On Thu, Sep 26, 2019 at 09:10:17AM +0200, Roger Pau Monné wrote:
> On Thu, Sep 26, 2019 at 06:16:06AM +0200, Marek Marczykowski-Górecki wrote:
> > On Wed, Sep 25, 2019 at 03:26:17PM +0200, Roger Pau Monné wrote:
> > > On Wed, Sep 25, 2019 at 02:29:41PM +0200, Marek Marczykowski-Górecki wrote:
> > > > On Wed, Sep 25, 2019 at 11:41:50AM +0200, Roger Pau Monné wrote:
> > > > > It would still be nice to get the missing bits (interrupt enabling),
> > > > > or else this patch is kind of pointless, since it still doesn't allow
> > > > > stubdomains to work correctly with passed through devices.
> > > > 
> > > > Well, that part, as discussed, doesn't need to be in Xen. For example
> > > > the solution deployed in current Qubes stable version is based on
> > > > pciback for this purpose.
> > > 
> > > Ack. Do you think it would make sense to submit that part to Linux
> > > then?
> > 
> > How would an interface with toolstack (when to allow enabling MSI)
> > should look like? Right now I have it as extra attribute in sysfs of
> > pciback and libxl writes to it. Or rather should it be in xenstore?
> 
> I think xenstore would be more suitable for this. There are already
> some flags passed to pciback there: msitranslate, power_mgmt and
> permissive (see libxl_pci.c:63).

Hmm, I see permissive is also set in sysfs
(libxl_pci.c:pci_add_dm_done). And I think that is used by pciback,
based on inspection of its source code.
In fact, I don't see anything in pciback parsing opts-%d xenstore entry
at all. It looks like it's only used by the toolstack to reconstruct
libxl_device_pci struct from xenstore.

> > Or maybe pciback should somehow detect itself if it's talking to
> > stubdomain while the device is assigned to a HVM domain, or to a target
> > PV domain itself?
> 
> I think doing it in the toolstack and just passing an option to
> pciback is likely easier than adding more logic to pciback.

Agree.

-- 
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?