Changeset 18597 in vbox

Timestamp:

Apr 1, 2009 1:34:19 PM (16 years ago)

Author:

vboxsync

svn:sync-xref-src-repo-rev:

45501

Message:

REM: Added a l0 map for PageDesc, this should fix the performance issue if our hunch is right about the cause. Profile tb_flush.

Location:

trunk/src/recompiler_new

Files:

• VBoxRecompiler.c (modified) (1 diff)
• exec.c (modified) (6 diffs)
• target-i386/cpu.h (modified) (2 diffs)

trunk/src/recompiler_new/VBoxRecompiler.c

@@ -390 +390 @@
     STAM_REG(pVM, &gStatSelOutOfSyncStateBack[4],   STAMTYPE_COUNTER,   "/REM/StateBack/SelOutOfSync/FS",   STAMUNIT_OCCURENCES, "FS out of sync");
     STAM_REG(pVM, &gStatSelOutOfSyncStateBack[5],   STAMTYPE_COUNTER,   "/REM/StateBack/SelOutOfSync/GS",   STAMUNIT_OCCURENCES, "GS out of sync");
+
+    STAM_REG(pVM, &pVM->rem.s.Env.StatTbFlush,      STAMTYPE_PROFILE,   "/REM/TbFlush",     STAMUNIT_TICKS_PER_CALL, "profiling tb_flush().");
 #endif /* VBOX_WITH_STATISTICS */
 

trunk/src/recompiler_new/exec.c

@@ -182 +182 @@
 #define L1_BITS (TARGET_VIRT_ADDR_SPACE_BITS - L2_BITS - TARGET_PAGE_BITS)
 #else
-# ifdef VBOX /* > 4GB please. */
-#define L1_BITS (TARGET_PHYS_ADDR_SPACE_BITS - L2_BITS - TARGET_PAGE_BITS)
-# else
 #define L1_BITS (32 - L2_BITS - TARGET_PAGE_BITS)
-# endif
-#endif
-
+#endif
+#ifdef VBOX
+/* Note: Not for PhysPageDesc, only to speed up page_flush_tb. */
+#define L0_BITS (TARGET_PHYS_ADDR_SPACE_BITS - 32)
+#endif
+
+#ifdef VBOX
+#define L0_SIZE (1 << L0_BITS)
+#endif
 #define L1_SIZE (1 << L1_BITS)
 #define L2_SIZE (1 << L2_BITS)
@@ -200 +203 @@
 
 /* XXX: for system emulation, it could just be an array */
+#ifndef VBOX
 static PageDesc *l1_map[L1_SIZE];
+#else
+static l0_map_max_used = 0;
+static PageDesc **l0_map[L0_SIZE];
+#endif
 static PhysPageDesc **l1_phys_map;
 
@@ -353 +361 @@
 #endif
 {
+#ifndef VBOX
 #if TARGET_LONG_BITS > 32
     /* Host memory outside guest VM.  For 32-bit targets we have already
        excluded high addresses.  */
-# ifndef VBOX
     if (index > ((target_ulong)L2_SIZE * L1_SIZE))
         return NULL;
-# else  /* VBOX */
-    AssertMsgReturn(index < (target_ulong)L2_SIZE * L1_SIZE,
-                    ("index=%RGp >= %RGp; L1_SIZE=%#x L2_SIZE=%#x\n",
-                     (RTGCPHYS)index, (RTGCPHYS)L2_SIZE * L1_SIZE, L1_SIZE, L2_SIZE),
+#endif
+    return &l1_map[index >> L2_BITS];
+#else  /* VBOX */
+    PageDesc **l1_map;
+    AssertMsgReturn(index < (target_ulong)L2_SIZE * L1_SIZE * L0_SIZE,
+                    ("index=%RGp >= %RGp; L1_SIZE=%#x L2_SIZE=%#x L0_SIZE=%#x\n",
+                     (RTGCPHYS)index, (RTGCPHYS)L2_SIZE * L1_SIZE, L1_SIZE, L2_SIZE, L0_SIZE),
                     NULL);
-# endif /* VBOX */
-
-#endif
-    return &l1_map[index >> L2_BITS];
+    l1_map = l0_map[index >> (L1_BITS + L2_BITS)];
+    if (RT_UNLIKELY(!l1_map))
+    {
+        unsigned i0 = index >> (L1_BITS + L2_BITS);
+        l0_map[i0] = l1_map = qemu_mallocz(sizeof(PageDesc *) * L1_SIZE);
+        if (RT_UNLIKELY(!l1_map))
+            return NULL;
+        if (i0 >= l0_map_max_used)
+            l0_map_max_used = i0 + 1;
+    }
+    return &l1_map[(index >> L2_BITS) & (L1_SIZE - 1)];
+#endif /* VBOX */
 }
 
@@ -703 +722 @@
     int i, j;
     PageDesc *p;
-
-    for(i = 0; i < L1_SIZE; i++) {
-        p = l1_map[i];
-        if (p) {
-            for(j = 0; j < L2_SIZE; j++) {
-                p->first_tb = NULL;
-                invalidate_page_bitmap(p);
-                p++;
+#ifdef VBOX
+    int k;
+#endif
+
+#ifdef VBOX
+    k = l0_map_max_used;
+    while (k-- > 0) {
+        PageDesc **l1_map = l0_map[k];
+        if (l1_map) {
+#endif
+            for(i = 0; i < L1_SIZE; i++) {
+                p = l1_map[i];
+                if (p) {
+                    for(j = 0; j < L2_SIZE; j++) {
+                        p->first_tb = NULL;
+                        invalidate_page_bitmap(p);
+                        p++;
+                    }
+                }
             }
-        }
-    }
+#ifdef VBOX
+        }
+    }
+#endif
 }
 
@@ -721 +753 @@
 {
     CPUState *env;
+#ifdef VBOX
+    STAM_PROFILE_START(&env1->StatTbFlush, a);
+#endif
 #if defined(DEBUG_FLUSH)
     printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n",
@@ -743 +778 @@
        expensive */
     tb_flush_count++;
+#ifdef VBOX
+    STAM_PROFILE_STOP(&env1->StatTbFlush, a);
+#endif
 }
 

trunk/src/recompiler_new/target-i386/cpu.h

@@ -62 +62 @@
 # include <iprt/asm.h>
 # include <VBox/vmm.h>
+# include <VBox/stam.h>
 #endif /* VBOX */
 
@@ -677 +678 @@
 #else
     uint32_t alignment2[3];
+    /** Profiling tb_flush. */
+    STAMPROFILE StatTbFlush;
 #endif
 } CPUX86State;