| Message ID | 1476468318-24422-7-git-send-email-boris.ostrovsky@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: Perhaps add in here: PVH is like PV in that there are no PCI devices - which HVM code would piggyback on to find the Xen PCI platform device and use its MMIO space to stash the grants in. For PVH we balloon out memory and stash the grants in there. (Which begs the next question - where and when do we balloon out the normal memory back in?) > Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> > --- > drivers/xen/grant-table.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > > diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c > index bb36b1e..d6786b8 100644 > --- a/drivers/xen/grant-table.c > +++ b/drivers/xen/grant-table.c > @@ -1146,13 +1146,13 @@ int gnttab_init(void) > > static int __gnttab_init(void) > { > + if (!xen_domain()) > + return -ENODEV; > + > /* Delay grant-table initialization in the PV on HVM case */ > - if (xen_hvm_domain()) > + if (xen_hvm_domain() && !xen_pvh_domain()) > return 0; > > - if (!xen_pv_domain()) > - return -ENODEV; > - > return gnttab_init(); > } > /* Starts after core_initcall so that xen_pvh_gnttab_setup can be called > -- > 1.8.3.1 > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xen.org > https://lists.xen.org/xen-devel
On 10/14/2016 03:19 PM, Konrad Rzeszutek Wilk wrote: > On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: > > Perhaps add in here: > > PVH is like PV in that there are no PCI devices - which HVM > code would piggyback on to find the Xen PCI platform device and > use its MMIO space to stash the grants in. > > For PVH we balloon out memory and stash the grants in there. > > (Which begs the next question - where and when do we balloon out the > normal memory back in?) Are you saying that we should get back memory that we gave to grant tables? -boris > >> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> >> --- >> drivers/xen/grant-table.c | 8 ++++---- >> 1 file changed, 4 insertions(+), 4 deletions(-) >> >> diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c >> index bb36b1e..d6786b8 100644 >> --- a/drivers/xen/grant-table.c >> +++ b/drivers/xen/grant-table.c >> @@ -1146,13 +1146,13 @@ int gnttab_init(void) >> >> static int __gnttab_init(void) >> { >> + if (!xen_domain()) >> + return -ENODEV; >> + >> /* Delay grant-table initialization in the PV on HVM case */ >> - if (xen_hvm_domain()) >> + if (xen_hvm_domain() && !xen_pvh_domain()) >> return 0; >> >> - if (!xen_pv_domain()) >> - return -ENODEV; >> - >> return gnttab_init(); >> } >> /* Starts after core_initcall so that xen_pvh_gnttab_setup can be called >> -- >> 1.8.3.1 >> >> >> _______________________________________________ >> Xen-devel mailing list >> Xen-devel@lists.xen.org >> https://lists.xen.org/xen-devel
On Fri, Oct 14, 2016 at 03:43:19PM -0400, Boris Ostrovsky wrote: > On 10/14/2016 03:19 PM, Konrad Rzeszutek Wilk wrote: > > On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: > > > > Perhaps add in here: > > > > PVH is like PV in that there are no PCI devices - which HVM > > code would piggyback on to find the Xen PCI platform device and > > use its MMIO space to stash the grants in. > > > > For PVH we balloon out memory and stash the grants in there. > > > > (Which begs the next question - where and when do we balloon out the > > normal memory back in?) > > Are you saying that we should get back memory that we gave to grant tables? Yes. In pure HVM that area is MMIO - which hvmloader has balloonned out. The hvmloader then balloons that number of pages back at the end of guest memory (after 4GB).
On 10/14/2016 03:51 PM, Konrad Rzeszutek Wilk wrote: > On Fri, Oct 14, 2016 at 03:43:19PM -0400, Boris Ostrovsky wrote: >> On 10/14/2016 03:19 PM, Konrad Rzeszutek Wilk wrote: >>> On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: >>> >>> Perhaps add in here: >>> >>> PVH is like PV in that there are no PCI devices - which HVM >>> code would piggyback on to find the Xen PCI platform device and >>> use its MMIO space to stash the grants in. >>> >>> For PVH we balloon out memory and stash the grants in there. >>> >>> (Which begs the next question - where and when do we balloon out the >>> normal memory back in?) >> Are you saying that we should get back memory that we gave to grant tables? > Yes. > > In pure HVM that area is MMIO - which hvmloader has balloonned out. > > The hvmloader then balloons that number of pages back at the end of > guest memory (after 4GB). We don't do this for PV though, do we? -boris
On 14/10/16 22:02, Boris Ostrovsky wrote: > On 10/14/2016 03:51 PM, Konrad Rzeszutek Wilk wrote: >> On Fri, Oct 14, 2016 at 03:43:19PM -0400, Boris Ostrovsky wrote: >>> On 10/14/2016 03:19 PM, Konrad Rzeszutek Wilk wrote: >>>> On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: >>>> >>>> Perhaps add in here: >>>> >>>> PVH is like PV in that there are no PCI devices - which HVM >>>> code would piggyback on to find the Xen PCI platform device and >>>> use its MMIO space to stash the grants in. >>>> >>>> For PVH we balloon out memory and stash the grants in there. >>>> >>>> (Which begs the next question - where and when do we balloon out the >>>> normal memory back in?) >>> Are you saying that we should get back memory that we gave to grant tables? >> Yes. >> >> In pure HVM that area is MMIO - which hvmloader has balloonned out. >> >> The hvmloader then balloons that number of pages back at the end of >> guest memory (after 4GB). > > We don't do this for PV though, do we? Uuh, kind of. We try to allocate granted pages from the ballooned area. See gnttab_alloc_pages(). So for PV(H) we don't need to balloon this memory back in as it was never shadowed by a grant. Juergen
On 10/18/2016 12:08 PM, Juergen Gross wrote: > On 14/10/16 22:02, Boris Ostrovsky wrote: >> On 10/14/2016 03:51 PM, Konrad Rzeszutek Wilk wrote: >>> On Fri, Oct 14, 2016 at 03:43:19PM -0400, Boris Ostrovsky wrote: >>>> On 10/14/2016 03:19 PM, Konrad Rzeszutek Wilk wrote: >>>>> On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: >>>>> >>>>> Perhaps add in here: >>>>> >>>>> PVH is like PV in that there are no PCI devices - which HVM >>>>> code would piggyback on to find the Xen PCI platform device and >>>>> use its MMIO space to stash the grants in. >>>>> >>>>> For PVH we balloon out memory and stash the grants in there. >>>>> >>>>> (Which begs the next question - where and when do we balloon out the >>>>> normal memory back in?) >>>> Are you saying that we should get back memory that we gave to grant tables? >>> Yes. >>> >>> In pure HVM that area is MMIO - which hvmloader has balloonned out. >>> >>> The hvmloader then balloons that number of pages back at the end of >>> guest memory (after 4GB). >> We don't do this for PV though, do we? > Uuh, kind of. We try to allocate granted pages from the ballooned area. > See gnttab_alloc_pages(). I meant that we don't give memory back for PV. > > So for PV(H) we don't need to balloon this memory back in as it was > never shadowed by a grant. Is it *never* or *may or may not be* shadowed? (I assume "shadowed" means "used for" here.) -boris
On 18/10/16 18:40, Boris Ostrovsky wrote: > On 10/18/2016 12:08 PM, Juergen Gross wrote: >> On 14/10/16 22:02, Boris Ostrovsky wrote: >>> On 10/14/2016 03:51 PM, Konrad Rzeszutek Wilk wrote: >>>> On Fri, Oct 14, 2016 at 03:43:19PM -0400, Boris Ostrovsky wrote: >>>>> On 10/14/2016 03:19 PM, Konrad Rzeszutek Wilk wrote: >>>>>> On Fri, Oct 14, 2016 at 02:05:16PM -0400, Boris Ostrovsky wrote: >>>>>> >>>>>> Perhaps add in here: >>>>>> >>>>>> PVH is like PV in that there are no PCI devices - which HVM >>>>>> code would piggyback on to find the Xen PCI platform device and >>>>>> use its MMIO space to stash the grants in. >>>>>> >>>>>> For PVH we balloon out memory and stash the grants in there. >>>>>> >>>>>> (Which begs the next question - where and when do we balloon out the >>>>>> normal memory back in?) >>>>> Are you saying that we should get back memory that we gave to grant tables? >>>> Yes. >>>> >>>> In pure HVM that area is MMIO - which hvmloader has balloonned out. >>>> >>>> The hvmloader then balloons that number of pages back at the end of >>>> guest memory (after 4GB). >>> We don't do this for PV though, do we? >> Uuh, kind of. We try to allocate granted pages from the ballooned area. >> See gnttab_alloc_pages(). > > > I meant that we don't give memory back for PV. That's right AFAIK. >> So for PV(H) we don't need to balloon this memory back in as it was >> never shadowed by a grant. > > > Is it *never* or *may or may not be* shadowed? (I assume "shadowed" > means "used for" here.) "shadowed" means a pte is being used for a granted page which was referencing a RAM page before. So the RAM page is unusable as long as the grant is active. A page is shadowed by a grant only if there is no ballooning space available, so ballooning that page out would serve no purpose as we would have no way to balloon it in at another address. Juergen
diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c index bb36b1e..d6786b8 100644 --- a/drivers/xen/grant-table.c +++ b/drivers/xen/grant-table.c @@ -1146,13 +1146,13 @@ int gnttab_init(void) static int __gnttab_init(void) { + if (!xen_domain()) + return -ENODEV; + /* Delay grant-table initialization in the PV on HVM case */ - if (xen_hvm_domain()) + if (xen_hvm_domain() && !xen_pvh_domain()) return 0; - if (!xen_pv_domain()) - return -ENODEV; - return gnttab_init(); } /* Starts after core_initcall so that xen_pvh_gnttab_setup can be called
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> --- drivers/xen/grant-table.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)