From 659cd52d5c705c9904521ef894b4e56f5a7598e1 Mon Sep 17 00:00:00 2001
From: Wenkai Du <43822138+wenkaidu@users.noreply.github.com>
Date: Fri, 8 Jul 2022 09:52:27 -0700
Subject: [PATCH] Add more constraints to enable GDR (#579)

* Add more constraints to enable GDR

* Revert deleted line

[ROCm/rccl commit: aa0d7ca8827acc9b8ad769f1ebd0e6806345c34b]
---
 projects/rccl/src/graph/xml.h         | 2 ++
 projects/rccl/src/transport/net_ib.cc | 9 +++++++++
 2 files changed, 11 insertions(+)

diff --git a/projects/rccl/src/graph/xml.h b/projects/rccl/src/graph/xml.h
index 0e35f473ba..64f0b4cb8f 100644
--- a/projects/rccl/src/graph/xml.h
+++ b/projects/rccl/src/graph/xml.h
@@ -56,6 +56,8 @@ ncclResult_t ncclTopoFillNet(struct ncclXml* xml, const char* pciPath, const cha
 /* Remove unneeded parts */
 ncclResult_t ncclTopoTrimXml(struct ncclXml* xml);
 
+ncclResult_t ncclTopoGetStrFromSys(const char* path, const char* fileName, char* strValue);
+
 /**************/
 /* XML Struct */
 /* Functions  */
diff --git a/projects/rccl/src/transport/net_ib.cc b/projects/rccl/src/transport/net_ib.cc
index ed0e5e7fbc..8b8849abc0 100644
--- a/projects/rccl/src/transport/net_ib.cc
+++ b/projects/rccl/src/transport/net_ib.cc
@@ -25,6 +25,7 @@
 #include "timer.h"
 
 #include "ibvwrap.h"
+#include "graph/xml.h"
 
 #define USE_RDMA_WRITE 1
 #define MAXNAMESIZE 64
@@ -276,6 +277,14 @@ ncclResult_t ncclIbGdrSupport(int ibDev) {
   if (moduleLoaded == -1) {
 #if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
     moduleLoaded = (access("/sys/kernel/mm/memory_peers/amdkfd/version", F_OK) == -1) ? 0 : 1;
+    char strValue[MAX_STR_LEN];
+    NCCLCHECK(ncclTopoGetStrFromSys("/sys/devices/virtual/dmi/id", "bios_version", strValue));
+    if (strncmp("Hyper-V UEFI Release", strValue, 20) == 0) {
+      int roMode = ncclParamIbPciRelaxedOrdering();
+      NCCLCHECK(ncclTopoGetStrFromSys("/proc/sys/kernel", "numa_balancing", strValue));
+      if (strcmp(strValue, "1") == 0 && roMode == 0)
+        moduleLoaded = 0;
+    }
 #else
     // Check for the nv_peer_mem module being loaded
     moduleLoaded = ((access("/sys/kernel/mm/memory_peers/nv_mem/version", F_OK) == -1) &&