Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 8 Jun 2023 11:47:56 +0200
From:   Stefano Garzarella <sgarzare@redhat.com>
To:     Jason Wang <jasowang@redhat.com>
Cc:     "Michael S. Tsirkin" <mst@redhat.com>,
        Shannon Nelson <shannon.nelson@amd.com>,
        virtualization@lists.linux-foundation.org, netdev@vger.kernel.org,
        Eugenio =?utf-8?B?UMOpcmV6?= <eperezma@redhat.com>,
        Tiwei Bie <tiwei.bie@intel.com>, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] vhost-vdpa: filter VIRTIO_F_RING_PACKED feature
Message-ID: <t6ci7ek54zwss2w3kxaduirfi7vp5df5ydjxjlnr5fhv4ji3c5@aw26xy66pjc7>
References: <20230606085643-mutt-send-email-mst@kernel.org>
 <CAGxU2F7fkgL-HpZdj=5ZEGNWcESCHQpgRAYQA3W2sPZaoEpNyQ@mail.gmail.com>
 <20230607054246-mutt-send-email-mst@kernel.org>
 <CACGkMEuUapKvUYiJiLwtsN+x941jafDKS9tuSkiNrvkrrSmQkg@mail.gmail.com>
 <20230608020111-mutt-send-email-mst@kernel.org>
 <CACGkMEt4=3BRVNX38AD+mJU8v3bmqO-CdNj5NkFP-SSvsuy2Hg@mail.gmail.com>
 <5giudxjp6siucr4l3i4tggrh2dpqiqhhihmdd34w3mq2pm5dlo@mrqpbwckpxai>
 <CACGkMEtqn1dbrQZn3i-W_7sVikY4sQjwLRC5xAhMnyqkc3jwOw@mail.gmail.com>
 <lw3nmkdszqo6jjtneyp4kjlmutooozz7xj2fqyxgh4v2ralptc@vkimgnbfafvi>
 <CACGkMEt1yRV9qOLBqtQQmJA_UoRLCpznT=Gvd5D51Uaz2jakHA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <CACGkMEt1yRV9qOLBqtQQmJA_UoRLCpznT=Gvd5D51Uaz2jakHA@mail.gmail.com>
Precedence: bulk

On Thu, Jun 08, 2023 at 05:29:58PM +0800, Jason Wang wrote:
>On Thu, Jun 8, 2023 at 5:21 PM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> On Thu, Jun 08, 2023 at 05:00:00PM +0800, Jason Wang wrote:
>> >On Thu, Jun 8, 2023 at 4:00 PM Stefano Garzarella <sgarzare@redhat.com> wrote:
>> >>
>> >> On Thu, Jun 08, 2023 at 03:46:00PM +0800, Jason Wang wrote:
>> >>
>> >> [...]
>> >>
>> >> >> > > > > I have a question though, what if down the road there
>> >> >> > > > > is a new feature that needs more changes? It will be
>> >> >> > > > > broken too just like PACKED no?
>> >> >> > > > > Shouldn't vdpa have an allowlist of features it knows how
>> >> >> > > > > to support?
>> >> >> > > >
>> >> >> > > > It looks like we had it, but we took it out (by the way, we were
>> >> >> > > > enabling packed even though we didn't support it):
>> >> >> > > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6234f80574d7569444d8718355fa2838e92b158b
>> >> >> > > >
>> >> >> > > > The only problem I see is that for each new feature we have to modify
>> >> >> > > > the kernel.
>> >> >> > > > Could we have new features that don't require handling by vhost-vdpa?
>> >> >> > > >
>> >> >> > > > Thanks,
>> >> >> > > > Stefano
>> >> >> > >
>> >> >> > > Jason what do you say to reverting this?
>> >> >> >
>> >> >> > I may miss something but I don't see any problem with vDPA core.
>> >> >> >
>> >> >> > It's the duty of the parents to advertise the features it has. For example,
>> >> >> >
>> >> >> > 1) If some kernel version that is packed is not supported via
>> >> >> > set_vq_state, parents should not advertise PACKED features in this
>> >> >> > case.
>> >> >> > 2) If the kernel has support packed set_vq_state(), but it's emulated
>> >> >> > cvq doesn't support, parents should not advertise PACKED as well
>> >> >> >
>> >> >> > If a parent violates the above 2, it looks like a bug of the parents.
>> >> >> >
>> >> >> > Thanks
>> >> >>
>> >> >> Yes but what about vhost_vdpa? Talking about that not the core.
>> >> >
>> >> >Not sure it's a good idea to workaround parent bugs via vhost-vDPA.
>> >>
>> >> Sorry, I'm getting lost...
>> >> We were talking about the fact that vhost-vdpa doesn't handle
>> >> SET_VRING_BASE/GET_VRING_BASE ioctls well for packed virtqueue before
>> >> that series [1], no?
>> >>
>> >> The parents seem okay, but maybe I missed a few things.
>> >>
>> >> [1] https://lore.kernel.org/virtualization/20230424225031.18947-1-shannon.nelson@amd.com/
>> >
>> >Yes, more below.
>> >
>> >>
>> >> >
>> >> >> Should that not have a whitelist of features
>> >> >> since it interprets ioctls differently depending on this?
>> >> >
>> >> >If there's a bug, it might only matter the following setup:
>> >> >
>> >> >SET_VRING_BASE/GET_VRING_BASE + VDUSE.
>> >> >
>> >> >This seems to be broken since VDUSE was introduced. If we really want
>> >> >to backport something, it could be a fix to filter out PACKED in
>> >> >VDUSE?
>> >>
>> >> mmm it doesn't seem to be a problem in VDUSE, but in vhost-vdpa.
>> >> I think VDUSE works fine with packed virtqueue using virtio-vdpa
>> >> (I haven't tried), so why should we filter PACKED in VDUSE?
>> >
>> >I don't think we need any filtering since:
>> >
>> >PACKED features has been advertised to userspace via uAPI since
>> >6234f80574d7569444d8718355fa2838e92b158b. Once we relax in uAPI, it
>> >would be very hard to restrict it again. For the userspace that tries
>> >to negotiate PACKED:
>> >
>> >1) if it doesn't use SET_VRING_BASE/GET_VRING_BASE, everything works well
>> >2) if it uses SET_VRING_BASE/GET_VRING_BASE. it might fail or break silently
>> >
>> >If we backport the fixes to -stable, we may break the application at
>> >least in the case 1).
>>
>> Okay, I see now, thanks for the details!
>>
>> Maybe instead of "break silently", we can return an explicit error for
>> SET_VRING_BASE/GET_VRING_BASE in stable branches.
>> But if there are not many cases, we can leave it like that.
>
>A second thought, if we need to do something for stable. is it better
>if we just backport Shannon's series to stable?

I tried to look at it, but it looks like we have to backport quite a few
patches, I wrote a few things here:

https://lore.kernel.org/virtualization/32ejjuvhvcicv7wjuetkv34qtlpa657n4zlow4eq3fsi2twozk@iqnd2t5tw2an/

But if you think it's the best way, though, we can take a better look
at how many patches are to backport and whether it's risky or not.

>
>>
>> I was just concerned about how does the user space understand that it
>> can use SET_VRING_BASE/GET_VRING_BASE for PACKED virtqueues in a given
>> kernel or not.
>
>My understanding is that if packed is advertised, the application
>should assume SET/GET_VRING_BASE work.
>

Same here. So as an alternative to backporting a large set of patches,
I proposed to completely disable packed for stable branches where
vhost-vdpa IOCTLs doesn't support them very well.

Thanks,
Stefano