drm/amdgpu: Add submit IB function for KFD

This can be used for flushing caches when not using the HWS.

Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>

5 files changed, 69 insertions, 0 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 5a881107a15a..8a23aa8f9c73 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -367,6 +367,61 @@ uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
         return amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 }
 
+int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
+                                uint32_t vmid, uint64_t gpu_addr,
+                                uint32_t *ib_cmd, uint32_t ib_len)
+{
+        struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+        struct amdgpu_job *job;
+        struct amdgpu_ib *ib;
+        struct amdgpu_ring *ring;
+        struct dma_fence *f = NULL;
+        int ret;
+
+        switch (engine) {
+        case KGD_ENGINE_MEC1:
+                ring = &adev->gfx.compute_ring[0];
+                break;
+        case KGD_ENGINE_SDMA1:
+                ring = &adev->sdma.instance[0].ring;
+                break;
+        case KGD_ENGINE_SDMA2:
+                ring = &adev->sdma.instance[1].ring;
+                break;
+        default:
+                pr_err("Invalid engine in IB submission: %d\n", engine);
+                ret = -EINVAL;
+                goto err;
+        }
+
+        ret = amdgpu_job_alloc(adev, 1, &job, NULL);
+        if (ret)
+                goto err;
+
+        ib = &job->ibs[0];
+        memset(ib, 0, sizeof(struct amdgpu_ib));
+
+        ib->gpu_addr = gpu_addr;
+        ib->ptr = ib_cmd;
+        ib->length_dw = ib_len;
+        /* This works for NO_HWS. TODO: need to handle without knowing VMID */
+        job->vmid = vmid;
+
+        ret = amdgpu_ib_schedule(ring, 1, ib, job, &f);
+        if (ret) {
+                DRM_ERROR("amdgpu: failed to schedule IB.\n");
+                goto err_ib_sched;
+        }
+
+        ret = dma_fence_wait(f, false);
+
+err_ib_sched:
+        dma_fence_put(f);
+        amdgpu_job_free(job);
+err:
+        return ret;
+}
+
 bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
 {
         if (adev->kfd) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 05a228d60241..d7509b706b26 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -124,6 +124,10 @@ void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev);
 void amdgpu_amdkfd_device_init(struct amdgpu_device *adev);
 void amdgpu_amdkfd_device_fini(struct amdgpu_device *adev);
 
+int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
+                                uint32_t vmid, uint64_t gpu_addr,
+                                uint32_t *ib_cmd, uint32_t ib_len);
+
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void);
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
index 65783d1eddca..7485c376b90e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
@@ -217,6 +217,7 @@ static const struct kfd2kgd_calls kfd2kgd = {
         .restore_process_bos = amdgpu_amdkfd_gpuvm_restore_process_bos,
         .invalidate_tlbs = invalidate_tlbs,
         .invalidate_tlbs_vmid = invalidate_tlbs_vmid,
+        .submit_ib = amdgpu_amdkfd_submit_ib,
 };
 
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index 1b5bf1353f0c..7be453494423 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -177,6 +177,7 @@ static const struct kfd2kgd_calls kfd2kgd = {
         .restore_process_bos = amdgpu_amdkfd_gpuvm_restore_process_bos,
         .invalidate_tlbs = invalidate_tlbs,
         .invalidate_tlbs_vmid = invalidate_tlbs_vmid,
+        .submit_ib = amdgpu_amdkfd_submit_ib,
 };
 
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void)
diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
index 5984fec78be4..1e5c22ceb256 100644
--- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
@@ -240,6 +240,10 @@ struct tile_config {
  *
  * @invalidate_tlbs_vmid: Invalidate TLBs for a specific VMID
  *
+ * @submit_ib: Submits an IB to the engine specified by inserting the
+ * IB to the corresponding ring (ring type). The IB is executed with the
+ * specified VMID in a user mode context.
+ *
  * This structure contains function pointers to services that the kgd driver
  * provides to amdkfd driver.
  *
@@ -352,6 +356,10 @@ struct kfd2kgd_calls {
 
         int (*invalidate_tlbs)(struct kgd_dev *kgd, uint16_t pasid);
         int (*invalidate_tlbs_vmid)(struct kgd_dev *kgd, uint16_t vmid);
+
+        int (*submit_ib)(struct kgd_dev *kgd, enum kgd_engine_type engine,
+                        uint32_t vmid, uint64_t gpu_addr,
+                        uint32_t *ib_cmd, uint32_t ib_len);
 };
 
 /**