Refactor super page shattering
v3 cover Refactor super page shattering
map_pages_to_xen and modify_xen_mappings use almost exactly the same page shattering logic, and the code is mingled with other PTE manipulations so it is less obvious that the intention is page shattering. Factor out the functions to make them reusable and to make the intention more obvious.
Of course, there is not much difference between the shattering logic of each level, so we could further turn the per-level functions into a single macro, although this is not that simple since we have per-level functions and macros all over the place and there are slight differences between levels. Keep it per-level for now.
tree:
https://xenbits.xen.org/git-http/people/hx242/xen.git shatter-refactor
---
Changes in v3:
- style and indentation fixes.
Changes in v2:
- rebase.
- improve asm code.
- avoid stale values when taking the lock.
- move allocation of PTE tables inside the shatter function.
Hongyan Xia (2):
x86/mm: factor out the code for shattering an l3 PTE x86/mm: factor out the code for shattering an l2 PTE
xen/arch/x86/mm.c | 194 +++++++++++++++++++++++----------------------- 1 file changed, 98 insertions(+), 96 deletions(-)
Signature · 7 lines
2.17.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 1/2 x86/mm: factor out the code for shattering an l3 PTE
map_pages_to_xen and modify_xen_mappings are performing almost exactly the same operations when shattering an l3 PTE, the only difference being whether we want to flush.
---
Changes in v3:
- style and indentation changes.
- return -ENOMEM instead of -1.
Changes in v2:
- improve asm.
- re-read pl3e from memory when taking the lock.
- move the allocation of l2t inside the shatter function.
--- xen/arch/x86/mm.c | 98 +++++++++++++++++++++++------------------------ 1 file changed, 49 insertions(+), 49 deletions(-)
xen/arch/x86/mm.c +49 -49
@@ -5151,6 +5151,52 @@ l1_pgentry_t *virt_to_xen_l1e(unsigned long v) flush_area_local((const void *)v, f) : \ flush_area_all((const void *)v, f)) +/* Shatter an l3 entry and populate l2. If virt is passed in, also do flush. */ +static int shatter_l3e (l3_pgentry_t *pl3e,unsigned long virt,bool locking) +{ +unsigned int i; +l3_pgentry_t ol3e = *pl3e; +l2_pgentry_t l2e =l2e_from_intpte (l3e_get_intpte (ol3e)); +l2_pgentry_t *l2t =alloc_xen_pagetable (); + +if ( !l2t ) +return -ENOMEM ; + +for ( i =0 ; i <L2_PAGETABLE_ENTRIES ; i++ ) + { +l2e_write (l2t + i, l2e); + l2e =l2e_from_intpte ( +l2e_get_intpte (l2e) + (PAGE_SIZE <<PAGETABLE_ORDER )); + } + +if ( locking ) +spin_lock (&map_pgdir_lock); +if ( (l3e_get_flags (*pl3e) & _PAGE_PRESENT) && + (l3e_get_flags (*pl3e) & _PAGE_PSE) ) + { +l3e_write_atomic (pl3e, +l3e_from_paddr ((paddr_t )virt_to_maddr (l2t), __PAGE_HYPERVISOR)); + l2t =NULL ; + } +if ( locking ) +spin_unlock (&map_pgdir_lock); + +if ( virt ) + { +unsigned int flush_flags = +FLUSH_TLB |FLUSH_ORDER (2 *PAGETABLE_ORDER ); + +if (l3e_get_flags (ol3e) & _PAGE_GLOBAL ) + flush_flags |=FLUSH_TLB_GLOBAL ; +flush_area (virt, flush_flags); + } + +if ( l2t ) +free_xen_pagetable (l2t); + +return 0 ; +} + int map_pages_to_xen( unsigned long virt, mfn_t mfn,
@@ -5244,9 +5290,6 @@ int map_pages_to_xen(if ( (l3e_get_flags (ol3e) & _PAGE_PRESENT) && (l3e_get_flags (ol3e) & _PAGE_PSE) ) { -unsigned int flush_flags = -FLUSH_TLB |FLUSH_ORDER (2 *PAGETABLE_ORDER ); -/* Skip this PTE if there is no change. */ if ( ((l3e_get_pfn(ol3e) & ~(L2_PAGETABLE_ENTRIES * L1_PAGETABLE_ENTRIES - 1)) +
@@ -5267,33 +5310,9 @@ int map_pages_to_xen(continue ; } - pl2e =alloc_xen_pagetable (); -if ( pl2e ==NULL ) +/* Pass virt to indicate we need to flush. */ +if (shatter_l3e (pl3e, virt, locking) )return -ENOMEM ; - -for ( i =0 ; i <L2_PAGETABLE_ENTRIES ; i++ ) -l2e_write (pl2e + i, -l2e_from_pfn (l3e_get_pfn (ol3e) + - (i <<PAGETABLE_ORDER ), -l3e_get_flags (ol3e))); - -if (l3e_get_flags (ol3e) & _PAGE_GLOBAL ) - flush_flags |=FLUSH_TLB_GLOBAL ; - -if ( locking ) -spin_lock (&map_pgdir_lock); -if ( (l3e_get_flags (*pl3e) & _PAGE_PRESENT) && - (l3e_get_flags (*pl3e) & _PAGE_PSE) ) - { -l3e_write_atomic (pl3e,l3e_from_mfn (virt_to_mfn (pl2e), - __PAGE_HYPERVISOR)); - pl2e =NULL ; - } -if ( locking ) -spin_unlock (&map_pgdir_lock); -flush_area (virt, flush_flags); -if ( pl2e ) -free_xen_pagetable (pl2e); } pl2e =virt_to_xen_l2e (virt);
@@ -5578,27 +5597,8 @@ int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) }/* PAGE1GB: shatter the superpage and fall through. */ - pl2e =alloc_xen_pagetable (); -if ( !pl2e ) +if (shatter_l3e (pl3e,0 , locking) )return -ENOMEM ; -for ( i =0 ; i <L2_PAGETABLE_ENTRIES ; i++ ) -l2e_write (pl2e + i, -l2e_from_pfn (l3e_get_pfn (*pl3e) + - (i <<PAGETABLE_ORDER ), -l3e_get_flags (*pl3e))); -if ( locking ) -spin_lock (&map_pgdir_lock); -if ( (l3e_get_flags (*pl3e) & _PAGE_PRESENT) && - (l3e_get_flags (*pl3e) & _PAGE_PSE) ) - { -l3e_write_atomic (pl3e,l3e_from_mfn (virt_to_mfn (pl2e), - __PAGE_HYPERVISOR)); - pl2e =NULL ; - } -if ( locking ) -spin_unlock (&map_pgdir_lock); -if ( pl2e ) -free_xen_pagetable (pl2e); }/*
Signature · 7 lines
2.17.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/2] x86/mm: factor out the code for shattering an l3 PTE
Quoted 33 lines · Hongyan Xia
map_pages_to_xenandmodify_xen_mappingsare performing almost exactly the same operations when shattering an l3 PTE, the only difference being whether we want to flush.Signed-off-by Hongyan Xia <[email protected]>---
Changes in v3:
- style and indentation changes.
- return-ENOMEMinstead of -1.Changes in v2:
- improve asm.
- re-read pl3e from memory when taking the lock.
- move the allocation of l2t inside the shatter function.--- xen/arch/x86/mm.c | 98 +++++++++++++++++++++++------------------------ 1 file changed, 49 insertions(+), 49 deletions(-)xen/arch/x86/mm.c +6
-0@@ -5151,6 +5151,52 @@ l1_pgentry_t *virt_to_xen_l1e(unsigned long v) flush_area_local((const void *)v, f) : \ flush_area_all((const void *)v, f)) +/* Shatter an l3 entry and populate l2. If virt is passed in, also do flush. */ +static int shatter_l3e (l3_pgentry_t *pl3e,unsigned long virt,bool locking) +{ +unsigned int i; +l3_pgentry_t ol3e = *pl3e; +l2_pgentry_t l2e =l2e_from_intpte (l3e_get_intpte (ol3e));
xen/arch/x86/mm.c l1_pgentry_t *virt_to_xen_l1e(unsigned long v) · line 5169
Quoted 10 lines
xen/arch/x86/mm.c +10
-0+l2_pgentry_t *l2t =alloc_xen_pagetable (); + +if ( !l2t ) +return -ENOMEM ; + +for ( i =0 ; i <L2_PAGETABLE_ENTRIES ; i++ ) + { +l2e_write (l2t + i, l2e); + l2e =l2e_from_intpte ( +l2e_get_intpte (l2e) + (PAGE_SIZE <<PAGETABLE_ORDER ));
Andrew - iirc you had suggested this (in some different form, but to the same effect) to improve code generation. If you're convinced that the downside of the loss of type safety is worth the win in generated code, I'm not going to stand in the way here, but it'll then need to be you to ack these two patches in their eventually final shape.
xen/arch/x86/mm.c l1_pgentry_t *virt_to_xen_l1e(unsigned long v) · line 5178
Quoted 9 lines
xen/arch/x86/mm.c +9
-0+ } + +if ( locking ) +spin_lock (&map_pgdir_lock); +if ( (l3e_get_flags (*pl3e) & _PAGE_PRESENT) && + (l3e_get_flags (*pl3e) & _PAGE_PSE) ) + { +l3e_write_atomic (pl3e, +l3e_from_paddr ((paddr_t )virt_to_maddr (l2t), __PAGE_HYPERVISOR));
Why the cast? (I'm sorry if this was there on v3 already and I didn't spot it. And if this remains the only thing to adjust, then I guess this could be taken care of while committing.)
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/2] x86/mm: factor out the code for shattering an l3 PTE
Quoted 11 lines · Jan Beulich
Quoted 10 lines
xen/arch/x86/mm.c +9
-0+ } + +if ( locking ) +spin_lock (&map_pgdir_lock); +if ( (l3e_get_flags (*pl3e) & _PAGE_PRESENT) && + (l3e_get_flags (*pl3e) & _PAGE_PSE) ) + { + l3e_write_atomic(pl3e, + l3e_from_paddr((paddr_t)virt_to_maddr(l2t),
__PAGE_HYPERVISOR));
Quoted 5 lines
Why the cast? (I'm sorry if this was there on v3 already and I didn't spot it. And if this remains the only thing to adjust, then I guess this could be taken care of while committing.)
Jan
Sadly there is no l3e_from_maddr or virt_to_paddr to call directly. Of course, paddr_t and maddr have the same underlying type (unsigned long), so it works without a cast. I just added the cast to make it explicit that these two are not exactly the same.
Hongyan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/2] x86/mm: factor out the code for shattering an l3 PTE
Quoted 12 lines · Hongyan
Quoted 11 lines · Jan Beulich
Quoted 10 lines
xen/arch/x86/mm.c +9
-0+ } + +if ( locking ) +spin_lock (&map_pgdir_lock); +if ( (l3e_get_flags (*pl3e) & _PAGE_PRESENT) && + (l3e_get_flags (*pl3e) & _PAGE_PSE) ) + { + l3e_write_atomic(pl3e, + l3e_from_paddr((paddr_t)virt_to_maddr(l2t),
__PAGE_HYPERVISOR));
Quoted 8 lines
Quoted 3 lines
Why the cast? (I'm sorry if this was there on v3 already and I didn't spot it. And if this remains the only thing to adjust, then I guess this could be taken care of while committing.)
Sadly there is no
l3e_from_maddrorvirt_to_paddrto call directly. Of course,paddr_tand maddr have the same underlying type (unsigned long), so it works without a cast. I just added the cast to make it explicit that these two are not exactly the same.
Yes, there continues to be a naming disconnect. But no, this is not a reason to add a cast. Casts should be used as sparingly as possible, since they tend to hide problems.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/2] x86/mm: factor out the code for shattering an l3 PTE
In reply to Hongyan Xiaxen/arch/x86/mm.c int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) · line 5601
Quoted 8 lines · Hongyan Xia
xen/arch/x86/mm.c +1
-2@@ -5578,27 +5597,8 @@ int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) }/* PAGE1GB: shatter the superpage and fall through. */ - pl2e =alloc_xen_pagetable (); -if ( !pl2e ) +if (shatter_l3e (pl3e,0 , locking) )return -ENOMEM ;
Hmm, I didn't expect I'd need to comment on this again: As per my v2 reply, you should hand on the return value from the function, not make up your own. This is so that in case the function gains another error path with a different error code, it wouldn't become indistinguishable to callers further up.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/2] x86/mm: factor out the code for shattering an l3 PTE
Quoted 5 lines · Jan Beulich
Quoted 4 lines · Hongyan Xia
(quoted diff) +0
-0@@ -5578,27 +5597,8 @@ int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) }
Quoted 12 lines
xen/arch/x86/mm.c int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) · line 5601
Quoted 5 lines
xen/arch/x86/mm.c +1
-2/* PAGE1GB: shatter the superpage and fall through. */ - pl2e =alloc_xen_pagetable (); -if ( !pl2e ) +if (shatter_l3e (pl3e,0 , locking) )return -ENOMEM ;Hmm, I didn't expect I'd need to comment on this again: As per my v2 reply, you should hand on the return value from the function, not make up your own. This is so that in case the function gains another error path with a different error code, it wouldn't become indistinguishable to callers further up.
I was basically thinking about the conversation we had that ENOMEM is probably the only error value map_pages_to_xen would return ever, and it is unlikely to gain another return value in the future, so initially I just let shatter return -1 and the caller return -ENOMEM. There is no problem for me if we want to change it to handle different error values.
Hongyan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/2] x86/mm: factor out the code for shattering an l3 PTE
Quoted 18 lines · Hongyan
Quoted 5 lines · Jan Beulich
Quoted 4 lines · Hongyan Xia
(quoted diff) +0
-0@@ -5578,27 +5597,8 @@ int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) }Quoted 11 lines
xen/arch/x86/mm.c int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf) · line 5601
Quoted 5 lines
xen/arch/x86/mm.c +1
-2/* PAGE1GB: shatter the superpage and fall through. */ - pl2e =alloc_xen_pagetable (); -if ( !pl2e ) +if (shatter_l3e (pl3e,0 , locking) )return -ENOMEM ;Hmm, I didn't expect I'd need to comment on this again: As per my v2 reply, you should hand on the return value from the function, not make up your own. This is so that in case the function gains another error path with a different error code, it wouldn't become indistinguishable to callers further up.
Quoted 6 lines
I was basically thinking about the conversation we had that ENOMEM is probably the only error value
map_pages_to_xenwould return ever, and it is unlikely to gain another return value in the future, so initially I just let shatter return -1 and the caller return-ENOMEM. There is no problem for me if we want to change it to handle different error values.
The alternative to your prior 0 / -1 returning would have been to have the function return bool. In this case "inventing" an error code here would be fine. The 0 / -1 approach would introduce another instance of what we're trying to get rid of elsewhere.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 2/2 x86/mm: factor out the code for shattering an l2 PTE
In reply to Hongyan Xiamap_pages_to_xen and modify_xen_mappings are performing almost exactly the same operations when shattering an l2 PTE, the only difference being whether we want to flush.
---
Changes in v3:
- style and indentation changes.
- return -ENOMEM instead of -1.
Changes in v2:
- improve asm.
- re-read pl2e from memory when taking the lock.
- move the allocation of l1t inside the shatter function.
--- xen/arch/x86/mm.c | 96 ++++++++++++++++++++++++----------------------- 1 file changed, 49 insertions(+), 47 deletions(-)
xen/arch/x86/mm.c +49 -47
@@ -5151,6 +5151,52 @@ l1_pgentry_t *virt_to_xen_l1e(unsigned long v) flush_area_local((const void *)v, f) : \ flush_area_all((const void *)v, f)) +/* Shatter an l2 entry and populate l1. If virt is passed in, also do flush. */ +static int shatter_l2e (l2_pgentry_t *pl2e,unsigned long virt,bool locking) +{ +unsigned int i; +l2_pgentry_t ol2e = *pl2e; +l1_pgentry_t l1e =l1e_from_paddr (l2e_get_paddr (ol2e), +lNf_to_l1f (l2e_get_flags (ol2e))); +l1_pgentry_t *l1t =alloc_xen_pagetable (); + +if ( !l1t ) +return -ENOMEM ; + +for ( i =0 ; i <L1_PAGETABLE_ENTRIES ; i++ ) + { +l1e_write (l1t + i, l1e); + l1e =l1e_from_intpte (l1e_get_intpte (l1e) +PAGE_SIZE ); + } + +if ( locking ) +spin_lock (&map_pgdir_lock); +if ( (l2e_get_flags (*pl2e) & _PAGE_PRESENT) && + (l2e_get_flags (*pl2e) & _PAGE_PSE) ) + { +l2e_write_atomic (pl2e, +l2e_from_paddr ((paddr_t )virt_to_maddr (l1t), __PAGE_HYPERVISOR)); + l1t =NULL ; + } +if ( locking ) +spin_unlock (&map_pgdir_lock); + +if ( virt ) + { +unsigned int flush_flags = +FLUSH_TLB |FLUSH_ORDER (PAGETABLE_ORDER ); + +if (l2e_get_flags (ol2e) & _PAGE_GLOBAL ) + flush_flags |=FLUSH_TLB_GLOBAL ; +flush_area (virt, flush_flags); + } + +if ( l1t ) +free_xen_pagetable (l1t); + +return 0 ; +} +/* Shatter an l3 entry and populate l2. If virt is passed in, also do flush. */ static int shatter_l3e (l3_pgentry_t *pl3e,unsigned long virt,bool locking) {
@@ -5364,9 +5410,6 @@ int map_pages_to_xen( }else if (l2e_get_flags (*pl2e) & _PAGE_PSE ) { -unsigned int flush_flags = -FLUSH_TLB |FLUSH_ORDER (PAGETABLE_ORDER ); -/* Skip this PTE if there is no change. */ if ( (((l2e_get_pfn(*pl2e) & ~(L1_PAGETABLE_ENTRIES - 1)) + l1_table_offset(virt)) == mfn_x(mfn)) &&
@@ -5385,32 +5428,9 @@ int map_pages_to_xen( goto check_l3; } - pl1e =alloc_xen_pagetable (); -if ( pl1e ==NULL ) +/* Pass virt to indicate we need to flush. */ +if (shatter_l2e (pl2e, virt, locking) )return -ENOMEM ; - -for ( i =0 ; i <L1_PAGETABLE_ENTRIES ; i++ ) -l1e_write (&pl1e[i], -l1e_from_pfn (l2e_get_pfn (*pl2e) + i, -lNf_to_l1f (l2e_get_flags (*pl2e)))); - -if (l2e_get_flags (*pl2e) & _PAGE_GLOBAL ) - flush_flags |=FLUSH_TLB_GLOBAL ; - -if ( locking ) -spin_lock (&map_pgdir_lock); -if ( (l2e_get_flags (*pl2e) & _PAGE_PRESENT) && - (l2e_get_flags (*pl2e) & _PAGE_PSE) ) - { -l2e_write_atomic (pl2e,l2e_from_mfn (virt_to_mfn (pl1e), - __PAGE_HYPERVISOR)); - pl1e =NULL ; - } -if ( locking ) -spin_unlock (&map_pgdir_lock); -flush_area (virt, flush_flags); -if ( pl1e ) -free_xen_pagetable (pl1e); } pl1e =l2e_to_l1e (*pl2e) +l1_table_offset (virt);
@@ -5633,26 +5653,8 @@ int modify_xen_mappings(unsigned long s, unsigned long e, unsigned int nf)else {/* PSE: shatter the superpage and try again. */ - pl1e =alloc_xen_pagetable (); -if ( !pl1e ) +if (shatter_l2e (pl2e,0 , locking) )return -ENOMEM ; -for ( i =0 ; i <L1_PAGETABLE_ENTRIES ; i++ ) -l1e_write (&pl1e[i], -l1e_from_pfn (l2e_get_pfn (*pl2e) + i, -l2e_get_flags (*pl2e) & ~_PAGE_PSE)); -if ( locking ) -spin_lock (&map_pgdir_lock); -if ( (l2e_get_flags (*pl2e) & _PAGE_PRESENT) && - (l2e_get_flags (*pl2e) & _PAGE_PSE) ) - { -l2e_write_atomic (pl2e,l2e_from_mfn (virt_to_mfn (pl1e), - __PAGE_HYPERVISOR)); - pl1e =NULL ; - } -if ( locking ) -spin_unlock (&map_pgdir_lock); -if ( pl1e ) -free_xen_pagetable (pl1e); } }else
Signature · 7 lines
2.17.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 2/2] x86/mm: factor out the code for shattering an l2 PTE
Quoted 5 lines · Hongyan Xia
map_pages_to_xenandmodify_xen_mappingsare performing almost exactly the same operations when shattering an l2 PTE, the only difference being whether we want to flush.Signed-off-by Hongyan Xia <[email protected]>
As before comments on patch 1 apply here as well.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel