[librm] Speed up protected-mode calls under KVM

When making a call from real mode to protected mode, we save and
restore the global and interrupt descriptor table registers.  The
restore currently takes place after returning to real mode, which
generates two EXCEPTION_NMIs and corresponding VM exits when running
under KVM on an Intel CPU.

Avoid the VM exits by restoring the descriptor table registers inside
prot_to_real, while still running in protected mode.

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

src/arch/i386/transitions/librm.S

 	/* Return to virtual address */
 	ret
 
-	/* Default real-mode interrupt descriptor table */
-	.section ".data", "aw", @progbits
-rm_idtr:
-	.word 0xffff /* limit */
-	.long 0 /* base */
-
 /****************************************************************************
  * prot_to_real (protected-mode near call, 32-bit real-mode return address)
  *
  *
  * Parameters: 
  *   %ecx : number of bytes to move from PM stack to RM stack
+ *   %esi : real-mode global and interrupt descriptor table registers
  *
  ****************************************************************************
  */
 	.section ".text", "ax", @progbits
 	.code32
 prot_to_real:
+	/* Copy real-mode global descriptor table register to RM code segment */
+	movl	text16, %edi
+	leal	rm_gdtr(%edi), %edi
+	movsw
+	movsl
+
+	/* Load real-mode interrupt descriptor table register */
+	lidt	(%esi)
+
 	/* Add return address to data to be moved to RM stack */
 	addl	$4, %ecx
 	
 	/* Record protected-mode %esp (after removal of data) */
 	movl	%esi, pm_esp
 
-	/* Reset IDTR to the real-mode defaults */
-	lidt	rm_idtr
-
 	/* Load real-mode segment limits */
 	movw	$REAL_DS, %ax
 	movw	%ax, %ds
 	.section ".text16", "ax", @progbits
 	.code16
 p2r_rmode:
+	/* Load real-mode GDT */
+	data32 lgdt %cs:rm_gdtr
 	/* Switch to real mode */
 	movl	%cr0, %eax
 	andb	$0!CR0_PE, %al
 	.globl rm_ds
 rm_ds:	.word 0
 
+	/* Real-mode global and interrupt descriptor table registers */
+	.section ".text16.data", "aw", @progbits
+rm_gdtr:
+	.word 0 /* Limit */
+	.long 0 /* Base */
+
 /****************************************************************************
  * prot_call (real-mode far call, 16-bit real-mode far return address)
  *
  */
 
 #define PC_OFFSET_GDT ( 0 )
-#define PC_OFFSET_IDT ( PC_OFFSET_GDT + 8 /* pad to 8 to keep alignment */ )
-#define PC_OFFSET_IX86 ( PC_OFFSET_IDT + 8 /* pad to 8 to keep alignment */ )
+#define PC_OFFSET_IDT ( PC_OFFSET_GDT + 6 )
+#define PC_OFFSET_IX86 ( PC_OFFSET_IDT + 6 )
 #define PC_OFFSET_RETADDR ( PC_OFFSET_IX86 + SIZEOF_I386_ALL_REGS )
 #define PC_OFFSET_FUNCTION ( PC_OFFSET_RETADDR + 4 )
 #define PC_OFFSET_END ( PC_OFFSET_FUNCTION + 4 )
 	pushw	%ds
 	pushw	%ss
 	pushw	%cs
-	subw	$16, %sp
+	subw	$PC_OFFSET_IX86, %sp
 	movw	%sp, %bp
-	sidt	8(%bp)
-	sgdt	(%bp)
+	sidt	PC_OFFSET_IDT(%bp)
+	sgdt	PC_OFFSET_GDT(%bp)
 
 	/* For sanity's sake, clear the direction flag as soon as possible */
 	cld
 
 	/* Switch to real mode and move register dump back to RM stack */
 	movl	$PC_OFFSET_END, %ecx
+	movl	%esp, %esi
 	pushl	$pc_rmode
 	jmp	prot_to_real
 	.section ".text16", "ax", @progbits
 	.code16
 pc_rmode:
-	/* Reload GDT and IDT, restore registers and flags and return */
-	movw	%sp, %bp
-	data32 lgdt (%bp)
-	data32 lidt 8(%bp)
-	addw	$20, %sp /* also skip %cs and %ss */
+	/* Restore registers and flags and return */
+	addw	$( PC_OFFSET_IX86 + 4 /* also skip %cs and %ss */ ), %sp
 	popw	%ds
 	popw	%es
 	popw	%fs
 	/* Switch to real mode and move register dump to RM stack  */
 	movl	$( RC_OFFSET_RETADDR + 4 /* function pointer copy */ ), %ecx
 	pushl	$rc_rmode
+	movl	$rm_default_gdtr_idtr, %esi
 	jmp	prot_to_real
 	.section ".text16", "ax", @progbits
 	.code16
 	.section ".data16", "aw", @progbits
 rc_function:	.word 0, 0
 
+	/* Default real-mode global and interrupt descriptor table registers */
+	.section ".data", "aw", @progbits
+rm_default_gdtr_idtr:
+	.word 0		/* Global descriptor table limit */
+	.long 0		/* Global descriptor table base */
+	.word 0x03ff	/* Interrupt descriptor table limit */
+	.long 0		/* Interrupt descriptor table base */
+
 /****************************************************************************
  * flatten_real_mode (real-mode near call)
  *