Re: [PATCH v6 4/4] PCI: qcom: Add RX margining settings for 16.0 GT/s

[Shashank Babu Chinta Venkata]

On 9/4/24 09:04, Manivannan Sadhasivam wrote:
> On Wed, Sep 04, 2024 at 11:53:42AM +0200, Johan Hovold wrote:
>> On Wed, Sep 04, 2024 at 12:42:00PM +0530, Manivannan Sadhasivam via B4 Relay wrote:
>>> From: Shashank Babu Chinta Venkata <quic_schintav@xxxxxxxxxxx>
>>>
>>> Add RX lane margining settings for 16.0 GT/s (GEN 4) data rate. These
>>> settings improve link stability while operating at high date rates and
>>> helps to improve signal quality.
>>>
>>> Signed-off-by: Shashank Babu Chinta Venkata <quic_schintav@xxxxxxxxxxx>
>>> Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
>>> [mani: dropped the code refactoring and minor changes]
>>> Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
>>> ---
>>>  drivers/pci/controller/dwc/pcie-designware.h  | 18 ++++++++++++++++
>>>  drivers/pci/controller/dwc/pcie-qcom-common.c | 31 +++++++++++++++++++++++++++
>>>  drivers/pci/controller/dwc/pcie-qcom-common.h |  1 +
>>>  drivers/pci/controller/dwc/pcie-qcom-ep.c     |  4 +++-
>>>  drivers/pci/controller/dwc/pcie-qcom.c        |  4 +++-
>>>  5 files changed, 56 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/pci/controller/dwc/pcie-designware.h b/drivers/pci/controller/dwc/pcie-designware.h
>>> index 51744ad25575..f5be99731f7e 100644
>>> --- a/drivers/pci/controller/dwc/pcie-designware.h
>>> +++ b/drivers/pci/controller/dwc/pcie-designware.h
>>> @@ -209,6 +209,24 @@
>>>  
>>>  #define PCIE_PL_CHK_REG_ERR_ADDR			0xB28
>>>  
>>> +/*
>>> + * 16.0 GT/s (GEN4) lane margining register definitions
>>
>> nit: Gen 4?
>>
>>> + */
>>> +#define GEN4_LANE_MARGINING_1_OFF		0xb80
>>
>> nit: upper case hex
>>
>>> +#define MARGINING_MAX_VOLTAGE_OFFSET		GENMASK(29, 24)
>>> +#define MARGINING_NUM_VOLTAGE_STEPS		GENMASK(22, 16)
>>> +#define MARGINING_MAX_TIMING_OFFSET		GENMASK(13, 8)
>>> +#define MARGINING_NUM_TIMING_STEPS		GENMASK(5, 0)
>>> +
>>> +#define GEN4_LANE_MARGINING_2_OFF		0xb84
>>
>> Same here
>>
>>> +#define MARGINING_IND_ERROR_SAMPLER		BIT(28)
>>> +#define MARGINING_SAMPLE_REPORTING_METHOD	BIT(27)
>>> +#define MARGINING_IND_LEFT_RIGHT_TIMING		BIT(26)
>>> +#define MARGINING_IND_UP_DOWN_VOLTAGE		BIT(25)
>>> +#define MARGINING_VOLTAGE_SUPPORTED		BIT(24)
>>> +#define MARGINING_MAXLANES			GENMASK(20, 16)
>>> +#define MARGINING_SAMPLE_RATE_TIMING		GENMASK(13, 8)
>>> +#define MARGINING_SAMPLE_RATE_VOLTAGE		GENMASK(5, 0)
>>>  /*
>>>   * iATU Unroll-specific register definitions
>>>   * From 4.80 core version the address translation will be made by unroll
>>> diff --git a/drivers/pci/controller/dwc/pcie-qcom-common.c b/drivers/pci/controller/dwc/pcie-qcom-common.c
>>> index dc7d93db9dc5..99b75e7f085d 100644
>>> --- a/drivers/pci/controller/dwc/pcie-qcom-common.c
>>> +++ b/drivers/pci/controller/dwc/pcie-qcom-common.c
>>> @@ -43,3 +43,34 @@ void qcom_pcie_common_set_16gt_eq_settings(struct dw_pcie *pci)
>>>  	dw_pcie_writel_dbi(pci, GEN3_EQ_CONTROL_OFF, reg);
>>>  }
>>>  EXPORT_SYMBOL_GPL(qcom_pcie_common_set_16gt_eq_settings);
>>> +
>>> +void qcom_pcie_common_set_16gt_rx_margining_settings(struct dw_pcie *pci)
>>
>> I'd try to find a shorter symbol name here, "settings" seems redundant
>> after "set". Perhaps just
>>
>> 	qcom_pcie_common_enable_lane_margining()
>>
>> or
>>
>> 	qcom_pcie_common_enable_16gt_lane_margining()?
>>
> 
> This one looks better. Since lane margining is implemented in the receiver, we
> don't really need 'rx' in the function name.
> 
>> if these settings are indeed specific to 16 GT/s. But perhaps it's
>> better to let the helper honour pci->max_link_speed if different
>> settings will later be needed for higher speeds:
>>
>> 	if (pcie_link_speed[pci->max_link_speed] >= PCIE_SPEED_16_0GT)
>> 		qcom_pcie_common_enable_lane_margining(pci)
>>
> 
> I did thought about it during the review, but this setting claims to be for 16
> GT/s only. So I wouldn't recommend applying it to other speeds without checking 
> with Qcom.
> 
> Unfortunately, I'm on vacation for 2 weeks and have limited access to Qcom
> internal docs/chat. So won't be able to check it soon. If Shashank could check
> it, it is fine. But on the conservative side, let's stick to 16 GT/s only?
> 
> - Mani

Yes Mani I think we have to stick to 16 GT/s only for now as we haven't characterized 32 GT/s yet.


> 
>>>  void qcom_pcie_common_set_16gt_eq_settings(struct dw_pcie *pci);
>>> +void qcom_pcie_common_set_16gt_rx_margining_settings(struct dw_pcie *pci);
>>
>> And maybe something similar for the eq settings for symmetry.
>>
>> Johan
>