diff options
Diffstat (limited to 'meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.19.8/2739-drm-amdkfd-Optimize-out-sdma-doorbell-array-in-kgd2k.patch')
| -rw-r--r-- | meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.19.8/2739-drm-amdkfd-Optimize-out-sdma-doorbell-array-in-kgd2k.patch | 83 |
1 files changed, 83 insertions, 0 deletions
diff --git a/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.19.8/2739-drm-amdkfd-Optimize-out-sdma-doorbell-array-in-kgd2k.patch b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.19.8/2739-drm-amdkfd-Optimize-out-sdma-doorbell-array-in-kgd2k.patch new file mode 100644 index 00000000..16db936f --- /dev/null +++ b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.19.8/2739-drm-amdkfd-Optimize-out-sdma-doorbell-array-in-kgd2k.patch @@ -0,0 +1,83 @@ +From e69e9007b78b1004c257d3d308bb837f4544a84a Mon Sep 17 00:00:00 2001 +From: Yong Zhao <Yong.Zhao@amd.com> +Date: Wed, 9 Jan 2019 23:31:14 -0500 +Subject: [PATCH 2739/2940] drm/amdkfd: Optimize out sdma doorbell array in + kgd2kfd_shared_resources + +We can directly calculate sdma doorbell indexes in the process doorbell +pages through the doorbell_index structure in amdgpu_device, so no need +to cache them in kgd2kfd_shared_resources any more. This alleviates the +adaptation needs when new SDMA configurations are introduced. + +Change-Id: Ic657799856ed0256f36b01e502ef0cab263b1f49 +Signed-off-by: Yong Zhao <Yong.Zhao@amd.com> +Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> +--- + drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 3 ++- + .../drm/amd/amdkfd/kfd_device_queue_manager.c | 16 +++++++++++----- + drivers/gpu/drm/amd/include/kgd_kfd_interface.h | 4 ++-- + 3 files changed, 15 insertions(+), 8 deletions(-) + +diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +index ea15e27f61f9..01b35b0455ca 100644 +--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c ++++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +@@ -145,7 +145,8 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) + .gpuvm_size = min(adev->vm_manager.max_pfn + << AMDGPU_GPU_PAGE_SHIFT, + AMDGPU_GMC_HOLE_START), +- .drm_render_minor = adev->ddev->render->index ++ .drm_render_minor = adev->ddev->render->index, ++ .sdma_doorbell_idx = adev->doorbell_index.sdma_engine, + }; + + /* this is going to have a few of the MSBs set that we need to +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +index 8372556b52eb..c6c9530e704e 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +@@ -134,12 +134,18 @@ static int allocate_doorbell(struct qcm_process_device *qpd, struct queue *q) + */ + q->doorbell_id = q->properties.queue_id; + } else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) { +- /* For SDMA queues on SOC15, use static doorbell +- * assignments based on the engine and queue. ++ /* For SDMA queues on SOC15 with 8-byte doorbell, use static ++ * doorbell assignments based on the engine and queue id. ++ * The doobell index distance between RLC (2*i) and (2*i+1) ++ * for a SDMA engine is 512. + */ +- q->doorbell_id = dev->shared_resources.sdma_doorbell +- [q->properties.sdma_engine_id] +- [q->properties.sdma_queue_id]; ++ uint32_t *idx_offset = ++ dev->shared_resources.sdma_doorbell_idx; ++ ++ q->doorbell_id = idx_offset[q->properties.sdma_engine_id] ++ + (q->properties.sdma_queue_id & 1) ++ * KFD_QUEUE_DOORBELL_MIRROR_OFFSET ++ + (q->properties.sdma_queue_id >> 1); + } else { + /* For CP queues on SOC15 reserve a free doorbell ID */ + unsigned int found; +diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h +index 7e44dcd15cdc..b71b684ad6fd 100644 +--- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h ++++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h +@@ -110,11 +110,11 @@ struct kgd2kfd_shared_resources { + /* Bit n == 1 means Queue n is available for KFD */ + DECLARE_BITMAP(queue_bitmap, KGD_MAX_QUEUES); + +- /* Doorbell assignments (SOC15 and later chips only). Only ++ /* SDMA doorbell assignments (SOC15 and later chips only). Only + * specific doorbells are routed to each SDMA engine. Others + * are routed to IH and VCN. They are not usable by the CP. + */ +- unsigned int sdma_doorbell[2][2]; ++ uint32_t *sdma_doorbell_idx; + + /* From SOC15 onward, the doorbell index range not usable for CP + * queues. +-- +2.17.1 + |