AI/rocm-systems
2
0
Fork 0
Código Incidencias Pull Requests Acciones Paquetes Proyectos Lanzamientos Wiki Actividad

libhsakmt: Map VRAM only on supported peer GPUs

Explorar el Código 
hsaKmtMapMemoryToGPU should not try to map VRAM on peer GPUs that don't
have an IO-Link to the memory. The new P2P mapping code in KFD will
fail otherwise.

Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Change-Id: I6d59b55651b98756865a0f69eafef3e386372cf3


[ROCm/ROCR-Runtime commit: 9ac2c75171]
Este commit está contenido en:
Felix Kuehling
2022-07-13 20:40:43 -04:00
padre e8ba7dc5c2
commit c1a3ea8a30
Se han modificado 3 ficheros con 96 adiciones y 20 borrados
Descargar archivo de parche Descargar archivo de diferencias Expandir todos los archivos Contraer todos los archivos
projects/rocr-runtime/src/fmm.c
+78 -13
Ver fichero
@@ -36,6 +36,7 @@
#include <sys/mman.h>
#include <sys/time.h>
#include <errno.h>
#include <assert.h>
#include <numa.h>
#include <numaif.h>
@@ -184,6 +185,8 @@ typedef struct {
*/
manageable_aperture_t gpuvm_aperture; /* used for GPUVM on APU, outsidethe canonical address range */
int drm_render_fd;
uint32_t usable_peer_id_num;
uint32_t *usable_peer_id_array;
} gpu_mem_t;
enum svm_aperture_type {
@@ -832,6 +835,17 @@ static int32_t gpu_mem_find_by_gpu_id(uint32_t gpu_id)
return -1;
}
static int32_t gpu_mem_find_by_node_id(uint32_t node_id)
{
uint32_t i;
for (i = 0 ; i < gpu_mem_count ; i++)
if (gpu_mem[i].node_id == node_id)
return i;
return -1;
}
static manageable_aperture_t *fmm_get_aperture(HsaApertureInfo info)
{
switch (info.type) {
@@ -2170,7 +2184,6 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
{
uint32_t i;
int32_t gpu_mem_id = 0;
HsaNodeProperties props;
struct kfd_process_device_apertures *process_apertures;
uint32_t num_of_sysfs_nodes;
HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;
@@ -2233,10 +2246,11 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
is_dgpu = false;
for (i = 0; i < NumNodes; i++) {
memset(&props, 0, sizeof(props));
HsaNodeProperties props;
ret = topology_get_node_props(i, &props);
if (ret != HSAKMT_STATUS_SUCCESS)
goto sysfs_parse_failed;
goto gpu_mem_init_failed;
topology_setup_is_dgpu_param(&props);
@@ -2245,9 +2259,18 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
int fd = open_drm_render_device(props.DrmRenderMinor);
if (fd <= 0) {
ret = HSAKMT_STATUS_ERROR;
goto sysfs_parse_failed;
goto gpu_mem_init_failed;
}
gpu_mem[gpu_mem_count].usable_peer_id_array =
calloc(NumNodes, sizeof(uint32_t));
if (!gpu_mem[gpu_mem_count].usable_peer_id_array) {
ret = HSAKMT_STATUS_NO_MEMORY;
goto gpu_mem_init_failed;
}
gpu_mem[gpu_mem_count].usable_peer_id_array[0] = props.KFDGpuID;
gpu_mem[gpu_mem_count].usable_peer_id_num = 1;
gpu_mem[gpu_mem_count].EngineId.ui32.Major = props.EngineId.ui32.Major;
gpu_mem[gpu_mem_count].EngineId.ui32.Minor = props.EngineId.ui32.Minor;
gpu_mem[gpu_mem_count].EngineId.ui32.Stepping = props.EngineId.ui32.Stepping;
@@ -2311,6 +2334,11 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
}
for (i = 0 ; i < num_of_sysfs_nodes ; i++) {
HsaNodeProperties nodeProps;
HsaIoLinkProperties linkProps[NumNodes];
uint32_t nodeId;
uint32_t j;
/* Map Kernel process device data node i <--> gpu_mem_id which
* indexes into gpu_mem[] based on gpu_id
*/
@@ -2320,10 +2348,37 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
if (all_gpu_id_array_size == gpu_mem_count) {
ret = HSAKMT_STATUS_ERROR;
goto invalid_gpu_id;
goto aperture_init_failed;
}
all_gpu_id_array[all_gpu_id_array_size++] = process_apertures[i].gpu_id;
/* Add this GPU to the usable_peer_id_arrays of all GPUs that
* this GPU has an IO link to. This GPU can map memory
* allocated on those GPUs.
*/
nodeId = gpu_mem[gpu_mem_id].node_id;
ret = topology_get_node_props(nodeId, &nodeProps);
if (ret != HSAKMT_STATUS_SUCCESS)
goto aperture_init_failed;
assert(nodeProps.NumIOLinks <= NumNodes);
ret = topology_get_iolink_props(nodeId, nodeProps.NumIOLinks,
linkProps);
if (ret != HSAKMT_STATUS_SUCCESS)
goto aperture_init_failed;
for (j = 0; j < nodeProps.NumIOLinks; j++) {
int32_t to_gpu_mem_id =
gpu_mem_find_by_node_id(linkProps[j].NodeTo);
uint32_t peer;
if (to_gpu_mem_id < 0)
continue;
assert(gpu_mem[to_gpu_mem_id].usable_peer_id_num < NumNodes);
peer = gpu_mem[to_gpu_mem_id].usable_peer_id_num++;
gpu_mem[to_gpu_mem_id].usable_peer_id_array[peer] =
gpu_mem[gpu_mem_id].gpu_id;
}
gpu_mem[gpu_mem_id].lds_aperture.base =
PORT_UINT64_TO_VPTR(process_apertures[i].lds_base);
gpu_mem[gpu_mem_id].lds_aperture.limit =
@@ -2374,7 +2429,7 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
ret = acquire_vm(gpu_mem[gpu_mem_id].gpu_id,
gpu_mem[gpu_mem_id].drm_render_fd);
if (ret != HSAKMT_STATUS_SUCCESS)
goto acquire_vm_failed;
goto aperture_init_failed;
}
all_gpu_id_array_size *= sizeof(uint32_t);
@@ -2438,15 +2493,15 @@ HSAKMT_STATUS fmm_init_process_apertures(unsigned int NumNodes)
free(process_apertures);
return ret;
invalid_gpu_id:
aperture_init_failed:
init_svm_failed:
acquire_vm_failed:
set_memory_policy_failed:
free(all_gpu_id_array);
all_gpu_id_array = NULL;
get_aperture_ioctl_failed:
free(process_apertures);
sysfs_parse_failed:
gpu_mem_init_failed:
fmm_destroy_process_apertures();
return ret;
}
@@ -2455,6 +2510,8 @@ void fmm_destroy_process_apertures(void)
{
release_mmio();
if (gpu_mem) {
while (gpu_mem_count-- > 0)
free(gpu_mem[gpu_mem_count].usable_peer_id_array);
free(gpu_mem);
gpu_mem = NULL;
}
@@ -2635,8 +2692,18 @@ static int _fmm_map_to_gpu(manageable_aperture_t *aperture,
sizeof(uint32_t);
} else {
/* not specified, not registered: map all GPUs */
args.device_ids_array_ptr = (uint64_t)all_gpu_id_array;
args.n_devices = all_gpu_id_array_size / sizeof(uint32_t);
int32_t gpu_mem_id = gpu_mem_find_by_node_id(obj->node_id);
if (!obj->userptr && get_device_id_by_node_id(obj->node_id) &&
gpu_mem_id >= 0) {
args.device_ids_array_ptr = (uint64_t)
gpu_mem[gpu_mem_id].usable_peer_id_array;
args.n_devices =
gpu_mem[gpu_mem_id].usable_peer_id_num;
} else {
args.device_ids_array_ptr = (uint64_t)all_gpu_id_array;
args.n_devices = all_gpu_id_array_size / sizeof(uint32_t);
}
}
args.n_success = 0;
@@ -3736,7 +3803,5 @@ void fmm_clear_all_mem(void)
fmm_clear_aperture(&gpu_mem[i].scratch_physical);
}
gpu_mem_count = 0;
free(gpu_mem);
gpu_mem = NULL;
fmm_destroy_process_apertures();
}
projects/rocr-runtime/src/libhsakmt.h
+3
Ver fichero
@@ -175,6 +175,9 @@ HSAKMT_STATUS validate_nodeid_array(uint32_t **gpu_id_array,
HSAKMT_STATUS topology_sysfs_get_system_props(HsaSystemProperties *props);
HSAKMT_STATUS topology_get_node_props(HSAuint32 NodeId,
HsaNodeProperties *NodeProperties);
HSAKMT_STATUS topology_get_iolink_props(HSAuint32 NodeId,
HSAuint32 NumIoLinks,
HsaIoLinkProperties *IoLinkProperties);
void topology_setup_is_dgpu_param(HsaNodeProperties *props);
bool topology_is_svm_needed(HSA_ENGINE_ID EngineId);
projects/rocr-runtime/src/topology.c
+15 -7
Ver fichero
@@ -2339,12 +2339,24 @@ out:
return err;
}
HSAKMT_STATUS topology_get_iolink_props(HSAuint32 NodeId,
HSAuint32 NumIoLinks,
HsaIoLinkProperties *IoLinkProperties)
{
if (!g_system || !g_props || NodeId >= g_system->NumNodes)
return HSAKMT_STATUS_ERROR;
memcpy(IoLinkProperties, g_props[NodeId].link,
NumIoLinks * sizeof(*IoLinkProperties));
return HSAKMT_STATUS_SUCCESS;
}
HSAKMT_STATUS HSAKMTAPI hsaKmtGetNodeIoLinkProperties(HSAuint32 NodeId,
HSAuint32 NumIoLinks,
HsaIoLinkProperties *IoLinkProperties)
{
HSAKMT_STATUS err;
uint32_t i;
if (!IoLinkProperties)
return HSAKMT_STATUS_INVALID_PARAMETER;
@@ -2364,12 +2376,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtGetNodeIoLinkProperties(HSAuint32 NodeId,
goto out;
}
for (i = 0; i < MIN(g_props[NodeId].node.NumIOLinks, NumIoLinks); i++) {
assert(g_props[NodeId].link);
IoLinkProperties[i] = g_props[NodeId].link[i];
}
err = HSAKMT_STATUS_SUCCESS;
assert(g_props[NodeId].link);
err = topology_get_iolink_props(NodeId, NumIoLinks, IoLinkProperties);
out:
pthread_mutex_unlock(&hsakmt_mutex);