x86/shadow: replace stale literal numbers in hash_{vcpu,domain}_foreach()
PATCH 1/1 x86/shadow: replace stale literal numbers in hash_{vcpu,domain}_foreach()
15 apparently once used to be the last valid type to request a callback for, and the dimension of the respective array. The arrays meanwhile are larger than this (in a benign way, i.e. no caller ever sets a mask bit higher than 15), dimensioned by SH_type_unused. Have the ASSERT()s follow suit and add build time checks at the call sites.
Also adjust a comment naming the wrong of the two functions.
---
The ASSERT()s being adjusted look redundant with the BUILD_BUG_ON()s being added, so I wonder whether dropping them wouldn't be the better route.
xen/arch/x86/mm/shadow/common.c +15 -5
@@ -1623,6 +1623,9 @@ void shadow_hash_delete(struct domain *dtypedef int (*hash_vcpu_callback_t )(struct vcpu *v,mfn_t smfn,mfn_t other_mfn);typedef int (*hash_domain_callback_t )(struct domain *d,mfn_t smfn,mfn_t other_mfn); +#define HASH_CALLBACKS_CHECK (mask) \ + BUILD_BUG_ON((mask) > (1U << ARRAY_SIZE(callbacks)) - 1) +static void hash_vcpu_foreach (struct vcpu *v,unsigned int callback_mask,const hash_vcpu_callback_t callbacks[],mfn_t callback_mfn)
@@ -1658,7 +1661,7 @@ static void hash_vcpu_foreach(struct vcp {if ( callback_mask & (1 << x->u.sh.type) ) { -ASSERT (x->u.sh.type <=15 ); +ASSERT (x->u.sh.type < SH_type_unused);ASSERT (callbacks[x->u.sh.type] !=NULL ); done = callbacks[x->u.sh.type](v,page_to_mfn (x), callback_mfn);
@@ -1705,7 +1708,7 @@ static void hash_domain_foreach(struct d {if ( callback_mask & (1 << x->u.sh.type) ) { -ASSERT (x->u.sh.type <=15 ); +ASSERT (x->u.sh.type < SH_type_unused);ASSERT (callbacks[x->u.sh.type] !=NULL ); done = callbacks[x->u.sh.type](d,page_to_mfn (x), callback_mfn);
@@ -2009,6 +2012,7 @@ int sh_remove_write_access(struct domainperfc_incr (shadow_writeable_bf_1); elseperfc_incr (shadow_writeable_bf ); +HASH_CALLBACKS_CHECK (callback_mask);hash_domain_foreach (d, callback_mask, callbacks, gmfn);/* If that didn't catch the mapping, then there's some non-pagetable
@@ -2080,6 +2084,7 @@ int sh_remove_all_mappings(struct domain/* Brute-force search of all the shadows, by walking the hash */ perfc_incr (shadow_mappings_bf); +HASH_CALLBACKS_CHECK (callback_mask);hash_domain_foreach (d, callback_mask, callbacks, gmfn);/* If that didn't catch the mapping, something is very wrong */
@@ -2246,10 +2251,12 @@ void sh_remove_shadows(struct domain *d,/* Search for this shadow in all appropriate shadows */ perfc_incr (shadow_unshadow); -/* Lower-level shadows need to be excised from upper-level shadows. -* This call to hash_vcpu_foreach() looks dangerous but is in fact OK: each +/* +* Lower-level shadows need to be excised from upper-level shadows. This +* call to hash_domain_foreach() looks dangerous but is in fact OK: each * call will remove at most one shadow, and terminate immediately when -* it does remove it, so we never walk the hash after doing a deletion. */ +* it does remove it, so we never walk the hash after doing a deletion. +*/ #define DO_UNSHADOW (_type) do { \ t = (_type); \ if( !(pg->count_info & PGC_page_table) \
@@ -2270,6 +2277,7 @@ void sh_remove_shadows(struct domain *d,if ( !fast \ && (pg->count_info & PGC_page_table) \ && (pg->shadow_flags & (1 << t)) ) \ +HASH_CALLBACKS_CHECK (SHF_page_type_mask); \hash_domain_foreach (d, masks[t], callbacks, smfn); \ }while (0 )
@@ -2370,6 +2378,7 @@ void sh_reset_l3_up_pointers(struct vcpu };static const unsigned int callback_mask =SHF_L3_64 ; +HASH_CALLBACKS_CHECK (callback_mask);hash_vcpu_foreach (v, callback_mask, callbacks,INVALID_MFN ); }
@@ -3420,6 +3429,7 @@ void shadow_audit_tables(struct vcpu *v) } } +HASH_CALLBACKS_CHECK (SHF_page_type_mask);hash_vcpu_foreach (v, mask, callbacks,INVALID_MFN ); }
Quoted 9 lines · Jan Beulich
15 apparently once used to be the last valid type to request a callback for, and the dimension of the respective array. The arrays meanwhile are larger than this (in a benign way, i.e. no caller ever sets a mask bit higher than 15), dimensioned by
SH_type_unused. Have the ASSERT()s follow suit and add build time checks at the call sites.Also adjust a comment naming the wrong of the two functions.
Signed-off-by Jan Beulich <[email protected]>
Quoted 4 lines
---
The ASSERT()s being adjusted look redundant with the BUILD_BUG_ON()s being added, so I wonder whether dropping them wouldn't be the better route.
I'm happy to keep both, as they do slightly different things.
Thanks for fixing this up!
Tim.
xen/arch/x86/mm/shadow/common.c void shadow_hash_delete(struct domain *d · line 1627
Quoted 8 lines · Jan Beulich
xen/arch/x86/mm/shadow/common.c +2
-0@@ -1623,6 +1623,9 @@ void shadow_hash_delete(struct domain *dtypedef int (*hash_vcpu_callback_t )(struct vcpu *v,mfn_t smfn,mfn_t other_mfn);typedef int (*hash_domain_callback_t )(struct domain *d,mfn_t smfn,mfn_t other_mfn); +#define HASH_CALLBACKS_CHECK (mask) \ + BUILD_BUG_ON((mask) > (1U << ARRAY_SIZE(callbacks)) - 1)
Sadly at least with Clang5 this doesn't work for some of the uses further down, e.g. ...
xen/arch/x86/mm/shadow/common.c int sh_remove_write_access(struct domain · line 2015
Quoted 5 lines
xen/arch/x86/mm/shadow/common.c +1
-0@@ -2009,6 +2012,7 @@ int sh_remove_write_access(struct domainperfc_incr (shadow_writeable_bf_1); elseperfc_incr (shadow_writeable_bf ); +HASH_CALLBACKS_CHECK (callback_mask);
... this one. I've made it
#ifndef __clang__ /* At least some versions dislike some of the uses. */
#define HASH_CALLBACKS_CHECK(mask) \
BUILD_BUG_ON((mask) > (1U << ARRAY_SIZE(callbacks)) - 1)#else
#define HASH_CALLBACKS_CHECK(mask) ((void)(mask))
#endif
for the time being - if anyone has any better idea, I'll be happy to take suggestions.
Jan