|
|
|
|
|
|
|
|
|
|
xen-devel
RE: [Xen-devel] Why gnttab_grant_foreign_access() can only grantaccess f
> I just read the source code. gnttab_grant_foreign_access() seems to
just
> set some data in an in-memory structure "shared". No hypercall is
needed
> here. Right?
> If so, whether calling this function in a batch mode does not make big
> difference to me. I thought a gnttab_grant_foreign_access() will cause
a
> hypercall and context switch, then
> whether we can do batch granting will be important to performance.
Yes, only one side of the interface requires hypercalls.
> Ian, you said "grant tables are not typically used to map machine
> contiguous regions". Does Xen provide any mechanism to map contiguous
> regions?
Not as yet as it hasn't been needed. Are you sure the pages you want to
map are machine (as opposed to just guest physical) contiguous.
> Another question: Grant table currently seems to only allow 1024
entries,
> which means a domain may grant remote access on at most 1024 pages.
> However, this may not be enough in some cases. For example, in my
case, I
> need to allow a backend VM to share many pages with frontend. I can
> certainly change the xen code to make the grant table larger. But does
Xen
> has a better way to handle this situation?
There should be a hypercall to grow a domain's grant table. It's been on
the todo list for a while. Want to knock something up?
Thanks,
Ian
> thanks,
> lily
>
>
> On 7/7/06, Ian Pratt <m+Ian.Pratt@xxxxxxxxxxxx> wrote:
>
> > Lily Huang wrote:
> >
> > > THis seems to be less efficient. Why cannot we allow this
> function
> > > to grant access for mulitiple pages just remap function?
> >
> > With what frequency are multiple page grants requested?
>
> You can always use multicalls for batches. In the current usage
> scenarios, grant tables are not typically used to map machine
> contiguous
> regions.
>
> Ian
>
>
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
|
|
|
|