Re: [PATCH V5 00/22] IOREQ feature (+ virtio-mmio) on Arm

From: Julien Grall <julien@xen.org>
To: Andrew Cooper <andrew.cooper3@citrix.com>,
	Stefano Stabellini <sstabellini@kernel.org>
Cc: "Oleksandr Tyshchenko" <olekstysh@gmail.com>,
	xen-devel@lists.xenproject.org,
	"Ian Jackson" <iwj@xenproject.org>,
	"Oleksandr Tyshchenko" <oleksandr_tyshchenko@epam.com>,
	"Paul Durrant" <paul@xen.org>, "Jan Beulich" <jbeulich@suse.com>,
	"Roger Pau Monné" <roger.pau@citrix.com>, "Wei Liu" <wl@xen.org>,
	"Julien Grall" <julien.grall@arm.com>,
	"George Dunlap" <george.dunlap@citrix.com>,
	"Jun Nakajima" <jun.nakajima@intel.com>,
	"Kevin Tian" <kevin.tian@intel.com>, "Tim Deegan" <tim@xen.org>,
	"Daniel De Graaf" <dgdegra@tycho.nsa.gov>,
	"Volodymyr Babchuk" <Volodymyr_Babchuk@epam.com>,
	"Bertrand Marquis" <bertrand.marquis@arm.com>,
	"Wei Chen" <Wei.Chen@arm.com>, "Kaly Xin" <Kaly.Xin@arm.com>,
	"Artem Mygaiev" <joculator@gmail.com>,
	"Alex Bennée" <alex.bennee@linaro.org>
Subject: Re: [PATCH V5 00/22] IOREQ feature (+ virtio-mmio) on Arm
Date: Thu, 28 Jan 2021 18:21:05 +0000	[thread overview]
Message-ID: <32349c24-dc9b-8c0a-d98e-3fc39eebb9ec@xen.org> (raw)
In-Reply-To: <9d0e2e0a-4409-ee37-636b-a0403e3571fc@xen.org>

On 28/01/2021 18:16, Julien Grall wrote:
> Hi Andrew,
> 
> On 28/01/2021 18:10, Andrew Cooper wrote:
>> On 28/01/2021 17:44, Julien Grall wrote:
>>>
>>>
>>> On 28/01/2021 17:24, Stefano Stabellini wrote:
>>>> On Thu, 28 Jan 2021, Julien Grall wrote:
>>>>> On 25/01/2021 19:08, Oleksandr Tyshchenko wrote:
>>>>>    > Patch series [8] was rebased on recent "staging branch"
>>>>>> (5e31789 tools/ocaml/libs/xb: Do not crash after xenbus is
>>>>>> unmapped) and
>>>>>> tested on
>>>>>> Renesas Salvator-X board + H3 ES3.0 SoC (Arm64) with virtio-mmio disk
>>>>>> backend [9]
>>>>>> running in driver domain and unmodified Linux Guest running on
>>>>>> existing
>>>>>> virtio-blk driver (frontend). No issues were observed. Guest domain
>>>>>> 'reboot/destroy'
>>>>>> use-cases work properly. Patch series was only build-tested on x86.
>>>>>
>>>>> I have done basic testing with a Linux guest on x86 and I didn't
>>>>> spot any
>>>>> regression.
>>>>>
>>>>> Unfortunately, I don't have a Windows VM in hand, so I can't confirm
>>>>> if there
>>>>> is no regression there. Can anyone give a try?
>>>>>
>>>>> On a separate topic, Andrew expressed on IRC some concern to expose 
>>>>> the
>>>>> bufioreq interface to other arch than x86. I will let him explain
>>>>> his view
>>>>> here.
>>>>>
>>>>> Given that IOREQ will be a tech preview on Arm (this implies the
>>>>> interface is
>>>>> not stable) on Arm, I think we can defer the discussion past the
>>>>> freeze.
>>>>
>>>> Yes, I agree that we can defer the discussion.
>>>>
>>>>
>>>>> For now, I would suggest to check if a Device Model is trying to
>>>>> create an
>>>>> IOREQ server with bufioreq is enabled (see ioreq_server_create()).
>>>>> This would
>>>>> not alleviate Andrew's concern, however it would at allow us to
>>>>> judge whether
>>>>> the buffering concept is going to be used on Arm (I can see some use
>>>>> for the
>>>>> Virtio doorbell).
>>>>>
>>>>> What do others think?
>>>>
>>>> Difficult to say. We don't have any uses today but who knows in the
>>>> future.
>>>
>>> I have some ideas, but Andrew so far objects on using the existing
>>> bufioreq interface for good reasons. It is using guest_cmpxchg() which
>>> is really expensive.
>>
>> Worse.  Patch 5 needs to switch cmpxchg() to guest_cmpxchg() to avoid
>> reintroducing XSA-295, but doesn't.
> 
> The memory is only shared with the emulator (we don't allow the legacy 
> interface on Arm). So why do you think it is re-introducing XSA-295?

Just for clarification, the swithc to guest_cmpxchg() is done as part of 
patch #13.

Cheers,

-- 
Julien Grall