There are two problems with meson8b_devm_clk_prepare_enable(),
introduced in a54dc4a49045 (net: stmmac: dwmac-meson8b: Make the clock
enabling code re-usable):

- It doesn't pass the clk argument, but instead always the
  rgmii_tx_clk of the device.

- It silently ignores the return value of devm_add_action_or_reset().

The former didn't become an actual bug until another user showed up in
the next commit, 9308c47640d5 (net: stmmac: dwmac-meson8b: add support
for the RX delay configuration). The latter means the callers could
end up with the clock not actually prepared/enabled.

Fixes: a54dc4a49045 (net: stmmac: dwmac-meson8b: Make the clock enabling code re-usable)
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
---
Not tested in any way, just stumbled on this which didn't seem right.

It seems that the timing_adj_clk could be changed (with a bit of
refactoring) to use devm_clk_get_enabled() - i.e. don't make it
optional, but only get/prepare/enable it in the (delay_config &
PRG_ETH0_ADJ_ENABLE) branch where the code actually requires it.

But since there's no official devm_clk_prepare_enable, and the
rgmii_tx_clk isn't obtained via a clk_get, I don't see a simple way to
avoid this private implmentation of devm_clk_prepare_enable().

drivers/net/ethernet/stmicro/stmmac/dwmac-meson8b.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-meson8b.c b/drivers/net/ethernet/stmicro/stmmac/dwmac-meson8b.c
index c7a6588d9398..e8b507f88fbc 100644
--- a/drivers/net/ethernet/stmicro/stmmac/dwmac-meson8b.c
+++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-meson8b.c
@@ -272,11 +272,9 @@ static int meson8b_devm_clk_prepare_enable(struct meson8b_dwmac *dwmac,
 	if (ret)
 		return ret;
 
-	devm_add_action_or_reset(dwmac->dev,
-				 (void(*)(void *))clk_disable_unprepare,
-				 dwmac->rgmii_tx_clk);
-
-	return 0;
+	return devm_add_action_or_reset(dwmac->dev,
+					(void(*)(void *))clk_disable_unprepare,
+					clk);
 }
 
 static int meson8b_init_rgmii_delays(struct meson8b_dwmac *dwmac)
-- 
2.37.2

Hello:

This patch was applied to netdev/net.git (master)
by Jakub Kicinski <kuba@kernel.org>:

On Fri,  4 Nov 2022 09:30:04 +0100 you wrote:
> There are two problems with meson8b_devm_clk_prepare_enable(),
> introduced in a54dc4a49045 (net: stmmac: dwmac-meson8b: Make the clock
> enabling code re-usable):
> 
> - It doesn't pass the clk argument, but instead always the
>   rgmii_tx_clk of the device.
> 
> [...]

Here is the summary with links:
  - net: stmmac: dwmac-meson8b: fix meson8b_devm_clk_prepare_enable()
    https://git.kernel.org/netdev/net/c/ed4314f77297

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Hello Rasmus,

Thanks for your patch!
It would be great if you could Cc linux-amlogic@lists.infradead.org in
v2 of this patch, so other Amlogic SoC maintainers can also provide
their feedback.

On Fri, Nov 4, 2022 at 9:30 AM Rasmus Villemoes
<linux@rasmusvillemoes.dk> wrote:
>
> There are two problems with meson8b_devm_clk_prepare_enable(),
> introduced in a54dc4a49045 (net: stmmac: dwmac-meson8b: Make the clock
> enabling code re-usable):
>
> - It doesn't pass the clk argument, but instead always the
>   rgmii_tx_clk of the device.
Indeed, this is a problem - thanks for fixing this!

> - It silently ignores the return value of devm_add_action_or_reset().
This second part is not so easy.
The thought process back when this code was implemented was: let's
continue loading of the driver even if devm_add_action_or_reset()
fails as this just means during shutdown/rmmod we don't disable all
clocks.
If we want to propagate the error code returned by
devm_add_action_or_reset() then we also need to do the clean up within
meson8b_devm_clk_prepare_enable(), meaning we need to call
clk_disable_unprepare() in case devm_add_action_or_reset() failed.
Your change just propagates the error code without disabling and
unpreparing the clock.

[...]
> The latter means the callers could
> end up with the clock not actually prepared/enabled.
In my opinion this statement is not correct: even if
devm_add_action_or_reset() fails the clock will be prepared and
enabled (by clk_prepare_enable() right before).

Personally I would just change the clk argument and keep the return 0.
What do you think?


Best regards,
Martin

Rasmus replied to me off-list yesterday. His reply is important so I
am sharing it here.

On Fri, Nov 4, 2022 at 10:02 PM Martin Blumenstingl
<martin.blumenstingl@googlemail.com> wrote:
[...]
> If we want to propagate the error code returned by
> devm_add_action_or_reset() then we also need to do the clean up within
> meson8b_devm_clk_prepare_enable(), meaning we need to call
> clk_disable_unprepare() in case devm_add_action_or_reset() failed.
> Your change just propagates the error code without disabling and
> unpreparing the clock.
Rasmus replied to me:
"devm_add_action_or_reset precisely calls the reset callback for you
if the add action falls..."

I was not aware of this but it's actually true, which makes the patch
perfectly valid. Thanks for the heads up Rasmus!

So this patch gets my:
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>

I suggest applying it to net-next (even though it carries a Fixes tag)
so the various kernel auto testing labs can try it out on as many
Amlogic SoCs as possible.


Best regards,
Martin