Skip to contentSkip to conversation

x86/HVM: get_pat_flags() is needed only by shadow code

PATCH In review Jan Beulich 4 participants 5 messages Jul 18, 2024 – Jul 19, 2024

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/HVM: get_pat_flags() is needed only by shadow code

Therefore with SHADOW_PAGING=n this is better compiled out, to avoid leaving around unreachable/dead code.

xen/arch/x86/hvm/mtrr.c +4 -0
@@ -271,6 +271,8 @@ int mtrr_get_type(const struct mtrr_stat
    return overlap_mtrr_pos;
 }
 
+#ifdef CONFIG_SHADOW_PAGING
+
 /*
  * return the memory type from PAT.
  * NOTE: valid only when paging is enabled.
@@ -359,6 +361,8 @@ uint32_t get_pat_flags(struct vcpu *v,
     return pat_type_2_pte_flags(pat_entry_value);
 }
 
+#endif /* CONFIG_SHADOW_PAGING */
+
 static inline bool valid_mtrr_type(uint8_t type)
 {
     switch ( type )
#3 lore raw
In reply to Jan Beulich
Quoted 17 lines · Jan Beulich

Therefore with SHADOW_PAGING=n this is better compiled out, to avoid leaving around unreachable/dead code.

xen/arch/x86/hvm/mtrr.c +2 -0
@@ -271,6 +271,8 @@ int mtrr_get_type(const struct mtrr_stat
    return overlap_mtrr_pos;
 }
 
+#ifdef CONFIG_SHADOW_PAGING
+
 /*
  * return the memory type from PAT.
  * NOTE: valid only when paging is enabled.

xen/arch/x86/hvm/mtrr.c uint32_t get_pat_flags(struct vcpu *v, · line 368

Quoted 8 lines
xen/arch/x86/hvm/mtrr.c +2 -0
@@ -359,6 +361,8 @@ uint32_t get_pat_flags(struct vcpu *v,
     return pat_type_2_pte_flags(pat_entry_value);
 }
 
+#endif /* CONFIG_SHADOW_PAGING */
+
 static inline bool valid_mtrr_type(uint8_t type)
 {
     switch ( type )

While I can see this is true, the fact it is indicates that we have bugs/problems elsewhere.

It is not only the shadow code that has to combine attributes like this, so we've either got opencoding elsewhere, or a bad abstraction.

(This is an observation, rather than a specific objection.)

~Andrew

#4 lore raw
Quoted 30 lines · Andrew Cooper
Quoted 17 lines · Jan Beulich

Therefore with SHADOW_PAGING=n this is better compiled out, to avoid leaving around unreachable/dead code.

xen/arch/x86/hvm/mtrr.c +2 -0
@@ -271,6 +271,8 @@ int mtrr_get_type(const struct mtrr_stat
    return overlap_mtrr_pos;
 }
 
+#ifdef CONFIG_SHADOW_PAGING
+
 /*
  * return the memory type from PAT.
  * NOTE: valid only when paging is enabled.

xen/arch/x86/hvm/mtrr.c uint32_t get_pat_flags(struct vcpu *v, · line 368

Quoted 8 lines
xen/arch/x86/hvm/mtrr.c +2 -0
@@ -359,6 +361,8 @@ uint32_t get_pat_flags(struct vcpu *v,
     return pat_type_2_pte_flags(pat_entry_value);
 }
 
+#endif /* CONFIG_SHADOW_PAGING */
+
 static inline bool valid_mtrr_type(uint8_t type)
 {
     switch ( type )

While I can see this is true, the fact it is indicates that we have bugs/problems elsewhere.

Quoted 4 lines

It is not only the shadow code that has to combine attributes like this, so we've either got opencoding elsewhere, or a bad abstraction.

(This is an observation, rather than a specific objection.)

Won't shadow always need a specific helper, in order to combine both MTRRs and guest PAT attributes, while HAP only needs to merge MTRR attributes?

Not that current code couldn't be better structured.

Thanks, Roger.