drm/amdkfd: use kiq to load the mqd of hiq queue for gfx v9 (v6)

There is an issue that CP will check the HIQ queue to be configured and mapped
with KIQ ring, otherwise, it will be unable to read back the secure buffer while
the gfxoff is enabled even with trusted IP blocks.

v1 -> v2:
- Fix to remove surplus set_resources packets.
- Fill the whole configuration in MQD.
- Change the author as Aaron because he addressed the key point of this issue.
- Add kiq ring lock.

v2 -> v3:
- Free the lock while in error return case.
- Remove the programming only needed by the queue is unmapped.

v3 -> v4:
- Remove doorbell programming because it's used for restarting queue.
- Remove CP scheduler programming because map_queue packet will handle this.

v4 -> v5:
- Remove cp_hqd_active because mec ucode will enable it while use map_queues.
- Revise goto out_unlock.
- Correct the right doorbell offset for HIQ that kfd driver assigned in the
packet.

v5 -> v6:
- Merge Arcturus fix into this patch because it will get oops in Arcturus
platform.

Reported-by: Lisa Saturday <[e-mail]>
Signed-off-by: Aaron Liu <[e-mail]>
Signed-off-by: Huang Rui <[e-mail]>
Reviewed-and-Tested-by: Aaron Liu <[e-mail]>
Reviewed-by: Felix Kuehling <[e-mail]>
Signed-off-by: Alex Deucher <[e-mail]>

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c

···
 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
 	.init_interrupts = kgd_gfx_v9_init_interrupts,
 	.hqd_load = kgd_gfx_v9_hqd_load,
+	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
 	.hqd_sdma_load = kgd_hqd_sdma_load,
 	.hqd_dump = kgd_gfx_v9_hqd_dump,
 	.hqd_sdma_dump = kgd_hqd_sdma_dump,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c

···
 	lock_srbm(kgd, mec, pipe, queue_id, 0);
 }
 
-static uint32_t get_queue_mask(struct amdgpu_device *adev,
+static uint64_t get_queue_mask(struct amdgpu_device *adev,
 			       uint32_t pipe_id, uint32_t queue_id)
 {
-	unsigned int bit = (pipe_id * adev->gfx.mec.num_queue_per_pipe +
-			    queue_id) & 31;
+	unsigned int bit = pipe_id * adev->gfx.mec.num_queue_per_pipe +
+			queue_id;
 
-	return ((uint32_t)1) << bit;
+	return 1ull << bit;
 }
 
 static void release_queue(struct kgd_dev *kgd)
···
 
 	acquire_queue(kgd, pipe_id, queue_id);
 
-	/* HIQ is set during driver init period with vmid set to 0*/
-	if (m->cp_hqd_vmid == 0) {
-		uint32_t value, mec, pipe;
-
-		mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
-		pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
-
-		pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
-			mec, pipe, queue_id);
-		value = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CP_SCHEDULERS));
-		value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1,
-			((mec << 5) | (pipe << 3) | queue_id | 0x80));
-		WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CP_SCHEDULERS), value);
-	}
-
 	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
 	mqd_hqd = &m->cp_mqd_base_addr_lo;
 	hqd_base = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
···
 		WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI),
 		       upper_32_bits((uintptr_t)wptr));
 		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_PQ_WPTR_POLL_CNTL1),
-		       get_queue_mask(adev, pipe_id, queue_id));
+		       (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
 	}
 
 	/* Start the EOP fetcher */
···
 	release_queue(kgd);
 
 	return 0;
+}
+
+int kgd_gfx_v9_hiq_mqd_load(struct kgd_dev *kgd, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
+	struct v9_mqd *m;
+	uint32_t mec, pipe;
+	int r;
+
+	m = get_mqd(mqd);
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+		 mec, pipe, queue_id);
+
+	spin_lock(&adev->gfx.kiq.ring_lock);
+	r = amdgpu_ring_alloc(kiq_ring, 7);
+	if (r) {
+		pr_err("Failed to alloc KIQ (%d).\n", r);
+		goto out_unlock;
+	}
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
+			  PACKET3_MAP_QUEUES_VMID(m->cp_hqd_vmid) | /* VMID */
+			  PACKET3_MAP_QUEUES_QUEUE(queue_id) |
+			  PACKET3_MAP_QUEUES_PIPE(pipe) |
+			  PACKET3_MAP_QUEUES_ME((mec - 1)) |
+			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
+			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
+			  PACKET3_MAP_QUEUES_ENGINE_SEL(1) | /* engine_sel: hiq */
+			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(doorbell_off));
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_hi);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_hi);
+	amdgpu_ring_commit(kiq_ring);
+
+out_unlock:
+	spin_unlock(&adev->gfx.kiq.ring_lock);
+	release_queue(kgd);
+
+	return r;
 }
 
 int kgd_gfx_v9_hqd_dump(struct kgd_dev *kgd,
···
 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
 	.init_interrupts = kgd_gfx_v9_init_interrupts,
 	.hqd_load = kgd_gfx_v9_hqd_load,
+	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
 	.hqd_sdma_load = kgd_hqd_sdma_load,
 	.hqd_dump = kgd_gfx_v9_hqd_dump,
 	.hqd_sdma_dump = kgd_hqd_sdma_dump,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h

···
 			uint32_t queue_id, uint32_t __user *wptr,
 			uint32_t wptr_shift, uint32_t wptr_mask,
 			struct mm_struct *mm);
+int kgd_gfx_v9_hiq_mqd_load(struct kgd_dev *kgd, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off);
 int kgd_gfx_v9_hqd_dump(struct kgd_dev *kgd,
 			uint32_t pipe_id, uint32_t queue_id,
 			uint32_t (**dump)[2], uint32_t *n_regs);

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c

···
 					  wptr_shift, 0, mms);
 }
 
+static int hiq_load_mqd_kiq(struct mqd_manager *mm, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    struct queue_properties *p, struct mm_struct *mms)
+{
+	return mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->kgd, mqd, pipe_id,
+					      queue_id, p->doorbell_off);
+}
+
 static void update_mqd(struct mqd_manager *mm, void *mqd,
 		      struct queue_properties *q)
 {
···
 		mqd->allocate_mqd = allocate_hiq_mqd;
 		mqd->init_mqd = init_mqd_hiq;
 		mqd->free_mqd = free_mqd_hiq_sdma;
-		mqd->load_mqd = load_mqd;
+		mqd->load_mqd = hiq_load_mqd_kiq;
 		mqd->update_mqd = update_mqd;
 		mqd->destroy_mqd = destroy_mqd;
 		mqd->is_occupied = is_occupied;

drivers/gpu/drm/amd/include/kgd_kfd_interface.h

···
 			uint32_t wptr_shift, uint32_t wptr_mask,
 			struct mm_struct *mm);
 
+	int (*hiq_mqd_load)(struct kgd_dev *kgd, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off);
+
 	int (*hqd_sdma_load)(struct kgd_dev *kgd, void *mqd,
 			     uint32_t __user *wptr, struct mm_struct *mm);