Re: [PATCH 1/2] xen/*/nospec: Provide common versions of evaluate_nospec/block_speculation

From: Jan Beulich <jbeulich@suse.com>
To: Julien Grall <julien@xen.org>
Cc: "Roger Pau Monné" <roger.pau@citrix.com>, "Wei Liu" <wl@xen.org>,
	"Stefano Stabellini" <sstabellini@kernel.org>,
	"Volodymyr Babchuk" <Volodymyr_Babchuk@epam.com>,
	"Bertrand Marquis" <bertrand.marquis@arm.com>,
	"Michal Orzel" <michal.orzel@amd.com>,
	"Oleksii Kurochko" <oleksii.kurochko@gmail.com>,
	"Shawn Anastasio" <sanastasio@raptorengineering.com>,
	Xen-devel <xen-devel@lists.xenproject.org>,
	"Andrew Cooper" <andrew.cooper3@citrix.com>
Subject: Re: [PATCH 1/2] xen/*/nospec: Provide common versions of evaluate_nospec/block_speculation
Date: Mon, 4 Mar 2024 17:55:22 +0100	[thread overview]
Message-ID: <141ed8a2-df4f-492c-a192-4ffa7f4c8384@suse.com> (raw)
In-Reply-To: <5c06c437-b62c-4bee-8694-1be597887718@xen.org>

On 04.03.2024 17:46, Julien Grall wrote:
> On 04/03/2024 16:41, Jan Beulich wrote:
>> On 04.03.2024 17:31, Julien Grall wrote:
>>> On 04/03/2024 16:10, Andrew Cooper wrote:
>>>> It is daft to require all architectures to provide empty implementations of
>>>> this functionality.
>>>
>>> Oleksii recenlty sent a similar patch [1]. This was pushed back because
>>> from naming, it sounds like the helpers ought to be non-empty on every
>>> architecture.
>>>
>>> It would be best if asm-generic provides a safe version of the helpers.
>>> So my preference is to not have this patch. This can of course change if
>>> I see an explanation why it is empty on Arm (I believe it should contain
>>> csdb) and other arch would want the same.
>>
>> Except that there's no new asm-generic/ header here (as opposed to how
>> Oleksii had it). Imo avoiding the need for empty stubs is okay this way,
>> when introducing an asm-generic/ header would not have been. Of course
>> if Arm wants to put something there rather sooner than later, then
>> perhaps the functions better wouldn't be removed from there, just to then
>> be put back pretty soon.
> 
> I am confused. I agree the patch is slightly different, but I thought 
> the fundamental problem was the block_speculation() implementation may 
> not be safe everywhere. And it was best to let each architecture decide 
> how they want to implement (vs Xen decide for us the default).
> 
> Reading the original thread, I thought you had agreed with that 
> statement. Did I misinterpret?
Yes and no. Whatever is put in asm-generic/ ought to be correct and safe
by default, imo. The same doesn't apply to fallbacks put in place in
headers in xen/: If an arch doesn't provide its own implementation, it
indicates that the default (fallback) is good enough. Still I can easily
see that other views are possible here ...

Jan