Allocate and map doorbells in SVM for discrete GPUs

Allocate doorbells for dGPUs in the SVM aperture and map them for
GPU access. This is necessary to allow GPU-initiated submissions to
user mode queues.

Depends on new doorbell BO allocation flag in KFD.

Change-Id: I0737bef4a4764bb4a66c43846707ead2108f6601


[ROCm/ROCR-Runtime commit: 2e0a6eb371]

projects/rocr-runtime/include/linux/kfd_ioctl.h

@@ -324,6 +324,8 @@ enum evict_type {
 
 #define KFD_IOC_ALLOC_MEM_FLAGS_USERPTR			(1 << 6)
 
+#define KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL		(1 << 7)
+
 struct kfd_ioctl_alloc_memory_of_gpu_new_args {
 	uint64_t va_addr;	/* to KFD */
 	uint64_t size;		/* to KFD */