Re: [Xen-devel] [PATCH v2 14/25] arm/altp2m: Make get_page_from_gva ready for altp2m.

[Julien Grall <julien.grall@xxxxxxx>]

On 06/08/2016 18:58, Sergej Proskurin wrote:
> Hi Julien,
Hello Sergej,

> On 08/06/2016 03:45 PM, Julien Grall wrote:
> > On 06/08/2016 11:38, Sergej Proskurin wrote:
> > > Hi Julien,
> > Hello Serge,
> >
> > > On 08/04/2016 01:59 PM, Julien Grall wrote:
> > > > Hello Sergej,
> > > >
> > > > On 01/08/16 18:10, Sergej Proskurin wrote:
> > > > > The function get_page_from_gva uses ARM's hardware support to
> > > > > translate
> > > > > gva's to machine addresses. This function is used, among others, for
> > > > > memory regulation purposes, e.g, within the context of memory
> > > > > ballooning.
> > > > > To ensure correct behavior while altp2m is in use, we use the
> > > > > host's p2m
> > > > > table for the associated gva to ma translation. This is required at
> > > > > this
> > > > > point, as altp2m lazily copies pages from the host's p2m and even
> > > > > might
> > > > > be flushed because of changes to the host's p2m (as it is done within
> > > > > the context of memory ballooning).
> > > > I was expecting to see some change in
> > > > p2m_mem_access_check_and_get_page. Is there any reason to not fix it?
> > > I did not yet encounter any issues with
> > > p2m_mem_access_check_and_get_page. According to ARM ARM, ATS1C** (see
> > > gva_to_ipa_par) translates VA to IPA in non-secure privilege levels (as
> > > it is the the case here). Thus, the 2nd level translation represented by
> > > the (alt)p2m is not really considered at this point and hence make an
> > > extension obsolete.
> > >
> > > Or did you have anything else in mind?
> > The stage-1 page tables are living in the guest memory. So every time
> > you access an entry in the page table, you have to translate the IPA
> > (guest physical address) into a PA.
> >
> > However, the underlying memory of those page table may have
> > restriction permission or does not exist in the altp2m at all. So the
> > translation will fail.
> Please correct me if I am wrong but as far as I understand: the function
> p2m_mem_access_check_and_get_page is called only from get_page_from_gva.
> Also it is called only if the page translation within the function
> get_page_from_gva was not successful. Because of the fact that we use
> the hostp2m's 2nd stage translation table including the original memory
> access permissions (please note the short sequence, where we temporarily
> reset the VTTBR_EL2 of the hostp2m if altp2m is active), potential
> faults (which would lead to the call of the function
> p2m_mem_access_check_and_get_page) must have reasons beyond altp2m.
The translation in get_page_from_gva may fail if the permission in the 
hostp2m has been restricted by memaccess (for instance because 
default_access is not p2m_access_rwx).
So you will fallback to p2m_mem_access_check_and_get_page. This function 
is calling gva_to_ipa that will use the altp2m to do the translation.
Therefore I think you need to modify p2m_mem_access_check_and_get_page 
to cope with altp2m.
Regards,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel