Andres Lagar-Cavilla
2012-Jan-09 21:41 UTC
[PATCH 0 of 2] x86/mm: Two hypervisor paging fixes
- Disallow for good paging_prep: it''s unsafe - Allow paging in of a page in paged-out state. This shortcuts the need to reference the page and trigger a populate event, thus saving a complete control stack round-trip. Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org> Acked-by: Tim Deegan <tim@xen.org> xen/arch/x86/mm/p2m.c | 15 +++++++++++---- xen/arch/x86/mm/p2m.c | 4 ++++ 2 files changed, 15 insertions(+), 4 deletions(-)
Andres Lagar-Cavilla
2012-Jan-09 21:41 UTC
[PATCH 1 of 2] x86/mm: Allow a page in p2m_ram_paged_out state to be loaded
xen/arch/x86/mm/p2m.c | 15 +++++++++++----
1 files changed, 11 insertions(+), 4 deletions(-)
This removes the need for a page to be accessed in order to be pageable
again. A pager can now page-in pages at will with no need to map them
in a separate thread.
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Tim Deegan <tim@xen.org>
diff -r 90f764bf02c3 -r f7c330d5b4b5 xen/arch/x86/mm/p2m.c
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -964,7 +964,7 @@ void p2m_mem_paging_populate(struct doma
int p2m_mem_paging_prep(struct domain *d, unsigned long gfn, uint64_t buffer)
{
struct page_info *page;
- p2m_type_t p2mt;
+ p2m_type_t p2mt, target_p2mt;
p2m_access_t a;
mfn_t mfn;
struct p2m_domain *p2m = p2m_get_hostp2m(d);
@@ -982,8 +982,8 @@ int p2m_mem_paging_prep(struct domain *d
mfn = p2m->get_entry(p2m, gfn, &p2mt, &a, p2m_query, NULL);
ret = -ENOENT;
- /* Allow only missing pages */
- if ( p2mt != p2m_ram_paging_in_start )
+ /* Allow missing pages */
+ if ( (p2mt != p2m_ram_paging_in_start) && (p2mt != p2m_ram_paged) )
goto out;
/* Allocate a page if the gfn does not have one yet */
@@ -1018,8 +1018,15 @@ int p2m_mem_paging_prep(struct domain *d
}
}
+ target_p2mt = (p2mt == p2m_ram_paging_in_start) ?
+ /* If we kicked the pager with a populate event, the pager will send
+ * a resume event back */
+ p2m_ram_paging_in :
+ /* If this was called asynchronously by the pager, then we can
+ * transition directly to the final guest-accessible type */
+ (paging_mode_log_dirty(d) ? p2m_ram_logdirty : p2m_ram_rw);
/* Fix p2m mapping */
- set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, p2m_ram_paging_in, a);
+ set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, target_p2mt, a);
atomic_dec(&d->paged_pages);
xen/arch/x86/mm/p2m.c | 4 ++++
1 files changed, 4 insertions(+), 0 deletions(-)
The only way to page-in a page is now the safe paging_load domctl.
(Unless the page was never paged out in the first place)
Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
Acked-by: Tim Deegan <tim@xen.org>
diff -r f7c330d5b4b5 -r d5e830891ee2 xen/arch/x86/mm/p2m.c
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -989,6 +989,10 @@ int p2m_mem_paging_prep(struct domain *d
/* Allocate a page if the gfn does not have one yet */
if ( !mfn_valid(mfn) )
{
+ /* If the user did not provide a buffer, we disallow */
+ ret = -EINVAL;
+ if ( unlikely(user_ptr == NULL) )
+ goto out;
/* Get a free page */
ret = -ENOMEM;
page = alloc_domheap_page(p2m->domain, 0);
Hongkaixing
2012-Jan-11 07:45 UTC
Re: [PATCH 1 of 2] x86/mm: Allow a page in p2m_ram_paged_out state to be loaded
I think it may have many unpredicted risks.
After p2mt is changed to p2m_ram_rw, Domain guest can access this page
unrestrictedly without being trapped in xen.
But at this time, the page is not prepared.
> -----Original Message-----
> From: xen-devel-bounces@lists.xensource.com
[mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Andres Lagar-Cavilla
> Sent: Tuesday, January 10, 2012 5:41 AM
> To: xen-devel@lists.xensource.com
> Cc: andres@gridcentric.ca; tim@xen.org; olaf@aepfle.de; adin@gridcentric.ca
> Subject: [Xen-devel] [PATCH 1 of 2] x86/mm: Allow a page in
p2m_ram_paged_out state to be loaded
>
> xen/arch/x86/mm/p2m.c | 15 +++++++++++----
> 1 files changed, 11 insertions(+), 4 deletions(-)
>
>
> This removes the need for a page to be accessed in order to be pageable
> again. A pager can now page-in pages at will with no need to map them
> in a separate thread.
>
> Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>
> Acked-by: Tim Deegan <tim@xen.org>
>
> diff -r 90f764bf02c3 -r f7c330d5b4b5 xen/arch/x86/mm/p2m.c
> --- a/xen/arch/x86/mm/p2m.c
> +++ b/xen/arch/x86/mm/p2m.c
> @@ -964,7 +964,7 @@ void p2m_mem_paging_populate(struct doma
> int p2m_mem_paging_prep(struct domain *d, unsigned long gfn, uint64_t
buffer)
> {
> struct page_info *page;
> - p2m_type_t p2mt;
> + p2m_type_t p2mt, target_p2mt;
> p2m_access_t a;
> mfn_t mfn;
> struct p2m_domain *p2m = p2m_get_hostp2m(d);
> @@ -982,8 +982,8 @@ int p2m_mem_paging_prep(struct domain *d
> mfn = p2m->get_entry(p2m, gfn, &p2mt, &a, p2m_query, NULL);
>
> ret = -ENOENT;
> - /* Allow only missing pages */
> - if ( p2mt != p2m_ram_paging_in_start )
> + /* Allow missing pages */
> + if ( (p2mt != p2m_ram_paging_in_start) && (p2mt !=
p2m_ram_paged) )
> goto out;
>
> /* Allocate a page if the gfn does not have one yet */
> @@ -1018,8 +1018,15 @@ int p2m_mem_paging_prep(struct domain *d
> }
> }
>
> + target_p2mt = (p2mt == p2m_ram_paging_in_start) ?
> + /* If we kicked the pager with a populate event, the pager will
send
> + * a resume event back */
> + p2m_ram_paging_in :
> + /* If this was called asynchronously by the pager, then we can
> + * transition directly to the final guest-accessible type */
> + (paging_mode_log_dirty(d) ? p2m_ram_logdirty : p2m_ram_rw);
> /* Fix p2m mapping */
> - set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, p2m_ram_paging_in, a);
> + set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, target_p2mt, a);
>
> atomic_dec(&d->paged_pages);
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
Andres Lagar-Cavilla
2012-Jan-11 14:57 UTC
Re: [PATCH 1 of 2] x86/mm: Allow a page in p2m_ram_paged_out state to be loaded
I think top-posting is frowned upon. Below...> I think it may have many unpredicted risks. > After p2mt is changed to p2m_ram_rw, Domain guest can access this page > unrestrictedly without being trapped in xen. > But at this time, the page is not prepared.Nope. The page has already been allocated and paged-in (copy_from_user out of user_ptr) by the time the p2mt is changed Andres> >> -----Original Message----- >> From: xen-devel-bounces@lists.xensource.com >> [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Andres >> Lagar-Cavilla >> Sent: Tuesday, January 10, 2012 5:41 AM >> To: xen-devel@lists.xensource.com >> Cc: andres@gridcentric.ca; tim@xen.org; olaf@aepfle.de; >> adin@gridcentric.ca >> Subject: [Xen-devel] [PATCH 1 of 2] x86/mm: Allow a page in >> p2m_ram_paged_out state to be loaded >> >> xen/arch/x86/mm/p2m.c | 15 +++++++++++---- >> 1 files changed, 11 insertions(+), 4 deletions(-) >> >> >> This removes the need for a page to be accessed in order to be pageable >> again. A pager can now page-in pages at will with no need to map them >> in a separate thread. >> >> Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org> >> Acked-by: Tim Deegan <tim@xen.org> >> >> diff -r 90f764bf02c3 -r f7c330d5b4b5 xen/arch/x86/mm/p2m.c >> --- a/xen/arch/x86/mm/p2m.c >> +++ b/xen/arch/x86/mm/p2m.c >> @@ -964,7 +964,7 @@ void p2m_mem_paging_populate(struct doma >> int p2m_mem_paging_prep(struct domain *d, unsigned long gfn, uint64_t >> buffer) >> { >> struct page_info *page; >> - p2m_type_t p2mt; >> + p2m_type_t p2mt, target_p2mt; >> p2m_access_t a; >> mfn_t mfn; >> struct p2m_domain *p2m = p2m_get_hostp2m(d); >> @@ -982,8 +982,8 @@ int p2m_mem_paging_prep(struct domain *d >> mfn = p2m->get_entry(p2m, gfn, &p2mt, &a, p2m_query, NULL); >> >> ret = -ENOENT; >> - /* Allow only missing pages */ >> - if ( p2mt != p2m_ram_paging_in_start ) >> + /* Allow missing pages */ >> + if ( (p2mt != p2m_ram_paging_in_start) && (p2mt != p2m_ram_paged) ) >> goto out; >> >> /* Allocate a page if the gfn does not have one yet */ >> @@ -1018,8 +1018,15 @@ int p2m_mem_paging_prep(struct domain *d >> } >> } >> >> + target_p2mt = (p2mt == p2m_ram_paging_in_start) ? >> + /* If we kicked the pager with a populate event, the pager will >> send >> + * a resume event back */ >> + p2m_ram_paging_in : >> + /* If this was called asynchronously by the pager, then we can >> + * transition directly to the final guest-accessible type */ >> + (paging_mode_log_dirty(d) ? p2m_ram_logdirty : p2m_ram_rw); >> /* Fix p2m mapping */ >> - set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, p2m_ram_paging_in, a); >> + set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, target_p2mt, a); >> >> atomic_dec(&d->paged_pages); >> >> >> _______________________________________________ >> Xen-devel mailing list >> Xen-devel@lists.xensource.com >> http://lists.xensource.com/xen-devel > >
Hongkaixing
2012-Jan-12 02:24 UTC
Re: [PATCH 1 of 2] x86/mm: Allow a page in p2m_ram_paged_out state to be loaded
> -----Original Message----- > From: Andres Lagar-Cavilla [mailto:andres@lagarcavilla.org] > Sent: Wednesday, January 11, 2012 10:57 PM > To: Hongkaixing > Cc: xen-devel@lists.xensource.com; andres@gridcentric.ca; tim@xen.org; olaf@aepfle.de; adin@gridcentric.ca; > yanqiangjun@huawei.com; bicky.shi@huawei.com; xiaowei.yang@huawei.com > Subject: RE: [Xen-devel] [PATCH 1 of 2] x86/mm: Allow a page in p2m_ram_paged_out state to be loaded > > I think top-posting is frowned upon. Below... > > I think it may have many unpredicted risks. > > After p2mt is changed to p2m_ram_rw, Domain guest can access this page > > unrestrictedly without being trapped in xen. > > But at this time, the page is not prepared. > > Nope. The page has already been allocated and paged-in (copy_from_user out > of user_ptr) by the time the p2mt is changedI have got it, first change p2mt to p2m_ram_paging_in, prepare a page, use copy_from_usr to copy, then change p2mt to ram_rw . It is a good idea.> Andres > > > >> -----Original Message----- > >> From: xen-devel-bounces@lists.xensource.com > >> [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Andres > >> Lagar-Cavilla > >> Sent: Tuesday, January 10, 2012 5:41 AM > >> To: xen-devel@lists.xensource.com > >> Cc: andres@gridcentric.ca; tim@xen.org; olaf@aepfle.de; > >> adin@gridcentric.ca > >> Subject: [Xen-devel] [PATCH 1 of 2] x86/mm: Allow a page in > >> p2m_ram_paged_out state to be loaded > >> > >> xen/arch/x86/mm/p2m.c | 15 +++++++++++---- > >> 1 files changed, 11 insertions(+), 4 deletions(-) > >> > >> > >> This removes the need for a page to be accessed in order to be pageable > >> again. A pager can now page-in pages at will with no need to map them > >> in a separate thread. > >> > >> Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org> > >> Acked-by: Tim Deegan <tim@xen.org> > >> > >> diff -r 90f764bf02c3 -r f7c330d5b4b5 xen/arch/x86/mm/p2m.c > >> --- a/xen/arch/x86/mm/p2m.c > >> +++ b/xen/arch/x86/mm/p2m.c > >> @@ -964,7 +964,7 @@ void p2m_mem_paging_populate(struct doma > >> int p2m_mem_paging_prep(struct domain *d, unsigned long gfn, uint64_t > >> buffer) > >> { > >> struct page_info *page; > >> - p2m_type_t p2mt; > >> + p2m_type_t p2mt, target_p2mt; > >> p2m_access_t a; > >> mfn_t mfn; > >> struct p2m_domain *p2m = p2m_get_hostp2m(d); > >> @@ -982,8 +982,8 @@ int p2m_mem_paging_prep(struct domain *d > >> mfn = p2m->get_entry(p2m, gfn, &p2mt, &a, p2m_query, NULL); > >> > >> ret = -ENOENT; > >> - /* Allow only missing pages */ > >> - if ( p2mt != p2m_ram_paging_in_start ) > >> + /* Allow missing pages */ > >> + if ( (p2mt != p2m_ram_paging_in_start) && (p2mt != p2m_ram_paged) ) > >> goto out; > >> > >> /* Allocate a page if the gfn does not have one yet */ > >> @@ -1018,8 +1018,15 @@ int p2m_mem_paging_prep(struct domain *d > >> } > >> } > >> > >> + target_p2mt = (p2mt == p2m_ram_paging_in_start) ? > >> + /* If we kicked the pager with a populate event, the pager will > >> send > >> + * a resume event back */ > >> + p2m_ram_paging_in : > >> + /* If this was called asynchronously by the pager, then we can > >> + * transition directly to the final guest-accessible type */ > >> + (paging_mode_log_dirty(d) ? p2m_ram_logdirty : p2m_ram_rw); > >> /* Fix p2m mapping */ > >> - set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, p2m_ram_paging_in, a); > >> + set_p2m_entry(p2m, gfn, mfn, PAGE_ORDER_4K, target_p2mt, a); > >> > >> atomic_dec(&d->paged_pages); > >> > >> > >> _______________________________________________ > >> Xen-devel mailing list > >> Xen-devel@lists.xensource.com > >> http://lists.xensource.com/xen-devel > > > >
At 16:41 -0500 on 09 Jan (1326127285), Andres Lagar-Cavilla wrote:> - Disallow for good paging_prep: it''s unsafe > - Allow paging in of a page in paged-out state. This shortcuts the > need to reference the page and trigger a populate event, thus saving > a complete control stack round-trip. > > Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org> > Acked-by: Tim Deegan <tim@xen.org>Applied, thanks. Tim.