/* atomic.S: Move this stuff here for better ICACHE hit rates.
 *
 * Copyright (C) 1996 David S. Miller (davem@caipfs.rutgers.edu)
 */

#include <asm/cprefix.h>
#include <asm/ptrace.h>
#include <asm/psr.h>

	.text
	.align	4

#ifndef __SMP__
	.globl	___xchg32_sun4c
___xchg32_sun4c:
	rd	%psr, %g3
	andcc	%g3, PSR_PIL, %g0
	bne	1f
	 nop
	wr	%g3, PSR_PIL, %psr
	nop; nop; nop
1:
	andcc	%g3, PSR_PIL, %g0
	ld	[%g1], %g7
	bne	1f
	 st	%g2, [%g1]
	wr	%g3, 0x0, %psr
	nop; nop; nop
1:
	mov	%g7, %g2
	jmpl	%o7 + 8, %g0
	 mov	%g4, %o7

	.globl	___xchg32_sun4md
___xchg32_sun4md:
	swap	[%g1], %g2
	jmpl	%o7 + 8, %g0
	 mov	%g4, %o7
#endif

	/* Read asm-sparc/atomic.h carefully to understand how this works for SMP.
	 * Really, some things here for SMP are overly clever, go read the header.
	 */
	.globl	___atomic_add
___atomic_add:
	rd	%psr, %g3		! Keep the code small, old way was stupid
	nop; nop; nop;			! Let the bits set
	or	%g3, PSR_PIL, %g7	! Disable interrupts
	wr	%g7, 0x0, %psr		! Set %psr
	nop; nop; nop;			! Let the bits set
#ifdef __SMP__
1:	ldstub	[%g1 + 3], %g7		! Spin on the byte lock for SMP.
	orcc	%g7, 0x0, %g0		! Did we get it?
	bne	1b			! Nope...
	 ld	[%g1], %g7		! Load locked atomic_t
	sra	%g7, 8, %g7		! Get signed 24-bit integer
	add	%g7, %g2, %g2		! Add in argument
	sll	%g2, 8, %g7		! Transpose back to atomic_t
	st	%g7, [%g1]		! Clever: This releases the lock as well.
#else
	ld	[%g1], %g7		! Load locked atomic_t
	add	%g7, %g2, %g2		! Add in argument
	st	%g2, [%g1]		! Store it back
#endif
	wr	%g3, 0x0, %psr		! Restore original PSR_PIL
	nop; nop; nop;			! Let the bits set
	jmpl	%o7, %g0		! NOTE: not + 8, see callers in atomic.h
	 mov	%g4, %o7		! Restore %o7

	.globl	___atomic_sub
___atomic_sub:
	rd	%psr, %g3		! Keep the code small, old way was stupid
	nop; nop; nop;			! Let the bits set
	or	%g3, PSR_PIL, %g7	! Disable interrupts
	wr	%g7, 0x0, %psr		! Set %psr
	nop; nop; nop;			! Let the bits set
#ifdef __SMP__
1:	ldstub	[%g1 + 3], %g7		! Spin on the byte lock for SMP.
	orcc	%g7, 0x0, %g0		! Did we get it?
	bne	1b			! Nope...
	 ld	[%g1], %g7		! Load locked atomic_t
	sra	%g7, 8, %g7		! Get signed 24-bit integer
	sub	%g7, %g2, %g2		! Subtract argument
	sll	%g2, 8, %g7		! Transpose back to atomic_t
	st	%g7, [%g1]		! Clever: This releases the lock as well
#else
	ld	[%g1], %g7		! Load locked atomic_t
	sub	%g7, %g2, %g2		! Subtract argument
	st	%g2, [%g1]		! Store it back
#endif
	wr	%g3, 0x0, %psr		! Restore original PSR_PIL
	nop; nop; nop;			! Let the bits set
	jmpl	%o7, %g0		! NOTE: not + 8, see callers in atomic.h
	 mov	%g4, %o7		! Restore %o7