Re: [PATCH 1/1] drm/msm: Expose uche trap base via uapi

[Danylo <danylo.piliaiev@gmail.com>]

On 12/3/24 13:52, Dmitry Baryshkov wrote:
> On Tue, Dec 03, 2024 at 10:59:20AM +0100, Danylo Piliaiev wrote:
>> This adds MSM_PARAM_UCHE_TRAP_BASE that will be used by Mesa
>> implementation for VK_KHR_shader_clock and GL_ARB_shader_clock.
> Could you please spend more words, describing what it is and why is it
> necessary for those extensions. For a5xx+ it looks like some kind of an
> internal address (what kind of?). For a4xx I'm completely lost.
Yes, my bad. On at least a6xx+, shader could read 64b ALWAYSON counter
from UCHE_TRAP_BASE+0 address. We use it to implement VK_KHR_shader_clock:
  "This extension advertises the SPIR-V ShaderClockKHR capability for 
Vulkan, which
   allows a shader to query a real-time or monotonically incrementing 
counter at
   the subgroup level or across the device level."
And there is no other proper way to get such counter. Same with 
GL_ARB_shader_clock.

Not sure what's there on older gens, I exposed the value on them for the
completeness sake. But I don't know whether it is desired or not (I 
don't expect
the value being read and used on a4xx/a5xx in Mesa).

>> Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
>> ---
>>   drivers/gpu/drm/msm/adreno/a4xx_gpu.c   |  6 ++++--
>>   drivers/gpu/drm/msm/adreno/a5xx_gpu.c   | 10 ++++++----
>>   drivers/gpu/drm/msm/adreno/a6xx_gpu.c   | 12 +++++++-----
>>   drivers/gpu/drm/msm/adreno/adreno_gpu.c |  3 +++
>>   drivers/gpu/drm/msm/adreno/adreno_gpu.h |  2 ++
>>   include/uapi/drm/msm_drm.h              |  1 +
>>   6 files changed, 23 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
>> index 50c490b492f0..f1b18a6663f7 100644
>> --- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
>> +++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
>> @@ -251,8 +251,8 @@ static int a4xx_hw_init(struct msm_gpu *gpu)
>>   		gpu_write(gpu, REG_A4XX_UCHE_CACHE_WAYS_VFD, 0x07);
>>   
>>   	/* Disable L2 bypass to avoid UCHE out of bounds errors */
>> -	gpu_write(gpu, REG_A4XX_UCHE_TRAP_BASE_LO, 0xffff0000);
>> -	gpu_write(gpu, REG_A4XX_UCHE_TRAP_BASE_HI, 0xffff0000);
>> +	gpu_write(gpu, REG_A4XX_UCHE_TRAP_BASE_LO, lower_32_bits(adreno_gpu->uche_trap_base));
>> +	gpu_write(gpu, REG_A4XX_UCHE_TRAP_BASE_HI, upper_32_bits(adreno_gpu->uche_trap_base));
>>   
>>   	gpu_write(gpu, REG_A4XX_CP_DEBUG, (1 << 25) |
>>   			(adreno_is_a420(adreno_gpu) ? (1 << 29) : 0));
>> @@ -693,6 +693,8 @@ struct msm_gpu *a4xx_gpu_init(struct drm_device *dev)
>>   	if (ret)
>>   		goto fail;
>>   
>> +	adreno_gpu->uche_trap_base = 0xffff0000ffff0000ull;
>> +
>>   	if (!gpu->aspace) {
>>   		/* TODO we think it is possible to configure the GPU to
>>   		 * restrict access to VRAM carveout.  But the required
>> diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
>> index ee89db72e36e..caf2c0a7a29f 100644
>> --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
>> +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
>> @@ -750,10 +750,10 @@ static int a5xx_hw_init(struct msm_gpu *gpu)
>>   	gpu_write(gpu, REG_A5XX_UCHE_CACHE_WAYS, 0x02);
>>   
>>   	/* Disable L2 bypass in the UCHE */
>> -	gpu_write(gpu, REG_A5XX_UCHE_TRAP_BASE_LO, 0xFFFF0000);
>> -	gpu_write(gpu, REG_A5XX_UCHE_TRAP_BASE_HI, 0x0001FFFF);
>> -	gpu_write(gpu, REG_A5XX_UCHE_WRITE_THRU_BASE_LO, 0xFFFF0000);
>> -	gpu_write(gpu, REG_A5XX_UCHE_WRITE_THRU_BASE_HI, 0x0001FFFF);
>> +	gpu_write(gpu, REG_A5XX_UCHE_TRAP_BASE_LO, lower_32_bits(adreno_gpu->uche_trap_base));
>> +	gpu_write(gpu, REG_A5XX_UCHE_TRAP_BASE_HI, upper_32_bits(adreno_gpu->uche_trap_base));
>> +	gpu_write(gpu, REG_A5XX_UCHE_WRITE_THRU_BASE_LO, lower_32_bits(adreno_gpu->uche_trap_base));
>> +	gpu_write(gpu, REG_A5XX_UCHE_WRITE_THRU_BASE_HI, upper_32_bits(adreno_gpu->uche_trap_base));
>>   
>>   	/* Set the GMEM VA range (0 to gpu->gmem) */
>>   	gpu_write(gpu, REG_A5XX_UCHE_GMEM_RANGE_MIN_LO, 0x00100000);
>> @@ -1805,5 +1805,7 @@ struct msm_gpu *a5xx_gpu_init(struct drm_device *dev)
>>   	adreno_gpu->ubwc_config.macrotile_mode = 0;
>>   	adreno_gpu->ubwc_config.ubwc_swizzle = 0x7;
>>   
>> +	adreno_gpu->uche_trap_base = 0x0001ffffffff0000ull;
>> +
>>   	return gpu;
>>   }
>> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
>> index 019610341df1..0ae29a7c8a4d 100644
>> --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
>> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
>> @@ -1123,12 +1123,12 @@ static int hw_init(struct msm_gpu *gpu)
>>   
>>   	/* Disable L2 bypass in the UCHE */
>>   	if (adreno_is_a7xx(adreno_gpu)) {
>> -		gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, 0x0001fffffffff000llu);
>> -		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, 0x0001fffffffff000llu);
>> +		gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, adreno_gpu->uche_trap_base);
>> +		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, adreno_gpu->uche_trap_base);
>>   	} else {
>> -		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX, 0x0001ffffffffffc0llu);
>> -		gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, 0x0001fffffffff000llu);
>> -		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, 0x0001fffffffff000llu);
>> +		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_RANGE_MAX, adreno_gpu->uche_trap_base + 0xfc0);
>> +		gpu_write64(gpu, REG_A6XX_UCHE_TRAP_BASE, adreno_gpu->uche_trap_base);
>> +		gpu_write64(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE, adreno_gpu->uche_trap_base);
>>   	}
>>   
>>   	if (!(adreno_is_a650_family(adreno_gpu) ||
>> @@ -2533,6 +2533,8 @@ struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
>>   		}
>>   	}
>>   
>> +	adreno_gpu->uche_trap_base = 0x1fffffffff000ull;
>> +
>>   	if (gpu->aspace)
>>   		msm_mmu_set_fault_handler(gpu->aspace->mmu, gpu,
>>   				a6xx_fault_handler);
>> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
>> index 076be0473eb5..774ff6eacb9f 100644
>> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
>> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
>> @@ -385,6 +385,9 @@ int adreno_get_param(struct msm_gpu *gpu, struct msm_file_private *ctx,
>>   	case MSM_PARAM_MACROTILE_MODE:
>>   		*value = adreno_gpu->ubwc_config.macrotile_mode;
>>   		return 0;
>> +	case MSM_PARAM_UCHE_TRAP_BASE:
>> +		*value = adreno_gpu->uche_trap_base;
>> +		return 0;
>>   	default:
>>   		DBG("%s: invalid param: %u", gpu->name, param);
>>   		return -EINVAL;
>> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
>> index e71f420f8b3a..9bd38dda4308 100644
>> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
>> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
>> @@ -253,6 +253,8 @@ struct adreno_gpu {
>>   	bool gmu_is_wrapper;
>>   
>>   	bool has_ray_tracing;
>> +
>> +	u64 uche_trap_base;
>>   };
>>   #define to_adreno_gpu(x) container_of(x, struct adreno_gpu, base)
>>   
>> diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h
>> index b916aab80dde..2342cb90857e 100644
>> --- a/include/uapi/drm/msm_drm.h
>> +++ b/include/uapi/drm/msm_drm.h
>> @@ -90,6 +90,7 @@ struct drm_msm_timespec {
>>   #define MSM_PARAM_RAYTRACING 0x11 /* RO */
>>   #define MSM_PARAM_UBWC_SWIZZLE 0x12 /* RO */
>>   #define MSM_PARAM_MACROTILE_MODE 0x13 /* RO */
>> +#define MSM_PARAM_UCHE_TRAP_BASE 0x14 /* RO */
>>   
>>   /* For backwards compat.  The original support for preemption was based on
>>    * a single ring per priority level so # of priority levels equals the #
>> -- 
>> 2.46.2
>>