From: Julien Grall <jgrall@amazon.com>
Subject: xen/x86: p2m: Add preemption in p2m_teardown()

The list p2m->pages contain all the pages used by the P2M. On large
instance this can be quite large and the time spent to call
d->arch.paging.free_page() will take more than 1ms for a 80GB guest
on a Xen running in nested environment on a c5.metal.

By extrapolation, it would take > 100ms for a 8TB guest (what we
current security support). So add some preemption in p2m_teardown()
and propagate to the callers. Note there are 3 places where
the preemption is not enabled:
    - hap_final_teardown()/shadow_final_teardown(): We are
      preventing update the P2M once the domain is dying (so
      no more pages could be allocated) and most of the P2M pages
      will be freed in preemptive manneer when relinquishing the
      resources. So this is fine to disable preemption.
    - shadow_enable(): This is fine because it will undo the allocation
      that may have been made by p2m_alloc_table() (so only the root
      page table).

The preemption is arbitrarily checked every 1024 iterations.

Note that with the current approach, Xen doesn't keep track on whether
the alt/nested P2Ms have been cleared. So there are some redundant work.
However, this is not expected to incurr too much overhead (the P2M lock
shouldn't be contended during teardown). So this is optimization is
left outside of the security event.

This is part of CVE-2022-33746 / XSA-410.

Signed-off-by: Julien Grall <jgrall@amazon.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- a/xen/include/asm-x86/p2m.h
+++ b/xen/include/asm-x86/p2m.h
@@ -574,7 +574,7 @@ int p2m_init(struct domain *d);
 int p2m_alloc_table(struct p2m_domain *p2m);
 
 /* Return all the p2m resources to Xen. */
-void p2m_teardown(struct p2m_domain *p2m, bool remove_root);
+void p2m_teardown(struct p2m_domain *p2m, bool remove_root, bool *preempted);
 void p2m_final_teardown(struct domain *d);
 
 /* Add a page to a domain's p2m table */
--- a/xen/arch/x86/mm/hap/hap.c
+++ b/xen/arch/x86/mm/hap/hap.c
@@ -548,17 +548,17 @@ void hap_final_teardown(struct domain *d
 
     if ( hvm_altp2m_supported() )
         for ( i = 0; i < MAX_ALTP2M; i++ )
-            p2m_teardown(d->arch.altp2m_p2m[i], true);
+            p2m_teardown(d->arch.altp2m_p2m[i], true, NULL);
 
     /* Destroy nestedp2m's first */
     for (i = 0; i < MAX_NESTEDP2M; i++) {
-        p2m_teardown(d->arch.nested_p2m[i], true);
+        p2m_teardown(d->arch.nested_p2m[i], true, NULL);
     }
 
     if ( d->arch.paging.hap.total_pages != 0 )
         hap_teardown(d, NULL);
 
-    p2m_teardown(p2m_get_hostp2m(d), true);
+    p2m_teardown(p2m_get_hostp2m(d), true, NULL);
     /* Free any memory that the p2m teardown released */
     paging_lock(d);
     hap_set_allocation(d, 0, NULL);
@@ -612,14 +612,24 @@ void hap_teardown(struct domain *d, bool
         FREE_XENHEAP_PAGE(d->arch.altp2m_visible_eptp);
 
         for ( i = 0; i < MAX_ALTP2M; i++ )
-            p2m_teardown(d->arch.altp2m_p2m[i], false);
+        {
+            p2m_teardown(d->arch.altp2m_p2m[i], false, preempted);
+            if ( preempted && *preempted )
+                return;
+        }
     }
 
     /* Destroy nestedp2m's after altp2m. */
     for ( i = 0; i < MAX_NESTEDP2M; i++ )
-        p2m_teardown(d->arch.nested_p2m[i], false);
+    {
+        p2m_teardown(d->arch.nested_p2m[i], false, preempted);
+        if ( preempted && *preempted )
+            return;
+    }
 
-    p2m_teardown(p2m_get_hostp2m(d), false);
+    p2m_teardown(p2m_get_hostp2m(d), false, preempted);
+    if ( preempted && *preempted )
+        return;
 
     paging_lock(d); /* Keep various asserts happy */
 
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -749,12 +749,13 @@ int p2m_alloc_table(struct p2m_domain *p
  * hvm fixme: when adding support for pvh non-hardware domains, this path must
  * cleanup any foreign p2m types (release refcnts on them).
  */
-void p2m_teardown(struct p2m_domain *p2m, bool remove_root)
+void p2m_teardown(struct p2m_domain *p2m, bool remove_root, bool *preempted)
 /* Return all the p2m pages to Xen.
  * We know we don't have any extra mappings to these pages */
 {
     struct page_info *pg, *root_pg = NULL;
     struct domain *d;
+    unsigned int i = 0;
 
     if (p2m == NULL)
         return;
@@ -773,8 +774,19 @@ void p2m_teardown(struct p2m_domain *p2m
     }
 
     while ( (pg = page_list_remove_head(&p2m->pages)) )
-        if ( pg != root_pg )
-            d->arch.paging.free_page(d, pg);
+    {
+        if ( pg == root_pg )
+            continue;
+
+        d->arch.paging.free_page(d, pg);
+
+        /* Arbitrarily check preemption every 1024 iterations */
+        if ( preempted && !(++i % 1024) && general_preempt_check() )
+        {
+            *preempted = true;
+            break;
+        }
+    }
 
     if ( root_pg )
         page_list_add(root_pg, &p2m->pages);
--- a/xen/arch/x86/mm/shadow/common.c
+++ b/xen/arch/x86/mm/shadow/common.c
@@ -2770,8 +2770,12 @@ int shadow_enable(struct domain *d, u32
  out_locked:
     paging_unlock(d);
  out_unlocked:
+    /*
+     * This is fine to ignore the preemption here because only the root
+     * will be allocated by p2m_alloc_table().
+     */
     if ( rv != 0 && !pagetable_is_null(p2m_get_pagetable(p2m)) )
-        p2m_teardown(p2m, true);
+        p2m_teardown(p2m, true, NULL);
     if ( rv != 0 && pg != NULL )
     {
         pg->count_info &= ~PGC_count_mask;
@@ -2824,7 +2828,9 @@ void shadow_teardown(struct domain *d, b
     for_each_vcpu ( d, v )
         shadow_vcpu_teardown(v);
 
-    p2m_teardown(p2m_get_hostp2m(d), false);
+    p2m_teardown(p2m_get_hostp2m(d), false, preempted);
+    if ( preempted && *preempted )
+        return;
 
     paging_lock(d);
 
@@ -2945,7 +2951,7 @@ void shadow_final_teardown(struct domain
         shadow_teardown(d, NULL);
 
     /* It is now safe to pull down the p2m map. */
-    p2m_teardown(p2m_get_hostp2m(d), true);
+    p2m_teardown(p2m_get_hostp2m(d), true, NULL);
     /* Free any shadow memory that the p2m teardown released */
     paging_lock(d);
     shadow_set_allocation(d, 0, NULL);