Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  Revert "[IA64] fix percpu warnings"
  [IA64] fix percpu warnings
  [IA64] SMT friendly version of spin_unlock_wait()
  [IA64] use printk_once() unaligned.c/io_common.c
  [IA64] Require SAL 3.2 in order to do extended config space ops
  [IA64] unsigned cannot be less than 0 in sn_hwperf_ioctl()
  [IA64] Restore registers in the stack on INIT
  [IA64] Re-implement spinaphores using ticket lock concepts
  [IA64] Squeeze ticket locks back into 4 bytes.

arch/ia64/include/asm/spinlock.h

@@ -25,61 +25,82 @@
  * by atomically noting the tail and incrementing it by one (thus adding
  * ourself to the queue and noting our position), then waiting until the head
  * becomes equal to the the initial value of the tail.
+ * The pad bits in the middle are used to prevent the next_ticket number
+ * overflowing into the now_serving number.
  *
- *   63                     32  31                      0
+ *   31             17  16    15  14                    0
  *  +----------------------------------------------------+
- *  |  next_ticket_number      |     now_serving         |
+ *  |  now_serving     | padding |   next_ticket         |
  *  +----------------------------------------------------+
  */
 
-#define TICKET_SHIFT	32
+#define TICKET_SHIFT	17
+#define TICKET_BITS	15
+#define	TICKET_MASK	((1 << TICKET_BITS) - 1)
 
 static __always_inline void __ticket_spin_lock(raw_spinlock_t *lock)
 {
-	int	*p = (int *)&lock->lock, turn, now_serving;
+	int	*p = (int *)&lock->lock, ticket, serve;
 
-	now_serving = *p;
-	turn = ia64_fetchadd(1, p+1, acq);
+	ticket = ia64_fetchadd(1, p, acq);
 
-	if (turn == now_serving)
+	if (!(((ticket >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
 		return;
 
-	do {
+	ia64_invala();
+
+	for (;;) {
+		asm volatile ("ld4.c.nc %0=[%1]" : "=r"(serve) : "r"(p) : "memory");
+
+		if (!(((serve >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
+			return;
 		cpu_relax();
-	} while (ACCESS_ONCE(*p) != turn);
+	}
 }
 
 static __always_inline int __ticket_spin_trylock(raw_spinlock_t *lock)
 {
-	long tmp = ACCESS_ONCE(lock->lock), try;
+	int tmp = ACCESS_ONCE(lock->lock);
 
-	if (!(((tmp >> TICKET_SHIFT) ^ tmp) & ((1L << TICKET_SHIFT) - 1))) {
-		try = tmp + (1L << TICKET_SHIFT);
-
-		return ia64_cmpxchg(acq, &lock->lock, tmp, try, sizeof (tmp)) == tmp;
-	}
+	if (!(((tmp >> TICKET_SHIFT) ^ tmp) & TICKET_MASK))
+		return ia64_cmpxchg(acq, &lock->lock, tmp, tmp + 1, sizeof (tmp)) == tmp;
 	return 0;
 }
 
 static __always_inline void __ticket_spin_unlock(raw_spinlock_t *lock)
 {
-	int	*p = (int *)&lock->lock;
+	unsigned short	*p = (unsigned short *)&lock->lock + 1, tmp;
 
-	(void)ia64_fetchadd(1, p, rel);
+	asm volatile ("ld2.bias %0=[%1]" : "=r"(tmp) : "r"(p));
+	ACCESS_ONCE(*p) = (tmp + 2) & ~1;
+}
+
+static __always_inline void __ticket_spin_unlock_wait(raw_spinlock_t *lock)
+{
+	int	*p = (int *)&lock->lock, ticket;
+
+	ia64_invala();
+
+	for (;;) {
+		asm volatile ("ld4.c.nc %0=[%1]" : "=r"(ticket) : "r"(p) : "memory");
+		if (!(((ticket >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
+			return;
+		cpu_relax();
+	}
 }
 
 static inline int __ticket_spin_is_locked(raw_spinlock_t *lock)
 {
 	long tmp = ACCESS_ONCE(lock->lock);
 
-	return !!(((tmp >> TICKET_SHIFT) ^ tmp) & ((1L << TICKET_SHIFT) - 1));
+	return !!(((tmp >> TICKET_SHIFT) ^ tmp) & TICKET_MASK);
 }
 
 static inline int __ticket_spin_is_contended(raw_spinlock_t *lock)
 {
 	long tmp = ACCESS_ONCE(lock->lock);
 
-	return (((tmp >> TICKET_SHIFT) - tmp) & ((1L << TICKET_SHIFT) - 1)) > 1;
+	return ((tmp - (tmp >> TICKET_SHIFT)) & TICKET_MASK) > 1;
 }
 
 static inline int __raw_spin_is_locked(raw_spinlock_t *lock)
@@ -116,8 +137,7 @@ static __always_inline void __raw_spin_lock_flags(raw_spinlock_t *lock,
 
 static inline void __raw_spin_unlock_wait(raw_spinlock_t *lock)
 {
-	while (__raw_spin_is_locked(lock))
-		cpu_relax();
+	__ticket_spin_unlock_wait(lock);
 }
 
 #define __raw_read_can_lock(rw)		(*(volatile int *)(rw) >= 0)

arch/ia64/include/asm/spinlock_types.h

@@ -6,7 +6,7 @@
 #endif
 
 typedef struct {
-	volatile unsigned long lock;
+	volatile unsigned int lock;
 } raw_spinlock_t;
 
 #define __RAW_SPIN_LOCK_UNLOCKED	{ 0 }

arch/ia64/kernel/mca.c

@@ -887,6 +887,60 @@ ia64_mca_modify_comm(const struct task_struct *previous_current)
 	memcpy(current->comm, comm, sizeof(current->comm));
 }
 
+static void
+finish_pt_regs(struct pt_regs *regs, const pal_min_state_area_t *ms,
+		unsigned long *nat)
+{
+	const u64 *bank;
+
+	/* If ipsr.ic then use pmsa_{iip,ipsr,ifs}, else use
+	 * pmsa_{xip,xpsr,xfs}
+	 */
+	if (ia64_psr(regs)->ic) {
+		regs->cr_iip = ms->pmsa_iip;
+		regs->cr_ipsr = ms->pmsa_ipsr;
+		regs->cr_ifs = ms->pmsa_ifs;
+	} else {
+		regs->cr_iip = ms->pmsa_xip;
+		regs->cr_ipsr = ms->pmsa_xpsr;
+		regs->cr_ifs = ms->pmsa_xfs;
+	}
+	regs->pr = ms->pmsa_pr;
+	regs->b0 = ms->pmsa_br0;
+	regs->ar_rsc = ms->pmsa_rsc;
+	copy_reg(&ms->pmsa_gr[1-1], ms->pmsa_nat_bits, &regs->r1, nat);
+	copy_reg(&ms->pmsa_gr[2-1], ms->pmsa_nat_bits, &regs->r2, nat);
+	copy_reg(&ms->pmsa_gr[3-1], ms->pmsa_nat_bits, &regs->r3, nat);
+	copy_reg(&ms->pmsa_gr[8-1], ms->pmsa_nat_bits, &regs->r8, nat);
+	copy_reg(&ms->pmsa_gr[9-1], ms->pmsa_nat_bits, &regs->r9, nat);
+	copy_reg(&ms->pmsa_gr[10-1], ms->pmsa_nat_bits, &regs->r10, nat);
+	copy_reg(&ms->pmsa_gr[11-1], ms->pmsa_nat_bits, &regs->r11, nat);
+	copy_reg(&ms->pmsa_gr[12-1], ms->pmsa_nat_bits, &regs->r12, nat);
+	copy_reg(&ms->pmsa_gr[13-1], ms->pmsa_nat_bits, &regs->r13, nat);
+	copy_reg(&ms->pmsa_gr[14-1], ms->pmsa_nat_bits, &regs->r14, nat);
+	copy_reg(&ms->pmsa_gr[15-1], ms->pmsa_nat_bits, &regs->r15, nat);
+	if (ia64_psr(regs)->bn)
+		bank = ms->pmsa_bank1_gr;
+	else
+		bank = ms->pmsa_bank0_gr;
+	copy_reg(&bank[16-16], ms->pmsa_nat_bits, &regs->r16, nat);
+	copy_reg(&bank[17-16], ms->pmsa_nat_bits, &regs->r17, nat);
+	copy_reg(&bank[18-16], ms->pmsa_nat_bits, &regs->r18, nat);
+	copy_reg(&bank[19-16], ms->pmsa_nat_bits, &regs->r19, nat);
+	copy_reg(&bank[20-16], ms->pmsa_nat_bits, &regs->r20, nat);
+	copy_reg(&bank[21-16], ms->pmsa_nat_bits, &regs->r21, nat);
+	copy_reg(&bank[22-16], ms->pmsa_nat_bits, &regs->r22, nat);
+	copy_reg(&bank[23-16], ms->pmsa_nat_bits, &regs->r23, nat);
+	copy_reg(&bank[24-16], ms->pmsa_nat_bits, &regs->r24, nat);
+	copy_reg(&bank[25-16], ms->pmsa_nat_bits, &regs->r25, nat);
+	copy_reg(&bank[26-16], ms->pmsa_nat_bits, &regs->r26, nat);
+	copy_reg(&bank[27-16], ms->pmsa_nat_bits, &regs->r27, nat);
+	copy_reg(&bank[28-16], ms->pmsa_nat_bits, &regs->r28, nat);
+	copy_reg(&bank[29-16], ms->pmsa_nat_bits, &regs->r29, nat);
+	copy_reg(&bank[30-16], ms->pmsa_nat_bits, &regs->r30, nat);
+	copy_reg(&bank[31-16], ms->pmsa_nat_bits, &regs->r31, nat);
+}
+
 /* On entry to this routine, we are running on the per cpu stack, see
  * mca_asm.h.  The original stack has not been touched by this event.  Some of
  * the original stack's registers will be in the RBS on this stack.  This stack
@@ -921,7 +975,6 @@ ia64_mca_modify_original_stack(struct pt_regs *regs,
 	u64 r12 = ms->pmsa_gr[12-1], r13 = ms->pmsa_gr[13-1];
 	u64 ar_bspstore = regs->ar_bspstore;
 	u64 ar_bsp = regs->ar_bspstore + (loadrs >> 16);
-	const u64 *bank;
 	const char *msg;
 	int cpu = smp_processor_id();
 
@@ -1024,54 +1077,9 @@ ia64_mca_modify_original_stack(struct pt_regs *regs,
 	p = (char *)r12 - sizeof(*regs);
 	old_regs = (struct pt_regs *)p;
 	memcpy(old_regs, regs, sizeof(*regs));
-	/* If ipsr.ic then use pmsa_{iip,ipsr,ifs}, else use
-	 * pmsa_{xip,xpsr,xfs}
-	 */
-	if (ia64_psr(regs)->ic) {
-		old_regs->cr_iip = ms->pmsa_iip;
-		old_regs->cr_ipsr = ms->pmsa_ipsr;
-		old_regs->cr_ifs = ms->pmsa_ifs;
-	} else {
-		old_regs->cr_iip = ms->pmsa_xip;
-		old_regs->cr_ipsr = ms->pmsa_xpsr;
-		old_regs->cr_ifs = ms->pmsa_xfs;
-	}
-	old_regs->pr = ms->pmsa_pr;
-	old_regs->b0 = ms->pmsa_br0;
 	old_regs->loadrs = loadrs;
-	old_regs->ar_rsc = ms->pmsa_rsc;
 	old_unat = old_regs->ar_unat;
-	copy_reg(&ms->pmsa_gr[1-1], ms->pmsa_nat_bits, &old_regs->r1, &old_unat);
-	copy_reg(&ms->pmsa_gr[2-1], ms->pmsa_nat_bits, &old_regs->r2, &old_unat);
-	copy_reg(&ms->pmsa_gr[3-1], ms->pmsa_nat_bits, &old_regs->r3, &old_unat);
-	copy_reg(&ms->pmsa_gr[8-1], ms->pmsa_nat_bits, &old_regs->r8, &old_unat);
-	copy_reg(&ms->pmsa_gr[9-1], ms->pmsa_nat_bits, &old_regs->r9, &old_unat);
-	copy_reg(&ms->pmsa_gr[10-1], ms->pmsa_nat_bits, &old_regs->r10, &old_unat);
-	copy_reg(&ms->pmsa_gr[11-1], ms->pmsa_nat_bits, &old_regs->r11, &old_unat);
-	copy_reg(&ms->pmsa_gr[12-1], ms->pmsa_nat_bits, &old_regs->r12, &old_unat);
-	copy_reg(&ms->pmsa_gr[13-1], ms->pmsa_nat_bits, &old_regs->r13, &old_unat);
-	copy_reg(&ms->pmsa_gr[14-1], ms->pmsa_nat_bits, &old_regs->r14, &old_unat);
-	copy_reg(&ms->pmsa_gr[15-1], ms->pmsa_nat_bits, &old_regs->r15, &old_unat);
-	if (ia64_psr(old_regs)->bn)
-		bank = ms->pmsa_bank1_gr;
-	else
-		bank = ms->pmsa_bank0_gr;
-	copy_reg(&bank[16-16], ms->pmsa_nat_bits, &old_regs->r16, &old_unat);
-	copy_reg(&bank[17-16], ms->pmsa_nat_bits, &old_regs->r17, &old_unat);
-	copy_reg(&bank[18-16], ms->pmsa_nat_bits, &old_regs->r18, &old_unat);
-	copy_reg(&bank[19-16], ms->pmsa_nat_bits, &old_regs->r19, &old_unat);
-	copy_reg(&bank[20-16], ms->pmsa_nat_bits, &old_regs->r20, &old_unat);
-	copy_reg(&bank[21-16], ms->pmsa_nat_bits, &old_regs->r21, &old_unat);
-	copy_reg(&bank[22-16], ms->pmsa_nat_bits, &old_regs->r22, &old_unat);
-	copy_reg(&bank[23-16], ms->pmsa_nat_bits, &old_regs->r23, &old_unat);
-	copy_reg(&bank[24-16], ms->pmsa_nat_bits, &old_regs->r24, &old_unat);
-	copy_reg(&bank[25-16], ms->pmsa_nat_bits, &old_regs->r25, &old_unat);
-	copy_reg(&bank[26-16], ms->pmsa_nat_bits, &old_regs->r26, &old_unat);
-	copy_reg(&bank[27-16], ms->pmsa_nat_bits, &old_regs->r27, &old_unat);
-	copy_reg(&bank[28-16], ms->pmsa_nat_bits, &old_regs->r28, &old_unat);
-	copy_reg(&bank[29-16], ms->pmsa_nat_bits, &old_regs->r29, &old_unat);
-	copy_reg(&bank[30-16], ms->pmsa_nat_bits, &old_regs->r30, &old_unat);
-	copy_reg(&bank[31-16], ms->pmsa_nat_bits, &old_regs->r31, &old_unat);
+	finish_pt_regs(old_regs, ms, &old_unat);
 
 	/* Next stack a struct switch_stack.  mca_asm.S built a partial
 	 * switch_stack, copy it and fill in the blanks using pt_regs and
@@ -1141,6 +1149,8 @@ ia64_mca_modify_original_stack(struct pt_regs *regs,
 no_mod:
 	mprintk(KERN_INFO "cpu %d, %s %s, original stack not modified\n",
 			smp_processor_id(), type, msg);
+	old_unat = regs->ar_unat;
+	finish_pt_regs(regs, ms, &old_unat);
 	return previous_current;
 }
 

arch/ia64/kernel/unaligned.c

@@ -60,7 +60,6 @@ dump (const char *str, void *vp, size_t len)
  */
 int no_unaligned_warning;
 int unaligned_dump_stack;
-static int noprint_warning;
 
 /*
  * For M-unit:
@@ -1357,9 +1356,8 @@ ia64_handle_unaligned (unsigned long ifa, struct pt_regs *regs)
 			/* watch for command names containing %s */
 			printk(KERN_WARNING "%s", buf);
 		} else {
-			if (no_unaligned_warning && !noprint_warning) {
-				noprint_warning = 1;
-				printk(KERN_WARNING "%s(%d) encountered an "
+			if (no_unaligned_warning) {
+				printk_once(KERN_WARNING "%s(%d) encountered an "
 				       "unaligned exception which required\n"
 				       "kernel assistance, which degrades "
 				       "the performance of the application.\n"

arch/ia64/mm/tlb.c

@@ -100,24 +100,36 @@ wrap_mmu_context (struct mm_struct *mm)
  * this primitive it can be moved up to a spinaphore.h header.
  */
 struct spinaphore {
-	atomic_t	cur;
+	unsigned long	ticket;
+	unsigned long	serve;
 };
 
 static inline void spinaphore_init(struct spinaphore *ss, int val)
 {
-	atomic_set(&ss->cur, val);
+	ss->ticket = 0;
+	ss->serve = val;
 }
 
 static inline void down_spin(struct spinaphore *ss)
 {
-	while (unlikely(!atomic_add_unless(&ss->cur, -1, 0)))
-		while (atomic_read(&ss->cur) == 0)
+	unsigned long t = ia64_fetchadd(1, &ss->ticket, acq), serve;
+
+	if (time_before(t, ss->serve))
+		return;
+
+	ia64_invala();
+
+	for (;;) {
+		asm volatile ("ld4.c.nc %0=[%1]" : "=r"(serve) : "r"(&ss->serve) : "memory");
+		if (time_before(t, serve))
+			return;
 		cpu_relax();
+	}
 }
 
 static inline void up_spin(struct spinaphore *ss)
 {
-	atomic_add(1, &ss->cur);
+	ia64_fetchadd(1, &ss->serve, rel);
 }
 
 static struct spinaphore ptcg_sem;

arch/ia64/pci/pci.c

@@ -56,10 +56,13 @@ int raw_pci_read(unsigned int seg, unsigned int bus, unsigned int devfn,
 	if ((seg | reg) <= 255) {
 		addr = PCI_SAL_ADDRESS(seg, bus, devfn, reg);
 		mode = 0;
-	} else {
+	} else if (sal_revision >= SAL_VERSION_CODE(3,2)) {
 		addr = PCI_SAL_EXT_ADDRESS(seg, bus, devfn, reg);
 		mode = 1;
+	} else {
+		return -EINVAL;
 	}
+
 	result = ia64_sal_pci_config_read(addr, mode, len, &data);
 	if (result != 0)
 		return -EINVAL;
@@ -80,9 +83,11 @@ int raw_pci_write(unsigned int seg, unsigned int bus, unsigned int devfn,
 	if ((seg | reg) <= 255) {
 		addr = PCI_SAL_ADDRESS(seg, bus, devfn, reg);
 		mode = 0;
-	} else {
+	} else if (sal_revision >= SAL_VERSION_CODE(3,2)) {
 		addr = PCI_SAL_EXT_ADDRESS(seg, bus, devfn, reg);
 		mode = 1;
+	} else {
+		return -EINVAL;
 	}
 	result = ia64_sal_pci_config_write(addr, mode, len, value);
 	if (result != 0)

arch/ia64/sn/kernel/io_common.c

@@ -119,7 +119,6 @@ sn_pcidev_info_get(struct pci_dev *dev)
  * Additionally note that the struct sn_flush_device_war also has to be
  * removed from arch/ia64/sn/include/xtalk/hubdev.h
  */
-static u8 war_implemented = 0;
 
 static s64 sn_device_fixup_war(u64 nasid, u64 widget, int device,
 			       struct sn_flush_device_common *common)
@@ -128,11 +127,8 @@ static s64 sn_device_fixup_war(u64 nasid, u64 widget, int device,
 	struct sn_flush_device_war *dev_entry;
 	struct ia64_sal_retval isrv = {0,0,0,0};
 
-	if (!war_implemented) {
-		printk(KERN_WARNING "PROM version < 4.50 -- implementing old "
-		       "PROM flush WAR\n");
-		war_implemented = 1;
-	}
+	printk_once(KERN_WARNING
+		"PROM version < 4.50 -- implementing old PROM flush WAR\n");
 
 	war_list = kzalloc(DEV_PER_WIDGET * sizeof(*war_list), GFP_KERNEL);
 	BUG_ON(!war_list);

arch/ia64/sn/kernel/sn2/sn_hwperf.c

@@ -786,17 +786,18 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, unsigned long arg)
 		break;
 
 	case SN_HWPERF_GET_OBJ_NODE:
-		if (a.sz != sizeof(u64) || a.arg < 0) {
+		i = a.arg;
+		if (a.sz != sizeof(u64) || i < 0) {
 			r = -EINVAL;
 			goto error;
 		}
 		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
-			if (a.arg >= nobj) {
+			if (i >= nobj) {
 				r = -EINVAL;
 				vfree(objs);
 				goto error;
 			}
-			if (objs[(i = a.arg)].id != a.arg) {
+			if (objs[i].id != a.arg) {
 				for (i = 0; i < nobj; i++) {
 					if (objs[i].id == a.arg)
 						break;