[prefix] Add A20-enabling code in libflat

iPXE currently insists on residing in an even megabyte.  This imposes
undesirably severe constraints upon our PMM allocation strategy, and
limits our options for mechanisms to access ROMs greater than 64kB in
size.

Add A20 handling code to libflat so that prefixes are able to access
memory even in odd megabytes.

The algorithms and tuning parameters in the new A20 handling code are
based upon a mixture of the existing iPXE A20 code and the A20 code
from the 2.6.32 Linux kernel.

Signed-off-by: Michael Brown <mcb30@ipxe.org>

src/arch/i386/prefix/libprefix.S

@@ -512,13 +512,24 @@ install_prealloc:
 
 	/* Open up access to payload */
 #ifndef KEEP_IT_REAL
-	/* Flatten real mode */
+	/* Access high memory */
 	pushw	%cs
 	pushw	$1f
 	pushw	%ax
-	pushw	$flatten_real_mode
+	pushw	$access_highmem
 	lret
-1:
+1:	/* Die if we could not access high memory */
+	jnc	3f
+	movw	$a20_death_message, %si
+	xorw	%di, %di
+	call	print_message
+2:	jmp	2b
+	.section ".prefix.data", "aw", @progbits
+a20_death_message:
+	.asciz	"Gate A20 stuck - cannot continue\n"
+	.size	a20_death_message, . - a20_death_message
+	.previous
+3:
 #endif
 
 	/* Calculate physical address of payload (i.e. first source) */
@@ -570,13 +581,13 @@ install_prealloc:
 	popl	%edx /* discard */
 
 	/* Copy code to new location */
+	pushl	%edi
 	xorw	%ax, %ax
 	movw	%ax, %es
-	movl	%ebp, %edi
 	es rep addr32 movsb
+	popl	%edi
 
 	/* Initialise librm at new location */
-	movl	%ebp, %edi
 	lcall	*init_librm_vector
 #endif
 

src/arch/i386/transitions/libflat.S

@@ -24,7 +24,7 @@ FILE_LICENCE ( GPL2_OR_LATER )
 #define CR0_PE 1
 
 /****************************************************************************
- * flatten_real_mode (real-mode far call)
+ * flatten_real_mode
  *
  * Set up 4GB segment limits
  *
@@ -63,7 +63,6 @@ flatten_saved_gdt:
 
 	.section ".text16.early", "awx", @progbits
 	.code16
-	.globl	flatten_real_mode
 flatten_real_mode:
 	/* Preserve registers and flags */
 	pushfl
@@ -126,7 +125,7 @@ flatten_real_mode:
 	popw	%si
 	popl	%eax
 	popfl
-	lret
+	ret
 	.size flatten_real_mode, . - flatten_real_mode
 
 	.section ".text16.early", "awx", @progbits
@@ -139,3 +138,281 @@ set_seg_base:
 	andb	$0x0f, 4(%si)
 	ret
 	.size set_seg_base, . - set_seg_base
+
+/****************************************************************************
+ * test_a20_short, test_a20_long
+ *
+ * Check to see if A20 line is enabled
+ *
+ * Parameters:
+ *   none
+ * Returns:
+ *   CF set if A20 line is not enabled
+ * Corrupts:
+ *   none
+ ****************************************************************************
+ */
+#define TEST_A20_SHORT_MAX_RETRIES 0x20
+#define TEST_A20_LONG_MAX_RETRIES 0x200000
+	.section ".text16.early", "awx", @progbits
+	.code16
+test_a20_short:
+	pushl	%ecx
+	movl	$TEST_A20_SHORT_MAX_RETRIES, %ecx
+	jmp	1f
+	.size	test_a20_short, . - test_a20_short
+test_a20_long:
+	pushl	%ecx
+	movl	$TEST_A20_LONG_MAX_RETRIES, %ecx
+1:	pushw	%ax
+
+	/* Flatten real mode so we can access the test pattern's 1MB offset */
+	call	flatten_real_mode
+
+2:	/* Modify and check test pattern; succeed if we see a difference */
+	incw	%cs:test_a20_data
+	addr32 movw %cs:(test_a20_data + 0x100000 ), %ax
+	cmpw	%cs:test_a20_data, %ax
+	clc
+	jnz	99f
+
+	/* Delay and retry */
+	outb	%al, $0x80
+	addr32 loop 2b
+	stc
+
+99:	/* Restore registers and return */
+	popw	%ax
+	popl	%ecx
+	ret
+	.size	test_a20_long, . - test_a20_long
+
+	.section ".text16.early.data", "aw", @progbits
+	.align	2
+test_a20_data:
+	.word	0xdead
+	.size	test_a20_data, . - test_a20_data
+
+/****************************************************************************
+ * enable_a20_bios
+ *
+ * Try enabling A20 line via BIOS
+ *
+ * Parameters:
+ *   none
+ * Returns:
+ *   CF set if A20 line is not enabled
+ * Corrupts:
+ *   none
+ ****************************************************************************
+ */
+	.section ".text16.early", "awx", @progbits
+	.code16
+enable_a20_bios:
+	/* Preserve registers */
+	pushw	%ax
+
+	/* Attempt INT 15,2401 */
+	movw	$0x2401, %ax
+	int	$0x15
+	jc	99f
+
+	/* Check that success was really successful */
+	call	test_a20_short
+
+99:	/* Restore registers and return */
+	popw	%ax
+	ret
+	.size	enable_a20_bios, . - enable_a20_bios
+
+/****************************************************************************
+ * enable_a20_kbc
+ *
+ * Try enabling A20 line via keyboard controller
+ *
+ * Parameters:
+ *   none
+ * Returns:
+ *   CF set if A20 line is not enabled
+ * Corrupts:
+ *   none
+ ****************************************************************************
+ */
+#define KC_RDWR		0x60
+#define KC_RDWR_SET_A20		0xdf
+#define	KC_CMD		0x64
+#define KC_CMD_WOUT		0xd1
+#define KC_CMD_NULL		0xff
+#define KC_STATUS	0x64
+#define KC_STATUS_OBUF_FULL	0x01
+#define KC_STATUS_IBUF_FULL	0x02
+#define KC_MAX_RETRIES	100000
+	.section ".text16.early", "awx", @progbits
+	.code16
+enable_a20_kbc:
+	/* Preserve registers */
+	pushw	%ax
+
+	/* Try keyboard controller */
+	call	empty_kbc
+	movb	$KC_CMD_WOUT, %al
+	outb	%al, $KC_CMD
+	call	empty_kbc
+	movb	$KC_RDWR_SET_A20, %al
+	outb	%al, $KC_RDWR
+	call	empty_kbc
+	movb	$KC_CMD_NULL, %al
+	outb	%al, $KC_CMD
+	call	empty_kbc
+
+	/* Check to see if it worked */
+	call	test_a20_long
+
+	/* Restore registers and return */
+	popw	%ax
+	ret
+	.size	enable_a20_kbc, . - enable_a20_kbc
+
+	.section ".text16.early", "awx", @progbits
+	.code16
+empty_kbc:
+	/* Preserve registers */
+	pushl	%ecx
+	pushw	%ax
+
+	/* Wait for KBC to become empty */
+	movl	$KC_MAX_RETRIES, %ecx
+1:	outb	%al, $0x80
+	inb	$KC_STATUS, %al
+	testb	$( KC_STATUS_OBUF_FULL | KC_STATUS_IBUF_FULL ), %al
+	jz	99f
+	testb	$KC_STATUS_OBUF_FULL, %al
+	jz	2f
+	outb	%al, $0x80
+	inb	$KC_RDWR, %al
+2:	addr32 loop 1b
+
+99:	/* Restore registers and return */
+	popw	%ax
+	popl	%ecx
+	ret
+	.size	empty_kbc, . - empty_kbc
+
+/****************************************************************************
+ * enable_a20_fast
+ *
+ * Try enabling A20 line via "Fast Gate A20"
+ *
+ * Parameters:
+ *   none
+ * Returns:
+ *   CF set if A20 line is not enabled
+ * Corrupts:
+ *   none
+ ****************************************************************************
+ */
+#define SCP_A 0x92
+	.section ".text16.early", "awx", @progbits
+	.code16
+enable_a20_fast:
+	/* Preserve registers */
+	pushw	%ax
+
+	/* Try "Fast Gate A20" */
+	inb	$SCP_A, %al
+	orb	$0x02, %al
+	andb	$~0x01, %al
+	outb	%al, $SCP_A
+
+	/* Check to see if it worked */
+	call	test_a20_long
+
+	/* Restore registers and return */
+	popw	%ax
+	ret
+	.size	enable_a20_fast, . - enable_a20_fast
+
+/****************************************************************************
+ * enable_a20
+ *
+ * Try enabling A20 line via any available method
+ *
+ * Parameters:
+ *   none
+ * Returns:
+ *   CF set if A20 line is not enabled
+ * Corrupts:
+ *   none
+ ****************************************************************************
+ */
+#define ENABLE_A20_RETRIES 255
+	.section ".text16.early", "awx", @progbits
+	.code16
+enable_a20:
+	/* Preserve registers */
+	pushl	%ecx
+	pushw	%ax
+
+	/* Check to see if A20 is already enabled */
+	call	test_a20_short
+	jnc	99f
+
+	/* Use known working method, if we have one */
+	movw	%cs:enable_a20_method, %ax
+	testw	%ax, %ax
+	jz	1f
+	call	*%ax
+	jmp	99f
+1:
+	/* Try all methods in turn until one works */
+	movl	$ENABLE_A20_RETRIES, %ecx
+2:	movw	$enable_a20_bios, %ax
+	movw	%ax, %cs:enable_a20_method
+	call	*%ax
+	jnc	99f
+	movw	$enable_a20_kbc, %ax
+	movw	%ax, %cs:enable_a20_method
+	call	*%ax
+	jnc	99f
+	movw	$enable_a20_fast, %ax
+	movw	%ax, %cs:enable_a20_method
+	call	*%ax
+	jnc	99f
+	addr32 loop 2b
+	/* Failure; exit with carry set */
+	movw	$0, %cs:enable_a20_method
+	stc
+
+99:	/* Restore registers and return */
+	popw	%ax
+	popl	%ecx
+	ret
+
+	.section ".text16.early.data", "aw", @progbits
+	.align	2
+enable_a20_method:
+	.word	0
+	.size	enable_a20_method, . - enable_a20_method
+
+/****************************************************************************
+ * access_highmem (real mode far call)
+ *
+ * Open up access to high memory in flat real mode with A20 enabled
+ *
+ * Parameters:
+ *   none
+ * Returns:
+ *   CF set if high memory could not be accessed
+ * Corrupts:
+ *   none
+ ****************************************************************************
+ */
+	.section ".text16.early", "awx", @progbits
+	.code16
+	.globl	access_highmem
+access_highmem:
+	/* Enable A20 line */
+	call	enable_a20
+	/* CPU will be in flat real mode as a result of this call */
+	lret
+	.size	access_highmem, . - access_highmem