From: bui duc phuc <phucduc.bui@gmail.com>

If regcache_sync() fails in vc5_resume(), the function only logs the
error, leaving regmap with cache_only disabled even though the register
cache has not been synchronized with the hardware.
Re-enable cache_only on failure to restore the state established by
vc5_suspend().

Signed-off-by: bui duc phuc <phucduc.bui@gmail.com>
---
 drivers/clk/clk-versaclock5.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/clk/clk-versaclock5.c b/drivers/clk/clk-versaclock5.c
index 913fcc5675f1..d87899118905 100644
--- a/drivers/clk/clk-versaclock5.c
+++ b/drivers/clk/clk-versaclock5.c
@@ -1240,8 +1240,10 @@ static int __maybe_unused vc5_resume(struct device *dev)
 
 	regcache_cache_only(vc5->regmap, false);
 	ret = regcache_sync(vc5->regmap);
-	if (ret)
+	if (ret) {
+		regcache_cache_only(vc5->regmap, true);
 		dev_err(dev, "Failed to restore register map: %d\n", ret);
+	}
 	return ret;
 }
 
-- 
2.43.0

Hi bui,

On Fri, Jul 24, 2026 at 02:46:03PM +0700, phucduc.bui@gmail.com wrote:
> From: bui duc phuc <phucduc.bui@gmail.com>
> 
> If regcache_sync() fails in vc5_resume(), the function only logs the
> error, leaving regmap with cache_only disabled even though the register
> cache has not been synchronized with the hardware.
> Re-enable cache_only on failure to restore the state established by
> vc5_suspend().
> 
> Signed-off-by: bui duc phuc <phucduc.bui@gmail.com>
> ---
>  drivers/clk/clk-versaclock5.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/clk-versaclock5.c b/drivers/clk/clk-versaclock5.c
> index 913fcc5675f1..d87899118905 100644
> --- a/drivers/clk/clk-versaclock5.c
> +++ b/drivers/clk/clk-versaclock5.c
> @@ -1240,8 +1240,10 @@ static int __maybe_unused vc5_resume(struct device *dev)
>  
>  	regcache_cache_only(vc5->regmap, false);
>  	ret = regcache_sync(vc5->regmap);
> -	if (ret)
> +	if (ret) {
> +		regcache_cache_only(vc5->regmap, true);
>  		dev_err(dev, "Failed to restore register map: %d\n", ret);
> +	}
>  	return ret;
>  }

I don't see precedent for this pattern anywhere else in the tree. Is
this an observed problem or something that could theoretically happen?

If this fails in the code today, the pm core will propagate the error,
and the device won't be marked as resumed.

Brian