On 16.03.16 at 14:15, konrad.wilk@oracle.com wrote:
On Wed, Mar 16, 2016 at 12:22:32PM +0000, Ian Campbell wrote:
On Wed, 2016-03-16 at 20:13 +0800, Shannon Zhao wrote:
On 2016/3/16 19:20, Ian Campbell wrote:
(nb, my citrix.com email is no longer valid) On Wed, 2016-03-16 at 11:33 +0800, Shannon Zhao wrote:
Hi,
I noticed there are some efforts to add Xen PV block device support in Seabios in a GSoC project and there is a wiki page [1] for it. I found some patches [2] to add Xenstore R/W support for Seabios. But I didn't find the patches to add PV block device driver in Seabios.
If you know please tell me where I can find these patches. And I noticed that the patches [2] and this GSoC project work didn't get applied to Seabios eventually, is there any reason? Does it mean that there are some difficulties to support Xen PV block device in Seabios?
This work was never finished, IIRC (and it was a long time ago so this might be a faulty recollection) the main stumbling block was that there is no reasonable BIOS level event which could be used to tear down the PV interfaces in order to hand them over to the OS (unlike, say, EFI where there is ExitBootServices).
Ian, thanks for your reply! It looks like the problem is how and when to clear PV resources in seabios before handing over to guest. But I wonder why virtio works in seabios. Does seabios using virtio need to clear things like vrings? Or seabios doesn't clear the things and guest just covers the configuration with new values?
I think virtio covered this use case from day 1 by having the reset, but also by not having a xenstore ring etc.
So making this work would require something like a complete set of parallel PV infrastructure (devices, corresponding xenstore nodes, grant table) for the use of the BIOS firmware, or perhaps a (tricky to retrofit in a backwards compatible manner) PV facility for the guest to reset everything before starting to use them.
Like guest front-end driver checks if the backend state is XenbusStateInitWait, if not, tell the backend to reset to XenbusStateInitWait state?
Before it can do this the guest needs to recover the xenbus ring (which was used by SeaBIOS) into a usable state -- so you need to be thinking at least one layer further down before you can start to think about individual devices, and don't forget the grant table (which may have in use entries from the BIOS) and event channels (which the BIOS may have setup).
I'm afraid I don't have any concrete answer for what exactly needs to be done to make this work, but I do know that it is a (IMHO very) non- trivial amount of investigation, prototyping and coding.
But it does work with OVMF. That is it implements the full PV driver for block. So it surely is possible.
And grub2, iirc.
Jan
On Wed, 2016-03-16 at 07:33 -0600, Jan Beulich wrote:
On 16.03.16 at 14:15, konrad.wilk@oracle.com wrote:
On Wed, Mar 16, 2016 at 12:22:32PM +0000, Ian Campbell wrote:
On Wed, 2016-03-16 at 20:13 +0800, Shannon Zhao wrote:
On 2016/3/16 19:20, Ian Campbell wrote:
(nb, my citrix.com email is no longer valid) On Wed, 2016-03-16 at 11:33 +0800, Shannon Zhao wrote:
> > > Hi, > > I noticed there are some efforts to add Xen PV block > device support in > Seabios in a GSoC project and there is a wiki page [1] > for it. I found > some patches [2] to add Xenstore R/W support for Seabios. > But I didn't > find the patches to add PV block device driver in > Seabios. > > If you know please tell me where I can find these > patches. And I noticed > that the patches [2] and this GSoC project work didn't > get applied to > Seabios eventually, is there any reason? Does it mean > that there are > some difficulties to support Xen PV block device in > Seabios?
This work was never finished, IIRC (and it was a long time ago so this might be a faulty recollection) the main stumbling block was that there is no reasonable BIOS level event which could be used to tear down the PV interfaces in order to hand them over to the OS (unlike, say, EFI where there is ExitBootServices).
Ian, thanks for your reply! It looks like the problem is how and when to clear PV resources in seabios before handing over to guest. But I wonder why virtio works in seabios. Does seabios using virtio need to clear things like vrings? Or seabios doesn't clear the things and guest just covers the configuration with new values?
I think virtio covered this use case from day 1 by having the reset, but also by not having a xenstore ring etc.
So making this work would require something like a complete set of parallel PV infrastructure (devices, corresponding xenstore nodes, grant table) for the use of the BIOS firmware, or perhaps a (tricky to retrofit in a backwards compatible manner) PV facility for the guest to reset everything before starting to use them.
Like guest front-end driver checks if the backend state is XenbusStateInitWait, if not, tell the backend to reset to XenbusStateInitWait state?
Before it can do this the guest needs to recover the xenbus ring (which was used by SeaBIOS) into a usable state -- so you need to be thinking at least one layer further down before you can start to think about individual devices, and don't forget the grant table (which may have in use entries from the BIOS) and event channels (which the BIOS may have setup).
I'm afraid I don't have any concrete answer for what exactly needs to be done to make this work, but I do know that it is a (IMHO very) non- trivial amount of investigation, prototyping and coding.
But it does work with OVMF. That is it implements the full PV driver for block. So it surely is possible.
And grub2, iirc.
Both of those have a defined handover point at which they can preemptively reset things for the guest (ExitBootServices for UEFI and right before calling the kernel for grub).
A BIOS doesn't have that, an OS could make any number of INT calls at any time after the BIOS hands over control and there is no "I'm done now" INT which the OS could use to ask for a reset.
Ian.
-
Ian Campbell -
Jan Beulich