Re: [PATCH v5 07/11] xen/domctl: Introduce XEN_DOMCTL_CDF_vpci flag

From: Stefano Stabellini <sstabellini@kernel.org>
To: "Roger Pau Monné" <roger.pau@citrix.com>
Cc: Bertrand Marquis <Bertrand.Marquis@arm.com>,
	 Michal Orzel <Michal.Orzel@arm.com>,
	Rahul Singh <Rahul.Singh@arm.com>,
	 "xen-devel@lists.xenproject.org"
	<xen-devel@lists.xenproject.org>,
	 Andre Przywara <Andre.Przywara@arm.com>,
	 Christian Lindig <christian.lindig@citrix.com>,
	 David Scott <dave@recoil.org>, Ian Jackson <iwj@xenproject.org>,
	 Wei Liu <wl@xen.org>, Andrew Cooper <andrew.cooper3@citrix.com>,
	 George Dunlap <george.dunlap@citrix.com>,
	Jan Beulich <jbeulich@suse.com>,  Julien Grall <julien@xen.org>,
	Stefano Stabellini <sstabellini@kernel.org>,
	 Volodymyr Babchuk <Volodymyr_Babchuk@epam.com>
Subject: Re: [PATCH v5 07/11] xen/domctl: Introduce XEN_DOMCTL_CDF_vpci flag
Date: Wed, 13 Oct 2021 13:53:33 -0700 (PDT)	[thread overview]
Message-ID: <alpine.DEB.2.21.2110131205410.9408@sstabellini-ThinkPad-T480s> (raw)
In-Reply-To: <YWa7eGr3D8Ax8VUX@MacBook-Air-de-Roger.local>

[-- Attachment #1: Type: text/plain, Size: 2646 bytes --]

On Wed, 13 Oct 2021, Roger Pau Monné wrote:
> > I think the second solution is the right one but it cannot be done so near from the
> > feature freeze.
> > 
> > The CDF flag as introduced right now is not creating any issue and will be used once
> > the emulation flag will be introduce. We will be able at this stage to set this properly
> > also on x86 (for dom0 pvh).
> > Moreover keeping it will allow to continue to merge the remaining part of the PCI
> > passthrough which are otherwise not possible to be done as they are dependent on this flag.
> > 
> > Can we agree on keep the DOMCTL_CDF_vpci flag and introduce the emulation
> > flag on Arm after 4.16 release ?
> 
> If vPCI for Arm on 4.16 is not going to be functional, why so much
> pressure in pushing those patches so fast? I understand the need to
> remove stuff from the queue, but I don't think it's worth the cost of
> introducing a broken interface deliberately on a release.
> 
> I think we need to at least settle on whether we want to keep
> CDF_vpci or use an arch specific signal mechanism in order to decide
> what to do regarding the release.

I wrote a longer separate email to provide more context about the
"pushing fast" comment.

I agree that we don't want to introduce a bad interface.

In regards to a way forward for 4.16, my suggestion is the following:

- revert this patch: do not change the interface in this series
    - do not change anything related to CDF_vpci for x86
    - for ARM, leave has_vpci(d) to { false } and do not set
      XEN_DOMCTL_CDF_vpci
    - we can discuss it in depth later on, no rush

- in patch #10, in libxl_arm.c:libxl__arch_domain_prepare_config
    - do not set XEN_DOMCTL_CDF_vpci
    - do not set b_info.arch_arm.vpci
    - do not define LIBXL_HAVE_BUILDINFO_ARM_VPCI in libxl.h
    - keep make_vpci_node and arch_arm.vpci

The other patches (#1, #8, #10), which don't change any interfaces, can
still make it for 4.16 if the review feedback is addressed on time, with
one open TODO item [1].

This way, we get all the essential infrastructure we are trying to
introduce without making any compromises on the external interfaces.
Still it is good to have patches #1 #8 #10 so that with a trival
oneliner patch on top of 4.16 we can enable PCI for ARM and do testing
in the community, in gitlab-ci, and OSSTest too. (We have been
discussing special OSSTest flights to valide PCI passthrough as we
complete development.)

If we think we need a snap decision on this topic, I am available for a
quick sync-up call or IRC meeting between 8AM and 10AM tomorrow (Oct
14).

[1] https://marc.info/?l=xen-devel&m=163412120531248