Re: [Xen-devel] [PATCH] xen/arm: Implement domain_get_maximum_gpfn

[Tamas K Lengyel <tamas.lengyel@xxxxxxxxxxxx>]

On Mon, Sep 8, 2014 at 10:47 PM, Tamas K Lengyel <tamas.lengyel@xxxxxxxxxxxx> wrote:

On Mon, Sep 8, 2014 at 10:43 PM, Julien Grall <julien.grall@xxxxxxxxxx> wrote:

Hello Tamas,

On 03/09/14 02:00, Tamas K Lengyel wrote:

On Wed, Sep 3, 2014 at 10:44 AM, Ian Campbell <Ian.Campbell@xxxxxxxxxx
<mailto:Ian.Campbell@citrix.com>> wrote:

    On Mon, 2014-09-01 at 17:32 -0400, Julien Grall wrote:
     > Hi Ian,
     >
     > On 16/07/14 12:02, Ian Campbell wrote:
     > > I'd much prefer to just have the fix to xc_dom_gnttab_hvm_seed
    for ARM
     > > and continue to punt on this interface until it is actually
    needed by
     > > something unavoidable on the guest side (and simultaneously
    hope that
     > > day never comes...).
     >
     > This patch is a requirement to make Xen Memory access working on ARM.
     > Could you reconsider the possibility to apply this patch on Xen?

    Needs more rationale as to why it is required for Xen Memory (do you
    mean xenaccess?). I assume I'll find that in the relevant thread once I
    get to it?


It's used in a non-critical sanity check for performance reasons, as
seen here:
https://github.com/tklengyel/xen/blob/arm_memaccess3/xen/common/mem_access.c#L75.
Without the sanity check we might attempt to set mem_access permissions
on gpfn's that don't exist for the guest. It wouldn't break anything to
do that but if we know beforehand that the gpfn is outside the scope of
what the guest has we can skip the entire thing.

It might be better if you carry this patch on your series.

Regards,

--
Julien Grall

As I have been looking at this more and more I'm coming to the conclusion that actually replacing the check in mem_access would be more sensible. The reason being that the user calling this code normally does it based on information gathered via getdomaininfo, and max_pages reported back to the user is higher then this implementation of domain_get_maximum_gpfn reports (on ARM at least). I switched xen-access to use tot_pages instead of max_pages, but tot_pages have been less then domain_get_maximum_gpfn reports. I'm not entirely sure why there is such a discrepancy, but enforcing a value that the user will ultimately 'guess' does not make much sense.

Tamas

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