Hardware Locality (hwloc)  PR-737-20250925.0822.gite8f69c77f
nvml.h
1 /*
2  * SPDX-License-Identifier: BSD-3-Clause
3  * Copyright © 2012-2024 Inria. All rights reserved.
4  * See COPYING in top-level directory.
5  */
6 
14 #ifndef HWLOC_NVML_H
15 #define HWLOC_NVML_H
16 
17 #include "hwloc.h"
18 #include "hwloc/autogen/config.h"
19 #include "hwloc/helper.h"
20 #ifdef HWLOC_LINUX_SYS
21 #include "hwloc/linux.h"
22 #endif
23 
24 #include <nvml.h>
25 
26 
27 #ifdef __cplusplus
28 extern "C" {
29 #endif
30 
31 
59 static __hwloc_inline int
60 hwloc_nvml_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
61  nvmlDevice_t device, hwloc_cpuset_t set)
62 {
63 #ifdef HWLOC_LINUX_SYS
64  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
65 #define HWLOC_NVML_DEVICE_SYSFS_PATH_MAX 128
66  char path[HWLOC_NVML_DEVICE_SYSFS_PATH_MAX];
67  nvmlReturn_t nvres;
68  nvmlPciInfo_t pci;
69 
70  if (!hwloc_topology_is_thissystem(topology)) {
71  errno = EINVAL;
72  return -1;
73  }
74 
75  nvres = nvmlDeviceGetPciInfo(device, &pci);
76  if (NVML_SUCCESS != nvres) {
77  errno = EINVAL;
78  return -1;
79  }
80 
81  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", pci.domain, pci.bus, pci.device);
82  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
83  || hwloc_bitmap_iszero(set))
85 #else
86  /* Non-Linux systems simply get a full cpuset */
88 #endif
89  return 0;
90 }
91 
105 static __hwloc_inline hwloc_obj_t
107 {
108  hwloc_obj_t osdev = NULL;
109  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
110  if ((osdev->attr->osdev.types & (HWLOC_OBJ_OSDEV_GPU|HWLOC_OBJ_OSDEV_COPROC)) /* assume future nvml devices will be at least GPU or COPROC */
111  && osdev->name
112  && !strncmp("nvml", osdev->name, 4)
113  && atoi(osdev->name + 4) == (int) idx)
114  return osdev;
115  }
116  return NULL;
117 }
118 
132 static __hwloc_inline hwloc_obj_t
133 hwloc_nvml_get_device_osdev(hwloc_topology_t topology, nvmlDevice_t device)
134 {
135  hwloc_obj_t osdev;
136  nvmlReturn_t nvres;
137  nvmlPciInfo_t pci;
138  char uuid[64];
139 
140  if (!hwloc_topology_is_thissystem(topology)) {
141  errno = EINVAL;
142  return NULL;
143  }
144 
145  nvres = nvmlDeviceGetPciInfo(device, &pci);
146  if (NVML_SUCCESS != nvres)
147  return NULL;
148 
149  nvres = nvmlDeviceGetUUID(device, uuid, sizeof(uuid));
150  if (NVML_SUCCESS != nvres)
151  uuid[0] = '\0';
152 
153  osdev = NULL;
154  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
155  hwloc_obj_t pcidev = osdev->parent;
156  const char *info;
157 
158  if (strncmp(osdev->name, "nvml", 4))
159  continue;
160 
161  if (pcidev
162  && pcidev->type == HWLOC_OBJ_PCI_DEVICE
163  && pcidev->attr->pcidev.domain == pci.domain
164  && pcidev->attr->pcidev.bus == pci.bus
165  && pcidev->attr->pcidev.dev == pci.device
166  && pcidev->attr->pcidev.func == 0)
167  return osdev;
168 
169  info = hwloc_obj_get_info_by_name(osdev, "NVIDIAUUID");
170  if (info && !strcmp(info, uuid))
171  return osdev;
172  }
173 
174  return NULL;
175 }
176 
180 #ifdef __cplusplus
181 } /* extern "C" */
182 #endif
183 
184 
185 #endif /* HWLOC_NVML_H */
HWLOC_OBJ_OSDEV_COPROC
@ HWLOC_OBJ_OSDEV_COPROC
Operating system co-processor device. For instance "opencl0d0" for a OpenCL device,...
Definition: hwloc.h:396
hwloc_bitmap_iszero
int hwloc_bitmap_iszero(hwloc_const_bitmap_t bitmap)
Test whether bitmap bitmap is empty.
hwloc_obj
Structure of a topology object.
Definition: hwloc.h:488
HWLOC_OBJ_OSDEV_GPU
@ HWLOC_OBJ_OSDEV_GPU
Operating system GPU device. For instance ":0.0" for a GL display, "card0" for a Linux DRM device,...
Definition: hwloc.h:389
hwloc_topology_t
struct hwloc_topology * hwloc_topology_t
Topology context.
Definition: hwloc.h:778
hwloc_linux_read_path_as_cpumask
int hwloc_linux_read_path_as_cpumask(const char *path, hwloc_bitmap_t set)
Convert a linux kernel cpumask file path into a hwloc bitmap set.
hwloc_nvml_get_device_osdev
hwloc_obj_t hwloc_nvml_get_device_osdev(hwloc_topology_t topology, nvmlDevice_t device)
Get the hwloc OS device object corresponding to NVML device device.
Definition: nvml.h:133
hwloc_obj_attr_u::hwloc_pcidev_attr_s::dev
unsigned char dev
Device number (zz in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:727
hwloc_obj_attr_u::hwloc_pcidev_attr_s::bus
unsigned char bus
Bus number (yy in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:726
hwloc_obj::parent
struct hwloc_obj * parent
Parent, NULL if root (Machine object)
Definition: hwloc.h:538
hwloc_obj_attr_u::osdev
struct hwloc_obj_attr_u::hwloc_osdev_attr_s osdev
hwloc_topology_get_complete_cpuset
hwloc_const_cpuset_t hwloc_topology_get_complete_cpuset(hwloc_topology_t topology)
Get complete CPU set.
hwloc_nvml_get_device_cpuset
int hwloc_nvml_get_device_cpuset(hwloc_topology_t topology, nvmlDevice_t device, hwloc_cpuset_t set)
Get the CPU set of processors that are physically close to NVML device device.
Definition: nvml.h:60
hwloc_obj::attr
union hwloc_obj_attr_u * attr
Object type-specific Attributes, may be NULL if no attribute value was found.
Definition: hwloc.h:507
hwloc_obj_attr_u::hwloc_osdev_attr_s::types
hwloc_obj_osdev_types_t types
OR'ed set of at least one hwloc_obj_osdev_type_e.
Definition: hwloc.h:761
hwloc_obj::name
char * name
Object-specific name if any. Mostly used for identifying OS devices and Misc objects where a name str...
Definition: hwloc.h:500
hwloc_cpuset_t
hwloc_bitmap_t hwloc_cpuset_t
A CPU set is a bitmap whose bits are set according to CPU physical OS indexes.
Definition: hwloc.h:162
hwloc_obj_attr_u::hwloc_pcidev_attr_s::domain
unsigned int domain
Domain number (xxxx in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:725
HWLOC_OBJ_PCI_DEVICE
@ HWLOC_OBJ_PCI_DEVICE
PCI device (filtered out by default).
Definition: hwloc.h:320
hwloc_bitmap_copy
int hwloc_bitmap_copy(hwloc_bitmap_t dst, hwloc_const_bitmap_t src)
Copy the contents of bitmap src into the already allocated bitmap dst.
hwloc_obj_attr_u::hwloc_pcidev_attr_s::func
unsigned char func
Function number (t in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:728
hwloc_obj_attr_u::pcidev
struct hwloc_obj_attr_u::hwloc_pcidev_attr_s pcidev
hwloc_obj::type
hwloc_obj_type_t type
Type of object.
Definition: hwloc.h:490
hwloc_get_next_osdev
hwloc_obj_t hwloc_get_next_osdev(hwloc_topology_t topology, hwloc_obj_t prev)
Get the next OS device in the system.
Definition: helper.h:1291
hwloc_obj_get_info_by_name
const char * hwloc_obj_get_info_by_name(hwloc_obj_t obj, const char *name)
Search the given name in the object array of infos and return the corresponding value.
hwloc_nvml_get_device_osdev_by_index
hwloc_obj_t hwloc_nvml_get_device_osdev_by_index(hwloc_topology_t topology, unsigned idx)
Get the hwloc OS device object corresponding to the NVML device whose index is idx.
Definition: nvml.h:106
hwloc_topology_is_thissystem
int hwloc_topology_is_thissystem(hwloc_topology_t restrict topology)
Does the topology context come from this system?