xref: /openbmc/linux/arch/sparc/kernel/irq_64.c (revision 82ced6fd)
1 /* irq.c: UltraSparc IRQ handling/init/registry.
2  *
3  * Copyright (C) 1997, 2007, 2008 David S. Miller (davem@davemloft.net)
4  * Copyright (C) 1998  Eddie C. Dost    (ecd@skynet.be)
5  * Copyright (C) 1998  Jakub Jelinek    (jj@ultra.linux.cz)
6  */
7 
8 #include <linux/module.h>
9 #include <linux/sched.h>
10 #include <linux/linkage.h>
11 #include <linux/ptrace.h>
12 #include <linux/errno.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/signal.h>
15 #include <linux/mm.h>
16 #include <linux/interrupt.h>
17 #include <linux/slab.h>
18 #include <linux/random.h>
19 #include <linux/init.h>
20 #include <linux/delay.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/bootmem.h>
24 #include <linux/irq.h>
25 
26 #include <asm/ptrace.h>
27 #include <asm/processor.h>
28 #include <asm/atomic.h>
29 #include <asm/system.h>
30 #include <asm/irq.h>
31 #include <asm/io.h>
32 #include <asm/iommu.h>
33 #include <asm/upa.h>
34 #include <asm/oplib.h>
35 #include <asm/prom.h>
36 #include <asm/timer.h>
37 #include <asm/smp.h>
38 #include <asm/starfire.h>
39 #include <asm/uaccess.h>
40 #include <asm/cache.h>
41 #include <asm/cpudata.h>
42 #include <asm/auxio.h>
43 #include <asm/head.h>
44 #include <asm/hypervisor.h>
45 #include <asm/cacheflush.h>
46 
47 #include "entry.h"
48 
49 #define NUM_IVECS	(IMAP_INR + 1)
50 
51 struct ino_bucket *ivector_table;
52 unsigned long ivector_table_pa;
53 
54 /* On several sun4u processors, it is illegal to mix bypass and
55  * non-bypass accesses.  Therefore we access all INO buckets
56  * using bypass accesses only.
57  */
58 static unsigned long bucket_get_chain_pa(unsigned long bucket_pa)
59 {
60 	unsigned long ret;
61 
62 	__asm__ __volatile__("ldxa	[%1] %2, %0"
63 			     : "=&r" (ret)
64 			     : "r" (bucket_pa +
65 				    offsetof(struct ino_bucket,
66 					     __irq_chain_pa)),
67 			       "i" (ASI_PHYS_USE_EC));
68 
69 	return ret;
70 }
71 
72 static void bucket_clear_chain_pa(unsigned long bucket_pa)
73 {
74 	__asm__ __volatile__("stxa	%%g0, [%0] %1"
75 			     : /* no outputs */
76 			     : "r" (bucket_pa +
77 				    offsetof(struct ino_bucket,
78 					     __irq_chain_pa)),
79 			       "i" (ASI_PHYS_USE_EC));
80 }
81 
82 static unsigned int bucket_get_virt_irq(unsigned long bucket_pa)
83 {
84 	unsigned int ret;
85 
86 	__asm__ __volatile__("lduwa	[%1] %2, %0"
87 			     : "=&r" (ret)
88 			     : "r" (bucket_pa +
89 				    offsetof(struct ino_bucket,
90 					     __virt_irq)),
91 			       "i" (ASI_PHYS_USE_EC));
92 
93 	return ret;
94 }
95 
96 static void bucket_set_virt_irq(unsigned long bucket_pa,
97 				unsigned int virt_irq)
98 {
99 	__asm__ __volatile__("stwa	%0, [%1] %2"
100 			     : /* no outputs */
101 			     : "r" (virt_irq),
102 			       "r" (bucket_pa +
103 				    offsetof(struct ino_bucket,
104 					     __virt_irq)),
105 			       "i" (ASI_PHYS_USE_EC));
106 }
107 
108 #define irq_work_pa(__cpu)	&(trap_block[(__cpu)].irq_worklist_pa)
109 
110 static struct {
111 	unsigned int dev_handle;
112 	unsigned int dev_ino;
113 	unsigned int in_use;
114 } virt_irq_table[NR_IRQS];
115 static DEFINE_SPINLOCK(virt_irq_alloc_lock);
116 
117 unsigned char virt_irq_alloc(unsigned int dev_handle,
118 			     unsigned int dev_ino)
119 {
120 	unsigned long flags;
121 	unsigned char ent;
122 
123 	BUILD_BUG_ON(NR_IRQS >= 256);
124 
125 	spin_lock_irqsave(&virt_irq_alloc_lock, flags);
126 
127 	for (ent = 1; ent < NR_IRQS; ent++) {
128 		if (!virt_irq_table[ent].in_use)
129 			break;
130 	}
131 	if (ent >= NR_IRQS) {
132 		printk(KERN_ERR "IRQ: Out of virtual IRQs.\n");
133 		ent = 0;
134 	} else {
135 		virt_irq_table[ent].dev_handle = dev_handle;
136 		virt_irq_table[ent].dev_ino = dev_ino;
137 		virt_irq_table[ent].in_use = 1;
138 	}
139 
140 	spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
141 
142 	return ent;
143 }
144 
145 #ifdef CONFIG_PCI_MSI
146 void virt_irq_free(unsigned int virt_irq)
147 {
148 	unsigned long flags;
149 
150 	if (virt_irq >= NR_IRQS)
151 		return;
152 
153 	spin_lock_irqsave(&virt_irq_alloc_lock, flags);
154 
155 	virt_irq_table[virt_irq].in_use = 0;
156 
157 	spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
158 }
159 #endif
160 
161 /*
162  * /proc/interrupts printing:
163  */
164 
165 int show_interrupts(struct seq_file *p, void *v)
166 {
167 	int i = *(loff_t *) v, j;
168 	struct irqaction * action;
169 	unsigned long flags;
170 
171 	if (i == 0) {
172 		seq_printf(p, "           ");
173 		for_each_online_cpu(j)
174 			seq_printf(p, "CPU%d       ",j);
175 		seq_putc(p, '\n');
176 	}
177 
178 	if (i < NR_IRQS) {
179 		spin_lock_irqsave(&irq_desc[i].lock, flags);
180 		action = irq_desc[i].action;
181 		if (!action)
182 			goto skip;
183 		seq_printf(p, "%3d: ",i);
184 #ifndef CONFIG_SMP
185 		seq_printf(p, "%10u ", kstat_irqs(i));
186 #else
187 		for_each_online_cpu(j)
188 			seq_printf(p, "%10u ", kstat_irqs_cpu(i, j));
189 #endif
190 		seq_printf(p, " %9s", irq_desc[i].chip->typename);
191 		seq_printf(p, "  %s", action->name);
192 
193 		for (action=action->next; action; action = action->next)
194 			seq_printf(p, ", %s", action->name);
195 
196 		seq_putc(p, '\n');
197 skip:
198 		spin_unlock_irqrestore(&irq_desc[i].lock, flags);
199 	} else if (i == NR_IRQS) {
200 		seq_printf(p, "NMI: ");
201 		for_each_online_cpu(j)
202 			seq_printf(p, "%10u ", cpu_data(j).__nmi_count);
203 		seq_printf(p, "     Non-maskable interrupts\n");
204 	}
205 	return 0;
206 }
207 
208 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
209 {
210 	unsigned int tid;
211 
212 	if (this_is_starfire) {
213 		tid = starfire_translate(imap, cpuid);
214 		tid <<= IMAP_TID_SHIFT;
215 		tid &= IMAP_TID_UPA;
216 	} else {
217 		if (tlb_type == cheetah || tlb_type == cheetah_plus) {
218 			unsigned long ver;
219 
220 			__asm__ ("rdpr %%ver, %0" : "=r" (ver));
221 			if ((ver >> 32UL) == __JALAPENO_ID ||
222 			    (ver >> 32UL) == __SERRANO_ID) {
223 				tid = cpuid << IMAP_TID_SHIFT;
224 				tid &= IMAP_TID_JBUS;
225 			} else {
226 				unsigned int a = cpuid & 0x1f;
227 				unsigned int n = (cpuid >> 5) & 0x1f;
228 
229 				tid = ((a << IMAP_AID_SHIFT) |
230 				       (n << IMAP_NID_SHIFT));
231 				tid &= (IMAP_AID_SAFARI |
232 					IMAP_NID_SAFARI);;
233 			}
234 		} else {
235 			tid = cpuid << IMAP_TID_SHIFT;
236 			tid &= IMAP_TID_UPA;
237 		}
238 	}
239 
240 	return tid;
241 }
242 
243 struct irq_handler_data {
244 	unsigned long	iclr;
245 	unsigned long	imap;
246 
247 	void		(*pre_handler)(unsigned int, void *, void *);
248 	void		*arg1;
249 	void		*arg2;
250 };
251 
252 #ifdef CONFIG_SMP
253 static int irq_choose_cpu(unsigned int virt_irq)
254 {
255 	cpumask_t mask;
256 	int cpuid;
257 
258 	cpumask_copy(&mask, irq_desc[virt_irq].affinity);
259 	if (cpus_equal(mask, CPU_MASK_ALL)) {
260 		static int irq_rover;
261 		static DEFINE_SPINLOCK(irq_rover_lock);
262 		unsigned long flags;
263 
264 		/* Round-robin distribution... */
265 	do_round_robin:
266 		spin_lock_irqsave(&irq_rover_lock, flags);
267 
268 		while (!cpu_online(irq_rover)) {
269 			if (++irq_rover >= nr_cpu_ids)
270 				irq_rover = 0;
271 		}
272 		cpuid = irq_rover;
273 		do {
274 			if (++irq_rover >= nr_cpu_ids)
275 				irq_rover = 0;
276 		} while (!cpu_online(irq_rover));
277 
278 		spin_unlock_irqrestore(&irq_rover_lock, flags);
279 	} else {
280 		cpumask_t tmp;
281 
282 		cpus_and(tmp, cpu_online_map, mask);
283 
284 		if (cpus_empty(tmp))
285 			goto do_round_robin;
286 
287 		cpuid = first_cpu(tmp);
288 	}
289 
290 	return cpuid;
291 }
292 #else
293 static int irq_choose_cpu(unsigned int virt_irq)
294 {
295 	return real_hard_smp_processor_id();
296 }
297 #endif
298 
299 static void sun4u_irq_enable(unsigned int virt_irq)
300 {
301 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
302 
303 	if (likely(data)) {
304 		unsigned long cpuid, imap, val;
305 		unsigned int tid;
306 
307 		cpuid = irq_choose_cpu(virt_irq);
308 		imap = data->imap;
309 
310 		tid = sun4u_compute_tid(imap, cpuid);
311 
312 		val = upa_readq(imap);
313 		val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
314 			 IMAP_AID_SAFARI | IMAP_NID_SAFARI);
315 		val |= tid | IMAP_VALID;
316 		upa_writeq(val, imap);
317 		upa_writeq(ICLR_IDLE, data->iclr);
318 	}
319 }
320 
321 static void sun4u_set_affinity(unsigned int virt_irq,
322 			       const struct cpumask *mask)
323 {
324 	sun4u_irq_enable(virt_irq);
325 }
326 
327 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
328  * handler_irq() will skip the handler call and that will leave the
329  * interrupt in the sent state.  The next ->enable() call will hit the
330  * ICLR register to reset the state machine.
331  *
332  * This scheme is necessary, instead of clearing the Valid bit in the
333  * IMAP register, to handle the case of IMAP registers being shared by
334  * multiple INOs (and thus ICLR registers).  Since we use a different
335  * virtual IRQ for each shared IMAP instance, the generic code thinks
336  * there is only one user so it prematurely calls ->disable() on
337  * free_irq().
338  *
339  * We have to provide an explicit ->disable() method instead of using
340  * NULL to get the default.  The reason is that if the generic code
341  * sees that, it also hooks up a default ->shutdown method which
342  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
343  */
344 static void sun4u_irq_disable(unsigned int virt_irq)
345 {
346 }
347 
348 static void sun4u_irq_eoi(unsigned int virt_irq)
349 {
350 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
351 	struct irq_desc *desc = irq_desc + virt_irq;
352 
353 	if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
354 		return;
355 
356 	if (likely(data))
357 		upa_writeq(ICLR_IDLE, data->iclr);
358 }
359 
360 static void sun4v_irq_enable(unsigned int virt_irq)
361 {
362 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
363 	unsigned long cpuid = irq_choose_cpu(virt_irq);
364 	int err;
365 
366 	err = sun4v_intr_settarget(ino, cpuid);
367 	if (err != HV_EOK)
368 		printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
369 		       "err(%d)\n", ino, cpuid, err);
370 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
371 	if (err != HV_EOK)
372 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
373 		       "err(%d)\n", ino, err);
374 	err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
375 	if (err != HV_EOK)
376 		printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
377 		       ino, err);
378 }
379 
380 static void sun4v_set_affinity(unsigned int virt_irq,
381 			       const struct cpumask *mask)
382 {
383 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
384 	unsigned long cpuid = irq_choose_cpu(virt_irq);
385 	int err;
386 
387 	err = sun4v_intr_settarget(ino, cpuid);
388 	if (err != HV_EOK)
389 		printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
390 		       "err(%d)\n", ino, cpuid, err);
391 }
392 
393 static void sun4v_irq_disable(unsigned int virt_irq)
394 {
395 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
396 	int err;
397 
398 	err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
399 	if (err != HV_EOK)
400 		printk(KERN_ERR "sun4v_intr_setenabled(%x): "
401 		       "err(%d)\n", ino, err);
402 }
403 
404 static void sun4v_irq_eoi(unsigned int virt_irq)
405 {
406 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
407 	struct irq_desc *desc = irq_desc + virt_irq;
408 	int err;
409 
410 	if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
411 		return;
412 
413 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
414 	if (err != HV_EOK)
415 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
416 		       "err(%d)\n", ino, err);
417 }
418 
419 static void sun4v_virq_enable(unsigned int virt_irq)
420 {
421 	unsigned long cpuid, dev_handle, dev_ino;
422 	int err;
423 
424 	cpuid = irq_choose_cpu(virt_irq);
425 
426 	dev_handle = virt_irq_table[virt_irq].dev_handle;
427 	dev_ino = virt_irq_table[virt_irq].dev_ino;
428 
429 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
430 	if (err != HV_EOK)
431 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
432 		       "err(%d)\n",
433 		       dev_handle, dev_ino, cpuid, err);
434 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
435 				    HV_INTR_STATE_IDLE);
436 	if (err != HV_EOK)
437 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
438 		       "HV_INTR_STATE_IDLE): err(%d)\n",
439 		       dev_handle, dev_ino, err);
440 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
441 				    HV_INTR_ENABLED);
442 	if (err != HV_EOK)
443 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
444 		       "HV_INTR_ENABLED): err(%d)\n",
445 		       dev_handle, dev_ino, err);
446 }
447 
448 static void sun4v_virt_set_affinity(unsigned int virt_irq,
449 				    const struct cpumask *mask)
450 {
451 	unsigned long cpuid, dev_handle, dev_ino;
452 	int err;
453 
454 	cpuid = irq_choose_cpu(virt_irq);
455 
456 	dev_handle = virt_irq_table[virt_irq].dev_handle;
457 	dev_ino = virt_irq_table[virt_irq].dev_ino;
458 
459 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
460 	if (err != HV_EOK)
461 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
462 		       "err(%d)\n",
463 		       dev_handle, dev_ino, cpuid, err);
464 }
465 
466 static void sun4v_virq_disable(unsigned int virt_irq)
467 {
468 	unsigned long dev_handle, dev_ino;
469 	int err;
470 
471 	dev_handle = virt_irq_table[virt_irq].dev_handle;
472 	dev_ino = virt_irq_table[virt_irq].dev_ino;
473 
474 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
475 				    HV_INTR_DISABLED);
476 	if (err != HV_EOK)
477 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
478 		       "HV_INTR_DISABLED): err(%d)\n",
479 		       dev_handle, dev_ino, err);
480 }
481 
482 static void sun4v_virq_eoi(unsigned int virt_irq)
483 {
484 	struct irq_desc *desc = irq_desc + virt_irq;
485 	unsigned long dev_handle, dev_ino;
486 	int err;
487 
488 	if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
489 		return;
490 
491 	dev_handle = virt_irq_table[virt_irq].dev_handle;
492 	dev_ino = virt_irq_table[virt_irq].dev_ino;
493 
494 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
495 				    HV_INTR_STATE_IDLE);
496 	if (err != HV_EOK)
497 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
498 		       "HV_INTR_STATE_IDLE): err(%d)\n",
499 		       dev_handle, dev_ino, err);
500 }
501 
502 static struct irq_chip sun4u_irq = {
503 	.typename	= "sun4u",
504 	.enable		= sun4u_irq_enable,
505 	.disable	= sun4u_irq_disable,
506 	.eoi		= sun4u_irq_eoi,
507 	.set_affinity	= sun4u_set_affinity,
508 };
509 
510 static struct irq_chip sun4v_irq = {
511 	.typename	= "sun4v",
512 	.enable		= sun4v_irq_enable,
513 	.disable	= sun4v_irq_disable,
514 	.eoi		= sun4v_irq_eoi,
515 	.set_affinity	= sun4v_set_affinity,
516 };
517 
518 static struct irq_chip sun4v_virq = {
519 	.typename	= "vsun4v",
520 	.enable		= sun4v_virq_enable,
521 	.disable	= sun4v_virq_disable,
522 	.eoi		= sun4v_virq_eoi,
523 	.set_affinity	= sun4v_virt_set_affinity,
524 };
525 
526 static void pre_flow_handler(unsigned int virt_irq,
527 				      struct irq_desc *desc)
528 {
529 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
530 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
531 
532 	data->pre_handler(ino, data->arg1, data->arg2);
533 
534 	handle_fasteoi_irq(virt_irq, desc);
535 }
536 
537 void irq_install_pre_handler(int virt_irq,
538 			     void (*func)(unsigned int, void *, void *),
539 			     void *arg1, void *arg2)
540 {
541 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
542 	struct irq_desc *desc = irq_desc + virt_irq;
543 
544 	data->pre_handler = func;
545 	data->arg1 = arg1;
546 	data->arg2 = arg2;
547 
548 	desc->handle_irq = pre_flow_handler;
549 }
550 
551 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
552 {
553 	struct ino_bucket *bucket;
554 	struct irq_handler_data *data;
555 	unsigned int virt_irq;
556 	int ino;
557 
558 	BUG_ON(tlb_type == hypervisor);
559 
560 	ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
561 	bucket = &ivector_table[ino];
562 	virt_irq = bucket_get_virt_irq(__pa(bucket));
563 	if (!virt_irq) {
564 		virt_irq = virt_irq_alloc(0, ino);
565 		bucket_set_virt_irq(__pa(bucket), virt_irq);
566 		set_irq_chip_and_handler_name(virt_irq,
567 					      &sun4u_irq,
568 					      handle_fasteoi_irq,
569 					      "IVEC");
570 	}
571 
572 	data = get_irq_chip_data(virt_irq);
573 	if (unlikely(data))
574 		goto out;
575 
576 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
577 	if (unlikely(!data)) {
578 		prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
579 		prom_halt();
580 	}
581 	set_irq_chip_data(virt_irq, data);
582 
583 	data->imap  = imap;
584 	data->iclr  = iclr;
585 
586 out:
587 	return virt_irq;
588 }
589 
590 static unsigned int sun4v_build_common(unsigned long sysino,
591 				       struct irq_chip *chip)
592 {
593 	struct ino_bucket *bucket;
594 	struct irq_handler_data *data;
595 	unsigned int virt_irq;
596 
597 	BUG_ON(tlb_type != hypervisor);
598 
599 	bucket = &ivector_table[sysino];
600 	virt_irq = bucket_get_virt_irq(__pa(bucket));
601 	if (!virt_irq) {
602 		virt_irq = virt_irq_alloc(0, sysino);
603 		bucket_set_virt_irq(__pa(bucket), virt_irq);
604 		set_irq_chip_and_handler_name(virt_irq, chip,
605 					      handle_fasteoi_irq,
606 					      "IVEC");
607 	}
608 
609 	data = get_irq_chip_data(virt_irq);
610 	if (unlikely(data))
611 		goto out;
612 
613 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
614 	if (unlikely(!data)) {
615 		prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
616 		prom_halt();
617 	}
618 	set_irq_chip_data(virt_irq, data);
619 
620 	/* Catch accidental accesses to these things.  IMAP/ICLR handling
621 	 * is done by hypervisor calls on sun4v platforms, not by direct
622 	 * register accesses.
623 	 */
624 	data->imap = ~0UL;
625 	data->iclr = ~0UL;
626 
627 out:
628 	return virt_irq;
629 }
630 
631 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
632 {
633 	unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
634 
635 	return sun4v_build_common(sysino, &sun4v_irq);
636 }
637 
638 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
639 {
640 	struct irq_handler_data *data;
641 	unsigned long hv_err, cookie;
642 	struct ino_bucket *bucket;
643 	struct irq_desc *desc;
644 	unsigned int virt_irq;
645 
646 	bucket = kzalloc(sizeof(struct ino_bucket), GFP_ATOMIC);
647 	if (unlikely(!bucket))
648 		return 0;
649 	__flush_dcache_range((unsigned long) bucket,
650 			     ((unsigned long) bucket +
651 			      sizeof(struct ino_bucket)));
652 
653 	virt_irq = virt_irq_alloc(devhandle, devino);
654 	bucket_set_virt_irq(__pa(bucket), virt_irq);
655 
656 	set_irq_chip_and_handler_name(virt_irq, &sun4v_virq,
657 				      handle_fasteoi_irq,
658 				      "IVEC");
659 
660 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
661 	if (unlikely(!data))
662 		return 0;
663 
664 	/* In order to make the LDC channel startup sequence easier,
665 	 * especially wrt. locking, we do not let request_irq() enable
666 	 * the interrupt.
667 	 */
668 	desc = irq_desc + virt_irq;
669 	desc->status |= IRQ_NOAUTOEN;
670 
671 	set_irq_chip_data(virt_irq, data);
672 
673 	/* Catch accidental accesses to these things.  IMAP/ICLR handling
674 	 * is done by hypervisor calls on sun4v platforms, not by direct
675 	 * register accesses.
676 	 */
677 	data->imap = ~0UL;
678 	data->iclr = ~0UL;
679 
680 	cookie = ~__pa(bucket);
681 	hv_err = sun4v_vintr_set_cookie(devhandle, devino, cookie);
682 	if (hv_err) {
683 		prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
684 			    "err=%lu\n", devhandle, devino, hv_err);
685 		prom_halt();
686 	}
687 
688 	return virt_irq;
689 }
690 
691 void ack_bad_irq(unsigned int virt_irq)
692 {
693 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
694 
695 	if (!ino)
696 		ino = 0xdeadbeef;
697 
698 	printk(KERN_CRIT "Unexpected IRQ from ino[%x] virt_irq[%u]\n",
699 	       ino, virt_irq);
700 }
701 
702 void *hardirq_stack[NR_CPUS];
703 void *softirq_stack[NR_CPUS];
704 
705 static __attribute__((always_inline)) void *set_hardirq_stack(void)
706 {
707 	void *orig_sp, *sp = hardirq_stack[smp_processor_id()];
708 
709 	__asm__ __volatile__("mov %%sp, %0" : "=r" (orig_sp));
710 	if (orig_sp < sp ||
711 	    orig_sp > (sp + THREAD_SIZE)) {
712 		sp += THREAD_SIZE - 192 - STACK_BIAS;
713 		__asm__ __volatile__("mov %0, %%sp" : : "r" (sp));
714 	}
715 
716 	return orig_sp;
717 }
718 static __attribute__((always_inline)) void restore_hardirq_stack(void *orig_sp)
719 {
720 	__asm__ __volatile__("mov %0, %%sp" : : "r" (orig_sp));
721 }
722 
723 void handler_irq(int irq, struct pt_regs *regs)
724 {
725 	unsigned long pstate, bucket_pa;
726 	struct pt_regs *old_regs;
727 	void *orig_sp;
728 
729 	clear_softint(1 << irq);
730 
731 	old_regs = set_irq_regs(regs);
732 	irq_enter();
733 
734 	/* Grab an atomic snapshot of the pending IVECs.  */
735 	__asm__ __volatile__("rdpr	%%pstate, %0\n\t"
736 			     "wrpr	%0, %3, %%pstate\n\t"
737 			     "ldx	[%2], %1\n\t"
738 			     "stx	%%g0, [%2]\n\t"
739 			     "wrpr	%0, 0x0, %%pstate\n\t"
740 			     : "=&r" (pstate), "=&r" (bucket_pa)
741 			     : "r" (irq_work_pa(smp_processor_id())),
742 			       "i" (PSTATE_IE)
743 			     : "memory");
744 
745 	orig_sp = set_hardirq_stack();
746 
747 	while (bucket_pa) {
748 		struct irq_desc *desc;
749 		unsigned long next_pa;
750 		unsigned int virt_irq;
751 
752 		next_pa = bucket_get_chain_pa(bucket_pa);
753 		virt_irq = bucket_get_virt_irq(bucket_pa);
754 		bucket_clear_chain_pa(bucket_pa);
755 
756 		desc = irq_desc + virt_irq;
757 
758 		if (!(desc->status & IRQ_DISABLED))
759 			desc->handle_irq(virt_irq, desc);
760 
761 		bucket_pa = next_pa;
762 	}
763 
764 	restore_hardirq_stack(orig_sp);
765 
766 	irq_exit();
767 	set_irq_regs(old_regs);
768 }
769 
770 void do_softirq(void)
771 {
772 	unsigned long flags;
773 
774 	if (in_interrupt())
775 		return;
776 
777 	local_irq_save(flags);
778 
779 	if (local_softirq_pending()) {
780 		void *orig_sp, *sp = softirq_stack[smp_processor_id()];
781 
782 		sp += THREAD_SIZE - 192 - STACK_BIAS;
783 
784 		__asm__ __volatile__("mov %%sp, %0\n\t"
785 				     "mov %1, %%sp"
786 				     : "=&r" (orig_sp)
787 				     : "r" (sp));
788 		__do_softirq();
789 		__asm__ __volatile__("mov %0, %%sp"
790 				     : : "r" (orig_sp));
791 	}
792 
793 	local_irq_restore(flags);
794 }
795 
796 #ifdef CONFIG_HOTPLUG_CPU
797 void fixup_irqs(void)
798 {
799 	unsigned int irq;
800 
801 	for (irq = 0; irq < NR_IRQS; irq++) {
802 		unsigned long flags;
803 
804 		spin_lock_irqsave(&irq_desc[irq].lock, flags);
805 		if (irq_desc[irq].action &&
806 		    !(irq_desc[irq].status & IRQ_PER_CPU)) {
807 			if (irq_desc[irq].chip->set_affinity)
808 				irq_desc[irq].chip->set_affinity(irq,
809 					irq_desc[irq].affinity);
810 		}
811 		spin_unlock_irqrestore(&irq_desc[irq].lock, flags);
812 	}
813 
814 	tick_ops->disable_irq();
815 }
816 #endif
817 
818 struct sun5_timer {
819 	u64	count0;
820 	u64	limit0;
821 	u64	count1;
822 	u64	limit1;
823 };
824 
825 static struct sun5_timer *prom_timers;
826 static u64 prom_limit0, prom_limit1;
827 
828 static void map_prom_timers(void)
829 {
830 	struct device_node *dp;
831 	const unsigned int *addr;
832 
833 	/* PROM timer node hangs out in the top level of device siblings... */
834 	dp = of_find_node_by_path("/");
835 	dp = dp->child;
836 	while (dp) {
837 		if (!strcmp(dp->name, "counter-timer"))
838 			break;
839 		dp = dp->sibling;
840 	}
841 
842 	/* Assume if node is not present, PROM uses different tick mechanism
843 	 * which we should not care about.
844 	 */
845 	if (!dp) {
846 		prom_timers = (struct sun5_timer *) 0;
847 		return;
848 	}
849 
850 	/* If PROM is really using this, it must be mapped by him. */
851 	addr = of_get_property(dp, "address", NULL);
852 	if (!addr) {
853 		prom_printf("PROM does not have timer mapped, trying to continue.\n");
854 		prom_timers = (struct sun5_timer *) 0;
855 		return;
856 	}
857 	prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
858 }
859 
860 static void kill_prom_timer(void)
861 {
862 	if (!prom_timers)
863 		return;
864 
865 	/* Save them away for later. */
866 	prom_limit0 = prom_timers->limit0;
867 	prom_limit1 = prom_timers->limit1;
868 
869 	/* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
870 	 * We turn both off here just to be paranoid.
871 	 */
872 	prom_timers->limit0 = 0;
873 	prom_timers->limit1 = 0;
874 
875 	/* Wheee, eat the interrupt packet too... */
876 	__asm__ __volatile__(
877 "	mov	0x40, %%g2\n"
878 "	ldxa	[%%g0] %0, %%g1\n"
879 "	ldxa	[%%g2] %1, %%g1\n"
880 "	stxa	%%g0, [%%g0] %0\n"
881 "	membar	#Sync\n"
882 	: /* no outputs */
883 	: "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
884 	: "g1", "g2");
885 }
886 
887 void notrace init_irqwork_curcpu(void)
888 {
889 	int cpu = hard_smp_processor_id();
890 
891 	trap_block[cpu].irq_worklist_pa = 0UL;
892 }
893 
894 /* Please be very careful with register_one_mondo() and
895  * sun4v_register_mondo_queues().
896  *
897  * On SMP this gets invoked from the CPU trampoline before
898  * the cpu has fully taken over the trap table from OBP,
899  * and it's kernel stack + %g6 thread register state is
900  * not fully cooked yet.
901  *
902  * Therefore you cannot make any OBP calls, not even prom_printf,
903  * from these two routines.
904  */
905 static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
906 {
907 	unsigned long num_entries = (qmask + 1) / 64;
908 	unsigned long status;
909 
910 	status = sun4v_cpu_qconf(type, paddr, num_entries);
911 	if (status != HV_EOK) {
912 		prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
913 			    "err %lu\n", type, paddr, num_entries, status);
914 		prom_halt();
915 	}
916 }
917 
918 void __cpuinit notrace sun4v_register_mondo_queues(int this_cpu)
919 {
920 	struct trap_per_cpu *tb = &trap_block[this_cpu];
921 
922 	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
923 			   tb->cpu_mondo_qmask);
924 	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
925 			   tb->dev_mondo_qmask);
926 	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
927 			   tb->resum_qmask);
928 	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
929 			   tb->nonresum_qmask);
930 }
931 
932 static void __init alloc_one_mondo(unsigned long *pa_ptr, unsigned long qmask)
933 {
934 	unsigned long size = PAGE_ALIGN(qmask + 1);
935 	void *p = __alloc_bootmem(size, size, 0);
936 	if (!p) {
937 		prom_printf("SUN4V: Error, cannot allocate mondo queue.\n");
938 		prom_halt();
939 	}
940 
941 	*pa_ptr = __pa(p);
942 }
943 
944 static void __init alloc_one_kbuf(unsigned long *pa_ptr, unsigned long qmask)
945 {
946 	unsigned long size = PAGE_ALIGN(qmask + 1);
947 	void *p = __alloc_bootmem(size, size, 0);
948 
949 	if (!p) {
950 		prom_printf("SUN4V: Error, cannot allocate kbuf page.\n");
951 		prom_halt();
952 	}
953 
954 	*pa_ptr = __pa(p);
955 }
956 
957 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
958 {
959 #ifdef CONFIG_SMP
960 	void *page;
961 
962 	BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
963 
964 	page = alloc_bootmem_pages(PAGE_SIZE);
965 	if (!page) {
966 		prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
967 		prom_halt();
968 	}
969 
970 	tb->cpu_mondo_block_pa = __pa(page);
971 	tb->cpu_list_pa = __pa(page + 64);
972 #endif
973 }
974 
975 /* Allocate mondo and error queues for all possible cpus.  */
976 static void __init sun4v_init_mondo_queues(void)
977 {
978 	int cpu;
979 
980 	for_each_possible_cpu(cpu) {
981 		struct trap_per_cpu *tb = &trap_block[cpu];
982 
983 		alloc_one_mondo(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
984 		alloc_one_mondo(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
985 		alloc_one_mondo(&tb->resum_mondo_pa, tb->resum_qmask);
986 		alloc_one_kbuf(&tb->resum_kernel_buf_pa, tb->resum_qmask);
987 		alloc_one_mondo(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
988 		alloc_one_kbuf(&tb->nonresum_kernel_buf_pa,
989 			       tb->nonresum_qmask);
990 	}
991 }
992 
993 static void __init init_send_mondo_info(void)
994 {
995 	int cpu;
996 
997 	for_each_possible_cpu(cpu) {
998 		struct trap_per_cpu *tb = &trap_block[cpu];
999 
1000 		init_cpu_send_mondo_info(tb);
1001 	}
1002 }
1003 
1004 static struct irqaction timer_irq_action = {
1005 	.name = "timer",
1006 };
1007 
1008 /* Only invoked on boot processor. */
1009 void __init init_IRQ(void)
1010 {
1011 	unsigned long size;
1012 
1013 	map_prom_timers();
1014 	kill_prom_timer();
1015 
1016 	size = sizeof(struct ino_bucket) * NUM_IVECS;
1017 	ivector_table = alloc_bootmem(size);
1018 	if (!ivector_table) {
1019 		prom_printf("Fatal error, cannot allocate ivector_table\n");
1020 		prom_halt();
1021 	}
1022 	__flush_dcache_range((unsigned long) ivector_table,
1023 			     ((unsigned long) ivector_table) + size);
1024 
1025 	ivector_table_pa = __pa(ivector_table);
1026 
1027 	if (tlb_type == hypervisor)
1028 		sun4v_init_mondo_queues();
1029 
1030 	init_send_mondo_info();
1031 
1032 	if (tlb_type == hypervisor) {
1033 		/* Load up the boot cpu's entries.  */
1034 		sun4v_register_mondo_queues(hard_smp_processor_id());
1035 	}
1036 
1037 	/* We need to clear any IRQ's pending in the soft interrupt
1038 	 * registers, a spurious one could be left around from the
1039 	 * PROM timer which we just disabled.
1040 	 */
1041 	clear_softint(get_softint());
1042 
1043 	/* Now that ivector table is initialized, it is safe
1044 	 * to receive IRQ vector traps.  We will normally take
1045 	 * one or two right now, in case some device PROM used
1046 	 * to boot us wants to speak to us.  We just ignore them.
1047 	 */
1048 	__asm__ __volatile__("rdpr	%%pstate, %%g1\n\t"
1049 			     "or	%%g1, %0, %%g1\n\t"
1050 			     "wrpr	%%g1, 0x0, %%pstate"
1051 			     : /* No outputs */
1052 			     : "i" (PSTATE_IE)
1053 			     : "g1");
1054 
1055 	irq_desc[0].action = &timer_irq_action;
1056 }
1057