From: Jonathan Kim <jonathan.kim@xxxxxxx>
If a CPU and GPU are xGMI connected but the GPU is hiveless with
respect to other GPUs, create a new CPU-GPU hive using the GPU's PCI
device location ID as the new hive ID to maintain fine grain memory
access usage.
Signed-off-by: Jonathan Kim <jonathan.kim@xxxxxxx>
Reviewed-by: Felix Kuehling <felix.kuehling@xxxxxxx>
Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
---
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 5373a79ac6a1..c7072fff778e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -1328,8 +1328,15 @@ static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev)
/* Include the CPU peer in GPU hive if connected over xGMI. */
if (!peer_dev->gpu &&
- link->iolink_type == CRAT_IOLINK_TYPE_XGMI)
+ link->iolink_type == CRAT_IOLINK_TYPE_XGMI) {
+ /*
+ * If the GPU is not part of a GPU hive, use its pci
+ * device location as the hive ID to bind with the CPU.
+ */
+ if (!dev->node_props.hive_id)
+ dev->node_props.hive_id = pci_dev_id(dev->gpu->adev->pdev);
peer_dev->node_props.hive_id = dev->node_props.hive_id;
+ }
list_for_each_entry(inbound_link, &peer_dev->io_link_props,
list) {
--
2.39.2
