[dpdk-dev] [PATCH] doc: announce removal of ABIs in PCI bus driver

[Xia, Chenbo]

Hi Thomas,

> -----Original Message-----
> From: Thomas Monjalon <thomas at monjalon.net>
> Sent: Sunday, August 1, 2021 4:44 AM
> To: Xia, Chenbo <chenbo.xia at intel.com>
> Cc: dev at dpdk.org; Yigit, Ferruh <ferruh.yigit at intel.com>; dev at dpdk.org;
> mdr at ashroe.eu; david.marchand at redhat.com; Richardson, Bruce
> <bruce.richardson at intel.com>; andrew.rybchenko at oktetlabs.ru; Ananyev,
> Konstantin <konstantin.ananyev at intel.com>
> Subject: Re: [dpdk-dev] [PATCH] doc: announce removal of ABIs in PCI bus
> driver
> 
> 27/07/2021 10:44, Bruce Richardson:
> > On Mon, Jul 26, 2021 at 05:56:17AM +0000, Xia, Chenbo wrote:
> > > From: Yigit, Ferruh <ferruh.yigit at intel.com>
> > > > On 7/23/2021 8:39 AM, Xia, Chenbo wrote:
> > > > > From: dev <dev-bounces at dpdk.org> On Behalf Of Chenbo Xia
> > > > >> +* pci: To reduce unnecessary ABIs exposed by DPDK bus driver,
> > > > "rte_bus_pci.h"
> > > > >> +  will be made internal in 21.11 and macros/data
> structures/functions
> > > > defined
> > > > >> +  in the header will not be considered as ABI anymore. This change
> is
> > > > >> inspired
> > > > >> +  by the RFC
> > > > https://patchwork.dpdk.org/project/dpdk/list/?series=17176.
> > > > >
> > > > > I see there's some ABI improvement work on-going and I think it could
> be
> > > > part of
> > > > > the work. If it makes sense to you, I'd like some ACKs.
> > > > >
> > > >
> > > > Acked-by: Ferruh Yigit <ferruh.yigit at intel.com>
> > > >
> > > > I am for reducing the public ABI as much as possible. How big will the
> > > > change
> > > > be? Is the 'rte_bus_pci.h' used other than './drivers/bus/pci/'?
> > >
> > > I don't see big change here. And I am not sure if I understand your second
> > > question. The rte_bus_pci.h will still be used by drivers (maybe remove
> the
> > > rte prefix and change the file name).
> > >
> > The file itself will still be exported in some cases, where the end-user
> > has their own drivers which need to be compiled, so I'd recommend keeping
> > the rte_ prefix. However, I think making all bus APIs internal-only to DPDK
> > is a good idea.
> 
> I don't understand how it can exported _and_ internal.

I think we can use the meson option 'enable_driver_sdk'. The first use case is in
lib ethdev for exporting internal APIs for out-of-tree drivers. For pci bus, I
think the use case is similar: users who want to build out-of-tree drivers can
set the option true to export pci header but the structs/functions are marked
internal. Make sense to you?

Thanks,
Chenbo

> And about the rte_ prefix, it should be kept even if it used only
> in internal drivers because it prevent from namespace clash with other
> libraries included by the drivers.
> As a rule we should always have rte_ prefix for each symbol used outside
> of its own library.
> 
> That said I am OK with the direction of hiding PCI bus API.
> 
> Applied, thanks.
> 
>