[PATCH v2 14/21] drm/msm/adreno: Support AQE engine

[Konrad Dybcio]

On 11/10/25 5:37 PM, Akhil P Oommen wrote:
> AQE (Applicaton Qrisc Engine) is a dedicated core inside CP which aides
> in Raytracing related workloads. Add support for loading the AQE firmware
> and initialize the necessary registers.
> 
> Since AQE engine has dependency on preemption context records, expose
> Raytracing support to userspace only when preemption is enabled.
> 
> Signed-off-by: Akhil P Oommen <akhilpo at oss.qualcomm.com>
> ---

Side question: is the AQE fully functional on any gen7 targets?
A750?

[...]

> -		adreno_gpu->has_ray_tracing =
> -			!!(fuse_val & A7XX_CX_MISC_SW_FUSE_VALUE_RAYTRACING);
> +		/*
> +		 * AQE requires preemption records, so disable raytracing
> +		 * if preemption is not supported
> +		 */
> +		if (gpu->nr_rings == 1)

"nr_rings > 1"?

> +			adreno_gpu->has_ray_tracing =
> +				!!(fuse_val & A7XX_CX_MISC_SW_FUSE_VALUE_RAYTRACING);
>  	} else if (adreno_is_a740(adreno_gpu)) {
>  		/* Raytracing is always enabled on a740 */
>  		adreno_gpu->has_ray_tracing = true;
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.h b/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
> index b507ff4e1756..7921b4a68937 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
> @@ -58,6 +58,8 @@ struct a6xx_gpu {
>  
>  	struct drm_gem_object *sqe_bo;
>  	uint64_t sqe_iova;
> +	struct drm_gem_object *aqe_bo;
> +	uint64_t aqe_iova;
>  
>  	struct msm_ringbuffer *cur_ring;
>  	struct msm_ringbuffer *next_ring;
> diff --git a/drivers/gpu/drm/msm/adreno/a8xx_gpu.c b/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
> index 78dcb9bc3377..2ef69161f1d0 100644
> --- a/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
> @@ -616,6 +616,9 @@ static int hw_init(struct msm_gpu *gpu)
>  		goto out;
>  
>  	gpu_write64(gpu, REG_A8XX_CP_SQE_INSTR_BASE, a6xx_gpu->sqe_iova);
> +	if (a6xx_gpu->aqe_iova)
> +		gpu_write64(gpu, REG_A8XX_CP_AQE_INSTR_BASE_0, a6xx_gpu->aqe_iova);

I believe you should also set CP_AQE_APRIV_CNTL per-pipe

Should we also enable AQE1 while at it, to reduce potential backwards
compatibility issues? Would that require solving the iommu woes?

Konrad