[SPARC64]: More sensible udelay implementation.
[safe/jmp/linux-2.6] / arch / sparc64 / kernel / irq.c
1 /* irq.c: UltraSparc IRQ handling/init/registry.
2  *
3  * Copyright (C) 1997, 2007  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/ptrace.h>
11 #include <linux/errno.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/signal.h>
14 #include <linux/mm.h>
15 #include <linux/interrupt.h>
16 #include <linux/slab.h>
17 #include <linux/random.h>
18 #include <linux/init.h>
19 #include <linux/delay.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/bootmem.h>
23 #include <linux/irq.h>
24 #include <linux/msi.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/sbus.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
47 /* UPA nodes send interrupt packet to UltraSparc with first data reg
48  * value low 5 (7 on Starfire) bits holding the IRQ identifier being
49  * delivered.  We must translate this into a non-vector IRQ so we can
50  * set the softint on this cpu.
51  *
52  * To make processing these packets efficient and race free we use
53  * an array of irq buckets below.  The interrupt vector handler in
54  * entry.S feeds incoming packets into per-cpu pil-indexed lists.
55  * The IVEC handler does not need to act atomically, the PIL dispatch
56  * code uses CAS to get an atomic snapshot of the list and clear it
57  * at the same time.
58  *
59  * If you make changes to ino_bucket, please update hand coded assembler
60  * of the vectored interrupt trap handler(s) in entry.S and sun4v_ivec.S
61  */
62 struct ino_bucket {
63         /* Next handler in per-CPU IRQ worklist.  We know that
64          * bucket pointers have the high 32-bits clear, so to
65          * save space we only store the bits we need.
66          */
67 /*0x00*/unsigned int irq_chain;
68
69         /* Virtual interrupt number assigned to this INO.  */
70 /*0x04*/unsigned int virt_irq;
71 };
72
73 #define NUM_IVECS       (IMAP_INR + 1)
74 struct ino_bucket ivector_table[NUM_IVECS] __attribute__ ((aligned (SMP_CACHE_BYTES)));
75
76 #define __irq_ino(irq) \
77         (((struct ino_bucket *)(unsigned long)(irq)) - &ivector_table[0])
78 #define __bucket(irq) ((struct ino_bucket *)(unsigned long)(irq))
79 #define __irq(bucket) ((unsigned int)(unsigned long)(bucket))
80
81 /* This has to be in the main kernel image, it cannot be
82  * turned into per-cpu data.  The reason is that the main
83  * kernel image is locked into the TLB and this structure
84  * is accessed from the vectored interrupt trap handler.  If
85  * access to this structure takes a TLB miss it could cause
86  * the 5-level sparc v9 trap stack to overflow.
87  */
88 #define irq_work(__cpu) &(trap_block[(__cpu)].irq_worklist)
89
90 static unsigned int virt_to_real_irq_table[NR_IRQS];
91
92 static unsigned char virt_irq_alloc(unsigned int real_irq)
93 {
94         unsigned char ent;
95
96         BUILD_BUG_ON(NR_IRQS >= 256);
97
98         for (ent = 1; ent < NR_IRQS; ent++) {
99                 if (!virt_to_real_irq_table[ent])
100                         break;
101         }
102         if (ent >= NR_IRQS) {
103                 printk(KERN_ERR "IRQ: Out of virtual IRQs.\n");
104                 return 0;
105         }
106
107         virt_to_real_irq_table[ent] = real_irq;
108
109         return ent;
110 }
111
112 #ifdef CONFIG_PCI_MSI
113 static void virt_irq_free(unsigned int virt_irq)
114 {
115         unsigned int real_irq;
116
117         if (virt_irq >= NR_IRQS)
118                 return;
119
120         real_irq = virt_to_real_irq_table[virt_irq];
121         virt_to_real_irq_table[virt_irq] = 0;
122
123         __bucket(real_irq)->virt_irq = 0;
124 }
125 #endif
126
127 static unsigned int virt_to_real_irq(unsigned char virt_irq)
128 {
129         return virt_to_real_irq_table[virt_irq];
130 }
131
132 /*
133  * /proc/interrupts printing:
134  */
135
136 int show_interrupts(struct seq_file *p, void *v)
137 {
138         int i = *(loff_t *) v, j;
139         struct irqaction * action;
140         unsigned long flags;
141
142         if (i == 0) {
143                 seq_printf(p, "           ");
144                 for_each_online_cpu(j)
145                         seq_printf(p, "CPU%d       ",j);
146                 seq_putc(p, '\n');
147         }
148
149         if (i < NR_IRQS) {
150                 spin_lock_irqsave(&irq_desc[i].lock, flags);
151                 action = irq_desc[i].action;
152                 if (!action)
153                         goto skip;
154                 seq_printf(p, "%3d: ",i);
155 #ifndef CONFIG_SMP
156                 seq_printf(p, "%10u ", kstat_irqs(i));
157 #else
158                 for_each_online_cpu(j)
159                         seq_printf(p, "%10u ", kstat_cpu(j).irqs[i]);
160 #endif
161                 seq_printf(p, " %9s", irq_desc[i].chip->typename);
162                 seq_printf(p, "  %s", action->name);
163
164                 for (action=action->next; action; action = action->next)
165                         seq_printf(p, ", %s", action->name);
166
167                 seq_putc(p, '\n');
168 skip:
169                 spin_unlock_irqrestore(&irq_desc[i].lock, flags);
170         }
171         return 0;
172 }
173
174 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
175 {
176         unsigned int tid;
177
178         if (this_is_starfire) {
179                 tid = starfire_translate(imap, cpuid);
180                 tid <<= IMAP_TID_SHIFT;
181                 tid &= IMAP_TID_UPA;
182         } else {
183                 if (tlb_type == cheetah || tlb_type == cheetah_plus) {
184                         unsigned long ver;
185
186                         __asm__ ("rdpr %%ver, %0" : "=r" (ver));
187                         if ((ver >> 32UL) == __JALAPENO_ID ||
188                             (ver >> 32UL) == __SERRANO_ID) {
189                                 tid = cpuid << IMAP_TID_SHIFT;
190                                 tid &= IMAP_TID_JBUS;
191                         } else {
192                                 unsigned int a = cpuid & 0x1f;
193                                 unsigned int n = (cpuid >> 5) & 0x1f;
194
195                                 tid = ((a << IMAP_AID_SHIFT) |
196                                        (n << IMAP_NID_SHIFT));
197                                 tid &= (IMAP_AID_SAFARI |
198                                         IMAP_NID_SAFARI);;
199                         }
200                 } else {
201                         tid = cpuid << IMAP_TID_SHIFT;
202                         tid &= IMAP_TID_UPA;
203                 }
204         }
205
206         return tid;
207 }
208
209 struct irq_handler_data {
210         unsigned long   iclr;
211         unsigned long   imap;
212
213         void            (*pre_handler)(unsigned int, void *, void *);
214         void            *pre_handler_arg1;
215         void            *pre_handler_arg2;
216 };
217
218 static inline struct ino_bucket *virt_irq_to_bucket(unsigned int virt_irq)
219 {
220         unsigned int real_irq = virt_to_real_irq(virt_irq);
221         struct ino_bucket *bucket = NULL;
222
223         if (likely(real_irq))
224                 bucket = __bucket(real_irq);
225
226         return bucket;
227 }
228
229 #ifdef CONFIG_SMP
230 static int irq_choose_cpu(unsigned int virt_irq)
231 {
232         cpumask_t mask = irq_desc[virt_irq].affinity;
233         int cpuid;
234
235         if (cpus_equal(mask, CPU_MASK_ALL)) {
236                 static int irq_rover;
237                 static DEFINE_SPINLOCK(irq_rover_lock);
238                 unsigned long flags;
239
240                 /* Round-robin distribution... */
241         do_round_robin:
242                 spin_lock_irqsave(&irq_rover_lock, flags);
243
244                 while (!cpu_online(irq_rover)) {
245                         if (++irq_rover >= NR_CPUS)
246                                 irq_rover = 0;
247                 }
248                 cpuid = irq_rover;
249                 do {
250                         if (++irq_rover >= NR_CPUS)
251                                 irq_rover = 0;
252                 } while (!cpu_online(irq_rover));
253
254                 spin_unlock_irqrestore(&irq_rover_lock, flags);
255         } else {
256                 cpumask_t tmp;
257
258                 cpus_and(tmp, cpu_online_map, mask);
259
260                 if (cpus_empty(tmp))
261                         goto do_round_robin;
262
263                 cpuid = first_cpu(tmp);
264         }
265
266         return cpuid;
267 }
268 #else
269 static int irq_choose_cpu(unsigned int virt_irq)
270 {
271         return real_hard_smp_processor_id();
272 }
273 #endif
274
275 static void sun4u_irq_enable(unsigned int virt_irq)
276 {
277         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
278
279         if (likely(data)) {
280                 unsigned long cpuid, imap, val;
281                 unsigned int tid;
282
283                 cpuid = irq_choose_cpu(virt_irq);
284                 imap = data->imap;
285
286                 tid = sun4u_compute_tid(imap, cpuid);
287
288                 val = upa_readq(imap);
289                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
290                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
291                 val |= tid | IMAP_VALID;
292                 upa_writeq(val, imap);
293         }
294 }
295
296 static void sun4u_irq_disable(unsigned int virt_irq)
297 {
298         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
299
300         if (likely(data)) {
301                 unsigned long imap = data->imap;
302                 u32 tmp = upa_readq(imap);
303
304                 tmp &= ~IMAP_VALID;
305                 upa_writeq(tmp, imap);
306         }
307 }
308
309 static void sun4u_irq_end(unsigned int virt_irq)
310 {
311         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
312         struct irq_desc *desc = irq_desc + virt_irq;
313
314         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
315                 return;
316
317         if (likely(data))
318                 upa_writeq(ICLR_IDLE, data->iclr);
319 }
320
321 static void sun4v_irq_enable(unsigned int virt_irq)
322 {
323         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
324         unsigned int ino = bucket - &ivector_table[0];
325
326         if (likely(bucket)) {
327                 unsigned long cpuid;
328                 int err;
329
330                 cpuid = irq_choose_cpu(virt_irq);
331
332                 err = sun4v_intr_settarget(ino, cpuid);
333                 if (err != HV_EOK)
334                         printk("sun4v_intr_settarget(%x,%lu): err(%d)\n",
335                                ino, cpuid, err);
336                 err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
337                 if (err != HV_EOK)
338                         printk("sun4v_intr_setstate(%x): "
339                                "err(%d)\n", ino, err);
340                 err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
341                 if (err != HV_EOK)
342                         printk("sun4v_intr_setenabled(%x): err(%d)\n",
343                                ino, err);
344         }
345 }
346
347 static void sun4v_irq_disable(unsigned int virt_irq)
348 {
349         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
350         unsigned int ino = bucket - &ivector_table[0];
351
352         if (likely(bucket)) {
353                 int err;
354
355                 err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
356                 if (err != HV_EOK)
357                         printk("sun4v_intr_setenabled(%x): "
358                                "err(%d)\n", ino, err);
359         }
360 }
361
362 #ifdef CONFIG_PCI_MSI
363 static void sun4v_msi_enable(unsigned int virt_irq)
364 {
365         sun4v_irq_enable(virt_irq);
366         unmask_msi_irq(virt_irq);
367 }
368
369 static void sun4v_msi_disable(unsigned int virt_irq)
370 {
371         mask_msi_irq(virt_irq);
372         sun4v_irq_disable(virt_irq);
373 }
374 #endif
375
376 static void sun4v_irq_end(unsigned int virt_irq)
377 {
378         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
379         unsigned int ino = bucket - &ivector_table[0];
380         struct irq_desc *desc = irq_desc + virt_irq;
381
382         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
383                 return;
384
385         if (likely(bucket)) {
386                 int err;
387
388                 err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
389                 if (err != HV_EOK)
390                         printk("sun4v_intr_setstate(%x): "
391                                "err(%d)\n", ino, err);
392         }
393 }
394
395 static void sun4v_virq_enable(unsigned int virt_irq)
396 {
397         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
398         unsigned int ino = bucket - &ivector_table[0];
399
400         if (likely(bucket)) {
401                 unsigned long cpuid, dev_handle, dev_ino;
402                 int err;
403
404                 cpuid = irq_choose_cpu(virt_irq);
405
406                 dev_handle = ino & IMAP_IGN;
407                 dev_ino = ino & IMAP_INO;
408
409                 err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
410                 if (err != HV_EOK)
411                         printk("sun4v_vintr_set_target(%lx,%lx,%lu): "
412                                "err(%d)\n",
413                                dev_handle, dev_ino, cpuid, err);
414                 err = sun4v_vintr_set_state(dev_handle, dev_ino,
415                                             HV_INTR_STATE_IDLE);
416                 if (err != HV_EOK)
417                         printk("sun4v_vintr_set_state(%lx,%lx,"
418                                 "HV_INTR_STATE_IDLE): err(%d)\n",
419                                dev_handle, dev_ino, err);
420                 err = sun4v_vintr_set_valid(dev_handle, dev_ino,
421                                             HV_INTR_ENABLED);
422                 if (err != HV_EOK)
423                         printk("sun4v_vintr_set_state(%lx,%lx,"
424                                "HV_INTR_ENABLED): err(%d)\n",
425                                dev_handle, dev_ino, err);
426         }
427 }
428
429 static void sun4v_virq_disable(unsigned int virt_irq)
430 {
431         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
432         unsigned int ino = bucket - &ivector_table[0];
433
434         if (likely(bucket)) {
435                 unsigned long dev_handle, dev_ino;
436                 int err;
437
438                 dev_handle = ino & IMAP_IGN;
439                 dev_ino = ino & IMAP_INO;
440
441                 err = sun4v_vintr_set_valid(dev_handle, dev_ino,
442                                             HV_INTR_DISABLED);
443                 if (err != HV_EOK)
444                         printk("sun4v_vintr_set_state(%lx,%lx,"
445                                "HV_INTR_DISABLED): err(%d)\n",
446                                dev_handle, dev_ino, err);
447         }
448 }
449
450 static void sun4v_virq_end(unsigned int virt_irq)
451 {
452         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
453         unsigned int ino = bucket - &ivector_table[0];
454         struct irq_desc *desc = irq_desc + virt_irq;
455
456         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
457                 return;
458
459         if (likely(bucket)) {
460                 unsigned long dev_handle, dev_ino;
461                 int err;
462
463                 dev_handle = ino & IMAP_IGN;
464                 dev_ino = ino & IMAP_INO;
465
466                 err = sun4v_vintr_set_state(dev_handle, dev_ino,
467                                             HV_INTR_STATE_IDLE);
468                 if (err != HV_EOK)
469                         printk("sun4v_vintr_set_state(%lx,%lx,"
470                                 "HV_INTR_STATE_IDLE): err(%d)\n",
471                                dev_handle, dev_ino, err);
472         }
473 }
474
475 static void run_pre_handler(unsigned int virt_irq)
476 {
477         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
478         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
479
480         if (likely(data->pre_handler)) {
481                 data->pre_handler(__irq_ino(__irq(bucket)),
482                                   data->pre_handler_arg1,
483                                   data->pre_handler_arg2);
484         }
485 }
486
487 static struct irq_chip sun4u_irq = {
488         .typename       = "sun4u",
489         .enable         = sun4u_irq_enable,
490         .disable        = sun4u_irq_disable,
491         .end            = sun4u_irq_end,
492 };
493
494 static struct irq_chip sun4u_irq_ack = {
495         .typename       = "sun4u+ack",
496         .enable         = sun4u_irq_enable,
497         .disable        = sun4u_irq_disable,
498         .ack            = run_pre_handler,
499         .end            = sun4u_irq_end,
500 };
501
502 static struct irq_chip sun4v_irq = {
503         .typename       = "sun4v",
504         .enable         = sun4v_irq_enable,
505         .disable        = sun4v_irq_disable,
506         .end            = sun4v_irq_end,
507 };
508
509 static struct irq_chip sun4v_irq_ack = {
510         .typename       = "sun4v+ack",
511         .enable         = sun4v_irq_enable,
512         .disable        = sun4v_irq_disable,
513         .ack            = run_pre_handler,
514         .end            = sun4v_irq_end,
515 };
516
517 #ifdef CONFIG_PCI_MSI
518 static struct irq_chip sun4v_msi = {
519         .typename       = "sun4v+msi",
520         .mask           = mask_msi_irq,
521         .unmask         = unmask_msi_irq,
522         .enable         = sun4v_msi_enable,
523         .disable        = sun4v_msi_disable,
524         .ack            = run_pre_handler,
525         .end            = sun4v_irq_end,
526 };
527 #endif
528
529 static struct irq_chip sun4v_virq = {
530         .typename       = "vsun4v",
531         .enable         = sun4v_virq_enable,
532         .disable        = sun4v_virq_disable,
533         .end            = sun4v_virq_end,
534 };
535
536 static struct irq_chip sun4v_virq_ack = {
537         .typename       = "vsun4v+ack",
538         .enable         = sun4v_virq_enable,
539         .disable        = sun4v_virq_disable,
540         .ack            = run_pre_handler,
541         .end            = sun4v_virq_end,
542 };
543
544 void irq_install_pre_handler(int virt_irq,
545                              void (*func)(unsigned int, void *, void *),
546                              void *arg1, void *arg2)
547 {
548         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
549         struct irq_chip *chip;
550
551         data->pre_handler = func;
552         data->pre_handler_arg1 = arg1;
553         data->pre_handler_arg2 = arg2;
554
555         chip = get_irq_chip(virt_irq);
556         if (chip == &sun4u_irq_ack ||
557             chip == &sun4v_irq_ack ||
558             chip == &sun4v_virq_ack
559 #ifdef CONFIG_PCI_MSI
560             || chip == &sun4v_msi
561 #endif
562             )
563                 return;
564
565         chip = (chip == &sun4u_irq ?
566                 &sun4u_irq_ack :
567                 (chip == &sun4v_irq ?
568                  &sun4v_irq_ack : &sun4v_virq_ack));
569         set_irq_chip(virt_irq, chip);
570 }
571
572 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
573 {
574         struct ino_bucket *bucket;
575         struct irq_handler_data *data;
576         int ino;
577
578         BUG_ON(tlb_type == hypervisor);
579
580         ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
581         bucket = &ivector_table[ino];
582         if (!bucket->virt_irq) {
583                 bucket->virt_irq = virt_irq_alloc(__irq(bucket));
584                 set_irq_chip(bucket->virt_irq, &sun4u_irq);
585         }
586
587         data = get_irq_chip_data(bucket->virt_irq);
588         if (unlikely(data))
589                 goto out;
590
591         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
592         if (unlikely(!data)) {
593                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
594                 prom_halt();
595         }
596         set_irq_chip_data(bucket->virt_irq, data);
597
598         data->imap  = imap;
599         data->iclr  = iclr;
600
601 out:
602         return bucket->virt_irq;
603 }
604
605 static unsigned int sun4v_build_common(unsigned long sysino,
606                                        struct irq_chip *chip)
607 {
608         struct ino_bucket *bucket;
609         struct irq_handler_data *data;
610
611         BUG_ON(tlb_type != hypervisor);
612
613         bucket = &ivector_table[sysino];
614         if (!bucket->virt_irq) {
615                 bucket->virt_irq = virt_irq_alloc(__irq(bucket));
616                 set_irq_chip(bucket->virt_irq, chip);
617         }
618
619         data = get_irq_chip_data(bucket->virt_irq);
620         if (unlikely(data))
621                 goto out;
622
623         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
624         if (unlikely(!data)) {
625                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
626                 prom_halt();
627         }
628         set_irq_chip_data(bucket->virt_irq, data);
629
630         /* Catch accidental accesses to these things.  IMAP/ICLR handling
631          * is done by hypervisor calls on sun4v platforms, not by direct
632          * register accesses.
633          */
634         data->imap = ~0UL;
635         data->iclr = ~0UL;
636
637 out:
638         return bucket->virt_irq;
639 }
640
641 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
642 {
643         unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
644
645         return sun4v_build_common(sysino, &sun4v_irq);
646 }
647
648 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
649 {
650         unsigned long sysino, hv_err;
651
652         BUG_ON(devhandle & ~IMAP_IGN);
653         BUG_ON(devino & ~IMAP_INO);
654
655         sysino = devhandle | devino;
656
657         hv_err = sun4v_vintr_set_cookie(devhandle, devino, sysino);
658         if (hv_err) {
659                 prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
660                             "err=%lu\n", devhandle, devino, hv_err);
661                 prom_halt();
662         }
663
664         return sun4v_build_common(sysino, &sun4v_virq);
665 }
666
667 #ifdef CONFIG_PCI_MSI
668 unsigned int sun4v_build_msi(u32 devhandle, unsigned int *virt_irq_p,
669                              unsigned int msi_start, unsigned int msi_end)
670 {
671         struct ino_bucket *bucket;
672         struct irq_handler_data *data;
673         unsigned long sysino;
674         unsigned int devino;
675
676         BUG_ON(tlb_type != hypervisor);
677
678         /* Find a free devino in the given range.  */
679         for (devino = msi_start; devino < msi_end; devino++) {
680                 sysino = sun4v_devino_to_sysino(devhandle, devino);
681                 bucket = &ivector_table[sysino];
682                 if (!bucket->virt_irq)
683                         break;
684         }
685         if (devino >= msi_end)
686                 return 0;
687
688         sysino = sun4v_devino_to_sysino(devhandle, devino);
689         bucket = &ivector_table[sysino];
690         bucket->virt_irq = virt_irq_alloc(__irq(bucket));
691         *virt_irq_p = bucket->virt_irq;
692         set_irq_chip(bucket->virt_irq, &sun4v_msi);
693
694         data = get_irq_chip_data(bucket->virt_irq);
695         if (unlikely(data))
696                 return devino;
697
698         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
699         if (unlikely(!data)) {
700                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
701                 prom_halt();
702         }
703         set_irq_chip_data(bucket->virt_irq, data);
704
705         data->imap = ~0UL;
706         data->iclr = ~0UL;
707
708         return devino;
709 }
710
711 void sun4v_destroy_msi(unsigned int virt_irq)
712 {
713         virt_irq_free(virt_irq);
714 }
715 #endif
716
717 void ack_bad_irq(unsigned int virt_irq)
718 {
719         struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
720         unsigned int ino = 0xdeadbeef;
721
722         if (bucket)
723                 ino = bucket - &ivector_table[0];
724
725         printk(KERN_CRIT "Unexpected IRQ from ino[%x] virt_irq[%u]\n",
726                ino, virt_irq);
727 }
728
729 void handler_irq(int irq, struct pt_regs *regs)
730 {
731         struct ino_bucket *bucket;
732         struct pt_regs *old_regs;
733
734         clear_softint(1 << irq);
735
736         old_regs = set_irq_regs(regs);
737         irq_enter();
738
739         /* Sliiiick... */
740         bucket = __bucket(xchg32(irq_work(smp_processor_id()), 0));
741         while (bucket) {
742                 struct ino_bucket *next = __bucket(bucket->irq_chain);
743
744                 bucket->irq_chain = 0;
745                 __do_IRQ(bucket->virt_irq);
746
747                 bucket = next;
748         }
749
750         irq_exit();
751         set_irq_regs(old_regs);
752 }
753
754 struct sun5_timer {
755         u64     count0;
756         u64     limit0;
757         u64     count1;
758         u64     limit1;
759 };
760
761 static struct sun5_timer *prom_timers;
762 static u64 prom_limit0, prom_limit1;
763
764 static void map_prom_timers(void)
765 {
766         struct device_node *dp;
767         const unsigned int *addr;
768
769         /* PROM timer node hangs out in the top level of device siblings... */
770         dp = of_find_node_by_path("/");
771         dp = dp->child;
772         while (dp) {
773                 if (!strcmp(dp->name, "counter-timer"))
774                         break;
775                 dp = dp->sibling;
776         }
777
778         /* Assume if node is not present, PROM uses different tick mechanism
779          * which we should not care about.
780          */
781         if (!dp) {
782                 prom_timers = (struct sun5_timer *) 0;
783                 return;
784         }
785
786         /* If PROM is really using this, it must be mapped by him. */
787         addr = of_get_property(dp, "address", NULL);
788         if (!addr) {
789                 prom_printf("PROM does not have timer mapped, trying to continue.\n");
790                 prom_timers = (struct sun5_timer *) 0;
791                 return;
792         }
793         prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
794 }
795
796 static void kill_prom_timer(void)
797 {
798         if (!prom_timers)
799                 return;
800
801         /* Save them away for later. */
802         prom_limit0 = prom_timers->limit0;
803         prom_limit1 = prom_timers->limit1;
804
805         /* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
806          * We turn both off here just to be paranoid.
807          */
808         prom_timers->limit0 = 0;
809         prom_timers->limit1 = 0;
810
811         /* Wheee, eat the interrupt packet too... */
812         __asm__ __volatile__(
813 "       mov     0x40, %%g2\n"
814 "       ldxa    [%%g0] %0, %%g1\n"
815 "       ldxa    [%%g2] %1, %%g1\n"
816 "       stxa    %%g0, [%%g0] %0\n"
817 "       membar  #Sync\n"
818         : /* no outputs */
819         : "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
820         : "g1", "g2");
821 }
822
823 void init_irqwork_curcpu(void)
824 {
825         int cpu = hard_smp_processor_id();
826
827         trap_block[cpu].irq_worklist = 0;
828 }
829
830 /* Please be very careful with register_one_mondo() and
831  * sun4v_register_mondo_queues().
832  *
833  * On SMP this gets invoked from the CPU trampoline before
834  * the cpu has fully taken over the trap table from OBP,
835  * and it's kernel stack + %g6 thread register state is
836  * not fully cooked yet.
837  *
838  * Therefore you cannot make any OBP calls, not even prom_printf,
839  * from these two routines.
840  */
841 static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
842 {
843         unsigned long num_entries = (qmask + 1) / 64;
844         unsigned long status;
845
846         status = sun4v_cpu_qconf(type, paddr, num_entries);
847         if (status != HV_EOK) {
848                 prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
849                             "err %lu\n", type, paddr, num_entries, status);
850                 prom_halt();
851         }
852 }
853
854 static void __cpuinit sun4v_register_mondo_queues(int this_cpu)
855 {
856         struct trap_per_cpu *tb = &trap_block[this_cpu];
857
858         register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
859                            tb->cpu_mondo_qmask);
860         register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
861                            tb->dev_mondo_qmask);
862         register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
863                            tb->resum_qmask);
864         register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
865                            tb->nonresum_qmask);
866 }
867
868 static void __cpuinit alloc_one_mondo(unsigned long *pa_ptr, unsigned long qmask, int use_bootmem)
869 {
870         unsigned long size = PAGE_ALIGN(qmask + 1);
871         unsigned long order = get_order(size);
872         void *p = NULL;
873
874         if (use_bootmem) {
875                 p = __alloc_bootmem_low(size, size, 0);
876         } else {
877                 struct page *page = alloc_pages(GFP_ATOMIC | __GFP_ZERO, order);
878                 if (page)
879                         p = page_address(page);
880         }
881
882         if (!p) {
883                 prom_printf("SUN4V: Error, cannot allocate mondo queue.\n");
884                 prom_halt();
885         }
886
887         *pa_ptr = __pa(p);
888 }
889
890 static void __cpuinit alloc_one_kbuf(unsigned long *pa_ptr, unsigned long qmask, int use_bootmem)
891 {
892         unsigned long size = PAGE_ALIGN(qmask + 1);
893         unsigned long order = get_order(size);
894         void *p = NULL;
895
896         if (use_bootmem) {
897                 p = __alloc_bootmem_low(size, size, 0);
898         } else {
899                 struct page *page = alloc_pages(GFP_ATOMIC | __GFP_ZERO, order);
900                 if (page)
901                         p = page_address(page);
902         }
903
904         if (!p) {
905                 prom_printf("SUN4V: Error, cannot allocate kbuf page.\n");
906                 prom_halt();
907         }
908
909         *pa_ptr = __pa(p);
910 }
911
912 static void __cpuinit init_cpu_send_mondo_info(struct trap_per_cpu *tb, int use_bootmem)
913 {
914 #ifdef CONFIG_SMP
915         void *page;
916
917         BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
918
919         if (use_bootmem)
920                 page = alloc_bootmem_low_pages(PAGE_SIZE);
921         else
922                 page = (void *) get_zeroed_page(GFP_ATOMIC);
923
924         if (!page) {
925                 prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
926                 prom_halt();
927         }
928
929         tb->cpu_mondo_block_pa = __pa(page);
930         tb->cpu_list_pa = __pa(page + 64);
931 #endif
932 }
933
934 /* Allocate and register the mondo and error queues for this cpu.  */
935 void __cpuinit sun4v_init_mondo_queues(int use_bootmem, int cpu, int alloc, int load)
936 {
937         struct trap_per_cpu *tb = &trap_block[cpu];
938
939         if (alloc) {
940                 alloc_one_mondo(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask, use_bootmem);
941                 alloc_one_mondo(&tb->dev_mondo_pa, tb->dev_mondo_qmask, use_bootmem);
942                 alloc_one_mondo(&tb->resum_mondo_pa, tb->resum_qmask, use_bootmem);
943                 alloc_one_kbuf(&tb->resum_kernel_buf_pa, tb->resum_qmask, use_bootmem);
944                 alloc_one_mondo(&tb->nonresum_mondo_pa, tb->nonresum_qmask, use_bootmem);
945                 alloc_one_kbuf(&tb->nonresum_kernel_buf_pa, tb->nonresum_qmask, use_bootmem);
946
947                 init_cpu_send_mondo_info(tb, use_bootmem);
948         }
949
950         if (load) {
951                 if (cpu != hard_smp_processor_id()) {
952                         prom_printf("SUN4V: init mondo on cpu %d not %d\n",
953                                     cpu, hard_smp_processor_id());
954                         prom_halt();
955                 }
956                 sun4v_register_mondo_queues(cpu);
957         }
958 }
959
960 static struct irqaction timer_irq_action = {
961         .name = "timer",
962 };
963
964 /* Only invoked on boot processor. */
965 void __init init_IRQ(void)
966 {
967         map_prom_timers();
968         kill_prom_timer();
969         memset(&ivector_table[0], 0, sizeof(ivector_table));
970
971         if (tlb_type == hypervisor)
972                 sun4v_init_mondo_queues(1, hard_smp_processor_id(), 1, 1);
973
974         /* We need to clear any IRQ's pending in the soft interrupt
975          * registers, a spurious one could be left around from the
976          * PROM timer which we just disabled.
977          */
978         clear_softint(get_softint());
979
980         /* Now that ivector table is initialized, it is safe
981          * to receive IRQ vector traps.  We will normally take
982          * one or two right now, in case some device PROM used
983          * to boot us wants to speak to us.  We just ignore them.
984          */
985         __asm__ __volatile__("rdpr      %%pstate, %%g1\n\t"
986                              "or        %%g1, %0, %%g1\n\t"
987                              "wrpr      %%g1, 0x0, %%pstate"
988                              : /* No outputs */
989                              : "i" (PSTATE_IE)
990                              : "g1");
991
992         irq_desc[0].action = &timer_irq_action;
993 }