libhsakmt: Support contiguous VRAM allocation flag

Add HsaMemFlags Contiguous bit for hsaKmtAllocMemory to allocate
contiguous VRAM, to support RDMA device with limited scatter-gather
ability.

Check KFD ioctl minor version >= 17.

Change-Id: I0db00dad125b2b7be523f343082641f59b850423
Signed-off-by: Philip Yang <Philip.Yang@amd.com>


[ROCm/ROCR-Runtime commit: 97497c7efc]

projects/rocr-runtime/include/hsakmt/hsakmttypes.h

@@ -569,7 +569,8 @@ typedef struct _HsaMemFlags
             unsigned int ExtendedCoherent: 1;  // system-scope coherence on atomic instructions
             unsigned int GTTAccess:     1;  // default = 0; If 1: The caller indicates this memory will be mapped to GART for MES
 					    // KFD will allocate GTT memory with the Preferred_node set as gpu_id for GART mapping
-            unsigned int Reserved:     10;
+            unsigned int Contiguous:	1; // Allocate contiguous VRAM
+            unsigned int Reserved:      9;
 
         } ui32;
         HSAuint32 Value;

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

@@ -1029,6 +1029,7 @@ struct kfd_ioctl_acquire_vm_args {
 #define KFD_IOC_ALLOC_MEM_FLAGS_COHERENT	(1 << 26)
 #define KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED	(1 << 25)
 #define KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT	(1 << 24)
+#define KFD_IOC_ALLOC_MEM_FLAGS_CONTIGUOUS_BEST_EFFORT	(1 << 23)
 
 /* Allocate memory for later SVM (shared virtual memory) mapping.
  *

projects/rocr-runtime/src/fmm.c

@@ -1579,6 +1579,9 @@ void *fmm_allocate_device(uint32_t gpu_id, uint32_t node_id, void *address,
 	if (mflags.ui32.ExtendedCoherent)
 		ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT;
 
+	if (mflags.ui32.Contiguous)
+		ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_CONTIGUOUS_BEST_EFFORT;
+
 	mem = __fmm_allocate_device(gpu_id, address, size, aperture, &mmap_offset,
 				    ioc_flags, &vm_obj);
 

projects/rocr-runtime/src/memory.c

@@ -115,6 +115,9 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtAllocMemory(HSAuint32 PreferredNode,
 
 	CHECK_KFD_OPEN();
 
+	if (MemFlags.ui32.Contiguous)
+		CHECK_KFD_MINOR_VERSION(17);
+
 	pr_debug("[%s] node %d\n", __func__, PreferredNode);
 
 	result = validate_nodeid(PreferredNode, &gpu_id);