx *phy)
>>>>> {
>>>>> struct v4l2_ctrl *ctrl;
>>>>> + s64 rate;
>>>>
>>>> and now it becomes a 64 bit value.
>>>>
>>>>>
>>>>> if (!phy->sensor)
>>>>> return -ENODEV;
>>>>> @@ -495,10 +495,10 @@ static int cal_camerarx_get_external_info(struct cal_camerarx *phy)
>>>>> return -EPIPE;
>>>>> }
>>>>>
>>>>> - phy->external_rate = v4l2_ctrl_g_ctrl_int64(ctrl);
>>>>> - phy_dbg(3, phy, "sensor Pixel Rate: %u\n", phy->external_rate);
>>>>> + rate = v4l2_ctrl_g_ctrl_int64(ctrl);
>>>>> + phy_dbg(3, phy, "sensor Pixel Rate: %llu\n", rate);
>>>>>
>>>>> - return 0;
>>>>> + return rate;
>>>>> }
>>>>>
>>>>> static void cal_camerarx_lane_config(struct cal_camerarx *phy)
>>>>> @@ -554,7 +554,7 @@ static void cal_camerarx_disable(struct cal_camerarx *phy)
>>>>> #define TCLK_MISS 1
>>>>> #define TCLK_SETTLE 14
>>>>>
>>>>> -static void cal_camerarx_config(struct cal_camerarx *phy,
>>>>> +static void cal_camerarx_config(struct cal_camerarx *phy, s64 external_rate,
>>>>> const struct cal_fmt *fmt)
>>>>> {
>>>>> unsigned int reg0, reg1;
>>>>> @@ -566,7 +566,7 @@ static void cal_camerarx_config(struct cal_camerarx *phy,
>>>>>
>>>>> /* DPHY timing configuration */
>>>>> /* CSI-2 is DDR and we only count used lanes. */
>>>>> - csi2_ddrclk_khz = phy->external_rate / 1000
>>>>> + csi2_ddrclk_khz = external_rate / 1000
>>>>> / (2 * num_lanes) * fmt->bpp;
>>>>
>>>> Which causes this calculation to fail on 32 bit ARM builds.
>>>> (I'm building for the DRA76-EVM).
>>>
>>> Oops :-/
>>>
>>>> I've got the following fix up on the top of your tree to solve this, but
>>>> I'm not particularly happy about having to break the calculation up (and
>>>> re-use external_rate) though the use of do_div.
>>>>
>>>> From ca6ce335a852e34364bc45cb4240f703e4ea4248 Mon Sep 17 00:00:00 2001
>>>> From: Kieran Bingham <kieran.bingham@xxxxxxxxxxxxxxxx>
>>>> Date: Tue, 16 Jun 2020 16:19:04 +0100
>>>> Subject: [PATCH] media: ti-vpe: cal: Use do_div() for 64 bit operations
>>>>
>>>> Support building the CAL driver on arm32 bit targets by updating the
>>>> CSI2 clock calculation (which uses a signed 64 bit input value from
>>>> the sensors pixel clock rate) to use the do_div() helpers.
>>>>
>>>> The calculation is split into distinct parts to maintain
>>>> order of operations while making use of the do_div macro and further
>>>> re-ordered to convert to kHz at the end to maintain precision.
>>>>
>>>> Signed-off-by: Kieran Bingham <kieran.bingham@xxxxxxxxxxxxxxxx>
>>>> ---
>>>> drivers/media/platform/ti-vpe/cal-camerarx.c | 22 +++++++++++++++++---
>>>> 1 file changed, 19 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/drivers/media/platform/ti-vpe/cal-camerarx.c
>>>> b/drivers/media/platform/ti-vpe/cal-camerarx.c
>>>> index 014ca46509db..0ef19a516902 100644
>>>> --- a/drivers/media/platform/ti-vpe/cal-camerarx.c
>>>> +++ b/drivers/media/platform/ti-vpe/cal-camerarx.c
>>>> @@ -126,9 +126,25 @@ static void cal_camerarx_config(struct cal_camerarx
>>>> *phy, s64 external_rate)
>>>> u32 num_lanes = mipi_csi2->num_data_lanes;
>>>>
>>>> /* DPHY timing configuration */
>>>> - /* CSI-2 is DDR and we only count used lanes. */
>>>> - csi2_ddrclk_khz = external_rate / 1000
>>>> - / (2 * num_lanes) * phy->fmtinfo->bpp;
>>>> +
>>>> + /*
>>>> + * CSI-2 is DDR and we only count used lanes.
>>>> + *
>>>> + * csi2_ddrclk_khz = external_rate / 1000
>>>> + * / (2 * num_lanes) * phy->fmtinfo->bpp;
>>>> + *
>>>> + * The equation is broken into separate statements to maintain
>>>> + * order of operations, and conversion to kHz is done last to
>>>> + * keep precision.
>>>> + *
>>>> + * The 64 bit external_rate is modified during this equation and
>>>> + * contains the result, not the original after calculation.
>>>> + */
>>>> + do_div(external_rate, 2 * num_lanes);
>>>> + external_rate *= phy->fmtinfo->bpp;
>>>> + do_div(external_rate, 1000);
>>>> + csi2_ddrclk_khz = external_rate;
>>>
>>> How about
>>>
>>> external_rate *= phy->fmtinfo->bpp;
>>> do_div(external_rate, 2 * num_lanes * 1000);
>>
>> Ah yes, that looks better indeed, and keeps the improved integer precision.
>>
>> I believe the s64 external_rate should cope with the * bpp operation
>> easily too, so I don't think there's any risk of an overflow there.
>>
>> Squash it ;-)
>
> Even better,
>
> csi2_ddrclk_khz = div_s64(external_rate * phy->fmtinfo->bpp,
> 2 * num_lanes * 1000);
Ah, here I was about to say "you can't put the rate*bpp together because
the do_div reuses the argument ... and now I see div_s64 ...
/me discovers yet another way to do divides :-S
Well that's better all round ;-)
--
Kieran
>>>> +
>>>> phy_dbg(1, phy, "csi2_ddrclk_khz: %d\n", csi2_ddrclk_khz);
>>>>
>>>> /* THS_TERM: Programmed value = floor(20 ns/DDRClk period) */
>>>> --
>>>> 2.25.1
>>>>
>>>>
>>>> If you have a better way to correctly calculate the rate (also noting
>>>> that I moved the /1000 to the end, I'm not sure if that's more correct,
>>>> or makes it stop following what the hardware would do) - please update
>>>> accordingly, or feel free to squash this patch in as you wish.
>>>>
>>>>
>>>> With the calculation corrected:
>>>>
>>>> Reviewed-by: Kieran Bingham <kieran.bingham@xxxxxxxxxxxxxxxx>
>>>>
>>>> --
>>>> Kieran
>>>>
>>>>
>>>>> phy_dbg(1, phy, "csi2_ddrclk_khz: %d\n", csi2_ddrclk_khz);
>>>>>
>>>>> @@ -667,13 +667,14 @@ static void cal_camerarx_wait_stop_state(struct cal_camerarx *phy)
>>>>> static int cal_camerarx_start(struct cal_camerarx *phy,
>>>>> const struct cal_fmt *fmt)
>>>>> {
>>>>> + s64 external_rate;
>>>>> u32 sscounter;
>>>>> u32 val;
>>>>> int ret;
>>>>>
>>>>> - ret = cal_camerarx_get_external_info(phy);
>>>>> - if (ret < 0)
>>>>> - return ret;
>>>>> + external_rate = cal_camerarx_get_external_rate(phy);
>>>>> + if (external_rate < 0)
>>>>> + return external_rate;
>>>>>
>>>>> ret = v4l2_subdev_call(phy->sensor, core, s_power, 1);
>>>>> if (ret < 0 && ret != -ENOIOCTLCMD && ret != -ENODEV) {
>>>>> @@ -719,7 +720,7 @@ static int cal_camerarx_start(struct cal_camerarx *phy,
>>>>> reg_read(phy, CAL_CSI2_PHY_REG0);
>>>>>
>>>>> /* Program the PHY timing parameters. */
>>>>> - cal_camerarx_config(phy, fmt);
>>>>> + cal_camerarx_config(phy, external_rate, fmt);
>>>>>
>>>>> /*
>>>>> * b. Assert the FORCERXMODE signal.
>>>>> @@ -1034,7 +1035,6 @@ static struct cal_camerarx *cal_camerarx_create(struct cal_dev *cal,
>>>>>
>>>>> phy->cal = cal;
>>>>> phy->instance = instance;
>>>>> - phy->external_rate = 192000000;
>>>>>
>>>>> phy->res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
>>>>> (instance == 0) ?
>
--
Regards
--
Kieran
