libhsakmt: Use node_id in topology_is_svm_needed()

Signed-off-by: Graham Sider <Graham.Sider@amd.com>
Change-Id: I7773ccb959e60e43b816477c5659f1fe835ecbbd

src/fmm.c

@@ -1318,7 +1318,8 @@ static void *__fmm_allocate_device(uint32_t gpu_id, void *address, uint64_t Memo
 	return mem;
 }
 
-void *fmm_allocate_device(uint32_t gpu_id, void *address, uint64_t MemorySizeInBytes, HsaMemFlags mflags)
+void *fmm_allocate_device(uint32_t gpu_id, uint32_t node_id, void *address,
+			  uint64_t MemorySizeInBytes, HsaMemFlags mflags)
 {
 	manageable_aperture_t *aperture;
 	int32_t gpu_mem_id;
@@ -1339,7 +1340,7 @@ void *fmm_allocate_device(uint32_t gpu_id, void *address, uint64_t MemorySizeInB
 
 	ioc_flags |= fmm_translate_hsa_to_ioc_flags(mflags);
 
-	if (topology_is_svm_needed(get_device_id_by_gpu_id(gpu_id))) {
+	if (topology_is_svm_needed(node_id)) {
 		aperture = svm.dgpu_aperture;
 		if (mflags.ui32.AQLQueueMemory)
 			size = MemorySizeInBytes * 2;
@@ -2386,7 +2387,7 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
 	fmm_init_rbtree();
 
 	for (gpu_mem_id = 0; (uint32_t)gpu_mem_id < gpu_mem_count; gpu_mem_id++) {
-		if (!topology_is_svm_needed(gpu_mem[gpu_mem_id].device_id))
+		if (!topology_is_svm_needed(gpu_mem[gpu_mem_id].node_id))
 			continue;
 		gpu_mem[gpu_mem_id].mmio_aperture.base = map_mmio(
 				gpu_mem[gpu_mem_id].node_id,
@@ -3170,7 +3171,7 @@ HSAKMT_STATUS fmm_register_graphics_handle(HSAuint64 GraphicsResourceHandle,
 	gpu_mem_id = gpu_mem_find_by_gpu_id(infoArgs.gpu_id);
 	if (gpu_mem_id < 0)
 		goto error_free_metadata;
-	if (topology_is_svm_needed(gpu_mem[gpu_mem_id].device_id)) {
+	if (topology_is_svm_needed(gpu_mem[gpu_mem_id].node_id)) {
 		aperture = svm.dgpu_aperture;
 		aperture_base = NULL;
 	} else {

src/fmm.h

@@ -50,7 +50,8 @@ void fmm_destroy_process_apertures(void);
 
 /* Memory interface */
 void *fmm_allocate_scratch(uint32_t gpu_id, void *address, uint64_t MemorySizeInBytes);
-void *fmm_allocate_device(uint32_t gpu_id, void *address, uint64_t MemorySizeInBytes, HsaMemFlags flags);
+void *fmm_allocate_device(uint32_t gpu_id, uint32_t node_id, void *address,
+			uint64_t MemorySizeInBytes, HsaMemFlags flags);
 void *fmm_allocate_doorbell(uint32_t gpu_id, uint64_t MemorySizeInBytes, uint64_t doorbell_offset);
 void *fmm_allocate_host(uint32_t node_id, void *address, uint64_t MemorySizeInBytes,
 			HsaMemFlags flags);

src/libhsakmt.h

@@ -214,7 +214,7 @@ HSAKMT_STATUS topology_sysfs_get_node_props(uint32_t node_id, HsaNodeProperties
 					bool *p2p_links, uint32_t *num_p2pLinks);
 HSAKMT_STATUS topology_sysfs_get_system_props(HsaSystemProperties *props);
 void topology_setup_is_dgpu_param(HsaNodeProperties *props);
-bool topology_is_svm_needed(uint16_t device_id);
+bool topology_is_svm_needed(uint32_t node_id);
 HSAKMT_STATUS topology_get_asic_family(uint16_t device_id,
 					enum asic_family_type *asic);
 

src/memory.c

@@ -171,7 +171,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtAllocMemory(HSAuint32 PreferredNode,
 	}
 
 	/* GPU allocated VRAM */
-	*MemoryAddress = fmm_allocate_device(gpu_id, *MemoryAddress, SizeInBytes, MemFlags);
+	*MemoryAddress = fmm_allocate_device(gpu_id, PreferredNode, *MemoryAddress,
+					     SizeInBytes, MemFlags);
 
 	if (!(*MemoryAddress)) {
 		pr_err("[%s] failed to allocate %lu bytes from device\n",

src/topology.c

@@ -798,16 +798,12 @@ void topology_setup_is_dgpu_param(HsaNodeProperties *props)
 		is_dgpu = true;
 }
 
-bool topology_is_svm_needed(uint16_t device_id)
+bool topology_is_svm_needed(uint32_t node_id)
 {
-	const struct hsa_gfxip_table *hsa_gfxip;
-
 	if (is_dgpu)
 		return true;
 
-	hsa_gfxip = find_hsa_gfxip_device(device_id);
-
-	if (hsa_gfxip && hsa_gfxip->asic_family >= CHIP_VEGA10)
+	if (get_gfxv_by_node_id(node_id) >= GFX_VERSION_VEGA10)
 		return true;
 
 	return false;