Re: [PATCH v02 1/7] arm: introduce remoteprocessor iommu module

[Julien Grall <julien.grall@linaro.org>]

On 07/22/2014 01:42 PM, Stefano Stabellini wrote:
> On Wed, 16 Jul 2014, Ian Campbell wrote:
>> On Fri, 2014-07-04 at 14:59 +0100, Stefano Stabellini wrote:
>>> On Thu, 26 Jun 2014, Andrii Tseglytskyi wrote:
>>>> This is a fisrst patch from patch series which was
>>>> developed to handle remote (external) processors
>>>> memory management units. Remote processors are
>>>> typically used for graphic rendering (GPUs) and
>>>> high quality video decoding (IPUs). They are typically
>>>> installed on such multimedia SoCs as OMAP4 / OMAP5.
>>>>
>>>> As soon as remoteprocessor MMU typically works with
>>>> pagetables filled by physical addresses, which are
>>>> allocated by domU kernel, it is almost impossible to
>>>> use them under Xen, intermediate physical addresses
>>>> allocated by kernel, need to be translated to machine
>>>> addresses.
>>>>
>>>> This patch introduces a simple framework to perform
>>>> pfn -> mfn translation for external MMUs.
>>>> It introduces basic data structures and algorithms
>>>> needed for translation.
>>>>
>>>> Typically, when MMU is configured, some it registers
>>>> are updated by new values. Introduced frameworks
>>>> uses traps as starting point of remoteproc MMUs
>>>> pagetables translation.
>>>>
>>>> Change-Id: Ia4d311a40289df46a003f5ae8706c150bee1885d
>>>> Signed-off-by: Andrii Tseglytskyi <andrii.tseglytskyi@globallogic.com>
>>>
>>> There is one problem with this patch: you need to find a way to "pin"
>>> the p2m entries for the pfns and mfns found in the pagetables translated
>>> by remoteproc_iommu.c.  Otherwise Xen might decide to change the pfn to
>>> mfn mappings for the domain, breaking the pagetables already translated
>>> by remoteproc_iommu.c.  pfn to mfn mappings are not guaranteed to be
>>> stable. At the moment Xen on ARM wouldn't change pfn to mfn mappings
>>> under the guest feet, but features like memory sharing or swapping,
>>> supported on x86, could cause it to happen.
>>
>> I've not fully grokked this patch but wouldn't it be easier and better
>> to have a hook to cause the remoteproc to throw away and rebuild its
>> mappings when this happens? It could be called from e.g. the TLB flush
>> associated with the p2m changing, I think.
>  
> I guess it depends on how often Xen or the Xen tools are going to change
> the p2m for a running domain. For example changing one page mapping a
> couple of times a second would be very expensive if we have to rebuild
> the remoteproc pagetables every time.

The kernel could also change the p2m mapping via the ballooning driver.

Regards,

-- 
Julien Grall