[SPARC64]: Send all device interrupts via one PIL.

This is the first in a series of cleanups that will hopefully
allow a seamless attempt at using the generic IRQ handling
infrastructure in the Linux kernel.

Define PIL_DEVICE_IRQ and vector all device interrupts through
there.

Get rid of the ugly pil0_dummy_{bucket,desc}, instead vector
the timer interrupt directly to a specific handler since the
timer interrupt is the only event that will be signaled on
PIL 14.

The irq_worklist is now in the per-cpu trap_block[].

Signed-off-by: David S. Miller <davem@davemloft.net>

8 files changed, 101 insertions, 142 deletions

diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S
index 6d0b3ed77a02..c87365e59e71 100644
--- a/arch/sparc64/kernel/entry.S
+++ b/arch/sparc64/kernel/entry.S
@@ -22,6 +22,7 @@
 #include <asm/estate.h>
 #include <asm/auxio.h>
 #include <asm/sfafsr.h>
+#include <asm/pil.h>
 
 #define curptr      g6
 
@@ -434,17 +435,13 @@ do_ivec:
 	sllx		%g3, 5, %g3
 	or		%g2, %lo(ivector_table), %g2
 	add		%g2, %g3, %g3
-	ldub		[%g3 + 0x04], %g4	/* pil */
-	mov		1, %g2
-	sllx		%g2, %g4, %g2
-	sllx		%g4, 2, %g4
 
 	TRAP_LOAD_IRQ_WORK(%g6, %g1)
 
-	lduw		[%g6 + %g4], %g5	/* g5 = irq_work(cpu, pil) */
+	lduw		[%g6], %g5		/* g5 = irq_work(cpu) */
 	stw		%g5, [%g3 + 0x00]	/* bucket->irq_chain = g5 */
-	stw		%g3, [%g6 + %g4]	/* irq_work(cpu, pil) = bucket */
-	wr		%g2, 0x0, %set_softint
+	stw		%g3, [%g6]		/* irq_work(cpu) = bucket */
+	wr		%g0, 1 << PIL_DEVICE_IRQ, %set_softint
 	retry
 do_ivec_xcall:
 	mov		0x50, %g1
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index 11e645c9ec50..0fb1738a4f3f 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -68,11 +68,7 @@ struct ino_bucket ivector_table[NUM_IVECS] __attribute__ ((aligned (SMP_CACHE_BY
  * access to this structure takes a TLB miss it could cause
  * the 5-level sparc v9 trap stack to overflow.
  */
-struct irq_work_struct {
-	unsigned int	irq_worklists[16];
-};
-struct irq_work_struct __irq_work[NR_CPUS];
-#define irq_work(__cpu, __pil)	&(__irq_work[(__cpu)].irq_worklists[(__pil)])
+#define irq_work(__cpu)	&(trap_block[(__cpu)].irq_worklist)
 
 static struct irqaction *irq_action[NR_IRQS+1];
 
@@ -91,10 +87,8 @@ static void register_irq_proc (unsigned int irq);
  */
 #define put_ino_in_irqaction(action, irq) \
 	action->flags &= 0xffffffffffffUL; \
-	if (__bucket(irq) == &pil0_dummy_bucket) \
-		action->flags |= 0xdeadUL << 48;  \
-	else \
-		action->flags |= __irq_ino(irq) << 48;
+	action->flags |= __irq_ino(irq) << 48;
+
 #define get_ino_in_irqaction(action)	(action->flags >> 48)
 
 #define put_smpaff_in_irqaction(action, smpaff)	(action)->mask = (smpaff)
@@ -251,15 +245,6 @@ void disable_irq(unsigned int irq)
 	}
 }
 
-/* The timer is the one "weird" interrupt which is generated by
- * the CPU %tick register and not by some normal vectored interrupt
- * source.  To handle this special case, we use this dummy INO bucket.
- */
-static struct irq_desc pil0_dummy_desc;
-static struct ino_bucket pil0_dummy_bucket = {
-	.irq_info	=	&pil0_dummy_desc,
-};
-
 static void build_irq_error(const char *msg, unsigned int ino, int pil, int inofixup,
 			    unsigned long iclr, unsigned long imap,
 			    struct ino_bucket *bucket)
@@ -276,15 +261,7 @@ unsigned int build_irq(int pil, int inofixup, unsigned long iclr, unsigned long
 	struct ino_bucket *bucket;
 	int ino;
 
-	if (pil == 0) {
-		if (iclr != 0UL || imap != 0UL) {
-			prom_printf("Invalid dummy bucket for PIL0 (%lx:%lx)\n",
-				    iclr, imap);
-			prom_halt();
-		}
-		return __irq(&pil0_dummy_bucket);
-	}
-
+	BUG_ON(pil == 0);
 	BUG_ON(tlb_type == hypervisor);
 
 	/* RULE: Both must be specified in all other cases. */
@@ -371,7 +348,7 @@ static void atomic_bucket_insert(struct ino_bucket *bucket)
 	__asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 	__asm__ __volatile__("wrpr %0, %1, %%pstate"
 			     : : "r" (pstate), "i" (PSTATE_IE));
-	ent = irq_work(smp_processor_id(), bucket->pil);
+	ent = irq_work(smp_processor_id());
 	bucket->irq_chain = *ent;
 	*ent = __irq(bucket);
 	__asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
@@ -437,7 +414,7 @@ int request_irq(unsigned int irq, irqreturn_t (*handler)(int, void *, struct pt_
 	if (unlikely(!bucket->irq_info))
 		return -ENODEV;
 
-	if ((bucket != &pil0_dummy_bucket) && (irqflags & SA_SAMPLE_RANDOM)) {
+	if (irqflags & SA_SAMPLE_RANDOM) {
 		/*
 	 	 * This function might sleep, we want to call it first,
 	 	 * outside of the atomic block. In SA_STATIC_ALLOC case,
@@ -465,12 +442,9 @@ int request_irq(unsigned int irq, irqreturn_t (*handler)(int, void *, struct pt_
 	}
 
 	bucket->flags |= IBF_ACTIVE;
-	pending = 0;
-	if (bucket != &pil0_dummy_bucket) {
-		pending = bucket->pending;
-		if (pending)
-			bucket->pending = 0;
-	}
+	pending = bucket->pending;
+	if (pending)
+		bucket->pending = 0;
 
 	action->handler = handler;
 	action->flags = irqflags;
@@ -487,13 +461,12 @@ int request_irq(unsigned int irq, irqreturn_t (*handler)(int, void *, struct pt_
 	/* We ate the IVEC already, this makes sure it does not get lost. */
 	if (pending) {
 		atomic_bucket_insert(bucket);
-		set_softint(1 << bucket->pil);
+		set_softint(1 << PIL_DEVICE_IRQ);
 	}
 
 	spin_unlock_irqrestore(&irq_action_lock, flags);
 
-	if (bucket != &pil0_dummy_bucket)
-		register_irq_proc(__irq_ino(irq));
+	register_irq_proc(__irq_ino(irq));
 
 #ifdef CONFIG_SMP
 	distribute_irqs();
@@ -533,7 +506,9 @@ void free_irq(unsigned int irq, void *dev_id)
 {
 	struct irqaction *action;
 	struct ino_bucket *bucket;
+	struct irq_desc *desc;
 	unsigned long flags;
+	int ent, i;
 
 	spin_lock_irqsave(&irq_action_lock, flags);
 
@@ -549,42 +524,39 @@ void free_irq(unsigned int irq, void *dev_id)
 	spin_lock_irqsave(&irq_action_lock, flags);
 
 	bucket = __bucket(irq);
-	if (bucket != &pil0_dummy_bucket) {
-		struct irq_desc *desc = bucket->irq_info;
-		int ent, i;
+	desc = bucket->irq_info;
 
-		for (i = 0; i < MAX_IRQ_DESC_ACTION; i++) {
-			struct irqaction *p = &desc->action[i];
+	for (i = 0; i < MAX_IRQ_DESC_ACTION; i++) {
+		struct irqaction *p = &desc->action[i];
 
-			if (p == action) {
-				desc->action_active_mask &= ~(1 << i);
-				break;
-			}
+		if (p == action) {
+			desc->action_active_mask &= ~(1 << i);
+			break;
 		}
+	}
 
-		if (!desc->action_active_mask) {
-			unsigned long imap = bucket->imap;
-
-			/* This unique interrupt source is now inactive. */
-			bucket->flags &= ~IBF_ACTIVE;
+	if (!desc->action_active_mask) {
+		unsigned long imap = bucket->imap;
 
-			/* See if any other buckets share this bucket's IMAP
-			 * and are still active.
-			 */
-			for (ent = 0; ent < NUM_IVECS; ent++) {
-				struct ino_bucket *bp = &ivector_table[ent];
-				if (bp != bucket	&&
-				    bp->imap == imap	&&
-				    (bp->flags & IBF_ACTIVE) != 0)
-					break;
-			}
+		/* This unique interrupt source is now inactive. */
+		bucket->flags &= ~IBF_ACTIVE;
 
-			/* Only disable when no other sub-irq levels of
-			 * the same IMAP are active.
-			 */
-			if (ent == NUM_IVECS)
-				disable_irq(irq);
+		/* See if any other buckets share this bucket's IMAP
+		 * and are still active.
+		 */
+		for (ent = 0; ent < NUM_IVECS; ent++) {
+			struct ino_bucket *bp = &ivector_table[ent];
+			if (bp != bucket	&&
+			    bp->imap == imap	&&
+			    (bp->flags & IBF_ACTIVE) != 0)
+				break;
 		}
+
+		/* Only disable when no other sub-irq levels of
+		 * the same IMAP are active.
+		 */
+		if (ent == NUM_IVECS)
+			disable_irq(irq);
 	}
 
 	spin_unlock_irqrestore(&irq_action_lock, flags);
@@ -625,7 +597,7 @@ void synchronize_irq(unsigned int irq)
 }
 #endif /* CONFIG_SMP */
 
-static void process_bucket(int irq, struct ino_bucket *bp, struct pt_regs *regs)
+static void process_bucket(struct ino_bucket *bp, struct pt_regs *regs)
 {
 	struct irq_desc *desc = bp->irq_info;
 	unsigned char flags = bp->flags;
@@ -676,51 +648,54 @@ static void process_bucket(int irq, struct ino_bucket *bp, struct pt_regs *regs)
 
 		/* Test and add entropy */
 		if (random & SA_SAMPLE_RANDOM)
-			add_interrupt_randomness(irq);
+			add_interrupt_randomness(bp->pil);
 	}
 out:
 	bp->flags &= ~IBF_INPROGRESS;
 }
 
+#ifndef CONFIG_SMP
+extern irqreturn_t timer_interrupt(int, void *, struct pt_regs *);
+
+void timer_irq(int irq, struct pt_regs *regs)
+{
+	unsigned long clr_mask = 1 << irq;
+	unsigned long tick_mask = tick_ops->softint_mask;
+
+	if (get_softint() & tick_mask) {
+		irq = 0;
+		clr_mask = tick_mask;
+	}
+	clear_softint(clr_mask);
+
+	irq_enter();
+	kstat_this_cpu.irqs[irq]++;
+	timer_interrupt(irq, NULL, regs);
+	irq_exit();
+}
+#endif
+
 void handler_irq(int irq, struct pt_regs *regs)
 {
 	struct ino_bucket *bp;
 	int cpu = smp_processor_id();
 
-#ifndef CONFIG_SMP
-	/*
-	 * Check for TICK_INT on level 14 softint.
+	/* XXX at this point we should be able to assert that
+	 * XXX irq is PIL_DEVICE_IRQ...
 	 */
-	{
-		unsigned long clr_mask = 1 << irq;
-		unsigned long tick_mask = tick_ops->softint_mask;
-
-		if ((irq == 14) && (get_softint() & tick_mask)) {
-			irq = 0;
-			clr_mask = tick_mask;
-		}
-		clear_softint(clr_mask);
-	}
-#else
 	clear_softint(1 << irq);
-#endif
 
 	irq_enter();
-	kstat_this_cpu.irqs[irq]++;
 
 	/* Sliiiick... */
-#ifndef CONFIG_SMP
-	bp = ((irq != 0) ?
-	      __bucket(xchg32(irq_work(cpu, irq), 0)) :
-	      &pil0_dummy_bucket);
-#else
-	bp = __bucket(xchg32(irq_work(cpu, irq), 0));
-#endif
+	bp = __bucket(xchg32(irq_work(cpu), 0));
 	while (bp) {
 		struct ino_bucket *nbp = __bucket(bp->irq_chain);
 
+		kstat_this_cpu.irqs[bp->pil]++;
+
 		bp->irq_chain = 0;
-		process_bucket(irq, bp, regs);
+		process_bucket(bp, regs);
 		bp = nbp;
 	}
 	irq_exit();
@@ -929,7 +904,7 @@ void init_irqwork_curcpu(void)
 {
 	int cpu = hard_smp_processor_id();
 
-	memset(__irq_work + cpu, 0, sizeof(struct irq_work_struct));
+	trap_block[cpu].irq_worklist = 0;
 }
 
 static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type)
diff --git a/arch/sparc64/kernel/sun4v_ivec.S b/arch/sparc64/kernel/sun4v_ivec.S
index b49a68bdda43..f70e4774649d 100644
--- a/arch/sparc64/kernel/sun4v_ivec.S
+++ b/arch/sparc64/kernel/sun4v_ivec.S
@@ -5,6 +5,7 @@
 
 #include <asm/cpudata.h>
 #include <asm/intr_queue.h>
+#include <asm/pil.h>
 
 	.text
 	.align	32
@@ -106,19 +107,13 @@ sun4v_dev_mondo:
 	or	%g4, %lo(ivector_table), %g4
 	add	%g4, %g3, %g4
 
-	/* Load IRQ %pil into %g5.  */
-	ldub	[%g4 + 0x04], %g5
-
 	/* Insert ivector_table[] entry into __irq_work[] queue.  */
-	sllx	%g5, 2, %g3
-	lduw	[%g1 + %g3], %g2	/* g2 = irq_work(cpu, pil) */
+	lduw	[%g1], %g2		/* g2 = irq_work(cpu) */
 	stw	%g2, [%g4 + 0x00]	/* bucket->irq_chain = g2 */
-	stw	%g4, [%g1 + %g3]	/* irq_work(cpu, pil) = bucket */
+	stw	%g4, [%g1]		/* irq_work(cpu) = bucket */
 
 	/* Signal the interrupt by setting (1 << pil) in %softint.  */
-	mov	1, %g2
-	sllx	%g2, %g5, %g2
-	wr	%g2, 0x0, %set_softint
+	wr	%g0, 1 << PIL_DEVICE_IRQ, %set_softint
 
 sun4v_dev_mondo_queue_empty:
 	retry
diff --git a/arch/sparc64/kernel/time.c b/arch/sparc64/kernel/time.c
index e55b5c6ece02..0f00a99927e9 100644
--- a/arch/sparc64/kernel/time.c
+++ b/arch/sparc64/kernel/time.c
@@ -457,7 +457,7 @@ static inline void timer_check_rtc(void)
 	}
 }
 
-static irqreturn_t timer_interrupt(int irq, void *dev_id, struct pt_regs * regs)
+irqreturn_t timer_interrupt(int irq, void *dev_id, struct pt_regs * regs)
 {
 	unsigned long ticks, compare, pstate;
 
@@ -1020,19 +1020,9 @@ static unsigned long sparc64_init_timers(void)
 	return clock;
 }
 
-static void sparc64_start_timers(irqreturn_t (*cfunc)(int, void *, struct pt_regs *))
+static void sparc64_start_timers(void)
 {
 	unsigned long pstate;
-	int err;
-
-	/* Register IRQ handler. */
-	err = request_irq(build_irq(0, 0, 0UL, 0UL), cfunc, 0,
-			  "timer", NULL);
-
-	if (err) {
-		prom_printf("Serious problem, cannot register TICK_INT\n");
-		prom_halt();
-	}
 
 	/* Guarantee that the following sequences execute
 	 * uninterrupted.
@@ -1116,7 +1106,7 @@ void __init time_init(void)
 	/* Now that the interpolator is registered, it is
 	 * safe to start the timer ticking.
 	 */
-	sparc64_start_timers(timer_interrupt);
+	sparc64_start_timers();
 
 	timer_ticks_per_nsec_quotient =
 		(((NSEC_PER_SEC << SPARC64_NSEC_PER_CYC_SHIFT) +
diff --git a/arch/sparc64/kernel/traps.c b/arch/sparc64/kernel/traps.c
index 563db528e031..5059cbd4feee 100644
--- a/arch/sparc64/kernel/traps.c
+++ b/arch/sparc64/kernel/traps.c
@@ -2544,7 +2544,9 @@ void __init trap_init(void)
 	    (TRAP_PER_CPU_TSB_HUGE !=
 	     offsetof(struct trap_per_cpu, tsb_huge)) ||
 	    (TRAP_PER_CPU_TSB_HUGE_TEMP !=
-	     offsetof(struct trap_per_cpu, tsb_huge_temp)))
+	     offsetof(struct trap_per_cpu, tsb_huge_temp)) ||
+	    (TRAP_PER_CPU_IRQ_WORKLIST !=
+	     offsetof(struct trap_per_cpu, irq_worklist)))
 		trap_per_cpu_offsets_are_bolixed_dave();
 
 	if ((TSB_CONFIG_TSB !=
diff --git a/arch/sparc64/kernel/ttable.S b/arch/sparc64/kernel/ttable.S
index 5d901519db55..ee45ca2d7a04 100644
--- a/arch/sparc64/kernel/ttable.S
+++ b/arch/sparc64/kernel/ttable.S
@@ -58,13 +58,11 @@ tl0_irq2:	BTRAP(0x42)
 tl0_irq3:	BTRAP(0x43)
 tl0_irq4:	BTRAP(0x44)
 #endif
-tl0_irq5:	TRAP_IRQ(handler_irq, 5)  TRAP_IRQ(handler_irq, 6)
-tl0_irq7:	TRAP_IRQ(handler_irq, 7)  TRAP_IRQ(handler_irq, 8)
-tl0_irq9:	TRAP_IRQ(handler_irq, 9)  TRAP_IRQ(handler_irq, 10)
-tl0_irq11:	TRAP_IRQ(handler_irq, 11) TRAP_IRQ(handler_irq, 12)
-tl0_irq13:	TRAP_IRQ(handler_irq, 13)
+tl0_irq5:	TRAP_IRQ(handler_irq, 5)
+tl0_irq6:	BTRAP(0x46) BTRAP(0x47) BTRAP(0x48) BTRAP(0x49)
+tl0_irq10:	BTRAP(0x4a) BTRAP(0x4b) BTRAP(0x4c) BTRAP(0x4d)
 #ifndef CONFIG_SMP
-tl0_irq14:	TRAP_IRQ(handler_irq, 14)
+tl0_irq14:	TRAP_IRQ(timer_irq, 14)
 #else
 tl0_irq14:	TICK_SMP_IRQ
 #endif
diff --git a/include/asm-sparc64/cpudata.h b/include/asm-sparc64/cpudata.h
index 9d6a6dbaf126..f2cc9411b4c7 100644
--- a/include/asm-sparc64/cpudata.h
+++ b/include/asm-sparc64/cpudata.h
@@ -74,8 +74,10 @@ struct trap_per_cpu {
 	unsigned long		tsb_huge;
 	unsigned long		tsb_huge_temp;
 
-/* Dcache line 8: Unused, needed to keep trap_block a power-of-2 in size.  */
-	unsigned long		__pad2[4];
+/* Dcache line 8: IRQ work list, and keep trap_block a power-of-2 in size.  */
+	unsigned int		irq_worklist;
+	unsigned int		__pad1;
+	unsigned long		__pad2[3];
 } __attribute__((aligned(64)));
 extern struct trap_per_cpu trap_block[NR_CPUS];
 extern void init_cur_cpu_trap(struct thread_info *);
@@ -119,6 +121,7 @@ extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
 #define TRAP_PER_CPU_CPU_LIST_PA	0xc8
 #define TRAP_PER_CPU_TSB_HUGE		0xd0
 #define TRAP_PER_CPU_TSB_HUGE_TEMP	0xd8
+#define TRAP_PER_CPU_IRQ_WORKLIST	0xe0
 
 #define TRAP_BLOCK_SZ_SHIFT		8
 
@@ -171,11 +174,8 @@ extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
 
 /* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
 #define TRAP_LOAD_IRQ_WORK(DEST, TMP)		\
-	__GET_CPUID(TMP)			\
-	sethi	%hi(__irq_work), DEST;		\
-	sllx	TMP, 6, TMP;			\
-	or	DEST, %lo(__irq_work), DEST;	\
-	add	DEST, TMP, DEST;
+	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	add	DEST, TRAP_PER_CPU_IRQ_WORKLIST, DEST;
 
 /* Clobbers TMP, loads DEST with current thread info pointer.  */
 #define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
@@ -211,9 +211,10 @@ extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
 	ldx	[DEST + TRAP_PER_CPU_PGD_PADDR], DEST;
 
+/* Clobbers TMP, loads local processor's IRQ work area into DEST.  */
 #define TRAP_LOAD_IRQ_WORK(DEST, TMP)		\
-	sethi	%hi(__irq_work), DEST;		\
-	or	DEST, %lo(__irq_work), DEST;
+	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
+	add	DEST, TRAP_PER_CPU_IRQ_WORKLIST, DEST;
 
 #define TRAP_LOAD_THREAD_REG(DEST, TMP)		\
 	TRAP_LOAD_TRAP_BLOCK(DEST, TMP)		\
diff --git a/include/asm-sparc64/pil.h b/include/asm-sparc64/pil.h
index 79f827eb3f5d..72927749aebf 100644
--- a/include/asm-sparc64/pil.h
+++ b/include/asm-sparc64/pil.h
@@ -5,9 +5,9 @@
 /* To avoid some locking problems, we hard allocate certain PILs
  * for SMP cross call messages that must do a etrap/rtrap.
  *
- * A cli() does not block the cross call delivery, so when SMP
- * locking is an issue we reschedule the event into a PIL interrupt
- * which is blocked by cli().
+ * A local_irq_disable() does not block the cross call delivery, so
+ * when SMP locking is an issue we reschedule the event into a PIL
+ * interrupt which is blocked by local_irq_disable().
  *
  * In fact any XCALL which has to etrap/rtrap has a problem because
  * it is difficult to prevent rtrap from running BH's, and that would
@@ -17,6 +17,7 @@
 #define PIL_SMP_RECEIVE_SIGNAL	2
 #define PIL_SMP_CAPTURE		3
 #define PIL_SMP_CTX_NEW_VERSION	4
+#define PIL_DEVICE_IRQ		5
 
 #ifndef __ASSEMBLY__
 #define PIL_RESERVED(PIL)	((PIL) == PIL_SMP_CALL_FUNC || \