[a20] Send a null command to the KBC after changing A20
authorH. Peter Anvin <hpa@zytor.com>
Fri, 27 Jun 2008 21:29:13 +0000 (22:29 +0100)
committerMichael Brown <mcb30@etherboot.org>
Fri, 27 Jun 2008 21:29:13 +0000 (22:29 +0100)
Send a null command, specifically "pulse outputs" with no outputs
selected, to the KBC after changing A20.  This was apparently done by DOS,
presumably as a synchronization hack, and the authors of the UHCI spec
thought it was inherent.  Therefore, there are systems out there (e.g. HP
DL360 G5) which will stop responsing to "legacy USB" unless they see the
null command, 0xFF, written to port 0x64 at the end of the A20 toggling
sequence.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
src/arch/i386/firmware/pcbios/gateA20.c

index 2caac89..a14e341 100644 (file)
@@ -12,6 +12,7 @@
 
 #define KC_CMD_WIN     0xd0            /* read  output port */
 #define KC_CMD_WOUT    0xd1            /* write output port */
+#define KC_CMD_NULL    0xff            /* null command ("pulse nothing") */
 #define KB_SET_A20     0xdf            /* enable A20,
                                           enable output buffer full interrupt
                                           enable data line
@@ -130,6 +131,8 @@ void gateA20_set ( void ) {
                        empty_8042();
                        outb ( KB_SET_A20, K_RDWR );
                        empty_8042();
+                       outb ( KC_CMD_NULL, K_CMD );
+                       empty_8042();
                        if ( gateA20_is_set ( A20_KBC_RETRIES ) ) {
                                DBG ( "Enabled gate A20 using "
                                      "keyboard controller\n" );