Merge branch 'cuse' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse
[linux-2.6] / arch / sparc / kernel / irq_64.c
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 int sun4u_set_affinity(unsigned int virt_irq,
322                                const struct cpumask *mask)
323 {
324         sun4u_irq_enable(virt_irq);
325
326         return 0;
327 }
328
329 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
330  * handler_irq() will skip the handler call and that will leave the
331  * interrupt in the sent state.  The next ->enable() call will hit the
332  * ICLR register to reset the state machine.
333  *
334  * This scheme is necessary, instead of clearing the Valid bit in the
335  * IMAP register, to handle the case of IMAP registers being shared by
336  * multiple INOs (and thus ICLR registers).  Since we use a different
337  * virtual IRQ for each shared IMAP instance, the generic code thinks
338  * there is only one user so it prematurely calls ->disable() on
339  * free_irq().
340  *
341  * We have to provide an explicit ->disable() method instead of using
342  * NULL to get the default.  The reason is that if the generic code
343  * sees that, it also hooks up a default ->shutdown method which
344  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
345  */
346 static void sun4u_irq_disable(unsigned int virt_irq)
347 {
348 }
349
350 static void sun4u_irq_eoi(unsigned int virt_irq)
351 {
352         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
353         struct irq_desc *desc = irq_desc + virt_irq;
354
355         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
356                 return;
357
358         if (likely(data))
359                 upa_writeq(ICLR_IDLE, data->iclr);
360 }
361
362 static void sun4v_irq_enable(unsigned int virt_irq)
363 {
364         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
365         unsigned long cpuid = irq_choose_cpu(virt_irq);
366         int err;
367
368         err = sun4v_intr_settarget(ino, cpuid);
369         if (err != HV_EOK)
370                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
371                        "err(%d)\n", ino, cpuid, err);
372         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
373         if (err != HV_EOK)
374                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
375                        "err(%d)\n", ino, err);
376         err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
377         if (err != HV_EOK)
378                 printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
379                        ino, err);
380 }
381
382 static int sun4v_set_affinity(unsigned int virt_irq,
383                                const struct cpumask *mask)
384 {
385         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
386         unsigned long cpuid = irq_choose_cpu(virt_irq);
387         int err;
388
389         err = sun4v_intr_settarget(ino, cpuid);
390         if (err != HV_EOK)
391                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
392                        "err(%d)\n", ino, cpuid, err);
393
394         return 0;
395 }
396
397 static void sun4v_irq_disable(unsigned int virt_irq)
398 {
399         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
400         int err;
401
402         err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
403         if (err != HV_EOK)
404                 printk(KERN_ERR "sun4v_intr_setenabled(%x): "
405                        "err(%d)\n", ino, err);
406 }
407
408 static void sun4v_irq_eoi(unsigned int virt_irq)
409 {
410         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
411         struct irq_desc *desc = irq_desc + virt_irq;
412         int err;
413
414         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
415                 return;
416
417         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
418         if (err != HV_EOK)
419                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
420                        "err(%d)\n", ino, err);
421 }
422
423 static void sun4v_virq_enable(unsigned int virt_irq)
424 {
425         unsigned long cpuid, dev_handle, dev_ino;
426         int err;
427
428         cpuid = irq_choose_cpu(virt_irq);
429
430         dev_handle = virt_irq_table[virt_irq].dev_handle;
431         dev_ino = virt_irq_table[virt_irq].dev_ino;
432
433         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
434         if (err != HV_EOK)
435                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
436                        "err(%d)\n",
437                        dev_handle, dev_ino, cpuid, err);
438         err = sun4v_vintr_set_state(dev_handle, dev_ino,
439                                     HV_INTR_STATE_IDLE);
440         if (err != HV_EOK)
441                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
442                        "HV_INTR_STATE_IDLE): err(%d)\n",
443                        dev_handle, dev_ino, err);
444         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
445                                     HV_INTR_ENABLED);
446         if (err != HV_EOK)
447                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
448                        "HV_INTR_ENABLED): err(%d)\n",
449                        dev_handle, dev_ino, err);
450 }
451
452 static int sun4v_virt_set_affinity(unsigned int virt_irq,
453                                     const struct cpumask *mask)
454 {
455         unsigned long cpuid, dev_handle, dev_ino;
456         int err;
457
458         cpuid = irq_choose_cpu(virt_irq);
459
460         dev_handle = virt_irq_table[virt_irq].dev_handle;
461         dev_ino = virt_irq_table[virt_irq].dev_ino;
462
463         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
464         if (err != HV_EOK)
465                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
466                        "err(%d)\n",
467                        dev_handle, dev_ino, cpuid, err);
468
469         return 0;
470 }
471
472 static void sun4v_virq_disable(unsigned int virt_irq)
473 {
474         unsigned long dev_handle, dev_ino;
475         int err;
476
477         dev_handle = virt_irq_table[virt_irq].dev_handle;
478         dev_ino = virt_irq_table[virt_irq].dev_ino;
479
480         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
481                                     HV_INTR_DISABLED);
482         if (err != HV_EOK)
483                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
484                        "HV_INTR_DISABLED): err(%d)\n",
485                        dev_handle, dev_ino, err);
486 }
487
488 static void sun4v_virq_eoi(unsigned int virt_irq)
489 {
490         struct irq_desc *desc = irq_desc + virt_irq;
491         unsigned long dev_handle, dev_ino;
492         int err;
493
494         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
495                 return;
496
497         dev_handle = virt_irq_table[virt_irq].dev_handle;
498         dev_ino = virt_irq_table[virt_irq].dev_ino;
499
500         err = sun4v_vintr_set_state(dev_handle, dev_ino,
501                                     HV_INTR_STATE_IDLE);
502         if (err != HV_EOK)
503                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
504                        "HV_INTR_STATE_IDLE): err(%d)\n",
505                        dev_handle, dev_ino, err);
506 }
507
508 static struct irq_chip sun4u_irq = {
509         .typename       = "sun4u",
510         .enable         = sun4u_irq_enable,
511         .disable        = sun4u_irq_disable,
512         .eoi            = sun4u_irq_eoi,
513         .set_affinity   = sun4u_set_affinity,
514 };
515
516 static struct irq_chip sun4v_irq = {
517         .typename       = "sun4v",
518         .enable         = sun4v_irq_enable,
519         .disable        = sun4v_irq_disable,
520         .eoi            = sun4v_irq_eoi,
521         .set_affinity   = sun4v_set_affinity,
522 };
523
524 static struct irq_chip sun4v_virq = {
525         .typename       = "vsun4v",
526         .enable         = sun4v_virq_enable,
527         .disable        = sun4v_virq_disable,
528         .eoi            = sun4v_virq_eoi,
529         .set_affinity   = sun4v_virt_set_affinity,
530 };
531
532 static void pre_flow_handler(unsigned int virt_irq,
533                                       struct irq_desc *desc)
534 {
535         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
536         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
537
538         data->pre_handler(ino, data->arg1, data->arg2);
539
540         handle_fasteoi_irq(virt_irq, desc);
541 }
542
543 void irq_install_pre_handler(int virt_irq,
544                              void (*func)(unsigned int, void *, void *),
545                              void *arg1, void *arg2)
546 {
547         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
548         struct irq_desc *desc = irq_desc + virt_irq;
549
550         data->pre_handler = func;
551         data->arg1 = arg1;
552         data->arg2 = arg2;
553
554         desc->handle_irq = pre_flow_handler;
555 }
556
557 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
558 {
559         struct ino_bucket *bucket;
560         struct irq_handler_data *data;
561         unsigned int virt_irq;
562         int ino;
563
564         BUG_ON(tlb_type == hypervisor);
565
566         ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
567         bucket = &ivector_table[ino];
568         virt_irq = bucket_get_virt_irq(__pa(bucket));
569         if (!virt_irq) {
570                 virt_irq = virt_irq_alloc(0, ino);
571                 bucket_set_virt_irq(__pa(bucket), virt_irq);
572                 set_irq_chip_and_handler_name(virt_irq,
573                                               &sun4u_irq,
574                                               handle_fasteoi_irq,
575                                               "IVEC");
576         }
577
578         data = get_irq_chip_data(virt_irq);
579         if (unlikely(data))
580                 goto out;
581
582         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
583         if (unlikely(!data)) {
584                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
585                 prom_halt();
586         }
587         set_irq_chip_data(virt_irq, data);
588
589         data->imap  = imap;
590         data->iclr  = iclr;
591
592 out:
593         return virt_irq;
594 }
595
596 static unsigned int sun4v_build_common(unsigned long sysino,
597                                        struct irq_chip *chip)
598 {
599         struct ino_bucket *bucket;
600         struct irq_handler_data *data;
601         unsigned int virt_irq;
602
603         BUG_ON(tlb_type != hypervisor);
604
605         bucket = &ivector_table[sysino];
606         virt_irq = bucket_get_virt_irq(__pa(bucket));
607         if (!virt_irq) {
608                 virt_irq = virt_irq_alloc(0, sysino);
609                 bucket_set_virt_irq(__pa(bucket), virt_irq);
610                 set_irq_chip_and_handler_name(virt_irq, chip,
611                                               handle_fasteoi_irq,
612                                               "IVEC");
613         }
614
615         data = get_irq_chip_data(virt_irq);
616         if (unlikely(data))
617                 goto out;
618
619         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
620         if (unlikely(!data)) {
621                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
622                 prom_halt();
623         }
624         set_irq_chip_data(virt_irq, data);
625
626         /* Catch accidental accesses to these things.  IMAP/ICLR handling
627          * is done by hypervisor calls on sun4v platforms, not by direct
628          * register accesses.
629          */
630         data->imap = ~0UL;
631         data->iclr = ~0UL;
632
633 out:
634         return virt_irq;
635 }
636
637 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
638 {
639         unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
640
641         return sun4v_build_common(sysino, &sun4v_irq);
642 }
643
644 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
645 {
646         struct irq_handler_data *data;
647         unsigned long hv_err, cookie;
648         struct ino_bucket *bucket;
649         struct irq_desc *desc;
650         unsigned int virt_irq;
651
652         bucket = kzalloc(sizeof(struct ino_bucket), GFP_ATOMIC);
653         if (unlikely(!bucket))
654                 return 0;
655         __flush_dcache_range((unsigned long) bucket,
656                              ((unsigned long) bucket +
657                               sizeof(struct ino_bucket)));
658
659         virt_irq = virt_irq_alloc(devhandle, devino);
660         bucket_set_virt_irq(__pa(bucket), virt_irq);
661
662         set_irq_chip_and_handler_name(virt_irq, &sun4v_virq,
663                                       handle_fasteoi_irq,
664                                       "IVEC");
665
666         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
667         if (unlikely(!data))
668                 return 0;
669
670         /* In order to make the LDC channel startup sequence easier,
671          * especially wrt. locking, we do not let request_irq() enable
672          * the interrupt.
673          */
674         desc = irq_desc + virt_irq;
675         desc->status |= IRQ_NOAUTOEN;
676
677         set_irq_chip_data(virt_irq, data);
678
679         /* Catch accidental accesses to these things.  IMAP/ICLR handling
680          * is done by hypervisor calls on sun4v platforms, not by direct
681          * register accesses.
682          */
683         data->imap = ~0UL;
684         data->iclr = ~0UL;
685
686         cookie = ~__pa(bucket);
687         hv_err = sun4v_vintr_set_cookie(devhandle, devino, cookie);
688         if (hv_err) {
689                 prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
690                             "err=%lu\n", devhandle, devino, hv_err);
691                 prom_halt();
692         }
693
694         return virt_irq;
695 }
696
697 void ack_bad_irq(unsigned int virt_irq)
698 {
699         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
700
701         if (!ino)
702                 ino = 0xdeadbeef;
703
704         printk(KERN_CRIT "Unexpected IRQ from ino[%x] virt_irq[%u]\n",
705                ino, virt_irq);
706 }
707
708 void *hardirq_stack[NR_CPUS];
709 void *softirq_stack[NR_CPUS];
710
711 static __attribute__((always_inline)) void *set_hardirq_stack(void)
712 {
713         void *orig_sp, *sp = hardirq_stack[smp_processor_id()];
714
715         __asm__ __volatile__("mov %%sp, %0" : "=r" (orig_sp));
716         if (orig_sp < sp ||
717             orig_sp > (sp + THREAD_SIZE)) {
718                 sp += THREAD_SIZE - 192 - STACK_BIAS;
719                 __asm__ __volatile__("mov %0, %%sp" : : "r" (sp));
720         }
721
722         return orig_sp;
723 }
724 static __attribute__((always_inline)) void restore_hardirq_stack(void *orig_sp)
725 {
726         __asm__ __volatile__("mov %0, %%sp" : : "r" (orig_sp));
727 }
728
729 void handler_irq(int irq, struct pt_regs *regs)
730 {
731         unsigned long pstate, bucket_pa;
732         struct pt_regs *old_regs;
733         void *orig_sp;
734
735         clear_softint(1 << irq);
736
737         old_regs = set_irq_regs(regs);
738         irq_enter();
739
740         /* Grab an atomic snapshot of the pending IVECs.  */
741         __asm__ __volatile__("rdpr      %%pstate, %0\n\t"
742                              "wrpr      %0, %3, %%pstate\n\t"
743                              "ldx       [%2], %1\n\t"
744                              "stx       %%g0, [%2]\n\t"
745                              "wrpr      %0, 0x0, %%pstate\n\t"
746                              : "=&r" (pstate), "=&r" (bucket_pa)
747                              : "r" (irq_work_pa(smp_processor_id())),
748                                "i" (PSTATE_IE)
749                              : "memory");
750
751         orig_sp = set_hardirq_stack();
752
753         while (bucket_pa) {
754                 struct irq_desc *desc;
755                 unsigned long next_pa;
756                 unsigned int virt_irq;
757
758                 next_pa = bucket_get_chain_pa(bucket_pa);
759                 virt_irq = bucket_get_virt_irq(bucket_pa);
760                 bucket_clear_chain_pa(bucket_pa);
761
762                 desc = irq_desc + virt_irq;
763
764                 if (!(desc->status & IRQ_DISABLED))
765                         desc->handle_irq(virt_irq, desc);
766
767                 bucket_pa = next_pa;
768         }
769
770         restore_hardirq_stack(orig_sp);
771
772         irq_exit();
773         set_irq_regs(old_regs);
774 }
775
776 void do_softirq(void)
777 {
778         unsigned long flags;
779
780         if (in_interrupt())
781                 return;
782
783         local_irq_save(flags);
784
785         if (local_softirq_pending()) {
786                 void *orig_sp, *sp = softirq_stack[smp_processor_id()];
787
788                 sp += THREAD_SIZE - 192 - STACK_BIAS;
789
790                 __asm__ __volatile__("mov %%sp, %0\n\t"
791                                      "mov %1, %%sp"
792                                      : "=&r" (orig_sp)
793                                      : "r" (sp));
794                 __do_softirq();
795                 __asm__ __volatile__("mov %0, %%sp"
796                                      : : "r" (orig_sp));
797         }
798
799         local_irq_restore(flags);
800 }
801
802 #ifdef CONFIG_HOTPLUG_CPU
803 void fixup_irqs(void)
804 {
805         unsigned int irq;
806
807         for (irq = 0; irq < NR_IRQS; irq++) {
808                 unsigned long flags;
809
810                 spin_lock_irqsave(&irq_desc[irq].lock, flags);
811                 if (irq_desc[irq].action &&
812                     !(irq_desc[irq].status & IRQ_PER_CPU)) {
813                         if (irq_desc[irq].chip->set_affinity)
814                                 irq_desc[irq].chip->set_affinity(irq,
815                                         irq_desc[irq].affinity);
816                 }
817                 spin_unlock_irqrestore(&irq_desc[irq].lock, flags);
818         }
819
820         tick_ops->disable_irq();
821 }
822 #endif
823
824 struct sun5_timer {
825         u64     count0;
826         u64     limit0;
827         u64     count1;
828         u64     limit1;
829 };
830
831 static struct sun5_timer *prom_timers;
832 static u64 prom_limit0, prom_limit1;
833
834 static void map_prom_timers(void)
835 {
836         struct device_node *dp;
837         const unsigned int *addr;
838
839         /* PROM timer node hangs out in the top level of device siblings... */
840         dp = of_find_node_by_path("/");
841         dp = dp->child;
842         while (dp) {
843                 if (!strcmp(dp->name, "counter-timer"))
844                         break;
845                 dp = dp->sibling;
846         }
847
848         /* Assume if node is not present, PROM uses different tick mechanism
849          * which we should not care about.
850          */
851         if (!dp) {
852                 prom_timers = (struct sun5_timer *) 0;
853                 return;
854         }
855
856         /* If PROM is really using this, it must be mapped by him. */
857         addr = of_get_property(dp, "address", NULL);
858         if (!addr) {
859                 prom_printf("PROM does not have timer mapped, trying to continue.\n");
860                 prom_timers = (struct sun5_timer *) 0;
861                 return;
862         }
863         prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
864 }
865
866 static void kill_prom_timer(void)
867 {
868         if (!prom_timers)
869                 return;
870
871         /* Save them away for later. */
872         prom_limit0 = prom_timers->limit0;
873         prom_limit1 = prom_timers->limit1;
874
875         /* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
876          * We turn both off here just to be paranoid.
877          */
878         prom_timers->limit0 = 0;
879         prom_timers->limit1 = 0;
880
881         /* Wheee, eat the interrupt packet too... */
882         __asm__ __volatile__(
883 "       mov     0x40, %%g2\n"
884 "       ldxa    [%%g0] %0, %%g1\n"
885 "       ldxa    [%%g2] %1, %%g1\n"
886 "       stxa    %%g0, [%%g0] %0\n"
887 "       membar  #Sync\n"
888         : /* no outputs */
889         : "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
890         : "g1", "g2");
891 }
892
893 void notrace init_irqwork_curcpu(void)
894 {
895         int cpu = hard_smp_processor_id();
896
897         trap_block[cpu].irq_worklist_pa = 0UL;
898 }
899
900 /* Please be very careful with register_one_mondo() and
901  * sun4v_register_mondo_queues().
902  *
903  * On SMP this gets invoked from the CPU trampoline before
904  * the cpu has fully taken over the trap table from OBP,
905  * and it's kernel stack + %g6 thread register state is
906  * not fully cooked yet.
907  *
908  * Therefore you cannot make any OBP calls, not even prom_printf,
909  * from these two routines.
910  */
911 static void __cpuinit register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
912 {
913         unsigned long num_entries = (qmask + 1) / 64;
914         unsigned long status;
915
916         status = sun4v_cpu_qconf(type, paddr, num_entries);
917         if (status != HV_EOK) {
918                 prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
919                             "err %lu\n", type, paddr, num_entries, status);
920                 prom_halt();
921         }
922 }
923
924 void __cpuinit notrace sun4v_register_mondo_queues(int this_cpu)
925 {
926         struct trap_per_cpu *tb = &trap_block[this_cpu];
927
928         register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
929                            tb->cpu_mondo_qmask);
930         register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
931                            tb->dev_mondo_qmask);
932         register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
933                            tb->resum_qmask);
934         register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
935                            tb->nonresum_qmask);
936 }
937
938 static void __init alloc_one_mondo(unsigned long *pa_ptr, unsigned long qmask)
939 {
940         unsigned long size = PAGE_ALIGN(qmask + 1);
941         void *p = __alloc_bootmem(size, size, 0);
942         if (!p) {
943                 prom_printf("SUN4V: Error, cannot allocate mondo queue.\n");
944                 prom_halt();
945         }
946
947         *pa_ptr = __pa(p);
948 }
949
950 static void __init alloc_one_kbuf(unsigned long *pa_ptr, unsigned long qmask)
951 {
952         unsigned long size = PAGE_ALIGN(qmask + 1);
953         void *p = __alloc_bootmem(size, size, 0);
954
955         if (!p) {
956                 prom_printf("SUN4V: Error, cannot allocate kbuf page.\n");
957                 prom_halt();
958         }
959
960         *pa_ptr = __pa(p);
961 }
962
963 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
964 {
965 #ifdef CONFIG_SMP
966         void *page;
967
968         BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
969
970         page = alloc_bootmem_pages(PAGE_SIZE);
971         if (!page) {
972                 prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
973                 prom_halt();
974         }
975
976         tb->cpu_mondo_block_pa = __pa(page);
977         tb->cpu_list_pa = __pa(page + 64);
978 #endif
979 }
980
981 /* Allocate mondo and error queues for all possible cpus.  */
982 static void __init sun4v_init_mondo_queues(void)
983 {
984         int cpu;
985
986         for_each_possible_cpu(cpu) {
987                 struct trap_per_cpu *tb = &trap_block[cpu];
988
989                 alloc_one_mondo(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
990                 alloc_one_mondo(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
991                 alloc_one_mondo(&tb->resum_mondo_pa, tb->resum_qmask);
992                 alloc_one_kbuf(&tb->resum_kernel_buf_pa, tb->resum_qmask);
993                 alloc_one_mondo(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
994                 alloc_one_kbuf(&tb->nonresum_kernel_buf_pa,
995                                tb->nonresum_qmask);
996         }
997 }
998
999 static void __init init_send_mondo_info(void)
1000 {
1001         int cpu;
1002
1003         for_each_possible_cpu(cpu) {
1004                 struct trap_per_cpu *tb = &trap_block[cpu];
1005
1006                 init_cpu_send_mondo_info(tb);
1007         }
1008 }
1009
1010 static struct irqaction timer_irq_action = {
1011         .name = "timer",
1012 };
1013
1014 /* Only invoked on boot processor. */
1015 void __init init_IRQ(void)
1016 {
1017         unsigned long size;
1018
1019         map_prom_timers();
1020         kill_prom_timer();
1021
1022         size = sizeof(struct ino_bucket) * NUM_IVECS;
1023         ivector_table = alloc_bootmem(size);
1024         if (!ivector_table) {
1025                 prom_printf("Fatal error, cannot allocate ivector_table\n");
1026                 prom_halt();
1027         }
1028         __flush_dcache_range((unsigned long) ivector_table,
1029                              ((unsigned long) ivector_table) + size);
1030
1031         ivector_table_pa = __pa(ivector_table);
1032
1033         if (tlb_type == hypervisor)
1034                 sun4v_init_mondo_queues();
1035
1036         init_send_mondo_info();
1037
1038         if (tlb_type == hypervisor) {
1039                 /* Load up the boot cpu's entries.  */
1040                 sun4v_register_mondo_queues(hard_smp_processor_id());
1041         }
1042
1043         /* We need to clear any IRQ's pending in the soft interrupt
1044          * registers, a spurious one could be left around from the
1045          * PROM timer which we just disabled.
1046          */
1047         clear_softint(get_softint());
1048
1049         /* Now that ivector table is initialized, it is safe
1050          * to receive IRQ vector traps.  We will normally take
1051          * one or two right now, in case some device PROM used
1052          * to boot us wants to speak to us.  We just ignore them.
1053          */
1054         __asm__ __volatile__("rdpr      %%pstate, %%g1\n\t"
1055                              "or        %%g1, %0, %%g1\n\t"
1056                              "wrpr      %%g1, 0x0, %%pstate"
1057                              : /* No outputs */
1058                              : "i" (PSTATE_IE)
1059                              : "g1");
1060
1061         irq_desc[0].action = &timer_irq_action;
1062 }