DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 2DB2461462
Subject: Re: [RFC] KVM: arm/arm64: Introduce KVM_DEV_ARM_ITS_CTRL_RESET
To: Marc Zyngier <marc.zyngier@arm.com>
References: <1505379448-19583-1-git-send-email-eric.auger@redhat.com>
 <864ls5rz0h.fsf@arm.com>
Cc: eric.auger.pro@gmail.com, linux-kernel@vger.kernel.org,
        kvm@vger.kernel.org, cdall@linaro.org, peter.maydell@linaro.org,
        andre.przywara@arm.com, wanghaibin.wang@huawei.com,
        wu.wubin@huawei.com
From: Auger Eric <eric.auger@redhat.com>
Message-ID: <2bb2cec8-ff29-e5e7-59dc-727924a51898@redhat.com>
Date: Fri, 15 Sep 2017 14:26:41 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101
 Thunderbird/45.4.0
MIME-Version: 1.0
In-Reply-To: <864ls5rz0h.fsf@arm.com>
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1693
Lines: 47

Hi,

On 14/09/2017 19:06, Marc Zyngier wrote:
> On Thu, Sep 14 2017 at 10:57:28 am BST, Eric Auger <eric.auger@redhat.com> wrote:
>> At the moment, the in-kernel emulated ITS is not properly reset.
>> On guest restart/reset some registers keep their old values and
>> internal structures like device, ITE, collection lists are not emptied.
>>
>> This may lead to various bugs. Among them, we can have incorrect state
>> backup or failure when saving the ITS state at early guest boot stage.
>>
>> This patch introduces a new attribute, KVM_DEV_ARM_ITS_CTRL_RESET in
>> the KVM_DEV_ARM_VGIC_GRP_CTRL group.
>>
>> Upon this action, we can invalidate the various memory structures
>> pointed by GITS_BASERn and GITS_CBASER, free the ITS internal caches
>> and reset the relevant registers.
>>
>> Signed-off-by: Eric Auger <eric.auger@redhat.com>
>>
>> ---
>>
>> An alternative would consist in having the userspace writing
>> individual registers with default values: GITS_BASERn, GITS_CBASER
>> and GITS_CTLR. On kernel side we would reset related lists when
>> detecting the valid bit is set to false.
> 
> I'm not sure this is necessarily a "either/or" situation. It looks to me
> that we're not completely doing the right thing when writing to the
> GITS_BASER registers, and that writing a new value (with the valid bit
> set or not) should have an action of some sort on the fate of the
> existing mappings.

I agree. I think whenever the GITS_BASERn or GITS_CBASER validity bit is
reset, we should empty the internal lists and assure the code does not
attempt to read the data structures in caches/RAM anymore.

I will follow up with some patches.

Thanks

Eric
> 
> Thoughts?
> 
> 	M.
>