Changeset 21006 in vbox

Timestamp:

Jun 28, 2009 7:07:20 PM (16 years ago)

Author:

vboxsync

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

49220

Message:

VHD:

• Fix calculation of block bitmap bit to test or set

Explanation: The code to avoid unaligned accesses for bt calculates the wrong bit to test.

If sector 0 in the block bitmap should be tested we have to test bit 7 in the first
byte the pointer of the data block bitmap points to.
The current code however would test the 31th bit which is wrong.
It would be correct to add the skipped bytes as number of bits to the bit position
like iBitInByte = ((8-1) - (cBlockBitmapEntry % 8)) + ((cBlockBitmapEntry / 8) & 3)*8
to ensure aligned access but I doubt that the additional instructions outweigh the
the slower memory access. So we restore the old behavior here and accept unaligned accesses.

• On 64bit Windows systems we have to add 8 bytes instead of 4 because BT might read 8 bytes from the memory operand

Many thanks to Huihong Luo for pointing that out!

File:

• trunk/src/VBox/Devices/Storage/VHDHDDCore.cpp (modified) (3 diffs)

trunk/src/VBox/Devices/Storage/VHDHDDCore.cpp

@@ -223 +223 @@
 
 /**
- * Internal: Allocates the block bitmap rounding up to the next 32bit boundary.
+ * Internal: Allocates the block bitmap rounding up to the next 32bit or 64bit boundary.
  *           Can be freed with RTMemFree. The memory is zeroed.
  */
 DECLINLINE(uint8_t *)vhdBlockBitmapAllocate(PVHDIMAGE pImage)
 {
-    return (uint8_t *)RTMemAllocZ(RT_ALIGN_T(pImage->cbDataBlockBitmap, 4, uint32_t));
+#ifdef RT_ARCH_AMD64
+    return (uint8_t *)RTMemAllocZ(RT_ALIGN_64(pImage->cbDataBlockBitmap, 8));
+#else
+    return (uint8_t *)RTMemAllocZ(RT_ALIGN_32(pImage->cbDataBlockBitmap, 4));
+#endif
 }
 
@@ -917 +921 @@
      * The most signifcant bit stands for a lower sector number.
      */
-    uint8_t  iBitInByte = (32-1) - (cBlockBitmapEntry % 32);
+    uint8_t  iBitInByte = (8-1) - (cBlockBitmapEntry % 8);
     uint8_t *puBitmap = pImage->pu8Bitmap + iBitmap;
 
@@ -937 +941 @@
      * The most signifcant bit stands for a lower sector number.
      */
-    uint8_t  iBitInByte = (32-1) - (cBlockBitmapEntry % 32);
+    uint8_t  iBitInByte = (8-1) - (cBlockBitmapEntry % 8);
     uint8_t  *puBitmap  = pImage->pu8Bitmap + iBitmap;