drm/amdkfd: Add 64-bit doorbell and wptr support to kernel queue

v2: Removed redundant 0x before %p.
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>

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

@@ -214,6 +214,16 @@ void write_kernel_doorbell(void __iomem *db, u32 value)
 	}
 }
 
+void write_kernel_doorbell64(void __iomem *db, u64 value)
+{
+	if (db) {
+		WARN(((unsigned long)db & 7) != 0,
+		     "Unaligned 64-bit doorbell");
+		writeq(value, (u64 __iomem *)db);
+		pr_debug("writing %llu to doorbell address %p\n", value, db);
+	}
+}
+
 unsigned int kfd_doorbell_id_to_offset(struct kfd_dev *kfd,
 					struct kfd_process *process,
 					unsigned int doorbell_id)

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

@@ -99,7 +99,7 @@ static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	kq->rptr_kernel = kq->rptr_mem->cpu_ptr;
 	kq->rptr_gpu_addr = kq->rptr_mem->gpu_addr;
 
-	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->wptr_kernel),
+	retval = kfd_gtt_sa_allocate(dev, dev->device_info->doorbell_size,
 					&kq->wptr_mem);
 
 	if (retval != 0)
@@ -208,6 +208,7 @@ static int acquire_packet_buffer(struct kernel_queue *kq,
 	size_t available_size;
 	size_t queue_size_dwords;
 	uint32_t wptr, rptr;
+	uint64_t wptr64;
 	unsigned int *queue_address;
 
 	/* When rptr == wptr, the buffer is empty.
@@ -216,7 +217,8 @@ static int acquire_packet_buffer(struct kernel_queue *kq,
 	 * the opposite. So we can only use up to queue_size_dwords - 1 dwords.
 	 */
 	rptr = *kq->rptr_kernel;
-	wptr = *kq->wptr_kernel;
+	wptr = kq->pending_wptr;
+	wptr64 = kq->pending_wptr64;
 	queue_address = (unsigned int *)kq->pq_kernel_addr;
 	queue_size_dwords = kq->queue->properties.queue_size / 4;
 
@@ -246,11 +248,13 @@ static int acquire_packet_buffer(struct kernel_queue *kq,
 		while (wptr > 0) {
 			queue_address[wptr] = kq->nop_packet;
 			wptr = (wptr + 1) % queue_size_dwords;
+			wptr64++;
 		}
 	}
 
 	*buffer_ptr = &queue_address[wptr];
 	kq->pending_wptr = wptr + packet_size_in_dwords;
+	kq->pending_wptr64 = wptr64 + packet_size_in_dwords;
 
 	return 0;
 
@@ -272,14 +276,18 @@ static void submit_packet(struct kernel_queue *kq)
 	pr_debug("\n");
 #endif
 
-	*kq->wptr_kernel = kq->pending_wptr;
-	write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
-				kq->pending_wptr);
+	kq->ops_asic_specific.submit_packet(kq);
 }
 
 static void rollback_packet(struct kernel_queue *kq)
 {
-	kq->pending_wptr = *kq->wptr_kernel;
+	if (kq->dev->device_info->doorbell_size == 8) {
+		kq->pending_wptr64 = *kq->wptr64_kernel;
+		kq->pending_wptr = *kq->wptr_kernel %
+			(kq->queue->properties.queue_size / 4);
+	} else {
+		kq->pending_wptr = *kq->wptr_kernel;
+	}
 }
 
 struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
@@ -310,6 +318,11 @@ struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
 	case CHIP_HAWAII:
 		kernel_queue_init_cik(&kq->ops_asic_specific);
 		break;
+
+	case CHIP_VEGA10:
+	case CHIP_RAVEN:
+		kernel_queue_init_v9(&kq->ops_asic_specific);
+		break;
 	default:
 		WARN(1, "Unexpected ASIC family %u",
 		     dev->device_info->asic_family);

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h

@@ -72,6 +72,7 @@ struct kernel_queue {
 	struct kfd_dev		*dev;
 	struct mqd_manager	*mqd;
 	struct queue		*queue;
+	uint64_t		pending_wptr64;
 	uint32_t		pending_wptr;
 	unsigned int		nop_packet;
 
@@ -79,7 +80,10 @@ struct kernel_queue {
 	uint32_t		*rptr_kernel;
 	uint64_t		rptr_gpu_addr;
 	struct kfd_mem_obj	*wptr_mem;
-	uint32_t		*wptr_kernel;
+	union {
+		uint64_t	*wptr64_kernel;
+		uint32_t	*wptr_kernel;
+	};
 	uint64_t		wptr_gpu_addr;
 	struct kfd_mem_obj	*pq;
 	uint64_t		pq_gpu_addr;
@@ -97,5 +101,6 @@ struct kernel_queue {
 
 void kernel_queue_init_cik(struct kernel_queue_ops *ops);
 void kernel_queue_init_vi(struct kernel_queue_ops *ops);
+void kernel_queue_init_v9(struct kernel_queue_ops *ops);
 
 #endif /* KFD_KERNEL_QUEUE_H_ */

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

@@ -26,11 +26,13 @@
 static bool initialize_cik(struct kernel_queue *kq, struct kfd_dev *dev,
 			enum kfd_queue_type type, unsigned int queue_size);
 static void uninitialize_cik(struct kernel_queue *kq);
+static void submit_packet_cik(struct kernel_queue *kq);
 
 void kernel_queue_init_cik(struct kernel_queue_ops *ops)
 {
 	ops->initialize = initialize_cik;
 	ops->uninitialize = uninitialize_cik;
+	ops->submit_packet = submit_packet_cik;
 }
 
 static bool initialize_cik(struct kernel_queue *kq, struct kfd_dev *dev,
@@ -42,3 +44,10 @@ static bool initialize_cik(struct kernel_queue *kq, struct kfd_dev *dev,
 static void uninitialize_cik(struct kernel_queue *kq)
 {
 }
+
+static void submit_packet_cik(struct kernel_queue *kq)
+{
+	*kq->wptr_kernel = kq->pending_wptr;
+	write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
+				kq->pending_wptr);
+}

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v9.c

@@ -29,11 +29,13 @@
 static bool initialize_v9(struct kernel_queue *kq, struct kfd_dev *dev,
 			enum kfd_queue_type type, unsigned int queue_size);
 static void uninitialize_v9(struct kernel_queue *kq);
+static void submit_packet_v9(struct kernel_queue *kq);
 
 void kernel_queue_init_v9(struct kernel_queue_ops *ops)
 {
 	ops->initialize = initialize_v9;
 	ops->uninitialize = uninitialize_v9;
+	ops->submit_packet = submit_packet_v9;
 }
 
 static bool initialize_v9(struct kernel_queue *kq, struct kfd_dev *dev,
@@ -58,6 +60,13 @@ static void uninitialize_v9(struct kernel_queue *kq)
 	kfd_gtt_sa_free(kq->dev, kq->eop_mem);
 }
 
+static void submit_packet_v9(struct kernel_queue *kq)
+{
+	*kq->wptr64_kernel = kq->pending_wptr64;
+	write_kernel_doorbell64(kq->queue->properties.doorbell_ptr,
+				kq->pending_wptr64);
+}
+
 static int pm_map_process_v9(struct packet_manager *pm,
 		uint32_t *buffer, struct qcm_process_device *qpd)
 {

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

@@ -29,11 +29,13 @@
 static bool initialize_vi(struct kernel_queue *kq, struct kfd_dev *dev,
 			enum kfd_queue_type type, unsigned int queue_size);
 static void uninitialize_vi(struct kernel_queue *kq);
+static void submit_packet_vi(struct kernel_queue *kq);
 
 void kernel_queue_init_vi(struct kernel_queue_ops *ops)
 {
 	ops->initialize = initialize_vi;
 	ops->uninitialize = uninitialize_vi;
+	ops->submit_packet = submit_packet_vi;
 }
 
 static bool initialize_vi(struct kernel_queue *kq, struct kfd_dev *dev,
@@ -58,6 +60,13 @@ static void uninitialize_vi(struct kernel_queue *kq)
 	kfd_gtt_sa_free(kq->dev, kq->eop_mem);
 }
 
+static void submit_packet_vi(struct kernel_queue *kq)
+{
+	*kq->wptr_kernel = kq->pending_wptr;
+	write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
+				kq->pending_wptr);
+}
+
 unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size)
 {
 	union PM4_MES_TYPE_3_HEADER header;

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -769,6 +769,7 @@ void __iomem *kfd_get_kernel_doorbell(struct kfd_dev *kfd,
 void kfd_release_kernel_doorbell(struct kfd_dev *kfd, u32 __iomem *db_addr);
 u32 read_kernel_doorbell(u32 __iomem *db);
 void write_kernel_doorbell(void __iomem *db, u32 value);
+void write_kernel_doorbell64(void __iomem *db, u64 value);
 unsigned int kfd_doorbell_id_to_offset(struct kfd_dev *kfd,
 					struct kfd_process *process,
 					unsigned int doorbell_id);