diff --git a/target/i386/tcg/sysemu/excp_helper.c b/target/i386/tcg/sysemu/excp_helper.c
index 61b8d2fdde..5999cdedf5 100644
--- a/target/i386/tcg/sysemu/excp_helper.c
+++ b/target/i386/tcg/sysemu/excp_helper.c
@@ -133,7 +133,6 @@ static inline bool ptw_setl(const PTETranslate *in, uint32_t old, uint32_t set)
 static bool mmu_translate(CPUX86State *env, const TranslateParams *in,
                           TranslateResult *out, TranslateFault *err)
 {
-    const int32_t a20_mask = x86_get_a20_mask(env);
     const target_ulong addr = in->addr;
     const int pg_mode = in->pg_mode;
     const bool is_user = (in->mmu_idx == MMU_USER_IDX);
@@ -415,10 +414,13 @@ do_check_protect_pse36:
         }
     }
 
-    /* align to page_size */
-    paddr = (pte & a20_mask & PG_ADDRESS_MASK & ~(page_size - 1))
-          | (addr & (page_size - 1));
+    /* merge offset within page */
+    paddr = (pte & PG_ADDRESS_MASK & ~(page_size - 1)) | (addr & (page_size - 1));
 
+    /*
+     * Note that NPT is walked (for both paging structures and final guest
+     * addresses) using the address with the A20 bit set.
+     */
     if (in->ptw_idx == MMU_NESTED_IDX) {
         CPUTLBEntryFull *full;
         int flags, nested_page_size;
@@ -457,7 +459,7 @@ do_check_protect_pse36:
         }
     }
 
-    out->paddr = paddr;
+    out->paddr = paddr & x86_get_a20_mask(env);
     out->prot = prot;
     out->page_size = page_size;
     return true;