Re: [PATCH v2 02/10] KVM: arm/arm64: vgic-its: Always allow clearing GITS_CREADR/CWRITER

From: Auger Eric
Date: Sat Oct 14 2017 - 04:53:20 EST


Hi Christoffer,

On 13/10/2017 19:54, Christoffer Dall wrote:
> On Fri, Oct 13, 2017 at 01:54:34PM +0200, Auger Eric wrote:
>> Hi Christoffer,
>> On 13/10/2017 13:44, Christoffer Dall wrote:
>>> On Fri, Oct 06, 2017 at 05:29:02PM +0200, Auger Eric wrote:
>>>> Hi,
>>>>
>>>> On 06/10/2017 16:37, Andre Przywara wrote:
>>>>> Hi,
>>>>>
>>>>> On 27/09/17 14:28, Eric Auger wrote:
>>>>>> If the GITS_CBASER Size field is 0, which can correspond to a
>>>>>> reset value, the userspace fails to set the GITS_CREADR/CWRITER
>>>>>> offsets to 0. This failure is not justified.
>>>>>>
>>>>>> Let's allow this setting which can also correspond to a reset value.
>>>>>
>>>>> But the GITS_CBASER size field has a +1 encoding, so can never be 0, if
>>>>> I understand the manual (and ITS_CMD_BUFFER_SIZE) correctly.
>>>> Oh OK, you're. I missed that. I would have sworn I've seen that failure
>>>> but I must have misinterpreted it.
>>>>
>>>
>>> What is CWRITER is written to a brand new ITS before the CBASER is
>>> written?
>>>
>>> I don't see us initializing the cbaser field anywhere?
>> in vgic_its_create() its is allocated with kzalloc so its->cbaser is
>> initialized at this moment.
>>
>
> Right, so it can be 0, and we still need your patch, contrary to Andre's
> comment. Am I missing something?

No Andre is right, cbaser = 0 but ITS_CMD_BUFFER_SIZE(its->cbaser)
returns 4kB.

#define ITS_CMD_BUFFER_SIZE(baser) ((((baser) & 0xff) + 1) << 12)

So the check doesn't fail.

Thanks

Eric


>
> Thanks,
> -Christoffer
>