Re: [Question about KVM on arm64] Consider putting VINVALL to deactivation

From: Marc Zyngier <maz@kernel.org>
To: Shaokun Zhang <zhangshaokun@hisilicon.com>
Cc: wangwudi <wangwudi@hisilicon.com>,
	Lizixian <lizixian@hisilicon.com>,
	jiayanlei@huawei.com, fanhenglong@huawei.com,
	kvmarm@lists.cs.columbia.edu,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [Question about KVM on arm64] Consider putting VINVALL to deactivation
Date: Wed, 15 Jan 2020 13:50:58 +0000	[thread overview]
Message-ID: <37126a877e1160ed50ee6d95a03d1574@kernel.org> (raw)
In-Reply-To: <bf470803-8e1c-d1c3-d5ac-731536196543@hisilicon.com>

Hi Shaokun,

On 2020-01-14 14:20, Shaokun Zhang wrote:
> Hi Marc,
> 
> On activation, VMAPP command is followed by a VINVALL, which could be
> quite expensive for the start-up of virtual machine. If a vpeid is 
> allowed
> successfully, it is not used in system.

How expensive? This is exactly similar to what happens on a physical 
machine
where we perform an INVALL on MAPC. And yet you don't complain about 
that.

Please provide numbers.

> We may consider put VINVALL to deactivation to ensure all cache of 
> certain
> vpeid is invalid, to simplify activation. We consider start-up may be 
> more
> common and more time-consuming-sensitive than shutdown process.

In my world, they cost the same thing, and happen just as often. Also, I 
want
guarantees that on VMAPP, there is no stale information even if this is 
the
first time we're using this VPEid (who knows what happens over kexec, 
for
example).

> Do you think it's all right?

I don't, for the reasons stated above. You also provide no numbers 
showing
how bad the overhead is, so I'm left guessing.

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm