Re: [Xen-devel] [PATCH v2 09/11] vpci: register as an internal ioreq server

[Paul Durrant <Paul.Durrant@citrix.com>]

> -----Original Message-----
> From: Roger Pau Monne <roger.pau@citrix.com>
> Sent: 27 September 2019 09:46
> To: Paul Durrant <Paul.Durrant@citrix.com>
> Cc: xen-devel@lists.xenproject.org; Ian Jackson <Ian.Jackson@citrix.com>; Wei Liu <wl@xen.org>; Andrew
> Cooper <Andrew.Cooper3@citrix.com>; George Dunlap <George.Dunlap@citrix.com>; Jan Beulich
> <jbeulich@suse.com>; Julien Grall <julien.grall@arm.com>; Konrad Rzeszutek Wilk
> <konrad.wilk@oracle.com>; Stefano Stabellini <sstabellini@kernel.org>; Tim (Xen.org) <tim@xen.org>
> Subject: Re: [PATCH v2 09/11] vpci: register as an internal ioreq server
> 
> On Fri, Sep 27, 2019 at 10:29:21AM +0200, Paul Durrant wrote:
> > > -----Original Message-----
> > > From: Roger Pau Monne <roger.pau@citrix.com>
> > > Sent: 26 September 2019 16:07
> > > To: Paul Durrant <Paul.Durrant@citrix.com>
> > > Cc: xen-devel@lists.xenproject.org; Ian Jackson <Ian.Jackson@citrix.com>; Wei Liu <wl@xen.org>;
> Andrew
> > > Cooper <Andrew.Cooper3@citrix.com>; George Dunlap <George.Dunlap@citrix.com>; Jan Beulich
> > > <jbeulich@suse.com>; Julien Grall <julien.grall@arm.com>; Konrad Rzeszutek Wilk
> > > <konrad.wilk@oracle.com>; Stefano Stabellini <sstabellini@kernel.org>; Tim (Xen.org) <tim@xen.org>
> > > Subject: Re: [PATCH v2 09/11] vpci: register as an internal ioreq server
> > >
> > > On Tue, Sep 10, 2019 at 03:49:41PM +0200, Paul Durrant wrote:
> > > > > -----Original Message-----
> > > > > From: Roger Pau Monne <roger.pau@citrix.com>
> > > > > Sent: 03 September 2019 17:14
> > > > > To: xen-devel@lists.xenproject.org
> > > > > Cc: Roger Pau Monne <roger.pau@citrix.com>; Ian Jackson <Ian.Jackson@citrix.com>; Wei Liu
> > > > > <wl@xen.org>; Andrew Cooper <Andrew.Cooper3@citrix.com>; George Dunlap
> <George.Dunlap@citrix.com>;
> > > Jan
> > > > > Beulich <jbeulich@suse.com>; Julien Grall <julien.grall@arm.com>; Konrad Rzeszutek Wilk
> > > > > <konrad.wilk@oracle.com>; Stefano Stabellini <sstabellini@kernel.org>; Tim (Xen.org)
> > > <tim@xen.org>;
> > > > > Paul Durrant <Paul.Durrant@citrix.com>
> > > > > Subject: [PATCH v2 09/11] vpci: register as an internal ioreq server
> > > > > @@ -478,6 +480,67 @@ void vpci_write(pci_sbdf_t sbdf, unsigned int reg, unsigned int size,
> > > > >      spin_unlock(&pdev->vpci->lock);
> > > > >  }
> > > > >
> > > > > +#ifdef __XEN__
> > > > > +static int ioreq_handler(struct vcpu *v, ioreq_t *req, void *data)
> > > > > +{
> > > > > +    pci_sbdf_t sbdf;
> > > > > +
> > > > > +    if ( req->type == IOREQ_TYPE_INVALIDATE )
> > > > > +        /*
> > > > > +         * Ignore invalidate requests, those can be received even without
> > > > > +         * having any memory ranges registered, see send_invalidate_req.
> > > > > +         */
> > > > > +        return X86EMUL_OKAY;
> > > >
> > > > In general, I wonder whether internal servers will ever need to deal with invalidate? The code
> only
> > > exists to get QEMU to drop its map cache after a decrease_reservation so that the page refs get
> > > dropped.
> > >
> > > I think the best solution here is to rename hvm_broadcast_ioreq to
> > > hvm_broadcast_ioreq_external and switch it's callers. Both
> > > send_timeoffset_req and send_invalidate_req seem only relevant to
> > > external ioreq servers.
> >
> > send_timeoffset_req() is relic which ought to be replaced with another mechanism IMO...
> >
> > When an HVM guest writes its RTC, a new 'timeoffset' value (offset of RTC from host time) is
> calculated (also applied to the PV wallclock) and advertised via this ioreq. In XenServer, this is
> picked up by QEMU, forwarded via QMP to XAPI and then written into the VM meta-data (which than causes
> it to be written into xenstore too). All this is so that that guest's RTC can be set correctly when it
> is rebooted... There has to be a better way (e.g. extracting RTC via hvm context and saving it before
> cleaning up the domain).
> >
> > send_invalidate_req() is relevant for any emulator maintaining a cache of guest->host memory
> mappings which, I guess, could include internal emulators even if this is not the case at the moment.
> 
> Maybe, but I would expect an internal emulator to get a reference on
> the gfn if it does need to keep it in some kind of cache, or else I
> don't think code in the hypervisor should be keeping such references.

Oh indeed, but that's not the issue. The issue is when to drop those refs... If the guest does a decrease_reservation on a gfn cached by the emulator then the emulator needs to drop its ref to allow the page to be freed.

  Paul

> IMO I would start by not forwarding invalidate requests to internal
> emulators. We can always change this in the future if we come up with
> a use-cases that needs it.
> 
> Thanks, Roger.

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