Re: [RFC 0/6] Regressions for "imply" behavior change

[Arnd Bergmann <arnd@arndb.de>]

On Tue, Apr 14, 2020 at 3:29 PM Jason Gunthorpe <jgg@ziepe.ca> wrote:
> On Fri, Apr 10, 2020 at 07:04:27PM +0000, Saeed Mahameed wrote:
> > On Fri, 2020-04-10 at 14:13 -0300, Jason Gunthorpe wrote:
> > > On Fri, Apr 10, 2020 at 02:40:42AM +0000, Saeed Mahameed wrote:
> > >
> > > > This assumes that the module using FOO has its own flag
> > > > representing
> > > > FOO which is not always the case.
> > > >
> > > > for example in mlx5 we use VXLAN config flag directly to compile
> > > > VXLAN related files:
> > > >
> > > > mlx5/core/Makefile:
> > > >
> > > > obj-$(CONFIG_MLX5_CORE) += mlx5_core.o
> > > >
> > > > mlx5_core-y := mlx5_core.o
> > > > mlx5_core-$(VXLAN) += mlx5_vxlan.o
> > > >
> > > > and in mlx5_main.o we do:
> > >
> > > Does this work if VXLAN = m ?
> >
> > Yes, if VXLAN IS_REACHABLE to MLX5, mlx5_vxlan.o will be
> > compiled/linked.
>
> So mlx5_core-m does the right thing somehow?

What happens with CONFIG_VXLAN=m is that the above turns into

mlx5_core-y := mlx5_core.o
mlx5_core-m += mlx5_vxlan.o

which in turn leads to mlx5_core.ko *not* containing mlx5_vxlan.o,
and in turn causing that link error against
mlx5_vxlan_create/mlx5_vxlan_destroy, unless the IS_ENABLED()
is changed to IS_REACHABLE().

> > > > if (IS_ENABLED(VXLAN))
> > > >        mlx5_vxlan_init()
> > > >
> > > > after the change in imply semantics:
> > > > our options are:
> > > >
> > > > 1) use IS_REACHABLE(VXLAN) instead of IS_ENABLED(VXLAN)
> > > >
> > > > 2) have MLX5_VXLAN in mlx5 Kconfig and use IS_ENABLED(MLX5_VXLAN)
> > > > config MLX5_VXLAN
> > > >   depends on VXLAN || !VXLAN
> > > >   bool
> > >
> > > Does this trick work when vxlan is a bool not a tristate?
> > >
> > > Why not just put the VXLAN || !VXLAN directly on MLX5_CORE?
> > >
> >
> > so force MLX5_CORE to n if vxlan is not reachable ?
>
> IIRC that isn't what the expression does, if vxlan is 'n' then
>   n || !n == true

It forces MLX5_CORE to 'm' or 'n' but not 'y' if VXLAN=m,
but allows any option if VXLAN=y

> The other version of this is (m || VXLAN != m)

Right, that should be the same, but is less common.

I later found that I also needed this one for the same
kind of dependency on PTP:

--- a/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
+++ b/drivers/net/ethernet/mellanox/mlx5/core/Kconfig
@@ -7,7 +7,7 @@ config MLX5_CORE
        tristate "Mellanox 5th generation network adapters (ConnectX
series) core driver"
        depends on PCI
        select NET_DEVLINK
-       imply PTP_1588_CLOCK
+       depends on PTP_1588_CLOCK || !PTP_1588_CLOCK
        depends on VXLAN || !VXLAN
        imply MLXFW
        imply PCI_HYPERV_INTERFACE