From: Arnd Bergmann <arnd@arndb.de>
Multiple switchdev drivers depend on CONFIG_NET_SWITCHDEV in Kconfig,
but have also gained a dependency on the bridge driver as they now
call switchdev_bridge_port_offload():
drivers/net/ethernet/microchip/sparx5/sparx5_switchdev.o: In function `sparx5_netdevice_event':
sparx5_switchdev.c:(.text+0x3cc): undefined reference to `switchdev_bridge_port_offload'
drivers/net/ethernet/ti/cpsw_new.o: In function `cpsw_netdevice_event':
cpsw_new.c:(.text+0x1098): undefined reference to `switchdev_bridge_port_offload'
Some of these drivers already have a 'BRIDGE || !BRIDGE' dependency
that avoids the link failure, but the 'rocker' driver was missing this
For MLXSW/MLX5, SPARX5_SWITCH, and TI_K3_AM65_CPSW_NUSS, the
driver can conditionally use switchdev support, which is then guarded
by another Kconfig symbol. For these, add a dependency on a new Kconfig
symbol NET_MAY_USE_SWITCHDEV that is defined to correctly model the
dependency: if switchdev support is enabled, these drivers cannot be
built-in when bridge support is in a module, but if either bridge or
switchdev is disabled, or both are built-in, there is no such restriction.
Fixes: 2f5dc00f7a3e ("net: bridge: switchdev: let drivers inform which bridge ports are offloaded")
Fixes: b0e81817629a ("net: build all switchdev drivers as modules when the bridge is a module")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
This version seems to pass my randconfig builds for the moment,
but that doesn't mean it's correct either. Please have a closer
look before this gets applied.
---
drivers/net/ethernet/mellanox/mlx5/core/Kconfig | 1 +
drivers/net/ethernet/mellanox/mlxsw/Kconfig | 1 +
drivers/net/ethernet/netronome/Kconfig | 1 +
drivers/net/ethernet/ti/Kconfig | 1 +
net/switchdev/Kconfig | 5 +++++
5 files changed, 9 insertions(+)
From: Simon Horman <hidden> Date: 2021-08-02 16:23:07
On Mon, Aug 02, 2021 at 04:47:28PM +0200, Arnd Bergmann wrote:
quoted hunk
From: Arnd Bergmann <arnd@arndb.de>
Multiple switchdev drivers depend on CONFIG_NET_SWITCHDEV in Kconfig,
but have also gained a dependency on the bridge driver as they now
call switchdev_bridge_port_offload():
drivers/net/ethernet/microchip/sparx5/sparx5_switchdev.o: In function `sparx5_netdevice_event':
sparx5_switchdev.c:(.text+0x3cc): undefined reference to `switchdev_bridge_port_offload'
drivers/net/ethernet/ti/cpsw_new.o: In function `cpsw_netdevice_event':
cpsw_new.c:(.text+0x1098): undefined reference to `switchdev_bridge_port_offload'
Some of these drivers already have a 'BRIDGE || !BRIDGE' dependency
that avoids the link failure, but the 'rocker' driver was missing this
For MLXSW/MLX5, SPARX5_SWITCH, and TI_K3_AM65_CPSW_NUSS, the
driver can conditionally use switchdev support, which is then guarded
by another Kconfig symbol. For these, add a dependency on a new Kconfig
symbol NET_MAY_USE_SWITCHDEV that is defined to correctly model the
dependency: if switchdev support is enabled, these drivers cannot be
built-in when bridge support is in a module, but if either bridge or
switchdev is disabled, or both are built-in, there is no such restriction.
Fixes: 2f5dc00f7a3e ("net: bridge: switchdev: let drivers inform which bridge ports are offloaded")
Fixes: b0e81817629a ("net: build all switchdev drivers as modules when the bridge is a module")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
This version seems to pass my randconfig builds for the moment,
but that doesn't mean it's correct either. Please have a closer
look before this gets applied.
---
drivers/net/ethernet/mellanox/mlx5/core/Kconfig | 1 +
drivers/net/ethernet/mellanox/mlxsw/Kconfig | 1 +
drivers/net/ethernet/netronome/Kconfig | 1 +
drivers/net/ethernet/ti/Kconfig | 1 +
net/switchdev/Kconfig | 5 +++++
5 files changed, 9 insertions(+)
I think it is MLXSW_SPECTRUM rather than MLXSW_CORE
that controls compilation of spectrum_switchdev.c
which calls switchdev_bridge_port_offload.
But MLXSW_SPECTRUM seems to already depend on BRIDGE || BRIDGE=n
I believe this has already been addressed by the following patch in net
b0e81817629a ("net: build all switchdev drivers as modules when the bridge is a module")
On Mon, Aug 2, 2021 at 6:23 PM Simon Horman [off-list ref] wrote:
On Mon, Aug 02, 2021 at 04:47:28PM +0200, Arnd Bergmann wrote:
quoted
---
This version seems to pass my randconfig builds for the moment,
but that doesn't mean it's correct either. Please have a closer
look before this gets applied.
Thank you for taking a look, it seems I have done a particularly bad
job rebasing
the patch on top of the previous fix, leaving only the wrong bits ;-)
MLX5_CORE does not appear to cover code that calls
switchdev_bridge_port_offload.
Ah right, I did get a link failure with my test build, but it was an
unrelated one:
ld: drivers/net/ethernet/mellanox/mlx5/core/esw/sample.o: in function
`mlx5_esw_sample_skb':
sample.c:(.text+0x5b4): undefined reference to `psample_sample_packet'
I think that one needs
I think it is MLXSW_SPECTRUM rather than MLXSW_CORE
that controls compilation of spectrum_switchdev.c
which calls switchdev_bridge_port_offload.
But MLXSW_SPECTRUM seems to already depend on BRIDGE || BRIDGE=n
This seems wrong, the NFP driver doesn't call
switchdev_bridge_port_offload()
Ah right, I actually noticed that earlier and then forgot to remove that hunk.
Also: is this actually intended or should the driver call
switchdev_bridge_port_offload() like the other switchdev drivers do?
I believe this has already been addressed by the following patch in net
b0e81817629a ("net: build all switchdev drivers as modules when the bridge is a module")
I think the fix was wrong here, and that hunk should be reverted.
The dependency was added to a bool option, where it does not have
the intended effect.
I think this is the only remaining thing needed from my patch, so
the NET_MAY_USE_SWITCHDEV option is not needed either,
and it could be written as:
From: Simon Horman <hidden> Date: 2021-08-02 18:51:34
On Mon, Aug 02, 2021 at 08:29:25PM +0200, Arnd Bergmann wrote:
On Mon, Aug 2, 2021 at 6:23 PM Simon Horman [off-list ref] wrote:
quoted
On Mon, Aug 02, 2021 at 04:47:28PM +0200, Arnd Bergmann wrote:
quoted
---
This version seems to pass my randconfig builds for the moment,
but that doesn't mean it's correct either. Please have a closer
look before this gets applied.
Thank you for taking a look, it seems I have done a particularly bad
job rebasing
the patch on top of the previous fix, leaving only the wrong bits ;-)
This seems wrong, the NFP driver doesn't call
switchdev_bridge_port_offload()
Ah right, I actually noticed that earlier and then forgot to remove that hunk.
Also: is this actually intended or should the driver call
switchdev_bridge_port_offload() like the other switchdev drivers do?
The NFP driver doesn't implement /use this part of the switchdev API.
It is intentional given the current scope of features implemented.
I believe this has already been addressed by the following patch in net
b0e81817629a ("net: build all switchdev drivers as modules when the bridge is a module")
I think the fix was wrong here, and that hunk should be reverted.
The dependency was added to a bool option, where it does not have
the intended effect.
I think this is the only remaining thing needed from my patch, so
the NET_MAY_USE_SWITCHDEV option is not needed either,
and it could be written as:
If this looks correct to you, I can submit it as a standalone patch.
Is there a conflict between your proposed change and the
presence of "depends on NET_SWITCHDEV" ?
BTW, I'm not sure I know how to compile TI_K3_AM65_CPSW_NUSS.
So I'm perhaps not the best arbiter of correctness here.
From: Vladimir Oltean <vladimir.oltean@nxp.com> Date: 2021-08-02 20:20:58
On Mon, Aug 02, 2021 at 09:55:20PM +0200, Arnd Bergmann wrote:
On Mon, Aug 2, 2021 at 9:05 PM Vladimir Oltean [off-list ref] wrote:
quoted
On Mon, Aug 02, 2021 at 08:29:25PM +0200, Arnd Bergmann wrote:
quoted
If this looks correct to you, I can submit it as a standalone patch.
I think it's easiest I just ask you to provide a .config that triggers
actual build failures and we can go from there.
This one is with an arm64 allmodconfig, plus
CONFIG_PTP_1588_CLOCK=y
CONFIG_TI_K3_AM65_CPTS=y
CONFIG_TI_K3_AM65_CPSW_NUSS=y
Yeah, ok, I remember now, I saw that TI_CPSW_SWITCHDEV is tristate, and
incorrectly thought that TI_K3_AM65_CPSW_SWITCHDEV (which is mostly a
copy-paste job of the main cpsw anyway, makes you cringe that they wrote
a separate driver for it) is tristate too.
The options are either to make TI_K3_AM65_CPSW_SWITCHDEV tristate like
TI_CPSW_SWITCHDEV is, and to edit the Makefile accordingly to make
am65-cpsw-switchdev.o part of obj-$(CONFIG_TI_K3_AM65_CPSW_SWITCHDEV),
or to extend the BRIDGE || BRIDGE=n dependency to TI_K3_AM65_CPSW_NUSS
which is the direct tristate dependency of CONFIG_TI_K3_AM65_CPSW_SWITCHDEV,
and to make CONFIG_TI_K3_AM65_CPSW_SWITCHDEV simply depend on BRIDGE.
On Mon, Aug 2, 2021 at 10:21 PM Vladimir Oltean [off-list ref] wrote:
On Mon, Aug 02, 2021 at 09:55:20PM +0200, Arnd Bergmann wrote:
quoted
On Mon, Aug 2, 2021 at 9:05 PM Vladimir Oltean [off-list ref] wrote:
quoted
On Mon, Aug 02, 2021 at 08:29:25PM +0200, Arnd Bergmann wrote:
quoted
If this looks correct to you, I can submit it as a standalone patch.
I think it's easiest I just ask you to provide a .config that triggers
actual build failures and we can go from there.
This one is with an arm64 allmodconfig, plus
CONFIG_PTP_1588_CLOCK=y
CONFIG_TI_K3_AM65_CPTS=y
CONFIG_TI_K3_AM65_CPSW_NUSS=y
Yeah, ok, I remember now, I saw that TI_CPSW_SWITCHDEV is tristate, and
incorrectly thought that TI_K3_AM65_CPSW_SWITCHDEV (which is mostly a
copy-paste job of the main cpsw anyway, makes you cringe that they wrote
a separate driver for it) is tristate too.
Right.
The options are either to make TI_K3_AM65_CPSW_SWITCHDEV tristate like
TI_CPSW_SWITCHDEV is, and to edit the Makefile accordingly to make
am65-cpsw-switchdev.o part of obj-$(CONFIG_TI_K3_AM65_CPSW_SWITCHDEV),
This probably won't work as this is currently part of the ti-am65-cpsw-nuss.ko
module, so I would assume that it's not easy to separate from the main module.
or to extend the BRIDGE || BRIDGE=n dependency to TI_K3_AM65_CPSW_NUSS
which is the direct tristate dependency of CONFIG_TI_K3_AM65_CPSW_SWITCHDEV,
That would work, but it's slightly more heavy-handed than my proposal, as this
prevents TI_K3_AM65_CPSW_NUSS from being built-in when BRIDGE is a module,
even when switchdev support is completely disabled.
and to make CONFIG_TI_K3_AM65_CPSW_SWITCHDEV simply depend on BRIDGE.
From: Vladimir Oltean <vladimir.oltean@nxp.com> Date: 2021-08-02 20:53:40
On Mon, Aug 02, 2021 at 10:44:04PM +0200, Arnd Bergmann wrote:
quoted
or to extend the BRIDGE || BRIDGE=n dependency to TI_K3_AM65_CPSW_NUSS
which is the direct tristate dependency of CONFIG_TI_K3_AM65_CPSW_SWITCHDEV,
That would work, but it's slightly more heavy-handed than my proposal, as this
prevents TI_K3_AM65_CPSW_NUSS from being built-in when BRIDGE is a module,
even when switchdev support is completely disabled.
quoted
and to make CONFIG_TI_K3_AM65_CPSW_SWITCHDEV simply depend on BRIDGE.
This would not be needed then I think.
I've tested this change according to your suggestion:
@@ -110,7 +111,6 @@ config TI_K3_AM65_CPSW_NUSS config TI_K3_AM65_CPSW_SWITCHDEV bool "TI K3 AM654x/J721E CPSW Switch mode support" depends on TI_K3_AM65_CPSW_NUSS- depends on BRIDGE || BRIDGE=n depends on NET_SWITCHDEV help This enables switchdev support for TI K3 CPSWxG Ethernet
and it does appear to work, so when you resubmit please feel free to add:
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Thanks and sorry for breaking the dependency chain yet again! It takes
time to learn this stuff.