ACPICA: Fix regression in FADT revision checks

commit 3e80acd1af upstream.

	commit 64b3db22c0 (2.6.39),
"Remove use of unreliable FADT revision field" causes regression
for old P4 systems because now cst_control and other fields are
not reset to 0.

	The effect is that acpi_processor_power_init will notice
cst_control != 0 and a write to CST_CNT register is performed
that should not happen. As result, the system oopses after the
"No _CST, giving up" message, sometimes in acpi_ns_internalize_name,
sometimes in acpi_ns_get_type, usually at random places. May be
during migration to CPU 1 in acpi_processor_get_throttling.

	Every one of these settings help to avoid this problem:
 - acpi=off
 - processor.nocst=1
 - maxcpus=1

	The fix is to update acpi_gbl_FADT.header.length after
the original value is used to check for old revisions.

https://bugzilla.kernel.org/show_bug.cgi?id=42700
https://bugzilla.redhat.com/show_bug.cgi?id=727865

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Cc: Jonathan Nieder <jrnieder@gmail.com>
Cc: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/acpi/acpica/tbfadt.c

@@ -350,10 +350,6 @@ static void acpi_tb_convert_fadt(void)
 	u32 address32;
 	u32 i;
 
-	/* Update the local FADT table header length */
-
-	acpi_gbl_FADT.header.length = sizeof(struct acpi_table_fadt);
-
 	/*
 	 * Expand the 32-bit FACS and DSDT addresses to 64-bit as necessary.
 	 * Later code will always use the X 64-bit field. Also, check for an
@@ -395,6 +391,10 @@ static void acpi_tb_convert_fadt(void)
 		acpi_gbl_FADT.boot_flags = 0;
 	}
 
+	/* Update the local FADT table header length */
+
+	acpi_gbl_FADT.header.length = sizeof(struct acpi_table_fadt);
+
 	/*
 	 * 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