From: TanZheng <tanzheng@kylinos.cn>

ghes_scan_system() sets system_scanned after the DMI walk. The unlock
path always kfree()s ghes_hw.dimms but left system_scanned and
num_dimms set, so a later ghes_edac_register() skipped the scan and
indexed a NULL dimms array.

Clear the scan flag (and num_dimms) in that shared unlock path when a
scan actually ran. The next register() will walk DMI again.

Signed-off-by: TanZheng <tanzheng@kylinos.cn>
---
 drivers/edac/ghes_edac.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/edac/ghes_edac.c b/drivers/edac/ghes_edac.c
index d80c88818691..b0630f1a2815 100644
--- a/drivers/edac/ghes_edac.c
+++ b/drivers/edac/ghes_edac.c
@@ -499,6 +499,10 @@ static int ghes_edac_register(struct device *dev)
 	/* Not needed anymore */
 	kfree(ghes_hw.dimms);
 	ghes_hw.dimms = NULL;
+	if (system_scanned) {
+		system_scanned = false;
+		ghes_hw.num_dimms = 0;
+	}
 
 	mutex_unlock(&ghes_reg_mutex);
 
-- 
2.25.1

On Wed, Sep 09, 2026 at 03:19:34PM +0800, kensanya@163.com wrote:
> From: TanZheng <tanzheng@kylinos.cn>
> 
> ghes_scan_system() sets system_scanned after the DMI walk. The unlock
> path always kfree()s ghes_hw.dimms but left system_scanned and
> num_dimms set, so a later ghes_edac_register() skipped the scan and
> indexed a NULL dimms array.
> 
> Clear the scan flag (and num_dimms) in that shared unlock path when a
> scan actually ran. The next register() will walk DMI again.
> 
> Signed-off-by: TanZheng <tanzheng@kylinos.cn>
> ---
>  drivers/edac/ghes_edac.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/edac/ghes_edac.c b/drivers/edac/ghes_edac.c
> index d80c88818691..b0630f1a2815 100644
> --- a/drivers/edac/ghes_edac.c
> +++ b/drivers/edac/ghes_edac.c
> @@ -499,6 +499,10 @@ static int ghes_edac_register(struct device *dev)
>  	/* Not needed anymore */
>  	kfree(ghes_hw.dimms);
>  	ghes_hw.dimms = NULL;
> +	if (system_scanned) {
> +		system_scanned = false;
> +		ghes_hw.num_dimms = 0;
> +	}
>  
>  	mutex_unlock(&ghes_reg_mutex);
>  
> -- 

That's not the right fix. On unlock, it returns an error which the caller
should check and then unwind by calling ghes_edac_unregister() on the already
successfully registered instances.

But I don't have the necessary hardware to test it on and if you don't have it
either, don't bother fixing it. Might as well obsolete this driver too.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette