| Message ID | 20220704153510.3859649-1-f.fainelli@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [stable,4.9] net: dsa: bcm_sf2: force pause link settings | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Guessing tree name failed - patch did not apply |
Hi Florian, On Mon, Jul 04, 2022 at 08:35:07AM -0700, Florian Fainelli wrote: > From: Doug Berger <opendmb@gmail.com> > > commit 7c97bc0128b2eecc703106112679a69d446d1a12 upstream > > The pause settings reported by the PHY should also be applied to the GMII port > status override otherwise the switch will not generate pause frames towards the > link partner despite the advertisement saying otherwise. > > Fixes: 246d7f773c13 ("net: dsa: add Broadcom SF2 switch driver") > Signed-off-by: Doug Berger <opendmb@gmail.com> > Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> > --- > drivers/net/dsa/bcm_sf2.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/drivers/net/dsa/bcm_sf2.c b/drivers/net/dsa/bcm_sf2.c > index 40b3adf7ad99..03f38c36e188 100644 > --- a/drivers/net/dsa/bcm_sf2.c > +++ b/drivers/net/dsa/bcm_sf2.c > @@ -671,6 +671,11 @@ static void bcm_sf2_sw_adjust_link(struct dsa_switch *ds, int port, > reg |= LINK_STS; > if (phydev->duplex == DUPLEX_FULL) > reg |= DUPLX_MODE; > + if (phydev->pause) { > + if (phydev->asym_pause) > + reg |= TXFLOW_CNTL; > + reg |= RXFLOW_CNTL; > + } Is this correct? phydev->pause and phydev->asym_pause keep the Pause and Asym_Pause bits advertised by the link partner. In other words, in this manual resolution you are ignoring what the local switch port has advertised. To give you an example (I'm looking at the infamous Table 28B–3—Pause resolution from IEEE 802.3). Your logic, simplified, says: enable TXFLOW_CNTL as local device resolution iff the link partner advertised PAUSE=1 && ASM_DIR=1, and RXFLOW_CNTL if the link partner advertised PAUSE=1 at all. The most trivial counter-example is if we (local device) advertise PAUSE=0 ASM_DIR=0 (this can be achieved with ethtool rx off tx off). That is one way to rig the negotiated flow control to off/off regardless of what the link partner has advertised. But your logic would enable local capabilities just because the link partner said so, ignoring what was advertised here. Another example is if the local station advertises PAUSE=1, ASM_DIR=1, and remote PAUSE=0, ASM_DIR=1. According to IEEE, the link partners should agree on "rx on tx off" for the local system, and "tx on rx off" for the link partner. But your logic does not even enter the first "if" condition, because phydev->pause (link partner's PAUSE advertisement) is 0. So neither TXFLOW_CNTL nor RXFLOW_CNTL will be enabled locally, despite the expectation and what is printed by phy_print_status(). > > core_writel(priv, reg, CORE_STS_OVERRIDE_GMIIP_PORT(port)); > > -- > 2.25.1 >
On Tue, Jul 05, 2022 at 02:34:57AM +0300, Vladimir Oltean wrote: > Hi Florian, > > On Mon, Jul 04, 2022 at 08:35:07AM -0700, Florian Fainelli wrote: > > From: Doug Berger <opendmb@gmail.com> > > > > commit 7c97bc0128b2eecc703106112679a69d446d1a12 upstream > > > > The pause settings reported by the PHY should also be applied to the GMII port > > status override otherwise the switch will not generate pause frames towards the > > link partner despite the advertisement saying otherwise. > > > > Fixes: 246d7f773c13 ("net: dsa: add Broadcom SF2 switch driver") > > Signed-off-by: Doug Berger <opendmb@gmail.com> > > Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> > > --- > > drivers/net/dsa/bcm_sf2.c | 5 +++++ > > 1 file changed, 5 insertions(+) > > > > diff --git a/drivers/net/dsa/bcm_sf2.c b/drivers/net/dsa/bcm_sf2.c > > index 40b3adf7ad99..03f38c36e188 100644 > > --- a/drivers/net/dsa/bcm_sf2.c > > +++ b/drivers/net/dsa/bcm_sf2.c > > @@ -671,6 +671,11 @@ static void bcm_sf2_sw_adjust_link(struct dsa_switch *ds, int port, > > reg |= LINK_STS; > > if (phydev->duplex == DUPLEX_FULL) > > reg |= DUPLX_MODE; > > + if (phydev->pause) { > > + if (phydev->asym_pause) > > + reg |= TXFLOW_CNTL; > > + reg |= RXFLOW_CNTL; > > + } > > Is this correct? phydev->pause and phydev->asym_pause keep the Pause and > Asym_Pause bits advertised by the link partner. In other words, in this > manual resolution you are ignoring what the local switch port has > advertised. linkmode_resolve_pause() is not used yet outside of phylink, but should help here. Andrew
On Tue, Jul 05, 2022 at 08:56:16AM +0200, Andrew Lunn wrote: > linkmode_resolve_pause() is not used yet outside of phylink, but > should help here. I don't think linkmode_resolve_pause() was present in 4.9. I'm looking at what gianfar was doing at the time (gfar_get_flowctrl_cfg) and I see mii_resolve_flowctrl_fdx() as a potentially relevant helper.
diff --git a/drivers/net/dsa/bcm_sf2.c b/drivers/net/dsa/bcm_sf2.c index 40b3adf7ad99..03f38c36e188 100644 --- a/drivers/net/dsa/bcm_sf2.c +++ b/drivers/net/dsa/bcm_sf2.c @@ -671,6 +671,11 @@ static void bcm_sf2_sw_adjust_link(struct dsa_switch *ds, int port, reg |= LINK_STS; if (phydev->duplex == DUPLEX_FULL) reg |= DUPLX_MODE; + if (phydev->pause) { + if (phydev->asym_pause) + reg |= TXFLOW_CNTL; + reg |= RXFLOW_CNTL; + } core_writel(priv, reg, CORE_STS_OVERRIDE_GMIIP_PORT(port));