Re: [Xen-devel] pvgrub2 is merged

[M A Young]

On Thu, 14 Nov 2013, Vladimir 'φ-coder/phcoder' Serbinenko wrote:

> On 14.11.2013 19:57, Vladimir 'φ-coder/phcoder' Serbinenko wrote:
> > On 14.11.2013 19:48, M A Young wrote:
> > > On Thu, 14 Nov 2013, Vladimir 'φ-coder/phcoder' Serbinenko wrote:
> > >
> > > > On 14.11.2013 18:03, M A Young wrote:
> > > > > On Thu, 14 Nov 2013, M A Young wrote:
> > > > >
> > > > > > On Wed, 13 Nov 2013, Vladimir 'φ-coder/phcoder' Serbinenko wrote:
> > > > > >
> > > > > > > On 13.11.2013 20:06, M A Young wrote:
> > > > > > > > It doesn't seem to understand sub-partitions. I can get it to work if
> > > > > > > > the boot files are in /dev/xvda but not in /dev/xvda1 .
> > > > > > > insmod part_msdos
> > > > > > > insmod part_gpt
> > > > > >
> > > > > > Right, if I add those to the embedded grub.cfg file I get to the
> > > > > > standard grub menu and the boot starts. However the boot doesn't get
> > > > > > very far - it loads the kernel and the initrd file and starts the
> > > > > > kernel but the kernel doesn't see the virtual disks so it doesn't get
> > > > > > very far.
> > > > >
> > > > > Using xenstore-ls from the dom0 on the guest when the boot stops the
> > > > > local/domain/2/device/vbd/51712 section looks like
> > > > >       backend = "/local/domain/0/backend/vbd/2/51712"
> > > > >       backend-id = "0"
> > > > >       state = "6\000"
> > > > >       virtual-device = "51712"
> > > > >       device-type = "disk"
> > > > >       ring-ref = "\000"
> > > > >       event-channel = "\000"
> > > > >       protocol = "x86_64-abi\000"
> > > > >
> > > > > As nothing else has null character endings I suspend that is wrong.
> > > > Good catch. Could you test following:
> > > > diff --git a/grub-core/kern/xen/init.c b/grub-core/kern/xen/init.c
> > > > index 3bfd99f..ab74543 100644
> > > > --- a/grub-core/kern/xen/init.c
> > > > +++ b/grub-core/kern/xen/init.c
> > > > @@ -256,11 +256,10 @@ grub_xenstore_write_file (const char *dir, const
> > > > void *buf, grub_size_t len)
> > > >
> > > >   grub_memset (&msg, 0, sizeof (msg));
> > > >   msg.type = XS_WRITE;
> > > > -  msg.len = dirlen + len + 1;
> > > > +  msg.len = dirlen + len;
> > > >   grub_xen_store_send (&msg, sizeof (msg));
> > > >   grub_xen_store_send (dir, dirlen);
> > > >   grub_xen_store_send (buf, len);
> > > > -  grub_xen_store_send ("", 1);
> > > >   grub_xen_store_recv (&msg, sizeof (msg));
> > > >   resp = grub_malloc (msg.len + 1);
> > > >   if (!resp)
> > >
> > > The section is tidied up, ie.
> > >       backend = "/local/domain/0/backend/vbd/4/51712"
> > >       backend-id = "0"
> > >       state = "6"
> > >       virtual-device = "51712"
> > >       device-type = "disk"
> > >       ring-ref = ""
> > >       event-channel = ""
> > >       protocol = "x86_64-abi"
> > >
> > > but unfortunately it doesn't help as the boot process sticks at the same
> > > point. I notice this section is in state 6 which apparently is "closed".
> > > I wonder if the kernel expecting something else.
> > Possible. I'd try this (on top of previous patch):
>
> Sorry, too tired. I meant:
> diff --git a/grub-core/disk/xen/xendisk.c b/grub-core/disk/xen/xendisk.c
> index c449848..9b71d3a 100644
> --- a/grub-core/disk/xen/xendisk.c
> +++ b/grub-core/disk/xen/xendisk.c
> @@ -449,5 +449,10 @@ grub_xendisk_fini (void)
>       grub_xen_free_shared_page (virtdisks[i].shared_page);
>
>       grub_xen_event_channel_op (EVTCHNOP_close, &close_op);
> +
> +      /* Prepare for handoff.  */
> +      grub_snprintf (fdir, sizeof (fdir), "%s/state",
> +                    virtdisks[i].frontend_dir);
> +      grub_xenstore_write_file (fdir, "0", 1);
>     }
> }

That doesn't work. However, according to the documentation state 0 is 
unknown, and the vif interface (while grub is running) is in state 1 
(initializing) so I thought I would try it, and if you replace "0" with 
"1" in the above patch then the kernel does boot.

        Michael Young