2024-04-05 08:42:37

by Konrad Dybcio

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

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



2024-04-06 03:23:59

by Dmitry Baryshkov

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

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 <[email protected]>
> ---
> 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

2024-04-06 10:33:48

by kernel test robot

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

Hi Konrad,

kernel test robot noticed the following build errors:

[auto build test ERROR on 2b3d5988ae2cb5cd945ddbc653f0a71706231fdd]

url: https://github.com/intel-lab-lkp/linux/commits/Konrad-Dybcio/soc-qcom-Move-some-socinfo-defines-to-the-header-expand-them/20240405-164231
base: 2b3d5988ae2cb5cd945ddbc653f0a71706231fdd
patch link: https://lore.kernel.org/r/20240405-topic-smem_speedbin-v1-4-ce2b864251b1%40linaro.org
patch subject: [PATCH 4/6] drm/msm/adreno: Implement SMEM-based speed bin
config: i386-buildonly-randconfig-001-20240406 (https://download.01.org/0day-ci/archive/20240406/[email protected]/config)
compiler: clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240406/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

>> drivers/gpu/drm/msm/adreno/adreno_gpu.c:1090:14: error: call to undeclared function 'FIELD_PREP'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
1090 | *speedbin = FIELD_PREP(ADRENO_SKU_ID_PCODE, pcode) |
| ^
1 error generated.


vim +/FIELD_PREP +1090 drivers/gpu/drm/msm/adreno/adreno_gpu.c

1062
1063 int adreno_read_speedbin(struct adreno_gpu *adreno_gpu,
1064 struct device *dev, u32 *speedbin)
1065 {
1066 u32 fcode, pcode;
1067 int ret;
1068
1069 /* Try reading the speedbin via a nvmem cell first */
1070 ret = nvmem_cell_read_variable_le_u32(dev, "speed_bin", speedbin);
1071 if (!ret && ret != -EINVAL)
1072 return ret;
1073
1074 ret = qcom_smem_get_feature_code(&fcode);
1075 if (ret) {
1076 dev_err(dev, "Couldn't get feature code from SMEM!\n");
1077 return ret;
1078 }
1079
1080 ret = qcom_smem_get_product_code(&pcode);
1081 if (ret) {
1082 dev_err(dev, "Couldn't get product code from SMEM!\n");
1083 return ret;
1084 }
1085
1086 /* Don't consider fcode for external feature codes */
1087 if (fcode <= SOCINFO_FC_EXT_RESERVE)
1088 fcode = SOCINFO_FC_UNKNOWN;
1089
> 1090 *speedbin = FIELD_PREP(ADRENO_SKU_ID_PCODE, pcode) |
1091 FIELD_PREP(ADRENO_SKU_ID_FCODE, fcode);
1092
1093 return ret;
1094 }
1095

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

2024-04-06 10:43:47

by kernel test robot

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

Hi Konrad,

kernel test robot noticed the following build errors:

[auto build test ERROR on 2b3d5988ae2cb5cd945ddbc653f0a71706231fdd]

url: https://github.com/intel-lab-lkp/linux/commits/Konrad-Dybcio/soc-qcom-Move-some-socinfo-defines-to-the-header-expand-them/20240405-164231
base: 2b3d5988ae2cb5cd945ddbc653f0a71706231fdd
patch link: https://lore.kernel.org/r/20240405-topic-smem_speedbin-v1-4-ce2b864251b1%40linaro.org
patch subject: [PATCH 4/6] drm/msm/adreno: Implement SMEM-based speed bin
config: i386-buildonly-randconfig-003-20240406 (https://download.01.org/0day-ci/archive/20240406/[email protected]/config)
compiler: gcc-7 (Ubuntu 7.5.0-6ubuntu2) 7.5.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240406/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

drivers/gpu/drm/msm/adreno/adreno_gpu.c: In function 'adreno_read_speedbin':
>> drivers/gpu/drm/msm/adreno/adreno_gpu.c:1090:14: error: implicit declaration of function 'FIELD_PREP'; did you mean 'NEED_PGE'? [-Werror=implicit-function-declaration]
*speedbin = FIELD_PREP(ADRENO_SKU_ID_PCODE, pcode) |
^~~~~~~~~~
NEED_PGE
cc1: some warnings being treated as errors


vim +1090 drivers/gpu/drm/msm/adreno/adreno_gpu.c

1062
1063 int adreno_read_speedbin(struct adreno_gpu *adreno_gpu,
1064 struct device *dev, u32 *speedbin)
1065 {
1066 u32 fcode, pcode;
1067 int ret;
1068
1069 /* Try reading the speedbin via a nvmem cell first */
1070 ret = nvmem_cell_read_variable_le_u32(dev, "speed_bin", speedbin);
1071 if (!ret && ret != -EINVAL)
1072 return ret;
1073
1074 ret = qcom_smem_get_feature_code(&fcode);
1075 if (ret) {
1076 dev_err(dev, "Couldn't get feature code from SMEM!\n");
1077 return ret;
1078 }
1079
1080 ret = qcom_smem_get_product_code(&pcode);
1081 if (ret) {
1082 dev_err(dev, "Couldn't get product code from SMEM!\n");
1083 return ret;
1084 }
1085
1086 /* Don't consider fcode for external feature codes */
1087 if (fcode <= SOCINFO_FC_EXT_RESERVE)
1088 fcode = SOCINFO_FC_UNKNOWN;
1089
> 1090 *speedbin = FIELD_PREP(ADRENO_SKU_ID_PCODE, pcode) |
1091 FIELD_PREP(ADRENO_SKU_ID_FCODE, fcode);
1092
1093 return ret;
1094 }
1095

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

2024-04-10 11:42:45

by Konrad Dybcio

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



On 4/6/24 05:23, Dmitry Baryshkov wrote:
> 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 <[email protected]>
>> ---

[...]

>> - 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.

Right, a better condition would be (!ret || ret != EINVAL)..


>
>> + 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.

OK, I had this in mind early on, but forgot to actually impl it.

>
>> + }
>> +
>> + 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.

No, the FCODE & PCODE can be interpreted differently across consumers.

>
>> +
>> + 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?

Ok, I can keep it, though realistically if this ever does anything
useful, it likely means the dt is wrong

Konrad

2024-04-10 19:26:51

by Dmitry Baryshkov

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

On Wed, Apr 10, 2024 at 01:42:33PM +0200, Konrad Dybcio wrote:
>
>
> On 4/6/24 05:23, Dmitry Baryshkov wrote:
> > 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 <[email protected]>
> > > ---
>
> [...]
>
> >
> > > + }
> > > +
> > > + 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.
>
> No, the FCODE & PCODE can be interpreted differently across consumers.

That's why I wrote about asking for 'gpu_bin'.

>
> >
> > > +
> > > + 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?
>
> Ok, I can keep it, though realistically if this ever does anything
> useful, it likely means the dt is wrong

Yes, but if DT is wrong, we should probably fail in a sensible way. I
just wanted to point out that previously we had this &0xffff, while your
patch silently removes it.

--
With best wishes
Dmitry

2024-04-11 21:35:58

by Konrad Dybcio

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



On 4/10/24 21:26, Dmitry Baryshkov wrote:
> On Wed, Apr 10, 2024 at 01:42:33PM +0200, Konrad Dybcio wrote:
>>
>>
>> On 4/6/24 05:23, Dmitry Baryshkov wrote:
>>> 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 <[email protected]>
>>>> ---
>>
>> [...]
>>
>>>
>>>> + }
>>>> +
>>>> + 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.
>>
>> No, the FCODE & PCODE can be interpreted differently across consumers.
>
> That's why I wrote about asking for 'gpu_bin'.

I'd rather keep the magic GPU LUTs inside the adreno driver, especially
since not all Snapdragons feature Adreno, but all Adrenos are on
Snapdragons (modulo a2xx but I refuse to make design decisions treating
these equally to e.g. a6xx)

>
>>
>>>
>>>> +
>>>> + 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?
>>
>> Ok, I can keep it, though realistically if this ever does anything
>> useful, it likely means the dt is wrong
>
> Yes, but if DT is wrong, we should probably fail in a sensible way. I
> just wanted to point out that previously we had this &0xffff, while your
> patch silently removes it.

Right, but I don't believe it actually matters.. If that AND ever did
anything, this was a silent failure with garbage data passed in anyway.

If you really insist, I can remove it separately.

Konrad

2024-04-11 21:46:56

by Dmitry Baryshkov

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

On Fri, 12 Apr 2024 at 00:35, Konrad Dybcio <[email protected]> wrote:
>
>
>
> On 4/10/24 21:26, Dmitry Baryshkov wrote:
> > On Wed, Apr 10, 2024 at 01:42:33PM +0200, Konrad Dybcio wrote:
> >>
> >>
> >> On 4/6/24 05:23, Dmitry Baryshkov wrote:
> >>> 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 <[email protected]>
> >>>> ---
> >>
> >> [...]
> >>
> >>>
> >>>> + }
> >>>> +
> >>>> + 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.
> >>
> >> No, the FCODE & PCODE can be interpreted differently across consumers.
> >
> > That's why I wrote about asking for 'gpu_bin'.
>
> I'd rather keep the magic GPU LUTs inside the adreno driver, especially
> since not all Snapdragons feature Adreno, but all Adrenos are on
> Snapdragons (modulo a2xx but I refuse to make design decisions treating
> these equally to e.g. a6xx)

LUTs - yes. I wanted to push (FC << a) | (PC << b) and all the RESERVE
/ UNKNOWN magic there.

>
> >
> >>
> >>>
> >>>> +
> >>>> + 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?
> >>
> >> Ok, I can keep it, though realistically if this ever does anything
> >> useful, it likely means the dt is wrong
> >
> > Yes, but if DT is wrong, we should probably fail in a sensible way. I
> > just wanted to point out that previously we had this &0xffff, while your
> > patch silently removes it.
>
> Right, but I don't believe it actually matters.. If that AND ever did
> anything, this was a silent failure with garbage data passed in anyway.
>
> If you really insist, I can remove it separately.

I'd say, up to Rob or up to your consideration.


--
With best wishes
Dmitry

2024-04-11 22:14:28

by Konrad Dybcio

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



On 4/11/24 23:46, Dmitry Baryshkov wrote:
> On Fri, 12 Apr 2024 at 00:35, Konrad Dybcio <[email protected]> wrote:
>>
>>
>>
>> On 4/10/24 21:26, Dmitry Baryshkov wrote:
>>> On Wed, Apr 10, 2024 at 01:42:33PM +0200, Konrad Dybcio wrote:
>>>>
>>>>
>>>> On 4/6/24 05:23, Dmitry Baryshkov wrote:
>>>>> 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 <[email protected]>
>>>>>> ---
>>>>
>>>> [...]
>>>>
>>>>>
>>>>>> + }
>>>>>> +
>>>>>> + 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.
>>>>
>>>> No, the FCODE & PCODE can be interpreted differently across consumers.
>>>
>>> That's why I wrote about asking for 'gpu_bin'.
>>
>> I'd rather keep the magic GPU LUTs inside the adreno driver, especially
>> since not all Snapdragons feature Adreno, but all Adrenos are on
>> Snapdragons (modulo a2xx but I refuse to make design decisions treating
>> these equally to e.g. a6xx)
>
> LUTs - yes. I wanted to push (FC << a) | (PC << b) and all the RESERVE
> / UNKNOWN magic there.

Ohh this specifically.. yeah I considered pushing that there as well,
but I realized this is specific to the GPU. The socinfo APIs should
only return a valid/unknown code for both P and F and let the consumer
figure out how to interpret these.

>
>>
>>>
>>>>
>>>>>
>>>>>> +
>>>>>> + 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?
>>>>
>>>> Ok, I can keep it, though realistically if this ever does anything
>>>> useful, it likely means the dt is wrong
>>>
>>> Yes, but if DT is wrong, we should probably fail in a sensible way. I
>>> just wanted to point out that previously we had this &0xffff, while your
>>> patch silently removes it.
>>
>> Right, but I don't believe it actually matters.. If that AND ever did
>> anything, this was a silent failure with garbage data passed in anyway.
>>
>> If you really insist, I can remove it separately.
>
> I'd say, up to Rob or up to your consideration.

Konrad