bitkeeper revision 1.1236.1.183 (424d1ccaDlUUQZ1KrVHubRYFS5AKQA)

Batch p.t. pin requests during guest restore (after being suspended, or
after migration). Allow L1 tables to be pinned even after their va
backptr is fixed (no longer mutable).
Signed-off-by: Keir Fraser <keir@xensource.com>

diff --git a/tools/libxc/xc_linux_restore.c b/tools/libxc/xc_linux_restore.c
index 220890d3461a9cb6b5e989c7eecf5ce79592e9ef..79f0c6ac455ac8f3bad0d0ae7283c0160bf914ae 100644 (file)
--- a/tools/libxc/xc_linux_restore.c
+++ b/tools/libxc/xc_linux_restore.c
@@ -106,6 +106,10 @@ int xc_linux_restore(int xc_handle, XcIOContext *ioctxt)
     /* used by debug verify code */
     unsigned long buf[PAGE_SIZE/sizeof(unsigned long)];
 
+#define MAX_PIN_BATCH 1024
+    struct mmuext_op pin[MAX_PIN_BATCH];
+    unsigned int nr_pins = 0;
+
     xcio_info(ioctxt, "xc_linux_restore start\n");
 
     if ( mlock(&ctxt, sizeof(ctxt) ) )
@@ -414,38 +418,33 @@ int xc_linux_restore(int xc_handle, XcIOContext *ioctxt)
 
     xcio_info(ioctxt, "Received all pages\n");
 
+    if ( finish_mmu_updates(xc_handle, mmu) )
+        goto out;
+
     /*
      * Pin page tables. Do this after writing to them as otherwise Xen
      * will barf when doing the type-checking.
      */
     for ( i = 0; i < nr_pfns; i++ )
     {
+        if ( (pfn_type[i] & LPINTAB) == 0 )
+            continue;
         if ( pfn_type[i] == (L1TAB|LPINTAB) )
+            pin[nr_pins].cmd = MMUEXT_PIN_L1_TABLE;
+        else /* pfn_type[i] == (L2TAB|LPINTAB) */
+            pin[nr_pins].cmd = MMUEXT_PIN_L2_TABLE;
+        pin[nr_pins].mfn = pfn_to_mfn_table[i];
+        if ( ++nr_pins == MAX_PIN_BATCH )
         {
-            if ( pin_table(xc_handle, MMUEXT_PIN_L1_TABLE,
-                           pfn_to_mfn_table[i], dom) ) {
-                printf("ERR pin L1 pfn=%lx mfn=%lx\n",
-                       (unsigned long)i, pfn_to_mfn_table[i]);
-                goto out;
-            }
-        }
-    }
-
-    /* must pin all L1's before L2's (need consistent va back ptr) */
-    for ( i = 0; i < nr_pfns; i++ )
-    {
-        if ( pfn_type[i] == (L2TAB|LPINTAB) )
-        {
-            if ( pin_table(xc_handle, MMUEXT_PIN_L2_TABLE,
-                           pfn_to_mfn_table[i], dom) ) {
-                printf("ERR pin L2 pfn=%lx mfn=%lx\n",
-                       (unsigned long)i, pfn_to_mfn_table[i]);
+            if ( do_mmuext_op(xc_handle, pin, nr_pins, dom) < 0 )
                 goto out;
-            }
+            nr_pins = 0;
         }
     }
 
-    if ( finish_mmu_updates(xc_handle, mmu) ) goto out;
+    if ( (nr_pins != 0) &&
+         (do_mmuext_op(xc_handle, pin, nr_pins, dom) < 0) )
+        goto out;
 
     xcio_info(ioctxt, "\b\b\b\b100%%\n");
     xcio_info(ioctxt, "Memory reloaded.\n");

diff --git a/tools/libxc/xc_private.c b/tools/libxc/xc_private.c
index 386a240178f364cb50d12c8b81ea7ce27be86a67..15327ccdbb1de7eefbe4bdf25d92c6ef94e22f1e 100644 (file)
--- a/tools/libxc/xc_private.c
+++ b/tools/libxc/xc_private.c
@@ -95,36 +95,15 @@ unsigned int get_pfn_type(int xc_handle,
 int pin_table(
     int xc_handle, unsigned int type, unsigned long mfn, domid_t dom)
 {
-    int err = 0;
     struct mmuext_op op;
-    privcmd_hypercall_t hypercall;
 
     op.cmd = type;
     op.mfn = mfn;
 
-    hypercall.op     = __HYPERVISOR_mmuext_op;
-    hypercall.arg[0] = (unsigned long)&op;
-    hypercall.arg[1] = 1;
-    hypercall.arg[2] = 0;
-    hypercall.arg[3] = dom;
-
-    if ( mlock(&op, sizeof(op)) != 0 )
-    {
-        PERROR("Could not lock mmuext_op");
-        err = 1;
-        goto out;
-    }
-
-    if ( do_xen_hypercall(xc_handle, &hypercall) < 0 )
-    {
-        ERROR("Failure when submitting mmu updates");
-        err = 1;
-    }
+    if ( do_mmuext_op(xc_handle, &op, 1, dom) < 0 )
+        return 1;
 
-    (void)munlock(&op, sizeof(op));
-
- out:
-    return err;
+    return 0;
 }
 
 static int flush_mmu_updates(int xc_handle, mmu_t *mmu)

diff --git a/tools/libxc/xc_private.h b/tools/libxc/xc_private.h
index 78f8bbe194c5c528976bb0513718b6969e980a9e..8445e75fbeee71c7669fa032bcd238df86fdb837 100644 (file)
--- a/tools/libxc/xc_private.h
+++ b/tools/libxc/xc_private.h
@@ -142,6 +142,38 @@ static inline int do_dom_mem_op(int            xc_handle,
  out1: return ret;
 }    
 
+static inline int do_mmuext_op(
+    int xc_handle,
+    struct mmuext_op *op,
+    unsigned int nr_ops,
+    domid_t dom)
+{
+    privcmd_hypercall_t hypercall;
+    long ret = -EINVAL;
+       
+    hypercall.op     = __HYPERVISOR_mmuext_op;
+    hypercall.arg[0] = (unsigned long)op;
+    hypercall.arg[1] = (unsigned long)nr_ops;
+    hypercall.arg[2] = (unsigned long)0;
+    hypercall.arg[3] = (unsigned long)dom;
+
+    if ( mlock(op, nr_ops*sizeof(*op)) != 0 )
+    {
+        PERROR("Could not lock memory for Xen hypercall");
+        goto out1;
+    }
+
+    if ( (ret = do_xen_hypercall(xc_handle, &hypercall)) < 0 )
+    {
+       fprintf(stderr, "Dom_mem operation failed (rc=%ld errno=%d)-- need to"
+                    " rebuild the user-space tool set?\n",ret,errno);
+        goto out2;
+    }
+
+ out2: (void)munlock(op, nr_ops*sizeof(*op));
+ out1: return ret;
+}    
+
 
 /*
  * PFN mapping.

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index b67458f521007e2591a67c2eaf62a46b2bcd5c6f..845fbc9f66ecada394001fef55c9a6543c46ba3f 100644 (file)
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -1186,7 +1186,8 @@ int get_page_type(struct pfn_info *page, u32 type)
                 nx &= ~PGT_va_mask;
                 nx |= type; /* we know the actual type is correct */
             }
-            else if ( unlikely((x & PGT_va_mask) != (type & PGT_va_mask)) )
+            else if ( ((type & PGT_va_mask) != PGT_va_mutable) &&
+                      ((type & PGT_va_mask) != (x & PGT_va_mask)) )
             {
                 /* This table is potentially mapped at multiple locations. */
                 nx &= ~PGT_va_mask;
@@ -1388,11 +1389,6 @@ int do_mmuext_op(
         switch ( op.cmd )
         {
         case MMUEXT_PIN_L1_TABLE:
-            /*
-             * We insist that, if you pin an L1 page, it's the first thing that
-             * you do to it. This is because we require the backptr to still be
-             * mutable. This assumption seems safe.
-             */
             type = PGT_l1_page_table | PGT_va_mutable;
 
         pin_page: