Fix memory leaks.
Fix crash in bootstrap error case.
Fix Collnet clean-up issue.
Make PCI switch vendor/device optional for XML injection.
Add support for nvidia-peermem module.
This commit is contained in:
Sylvain Jeaugey
2021-04-26 14:24:50 -07:00
parent a46ea10583
commit ca8485b0d0
9 changed files with 35 additions and 20 deletions
+4 -1
View File
@@ -220,7 +220,10 @@ ncclResult_t ncclIbDevices(int* ndev) {
ncclResult_t ncclIbGdrSupport(int ibDev) {
static int moduleLoaded = -1;
if (moduleLoaded == -1) {
moduleLoaded = (access("/sys/kernel/mm/memory_peers/nv_mem/version", F_OK) == -1) ? 0 : 1;
// Check for the nv_peer_mem module being loaded
moduleLoaded = ((access("/sys/kernel/mm/memory_peers/nv_mem/version", F_OK) == -1) &&
// Also support the new nvidia-peermem module
(access("/sys/kernel/mm/memory_peers/nvidia-peermem/version", F_OK) == -1)) ? 0 : 1;
}
if (moduleLoaded == 0) return ncclSystemError;
return ncclSuccess;