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

Re: [Xen-devel] [PATCH RFC 00/20] Change parts of the shadow interface to be domain based

At 13:35 +0000 on 16 Feb (1424090147), Andrew Cooper wrote:
> On 16/02/15 12:29, Tim Deegan wrote:
> > Hi,
> >
> > Thanks for this; I think the code looks a lot saner with all the
> > vcpu<->domain fiddling removed. :)
> >
> > At 17:15 +0000 on 12 Feb (1423757759), Andrew Cooper wrote:
> >> The purpose of this series is to prevent toolstack entry points into the
> >> shadow code from passing d->vcpu[0] for actions which are inherenly domain
> >> wide.  It also fixes the fact that shadow heuristics were being applied to
> >> vcpu 0 for toolstack-initiated actions.
> >>
> >> This series is composed mostly of mechanical changes.  The only patches 
> >> which
> >> have a practical difference on shadow execution are patches 4 and 20
> > So having read the series, you can add my Reviewed-by: to all except 4
> > and 20.  20 looks probably OK but I want to give it a second pass.
> 20 is no practical change.

It is if it's correct. :)

> > 4 definitely needs some thought -- in particular v == current is not the
> > same as d == current->domain.  I wonder if there's some better test;
> > maybe also checking that current is in the right mode will help.
> The result is still the same in that you can derive a valid v when
> running in the context of the target d.

It's different if there's any path in the existing code where these
functions can be called from another vcpu of the same guest.  I don't
_think_ there is such a path, though.

> The change specifically causes codepaths running in the context of the
> toolstack domain to not perform shadow shootdown heuristics.
> (It was you who suggested this, although I believe that we were down the
> pub at the time ;) )

Very likely. :)  Let me have another look at it later in the week. 

> > Let me think about it and I'll review again later in the week.  I'll want
> > to have a look at the remaining users of the vcpu variant hash_foreach
> > as well.
> There are certainly more codepaths which could be turned to taking a
> domain.  I limited this series just to adjusting the toolstack
> entrypoints into the shadow code.




Xen-devel mailing list



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