Update p2m->need_flush if page was present before (requiring a tlb flush).
This causes p2m->flush_tlb() to be now be reachable, make sure we call it
only when set.

Signed-off-by: Teddy Astie <teddy.astie@vates.tech>
---
This change is important for the next patch.
Would it be better to merge it with the next patch ?

 xen/arch/x86/mm/p2m-pt.c | 3 +++
 xen/arch/x86/mm/p2m.c    | 6 ++++--
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/xen/arch/x86/mm/p2m-pt.c b/xen/arch/x86/mm/p2m-pt.c
index 5a6ce2f8bc..1fea3884be 100644
--- a/xen/arch/x86/mm/p2m-pt.c
+++ b/xen/arch/x86/mm/p2m-pt.c
@@ -143,6 +143,9 @@ static int write_p2m_entry(struct p2m_domain *p2m, unsigned long gfn,
         if ( p2m->write_p2m_entry_post )
             p2m->write_p2m_entry_post(p2m, oflags);
 
+        if ( oflags & _PAGE_PRESENT )
+            p2m->need_flush = true;
+
         paging_unlock(d);
 
         if ( nestedhvm_enabled(d) && !p2m_is_nestedp2m(p2m) &&
diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
index e2a00a0efd..98f8272270 100644
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -274,7 +274,8 @@ void p2m_tlb_flush_sync(struct p2m_domain *p2m)
     if ( p2m->need_flush )
     {
         p2m->need_flush = 0;
-        p2m->tlb_flush(p2m);
+        if ( p2m->tlb_flush )
+            p2m->tlb_flush(p2m);
     }
 }
 
@@ -287,7 +288,8 @@ void p2m_unlock_and_tlb_flush(struct p2m_domain *p2m)
     {
         p2m->need_flush = 0;
         mm_write_unlock(&p2m->lock);
-        p2m->tlb_flush(p2m);
+        if ( p2m->tlb_flush )
+            p2m->tlb_flush(p2m);
     } else
         mm_write_unlock(&p2m->lock);
 }
-- 
2.51.2



--
Teddy Astie | Vates XCP-ng Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech

On 27.11.2025 14:39, Teddy Astie wrote:
> Update p2m->need_flush if page was present before (requiring a tlb flush).
> This causes p2m->flush_tlb() to be now be reachable, make sure we call it
> only when set.

I don't follow. Why would p2m->flush_tlb() not have been reachable? There are
other places where ->need_flush is set to true.

> Signed-off-by: Teddy Astie <teddy.astie@vates.tech>
> ---
> This change is important for the next patch.
> Would it be better to merge it with the next patch ?

I'm inclined to say yes, but without understanding the description here I
cannot be sure.

Hmm, wait. The sole place in x86 where the flag is set is in EPT code. So
do you perhaps mean that for NPT or shadow those call sites can now be
reached, and hence - with the hook being NULL there - checks need adding?
Then, by its title, the next patch actually may be making use of the hook,
so the checks then can be dropped again?

Jan

Le 27/11/2025 à 15:08, Jan Beulich a écrit :
> On 27.11.2025 14:39, Teddy Astie wrote:
>> Update p2m->need_flush if page was present before (requiring a tlb flush).
>> This causes p2m->flush_tlb() to be now be reachable, make sure we call it
>> only when set.
> 
> I don't follow. Why would p2m->flush_tlb() not have been reachable? There are
> other places where ->need_flush is set to true.
> 
>> Signed-off-by: Teddy Astie <teddy.astie@vates.tech>
>> ---
>> This change is important for the next patch.
>> Would it be better to merge it with the next patch ?
> 
> I'm inclined to say yes, but without understanding the description here I
> cannot be sure.
> 
> Hmm, wait. The sole place in x86 where the flag is set is in EPT code. So
> do you perhaps mean that for NPT or shadow those call sites can now be
> reached, and hence - with the hook being NULL there - checks need adding?
> Then, by its title, the next patch actually may be making use of the hook,
> so the checks then can be dropped again?
> 

Even with the next patch, in the shadow case, p2m->flush_tlb is still 
NULL as only HAP-NPT uses hap_p2m_tlb_flush().

> Jan



--
Teddy Astie | Vates XCP-ng Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech