Message-ID: <8174bfcd-76bb-4b5a-a253-ae3edec54048@ideasonboard.com>
Date: Mon, 2 Jun 2025 10:16:49 +0300
From: Tomi Valkeinen <tomi.valkeinen@...asonboard.com>
To: Yemike Abhilash Chandra <y-abhilashchandra@...com>
Cc: hverkuil@...all.nl, sakari.ailus@...ux.intel.com,
 laurent.pinchart@...asonboard.com, vaishnav.a@...com, u-kumar1@...com,
 jai.luthra@...ux.dev, linux-media@...r.kernel.org,
 devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
 mchehab@...nel.org, robh@...nel.org, krzk+dt@...nel.org, conor+dt@...nel.org
Subject: Re: [PATCH 2/2] media: i2c: ds90ub960: Add support for DS90UB954-Q1

Hi,

On 28/05/2025 09:25, Yemike Abhilash Chandra wrote:
> Hi Tomi,
> 
> Thanks for the review.
> 
> On 27/05/25 11:10, Tomi Valkeinen wrote:
>> Hi,
>>
>> On 23/05/2025 11:36, Yemike Abhilash Chandra wrote:
>>> DS90UB954-Q1 is an FPDLink-III deserializer that is mostly register
>>> compatible with DS90UB960-Q1. The main difference is that it supports
>>> half of the RX and TX ports, i.e. 2x FPDLink RX ports and 1x CSI TX
>>> port.
>>>
>>> Some other registers are marked as reserved in the datasheet as well,
>>> notably around CSI-TX frame and line-count monitoring and some other
>>
>> Hmm what does that mean? That in log_status we show random data (or
>> maybe always 0) for these?
>>
> 
> It seems like it is showing 0's for these. I streamed around 100 frames.
> But the frame counter and line counter returned is 0. Please find the
> logs at [1].

If the registers are marked as reserved and don't function, we should
not use them. Here it doesn't do any harm when running the code, but it
does decrease the usefulness of log_status if the user is shown data
that is wrong (and the user most likely doesn't know it's wrong).

>>> status registers. The datasheet also does not mention anything about
>>> setting strobe position, and fails to lock the RX ports if we forcefully
>>> set it, so disable it through the hw_data.
>>
>> This app-note has some details:
>>
>> https://www.ti.com/lit/an/snla301/snla301.pdf
>>
>>> Link: https://www.ti.com/lit/gpn/ds90ub954-q1
>>> Signed-off-by: Yemike Abhilash Chandra <y-abhilashchandra@...com>
>>> ---
>>>   drivers/media/i2c/Kconfig     |  2 +-
>>>   drivers/media/i2c/ds90ub960.c | 46 +++++++++++++++++++++++++++++++++++
>>>   2 files changed, 47 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/media/i2c/Kconfig b/drivers/media/i2c/Kconfig
>>> index e68202954a8f..6e265e1cec20 100644
>>> --- a/drivers/media/i2c/Kconfig
>>> +++ b/drivers/media/i2c/Kconfig
>>> @@ -1662,7 +1662,7 @@ config VIDEO_DS90UB960
>>>       select V4L2_FWNODE
>>>       select VIDEO_V4L2_SUBDEV_API
>>>       help
>>> -      Device driver for the Texas Instruments DS90UB960
>>> +      Device driver for the Texas Instruments DS90UB954/DS90UB960
>>>         FPD-Link III Deserializer and DS90UB9702 FPD-Link IV
>>> Deserializer.
>>>     config VIDEO_MAX96714
>>> diff --git a/drivers/media/i2c/ds90ub960.c b/drivers/media/i2c/
>>> ds90ub960.c
>>> index ed2cf9d247d1..38e4f006d098 100644
>>> --- a/drivers/media/i2c/ds90ub960.c
>>> +++ b/drivers/media/i2c/ds90ub960.c
>>> @@ -460,6 +460,7 @@ struct ub960_hw_data {
>>>       u8 num_txports;
>>>       bool is_ub9702;
>>>       bool is_fpdlink4;
>>> +    bool is_ub954;
>>
>> No, let's not add any more of these. We should have enums for the device
>> model and the "family" (ub954/ub960 are clearly of the same family,
>> whereas ub9702 is of a newer one).
>>
> 
> Got it. I will add enums in the next revision.
> 
>>>   };
>>>     enum ub960_rxport_mode {
>>> @@ -982,6 +983,10 @@ static int ub960_txport_select(struct ub960_data
>>> *priv, u8 nport)
>>>         lockdep_assert_held(&priv->reg_lock);
>>>   +    /* TX port registers are shared for UB954*/
>>
>> Space missing at the end. What does the comment mean? "registers are
>> shared"?
>>
> 
> Apologies for the inaccurate comment description, My intention to
> comment that the tx_port_select function does not make sense for
> UB954, since we have only 1 CSI TX. May be I can have something
> like below.
> 
> /** UB954 has only 1 CSI TX. Hence, no need to select **/
> 
>> I think it's good to have this after the lockdep assert. The lock rules
>> are in place, even if on ub954 we don't do anything here.
>>
>>> +    if (priv->hw_data->is_ub954)
>>> +        return 0;
>>> +
>>>       if (priv->reg_current.txport == nport)
>>>           return 0;
>>>   @@ -1415,6 +1420,13 @@ static int ub960_parse_dt_txport(struct
>>> ub960_data *priv,
>>>           goto err_free_vep;
>>>       }
>>>   +    /* UB954 does not support 1.2 Gbps */
>>> +    if (priv->tx_data_rate == MHZ(1200) && priv->hw_data->is_ub954) {
>>
>> Test for ub954 first, 1200 MHz second. It's more logical for the reader
>> that way.
>>
> 
> Noted, will do that in the next revision.
> 
>>> +        dev_err(dev, "tx%u: invalid 'link-frequencies' value\n",
>>> nport);
>>> +        ret = -EINVAL;
>>> +        goto err_free_vep;
>>> +    }
>>> +
>>>       v4l2_fwnode_endpoint_free(&vep);
>>>         priv->txports[nport] = txport;
>>> @@ -1572,6 +1584,10 @@ static int ub960_rxport_set_strobe_pos(struct
>>> ub960_data *priv,
>>>       u8 clk_delay, data_delay;
>>>       int ret = 0;
>>>   +    /* FIXME: After writing to this area the UB954 chip no longer
>>> responds */
>>> +    if (priv->hw_data->is_ub954)
>>> +        return 0;
>>> +
>>
>> Check the app note. It would be nice to have this working, as, afaik,
>> the HW functionality should be the same on ub954 and ub960.
>>
> 
> I tried referring the app note and changed the strobe position values
> accordingly, but it did not help.
> 
> Since the app note also specifies the below at Table 2 Strobe Adaption
> Modes
> 
> "
> AEQ Adaption Mode--> Strobe position is selected as part of AEQ. This is
> the default mode.
> 
> Manual Adaption Mode --> The strobe position is selected manually and
> will remain at
> the specified position until a new one is chosen. This mode is
> recommended as an
> evaluation and debugging mode "
> 
> Since, under the default settings, the strobe position is selected as
> part of the AEQ process.
> Can we limit the ub960_rxport_set_strobe_pos function to only UB960 and
> UB9702.

Ok. But it doesn't sound good if we just skip the
ub960_rxport_set_strobe_pos(), but keep all the other EQ related writes.
I.e. we do the EQ config partially, and leave out parts that, for
unknown reasons, seem to cause problems...

So probably the check should be in ub960_rxport_config_eq(). With a
FIXME comment, and a short note where it fails.

That said, if everyone (?) agrees that the HW should support this, it
would be really nice if you can keep poking the FPD-Link people in TI
and try to get clarification on what's going on (what's the diff between
ub960 and ub954).

Btw, did you look at the other EQ related writes, and check if they're
valid for ub954?

 Tomi

>>>       clk_delay = UB960_IR_RX_ANA_STROBE_SET_CLK_NO_EXTRA_DELAY;
>>>       data_delay = UB960_IR_RX_ANA_STROBE_SET_DATA_NO_EXTRA_DELAY;
>>>   @@ -5021,6 +5037,27 @@ static int ub960_enable_core_hw(struct
>>> ub960_data *priv)
>>>       if (priv->hw_data->is_ub9702)
>>>           ret = ub960_read(priv, UB9702_SR_REFCLK_FREQ, &refclk_freq,
>>>                    NULL);
>>> +    else if (priv->hw_data->is_ub954) {
>>> +        /* From DS90UB954-Q1 datasheet:
>>> +         * "REFCLK_FREQ measurement is not synchronized. Value in this
>>> +         * register should read twice and only considered valid if
>>> +         * REFCLK_FREQ is unchanged between reads."
>>> +         */
>>> +        unsigned long timeout = jiffies + msecs_to_jiffies(100);
>>> +
>>> +        do {
>>> +            u8 refclk_new;
>>> +
>>> +            ret = ub960_read(priv, UB960_XR_REFCLK_FREQ, &refclk_new,
>>> +                     NULL);
>>> +            if (ret)
>>> +                goto err_pd_gpio;
>>> +
>>> +            if (refclk_new == refclk_freq)
>>> +                break;
>>> +            refclk_freq = refclk_new;
>>> +        } while (time_before(jiffies, timeout));
>>> +    }
>>
>> This feels a bit too much for a not-that-important debug print... As the
>> tests show that a single read is (practically always?) enough, I think
>> we can just use the same code as for ub960. Maybe add a comment about
>> it, though.
>>
> 
> okay, I will use the same code that is being used for UB960 and will add
> a comment
> about that.
> 
> Thanks and Regards,
> Abhilash Chandra
> 
> [1]: https://gist.github.com/Yemike-Abhilash-Chandra/
> c6b3da2a10586567a3a4179a2b20d21b
> 
>>   Tomi
>>
>>>       else
>>>           ret = ub960_read(priv, UB960_XR_REFCLK_FREQ, &refclk_freq,
>>>                    NULL);
>>> @@ -5177,6 +5214,13 @@ static void ub960_remove(struct i2c_client
>>> *client)
>>>       mutex_destroy(&priv->reg_lock);
>>>   }
>>>   +static const struct ub960_hw_data ds90ub954_hw = {
>>> +    .model = "ub954",
>>> +    .num_rxports = 2,
>>> +    .num_txports = 1,
>>> +    .is_ub954 = true,
>>> +};
>>> +
>>>   static const struct ub960_hw_data ds90ub960_hw = {
>>>       .model = "ub960",
>>>       .num_rxports = 4,
>>> @@ -5192,6 +5236,7 @@ static const struct ub960_hw_data ds90ub9702_hw
>>> = {
>>>   };
>>>     static const struct i2c_device_id ub960_id[] = {
>>> +    { "ds90ub954-q1", (kernel_ulong_t)&ds90ub954_hw },
>>>       { "ds90ub960-q1", (kernel_ulong_t)&ds90ub960_hw },
>>>       { "ds90ub9702-q1", (kernel_ulong_t)&ds90ub9702_hw },
>>>       {}
>>> @@ -5199,6 +5244,7 @@ static const struct i2c_device_id ub960_id[] = {
>>>   MODULE_DEVICE_TABLE(i2c, ub960_id);
>>>     static const struct of_device_id ub960_dt_ids[] = {
>>> +    { .compatible = "ti,ds90ub954-q1", .data = &ds90ub954_hw },
>>>       { .compatible = "ti,ds90ub960-q1", .data = &ds90ub960_hw },
>>>       { .compatible = "ti,ds90ub9702-q1", .data = &ds90ub9702_hw },
>>>       {}
>>

Powered by blists - more mailing lists