Re: [Intel-gfx] [PATCH 2/5] drm/i915/display/adlp: Add new PSR2 workarounds

[Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>]

On 9/13/21 8:00 PM, Souza, Jose wrote:
> On Mon, 2021-09-13 at 19:09 +0300, Gwan-gyeong Mun wrote:
>>
>> On 9/10/21 7:29 PM, Souza, Jose wrote:
>>> On Fri, 2021-09-10 at 16:38 +0300, Gwan-gyeong Mun wrote:
>>>>
>>>> On 9/10/21 2:07 AM, José Roberto de Souza wrote:
>>>>> Wa_16014451276 fixes the starting coordinate for PSR2 selective
>>>>> updates. CHICKEN_TRANS definition of the workaround bit has a wrong
>>>>> name based on workaround definition and HSD.
>>>>>
>>>>> Wa_14014971508 allows the screen to continue to be updated when
>>>>> coming back from DC5/DC6 and SF_SINGLE_FULL_FRAME bit is not kept
>>>>> set in PSR2_MAN_TRK_CTL.
>>>>>
>>>>> Wa_16012604467 fixes underruns when exiting PSR2 when it is in one
>>>>> of its internal states.
>>>>>
>>>>> Wa_14014971508 is still in pending status in BSpec but by
>>>>> the time this is reviewed and ready to be merged it will be finalized.
>>>>>
>>>>> BSpec: 54369
>>>>> BSpec: 50054
>>>>> Cc: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
>>>>> Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
>>>>> ---
>>>>>     drivers/gpu/drm/i915/display/intel_psr.c | 23 ++++++++++++++++++++++-
>>>>>     drivers/gpu/drm/i915/i915_reg.h          |  4 ++++
>>>>>     2 files changed, 26 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/i915/display/intel_psr.c b/drivers/gpu/drm/i915/display/intel_psr.c
>>>>> index 36816abb3bcc0..92c0b2159559f 100644
>>>>> --- a/drivers/gpu/drm/i915/display/intel_psr.c
>>>>> +++ b/drivers/gpu/drm/i915/display/intel_psr.c
>>>>> @@ -1086,6 +1086,12 @@ static void intel_psr_enable_source(struct intel_dp *intel_dp)
>>>>>     intel_de_write(dev_priv, reg, chicken);
>>>>>     }
>>>>>
>>>>> +/* Wa_16014451276:adlp */
>>>>> +if (IS_ALDERLAKE_P(dev_priv) &&
>>>>> +    intel_dp->psr.psr2_enabled)
>>>>> +intel_de_rmw(dev_priv, CHICKEN_TRANS(cpu_transcoder), 0,
>>>>> +     D13_1_BASED_X_GRANULARITY);
>>>> Depending on the capability of the PSR panel, the following setting may
>>>> not be necessary, could you add some comments such as "force enable
>>>> 1-based X granularity on PSR2 VSC SDP"?
>>>
>>> It was made sure that all alderlake-P BOM panels will have 1-based X granularity, I can add something like that.
>>>
>>>
>>>>> +
>>>>>     /*
>>>>>      * Per Spec: Avoid continuous PSR exit by masking MEMUP and HPD also
>>>>>      * mask LPSP to avoid dependency on other drivers that might block
>>>>> @@ -1131,6 +1137,11 @@ static void intel_psr_enable_source(struct intel_dp *intel_dp)
>>>>>          TRANS_SET_CONTEXT_LATENCY(intel_dp->psr.transcoder),
>>>>>          TRANS_SET_CONTEXT_LATENCY_MASK,
>>>>>          TRANS_SET_CONTEXT_LATENCY_VALUE(1));
>>>>> +
>>>>> +/* Wa_16012604467:adlp */
>>>>> +if (IS_ALDERLAKE_P(dev_priv) && intel_dp->psr.psr2_enabled)
>>>>> +intel_de_rmw(dev_priv, CLKGATE_DIS_MISC, 0,
>>>>> +     CLKGATE_DIS_MISC_DMASC_GATING_DIS);
>>>>>     }
>>>>>
>>>>>     static bool psr_interrupt_error_check(struct intel_dp *intel_dp)
>>>>> @@ -1320,6 +1331,11 @@ static void intel_psr_disable_locked(struct intel_dp *intel_dp)
>>>>>          TRANS_SET_CONTEXT_LATENCY(intel_dp->psr.transcoder),
>>>>>          TRANS_SET_CONTEXT_LATENCY_MASK, 0);
>>>>>
>>>>> +/* Wa_16012604467:adlp */
>>>>> +if (IS_ALDERLAKE_P(dev_priv) && intel_dp->psr.psr2_enabled)
>>>>> +intel_de_rmw(dev_priv, CLKGATE_DIS_MISC,
>>>>> +     CLKGATE_DIS_MISC_DMASC_GATING_DIS, 0);
>>>>> +
>>>>>     intel_snps_phy_update_psr_power_state(dev_priv, phy, false);
>>>>>
>>>>>     /* Disable PSR on Sink */
>>>>> @@ -1488,8 +1504,13 @@ static void psr2_man_trk_ctl_calc(struct intel_crtc_state *crtc_state,
>>>>>     u32 val = PSR2_MAN_TRK_CTL_ENABLE;
>>>>>
>>>>>     if (full_update) {
>>>>> +/*
>>>>> + * Wa_14014971508:adlp
>>>>> + * SINGLE_FULL_FRAME bit is not hold in register so can not be
>>>>> + * restored by DMC, so using CONTINUOS_FULL_FRAME to mimic that
>>>>> + */
>>>>>     if (IS_ALDERLAKE_P(dev_priv))
>>>>> -val |= ADLP_PSR2_MAN_TRK_CTL_SF_SINGLE_FULL_FRAME;
>>>>> +val |= ADLP_PSR2_MAN_TRK_CTL_SF_CONTINUOS_FULL_FRAME;
>>>>>     else
>>>>>     val |= PSR2_MAN_TRK_CTL_SF_SINGLE_FULL_FRAME;
>>>>>
>>>>> diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
>>>>> index c2853cc005ee6..0de2f7541da6c 100644
>>>>> --- a/drivers/gpu/drm/i915/i915_reg.h
>>>>> +++ b/drivers/gpu/drm/i915/i915_reg.h
>>>>> @@ -8235,6 +8235,7 @@ enum {
>>>>>     #define  VSC_DATA_SEL_SOFTWARE_CONTROLREG_BIT(25) /* GLK */
>>>>>     #define  FECSTALL_DIS_DPTSTREAM_DPTTGREG_BIT(23)
>>>>>     #define  DDI_TRAINING_OVERRIDE_ENABLEREG_BIT(19)
>>>>> +#define  D13_1_BASED_X_GRANULARITYREG_BIT(18)
>>>> The meaning of this macro is to set "force enable 1-based X granularity
>>>> on PSR2 VSC SDP" in Display 13.1 ADL, so the meaning of the macro may be
>>>> a little ambiguous.
>>>
>>> The name of registers are set to match specification name as close as possible not the use or meaning.
>> Yes, just looking at the macro, I thought that it could be interpreted
>> in two ways: D13 / 1_BASED_X_GRANULARITY or D13_1 / BASED_X_GRANULARITY.
>> If our macro naming convention is fine in this case, then I don't think
>> the code is the problem either.
> 
> Okay yes someone could interpret into those 2 ways but checking bspec makes it clears that it is the first one.
> I can rename to ADLP_1_BASED_X_GRANULARITY if you think it would make it better.
> 
the ADLP_1_BASED_X_GRANULARITY you suggested looks better.
>>>
>>>>>     #define  DDI_TRAINING_OVERRIDE_VALUEREG_BIT(18)
>>>>>     #define  DDIE_TRAINING_OVERRIDE_ENABLEREG_BIT(17) /* CHICKEN_TRANS_A only */
>>>>>     #define  DDIE_TRAINING_OVERRIDE_VALUEREG_BIT(16) /* CHICKEN_TRANS_A only */
>>>>> @@ -12789,4 +12790,7 @@ enum skl_power_gate {
>>>>>     #define CLKREQ_POLICY_MMIO(0x101038)
>>>>>     #define  CLKREQ_POLICY_MEM_UP_OVRDREG_BIT(1)
>>>>>
>>>>> +#define CLKGATE_DIS_MISC_MMIO(0x46534)
>>>>> +#define  CLKGATE_DIS_MISC_DMASC_GATING_DISREG_BIT(21)
>>>>> +
>>>>>     #endif /* _I915_REG_H_ */
>>>>>
>>>
>