Re: [Xen-devel] [PATCH RFC v1 52/74] xen: mark xenstore/console pages as RAM and add them to dom_io

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 09.01.18 at 12:26, <roger.pau@xxxxxxxxxx> wrote:
> On Tue, Jan 09, 2018 at 04:03:25AM -0700, Jan Beulich wrote:
>> >>> On 09.01.18 at 10:25, <roger.pau@xxxxxxxxxx> wrote:
>> > On Mon, Jan 08, 2018 at 06:49:21AM -0700, Jan Beulich wrote:
>> >> >>> On 04.01.18 at 14:06, <wei.liu2@xxxxxxxxxx> wrote:
>> >> > +void __init hypervisor_init_memory(void)
>> >> > +{
>> >> > +    uint64_t pfn = 0;
>> >> > +    long rc;
>> >> > +
>> >> > +    if ( !xen_guest )
>> >> > +        return;
>> >> > +
>> >> > +#define SHARE_PARAM(p) ({                                              
>> >> >      \
>> >> > +    rc = xen_hypercall_hvm_get_param(p, &pfn);                         
>> >> >     \
>> >> > +    if ( rc )                                                          
>> >> >      \
>> >> > +        panic("Unable to get " #p);                                    
>> >> >      \
>> >> > +    share_xen_page_with_guest(mfn_to_page(pfn), dom_io, 
>> >> > XENSHARE_writable); \
>> >> 
>> >> Why dom_io rather than the client domain?
>> > 
>> > The client domain is not yet created at this point. This is exactly
>> > the same that Xen does for the low 1MiB for example.
>> 
>> The low 1Mb is being treated as MMIO, hence remains assigned
>> to dom_io.
>> 
>> >> The more that dom_io
>> >> pages can only be mapped by privileged guests (and hence I
>> >> assume you need another tweak somewhere this way).
>> > 
>> > I just use unshare_xen_page and share it again with the guest.
>> 
>> And there is no option of simply doing the sharing here later,
>> when the domain is already in existence?
> 
> I'm afraid that if I don't add the pages to dom_io at this point they
> would be added to the free memory pool, and thus might be used for
> anything. Maybe I'm missing something, but I didn't find any other way
> to deal with this given the short time.

The first thing you do is mark these pages as E820 RAM. If that
wasn't done, I don't think they'd end up in the allocator, and
hence could be shared later. I guess there may nevertheless be
a reason to do that early E820 manipulation, but with the patch
having no description I cannot guess what that reason might be
(and hence I can't think of alternatives).

Anyway - in the interest of quick progress I'm fine with this being
left as is for now, as long as revisiting it is being put on a todo
list.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel