Re: [PATCH 4/6] drm/msm/adreno: Implement SMEM-based speed bin

[Dmitry Baryshkov]

On Fri, Apr 05, 2024 at 10:41:32AM +0200, Konrad Dybcio wrote:
> On recent (SM8550+) Snapdragon platforms, the GPU speed bin data is
> abstracted through SMEM, instead of being directly available in a fuse.
> 
> Add support for SMEM-based speed binning, which includes getting
> "feature code" and "product code" from said source and parsing them
> to form something that lets us match OPPs against.
> 
> Signed-off-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxx>
> ---
>  drivers/gpu/drm/msm/adreno/a6xx_gpu.c      |  8 +++---
>  drivers/gpu/drm/msm/adreno/adreno_device.c |  2 ++
>  drivers/gpu/drm/msm/adreno/adreno_gpu.c    | 39 +++++++++++++++++++++++++++---
>  drivers/gpu/drm/msm/adreno/adreno_gpu.h    | 12 ++++++---
>  4 files changed, 51 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> index 4cbdfabbcee5..6776fd80f7a6 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> @@ -2890,13 +2890,15 @@ static u32 fuse_to_supp_hw(const struct adreno_info *info, u32 fuse)
>  	return UINT_MAX;
>  }
>  
> -static int a6xx_set_supported_hw(struct device *dev, const struct adreno_info *info)
> +static int a6xx_set_supported_hw(struct adreno_gpu *adreno_gpu,
> +				 struct device *dev,
> +				 const struct adreno_info *info)
>  {
>  	u32 supp_hw;
>  	u32 speedbin;
>  	int ret;
>  
> -	ret = adreno_read_speedbin(dev, &speedbin);
> +	ret = adreno_read_speedbin(adreno_gpu, dev, &speedbin);
>  	/*
>  	 * -ENOENT means that the platform doesn't support speedbin which is
>  	 * fine
> @@ -3056,7 +3058,7 @@ struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
>  
>  	a6xx_llc_slices_init(pdev, a6xx_gpu, is_a7xx);
>  
> -	ret = a6xx_set_supported_hw(&pdev->dev, config->info);
> +	ret = a6xx_set_supported_hw(adreno_gpu, &pdev->dev, config->info);
>  	if (ret) {
>  		a6xx_destroy(&(a6xx_gpu->base.base));
>  		return ERR_PTR(ret);
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_device.c b/drivers/gpu/drm/msm/adreno/adreno_device.c
> index c3703a51287b..901ef767e491 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_device.c
> +++ b/drivers/gpu/drm/msm/adreno/adreno_device.c
> @@ -6,6 +6,8 @@
>   * Copyright (c) 2014,2017 The Linux Foundation. All rights reserved.
>   */
>  
> +#include <linux/soc/qcom/socinfo.h>
> +
>  #include "adreno_gpu.h"
>  
>  bool hang_debug = false;
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> index 074fb498706f..0e4ff532ac3c 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> @@ -21,6 +21,9 @@
>  #include "msm_gem.h"
>  #include "msm_mmu.h"
>  
> +#include <linux/soc/qcom/smem.h>
> +#include <linux/soc/qcom/socinfo.h>
> +
>  static u64 address_space_size = 0;
>  MODULE_PARM_DESC(address_space_size, "Override for size of processes private GPU address space");
>  module_param(address_space_size, ullong, 0600);
> @@ -1057,9 +1060,37 @@ void adreno_gpu_ocmem_cleanup(struct adreno_ocmem *adreno_ocmem)
>  			   adreno_ocmem->hdl);
>  }
>  
> -int adreno_read_speedbin(struct device *dev, u32 *speedbin)
> +int adreno_read_speedbin(struct adreno_gpu *adreno_gpu,
> +			 struct device *dev, u32 *speedbin)
>  {
> -	return nvmem_cell_read_variable_le_u32(dev, "speed_bin", speedbin);
> +	u32 fcode, pcode;
> +	int ret;
> +
> +	/* Try reading the speedbin via a nvmem cell first */
> +	ret = nvmem_cell_read_variable_le_u32(dev, "speed_bin", speedbin);
> +	if (!ret && ret != -EINVAL)

This is always false.

> +		return ret;
> +
> +	ret = qcom_smem_get_feature_code(&fcode);
> +	if (ret) {
> +		dev_err(dev, "Couldn't get feature code from SMEM!\n");
> +		return ret;

This brings in QCOM_SMEM dependency (which is not mentioned in the
Kconfig). Please keep iMX5 hardware in mind, so the dependency should be
optional. Respective functions should be stubbed in the header.

> +	}
> +
> +	ret = qcom_smem_get_product_code(&pcode);
> +	if (ret) {
> +		dev_err(dev, "Couldn't get product code from SMEM!\n");
> +		return ret;
> +	}
> +
> +	/* Don't consider fcode for external feature codes */
> +	if (fcode <= SOCINFO_FC_EXT_RESERVE)
> +		fcode = SOCINFO_FC_UNKNOWN;
> +
> +	*speedbin = FIELD_PREP(ADRENO_SKU_ID_PCODE, pcode) |
> +		    FIELD_PREP(ADRENO_SKU_ID_FCODE, fcode);

What about just asking the qcom_smem for the 'gpu_bin' and hiding gory
details there? It almost feels that handling raw PCODE / FCODE here is
too low-level and a subject to change depending on the socinfo format.

> +
> +	return ret;
>  }
>  
>  int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
> @@ -1098,9 +1129,9 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
>  			devm_pm_opp_set_clkname(dev, "core");
>  	}
>  
> -	if (adreno_read_speedbin(dev, &speedbin) || !speedbin)
> +	if (adreno_read_speedbin(adreno_gpu, dev, &speedbin) || !speedbin)
>  		speedbin = 0xffff;
> -	adreno_gpu->speedbin = (uint16_t) (0xffff & speedbin);

the &= 0xffff should probably go to the adreno_read_speedbin / nvmem
case. WDYT?

> +	adreno_gpu->speedbin = speedbin;
>  
>  	gpu_name = devm_kasprintf(dev, GFP_KERNEL, "%"ADRENO_CHIPID_FMT,
>  			ADRENO_CHIPID_ARGS(config->chip_id));
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> index 460b399be37b..1770a9e20484 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> @@ -81,7 +81,12 @@ extern const struct adreno_reglist a612_hwcg[], a615_hwcg[], a630_hwcg[], a640_h
>  extern const struct adreno_reglist a660_hwcg[], a690_hwcg[], a702_hwcg[], a730_hwcg[], a740_hwcg[];
>  
>  struct adreno_speedbin {
> -	uint16_t fuse;
> +	/* <= 16-bit for NVMEM fuses, 32b for SOCID values */
> +	uint32_t fuse;
> +#define ADRENO_SKU_ID_PCODE		GENMASK(31, 16)
> +#define ADRENO_SKU_ID_FCODE		GENMASK(15, 0)
> +#define ADRENO_SKU_ID(pcode, fcode)	(pcode << 16 | fcode)
> +
>  	uint16_t speedbin;
>  };
>  
> @@ -137,7 +142,7 @@ struct adreno_gpu {
>  	struct msm_gpu base;
>  	const struct adreno_info *info;
>  	uint32_t chip_id;
> -	uint16_t speedbin;
> +	uint32_t speedbin;
>  	const struct adreno_gpu_funcs *funcs;
>  
>  	/* interesting register offsets to dump: */
> @@ -520,7 +525,8 @@ int adreno_fault_handler(struct msm_gpu *gpu, unsigned long iova, int flags,
>  			 struct adreno_smmu_fault_info *info, const char *block,
>  			 u32 scratch[4]);
>  
> -int adreno_read_speedbin(struct device *dev, u32 *speedbin);
> +int adreno_read_speedbin(struct adreno_gpu *adreno_gpu,
> +			 struct device *dev, u32 *speedbin);
>  
>  /*
>   * For a5xx and a6xx targets load the zap shader that is used to pull the GPU
> 
> -- 
> 2.40.1
> 

-- 
With best wishes
Dmitry