diff mbox series

[1/2] net: phylink: add a function to resume phy alone to fix resume issue with WoL enabled

Message ID 20221221080144.2549125-2-xiaoning.wang@nxp.com
State New
Headers
Series fix mac not working after system resumed with WoL enabled |

Commit Message

Clark Wang Dec. 21, 2022, 8:01 a.m. UTC 
  Issue we met:
On some platforms, mac cannot work after resumed from the suspend with WoL
enabled.

The cause of the issue:
1. phylink_resolve() is in a workqueue which will not be executed immediately.
   This is the call sequence:
       phylink_resolve()->phylink_link_up()->pl->mac_ops->mac_link_up()
   For stmmac driver, mac_link_up() will set the correct speed/duplex...
   values which are from link_state.
2. In stmmac_resume(), it will call stmmac_hw_setup() after called the
   phylink_resume(), because mac need phy rx_clk to do the reset.
   stmmac_core_init() is called in function stmmac_hw_setup(), which will
   reset the mac and set the speed/duplex... to default value.
Conclusion: Because phylink_resolve() cannot determine when it is called, it
            cannot be guaranteed to be called after stmmac_core_init().
	    Once stmmac_core_init() is called after phylink_resolve(),
	    the mac will be misconfigured and cannot be used.

In order to avoid this problem, add a function called phylink_phy_resume()
to resume phy separately. This eliminates the need to call phylink_resume()
before stmmac_hw_setup().

Add another judgement before called phy_start() in phylink_start(). This way
phy_start() will not be called multiple times when resumes. At the same time,
it may not affect other drivers that do not use phylink_phy_resume().

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
---
 drivers/net/phy/phylink.c | 21 ++++++++++++++++++++-
 include/linux/phylink.h   |  1 +
 2 files changed, 21 insertions(+), 1 deletion(-)

Comments

Piotr Raczynski Dec. 22, 2022, 8:32 p.m. UTC | #1 
On Wed, Dec 21, 2022 at 04:01:43PM +0800, Clark Wang wrote:
> Issue we met:
> On some platforms, mac cannot work after resumed from the suspend with WoL
> enabled.
> 
> The cause of the issue:
> 1. phylink_resolve() is in a workqueue which will not be executed immediately.
>    This is the call sequence:
>        phylink_resolve()->phylink_link_up()->pl->mac_ops->mac_link_up()
>    For stmmac driver, mac_link_up() will set the correct speed/duplex...
>    values which are from link_state.
> 2. In stmmac_resume(), it will call stmmac_hw_setup() after called the
>    phylink_resume(), because mac need phy rx_clk to do the reset.
>    stmmac_core_init() is called in function stmmac_hw_setup(), which will
>    reset the mac and set the speed/duplex... to default value.
> Conclusion: Because phylink_resolve() cannot determine when it is called, it
>             cannot be guaranteed to be called after stmmac_core_init().
> 	    Once stmmac_core_init() is called after phylink_resolve(),
> 	    the mac will be misconfigured and cannot be used.
> 
> In order to avoid this problem, add a function called phylink_phy_resume()
> to resume phy separately. This eliminates the need to call phylink_resume()
> before stmmac_hw_setup().
> 
> Add another judgement before called phy_start() in phylink_start(). This way
> phy_start() will not be called multiple times when resumes. At the same time,
> it may not affect other drivers that do not use phylink_phy_resume().
> 
It'd be nice to see Fixes tag.

> Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
> ---
>  drivers/net/phy/phylink.c | 21 ++++++++++++++++++++-
>  include/linux/phylink.h   |  1 +
>  2 files changed, 21 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c
> index 09cc65c0da93..5bab59142579 100644
> --- a/drivers/net/phy/phylink.c
> +++ b/drivers/net/phy/phylink.c
> @@ -1939,7 +1939,7 @@ void phylink_start(struct phylink *pl)
>  	}
>  	if (poll)
>  		mod_timer(&pl->link_poll, jiffies + HZ);
> -	if (pl->phydev)
> +	if (pl->phydev && pl->phydev->state < PHY_UP)
>  		phy_start(pl->phydev);
>  	if (pl->sfp_bus)
>  		sfp_upstream_start(pl->sfp_bus);
> @@ -2020,6 +2020,25 @@ void phylink_suspend(struct phylink *pl, bool mac_wol)
>  }
>  EXPORT_SYMBOL_GPL(phylink_suspend);
>  
> +/**
> + * phylink_phy_resume() - resume phy alone
> + * @pl: a pointer to a &struct phylink returned from phylink_create()
> + *
> + * In the MAC driver using phylink, if the MAC needs the clock of the phy
> + * when it resumes, can call this function to resume the phy separately.
> + * Then proceed to MAC resume operations.
> + */
> +void phylink_phy_resume(struct phylink *pl)
> +{
> +	ASSERT_RTNL();
> +
> +	if (!test_bit(PHYLINK_DISABLE_MAC_WOL, &pl->phylink_disable_state)
> +	    && pl->phydev)
you can fit && at the end of the previous line.
> +		phy_start(pl->phydev);
> +
this blank line is not necessary.
> +}
> +EXPORT_SYMBOL_GPL(phylink_phy_resume);
> +
>  /**
>   * phylink_resume() - handle a network device resume event
>   * @pl: a pointer to a &struct phylink returned from phylink_create()
> diff --git a/include/linux/phylink.h b/include/linux/phylink.h
> index c492c26202b5..6edfab5f754c 100644
> --- a/include/linux/phylink.h
> +++ b/include/linux/phylink.h
> @@ -589,6 +589,7 @@ void phylink_stop(struct phylink *);
>  
>  void phylink_suspend(struct phylink *pl, bool mac_wol);
>  void phylink_resume(struct phylink *pl);
> +void phylink_phy_resume(struct phylink *pl);
>  
>  void phylink_ethtool_get_wol(struct phylink *, struct ethtool_wolinfo *);
>  int phylink_ethtool_set_wol(struct phylink *, struct ethtool_wolinfo *);
> -- 
> 2.34.1
>
Russell King (Oracle) Jan. 3, 2023, 9:46 a.m. UTC | #2 
On Wed, Dec 21, 2022 at 04:01:43PM +0800, Clark Wang wrote:
> Issue we met:
> On some platforms, mac cannot work after resumed from the suspend with WoL
> enabled.
> 
> The cause of the issue:
> 1. phylink_resolve() is in a workqueue which will not be executed immediately.
>    This is the call sequence:
>        phylink_resolve()->phylink_link_up()->pl->mac_ops->mac_link_up()
>    For stmmac driver, mac_link_up() will set the correct speed/duplex...
>    values which are from link_state.
> 2. In stmmac_resume(), it will call stmmac_hw_setup() after called the
>    phylink_resume(), because mac need phy rx_clk to do the reset.
>    stmmac_core_init() is called in function stmmac_hw_setup(), which will
>    reset the mac and set the speed/duplex... to default value.
> Conclusion: Because phylink_resolve() cannot determine when it is called, it
>             cannot be guaranteed to be called after stmmac_core_init().
> 	    Once stmmac_core_init() is called after phylink_resolve(),
> 	    the mac will be misconfigured and cannot be used.
> 
> In order to avoid this problem, add a function called phylink_phy_resume()
> to resume phy separately. This eliminates the need to call phylink_resume()
> before stmmac_hw_setup().
> 
> Add another judgement before called phy_start() in phylink_start(). This way
> phy_start() will not be called multiple times when resumes. At the same time,
> it may not affect other drivers that do not use phylink_phy_resume().
> 
> Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
> ---
>  drivers/net/phy/phylink.c | 21 ++++++++++++++++++++-
>  include/linux/phylink.h   |  1 +
>  2 files changed, 21 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c
> index 09cc65c0da93..5bab59142579 100644
> --- a/drivers/net/phy/phylink.c
> +++ b/drivers/net/phy/phylink.c
> @@ -1939,7 +1939,7 @@ void phylink_start(struct phylink *pl)
>  	}
>  	if (poll)
>  		mod_timer(&pl->link_poll, jiffies + HZ);
> -	if (pl->phydev)
> +	if (pl->phydev && pl->phydev->state < PHY_UP)

I'm really not happy with this - not only does this subvert the checks in
phy_start(), it's a layering violation, and it delves into internals of
phylib in an unprotected way.
diff mbox series

Patch

diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c
index 09cc65c0da93..5bab59142579 100644
--- a/drivers/net/phy/phylink.c
+++ b/drivers/net/phy/phylink.c
@@ -1939,7 +1939,7 @@  void phylink_start(struct phylink *pl)
 	}
 	if (poll)
 		mod_timer(&pl->link_poll, jiffies + HZ);
-	if (pl->phydev)
+	if (pl->phydev && pl->phydev->state < PHY_UP)
 		phy_start(pl->phydev);
 	if (pl->sfp_bus)
 		sfp_upstream_start(pl->sfp_bus);
@@ -2020,6 +2020,25 @@  void phylink_suspend(struct phylink *pl, bool mac_wol)
 }
 EXPORT_SYMBOL_GPL(phylink_suspend);
 
+/**
+ * phylink_phy_resume() - resume phy alone
+ * @pl: a pointer to a &struct phylink returned from phylink_create()
+ *
+ * In the MAC driver using phylink, if the MAC needs the clock of the phy
+ * when it resumes, can call this function to resume the phy separately.
+ * Then proceed to MAC resume operations.
+ */
+void phylink_phy_resume(struct phylink *pl)
+{
+	ASSERT_RTNL();
+
+	if (!test_bit(PHYLINK_DISABLE_MAC_WOL, &pl->phylink_disable_state)
+	    && pl->phydev)
+		phy_start(pl->phydev);
+
+}
+EXPORT_SYMBOL_GPL(phylink_phy_resume);
+
 /**
  * phylink_resume() - handle a network device resume event
  * @pl: a pointer to a &struct phylink returned from phylink_create()
diff --git a/include/linux/phylink.h b/include/linux/phylink.h
index c492c26202b5..6edfab5f754c 100644
--- a/include/linux/phylink.h
+++ b/include/linux/phylink.h
@@ -589,6 +589,7 @@  void phylink_stop(struct phylink *);
 
 void phylink_suspend(struct phylink *pl, bool mac_wol);
 void phylink_resume(struct phylink *pl);
+void phylink_phy_resume(struct phylink *pl);
 
 void phylink_ethtool_get_wol(struct phylink *, struct ethtool_wolinfo *);
 int phylink_ethtool_set_wol(struct phylink *, struct ethtool_wolinfo *);