[PATCH] x86/np2m: flush the np2m on an SVM nested TLB flush

[Stephen Cheng <stephen.cheng@xxxxxxxxxx>]

On SVM, L1 flushes what is cached from its nested page tables by setting
TLB_CONTROL or using a new ASID on VMRUN.  Either sets nv_flushp2m, which
p2m_get_nestedp2m_locked() handles by only dropping nv_p2m.  Since np2ms
are shared between vCPUs, the lookup then finds the np2m for the same
N_CR3 and reuses it as is, so L2 keeps the mappings from before L1 changed
its NPT.

Flush that np2m before reusing it, as nested VMX does via
np2m_flush_base() on INVEPT.

Fixes: 6df4b481b0c5 ("x86/np2m: implement sharing of np2m between vCPUs")
Signed-off-by: Stephen Cheng <stephen.cheng@xxxxxxxxxx>
Assisted-by: Claude:claude-opus-5-5
---
Ross Lagerwall has a partial fix for this in the XenServer internal patch
queue, which was also part of v1 of Chunjie Zhu's "x86/nestedsvm: rework
nested VMRUN/VMEXIT handling" series (patch 5/5, dropped in v2):
https://lore.kernel.org/xen-devel/20260924091605.344577-1-chunjie.zhu@xxxxxxxxxx/

It flushes nv_p2m, the np2m this vCPU used last, before dropping it:

         if ( nv->nv_flushp2m && nv->nv_p2m )
    +    {
    +        p2m = nv->nv_p2m;
    +        if ( p2m )
    +            p2m_flush_table(p2m);
             nv->nv_p2m = NULL;
    +    }

That covers a vCPU re-entering the same N_CR3, but not one entering an
N_CR3 whose np2m another vCPU filled: on its first VMRUN (nv_p2m is NULL),
with an ASID new to it, or when switching N_CR3.  This patch flushes the
np2m for the incoming N_CR3 instead, which covers all of these.

Tested with an XTF test that is not yet upstream: L1 repoints an NPT
entry, then enters L2 on another vCPU or N_CR3 under a new ASID.  L2 saw
the old frame before this patch and sees the new one with it.

 xen/arch/x86/mm/p2m.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
index 027b9ae69b..52cadd9b25 100644
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -1547,7 +1547,11 @@ p2m_get_nestedp2m_locked(struct vcpu *v)
         p2m_lock(p2m);
 
         if ( p2m->np2m_base == np2m_base )
+        {
+            if ( nv->nv_flushp2m )
+                p2m_flush_table_locked(p2m);
             goto found;
+        }
 
         p2m_unlock(p2m);
     }
-- 
2.49.0