[PATCH net-next 0/2] net: phy: at803x: paging support

[PATCH net-next 0/2] net: phy: at803x: paging support

[Michael Walle]

Add paging support to the QCA AR8031/33 PHY. This will be needed if we
add support for the .config_inband_aneg callback, see series [1]. But it
also turns out, that the driver already accessed the fiber page all along
without proper locking. Patch 2 will fix that.

[1] https://lore.kernel.org/netdev/20210212172341.3489046-1-olteanv@gmail.com/

Michael Walle (2):
  net: phy: at803x: add pages support to AR8031/33
  net: phy: at803x: use proper locking in at803x_aneg_done()

 drivers/net/phy/at803x.c | 72 ++++++++++++++++++++++++++++------------
 1 file changed, 51 insertions(+), 21 deletions(-)

-- 
2.20.1

[PATCH net-next 1/2] net: phy: at803x: add pages support to AR8031/33

[Michael Walle]

The AR8031 has two register sets: Copper and Fiber. The fiber page is
used in case of 100Base-FX and 1000Base-X. But more importantly it is
also used for the SGMII link. Add support to switch between these two.

Signed-off-by: Michael Walle <michael@walle.cc>
---
 drivers/net/phy/at803x.c | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/drivers/net/phy/at803x.c b/drivers/net/phy/at803x.c
index d67bddc111e3..a3aa10f14638 100644
--- a/drivers/net/phy/at803x.c
+++ b/drivers/net/phy/at803x.c
@@ -144,6 +144,9 @@
 #define ATH8035_PHY_ID 0x004dd072
 #define AT8030_PHY_ID_MASK			0xffffffef
 
+#define AT803X_FIBER_PAGE 0
+#define AT803X_COPPER_PAGE 1
+
 MODULE_DESCRIPTION("Qualcomm Atheros AR803x PHY driver");
 MODULE_AUTHOR("Matus Ujhelyi");
 MODULE_LICENSE("GPL");
@@ -1143,6 +1146,36 @@ static int at803x_cable_test_start(struct phy_device *phydev)
 	return 0;
 }
 
+static int at803x_read_page(struct phy_device *phydev)
+{
+	int val;
+
+	val = __phy_read(phydev, AT803X_REG_CHIP_CONFIG);
+	if (val < 0)
+		return val;
+
+	return (val & AT803X_BT_BX_REG_SEL) ? AT803X_COPPER_PAGE : AT803X_FIBER_PAGE;
+}
+
+static int at803x_write_page(struct phy_device *phydev, int page)
+{
+	u16 sel;
+
+	switch (page) {
+	case AT803X_FIBER_PAGE:
+		sel = 0;
+		break;
+	case AT803X_COPPER_PAGE:
+		sel = AT803X_BT_BX_REG_SEL;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return __phy_modify(phydev, AT803X_REG_CHIP_CONFIG,
+			    AT803X_BT_BX_REG_SEL, sel);
+}
+
 static struct phy_driver at803x_driver[] = {
 {
 	/* Qualcomm Atheros AR8035 */
@@ -1189,6 +1222,8 @@ static struct phy_driver at803x_driver[] = {
 	.flags			= PHY_POLL_CABLE_TEST,
 	.probe			= at803x_probe,
 	.remove			= at803x_remove,
+	.read_page		= at803x_read_page,
+	.write_page		= at803x_write_page,
 	.config_init		= at803x_config_init,
 	.soft_reset		= genphy_soft_reset,
 	.set_wol		= at803x_set_wol,
-- 
2.20.1

[PATCH net-next 2/2] net: phy: at803x: use proper locking in at803x_aneg_done()

[Michael Walle]

at803x_aneg_done() checks if auto-negotiation is completed on the SGMII
side. This doesn't take the mdio bus lock and the page switching is
open-coded. Now that we have proper page support, just use
phy_read_paged(). Also use phydev->interface to check if we have an
SGMII link instead of reading the mode register and be a bit more
precise on the warning message.

Signed-off-by: Michael Walle <michael@walle.cc>
---
 drivers/net/phy/at803x.c | 37 ++++++++++++++++---------------------
 1 file changed, 16 insertions(+), 21 deletions(-)

diff --git a/drivers/net/phy/at803x.c b/drivers/net/phy/at803x.c
index a3aa10f14638..8abaea7ae6bd 100644
--- a/drivers/net/phy/at803x.c
+++ b/drivers/net/phy/at803x.c
@@ -756,32 +756,27 @@ static void at803x_link_change_notify(struct phy_device *phydev)
 
 static int at803x_aneg_done(struct phy_device *phydev)
 {
-	int ccr;
-
-	int aneg_done = genphy_aneg_done(phydev);
-	if (aneg_done != BMSR_ANEGCOMPLETE)
-		return aneg_done;
+	int ret, val;
 
-	/*
-	 * in SGMII mode, if copper side autoneg is successful,
-	 * also check SGMII side autoneg result
-	 */
-	ccr = phy_read(phydev, AT803X_REG_CHIP_CONFIG);
-	if ((ccr & AT803X_MODE_CFG_MASK) != AT803X_MODE_CFG_SGMII)
-		return aneg_done;
 
-	/* switch to SGMII/fiber page */
-	phy_write(phydev, AT803X_REG_CHIP_CONFIG, ccr & ~AT803X_BT_BX_REG_SEL);
+	ret = genphy_aneg_done(phydev);
 
-	/* check if the SGMII link is OK. */
-	if (!(phy_read(phydev, AT803X_PSSR) & AT803X_PSSR_MR_AN_COMPLETE)) {
-		phydev_warn(phydev, "803x_aneg_done: SGMII link is not ok\n");
-		aneg_done = 0;
+	/* In SGMII mode, if copper side autoneg is successful, also check
+	 * SGMII side autoneg result.
+	 */
+	if (phydev->interface == PHY_INTERFACE_MODE_SGMII &&
+	    ret == BMSR_ANEGCOMPLETE) {
+		val = phy_read_paged(phydev, AT803X_FIBER_PAGE, AT803X_PSSR);
+		if (val < 0)
+			return val;
+
+		if (!(val & AT803X_PSSR_MR_AN_COMPLETE)) {
+			phydev_warn(phydev, "SGMII autoneg isn't completed\n");
+			return 0;
+		}
 	}
-	/* switch back to copper page */
-	phy_write(phydev, AT803X_REG_CHIP_CONFIG, ccr | AT803X_BT_BX_REG_SEL);
 
-	return aneg_done;
+	return ret;
 }
 
 static int at803x_read_status(struct phy_device *phydev)
-- 
2.20.1

Re: [PATCH net-next 2/2] net: phy: at803x: use proper locking in at803x_aneg_done()

[Vladimir Oltean]

Hi Michael,

On Sun, Feb 14, 2021 at 02:04:05AM +0100, Michael Walle wrote:
> at803x_aneg_done() checks if auto-negotiation is completed on the SGMII
> side. This doesn't take the mdio bus lock and the page switching is
> open-coded. Now that we have proper page support, just use
> phy_read_paged(). Also use phydev->interface to check if we have an
> SGMII link instead of reading the mode register and be a bit more
> precise on the warning message.
>
> Signed-off-by: Michael Walle <michael@walle.cc>
> ---

How did you test this patch?

Re: [PATCH net-next 2/2] net: phy: at803x: use proper locking in at803x_aneg_done()

[Michael Walle]

Am 14. Februar 2021 02:57:33 MEZ schrieb Vladimir Oltean <olteanv@gmail.com>:
>Hi Michael,
>
>On Sun, Feb 14, 2021 at 02:04:05AM +0100, Michael Walle wrote:
>> at803x_aneg_done() checks if auto-negotiation is completed on the
>SGMII
>> side. This doesn't take the mdio bus lock and the page switching is
>> open-coded. Now that we have proper page support, just use
>> phy_read_paged(). Also use phydev->interface to check if we have an
>> SGMII link instead of reading the mode register and be a bit more
>> precise on the warning message.
>>
>> Signed-off-by: Michael Walle <michael@walle.cc>
>> ---
>
>How did you test this patch?

I'm afraid it's just compile time tested.

-michael 

Re: [PATCH net-next 2/2] net: phy: at803x: use proper locking in at803x_aneg_done()

[Vladimir Oltean]

On Sun, Feb 14, 2021 at 03:18:49AM +0100, Michael Walle wrote:
> Am 14. Februar 2021 02:57:33 MEZ schrieb Vladimir Oltean <olteanv@gmail.com>:
> >Hi Michael,
> >
> >On Sun, Feb 14, 2021 at 02:04:05AM +0100, Michael Walle wrote:
> >> at803x_aneg_done() checks if auto-negotiation is completed on the
> >SGMII
> >> side. This doesn't take the mdio bus lock and the page switching is
> >> open-coded. Now that we have proper page support, just use
> >> phy_read_paged(). Also use phydev->interface to check if we have an
> >> SGMII link instead of reading the mode register and be a bit more
> >> precise on the warning message.
> >>
> >> Signed-off-by: Michael Walle <michael@walle.cc>
> >> ---
> >
> >How did you test this patch?
>
> I'm afraid it's just compile time tested.

I'm asking because at803x_aneg_done has been dead code for more than 2
years now. Unreachable. And while it was reachable it was buggy and an
abuse of the phylib API. So you might want to just delete this function
instead. Context:
https://lkml.org/lkml/2020/5/30/375

Re: [PATCH net-next 2/2] net: phy: at803x: use proper locking in at803x_aneg_done()

[Michael Walle]

Am 2021-02-14 03:24, schrieb Vladimir Oltean:
> On Sun, Feb 14, 2021 at 03:18:49AM +0100, Michael Walle wrote:
>> Am 14. Februar 2021 02:57:33 MEZ schrieb Vladimir Oltean 
>> <olteanv@gmail.com>:
>> >Hi Michael,
>> >
>> >On Sun, Feb 14, 2021 at 02:04:05AM +0100, Michael Walle wrote:
>> >> at803x_aneg_done() checks if auto-negotiation is completed on the
>> >SGMII
>> >> side. This doesn't take the mdio bus lock and the page switching is
>> >> open-coded. Now that we have proper page support, just use
>> >> phy_read_paged(). Also use phydev->interface to check if we have an
>> >> SGMII link instead of reading the mode register and be a bit more
>> >> precise on the warning message.
>> >>
>> >> Signed-off-by: Michael Walle <michael@walle.cc>
>> >> ---
>> >
>> >How did you test this patch?
>> 
>> I'm afraid it's just compile time tested.
> 
> I'm asking because at803x_aneg_done has been dead code for more than 2
> years now. Unreachable. And while it was reachable it was buggy and an
> abuse of the phylib API. So you might want to just delete this function
> instead. Context:
> https://lkml.org/lkml/2020/5/30/375

Are you sure? While it isn't called from phylib, it might be called from
some drivers directly or indirectly if they use phy_speed_down(). But
it is questionable if this is much of a use then.

That being said, if no one objects, I'd remove it, too.

-michael

Re: [PATCH net-next 2/2] net: phy: at803x: use proper locking in at803x_aneg_done()

[Vladimir Oltean]

On Sun, Feb 14, 2021 at 09:48:53PM +0100, Michael Walle wrote:
> Am 2021-02-14 03:24, schrieb Vladimir Oltean:
> > On Sun, Feb 14, 2021 at 03:18:49AM +0100, Michael Walle wrote:
> > > Am 14. Februar 2021 02:57:33 MEZ schrieb Vladimir Oltean <olteanv@gmail.com>:
> > > >Hi Michael,
> > > >
> > > >On Sun, Feb 14, 2021 at 02:04:05AM +0100, Michael Walle wrote:
> > > >> at803x_aneg_done() checks if auto-negotiation is completed on the SGMII
> > > >> side. This doesn't take the mdio bus lock and the page switching is
> > > >> open-coded. Now that we have proper page support, just use
> > > >> phy_read_paged(). Also use phydev->interface to check if we have an
> > > >> SGMII link instead of reading the mode register and be a bit more
> > > >> precise on the warning message.
> > > >>
> > > >> Signed-off-by: Michael Walle <michael@walle.cc>
> > > >> ---
> > > >
> > > >How did you test this patch?
> > >
> > > I'm afraid it's just compile time tested.
> >
> > I'm asking because at803x_aneg_done has been dead code for more than 2
> > years now. Unreachable. And while it was reachable it was buggy and an
> > abuse of the phylib API. So you might want to just delete this function
> > instead. Context:
> > https://lkml.org/lkml/2020/5/30/375
>
> Are you sure? While it isn't called from phylib, it might be called from
> some drivers directly or indirectly if they use phy_speed_down().
> But it is questionable if this is much of a use then.
>
> That being said, if no one objects, I'd remove it, too.

It might, true, but it is certainly of no use.