Jan Beulich
2008-Nov-03 14:44 UTC
[Xen-devel] [PATCH] x86: allow grant table originating pte-s to have their guest available bits set
In order to allow 2.6.27''s get_user_pages_fast() to correctly detect granted pages (so they would be passed to get_user_pages()), it must be possible to tell the grant table code in Xen to set the three available bits in the pte-s it creates. While not applicable to the 2.6.18 tree, I''m nevertheless attaching our Linux side patch for reference. Signed-off-by: Jan Beulich <jbeulich@novell.com> Index: 2008-10-27/xen/arch/x86/mm.c ==================================================================--- 2008-10-27.orig/xen/arch/x86/mm.c 2008-10-31 16:13:56.000000000 +0100 +++ 2008-10-27/xen/arch/x86/mm.c 2008-10-31 17:41:12.000000000 +0100 @@ -3326,6 +3326,10 @@ int create_grant_host_mapping(uint64_t a if ( !(flags & GNTMAP_readonly) ) l1e_add_flags(pte,_PAGE_RW); + l1e_add_flags(pte, + ((flags >> _GNTMAP_guest_avail0) * _PAGE_AVAIL0) + & _PAGE_AVAIL); + l1e_add_flags(pte, cacheattr_to_pte_flags(cache_flags >> 5)); if ( flags & GNTMAP_contains_pte ) Index: 2008-10-27/xen/common/kernel.c ==================================================================--- 2008-10-27.orig/xen/common/kernel.c 2008-10-27 11:25:15.000000000 +0100 +++ 2008-10-27/xen/common/kernel.c 2008-10-31 17:17:30.000000000 +0100 @@ -222,7 +222,8 @@ DO(xen_version)(int cmd, XEN_GUEST_HANDL #ifdef CONFIG_X86 if ( !is_hvm_vcpu(current) ) fi.submap |= (1U << XENFEAT_mmu_pt_update_preserve_ad) | - (1U << XENFEAT_highmem_assist); + (1U << XENFEAT_highmem_assist) | + (1U << XENFEAT_gnttab_map_avail_bits); #endif break; default: Index: 2008-10-27/xen/include/public/features.h ==================================================================--- 2008-10-27.orig/xen/include/public/features.h 2008-10-27 11:24:26.000000000 +0100 +++ 2008-10-27/xen/include/public/features.h 2008-10-31 17:03:12.000000000 +0100 @@ -62,6 +62,12 @@ /* x86: Does this Xen host support the MMU_{CLEAR,COPY}_PAGE hypercall? */ #define XENFEAT_highmem_assist 6 +/* + * If set, GNTTABOP_map_grant_ref honors flags to be placed into guest kernel + * available pte bits. + */ +#define XENFEAT_gnttab_map_avail_bits 7 + #define XENFEAT_NR_SUBMAPS 1 #endif /* __XEN_PUBLIC_FEATURES_H__ */ Index: 2008-10-27/xen/include/public/grant_table.h ==================================================================--- 2008-10-27.orig/xen/include/public/grant_table.h 2007-12-07 11:51:50.000000000 +0100 +++ 2008-10-27/xen/include/public/grant_table.h 2008-10-31 17:10:21.000000000 +0100 @@ -360,7 +360,7 @@ DEFINE_XEN_GUEST_HANDLE(gnttab_unmap_and /* - * Bitfield values for update_pin_status.flags. + * Bitfield values for gnttab_map_grant_ref.flags. */ /* Map the grant entry for access by I/O devices. */ #define _GNTMAP_device_map (0) @@ -388,6 +388,13 @@ DEFINE_XEN_GUEST_HANDLE(gnttab_unmap_and #define GNTMAP_contains_pte (1<<_GNTMAP_contains_pte) /* + * Bits to be placed in guest kernel available PTE bits (architecture + * dependent; only supported when XENFEAT_gnttab_map_avail_bits is set). + */ +#define _GNTMAP_guest_avail0 (16) +#define GNTMAP_guest_avail_mask ((uint32_t)~0 << _GNTMAP_guest_avail0) + +/* * Values for error status returns. All errors are -ve. */ #define GNTST_okay (0) /* Normal return. */ _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir Fraser
2008-Nov-04 11:46 UTC
Re: [Xen-devel] [PATCH] x86: allow grant table originating pte-s to have their guest available bits set
Applied, thanks. I assume backport at least to 3.3 would be desirable? -- Keir On 3/11/08 14:44, "Jan Beulich" <jbeulich@novell.com> wrote:> In order to allow 2.6.27''s get_user_pages_fast() to correctly detect > granted pages (so they would be passed to get_user_pages()), it must be > possible to tell the grant table code in Xen to set the three available > bits in the pte-s it creates. > > While not applicable to the 2.6.18 tree, I''m nevertheless attaching our > Linux side patch for reference. > > Signed-off-by: Jan Beulich <jbeulich@novell.com> > > Index: 2008-10-27/xen/arch/x86/mm.c > ==================================================================> --- 2008-10-27.orig/xen/arch/x86/mm.c 2008-10-31 16:13:56.000000000 +0100 > +++ 2008-10-27/xen/arch/x86/mm.c 2008-10-31 17:41:12.000000000 +0100 > @@ -3326,6 +3326,10 @@ int create_grant_host_mapping(uint64_t a > if ( !(flags & GNTMAP_readonly) ) > l1e_add_flags(pte,_PAGE_RW); > > + l1e_add_flags(pte, > + ((flags >> _GNTMAP_guest_avail0) * _PAGE_AVAIL0) > + & _PAGE_AVAIL); > + > l1e_add_flags(pte, cacheattr_to_pte_flags(cache_flags >> 5)); > > if ( flags & GNTMAP_contains_pte ) > Index: 2008-10-27/xen/common/kernel.c > ==================================================================> --- 2008-10-27.orig/xen/common/kernel.c 2008-10-27 11:25:15.000000000 +0100 > +++ 2008-10-27/xen/common/kernel.c 2008-10-31 17:17:30.000000000 +0100 > @@ -222,7 +222,8 @@ DO(xen_version)(int cmd, XEN_GUEST_HANDL > #ifdef CONFIG_X86 > if ( !is_hvm_vcpu(current) ) > fi.submap |= (1U << XENFEAT_mmu_pt_update_preserve_ad) | > - (1U << XENFEAT_highmem_assist); > + (1U << XENFEAT_highmem_assist) | > + (1U << XENFEAT_gnttab_map_avail_bits); > #endif > break; > default: > Index: 2008-10-27/xen/include/public/features.h > ==================================================================> --- 2008-10-27.orig/xen/include/public/features.h 2008-10-27 > 11:24:26.000000000 +0100 > +++ 2008-10-27/xen/include/public/features.h 2008-10-31 17:03:12.000000000 > +0100 > @@ -62,6 +62,12 @@ > /* x86: Does this Xen host support the MMU_{CLEAR,COPY}_PAGE hypercall? */ > #define XENFEAT_highmem_assist 6 > > +/* > + * If set, GNTTABOP_map_grant_ref honors flags to be placed into guest kernel > + * available pte bits. > + */ > +#define XENFEAT_gnttab_map_avail_bits 7 > + > #define XENFEAT_NR_SUBMAPS 1 > > #endif /* __XEN_PUBLIC_FEATURES_H__ */ > Index: 2008-10-27/xen/include/public/grant_table.h > ==================================================================> --- 2008-10-27.orig/xen/include/public/grant_table.h 2007-12-07 > 11:51:50.000000000 +0100 > +++ 2008-10-27/xen/include/public/grant_table.h 2008-10-31 17:10:21.000000000 > +0100 > @@ -360,7 +360,7 @@ DEFINE_XEN_GUEST_HANDLE(gnttab_unmap_and > > > /* > - * Bitfield values for update_pin_status.flags. > + * Bitfield values for gnttab_map_grant_ref.flags. > */ > /* Map the grant entry for access by I/O devices. */ > #define _GNTMAP_device_map (0) > @@ -388,6 +388,13 @@ DEFINE_XEN_GUEST_HANDLE(gnttab_unmap_and > #define GNTMAP_contains_pte (1<<_GNTMAP_contains_pte) > > /* > + * Bits to be placed in guest kernel available PTE bits (architecture > + * dependent; only supported when XENFEAT_gnttab_map_avail_bits is set). > + */ > +#define _GNTMAP_guest_avail0 (16) > +#define GNTMAP_guest_avail_mask ((uint32_t)~0 << _GNTMAP_guest_avail0) > + > +/* > * Values for error status returns. All errors are -ve. > */ > #define GNTST_okay (0) /* Normal return. > */ > > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jan Beulich
2008-Nov-04 11:58 UTC
Re: [Xen-devel] [PATCH] x86: allow grant table originating pte-sto have their guest available bits set
>>> Keir Fraser <keir.fraser@eu.citrix.com> 04.11.08 12:46 >>> >Applied, thanks. I assume backport at least to 3.3 would be desirable?We would of course very much appreciate if you would do that, but to be honest I didn''t really expect you to (as it''s not a bug fix in that sense, i.e. the kernel can get away without this, though of course not without some extra effort and overhead). Jan On 3/11/08 14:44, "Jan Beulich" <jbeulich@novell.com> wrote:> In order to allow 2.6.27''s get_user_pages_fast() to correctly detect > granted pages (so they would be passed to get_user_pages()), it must be > possible to tell the grant table code in Xen to set the three available > bits in the pte-s it creates. > > While not applicable to the 2.6.18 tree, I''m nevertheless attaching our > Linux side patch for reference. > > Signed-off-by: Jan Beulich <jbeulich@novell.com> > > Index: 2008-10-27/xen/arch/x86/mm.c > ==================================================================> --- 2008-10-27.orig/xen/arch/x86/mm.c 2008-10-31 16:13:56.000000000 +0100 > +++ 2008-10-27/xen/arch/x86/mm.c 2008-10-31 17:41:12.000000000 +0100 > @@ -3326,6 +3326,10 @@ int create_grant_host_mapping(uint64_t a > if ( !(flags & GNTMAP_readonly) ) > l1e_add_flags(pte,_PAGE_RW); > > + l1e_add_flags(pte, > + ((flags >> _GNTMAP_guest_avail0) * _PAGE_AVAIL0) > + & _PAGE_AVAIL); > + > l1e_add_flags(pte, cacheattr_to_pte_flags(cache_flags >> 5)); > > if ( flags & GNTMAP_contains_pte ) > Index: 2008-10-27/xen/common/kernel.c > ==================================================================> --- 2008-10-27.orig/xen/common/kernel.c 2008-10-27 11:25:15.000000000 +0100 > +++ 2008-10-27/xen/common/kernel.c 2008-10-31 17:17:30.000000000 +0100 > @@ -222,7 +222,8 @@ DO(xen_version)(int cmd, XEN_GUEST_HANDL > #ifdef CONFIG_X86 > if ( !is_hvm_vcpu(current) ) > fi.submap |= (1U << XENFEAT_mmu_pt_update_preserve_ad) | > - (1U << XENFEAT_highmem_assist); > + (1U << XENFEAT_highmem_assist) | > + (1U << XENFEAT_gnttab_map_avail_bits); > #endif > break; > default: > Index: 2008-10-27/xen/include/public/features.h > ==================================================================> --- 2008-10-27.orig/xen/include/public/features.h 2008-10-27 > 11:24:26.000000000 +0100 > +++ 2008-10-27/xen/include/public/features.h 2008-10-31 17:03:12.000000000 > +0100 > @@ -62,6 +62,12 @@ > /* x86: Does this Xen host support the MMU_{CLEAR,COPY}_PAGE hypercall? */ > #define XENFEAT_highmem_assist 6 > > +/* > + * If set, GNTTABOP_map_grant_ref honors flags to be placed into guest kernel > + * available pte bits. > + */ > +#define XENFEAT_gnttab_map_avail_bits 7 > + > #define XENFEAT_NR_SUBMAPS 1 > > #endif /* __XEN_PUBLIC_FEATURES_H__ */ > Index: 2008-10-27/xen/include/public/grant_table.h > ==================================================================> --- 2008-10-27.orig/xen/include/public/grant_table.h 2007-12-07 > 11:51:50.000000000 +0100 > +++ 2008-10-27/xen/include/public/grant_table.h 2008-10-31 17:10:21.000000000 > +0100 > @@ -360,7 +360,7 @@ DEFINE_XEN_GUEST_HANDLE(gnttab_unmap_and > > > /* > - * Bitfield values for update_pin_status.flags. > + * Bitfield values for gnttab_map_grant_ref.flags. > */ > /* Map the grant entry for access by I/O devices. */ > #define _GNTMAP_device_map (0) > @@ -388,6 +388,13 @@ DEFINE_XEN_GUEST_HANDLE(gnttab_unmap_and > #define GNTMAP_contains_pte (1<<_GNTMAP_contains_pte) > > /* > + * Bits to be placed in guest kernel available PTE bits (architecture > + * dependent; only supported when XENFEAT_gnttab_map_avail_bits is set). > + */ > +#define _GNTMAP_guest_avail0 (16) > +#define GNTMAP_guest_avail_mask ((uint32_t)~0 << _GNTMAP_guest_avail0) > + > +/* > * Values for error status returns. All errors are -ve. > */ > #define GNTST_okay (0) /* Normal return. > */ > > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir Fraser
2008-Nov-04 12:09 UTC
Re: [Xen-devel] [PATCH] x86: allow grant table originating pte-sto have their guest available bits set
On 4/11/08 11:58, "Jan Beulich" <jbeulich@novell.com> wrote:>>>> Keir Fraser <keir.fraser@eu.citrix.com> 04.11.08 12:46 >>> >> Applied, thanks. I assume backport at least to 3.3 would be desirable? > > We would of course very much appreciate if you would do that, but to be > honest I didn''t really expect you to (as it''s not a bug fix in that sense, > i.e. > the kernel can get away without this, though of course not without some > extra effort and overhead).I assumed it might be required for your 2.6.27 to work. Still, 3.3.1 is going to be a big release (in terms of churn) as all our x.y.1 have been relative to x.y.0. And the change is pretty small... -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel