Re: [dpdk-dev] [PATCH v15 09/12] build: disable drivers in Arm builds

[Thomas Monjalon <thomas@monjalon.net>]

25/01/2021 15:58, Honnappa Nagarahalli:
> > 22/01/2021 10:07, Jerin Jacob:
> > > On Fri, Jan 22, 2021 at 2:28 PM Thomas Monjalon <thomas@monjalon.net>
> > wrote:
> > > > 22/01/2021 09:39, Juraj Linkeš:
> > > > > > > > > disabled drivers, similarly how the command line option
> > > > > > > > > works and remove unneeded driver options ported from the
> > > > > > > > > old makefile system, since they don't work in the current Meson
> > build system.
> > > > > > > > > Add support for removing drivers for cross builds so that
> > > > > > > > > we can disable them in cross files.
> > > > > > > >
> > > > > > > > Why disabling them?
> > > > > > > > If a driver is not supported it should disable itseld in its meson file.
> > > > > > > >
> > > > > > >
> > > > > > > This is helpful when building for an SoC where we don't want
> > > > > > > to build to build a driver, but the build machine actually supports
> > the driver.
> > > > > > > I believe in this case the meson build system would find the
> > > > > > > dependencies and designate the driver to be build, but we
> > > > > > > don't want to build
> > > > > > the driver for that SoC.
> > > > > > >
> > > > > > > There may be other reasons as well - Honnappa or others from
> > > > > > > the Arm community may shed more light on this.
> > > > > > IMO, the assumption should be everything compiles on all the
> > > > > > platforms. Hence, the disables should be applied to the
> > > > > > platforms where the drivers do not compile.
> > > >
> > > > If a driver does not compile, it can disable itself.
> > > > No need for a configuration.
> > > >
> > > > > Would it be okay to leave the disabled as they're in this commit and
> > leave the updates to the plaform owners? Thomas, what do you think?
> > > >
> > > > I think this patch should not disable drivers but just add the infra to do it.
> > >
> > > IMO, If the SOC has "fixed" set of dpdk devices, probably better to
> > > have positive logic to enable only those in config file.
> > > I think, that will be portable and useful.
> > > IMO, We can have infrastructure code enable only specific drivers and
> > > config owners can later enable the required set.
> > 
> > Yes you're right, enabling makes more sense than disabling for SoCs.
> Every SoC also supports PCIe interfaces. This means, one could use them with a PCIe based NIC (we do use these interfaces internally at Arm, I am not sure from a deployment perspective).
> 
> If we use the enable logic, the list will be huge?

That's also right.
The last to talk will be right :)