[dpdk-dev] [PATCH v2 00/14] Move PCI away from the EAL

[Gaëtan Rivet]

Hello Pablo,

On Mon, Oct 23, 2017 at 08:44:31AM +0000, De Lara Guarch, Pablo wrote:
> Hi Gaetan,
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Gaëtan Rivet
> > Sent: Wednesday, October 11, 2017 3:32 PM
> > To: Doherty, Declan <declan.doherty at intel.com>
> > Cc: dev at dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v2 00/14] Move PCI away from the EAL
> > 
> > On Wed, Oct 11, 2017 at 03:19:31PM +0100, Doherty, Declan wrote:
> > > On 18/09/2017 10:31 AM, Gaetan Rivet wrote:
> > > >Hi all,
> > > >
> > > >Here is a new version of the PCI bus move out of the EAL.
> > > >
> > > >The EAL PCI implementation is divided in two parts:
> > > >
> > > >   - librte_pci: library offering helpers to handle PCI objects
> > > >   - librte_bus_pci: bus driver for PCI devices
> > > >
> > > >This allows other libraries / tools to use PCI elements (location,
> > > >mappings, parsing operations, etc) without forcing a dependency on a
> > bus driver.
> > > >
> > > >The latter should not have to export helpers that others might need.
> > > >It is focused on defining the rte_pci_device, rte_pci_driver objects
> > > >and their handling.
> > > >
> > > >The cryptodev library has hard dependencies on rte_pci_devices (used
> > > >by generic probe function). Other similar libs (ether and eventdev)
> > > >avoided the issue by inlining such functions and expecting users to
> > > >include the relevant headers once the PCI bus has already been built.
> > > >
> > > >@Declan:
> > > >I proposed a solution that would avoid inlining those functions,
> > > >which does not feel right. Let me know what you think of it or if you
> > > >think of a better solution. I think it would be best to have
> > > >cryptodev completely independent from PCI / vdev as far as the lib in
> > > >concerned (the vdev bus will move as well).
> > > >
> > >
> > >
> > > Hey Gaetan, apologies for the delay in getting back to you on this, I
> > > had been looking at this but got sidelined onto other issues before
> > > usersapce and I'm only getting back to it now. I think that while your
> > > solution works it just highlights the dependency which probably
> > > shouldn't be there between the cryptodev library and PCI devices. I've
> > > had a look, and the functions in the cryptodev which you moved don't
> > > really provide that much useful functionality. I done some testing and
> > > completely removed them and just update the QAT PMD which is the only
> > > crypto PMD which was using them and it seems much cleaner to me. I'll
> > > push a patch for this change later today and it will allow you to drop
> > > the patch "cryptodev: move PCI specific helpers to drivers/crypto" from
> > this set.
> > >
> > > Regards
> > > Declan
> > >
> > >
> > >
> > 
> > Hi Declan,
> > 
> > All right, seems good from my PoV. I will rebase onto your patch once it is
> > sent.
> > 
> 
> Could you take a look at Declan's patchset and tell him what you think?
> http://dpdk.org/ml/archives/dev/2017-October/079827.html
> 
> It might be a bit late, but if we think it is a better approach for cryptodev,
> we could avoid creating a new PCI folder under drivers/crypto.
> 
> Thanks!
> Pablo
> 

Thanks for the heads up :)

I was working with a stand-in patch on my PCI move series, I will use
yours instead and see if I have any problem.

I am bound to send a new version soon, if all is fine I will add the
dependency on your series.

-- 
Gaëtan Rivet
6WIND