Re: [PATCH] net: dsa: mv88e6xxx: rework in-chip bridging
On 09/10/17 14:00, Florian Fainelli wrote: > Le 10/08/17 à 20:23, Greg Ungerer a écrit : >> On 07/10/17 13:04, Florian Fainelli wrote: >>> Le 10/03/17 à 23:20, Greg Ungerer a écrit : On Wed, Mar 29, 2017 at 04:30:16PM -0400, Vivien Didelot wrote: > All ports -- internal and external, for chips featuring a PVT -- have a > mask restricting to which internal ports a frame is allowed to egress. > > Now that DSA exposes the number of ports and their bridge devices, it is > possible to extract the code generating the VLAN map and make it generic > so that it can be shared later with the cross-chip bridging code. This patch changes the behavior of interfaces on startup if they are not part of a bridge. I have a board with a Marvell 6350 switch with a device tree that sets up the 5 ports as lan1, lan2, lan3, lan4, wan. With kernels before this patch (so linux-4.12 and older) after system startup I could do: ifconfig lan1 192.168.0.1 And then ping out that interface with no problems. After this patch is applied (effects linux-4.13 and newer) then the ping fails: PING 192.168.0.22 (192.168.0.22) 56(84) bytes of data. From 192.168.0.1 icmp_seq=1 Destination Host Unreachable From 192.168.0.1 icmp_seq=2 Destination Host Unreachable From 192.168.0.1 icmp_seq=3 Destination Host Unreachable If I incorporate an interface into a bridge then it all works ok. So simply: brctl addbr br0 brctl addif br0 lan1 ifconfig lan1 up ifconfig br0 192.168.0.1 Then pings out work as expected. And if I now remove that lan1 interface from the bridge and use it alone again then it will now work ok: ifconfig br0 down brctl delif br0 lan1 ifconfig lan1 192.168.0.1 And that now pings ok. I fixed this with the attached patch. It is probably not the correct approach, but it does restore the older behavior. What do you think? >>> >>> This is strange, the dsa_switch_tree and its associated dsa_switch >>> instances should be fully setup by the time ops->setup() is running in >>> your driver but your patch suggests this may not be happening? >> >> That is what I am seeing, yep. >> >> >>> Are you using the new style Device Tree binding or the old style Device >>> Tree binding out of curiosity? >> >> This is my device tree fragment for the switch: >> >> dsa@0 { >> compatible = "marvell,dsa"; >> #address-cells = <2>; >> #size-cells = <0>; >> >> dsa,ethernet = <>; >> dsa,mii-bus = <>; >> >> switch@0 { >> #address-cells = <1>; >> #size-cells = <0>; >> reg = <0x11 0>; >> >> port@0 { >> reg = <0>; >> label = "lan1"; >> }; >> port@1 { >> reg = <1>; >> label = "lan2"; >> }; >> port@2 { >> reg = <2>; >> label = "lan3"; >> }; >> port@3 { >> reg = <3>; >> label = "lan4"; >> }; >> port@4 { >> reg = <4>; >> label = "wan"; >> }; >> port@5 { >> reg = <5>; >> label = "cpu"; >> }; >> }; >> }; >> >> The board I am using is based around an Marvell Armada 370. This device tree >> setup looks pretty similar to the other Marvell boards using marvell,dsa. > > This is the old Device Tree binding which goes through an unfortunately > different code path while initializing all the dsa_switch_tree and > dsa_switch structures, while we should definitively look into fixing > this, would you mind trying to update your board using something similar > to this commit: > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=4cb2ec8cad8c82cd7cfd19edcacd846861d6e703 Ok, converted the devicetree to use this form of dsa setup. Does not show the problem, on first boot lan1 works properly. > This would make you go through net/dsa/dsa2.c which is what most of us > usually test. In the meantime we should probably start issuing warning > messages when people use the old Device Tree binding to encourage them > to migrate other. Yeah, I had not noticed that the devicetree dsa setup had changed, I have been using my devicetree for quite a few kernel
Re: [PATCH] net: dsa: mv88e6xxx: rework in-chip bridging
Le 10/08/17 à 20:23, Greg Ungerer a écrit : > Hi Florian, > > On 07/10/17 13:04, Florian Fainelli wrote: >> Le 10/03/17 à 23:20, Greg Ungerer a écrit : >>> On Wed, Mar 29, 2017 at 04:30:16PM -0400, Vivien Didelot wrote: All ports -- internal and external, for chips featuring a PVT -- have a mask restricting to which internal ports a frame is allowed to egress. Now that DSA exposes the number of ports and their bridge devices, it is possible to extract the code generating the VLAN map and make it generic so that it can be shared later with the cross-chip bridging code. >>> >>> This patch changes the behavior of interfaces on startup if they are >>> not part of a bridge. >>> >>> I have a board with a Marvell 6350 switch with a device tree that sets >>> up the 5 ports as lan1, lan2, lan3, lan4, wan. With kernels before >>> this patch (so linux-4.12 and older) after system startup I could do: >>> >>> ifconfig lan1 192.168.0.1 >>> >>> And then ping out that interface with no problems. >>> >>> After this patch is applied (effects linux-4.13 and newer) then the >>> ping fails: >>> >>> PING 192.168.0.22 (192.168.0.22) 56(84) bytes of data. >>> From 192.168.0.1 icmp_seq=1 Destination Host Unreachable >>> From 192.168.0.1 icmp_seq=2 Destination Host Unreachable >>> From 192.168.0.1 icmp_seq=3 Destination Host Unreachable >>> >>> If I incorporate an interface into a bridge then it all works ok. >>> So simply: >>> >>> brctl addbr br0 >>> brctl addif br0 lan1 >>> ifconfig lan1 up >>> ifconfig br0 192.168.0.1 >>> >>> Then pings out work as expected. And if I now remove that lan1 >>> interface from the bridge and use it alone again then it will >>> now work ok: >>> >>> ifconfig br0 down >>> brctl delif br0 lan1 >>> ifconfig lan1 192.168.0.1 >>> >>> And that now pings ok. >>> >>> I fixed this with the attached patch. It is probably not the correct >>> approach, but it does restore the older behavior. >>> >>> What do you think? >> >> This is strange, the dsa_switch_tree and its associated dsa_switch >> instances should be fully setup by the time ops->setup() is running in >> your driver but your patch suggests this may not be happening? > > That is what I am seeing, yep. > > >> Are you using the new style Device Tree binding or the old style Device >> Tree binding out of curiosity? > > This is my device tree fragment for the switch: > > dsa@0 { > compatible = "marvell,dsa"; > #address-cells = <2>; > #size-cells = <0>; > > dsa,ethernet = <>; > dsa,mii-bus = <>; > > switch@0 { > #address-cells = <1>; > #size-cells = <0>; > reg = <0x11 0>; > > port@0 { > reg = <0>; > label = "lan1"; > }; > port@1 { > reg = <1>; > label = "lan2"; > }; > port@2 { > reg = <2>; > label = "lan3"; > }; > port@3 { > reg = <3>; > label = "lan4"; > }; > port@4 { > reg = <4>; > label = "wan"; > }; > port@5 { > reg = <5>; > label = "cpu"; > }; > }; > }; > > The board I am using is based around an Marvell Armada 370. This device tree > setup looks pretty similar to the other Marvell boards using marvell,dsa. This is the old Device Tree binding which goes through an unfortunately different code path while initializing all the dsa_switch_tree and dsa_switch structures, while we should definitively look into fixing this, would you mind trying to update your board using something similar to this commit: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=4cb2ec8cad8c82cd7cfd19edcacd846861d6e703 This would make you go through net/dsa/dsa2.c which is what most of us usually test. In the meantime we should probably start issuing warning messages when people use the old Device Tree binding to encourage them to migrate other. -- Florian
Re: [PATCH] net: dsa: mv88e6xxx: rework in-chip bridging
Hi Florian, On 07/10/17 13:04, Florian Fainelli wrote: > Le 10/03/17 à 23:20, Greg Ungerer a écrit : >> On Wed, Mar 29, 2017 at 04:30:16PM -0400, Vivien Didelot wrote: >>> All ports -- internal and external, for chips featuring a PVT -- have a >>> mask restricting to which internal ports a frame is allowed to egress. >>> >>> Now that DSA exposes the number of ports and their bridge devices, it is >>> possible to extract the code generating the VLAN map and make it generic >>> so that it can be shared later with the cross-chip bridging code. >> >> This patch changes the behavior of interfaces on startup if they are >> not part of a bridge. >> >> I have a board with a Marvell 6350 switch with a device tree that sets >> up the 5 ports as lan1, lan2, lan3, lan4, wan. With kernels before >> this patch (so linux-4.12 and older) after system startup I could do: >> >> ifconfig lan1 192.168.0.1 >> >> And then ping out that interface with no problems. >> >> After this patch is applied (effects linux-4.13 and newer) then the >> ping fails: >> >> PING 192.168.0.22 (192.168.0.22) 56(84) bytes of data. >> From 192.168.0.1 icmp_seq=1 Destination Host Unreachable >> From 192.168.0.1 icmp_seq=2 Destination Host Unreachable >> From 192.168.0.1 icmp_seq=3 Destination Host Unreachable >> >> If I incorporate an interface into a bridge then it all works ok. >> So simply: >> >> brctl addbr br0 >> brctl addif br0 lan1 >> ifconfig lan1 up >> ifconfig br0 192.168.0.1 >> >> Then pings out work as expected. And if I now remove that lan1 >> interface from the bridge and use it alone again then it will >> now work ok: >> >> ifconfig br0 down >> brctl delif br0 lan1 >> ifconfig lan1 192.168.0.1 >> >> And that now pings ok. >> >> I fixed this with the attached patch. It is probably not the correct >> approach, but it does restore the older behavior. >> >> What do you think? > > This is strange, the dsa_switch_tree and its associated dsa_switch > instances should be fully setup by the time ops->setup() is running in > your driver but your patch suggests this may not be happening? That is what I am seeing, yep. > Are you using the new style Device Tree binding or the old style Device > Tree binding out of curiosity? This is my device tree fragment for the switch: dsa@0 { compatible = "marvell,dsa"; #address-cells = <2>; #size-cells = <0>; dsa,ethernet = <>; dsa,mii-bus = <>; switch@0 { #address-cells = <1>; #size-cells = <0>; reg = <0x11 0>; port@0 { reg = <0>; label = "lan1"; }; port@1 { reg = <1>; label = "lan2"; }; port@2 { reg = <2>; label = "lan3"; }; port@3 { reg = <3>; label = "lan4"; }; port@4 { reg = <4>; label = "wan"; }; port@5 { reg = <5>; label = "cpu"; }; }; }; The board I am using is based around an Marvell Armada 370. This device tree setup looks pretty similar to the other Marvell boards using marvell,dsa. Regards Greg >>> Signed-off-by: Vivien Didelot>>> --- >>> drivers/net/dsa/mv88e6xxx/chip.c | 53 >>> ++-- >>> 1 file changed, 34 insertions(+), 19 deletions(-) >>> >>> diff --git a/drivers/net/dsa/mv88e6xxx/chip.c >>> b/drivers/net/dsa/mv88e6xxx/chip.c >>> index b114bf8e6a11..e5165831e8b5 100644 >>> --- a/drivers/net/dsa/mv88e6xxx/chip.c >>> +++ b/drivers/net/dsa/mv88e6xxx/chip.c >>> @@ -1123,27 +1123,42 @@ static int mv88e6xxx_set_eee(struct dsa_switch *ds, >>> int >>> port, >>> return err; >>> } >>> >>> +static u16 mv88e6xxx_port_vlan(struct mv88e6xxx_chip *chip, int dev, int >>> port) >>> +{ >>> + struct dsa_switch *ds = NULL; >>> + struct net_device *br; >>> + u16 pvlan; >>> + int i; >>> + >>> + if (dev < DSA_MAX_SWITCHES) >>> + ds = chip->ds->dst->ds[dev]; >>> + >>> + /* Prevent frames from unknown switch or port */ >>> + if (!ds || port >= ds->num_ports) >>> + return 0; >>> + >>> + /* Frames from DSA links and CPU ports can egress any local port */ >>> + if (dsa_is_cpu_port(ds, port) || dsa_is_dsa_port(ds, port)) >>> + return
Re: [PATCH] net: dsa: mv88e6xxx: rework in-chip bridging
Le 10/03/17 à 23:20, Greg Ungerer a écrit : > Hi Vivien, > > On Wed, Mar 29, 2017 at 04:30:16PM -0400, Vivien Didelot wrote: >> All ports -- internal and external, for chips featuring a PVT -- have a >> mask restricting to which internal ports a frame is allowed to egress. >> >> Now that DSA exposes the number of ports and their bridge devices, it is >> possible to extract the code generating the VLAN map and make it generic >> so that it can be shared later with the cross-chip bridging code. > > This patch changes the behavior of interfaces on startup if they are > not part of a bridge. > > I have a board with a Marvell 6350 switch with a device tree that sets > up the 5 ports as lan1, lan2, lan3, lan4, wan. With kernels before > this patch (so linux-4.12 and older) after system startup I could do: > > ifconfig lan1 192.168.0.1 > > And then ping out that interface with no problems. > > After this patch is applied (effects linux-4.13 and newer) then the > ping fails: > > PING 192.168.0.22 (192.168.0.22) 56(84) bytes of data. > From 192.168.0.1 icmp_seq=1 Destination Host Unreachable > From 192.168.0.1 icmp_seq=2 Destination Host Unreachable > From 192.168.0.1 icmp_seq=3 Destination Host Unreachable > > If I incorporate an interface into a bridge then it all works ok. > So simply: > > brctl addbr br0 > brctl addif br0 lan1 > ifconfig lan1 up > ifconfig br0 192.168.0.1 > > Then pings out work as expected. And if I now remove that lan1 > interface from the bridge and use it alone again then it will > now work ok: > > ifconfig br0 down > brctl delif br0 lan1 > ifconfig lan1 192.168.0.1 > > And that now pings ok. > > I fixed this with the attached patch. It is probably not the correct > approach, but it does restore the older behavior. > > What do you think? This is strange, the dsa_switch_tree and its associated dsa_switch instances should be fully setup by the time ops->setup() is running in your driver but your patch suggests this may not be happening? Are you using the new style Device Tree binding or the old style Device Tree binding out of curiosity? > > Regards > Greg > > > >> Signed-off-by: Vivien Didelot>> --- >> drivers/net/dsa/mv88e6xxx/chip.c | 53 >> ++-- >> 1 file changed, 34 insertions(+), 19 deletions(-) >> >> diff --git a/drivers/net/dsa/mv88e6xxx/chip.c >> b/drivers/net/dsa/mv88e6xxx/chip.c >> index b114bf8e6a11..e5165831e8b5 100644 >> --- a/drivers/net/dsa/mv88e6xxx/chip.c >> +++ b/drivers/net/dsa/mv88e6xxx/chip.c >> @@ -1123,27 +1123,42 @@ static int mv88e6xxx_set_eee(struct dsa_switch *ds, >> int >> port, >> return err; >> } >> >> +static u16 mv88e6xxx_port_vlan(struct mv88e6xxx_chip *chip, int dev, int >> port) >> +{ >> + struct dsa_switch *ds = NULL; >> + struct net_device *br; >> + u16 pvlan; >> + int i; >> + >> + if (dev < DSA_MAX_SWITCHES) >> + ds = chip->ds->dst->ds[dev]; >> + >> + /* Prevent frames from unknown switch or port */ >> + if (!ds || port >= ds->num_ports) >> + return 0; >> + >> + /* Frames from DSA links and CPU ports can egress any local port */ >> + if (dsa_is_cpu_port(ds, port) || dsa_is_dsa_port(ds, port)) >> + return mv88e6xxx_port_mask(chip); >> + >> + br = ds->ports[port].bridge_dev; >> + pvlan = 0; >> + >> + /* Frames from user ports can egress any local DSA links and CPU >> ports, >> +* as well as any local member of their bridge group. >> +*/ >> + for (i = 0; i < mv88e6xxx_num_ports(chip); ++i) >> + if (dsa_is_cpu_port(chip->ds, i) || >> + dsa_is_dsa_port(chip->ds, i) || >> + (br && chip->ds->ports[i].bridge_dev == br)) >> + pvlan |= BIT(i); >> + >> + return pvlan; >> +} >> + >> static int _mv88e6xxx_port_based_vlan_map(struct mv88e6xxx_chip *chip, int >> port) >> { >> - struct dsa_switch *ds = chip->ds; >> - struct net_device *bridge = ds->ports[port].bridge_dev; >> - u16 output_ports = 0; >> - int i; >> - >> - /* allow CPU port or DSA link(s) to send frames to every port */ >> - if (dsa_is_cpu_port(ds, port) || dsa_is_dsa_port(ds, port)) { >> - output_ports = ~0; >> - } else { >> - for (i = 0; i < mv88e6xxx_num_ports(chip); ++i) { >> - /* allow sending frames to every group member */ >> - if (bridge && ds->ports[i].bridge_dev == bridge) >> - output_ports |= BIT(i); >> - >> - /* allow sending frames to CPU port and DSA link(s) >> */ >> - if (dsa_is_cpu_port(ds, i) || dsa_is_dsa_port(ds, i)) >> - output_ports |= BIT(i); >> - } >> - } >> + u16 output_ports =
Re: [PATCH] net: dsa: mv88e6xxx: rework in-chip bridging
Hi Vivien, On Wed, Mar 29, 2017 at 04:30:16PM -0400, Vivien Didelot wrote: > All ports -- internal and external, for chips featuring a PVT -- have a > mask restricting to which internal ports a frame is allowed to egress. > > Now that DSA exposes the number of ports and their bridge devices, it is > possible to extract the code generating the VLAN map and make it generic > so that it can be shared later with the cross-chip bridging code. This patch changes the behavior of interfaces on startup if they are not part of a bridge. I have a board with a Marvell 6350 switch with a device tree that sets up the 5 ports as lan1, lan2, lan3, lan4, wan. With kernels before this patch (so linux-4.12 and older) after system startup I could do: ifconfig lan1 192.168.0.1 And then ping out that interface with no problems. After this patch is applied (effects linux-4.13 and newer) then the ping fails: PING 192.168.0.22 (192.168.0.22) 56(84) bytes of data. From 192.168.0.1 icmp_seq=1 Destination Host Unreachable From 192.168.0.1 icmp_seq=2 Destination Host Unreachable From 192.168.0.1 icmp_seq=3 Destination Host Unreachable If I incorporate an interface into a bridge then it all works ok. So simply: brctl addbr br0 brctl addif br0 lan1 ifconfig lan1 up ifconfig br0 192.168.0.1 Then pings out work as expected. And if I now remove that lan1 interface from the bridge and use it alone again then it will now work ok: ifconfig br0 down brctl delif br0 lan1 ifconfig lan1 192.168.0.1 And that now pings ok. I fixed this with the attached patch. It is probably not the correct approach, but it does restore the older behavior. What do you think? Regards Greg > Signed-off-by: Vivien Didelot> --- > drivers/net/dsa/mv88e6xxx/chip.c | 53 > ++-- > 1 file changed, 34 insertions(+), 19 deletions(-) > > diff --git a/drivers/net/dsa/mv88e6xxx/chip.c > b/drivers/net/dsa/mv88e6xxx/chip.c > index b114bf8e6a11..e5165831e8b5 100644 > --- a/drivers/net/dsa/mv88e6xxx/chip.c > +++ b/drivers/net/dsa/mv88e6xxx/chip.c > @@ -1123,27 +1123,42 @@ static int mv88e6xxx_set_eee(struct dsa_switch *ds, > int > port, > return err; > } > > +static u16 mv88e6xxx_port_vlan(struct mv88e6xxx_chip *chip, int dev, int > port) > +{ > + struct dsa_switch *ds = NULL; > + struct net_device *br; > + u16 pvlan; > + int i; > + > + if (dev < DSA_MAX_SWITCHES) > + ds = chip->ds->dst->ds[dev]; > + > + /* Prevent frames from unknown switch or port */ > + if (!ds || port >= ds->num_ports) > + return 0; > + > + /* Frames from DSA links and CPU ports can egress any local port */ > + if (dsa_is_cpu_port(ds, port) || dsa_is_dsa_port(ds, port)) > + return mv88e6xxx_port_mask(chip); > + > + br = ds->ports[port].bridge_dev; > + pvlan = 0; > + > + /* Frames from user ports can egress any local DSA links and CPU > ports, > +* as well as any local member of their bridge group. > +*/ > + for (i = 0; i < mv88e6xxx_num_ports(chip); ++i) > + if (dsa_is_cpu_port(chip->ds, i) || > + dsa_is_dsa_port(chip->ds, i) || > + (br && chip->ds->ports[i].bridge_dev == br)) > + pvlan |= BIT(i); > + > + return pvlan; > +} > + > static int _mv88e6xxx_port_based_vlan_map(struct mv88e6xxx_chip *chip, int > port) > { > - struct dsa_switch *ds = chip->ds; > - struct net_device *bridge = ds->ports[port].bridge_dev; > - u16 output_ports = 0; > - int i; > - > - /* allow CPU port or DSA link(s) to send frames to every port */ > - if (dsa_is_cpu_port(ds, port) || dsa_is_dsa_port(ds, port)) { > - output_ports = ~0; > - } else { > - for (i = 0; i < mv88e6xxx_num_ports(chip); ++i) { > - /* allow sending frames to every group member */ > - if (bridge && ds->ports[i].bridge_dev == bridge) > - output_ports |= BIT(i); > - > - /* allow sending frames to CPU port and DSA link(s) */ > - if (dsa_is_cpu_port(ds, i) || dsa_is_dsa_port(ds, i)) > - output_ports |= BIT(i); > - } > - } > + u16 output_ports = mv88e6xxx_port_vlan(chip, chip->ds->index, port); > > /* prevent frames from going back out of the port they came in on */ > output_ports &= ~BIT(port); > -- >From d81c15dda9fd84895caa67e4c187ad0b6b33af1f Mon Sep 17 00:00:00 2001 From: Greg Ungerer Date: Thu, 21 Sep 2017 13:35:35 +1000 Subject: [PATCH] net: dsa: mv88e6xxx: fix first-use port breakage The very first time you configure a switch port for use - as a stand alone port - the interface cannot pass packets. If you configure the interface as part