ACPICA: FADT: Favor 32-bit register addresses for compatibility
authorBob Moore <robert.moore@intel.com>
Thu, 19 Mar 2009 02:12:13 +0000 (10:12 +0800)
committerLen Brown <len.brown@intel.com>
Fri, 27 Mar 2009 16:11:05 +0000 (12:11 -0400)
Use the 32-bit register addresses whenever they are non-zero. This
means that the 32-bit addresses are favored over the 64-bit
(GAS) addresses. The 64-bit addresses are only used if the 32-bit
addresses are zero. This change provides compatibility with all
versions of Windows. The worst case that this solves is when both
the 32-bit and 64-bit addresses are non-zero, but only the 32-bit
addresses are actually valid. This appears to happen in some
BIOSes because in this case, Windows uses the 32-bit addresses.

Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
drivers/acpi/acpica/tbfadt.c

index ff89cfee0e7edb2b2263b5f8c656f986006d05fb..f87bfb259ef6b2c4a964a14b3154a53279e98875 100644 (file)
@@ -320,29 +320,35 @@ void acpi_tb_create_local_fadt(struct acpi_table_header *table, u32 length)
  * RETURN:      None
  *
  * DESCRIPTION: Converts all versions of the FADT to a common internal format.
- *              Expand all 32-bit addresses to 64-bit.
+ *              Expand 32-bit addresses to 64-bit as necessary.
  *
  * NOTE:        acpi_gbl_FADT must be of size (struct acpi_table_fadt),
  *              and must contain a copy of the actual FADT.
  *
- * ACPICA will use the "X" fields of the FADT for all addresses.
+ * Notes on 64-bit register addresses:
  *
- * "X" fields are optional extensions to the original V1.0 fields. Even if
- * they are present in the structure, they can be optionally not used by
- * setting them to zero. Therefore, we must selectively expand V1.0 fields
- * if the corresponding X field is zero.
+ * After this FADT conversion, later ACPICA code will only use the 64-bit "X"
+ * fields of the FADT for all ACPI register addresses.
  *
- * For ACPI 1.0 FADTs, all address fields are expanded to the corresponding
- * "X" fields.
+ * The 64-bit "X" fields are optional extensions to the original 32-bit FADT
+ * V1.0 fields. Even if they are present in the FADT, they are optional and
+ * are unused if the BIOS sets them to zero. Therefore, we must copy/expand
+ * 32-bit V1.0 fields if the corresponding X field is zero.
  *
- * For ACPI 2.0 FADTs, any "X" fields that are NULL are filled in by
- * expanding the corresponding ACPI 1.0 field.
+ * For ACPI 1.0 FADTs, all 32-bit address fields are expanded to the
+ * corresponding "X" fields in the internal FADT.
+ *
+ * For ACPI 2.0+ FADTs, all valid (non-zero) 32-bit address fields are expanded
+ * to the corresponding 64-bit X fields. For compatibility with other ACPI
+ * implementations, we ignore the 64-bit field if the 32-bit field is valid,
+ * regardless of whether the host OS is 32-bit or 64-bit.
  *
  ******************************************************************************/
 
 static void acpi_tb_convert_fadt(void)
 {
-       struct acpi_generic_address *target64;
+       struct acpi_generic_address *address64;
+       u32 address32;
        u32 i;
 
        /* Update the local FADT table header length */
@@ -391,29 +397,51 @@ static void acpi_tb_convert_fadt(void)
         * Expand the ACPI 1.0 32-bit addresses to the ACPI 2.0 64-bit "X"
         * generic address structures as necessary. Later code will always use
         * the 64-bit address structures.
+        *
+        * March 2009:
+        * We now always use the 32-bit address if it is valid (non-null). This
+        * is not in accordance with the ACPI specification which states that
+        * the 64-bit address supersedes the 32-bit version, but we do this for
+        * compatibility with other ACPI implementations. Most notably, in the
+        * case where both the 32 and 64 versions are non-null, we use the 32-bit
+        * version. This is the only address that is guaranteed to have been
+        * tested by the BIOS manufacturer.
         */
        for (i = 0; i < ACPI_FADT_INFO_ENTRIES; i++) {
-               target64 =
-                   ACPI_ADD_PTR(struct acpi_generic_address, &acpi_gbl_FADT,
-                                fadt_info_table[i].address64);
+               address32 = *ACPI_ADD_PTR(u32,
+                                         &acpi_gbl_FADT,
+                                         fadt_info_table[i].address32);
 
-               /* Expand only if the 64-bit X target is null */
+               address64 = ACPI_ADD_PTR(struct acpi_generic_address,
+                                        &acpi_gbl_FADT,
+                                        fadt_info_table[i].address64);
 
-               if (!target64->address) {
+               /*
+                * If both 32- and 64-bit addresses are valid (non-zero),
+                * they must match.
+                */
+               if (address64->address && address32 &&
+                   (address64->address != (u64) address32)) {
+                       ACPI_ERROR((AE_INFO,
+                                   "32/64X address mismatch in %s: %8.8X/%8.8X%8.8X, using 32",
+                                   fadt_info_table[i].name, address32,
+                                   ACPI_FORMAT_UINT64(address64->address)));
+               }
 
-                       /* The space_id is always I/O for the 32-bit legacy address fields */
+               /* Always use 32-bit address if it is valid (non-null) */
 
-                       acpi_tb_init_generic_address(target64,
+               if (address32) {
+                       /*
+                        * Copy the 32-bit address to the 64-bit GAS structure. The
+                        * Space ID is always I/O for 32-bit legacy address fields
+                       */
+                       acpi_tb_init_generic_address(address64,
                                                     ACPI_ADR_SPACE_SYSTEM_IO,
                                                     *ACPI_ADD_PTR(u8,
                                                                   &acpi_gbl_FADT,
                                                                   fadt_info_table
                                                                   [i].length),
-                                                    (u64) * ACPI_ADD_PTR(u32,
-                                                                         &acpi_gbl_FADT,
-                                                                         fadt_info_table
-                                                                         [i].
-                                                                         address32));
+                                                    address32);
                }
        }
 }
@@ -530,18 +558,6 @@ static void acpi_tb_validate_fadt(void)
                                              length));
                        }
                }
-
-               /*
-                * If both 32- and 64-bit addresses are valid (non-zero),
-                * they must match
-                */
-               if (address64->address && *address32 &&
-                   (address64->address != (u64) * address32)) {
-                       ACPI_ERROR((AE_INFO,
-                                   "32/64X address mismatch in %s: %8.8X/%8.8X%8.8X, using 64X",
-                                   name, *address32,
-                                   ACPI_FORMAT_UINT64(address64->address)));
-               }
        }
 }