x86: Add support for running option ROMs natively

On x86 machines we can use an emulator to run option ROMS as with other
architectures. But with some additional effort (mostly due to the 16-bit
nature of option ROMs) we can run them natively. Add support for this.

Signed-off-by: Simon Glass <sjg@chromium.org>
diff --git a/arch/x86/lib/bios_asm.S b/arch/x86/lib/bios_asm.S
new file mode 100644
index 0000000..4faa70e
--- /dev/null
+++ b/arch/x86/lib/bios_asm.S
@@ -0,0 +1,281 @@
+/*
+ * From coreboot x86_asm.S, cleaned up substantially
+ *
+ * Copyright (C) 2009-2010 coresystems GmbH
+ *
+ * SPDX-License-Identifier:	GPL-2.0
+ */
+
+#include <asm/processor.h>
+#include <asm/processor-flags.h>
+#include "bios.h"
+
+#define SEG(segment)	$segment * X86_GDT_ENTRY_SIZE
+
+/*
+ * This is the interrupt handler stub code. It gets copied to the IDT and
+ * to some fixed addresses in the F segment. Before the code can used,
+ * it gets patched up by the C function copying it: byte 3 (the $0 in
+ * movb $0, %al) is overwritten with the interrupt numbers.
+ */
+
+	.code16
+	.globl __idt_handler
+__idt_handler:
+	pushal
+	movb 	$0, %al /* This instruction gets modified */
+	ljmp 	$0, $__interrupt_handler_16bit
+	.globl __idt_handler_size
+__idt_handler_size:
+	.long  . - __idt_handler
+
+.macro setup_registers
+	/* initial register values */
+	movl	44(%ebp), %eax
+	movl	%eax, __registers +  0 /* eax */
+	movl	48(%ebp), %eax
+	movl	%eax, __registers +  4 /* ebx */
+	movl	52(%ebp), %eax
+	movl	%eax, __registers +  8 /* ecx */
+	movl	56(%ebp), %eax
+	movl	%eax, __registers + 12 /* edx */
+	movl	60(%ebp), %eax
+	movl	%eax, __registers + 16 /* esi */
+	movl	64(%ebp), %eax
+	movl	%eax, __registers + 20 /* edi */
+.endm
+
+.macro	enter_real_mode
+	/* Activate the right segment descriptor real mode. */
+	ljmp	SEG(X86_GDT_ENTRY_16BIT_CS), $PTR_TO_REAL_MODE(1f)
+1:
+.code16
+	/*
+	 * Load the segment registers with properly configured segment
+	 * descriptors. They will retain these configurations (limits,
+	 * writability, etc.) once protected mode is turned off.
+	 */
+	mov	SEG(X86_GDT_ENTRY_16BIT_DS), %ax
+	mov	%ax, %ds
+	mov	%ax, %es
+	mov	%ax, %fs
+	mov	%ax, %gs
+	mov	%ax, %ss
+
+	/* Turn off protection */
+	movl	%cr0, %eax
+	andl	$~X86_CR0_PE, %eax
+	movl	%eax, %cr0
+
+	/* Now really going into real mode */
+	ljmp	$0, $PTR_TO_REAL_MODE(1f)
+1:
+	/*
+	 * Set up a stack: Put the stack at the end of page zero. That way
+	 * we can easily share it between real and protected, since the
+	 * 16-bit ESP at segment 0 will work for any case.
+	 */
+	mov	$0x0, %ax
+	mov	%ax, %ss
+
+	/* Load 16 bit IDT */
+	xor	%ax, %ax
+	mov	%ax, %ds
+	lidt	__realmode_idt
+
+.endm
+
+.macro	prepare_for_irom
+	movl	$0x1000, %eax
+	movl	%eax, %esp
+
+	/* Initialise registers for option rom lcall */
+	movl	__registers +  0, %eax
+	movl	__registers +  4, %ebx
+	movl	__registers +  8, %ecx
+	movl	__registers + 12, %edx
+	movl	__registers + 16, %esi
+	movl	__registers + 20, %edi
+
+	/* Set all segments to 0x0000, ds to 0x0040 */
+	push	%ax
+	xor	%ax, %ax
+	mov	%ax, %es
+	mov	%ax, %fs
+	mov	%ax, %gs
+	mov	SEG(X86_GDT_ENTRY_16BIT_FLAT_DS), %ax
+	mov	%ax, %ds
+	pop	%ax
+
+.endm
+
+.macro	enter_protected_mode
+	/* Go back to protected mode */
+	movl	%cr0, %eax
+	orl	$X86_CR0_PE, %eax
+	movl	%eax, %cr0
+
+	/* Now that we are in protected mode jump to a 32 bit code segment */
+	data32	ljmp	SEG(X86_GDT_ENTRY_32BIT_CS), $PTR_TO_REAL_MODE(1f)
+1:
+	.code32
+	mov	SEG(X86_GDT_ENTRY_32BIT_DS), %ax
+	mov	%ax, %ds
+	mov	%ax, %es
+	mov	%ax, %gs
+	mov	%ax, %ss
+	mov	SEG(X86_GDT_ENTRY_32BIT_FS), %ax
+	mov	%ax, %fs
+
+	/* restore proper idt */
+	lidt	idt_ptr
+.endm
+
+/*
+ * In order to be independent of U-Boot's position in RAM we relocate a part
+ * of the code to the first megabyte of RAM, so the CPU can use it in
+ * real-mode. This code lives at asm_realmode_code.
+ */
+	.globl asm_realmode_code
+asm_realmode_code:
+
+/* Realmode IDT pointer structure. */
+__realmode_idt = PTR_TO_REAL_MODE(.)
+	.word 1023	/* 16 bit limit */
+	.long 0		/* 24 bit base */
+	.word 0
+
+/* Preserve old stack */
+__stack = PTR_TO_REAL_MODE(.)
+	.long 0
+
+/* Register store for realmode_call and realmode_interrupt */
+__registers = PTR_TO_REAL_MODE(.)
+	.long 0 /*  0 - EAX */
+	.long 0 /*  4 - EBX */
+	.long 0 /*  8 - ECX */
+	.long 0 /* 12 - EDX */
+	.long 0 /* 16 - ESI */
+	.long 0 /* 20 - EDI */
+
+/* 256 byte buffer, used by int10 */
+	.globl asm_realmode_buffer
+asm_realmode_buffer:
+	.skip 256
+
+	.code32
+	.globl asm_realmode_call
+asm_realmode_call:
+	/* save all registers to the stack */
+	pusha
+	pushf
+	movl	%esp, __stack
+	movl	%esp, %ebp
+
+	/*
+	 * This function is called with regparm=0 and we have to skip the
+	 * 36 bytes from pushf+pusha. Hence start at 40.
+	 * Set up our call instruction.
+	 */
+	movl	40(%ebp), %eax
+	mov	%ax, __lcall_instr + 1
+	andl	$0xffff0000, %eax
+	shrl	$4, %eax
+	mov	%ax, __lcall_instr + 3
+
+	wbinvd
+
+	setup_registers
+	enter_real_mode
+	prepare_for_irom
+
+__lcall_instr = PTR_TO_REAL_MODE(.)
+	.byte 0x9a
+	.word 0x0000, 0x0000
+
+	enter_protected_mode
+
+	/* restore stack pointer, eflags and register values and exit */
+	movl	__stack, %esp
+	popf
+	popa
+	ret
+
+	.globl __realmode_interrupt
+__realmode_interrupt:
+	/* save all registers to the stack and store the stack pointer */
+	pusha
+	pushf
+	movl	%esp, __stack
+	movl	%esp, %ebp
+
+	/*
+	 * This function is called with regparm=0 and we have to skip the
+	 * 36 bytes from pushf+pusha. Hence start at 40.
+	 * Prepare interrupt calling code.
+	 */
+	movl	40(%ebp), %eax
+	movb	%al, __intXX_instr + 1 /* intno */
+
+	setup_registers
+	enter_real_mode
+	prepare_for_irom
+
+__intXX_instr = PTR_TO_REAL_MODE(.)
+	.byte 0xcd, 0x00 /* This becomes intXX */
+
+	enter_protected_mode
+
+	/* restore stack pointer, eflags and register values and exit */
+	movl	__stack, %esp
+	popf
+	popa
+	ret
+
+/*
+ * This is the 16-bit interrupt entry point called by the IDT stub code.
+ *
+ * Before this code code is called, %eax is pushed to the stack, and the
+ * interrupt number is loaded into %al. On return this function cleans up
+ * for its caller.
+ */
+	.code16
+__interrupt_handler_16bit = PTR_TO_REAL_MODE(.)
+	push	%ds
+	push	%es
+	push	%fs
+	push	%gs
+
+	/* Clear DF to not break ABI assumptions */
+	cld
+
+	/*
+	 * Clean up the interrupt number. We could do this in the stub, but
+	 * it would cost two more bytes per stub entry.
+	 */
+	andl	$0xff, %eax
+	pushl	%eax		/* ... and make it the first parameter */
+
+	enter_protected_mode
+
+	/* Call the C interrupt handler */
+	movl	$interrupt_handler, %eax
+	call	*%eax
+
+	enter_real_mode
+
+	/*
+	 * Restore all registers, including those manipulated by the C
+	 * handler
+	 */
+	popl	%eax
+	pop	%gs
+	pop	%fs
+	pop	%es
+	pop	%ds
+	popal
+	iret
+
+	.globl asm_realmode_code_size
+asm_realmode_code_size:
+	.long  . - asm_realmode_code