[POWERPC] Fix MMIO ops to provide expected barrier behaviour

This changes the writeX family of functions to have a sync instruction
before the MMIO store rather than after, because the generally expected
behaviour is that the device receiving the MMIO store can be guaranteed
to see the effects of any preceding writes to normal memory.

To preserve ordering between writeX and readX, and to preserve ordering
between preceding stores and the readX, the readX family of functions
have had an sync added before the load.

Although writeX followed by spin_unlock is not officially guaranteed
to keep the writeX inside the spin-locked region unless an mmiowb()
is used, there are currently drivers that depend on the previous
behaviour on powerpc, which was that the mmiowb wasn't actually required.
Therefore we have a per-cpu flag that is set by writeX, cleared by
__raw_spin_lock and mmiowb, and tested by __raw_spin_unlock.  If it is
set, __raw_spin_unlock does a sync and clears it.

This changes both 32-bit and 64-bit readX/writeX.  32-bit already has a
sync in __raw_spin_unlock (since lwsync doesn't exist on 32-bit), and thus
doesn't need the per-cpu flag.

Tested on G5 (PPC970) and POWER5.

Signed-off-by: Paul Mackerras <paulus@samba.org>

arch/powerpc/kernel/misc.S

@@ -17,15 +17,6 @@
 
 	.text
 
-#ifdef CONFIG_PPC64
-#define IN_SYNC		twi	0,r5,0; isync
-#define EIEIO_32
-#define SYNC_64		sync
-#else /* CONFIG_PPC32 */
-#define IN_SYNC
-#define EIEIO_32	eieio
-#define SYNC_64
-#endif
 /*
  * Returns (address we are running at) - (address we were linked at)
  * for use before the text and data are mapped to KERNELBASE.
@@ -70,6 +61,7 @@ _GLOBAL(add_reloc_offset)
  * The *_ns versions don't do byte-swapping.
  */
 _GLOBAL(_insb)
+	sync
 	cmpwi	0,r5,0
 	mtctr	r5
 	subi	r4,r4,1
@@ -78,7 +70,8 @@ _GLOBAL(_insb)
 	eieio
 	stbu	r5,1(r4)
 	bdnz	00b
-	IN_SYNC
+	twi	0,r5,0
+	isync
 	blr
 
 _GLOBAL(_outsb)
@@ -86,14 +79,15 @@ _GLOBAL(_outsb)
 	mtctr	r5
 	subi	r4,r4,1
 	blelr-
+	sync
 00:	lbzu	r5,1(r4)
 	stb	r5,0(r3)
-	EIEIO_32
 	bdnz	00b
-	SYNC_64
+	sync
 	blr
 
 _GLOBAL(_insw)
+	sync
 	cmpwi	0,r5,0
 	mtctr	r5
 	subi	r4,r4,2
@@ -102,7 +96,8 @@ _GLOBAL(_insw)
 	eieio
 	sthu	r5,2(r4)
 	bdnz	00b
-	IN_SYNC
+	twi	0,r5,0
+	isync
 	blr
 
 _GLOBAL(_outsw)
@@ -110,14 +105,15 @@ _GLOBAL(_outsw)
 	mtctr	r5
 	subi	r4,r4,2
 	blelr-
+	sync
 00:	lhzu	r5,2(r4)
-	EIEIO_32
 	sthbrx	r5,0,r3
 	bdnz	00b
-	SYNC_64
+	sync
 	blr
 
 _GLOBAL(_insl)
+	sync
 	cmpwi	0,r5,0
 	mtctr	r5
 	subi	r4,r4,4
@@ -126,7 +122,8 @@ _GLOBAL(_insl)
 	eieio
 	stwu	r5,4(r4)
 	bdnz	00b
-	IN_SYNC
+	twi	0,r5,0
+	isync
 	blr
 
 _GLOBAL(_outsl)
@@ -134,17 +131,18 @@ _GLOBAL(_outsl)
 	mtctr	r5
 	subi	r4,r4,4
 	blelr-
+	sync
 00:	lwzu	r5,4(r4)
 	stwbrx	r5,0,r3
-	EIEIO_32
 	bdnz	00b
-	SYNC_64
+	sync
 	blr
 
 #ifdef CONFIG_PPC32
 _GLOBAL(__ide_mm_insw)
 #endif
 _GLOBAL(_insw_ns)
+	sync
 	cmpwi	0,r5,0
 	mtctr	r5
 	subi	r4,r4,2
@@ -153,7 +151,8 @@ _GLOBAL(_insw_ns)
 	eieio
 	sthu	r5,2(r4)
 	bdnz	00b
-	IN_SYNC
+	twi	0,r5,0
+	isync
 	blr
 
 #ifdef CONFIG_PPC32
@@ -164,17 +163,18 @@ _GLOBAL(_outsw_ns)
 	mtctr	r5
 	subi	r4,r4,2
 	blelr-
+	sync
 00:	lhzu	r5,2(r4)
 	sth	r5,0(r3)
-	EIEIO_32
 	bdnz	00b
-	SYNC_64
+	sync
 	blr
 
 #ifdef CONFIG_PPC32
 _GLOBAL(__ide_mm_insl)
 #endif
 _GLOBAL(_insl_ns)
+	sync
 	cmpwi	0,r5,0
 	mtctr	r5
 	subi	r4,r4,4
@@ -183,7 +183,8 @@ _GLOBAL(_insl_ns)
 	eieio
 	stwu	r5,4(r4)
 	bdnz	00b
-	IN_SYNC
+	twi	0,r5,0
+	isync
 	blr
 
 #ifdef CONFIG_PPC32
@@ -194,10 +195,10 @@ _GLOBAL(_outsl_ns)
 	mtctr	r5
 	subi	r4,r4,4
 	blelr-
+	sync
 00:	lwzu	r5,4(r4)
 	stw	r5,0(r3)
-	EIEIO_32
 	bdnz	00b
-	SYNC_64
+	sync
 	blr
 

include/asm-powerpc/eeh.h

@@ -205,6 +205,7 @@ static inline void eeh_memset_io(volatile void __iomem *addr, int c,
 	lc |= lc << 8;
 	lc |= lc << 16;
 
+	__asm__ __volatile__ ("sync" : : : "memory");
 	while(n && !EEH_CHECK_ALIGN(p, 4)) {
 		*((volatile u8 *)p) = c;
 		p++;
@@ -229,6 +230,7 @@ static inline void eeh_memcpy_fromio(void *dest, const volatile void __iomem *sr
 	void *destsave = dest;
 	unsigned long nsave = n;
 
+	__asm__ __volatile__ ("sync" : : : "memory");
 	while(n && (!EEH_CHECK_ALIGN(vsrc, 4) || !EEH_CHECK_ALIGN(dest, 4))) {
 		*((u8 *)dest) = *((volatile u8 *)vsrc);
 		__asm__ __volatile__ ("eieio" : : : "memory");
@@ -266,6 +268,7 @@ static inline void eeh_memcpy_toio(volatile void __iomem *dest, const void *src,
 {
 	void *vdest = (void __force *) dest;
 
+	__asm__ __volatile__ ("sync" : : : "memory");
 	while(n && (!EEH_CHECK_ALIGN(vdest, 4) || !EEH_CHECK_ALIGN(src, 4))) {
 		*((volatile u8 *)vdest) = *((u8 *)src);
 		src++;

include/asm-powerpc/io.h

@@ -19,6 +19,7 @@ extern int check_legacy_ioport(unsigned long base_port);
 #include <linux/compiler.h>
 #include <asm/page.h>
 #include <asm/byteorder.h>
+#include <asm/paca.h>
 #ifdef CONFIG_PPC_ISERIES 
 #include <asm/iseries/iseries_io.h>
 #endif  
@@ -162,7 +163,11 @@ extern void _outsw_ns(volatile u16 __iomem *port, const void *buf, int ns);
 extern void _insl_ns(volatile u32 __iomem *port, void *buf, int nl);
 extern void _outsl_ns(volatile u32 __iomem *port, const void *buf, int nl);
 
-#define mmiowb()
+static inline void mmiowb(void)
+{
+	__asm__ __volatile__ ("sync" : : : "memory");
+	get_paca()->io_sync = 0;
+}
 
 /*
  * output pause versions need a delay at least for the
@@ -278,22 +283,23 @@ static inline int in_8(const volatile unsigned char __iomem *addr)
 {
 	int ret;
 
-	__asm__ __volatile__("lbz%U1%X1 %0,%1; twi 0,%0,0; isync"
+	__asm__ __volatile__("sync; lbz%U1%X1 %0,%1; twi 0,%0,0; isync"
 			     : "=r" (ret) : "m" (*addr));
 	return ret;
 }
 
 static inline void out_8(volatile unsigned char __iomem *addr, int val)
 {
-	__asm__ __volatile__("stb%U0%X0 %1,%0; sync"
+	__asm__ __volatile__("sync; stb%U0%X0 %1,%0"
 			     : "=m" (*addr) : "r" (val));
+	get_paca()->io_sync = 1;
 }
 
 static inline int in_le16(const volatile unsigned short __iomem *addr)
 {
 	int ret;
 
-	__asm__ __volatile__("lhbrx %0,0,%1; twi 0,%0,0; isync"
+	__asm__ __volatile__("sync; lhbrx %0,0,%1; twi 0,%0,0; isync"
 			     : "=r" (ret) : "r" (addr), "m" (*addr));
 	return ret;
 }
@@ -302,28 +308,30 @@ static inline int in_be16(const volatile unsigned short __iomem *addr)
 {
 	int ret;
 
-	__asm__ __volatile__("lhz%U1%X1 %0,%1; twi 0,%0,0; isync"
+	__asm__ __volatile__("sync; lhz%U1%X1 %0,%1; twi 0,%0,0; isync"
 			     : "=r" (ret) : "m" (*addr));
 	return ret;
 }
 
 static inline void out_le16(volatile unsigned short __iomem *addr, int val)
 {
-	__asm__ __volatile__("sthbrx %1,0,%2; sync"
+	__asm__ __volatile__("sync; sthbrx %1,0,%2"
 			     : "=m" (*addr) : "r" (val), "r" (addr));
+	get_paca()->io_sync = 1;
 }
 
 static inline void out_be16(volatile unsigned short __iomem *addr, int val)
 {
-	__asm__ __volatile__("sth%U0%X0 %1,%0; sync"
+	__asm__ __volatile__("sync; sth%U0%X0 %1,%0"
 			     : "=m" (*addr) : "r" (val));
+	get_paca()->io_sync = 1;
 }
 
 static inline unsigned in_le32(const volatile unsigned __iomem *addr)
 {
 	unsigned ret;
 
-	__asm__ __volatile__("lwbrx %0,0,%1; twi 0,%0,0; isync"
+	__asm__ __volatile__("sync; lwbrx %0,0,%1; twi 0,%0,0; isync"
 			     : "=r" (ret) : "r" (addr), "m" (*addr));
 	return ret;
 }
@@ -332,21 +340,23 @@ static inline unsigned in_be32(const volatile unsigned __iomem *addr)
 {
 	unsigned ret;
 
-	__asm__ __volatile__("lwz%U1%X1 %0,%1; twi 0,%0,0; isync"
+	__asm__ __volatile__("sync; lwz%U1%X1 %0,%1; twi 0,%0,0; isync"
 			     : "=r" (ret) : "m" (*addr));
 	return ret;
 }
 
 static inline void out_le32(volatile unsigned __iomem *addr, int val)
 {
-	__asm__ __volatile__("stwbrx %1,0,%2; sync" : "=m" (*addr)
+	__asm__ __volatile__("sync; stwbrx %1,0,%2" : "=m" (*addr)
 			     : "r" (val), "r" (addr));
+	get_paca()->io_sync = 1;
 }
 
 static inline void out_be32(volatile unsigned __iomem *addr, int val)
 {
-	__asm__ __volatile__("stw%U0%X0 %1,%0; sync"
+	__asm__ __volatile__("sync; stw%U0%X0 %1,%0"
 			     : "=m" (*addr) : "r" (val));
+	get_paca()->io_sync = 1;
 }
 
 static inline unsigned long in_le64(const volatile unsigned long __iomem *addr)
@@ -354,6 +364,7 @@ static inline unsigned long in_le64(const volatile unsigned long __iomem *addr)
 	unsigned long tmp, ret;
 
 	__asm__ __volatile__(
+			     "sync\n"
 			     "ld %1,0(%2)\n"
 			     "twi 0,%1,0\n"
 			     "isync\n"
@@ -372,7 +383,7 @@ static inline unsigned long in_be64(const volatile unsigned long __iomem *addr)
 {
 	unsigned long ret;
 
-	__asm__ __volatile__("ld%U1%X1 %0,%1; twi 0,%0,0; isync"
+	__asm__ __volatile__("sync; ld%U1%X1 %0,%1; twi 0,%0,0; isync"
 			     : "=r" (ret) : "m" (*addr));
 	return ret;
 }
@@ -389,14 +400,16 @@ static inline void out_le64(volatile unsigned long __iomem *addr, unsigned long
 			     "rldicl %1,%1,32,0\n"
 			     "rlwimi %0,%1,8,8,31\n"
 			     "rlwimi %0,%1,24,16,23\n"
-			     "std %0,0(%3)\n"
+			     "sync\n"
-			     "sync"
+			     "std %0,0(%3)"
 			     : "=&r" (tmp) , "=&r" (val) : "1" (val) , "b" (addr) , "m" (*addr));
+	get_paca()->io_sync = 1;
 }
 
 static inline void out_be64(volatile unsigned long __iomem *addr, unsigned long val)
 {
-	__asm__ __volatile__("std%U0%X0 %1,%0; sync" : "=m" (*addr) : "r" (val));
+	__asm__ __volatile__("sync; std%U0%X0 %1,%0" : "=m" (*addr) : "r" (val));
+	get_paca()->io_sync = 1;
 }
 
 #ifndef CONFIG_PPC_ISERIES 

include/asm-powerpc/paca.h

@@ -93,6 +93,7 @@ struct paca_struct {
 	u64 saved_r1;			/* r1 save for RTAS calls */
 	u64 saved_msr;			/* MSR saved here by enter_rtas */
 	u8 proc_enabled;		/* irq soft-enable flag */
+	u8 io_sync;			/* writel() needs spin_unlock sync */
 
 	/* Stuff for accurate time accounting */
 	u64 user_time;			/* accumulated usermode TB ticks */

include/asm-powerpc/spinlock.h

@@ -36,6 +36,19 @@
 #define LOCK_TOKEN	1
 #endif
 
+#if defined(CONFIG_PPC64) && defined(CONFIG_SMP)
+#define CLEAR_IO_SYNC	(get_paca()->io_sync = 0)
+#define SYNC_IO		do {						\
+				if (unlikely(get_paca()->io_sync)) {	\
+					mb();				\
+					get_paca()->io_sync = 0;	\
+				}					\
+			} while (0)
+#else
+#define CLEAR_IO_SYNC
+#define SYNC_IO
+#endif
+
 /*
  * This returns the old value in the lock, so we succeeded
  * in getting the lock if the return value is 0.
@@ -61,6 +74,7 @@ static __inline__ unsigned long __spin_trylock(raw_spinlock_t *lock)
 
 static int __inline__ __raw_spin_trylock(raw_spinlock_t *lock)
 {
+	CLEAR_IO_SYNC;
 	return __spin_trylock(lock) == 0;
 }
 
@@ -91,6 +105,7 @@ extern void __rw_yield(raw_rwlock_t *lock);
 
 static void __inline__ __raw_spin_lock(raw_spinlock_t *lock)
 {
+	CLEAR_IO_SYNC;
 	while (1) {
 		if (likely(__spin_trylock(lock) == 0))
 			break;
@@ -107,6 +122,7 @@ static void __inline__ __raw_spin_lock_flags(raw_spinlock_t *lock, unsigned long
 {
 	unsigned long flags_dis;
 
+	CLEAR_IO_SYNC;
 	while (1) {
 		if (likely(__spin_trylock(lock) == 0))
 			break;
@@ -124,6 +140,7 @@ static void __inline__ __raw_spin_lock_flags(raw_spinlock_t *lock, unsigned long
 
 static __inline__ void __raw_spin_unlock(raw_spinlock_t *lock)
 {
+	SYNC_IO;
 	__asm__ __volatile__("# __raw_spin_unlock\n\t"
 				LWSYNC_ON_SMP: : :"memory");
 	lock->slock = 0;

include/asm-ppc/io.h

@@ -63,7 +63,7 @@ extern inline int in_8(const volatile unsigned char __iomem *addr)
 	int ret;
 
 	__asm__ __volatile__(
-		"lbz%U1%X1 %0,%1;\n"
+		"sync; lbz%U1%X1 %0,%1;\n"
 		"twi 0,%0,0;\n"
 		"isync" : "=r" (ret) : "m" (*addr));
 	return ret;
@@ -78,7 +78,7 @@ extern inline int in_le16(const volatile unsigned short __iomem *addr)
 {
 	int ret;
 
-	__asm__ __volatile__("lhbrx %0,0,%1;\n"
+	__asm__ __volatile__("sync; lhbrx %0,0,%1;\n"
 			     "twi 0,%0,0;\n"
 			     "isync" : "=r" (ret) :
 			      "r" (addr), "m" (*addr));
@@ -89,7 +89,7 @@ extern inline int in_be16(const volatile unsigned short __iomem *addr)
 {
 	int ret;
 
-	__asm__ __volatile__("lhz%U1%X1 %0,%1;\n"
+	__asm__ __volatile__("sync; lhz%U1%X1 %0,%1;\n"
 			     "twi 0,%0,0;\n"
 			     "isync" : "=r" (ret) : "m" (*addr));
 	return ret;
@@ -97,20 +97,20 @@ extern inline int in_be16(const volatile unsigned short __iomem *addr)
 
 extern inline void out_le16(volatile unsigned short __iomem *addr, int val)
 {
-	__asm__ __volatile__("sthbrx %1,0,%2; eieio" : "=m" (*addr) :
+	__asm__ __volatile__("sync; sthbrx %1,0,%2" : "=m" (*addr) :
 			      "r" (val), "r" (addr));
 }
 
 extern inline void out_be16(volatile unsigned short __iomem *addr, int val)
 {
-	__asm__ __volatile__("sth%U0%X0 %1,%0; eieio" : "=m" (*addr) : "r" (val));
+	__asm__ __volatile__("sync; sth%U0%X0 %1,%0" : "=m" (*addr) : "r" (val));
 }
 
 extern inline unsigned in_le32(const volatile unsigned __iomem *addr)
 {
 	unsigned ret;
 
-	__asm__ __volatile__("lwbrx %0,0,%1;\n"
+	__asm__ __volatile__("sync; lwbrx %0,0,%1;\n"
 			     "twi 0,%0,0;\n"
 			     "isync" : "=r" (ret) :
 			     "r" (addr), "m" (*addr));
@@ -121,7 +121,7 @@ extern inline unsigned in_be32(const volatile unsigned __iomem *addr)
 {
 	unsigned ret;
 
-	__asm__ __volatile__("lwz%U1%X1 %0,%1;\n"
+	__asm__ __volatile__("sync; lwz%U1%X1 %0,%1;\n"
 			     "twi 0,%0,0;\n"
 			     "isync" : "=r" (ret) : "m" (*addr));
 	return ret;
@@ -129,13 +129,13 @@ extern inline unsigned in_be32(const volatile unsigned __iomem *addr)
 
 extern inline void out_le32(volatile unsigned __iomem *addr, int val)
 {
-	__asm__ __volatile__("stwbrx %1,0,%2; eieio" : "=m" (*addr) :
+	__asm__ __volatile__("sync; stwbrx %1,0,%2" : "=m" (*addr) :
 			     "r" (val), "r" (addr));
 }
 
 extern inline void out_be32(volatile unsigned __iomem *addr, int val)
 {
-	__asm__ __volatile__("stw%U0%X0 %1,%0; eieio" : "=m" (*addr) : "r" (val));
+	__asm__ __volatile__("sync; stw%U0%X0 %1,%0" : "=m" (*addr) : "r" (val));
 }
 #if defined (CONFIG_8260_PCI9)
 #define readb(addr) in_8((volatile u8 *)(addr))
@@ -259,6 +259,7 @@ extern __inline__ unsigned int name(unsigned int port)	\
 {							\
 	unsigned int x;					\
 	__asm__ __volatile__(				\
+		"sync\n"				\
 		"0:"	op "	%0,0,%1\n"		\
 		"1:	twi	0,%0,0\n"		\
 		"2:	isync\n"			\
@@ -284,6 +285,7 @@ extern __inline__ unsigned int name(unsigned int port)	\
 extern __inline__ void name(unsigned int val, unsigned int port) \
 {							\
 	__asm__ __volatile__(				\
+		"sync\n"				\
 		"0:" op " %0,0,%1\n"			\
 		"1:	sync\n"				\
 		"2:\n"					\