xref: /openbmc/linux/arch/sparc/kernel/irq_64.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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/ftrace.h>
24 #include <linux/irq.h>
25 #include <linux/kmemleak.h>
26 
27 #include <asm/ptrace.h>
28 #include <asm/processor.h>
29 #include <asm/atomic.h>
30 #include <asm/system.h>
31 #include <asm/irq.h>
32 #include <asm/io.h>
33 #include <asm/iommu.h>
34 #include <asm/upa.h>
35 #include <asm/oplib.h>
36 #include <asm/prom.h>
37 #include <asm/timer.h>
38 #include <asm/smp.h>
39 #include <asm/starfire.h>
40 #include <asm/uaccess.h>
41 #include <asm/cache.h>
42 #include <asm/cpudata.h>
43 #include <asm/auxio.h>
44 #include <asm/head.h>
45 #include <asm/hypervisor.h>
46 #include <asm/cacheflush.h>
47 
48 #include "entry.h"
49 #include "cpumap.h"
50 #include "kstack.h"
51 
52 #define NUM_IVECS	(IMAP_INR + 1)
53 
54 struct ino_bucket *ivector_table;
55 unsigned long ivector_table_pa;
56 
57 /* On several sun4u processors, it is illegal to mix bypass and
58  * non-bypass accesses.  Therefore we access all INO buckets
59  * using bypass accesses only.
60  */
61 static unsigned long bucket_get_chain_pa(unsigned long bucket_pa)
62 {
63 	unsigned long ret;
64 
65 	__asm__ __volatile__("ldxa	[%1] %2, %0"
66 			     : "=&r" (ret)
67 			     : "r" (bucket_pa +
68 				    offsetof(struct ino_bucket,
69 					     __irq_chain_pa)),
70 			       "i" (ASI_PHYS_USE_EC));
71 
72 	return ret;
73 }
74 
75 static void bucket_clear_chain_pa(unsigned long bucket_pa)
76 {
77 	__asm__ __volatile__("stxa	%%g0, [%0] %1"
78 			     : /* no outputs */
79 			     : "r" (bucket_pa +
80 				    offsetof(struct ino_bucket,
81 					     __irq_chain_pa)),
82 			       "i" (ASI_PHYS_USE_EC));
83 }
84 
85 static unsigned int bucket_get_virt_irq(unsigned long bucket_pa)
86 {
87 	unsigned int ret;
88 
89 	__asm__ __volatile__("lduwa	[%1] %2, %0"
90 			     : "=&r" (ret)
91 			     : "r" (bucket_pa +
92 				    offsetof(struct ino_bucket,
93 					     __virt_irq)),
94 			       "i" (ASI_PHYS_USE_EC));
95 
96 	return ret;
97 }
98 
99 static void bucket_set_virt_irq(unsigned long bucket_pa,
100 				unsigned int virt_irq)
101 {
102 	__asm__ __volatile__("stwa	%0, [%1] %2"
103 			     : /* no outputs */
104 			     : "r" (virt_irq),
105 			       "r" (bucket_pa +
106 				    offsetof(struct ino_bucket,
107 					     __virt_irq)),
108 			       "i" (ASI_PHYS_USE_EC));
109 }
110 
111 #define irq_work_pa(__cpu)	&(trap_block[(__cpu)].irq_worklist_pa)
112 
113 static struct {
114 	unsigned int dev_handle;
115 	unsigned int dev_ino;
116 	unsigned int in_use;
117 } virt_irq_table[NR_IRQS];
118 static DEFINE_SPINLOCK(virt_irq_alloc_lock);
119 
120 unsigned char virt_irq_alloc(unsigned int dev_handle,
121 			     unsigned int dev_ino)
122 {
123 	unsigned long flags;
124 	unsigned char ent;
125 
126 	BUILD_BUG_ON(NR_IRQS >= 256);
127 
128 	spin_lock_irqsave(&virt_irq_alloc_lock, flags);
129 
130 	for (ent = 1; ent < NR_IRQS; ent++) {
131 		if (!virt_irq_table[ent].in_use)
132 			break;
133 	}
134 	if (ent >= NR_IRQS) {
135 		printk(KERN_ERR "IRQ: Out of virtual IRQs.\n");
136 		ent = 0;
137 	} else {
138 		virt_irq_table[ent].dev_handle = dev_handle;
139 		virt_irq_table[ent].dev_ino = dev_ino;
140 		virt_irq_table[ent].in_use = 1;
141 	}
142 
143 	spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
144 
145 	return ent;
146 }
147 
148 #ifdef CONFIG_PCI_MSI
149 void virt_irq_free(unsigned int virt_irq)
150 {
151 	unsigned long flags;
152 
153 	if (virt_irq >= NR_IRQS)
154 		return;
155 
156 	spin_lock_irqsave(&virt_irq_alloc_lock, flags);
157 
158 	virt_irq_table[virt_irq].in_use = 0;
159 
160 	spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
161 }
162 #endif
163 
164 /*
165  * /proc/interrupts printing:
166  */
167 
168 int show_interrupts(struct seq_file *p, void *v)
169 {
170 	int i = *(loff_t *) v, j;
171 	struct irqaction * action;
172 	unsigned long flags;
173 
174 	if (i == 0) {
175 		seq_printf(p, "           ");
176 		for_each_online_cpu(j)
177 			seq_printf(p, "CPU%d       ",j);
178 		seq_putc(p, '\n');
179 	}
180 
181 	if (i < NR_IRQS) {
182 		raw_spin_lock_irqsave(&irq_desc[i].lock, flags);
183 		action = irq_desc[i].action;
184 		if (!action)
185 			goto skip;
186 		seq_printf(p, "%3d: ",i);
187 #ifndef CONFIG_SMP
188 		seq_printf(p, "%10u ", kstat_irqs(i));
189 #else
190 		for_each_online_cpu(j)
191 			seq_printf(p, "%10u ", kstat_irqs_cpu(i, j));
192 #endif
193 		seq_printf(p, " %9s", irq_desc[i].chip->name);
194 		seq_printf(p, "  %s", action->name);
195 
196 		for (action=action->next; action; action = action->next)
197 			seq_printf(p, ", %s", action->name);
198 
199 		seq_putc(p, '\n');
200 skip:
201 		raw_spin_unlock_irqrestore(&irq_desc[i].lock, flags);
202 	} else if (i == NR_IRQS) {
203 		seq_printf(p, "NMI: ");
204 		for_each_online_cpu(j)
205 			seq_printf(p, "%10u ", cpu_data(j).__nmi_count);
206 		seq_printf(p, "     Non-maskable interrupts\n");
207 	}
208 	return 0;
209 }
210 
211 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
212 {
213 	unsigned int tid;
214 
215 	if (this_is_starfire) {
216 		tid = starfire_translate(imap, cpuid);
217 		tid <<= IMAP_TID_SHIFT;
218 		tid &= IMAP_TID_UPA;
219 	} else {
220 		if (tlb_type == cheetah || tlb_type == cheetah_plus) {
221 			unsigned long ver;
222 
223 			__asm__ ("rdpr %%ver, %0" : "=r" (ver));
224 			if ((ver >> 32UL) == __JALAPENO_ID ||
225 			    (ver >> 32UL) == __SERRANO_ID) {
226 				tid = cpuid << IMAP_TID_SHIFT;
227 				tid &= IMAP_TID_JBUS;
228 			} else {
229 				unsigned int a = cpuid & 0x1f;
230 				unsigned int n = (cpuid >> 5) & 0x1f;
231 
232 				tid = ((a << IMAP_AID_SHIFT) |
233 				       (n << IMAP_NID_SHIFT));
234 				tid &= (IMAP_AID_SAFARI |
235 					IMAP_NID_SAFARI);
236 			}
237 		} else {
238 			tid = cpuid << IMAP_TID_SHIFT;
239 			tid &= IMAP_TID_UPA;
240 		}
241 	}
242 
243 	return tid;
244 }
245 
246 struct irq_handler_data {
247 	unsigned long	iclr;
248 	unsigned long	imap;
249 
250 	void		(*pre_handler)(unsigned int, void *, void *);
251 	void		*arg1;
252 	void		*arg2;
253 };
254 
255 #ifdef CONFIG_SMP
256 static int irq_choose_cpu(unsigned int virt_irq, const struct cpumask *affinity)
257 {
258 	cpumask_t mask;
259 	int cpuid;
260 
261 	cpumask_copy(&mask, affinity);
262 	if (cpus_equal(mask, cpu_online_map)) {
263 		cpuid = map_to_cpu(virt_irq);
264 	} else {
265 		cpumask_t tmp;
266 
267 		cpus_and(tmp, cpu_online_map, mask);
268 		cpuid = cpus_empty(tmp) ? map_to_cpu(virt_irq) : first_cpu(tmp);
269 	}
270 
271 	return cpuid;
272 }
273 #else
274 #define irq_choose_cpu(virt_irq, affinity)	\
275 	real_hard_smp_processor_id()
276 #endif
277 
278 static void sun4u_irq_enable(unsigned int virt_irq)
279 {
280 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
281 
282 	if (likely(data)) {
283 		unsigned long cpuid, imap, val;
284 		unsigned int tid;
285 
286 		cpuid = irq_choose_cpu(virt_irq,
287 				       irq_desc[virt_irq].affinity);
288 		imap = data->imap;
289 
290 		tid = sun4u_compute_tid(imap, cpuid);
291 
292 		val = upa_readq(imap);
293 		val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
294 			 IMAP_AID_SAFARI | IMAP_NID_SAFARI);
295 		val |= tid | IMAP_VALID;
296 		upa_writeq(val, imap);
297 		upa_writeq(ICLR_IDLE, data->iclr);
298 	}
299 }
300 
301 static int sun4u_set_affinity(unsigned int virt_irq,
302 			       const struct cpumask *mask)
303 {
304 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
305 
306 	if (likely(data)) {
307 		unsigned long cpuid, imap, val;
308 		unsigned int tid;
309 
310 		cpuid = irq_choose_cpu(virt_irq, mask);
311 		imap = data->imap;
312 
313 		tid = sun4u_compute_tid(imap, cpuid);
314 
315 		val = upa_readq(imap);
316 		val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
317 			 IMAP_AID_SAFARI | IMAP_NID_SAFARI);
318 		val |= tid | IMAP_VALID;
319 		upa_writeq(val, imap);
320 		upa_writeq(ICLR_IDLE, data->iclr);
321 	}
322 
323 	return 0;
324 }
325 
326 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
327  * handler_irq() will skip the handler call and that will leave the
328  * interrupt in the sent state.  The next ->enable() call will hit the
329  * ICLR register to reset the state machine.
330  *
331  * This scheme is necessary, instead of clearing the Valid bit in the
332  * IMAP register, to handle the case of IMAP registers being shared by
333  * multiple INOs (and thus ICLR registers).  Since we use a different
334  * virtual IRQ for each shared IMAP instance, the generic code thinks
335  * there is only one user so it prematurely calls ->disable() on
336  * free_irq().
337  *
338  * We have to provide an explicit ->disable() method instead of using
339  * NULL to get the default.  The reason is that if the generic code
340  * sees that, it also hooks up a default ->shutdown method which
341  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
342  */
343 static void sun4u_irq_disable(unsigned int virt_irq)
344 {
345 }
346 
347 static void sun4u_irq_eoi(unsigned int virt_irq)
348 {
349 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
350 	struct irq_desc *desc = irq_desc + virt_irq;
351 
352 	if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
353 		return;
354 
355 	if (likely(data))
356 		upa_writeq(ICLR_IDLE, data->iclr);
357 }
358 
359 static void sun4v_irq_enable(unsigned int virt_irq)
360 {
361 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
362 	unsigned long cpuid = irq_choose_cpu(virt_irq,
363 					     irq_desc[virt_irq].affinity);
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 int 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, mask);
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 	return 0;
393 }
394 
395 static void sun4v_irq_disable(unsigned int virt_irq)
396 {
397 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
398 	int err;
399 
400 	err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
401 	if (err != HV_EOK)
402 		printk(KERN_ERR "sun4v_intr_setenabled(%x): "
403 		       "err(%d)\n", ino, err);
404 }
405 
406 static void sun4v_irq_eoi(unsigned int virt_irq)
407 {
408 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
409 	struct irq_desc *desc = irq_desc + virt_irq;
410 	int err;
411 
412 	if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
413 		return;
414 
415 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
416 	if (err != HV_EOK)
417 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
418 		       "err(%d)\n", ino, err);
419 }
420 
421 static void sun4v_virq_enable(unsigned int virt_irq)
422 {
423 	unsigned long cpuid, dev_handle, dev_ino;
424 	int err;
425 
426 	cpuid = irq_choose_cpu(virt_irq, irq_desc[virt_irq].affinity);
427 
428 	dev_handle = virt_irq_table[virt_irq].dev_handle;
429 	dev_ino = virt_irq_table[virt_irq].dev_ino;
430 
431 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
432 	if (err != HV_EOK)
433 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
434 		       "err(%d)\n",
435 		       dev_handle, dev_ino, cpuid, err);
436 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
437 				    HV_INTR_STATE_IDLE);
438 	if (err != HV_EOK)
439 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
440 		       "HV_INTR_STATE_IDLE): err(%d)\n",
441 		       dev_handle, dev_ino, err);
442 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
443 				    HV_INTR_ENABLED);
444 	if (err != HV_EOK)
445 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
446 		       "HV_INTR_ENABLED): err(%d)\n",
447 		       dev_handle, dev_ino, err);
448 }
449 
450 static int sun4v_virt_set_affinity(unsigned int virt_irq,
451 				    const struct cpumask *mask)
452 {
453 	unsigned long cpuid, dev_handle, dev_ino;
454 	int err;
455 
456 	cpuid = irq_choose_cpu(virt_irq, mask);
457 
458 	dev_handle = virt_irq_table[virt_irq].dev_handle;
459 	dev_ino = virt_irq_table[virt_irq].dev_ino;
460 
461 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
462 	if (err != HV_EOK)
463 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
464 		       "err(%d)\n",
465 		       dev_handle, dev_ino, cpuid, err);
466 
467 	return 0;
468 }
469 
470 static void sun4v_virq_disable(unsigned int virt_irq)
471 {
472 	unsigned long dev_handle, dev_ino;
473 	int err;
474 
475 	dev_handle = virt_irq_table[virt_irq].dev_handle;
476 	dev_ino = virt_irq_table[virt_irq].dev_ino;
477 
478 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
479 				    HV_INTR_DISABLED);
480 	if (err != HV_EOK)
481 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
482 		       "HV_INTR_DISABLED): err(%d)\n",
483 		       dev_handle, dev_ino, err);
484 }
485 
486 static void sun4v_virq_eoi(unsigned int virt_irq)
487 {
488 	struct irq_desc *desc = irq_desc + virt_irq;
489 	unsigned long dev_handle, dev_ino;
490 	int err;
491 
492 	if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
493 		return;
494 
495 	dev_handle = virt_irq_table[virt_irq].dev_handle;
496 	dev_ino = virt_irq_table[virt_irq].dev_ino;
497 
498 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
499 				    HV_INTR_STATE_IDLE);
500 	if (err != HV_EOK)
501 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
502 		       "HV_INTR_STATE_IDLE): err(%d)\n",
503 		       dev_handle, dev_ino, err);
504 }
505 
506 static struct irq_chip sun4u_irq = {
507 	.name		= "sun4u",
508 	.enable		= sun4u_irq_enable,
509 	.disable	= sun4u_irq_disable,
510 	.eoi		= sun4u_irq_eoi,
511 	.set_affinity	= sun4u_set_affinity,
512 };
513 
514 static struct irq_chip sun4v_irq = {
515 	.name		= "sun4v",
516 	.enable		= sun4v_irq_enable,
517 	.disable	= sun4v_irq_disable,
518 	.eoi		= sun4v_irq_eoi,
519 	.set_affinity	= sun4v_set_affinity,
520 };
521 
522 static struct irq_chip sun4v_virq = {
523 	.name		= "vsun4v",
524 	.enable		= sun4v_virq_enable,
525 	.disable	= sun4v_virq_disable,
526 	.eoi		= sun4v_virq_eoi,
527 	.set_affinity	= sun4v_virt_set_affinity,
528 };
529 
530 static void pre_flow_handler(unsigned int virt_irq,
531 				      struct irq_desc *desc)
532 {
533 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
534 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
535 
536 	data->pre_handler(ino, data->arg1, data->arg2);
537 
538 	handle_fasteoi_irq(virt_irq, desc);
539 }
540 
541 void irq_install_pre_handler(int virt_irq,
542 			     void (*func)(unsigned int, void *, void *),
543 			     void *arg1, void *arg2)
544 {
545 	struct irq_handler_data *data = get_irq_chip_data(virt_irq);
546 	struct irq_desc *desc = irq_desc + virt_irq;
547 
548 	data->pre_handler = func;
549 	data->arg1 = arg1;
550 	data->arg2 = arg2;
551 
552 	desc->handle_irq = pre_flow_handler;
553 }
554 
555 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
556 {
557 	struct ino_bucket *bucket;
558 	struct irq_handler_data *data;
559 	unsigned int virt_irq;
560 	int ino;
561 
562 	BUG_ON(tlb_type == hypervisor);
563 
564 	ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
565 	bucket = &ivector_table[ino];
566 	virt_irq = bucket_get_virt_irq(__pa(bucket));
567 	if (!virt_irq) {
568 		virt_irq = virt_irq_alloc(0, ino);
569 		bucket_set_virt_irq(__pa(bucket), virt_irq);
570 		set_irq_chip_and_handler_name(virt_irq,
571 					      &sun4u_irq,
572 					      handle_fasteoi_irq,
573 					      "IVEC");
574 	}
575 
576 	data = get_irq_chip_data(virt_irq);
577 	if (unlikely(data))
578 		goto out;
579 
580 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
581 	if (unlikely(!data)) {
582 		prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
583 		prom_halt();
584 	}
585 	set_irq_chip_data(virt_irq, data);
586 
587 	data->imap  = imap;
588 	data->iclr  = iclr;
589 
590 out:
591 	return virt_irq;
592 }
593 
594 static unsigned int sun4v_build_common(unsigned long sysino,
595 				       struct irq_chip *chip)
596 {
597 	struct ino_bucket *bucket;
598 	struct irq_handler_data *data;
599 	unsigned int virt_irq;
600 
601 	BUG_ON(tlb_type != hypervisor);
602 
603 	bucket = &ivector_table[sysino];
604 	virt_irq = bucket_get_virt_irq(__pa(bucket));
605 	if (!virt_irq) {
606 		virt_irq = virt_irq_alloc(0, sysino);
607 		bucket_set_virt_irq(__pa(bucket), virt_irq);
608 		set_irq_chip_and_handler_name(virt_irq, chip,
609 					      handle_fasteoi_irq,
610 					      "IVEC");
611 	}
612 
613 	data = get_irq_chip_data(virt_irq);
614 	if (unlikely(data))
615 		goto out;
616 
617 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
618 	if (unlikely(!data)) {
619 		prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
620 		prom_halt();
621 	}
622 	set_irq_chip_data(virt_irq, data);
623 
624 	/* Catch accidental accesses to these things.  IMAP/ICLR handling
625 	 * is done by hypervisor calls on sun4v platforms, not by direct
626 	 * register accesses.
627 	 */
628 	data->imap = ~0UL;
629 	data->iclr = ~0UL;
630 
631 out:
632 	return virt_irq;
633 }
634 
635 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
636 {
637 	unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
638 
639 	return sun4v_build_common(sysino, &sun4v_irq);
640 }
641 
642 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
643 {
644 	struct irq_handler_data *data;
645 	unsigned long hv_err, cookie;
646 	struct ino_bucket *bucket;
647 	struct irq_desc *desc;
648 	unsigned int virt_irq;
649 
650 	bucket = kzalloc(sizeof(struct ino_bucket), GFP_ATOMIC);
651 	if (unlikely(!bucket))
652 		return 0;
653 
654 	/* The only reference we store to the IRQ bucket is
655 	 * by physical address which kmemleak can't see, tell
656 	 * it that this object explicitly is not a leak and
657 	 * should be scanned.
658 	 */
659 	kmemleak_not_leak(bucket);
660 
661 	__flush_dcache_range((unsigned long) bucket,
662 			     ((unsigned long) bucket +
663 			      sizeof(struct ino_bucket)));
664 
665 	virt_irq = virt_irq_alloc(devhandle, devino);
666 	bucket_set_virt_irq(__pa(bucket), virt_irq);
667 
668 	set_irq_chip_and_handler_name(virt_irq, &sun4v_virq,
669 				      handle_fasteoi_irq,
670 				      "IVEC");
671 
672 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
673 	if (unlikely(!data))
674 		return 0;
675 
676 	/* In order to make the LDC channel startup sequence easier,
677 	 * especially wrt. locking, we do not let request_irq() enable
678 	 * the interrupt.
679 	 */
680 	desc = irq_desc + virt_irq;
681 	desc->status |= IRQ_NOAUTOEN;
682 
683 	set_irq_chip_data(virt_irq, data);
684 
685 	/* Catch accidental accesses to these things.  IMAP/ICLR handling
686 	 * is done by hypervisor calls on sun4v platforms, not by direct
687 	 * register accesses.
688 	 */
689 	data->imap = ~0UL;
690 	data->iclr = ~0UL;
691 
692 	cookie = ~__pa(bucket);
693 	hv_err = sun4v_vintr_set_cookie(devhandle, devino, cookie);
694 	if (hv_err) {
695 		prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
696 			    "err=%lu\n", devhandle, devino, hv_err);
697 		prom_halt();
698 	}
699 
700 	return virt_irq;
701 }
702 
703 void ack_bad_irq(unsigned int virt_irq)
704 {
705 	unsigned int ino = virt_irq_table[virt_irq].dev_ino;
706 
707 	if (!ino)
708 		ino = 0xdeadbeef;
709 
710 	printk(KERN_CRIT "Unexpected IRQ from ino[%x] virt_irq[%u]\n",
711 	       ino, virt_irq);
712 }
713 
714 void *hardirq_stack[NR_CPUS];
715 void *softirq_stack[NR_CPUS];
716 
717 void __irq_entry handler_irq(int irq, struct pt_regs *regs)
718 {
719 	unsigned long pstate, bucket_pa;
720 	struct pt_regs *old_regs;
721 	void *orig_sp;
722 
723 	clear_softint(1 << irq);
724 
725 	old_regs = set_irq_regs(regs);
726 	irq_enter();
727 
728 	/* Grab an atomic snapshot of the pending IVECs.  */
729 	__asm__ __volatile__("rdpr	%%pstate, %0\n\t"
730 			     "wrpr	%0, %3, %%pstate\n\t"
731 			     "ldx	[%2], %1\n\t"
732 			     "stx	%%g0, [%2]\n\t"
733 			     "wrpr	%0, 0x0, %%pstate\n\t"
734 			     : "=&r" (pstate), "=&r" (bucket_pa)
735 			     : "r" (irq_work_pa(smp_processor_id())),
736 			       "i" (PSTATE_IE)
737 			     : "memory");
738 
739 	orig_sp = set_hardirq_stack();
740 
741 	while (bucket_pa) {
742 		struct irq_desc *desc;
743 		unsigned long next_pa;
744 		unsigned int virt_irq;
745 
746 		next_pa = bucket_get_chain_pa(bucket_pa);
747 		virt_irq = bucket_get_virt_irq(bucket_pa);
748 		bucket_clear_chain_pa(bucket_pa);
749 
750 		desc = irq_desc + virt_irq;
751 
752 		if (!(desc->status & IRQ_DISABLED))
753 			desc->handle_irq(virt_irq, desc);
754 
755 		bucket_pa = next_pa;
756 	}
757 
758 	restore_hardirq_stack(orig_sp);
759 
760 	irq_exit();
761 	set_irq_regs(old_regs);
762 }
763 
764 void do_softirq(void)
765 {
766 	unsigned long flags;
767 
768 	if (in_interrupt())
769 		return;
770 
771 	local_irq_save(flags);
772 
773 	if (local_softirq_pending()) {
774 		void *orig_sp, *sp = softirq_stack[smp_processor_id()];
775 
776 		sp += THREAD_SIZE - 192 - STACK_BIAS;
777 
778 		__asm__ __volatile__("mov %%sp, %0\n\t"
779 				     "mov %1, %%sp"
780 				     : "=&r" (orig_sp)
781 				     : "r" (sp));
782 		__do_softirq();
783 		__asm__ __volatile__("mov %0, %%sp"
784 				     : : "r" (orig_sp));
785 	}
786 
787 	local_irq_restore(flags);
788 }
789 
790 #ifdef CONFIG_HOTPLUG_CPU
791 void fixup_irqs(void)
792 {
793 	unsigned int irq;
794 
795 	for (irq = 0; irq < NR_IRQS; irq++) {
796 		unsigned long flags;
797 
798 		raw_spin_lock_irqsave(&irq_desc[irq].lock, flags);
799 		if (irq_desc[irq].action &&
800 		    !(irq_desc[irq].status & IRQ_PER_CPU)) {
801 			if (irq_desc[irq].chip->set_affinity)
802 				irq_desc[irq].chip->set_affinity(irq,
803 					irq_desc[irq].affinity);
804 		}
805 		raw_spin_unlock_irqrestore(&irq_desc[irq].lock, flags);
806 	}
807 
808 	tick_ops->disable_irq();
809 }
810 #endif
811 
812 struct sun5_timer {
813 	u64	count0;
814 	u64	limit0;
815 	u64	count1;
816 	u64	limit1;
817 };
818 
819 static struct sun5_timer *prom_timers;
820 static u64 prom_limit0, prom_limit1;
821 
822 static void map_prom_timers(void)
823 {
824 	struct device_node *dp;
825 	const unsigned int *addr;
826 
827 	/* PROM timer node hangs out in the top level of device siblings... */
828 	dp = of_find_node_by_path("/");
829 	dp = dp->child;
830 	while (dp) {
831 		if (!strcmp(dp->name, "counter-timer"))
832 			break;
833 		dp = dp->sibling;
834 	}
835 
836 	/* Assume if node is not present, PROM uses different tick mechanism
837 	 * which we should not care about.
838 	 */
839 	if (!dp) {
840 		prom_timers = (struct sun5_timer *) 0;
841 		return;
842 	}
843 
844 	/* If PROM is really using this, it must be mapped by him. */
845 	addr = of_get_property(dp, "address", NULL);
846 	if (!addr) {
847 		prom_printf("PROM does not have timer mapped, trying to continue.\n");
848 		prom_timers = (struct sun5_timer *) 0;
849 		return;
850 	}
851 	prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
852 }
853 
854 static void kill_prom_timer(void)
855 {
856 	if (!prom_timers)
857 		return;
858 
859 	/* Save them away for later. */
860 	prom_limit0 = prom_timers->limit0;
861 	prom_limit1 = prom_timers->limit1;
862 
863 	/* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
864 	 * We turn both off here just to be paranoid.
865 	 */
866 	prom_timers->limit0 = 0;
867 	prom_timers->limit1 = 0;
868 
869 	/* Wheee, eat the interrupt packet too... */
870 	__asm__ __volatile__(
871 "	mov	0x40, %%g2\n"
872 "	ldxa	[%%g0] %0, %%g1\n"
873 "	ldxa	[%%g2] %1, %%g1\n"
874 "	stxa	%%g0, [%%g0] %0\n"
875 "	membar	#Sync\n"
876 	: /* no outputs */
877 	: "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
878 	: "g1", "g2");
879 }
880 
881 void notrace init_irqwork_curcpu(void)
882 {
883 	int cpu = hard_smp_processor_id();
884 
885 	trap_block[cpu].irq_worklist_pa = 0UL;
886 }
887 
888 /* Please be very careful with register_one_mondo() and
889  * sun4v_register_mondo_queues().
890  *
891  * On SMP this gets invoked from the CPU trampoline before
892  * the cpu has fully taken over the trap table from OBP,
893  * and it's kernel stack + %g6 thread register state is
894  * not fully cooked yet.
895  *
896  * Therefore you cannot make any OBP calls, not even prom_printf,
897  * from these two routines.
898  */
899 static void __cpuinit notrace register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
900 {
901 	unsigned long num_entries = (qmask + 1) / 64;
902 	unsigned long status;
903 
904 	status = sun4v_cpu_qconf(type, paddr, num_entries);
905 	if (status != HV_EOK) {
906 		prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
907 			    "err %lu\n", type, paddr, num_entries, status);
908 		prom_halt();
909 	}
910 }
911 
912 void __cpuinit notrace sun4v_register_mondo_queues(int this_cpu)
913 {
914 	struct trap_per_cpu *tb = &trap_block[this_cpu];
915 
916 	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
917 			   tb->cpu_mondo_qmask);
918 	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
919 			   tb->dev_mondo_qmask);
920 	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
921 			   tb->resum_qmask);
922 	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
923 			   tb->nonresum_qmask);
924 }
925 
926 /* Each queue region must be a power of 2 multiple of 64 bytes in
927  * size.  The base real address must be aligned to the size of the
928  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
929  */
930 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
931 {
932 	unsigned long size = PAGE_ALIGN(qmask + 1);
933 	unsigned long order = get_order(size);
934 	unsigned long p;
935 
936 	p = __get_free_pages(GFP_KERNEL, order);
937 	if (!p) {
938 		prom_printf("SUN4V: Error, cannot allocate queue.\n");
939 		prom_halt();
940 	}
941 
942 	*pa_ptr = __pa(p);
943 }
944 
945 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
946 {
947 #ifdef CONFIG_SMP
948 	unsigned long page;
949 
950 	BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
951 
952 	page = get_zeroed_page(GFP_KERNEL);
953 	if (!page) {
954 		prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
955 		prom_halt();
956 	}
957 
958 	tb->cpu_mondo_block_pa = __pa(page);
959 	tb->cpu_list_pa = __pa(page + 64);
960 #endif
961 }
962 
963 /* Allocate mondo and error queues for all possible cpus.  */
964 static void __init sun4v_init_mondo_queues(void)
965 {
966 	int cpu;
967 
968 	for_each_possible_cpu(cpu) {
969 		struct trap_per_cpu *tb = &trap_block[cpu];
970 
971 		alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
972 		alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
973 		alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
974 		alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
975 		alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
976 		alloc_one_queue(&tb->nonresum_kernel_buf_pa,
977 				tb->nonresum_qmask);
978 	}
979 }
980 
981 static void __init init_send_mondo_info(void)
982 {
983 	int cpu;
984 
985 	for_each_possible_cpu(cpu) {
986 		struct trap_per_cpu *tb = &trap_block[cpu];
987 
988 		init_cpu_send_mondo_info(tb);
989 	}
990 }
991 
992 static struct irqaction timer_irq_action = {
993 	.name = "timer",
994 };
995 
996 /* Only invoked on boot processor. */
997 void __init init_IRQ(void)
998 {
999 	unsigned long size;
1000 
1001 	map_prom_timers();
1002 	kill_prom_timer();
1003 
1004 	size = sizeof(struct ino_bucket) * NUM_IVECS;
1005 	ivector_table = kzalloc(size, GFP_KERNEL);
1006 	if (!ivector_table) {
1007 		prom_printf("Fatal error, cannot allocate ivector_table\n");
1008 		prom_halt();
1009 	}
1010 	__flush_dcache_range((unsigned long) ivector_table,
1011 			     ((unsigned long) ivector_table) + size);
1012 
1013 	ivector_table_pa = __pa(ivector_table);
1014 
1015 	if (tlb_type == hypervisor)
1016 		sun4v_init_mondo_queues();
1017 
1018 	init_send_mondo_info();
1019 
1020 	if (tlb_type == hypervisor) {
1021 		/* Load up the boot cpu's entries.  */
1022 		sun4v_register_mondo_queues(hard_smp_processor_id());
1023 	}
1024 
1025 	/* We need to clear any IRQ's pending in the soft interrupt
1026 	 * registers, a spurious one could be left around from the
1027 	 * PROM timer which we just disabled.
1028 	 */
1029 	clear_softint(get_softint());
1030 
1031 	/* Now that ivector table is initialized, it is safe
1032 	 * to receive IRQ vector traps.  We will normally take
1033 	 * one or two right now, in case some device PROM used
1034 	 * to boot us wants to speak to us.  We just ignore them.
1035 	 */
1036 	__asm__ __volatile__("rdpr	%%pstate, %%g1\n\t"
1037 			     "or	%%g1, %0, %%g1\n\t"
1038 			     "wrpr	%%g1, 0x0, %%pstate"
1039 			     : /* No outputs */
1040 			     : "i" (PSTATE_IE)
1041 			     : "g1");
1042 
1043 	irq_desc[0].action = &timer_irq_action;
1044 }
1045