[Intel-gfx] [PATCH 13/22] drm/i915: CHV: Pipe level Gamma correction

[Sharma, Shashank]

Regards
Shashank

On 10/10/2015 4:37 AM, Emil Velikov wrote:
> Hi Shashank,
>
> On 9 October 2015 at 20:29, Shashank Sharma <shashank.sharma at intel.com> wrote:
>> CHV/BSW platform supports two different pipe level gamma
>> correction modes, which are:
>> 1. Legacy 8-bit mode
>> 2. 10-bit CGM (Color Gamut Mapping) mode
>>
>> This patch does the following:
>> 1. Attaches Gamma property to CRTC
>> 3. Adds the core Gamma correction function for CHV/BSW
>> 4. Adds Gamma correction macros
>>
>> Signed-off-by: Shashank Sharma <shashank.sharma at intel.com>
>> Signed-off-by: Kausal Malladi <kausalmalladi at gmail.com>
>> ---
>>   drivers/gpu/drm/i915/i915_reg.h            |  12 +++
>>   drivers/gpu/drm/i915/intel_color_manager.c | 114 +++++++++++++++++++++++++++++
>>   drivers/gpu/drm/i915/intel_color_manager.h |  13 ++++
>>   3 files changed, 139 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
>> index 60e6a73..885ac8a 100644
>> --- a/drivers/gpu/drm/i915/i915_reg.h
>> +++ b/drivers/gpu/drm/i915/i915_reg.h
>> @@ -8038,4 +8038,16 @@ enum skl_disp_power_wells {
>>   #define GEN9_VEBOX_MOCS_0      0xcb00  /* Video MOCS base register*/
>>   #define GEN9_BLT_MOCS_0                0xcc00  /* Blitter MOCS base register*/
>>
>> +/* Color Management */
>> +#define PIPEA_CGM_CONTROL                      (VLV_DISPLAY_BASE + 0x67A00)
>> +#define PIPEB_CGM_CONTROL                      (VLV_DISPLAY_BASE + 0x69A00)
>> +#define PIPEC_CGM_CONTROL                      (VLV_DISPLAY_BASE + 0x6BA00)
>> +#define PIPEA_CGM_GAMMA                        (VLV_DISPLAY_BASE + 0x67000)
>> +#define PIPEB_CGM_GAMMA                        (VLV_DISPLAY_BASE + 0x69000)
>> +#define PIPEC_CGM_GAMMA                        (VLV_DISPLAY_BASE + 0x6B000)
>> +#define _PIPE_CGM_CONTROL(pipe) \
>> +       (_PIPE3(pipe, PIPEA_CGM_CONTROL, PIPEB_CGM_CONTROL, PIPEC_CGM_CONTROL))
>> +#define _PIPE_GAMMA_BASE(pipe) \
>> +       (_PIPE3(pipe, PIPEA_CGM_GAMMA, PIPEB_CGM_GAMMA, PIPEC_CGM_GAMMA))
>> +
>>   #endif /* _I915_REG_H_ */
>> diff --git a/drivers/gpu/drm/i915/intel_color_manager.c b/drivers/gpu/drm/i915/intel_color_manager.c
>> index e466748..cf381b8 100644
>> --- a/drivers/gpu/drm/i915/intel_color_manager.c
>> +++ b/drivers/gpu/drm/i915/intel_color_manager.c
>> @@ -27,6 +27,112 @@
>>
>>   #include "intel_color_manager.h"
>>
>> +static int chv_set_gamma(struct drm_device *dev, struct drm_property_blob *blob,
>> +               struct drm_crtc *crtc)
>> +{
>> +       bool flag = false;
>> +       enum pipe pipe;
>> +       u16 red_fract, green_fract, blue_fract;
>> +       u32 red, green, blue, num_samples;
>> +       u32 word = 0;
>> +       u32 count = 0;
>> +       u32 cgm_gamma_reg = 0;
>> +       u32 cgm_control_reg = 0;
>> +       int ret = 0, length;
>> +       struct drm_r32g32b32 *correction_values = NULL;
> You can drop the useless initialization of correction_values. Same
> goes for patches 19 and 21.
Agree, thanks for pointing it out.
>
>> +       struct drm_palette *gamma_data;
>> +       struct drm_i915_private *dev_priv = dev->dev_private;
>> +
>> +       if (WARN_ON(!blob))
>> +               return -EINVAL;
>> +
>> +       gamma_data = (struct drm_palette *)blob->data;
>> +       pipe = to_intel_crtc(crtc)->pipe;
>> +       num_samples = gamma_data->num_samples;
>> +       length = num_samples * sizeof(struct drm_r32g32b32);
> Calculation can overflow.
good catch, will take care of this.
>
>> +
>> +       switch (num_samples) {
>> +       case GAMMA_DISABLE_VALS:
>> +
>> +               /* Disable Gamma functionality on Pipe - CGM Block */
>> +               cgm_control_reg = I915_READ(_PIPE_CGM_CONTROL(pipe));
>> +               cgm_control_reg &= ~CGM_GAMMA_EN;
>> +               I915_WRITE(_PIPE_CGM_CONTROL(pipe), cgm_control_reg);
>> +
>> +               DRM_DEBUG_DRIVER("Gamma disabled on Pipe %c\n",
>> +               pipe_name(pipe));
> Indentation looks wrong here.
>
Oops, bloody SI :).
>> +               ret = 0;
> Drop the variable and return 0, at the bottom of the function ?
>
Let me check this out.
>> +               break;
>> +
>> +       case CHV_8BIT_GAMMA_MAX_VALS:
>> +       case CHV_10BIT_GAMMA_MAX_VALS:
>> +
>> +               count = 0;
>> +               cgm_gamma_reg = _PIPE_GAMMA_BASE(pipe);
>> +               correction_values = (struct drm_r32g32b32 *)&gamma_data->lut;
>> +
>> +               while (count < num_samples) {
> Using for(i = 0;....) loop seems the more common approach ?
Nah, we are good with while. The whole color management series prefers 
while (and me too.... :))
>
> [snip]
>> +                       /*
>> +                       * On CHV, the best supported Gamma capability is
>> +                       * CGM block, that takes 257 samples.
>> +                       * If user gives 256 samples (legacy mode), then
>> +                       * duplicate the 256th value to 257th.
>> +                       * "flag" is used to make sure it happens only once
>> +                       */
>> +                       if (num_samples == CHV_8BIT_GAMMA_MAX_VALS &&
>> +                               count == CHV_8BIT_GAMMA_MAX_VALS && !flag) {
>> +                               count--;
>> +                               flag = true;
>> +                       }
> There is little point in going over this if statement 256 odd times.
> Split it out of the loop ?
Makes sense for sure, let me try this out.
>
>> +               }
>> +
>> +               /* Enable (CGM) Gamma on Pipe */
>> +               I915_WRITE(_PIPE_CGM_CONTROL(pipe),
>> +               I915_READ(_PIPE_CGM_CONTROL(pipe)) | CGM_GAMMA_EN);
>> +               DRM_DEBUG_DRIVER("CGM Gamma enabled on Pipe %c\n",
>> +               pipe_name(pipe));
> Indentation.
Guilty.
>
>> +               ret = 0;
> Drop the variable ?
Agree
>
>> +               break;
>> +
>> +       default:
>> +               DRM_ERROR("Invalid number of samples for Gamma LUT\n");
>> +               ret = -EINVAL;
> return -EINVAL;
Agree
>
>> +       }
>> +
>> +       return ret;
> return 0;
Agree
>
>
> Regards,
> Emil
>