[Public]

Seems reasonable to me, but I'd want Felix (unlikely due to PTO) or Harish to 
weigh in to make sure I am not missing something obvious.

>From what I can tell, it was called way back when KFD Topology was first 
>written, back when it was for APUs-only. So they'd do 2 calls to update sysfs, 
>once for the CPU and once for the GPU. I think the logic just never got 
>touched because it still works, even though we've changed things a bit now.

 Kent

> -----Original Message-----
> From: amd-gfx <[email protected]> On Behalf Of Mario
> Limonciello (AMD)
> Sent: Wednesday, December 10, 2025 3:15 PM
> To: [email protected]
> Cc: Mario Limonciello (AMD) <[email protected]>
> Subject: [PATCH v2 2/3] amdkfd: Don't rebuild node tree when calling
> kfd_topology_update_sysfs()
>
> There is no need to remove all the nodes and rebuild them.  The content
> will be the same.  Instead check whether the node was created and skip
> the creation.
>
> Signed-off-by: Mario Limonciello (AMD) <[email protected]>
> ---
>  drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 6 ++----
>  1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> index a0990dd2378c1..b65f29294e2d6 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
> @@ -650,8 +650,8 @@ static int kfd_build_sysfs_node_entry(struct
> kfd_topology_device *dev,
>       uint32_t i, num_attrs;
>       struct attribute **attrs;
>
> -     if (WARN_ON(dev->kobj_node))
> -             return -EEXIST;
> +     if (dev->kobj_node)
> +             return 0;
>
>       /*
>        * Creating the sysfs folders
> @@ -888,8 +888,6 @@ static int kfd_topology_update_sysfs(void)
>                       return ret;
>       }
>
> -     kfd_remove_sysfs_node_tree();
> -
>       return kfd_build_sysfs_node_tree();
>  }
>
> --
> 2.43.0

Reply via email to