[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] q35 support in Xen


On Thu, 14 Sep 2017 16:39:32 +0800
Yi Sun <yi.y.sun@xxxxxxxxxxxxxxx> wrote:

> Hi, Alexey,
> Have you submitted the patches? If yes, could you please share the link?

Sorry, had a lot of work recently -- so far I've managed to submit only the
bugfix for mentioned xen-mapcache issue with emulated AHCI DMA I/O, which
was a main prerequisite for Q35 on Xen.

Right now I need to rebase Q35 support patches on recent changes -- there
were multiple commits to related parts of code in both Xen and QEMU while
Q35 patches were initially made for Xen 4.8.0 (which I'm still using at the
moment). I'll try rebasing Q35 patches on this weekend, hopefully there
will be no big differences with 4.8.

> > > I saw Anthony's patch, but your extension patch seems still in
> > > development. Do you have plan to upstream it? I'm also interested in
> > > this basically I want full PCI-e passthru capability (Current Xen does
> > > support passthru a PCI-e device but guest can't see configuration
> > > offset 256-4095 for example). I'm glad to collaborate on this.  
> > 
> > Yes, I have plans to send patches for Q35 to the list. I've never
> > contributed to Xen/QEMU so far but I guess it's worth to try. It might
> > be a good idea to send them in batches -- split to separate parts for
> > libacpi, hvmloader and QEMU. There is also a number of minor
> > prerequisites which are required for Q35 support, ex. separating Xen
> > Platform device support from a selected machine (as it implemented
> > currently). It should be an independent option, not to be bound to a
> > pc/xenfv/etc machine. 
> > 
> > Right now many features require the emulation of something newer than a
> > i440 system, ex. MMCONFIG support will benefit from Q35 (or some other
> > PCIe-specific feature).
> > 
> > There still a lot of work towards a complete Q35 support in Xen of
> > course, but until we have a working minimum to move from there probably
> > will be no progress. So upstreaming a possibility to turn on the Q35
> > emulation and actually run a guest on a Q35 system with some PCIe
> > device passed through might be a good start (if there will be no
> > objections from maintainers).
> > 
> > Fixing (well, testing actually) the xen-mapcache DMA bug or validating
> > Stefano's patch for it is the first goal. The bug naturally affects Q35
> > but in theory might be reproduced using a pc/xenfv machine (much harder
> > though), so it's a good candidate to start with.

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.