Merge branch 'kvm-updates/2.6.39' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[pandora-kernel.git] / 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/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_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                                              __irq)),
94                                "i" (ASI_PHYS_USE_EC));
95
96         return ret;
97 }
98
99 static void bucket_set_irq(unsigned long bucket_pa, unsigned int irq)
100 {
101         __asm__ __volatile__("stwa      %0, [%1] %2"
102                              : /* no outputs */
103                              : "r" (irq),
104                                "r" (bucket_pa +
105                                     offsetof(struct ino_bucket,
106                                              __irq)),
107                                "i" (ASI_PHYS_USE_EC));
108 }
109
110 #define irq_work_pa(__cpu)      &(trap_block[(__cpu)].irq_worklist_pa)
111
112 static struct {
113         unsigned int dev_handle;
114         unsigned int dev_ino;
115         unsigned int in_use;
116 } irq_table[NR_IRQS];
117 static DEFINE_SPINLOCK(irq_alloc_lock);
118
119 unsigned char irq_alloc(unsigned int dev_handle, unsigned int dev_ino)
120 {
121         unsigned long flags;
122         unsigned char ent;
123
124         BUILD_BUG_ON(NR_IRQS >= 256);
125
126         spin_lock_irqsave(&irq_alloc_lock, flags);
127
128         for (ent = 1; ent < NR_IRQS; ent++) {
129                 if (!irq_table[ent].in_use)
130                         break;
131         }
132         if (ent >= NR_IRQS) {
133                 printk(KERN_ERR "IRQ: Out of virtual IRQs.\n");
134                 ent = 0;
135         } else {
136                 irq_table[ent].dev_handle = dev_handle;
137                 irq_table[ent].dev_ino = dev_ino;
138                 irq_table[ent].in_use = 1;
139         }
140
141         spin_unlock_irqrestore(&irq_alloc_lock, flags);
142
143         return ent;
144 }
145
146 #ifdef CONFIG_PCI_MSI
147 void irq_free(unsigned int irq)
148 {
149         unsigned long flags;
150
151         if (irq >= NR_IRQS)
152                 return;
153
154         spin_lock_irqsave(&irq_alloc_lock, flags);
155
156         irq_table[irq].in_use = 0;
157
158         spin_unlock_irqrestore(&irq_alloc_lock, flags);
159 }
160 #endif
161
162 /*
163  * /proc/interrupts printing:
164  */
165
166 int show_interrupts(struct seq_file *p, void *v)
167 {
168         int i = *(loff_t *) v, j;
169         struct irqaction * action;
170         unsigned long flags;
171
172         if (i == 0) {
173                 seq_printf(p, "           ");
174                 for_each_online_cpu(j)
175                         seq_printf(p, "CPU%d       ",j);
176                 seq_putc(p, '\n');
177         }
178
179         if (i < NR_IRQS) {
180                 raw_spin_lock_irqsave(&irq_desc[i].lock, flags);
181                 action = irq_desc[i].action;
182                 if (!action)
183                         goto skip;
184                 seq_printf(p, "%3d: ",i);
185 #ifndef CONFIG_SMP
186                 seq_printf(p, "%10u ", kstat_irqs(i));
187 #else
188                 for_each_online_cpu(j)
189                         seq_printf(p, "%10u ", kstat_irqs_cpu(i, j));
190 #endif
191                 seq_printf(p, " %9s", irq_desc[i].irq_data.chip->name);
192                 seq_printf(p, "  %s", action->name);
193
194                 for (action=action->next; action; action = action->next)
195                         seq_printf(p, ", %s", action->name);
196
197                 seq_putc(p, '\n');
198 skip:
199                 raw_spin_unlock_irqrestore(&irq_desc[i].lock, flags);
200         } else if (i == NR_IRQS) {
201                 seq_printf(p, "NMI: ");
202                 for_each_online_cpu(j)
203                         seq_printf(p, "%10u ", cpu_data(j).__nmi_count);
204                 seq_printf(p, "     Non-maskable interrupts\n");
205         }
206         return 0;
207 }
208
209 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
210 {
211         unsigned int tid;
212
213         if (this_is_starfire) {
214                 tid = starfire_translate(imap, cpuid);
215                 tid <<= IMAP_TID_SHIFT;
216                 tid &= IMAP_TID_UPA;
217         } else {
218                 if (tlb_type == cheetah || tlb_type == cheetah_plus) {
219                         unsigned long ver;
220
221                         __asm__ ("rdpr %%ver, %0" : "=r" (ver));
222                         if ((ver >> 32UL) == __JALAPENO_ID ||
223                             (ver >> 32UL) == __SERRANO_ID) {
224                                 tid = cpuid << IMAP_TID_SHIFT;
225                                 tid &= IMAP_TID_JBUS;
226                         } else {
227                                 unsigned int a = cpuid & 0x1f;
228                                 unsigned int n = (cpuid >> 5) & 0x1f;
229
230                                 tid = ((a << IMAP_AID_SHIFT) |
231                                        (n << IMAP_NID_SHIFT));
232                                 tid &= (IMAP_AID_SAFARI |
233                                         IMAP_NID_SAFARI);
234                         }
235                 } else {
236                         tid = cpuid << IMAP_TID_SHIFT;
237                         tid &= IMAP_TID_UPA;
238                 }
239         }
240
241         return tid;
242 }
243
244 struct irq_handler_data {
245         unsigned long   iclr;
246         unsigned long   imap;
247
248         void            (*pre_handler)(unsigned int, void *, void *);
249         void            *arg1;
250         void            *arg2;
251 };
252
253 #ifdef CONFIG_SMP
254 static int irq_choose_cpu(unsigned int irq, const struct cpumask *affinity)
255 {
256         cpumask_t mask;
257         int cpuid;
258
259         cpumask_copy(&mask, affinity);
260         if (cpus_equal(mask, cpu_online_map)) {
261                 cpuid = map_to_cpu(irq);
262         } else {
263                 cpumask_t tmp;
264
265                 cpus_and(tmp, cpu_online_map, mask);
266                 cpuid = cpus_empty(tmp) ? map_to_cpu(irq) : first_cpu(tmp);
267         }
268
269         return cpuid;
270 }
271 #else
272 #define irq_choose_cpu(irq, affinity)   \
273         real_hard_smp_processor_id()
274 #endif
275
276 static void sun4u_irq_enable(struct irq_data *data)
277 {
278         struct irq_handler_data *handler_data = data->handler_data;
279
280         if (likely(handler_data)) {
281                 unsigned long cpuid, imap, val;
282                 unsigned int tid;
283
284                 cpuid = irq_choose_cpu(data->irq, data->affinity);
285                 imap = handler_data->imap;
286
287                 tid = sun4u_compute_tid(imap, cpuid);
288
289                 val = upa_readq(imap);
290                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
291                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
292                 val |= tid | IMAP_VALID;
293                 upa_writeq(val, imap);
294                 upa_writeq(ICLR_IDLE, handler_data->iclr);
295         }
296 }
297
298 static int sun4u_set_affinity(struct irq_data *data,
299                                const struct cpumask *mask, bool force)
300 {
301         struct irq_handler_data *handler_data = data->handler_data;
302
303         if (likely(handler_data)) {
304                 unsigned long cpuid, imap, val;
305                 unsigned int tid;
306
307                 cpuid = irq_choose_cpu(data->irq, mask);
308                 imap = handler_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, handler_data->iclr);
318         }
319
320         return 0;
321 }
322
323 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
324  * handler_irq() will skip the handler call and that will leave the
325  * interrupt in the sent state.  The next ->enable() call will hit the
326  * ICLR register to reset the state machine.
327  *
328  * This scheme is necessary, instead of clearing the Valid bit in the
329  * IMAP register, to handle the case of IMAP registers being shared by
330  * multiple INOs (and thus ICLR registers).  Since we use a different
331  * virtual IRQ for each shared IMAP instance, the generic code thinks
332  * there is only one user so it prematurely calls ->disable() on
333  * free_irq().
334  *
335  * We have to provide an explicit ->disable() method instead of using
336  * NULL to get the default.  The reason is that if the generic code
337  * sees that, it also hooks up a default ->shutdown method which
338  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
339  */
340 static void sun4u_irq_disable(struct irq_data *data)
341 {
342 }
343
344 static void sun4u_irq_eoi(struct irq_data *data)
345 {
346         struct irq_handler_data *handler_data = data->handler_data;
347         struct irq_desc *desc = irq_desc + data->irq;
348
349         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
350                 return;
351
352         if (likely(handler_data))
353                 upa_writeq(ICLR_IDLE, handler_data->iclr);
354 }
355
356 static void sun4v_irq_enable(struct irq_data *data)
357 {
358         unsigned int ino = irq_table[data->irq].dev_ino;
359         unsigned long cpuid = irq_choose_cpu(data->irq, data->affinity);
360         int err;
361
362         err = sun4v_intr_settarget(ino, cpuid);
363         if (err != HV_EOK)
364                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
365                        "err(%d)\n", ino, cpuid, err);
366         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
367         if (err != HV_EOK)
368                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
369                        "err(%d)\n", ino, err);
370         err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
371         if (err != HV_EOK)
372                 printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
373                        ino, err);
374 }
375
376 static int sun4v_set_affinity(struct irq_data *data,
377                                const struct cpumask *mask, bool force)
378 {
379         unsigned int ino = irq_table[data->irq].dev_ino;
380         unsigned long cpuid = irq_choose_cpu(data->irq, mask);
381         int err;
382
383         err = sun4v_intr_settarget(ino, cpuid);
384         if (err != HV_EOK)
385                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
386                        "err(%d)\n", ino, cpuid, err);
387
388         return 0;
389 }
390
391 static void sun4v_irq_disable(struct irq_data *data)
392 {
393         unsigned int ino = irq_table[data->irq].dev_ino;
394         int err;
395
396         err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
397         if (err != HV_EOK)
398                 printk(KERN_ERR "sun4v_intr_setenabled(%x): "
399                        "err(%d)\n", ino, err);
400 }
401
402 static void sun4v_irq_eoi(struct irq_data *data)
403 {
404         unsigned int ino = irq_table[data->irq].dev_ino;
405         struct irq_desc *desc = irq_desc + data->irq;
406         int err;
407
408         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
409                 return;
410
411         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
412         if (err != HV_EOK)
413                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
414                        "err(%d)\n", ino, err);
415 }
416
417 static void sun4v_virq_enable(struct irq_data *data)
418 {
419         unsigned long cpuid, dev_handle, dev_ino;
420         int err;
421
422         cpuid = irq_choose_cpu(data->irq, data->affinity);
423
424         dev_handle = irq_table[data->irq].dev_handle;
425         dev_ino = irq_table[data->irq].dev_ino;
426
427         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
428         if (err != HV_EOK)
429                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
430                        "err(%d)\n",
431                        dev_handle, dev_ino, cpuid, err);
432         err = sun4v_vintr_set_state(dev_handle, dev_ino,
433                                     HV_INTR_STATE_IDLE);
434         if (err != HV_EOK)
435                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
436                        "HV_INTR_STATE_IDLE): err(%d)\n",
437                        dev_handle, dev_ino, err);
438         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
439                                     HV_INTR_ENABLED);
440         if (err != HV_EOK)
441                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
442                        "HV_INTR_ENABLED): err(%d)\n",
443                        dev_handle, dev_ino, err);
444 }
445
446 static int sun4v_virt_set_affinity(struct irq_data *data,
447                                     const struct cpumask *mask, bool force)
448 {
449         unsigned long cpuid, dev_handle, dev_ino;
450         int err;
451
452         cpuid = irq_choose_cpu(data->irq, mask);
453
454         dev_handle = irq_table[data->irq].dev_handle;
455         dev_ino = irq_table[data->irq].dev_ino;
456
457         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
458         if (err != HV_EOK)
459                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
460                        "err(%d)\n",
461                        dev_handle, dev_ino, cpuid, err);
462
463         return 0;
464 }
465
466 static void sun4v_virq_disable(struct irq_data *data)
467 {
468         unsigned long dev_handle, dev_ino;
469         int err;
470
471         dev_handle = irq_table[data->irq].dev_handle;
472         dev_ino = irq_table[data->irq].dev_ino;
473
474         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
475                                     HV_INTR_DISABLED);
476         if (err != HV_EOK)
477                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
478                        "HV_INTR_DISABLED): err(%d)\n",
479                        dev_handle, dev_ino, err);
480 }
481
482 static void sun4v_virq_eoi(struct irq_data *data)
483 {
484         struct irq_desc *desc = irq_desc + data->irq;
485         unsigned long dev_handle, dev_ino;
486         int err;
487
488         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
489                 return;
490
491         dev_handle = irq_table[data->irq].dev_handle;
492         dev_ino = irq_table[data->irq].dev_ino;
493
494         err = sun4v_vintr_set_state(dev_handle, dev_ino,
495                                     HV_INTR_STATE_IDLE);
496         if (err != HV_EOK)
497                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
498                        "HV_INTR_STATE_IDLE): err(%d)\n",
499                        dev_handle, dev_ino, err);
500 }
501
502 static struct irq_chip sun4u_irq = {
503         .name                   = "sun4u",
504         .irq_enable             = sun4u_irq_enable,
505         .irq_disable            = sun4u_irq_disable,
506         .irq_eoi                = sun4u_irq_eoi,
507         .irq_set_affinity       = sun4u_set_affinity,
508 };
509
510 static struct irq_chip sun4v_irq = {
511         .name                   = "sun4v",
512         .irq_enable             = sun4v_irq_enable,
513         .irq_disable            = sun4v_irq_disable,
514         .irq_eoi                = sun4v_irq_eoi,
515         .irq_set_affinity       = sun4v_set_affinity,
516 };
517
518 static struct irq_chip sun4v_virq = {
519         .name                   = "vsun4v",
520         .irq_enable             = sun4v_virq_enable,
521         .irq_disable            = sun4v_virq_disable,
522         .irq_eoi                = sun4v_virq_eoi,
523         .irq_set_affinity       = sun4v_virt_set_affinity,
524 };
525
526 static void pre_flow_handler(unsigned int irq, struct irq_desc *desc)
527 {
528         struct irq_handler_data *handler_data = get_irq_data(irq);
529         unsigned int ino = irq_table[irq].dev_ino;
530
531         handler_data->pre_handler(ino, handler_data->arg1, handler_data->arg2);
532
533         handle_fasteoi_irq(irq, desc);
534 }
535
536 void irq_install_pre_handler(int irq,
537                              void (*func)(unsigned int, void *, void *),
538                              void *arg1, void *arg2)
539 {
540         struct irq_handler_data *handler_data = get_irq_data(irq);
541         struct irq_desc *desc = irq_desc + irq;
542
543         handler_data->pre_handler = func;
544         handler_data->arg1 = arg1;
545         handler_data->arg2 = arg2;
546
547         desc->handle_irq = pre_flow_handler;
548 }
549
550 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
551 {
552         struct ino_bucket *bucket;
553         struct irq_handler_data *handler_data;
554         unsigned int irq;
555         int ino;
556
557         BUG_ON(tlb_type == hypervisor);
558
559         ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
560         bucket = &ivector_table[ino];
561         irq = bucket_get_irq(__pa(bucket));
562         if (!irq) {
563                 irq = irq_alloc(0, ino);
564                 bucket_set_irq(__pa(bucket), irq);
565                 set_irq_chip_and_handler_name(irq,
566                                               &sun4u_irq,
567                                               handle_fasteoi_irq,
568                                               "IVEC");
569         }
570
571         handler_data = get_irq_data(irq);
572         if (unlikely(handler_data))
573                 goto out;
574
575         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
576         if (unlikely(!handler_data)) {
577                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
578                 prom_halt();
579         }
580         set_irq_data(irq, handler_data);
581
582         handler_data->imap  = imap;
583         handler_data->iclr  = iclr;
584
585 out:
586         return irq;
587 }
588
589 static unsigned int sun4v_build_common(unsigned long sysino,
590                                        struct irq_chip *chip)
591 {
592         struct ino_bucket *bucket;
593         struct irq_handler_data *handler_data;
594         unsigned int irq;
595
596         BUG_ON(tlb_type != hypervisor);
597
598         bucket = &ivector_table[sysino];
599         irq = bucket_get_irq(__pa(bucket));
600         if (!irq) {
601                 irq = irq_alloc(0, sysino);
602                 bucket_set_irq(__pa(bucket), irq);
603                 set_irq_chip_and_handler_name(irq, chip,
604                                               handle_fasteoi_irq,
605                                               "IVEC");
606         }
607
608         handler_data = get_irq_data(irq);
609         if (unlikely(handler_data))
610                 goto out;
611
612         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
613         if (unlikely(!handler_data)) {
614                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
615                 prom_halt();
616         }
617         set_irq_data(irq, handler_data);
618
619         /* Catch accidental accesses to these things.  IMAP/ICLR handling
620          * is done by hypervisor calls on sun4v platforms, not by direct
621          * register accesses.
622          */
623         handler_data->imap = ~0UL;
624         handler_data->iclr = ~0UL;
625
626 out:
627         return irq;
628 }
629
630 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
631 {
632         unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
633
634         return sun4v_build_common(sysino, &sun4v_irq);
635 }
636
637 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
638 {
639         struct irq_handler_data *handler_data;
640         unsigned long hv_err, cookie;
641         struct ino_bucket *bucket;
642         struct irq_desc *desc;
643         unsigned int irq;
644
645         bucket = kzalloc(sizeof(struct ino_bucket), GFP_ATOMIC);
646         if (unlikely(!bucket))
647                 return 0;
648
649         /* The only reference we store to the IRQ bucket is
650          * by physical address which kmemleak can't see, tell
651          * it that this object explicitly is not a leak and
652          * should be scanned.
653          */
654         kmemleak_not_leak(bucket);
655
656         __flush_dcache_range((unsigned long) bucket,
657                              ((unsigned long) bucket +
658                               sizeof(struct ino_bucket)));
659
660         irq = irq_alloc(devhandle, devino);
661         bucket_set_irq(__pa(bucket), irq);
662
663         set_irq_chip_and_handler_name(irq, &sun4v_virq,
664                                       handle_fasteoi_irq,
665                                       "IVEC");
666
667         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
668         if (unlikely(!handler_data))
669                 return 0;
670
671         /* In order to make the LDC channel startup sequence easier,
672          * especially wrt. locking, we do not let request_irq() enable
673          * the interrupt.
674          */
675         desc = irq_desc + irq;
676         desc->status |= IRQ_NOAUTOEN;
677
678         set_irq_data(irq, handler_data);
679
680         /* Catch accidental accesses to these things.  IMAP/ICLR handling
681          * is done by hypervisor calls on sun4v platforms, not by direct
682          * register accesses.
683          */
684         handler_data->imap = ~0UL;
685         handler_data->iclr = ~0UL;
686
687         cookie = ~__pa(bucket);
688         hv_err = sun4v_vintr_set_cookie(devhandle, devino, cookie);
689         if (hv_err) {
690                 prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
691                             "err=%lu\n", devhandle, devino, hv_err);
692                 prom_halt();
693         }
694
695         return irq;
696 }
697
698 void ack_bad_irq(unsigned int irq)
699 {
700         unsigned int ino = irq_table[irq].dev_ino;
701
702         if (!ino)
703                 ino = 0xdeadbeef;
704
705         printk(KERN_CRIT "Unexpected IRQ from ino[%x] irq[%u]\n",
706                ino, irq);
707 }
708
709 void *hardirq_stack[NR_CPUS];
710 void *softirq_stack[NR_CPUS];
711
712 void __irq_entry handler_irq(int pil, struct pt_regs *regs)
713 {
714         unsigned long pstate, bucket_pa;
715         struct pt_regs *old_regs;
716         void *orig_sp;
717
718         clear_softint(1 << pil);
719
720         old_regs = set_irq_regs(regs);
721         irq_enter();
722
723         /* Grab an atomic snapshot of the pending IVECs.  */
724         __asm__ __volatile__("rdpr      %%pstate, %0\n\t"
725                              "wrpr      %0, %3, %%pstate\n\t"
726                              "ldx       [%2], %1\n\t"
727                              "stx       %%g0, [%2]\n\t"
728                              "wrpr      %0, 0x0, %%pstate\n\t"
729                              : "=&r" (pstate), "=&r" (bucket_pa)
730                              : "r" (irq_work_pa(smp_processor_id())),
731                                "i" (PSTATE_IE)
732                              : "memory");
733
734         orig_sp = set_hardirq_stack();
735
736         while (bucket_pa) {
737                 struct irq_desc *desc;
738                 unsigned long next_pa;
739                 unsigned int irq;
740
741                 next_pa = bucket_get_chain_pa(bucket_pa);
742                 irq = bucket_get_irq(bucket_pa);
743                 bucket_clear_chain_pa(bucket_pa);
744
745                 desc = irq_desc + irq;
746
747                 if (!(desc->status & IRQ_DISABLED))
748                         desc->handle_irq(irq, desc);
749
750                 bucket_pa = next_pa;
751         }
752
753         restore_hardirq_stack(orig_sp);
754
755         irq_exit();
756         set_irq_regs(old_regs);
757 }
758
759 void do_softirq(void)
760 {
761         unsigned long flags;
762
763         if (in_interrupt())
764                 return;
765
766         local_irq_save(flags);
767
768         if (local_softirq_pending()) {
769                 void *orig_sp, *sp = softirq_stack[smp_processor_id()];
770
771                 sp += THREAD_SIZE - 192 - STACK_BIAS;
772
773                 __asm__ __volatile__("mov %%sp, %0\n\t"
774                                      "mov %1, %%sp"
775                                      : "=&r" (orig_sp)
776                                      : "r" (sp));
777                 __do_softirq();
778                 __asm__ __volatile__("mov %0, %%sp"
779                                      : : "r" (orig_sp));
780         }
781
782         local_irq_restore(flags);
783 }
784
785 #ifdef CONFIG_HOTPLUG_CPU
786 void fixup_irqs(void)
787 {
788         unsigned int irq;
789
790         for (irq = 0; irq < NR_IRQS; irq++) {
791                 unsigned long flags;
792
793                 raw_spin_lock_irqsave(&irq_desc[irq].lock, flags);
794                 if (irq_desc[irq].action &&
795                     !(irq_desc[irq].status & IRQ_PER_CPU)) {
796                         struct irq_data *data = irq_get_irq_data(irq);
797
798                         if (data->chip->irq_set_affinity)
799                                 data->chip->irq_set_affinity(data,
800                                                              data->affinity,
801                                                              false);
802                 }
803                 raw_spin_unlock_irqrestore(&irq_desc[irq].lock, flags);
804         }
805
806         tick_ops->disable_irq();
807 }
808 #endif
809
810 struct sun5_timer {
811         u64     count0;
812         u64     limit0;
813         u64     count1;
814         u64     limit1;
815 };
816
817 static struct sun5_timer *prom_timers;
818 static u64 prom_limit0, prom_limit1;
819
820 static void map_prom_timers(void)
821 {
822         struct device_node *dp;
823         const unsigned int *addr;
824
825         /* PROM timer node hangs out in the top level of device siblings... */
826         dp = of_find_node_by_path("/");
827         dp = dp->child;
828         while (dp) {
829                 if (!strcmp(dp->name, "counter-timer"))
830                         break;
831                 dp = dp->sibling;
832         }
833
834         /* Assume if node is not present, PROM uses different tick mechanism
835          * which we should not care about.
836          */
837         if (!dp) {
838                 prom_timers = (struct sun5_timer *) 0;
839                 return;
840         }
841
842         /* If PROM is really using this, it must be mapped by him. */
843         addr = of_get_property(dp, "address", NULL);
844         if (!addr) {
845                 prom_printf("PROM does not have timer mapped, trying to continue.\n");
846                 prom_timers = (struct sun5_timer *) 0;
847                 return;
848         }
849         prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
850 }
851
852 static void kill_prom_timer(void)
853 {
854         if (!prom_timers)
855                 return;
856
857         /* Save them away for later. */
858         prom_limit0 = prom_timers->limit0;
859         prom_limit1 = prom_timers->limit1;
860
861         /* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
862          * We turn both off here just to be paranoid.
863          */
864         prom_timers->limit0 = 0;
865         prom_timers->limit1 = 0;
866
867         /* Wheee, eat the interrupt packet too... */
868         __asm__ __volatile__(
869 "       mov     0x40, %%g2\n"
870 "       ldxa    [%%g0] %0, %%g1\n"
871 "       ldxa    [%%g2] %1, %%g1\n"
872 "       stxa    %%g0, [%%g0] %0\n"
873 "       membar  #Sync\n"
874         : /* no outputs */
875         : "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
876         : "g1", "g2");
877 }
878
879 void notrace init_irqwork_curcpu(void)
880 {
881         int cpu = hard_smp_processor_id();
882
883         trap_block[cpu].irq_worklist_pa = 0UL;
884 }
885
886 /* Please be very careful with register_one_mondo() and
887  * sun4v_register_mondo_queues().
888  *
889  * On SMP this gets invoked from the CPU trampoline before
890  * the cpu has fully taken over the trap table from OBP,
891  * and it's kernel stack + %g6 thread register state is
892  * not fully cooked yet.
893  *
894  * Therefore you cannot make any OBP calls, not even prom_printf,
895  * from these two routines.
896  */
897 static void __cpuinit notrace register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
898 {
899         unsigned long num_entries = (qmask + 1) / 64;
900         unsigned long status;
901
902         status = sun4v_cpu_qconf(type, paddr, num_entries);
903         if (status != HV_EOK) {
904                 prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
905                             "err %lu\n", type, paddr, num_entries, status);
906                 prom_halt();
907         }
908 }
909
910 void __cpuinit notrace sun4v_register_mondo_queues(int this_cpu)
911 {
912         struct trap_per_cpu *tb = &trap_block[this_cpu];
913
914         register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
915                            tb->cpu_mondo_qmask);
916         register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
917                            tb->dev_mondo_qmask);
918         register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
919                            tb->resum_qmask);
920         register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
921                            tb->nonresum_qmask);
922 }
923
924 /* Each queue region must be a power of 2 multiple of 64 bytes in
925  * size.  The base real address must be aligned to the size of the
926  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
927  */
928 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
929 {
930         unsigned long size = PAGE_ALIGN(qmask + 1);
931         unsigned long order = get_order(size);
932         unsigned long p;
933
934         p = __get_free_pages(GFP_KERNEL, order);
935         if (!p) {
936                 prom_printf("SUN4V: Error, cannot allocate queue.\n");
937                 prom_halt();
938         }
939
940         *pa_ptr = __pa(p);
941 }
942
943 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
944 {
945 #ifdef CONFIG_SMP
946         unsigned long page;
947
948         BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
949
950         page = get_zeroed_page(GFP_KERNEL);
951         if (!page) {
952                 prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
953                 prom_halt();
954         }
955
956         tb->cpu_mondo_block_pa = __pa(page);
957         tb->cpu_list_pa = __pa(page + 64);
958 #endif
959 }
960
961 /* Allocate mondo and error queues for all possible cpus.  */
962 static void __init sun4v_init_mondo_queues(void)
963 {
964         int cpu;
965
966         for_each_possible_cpu(cpu) {
967                 struct trap_per_cpu *tb = &trap_block[cpu];
968
969                 alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
970                 alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
971                 alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
972                 alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
973                 alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
974                 alloc_one_queue(&tb->nonresum_kernel_buf_pa,
975                                 tb->nonresum_qmask);
976         }
977 }
978
979 static void __init init_send_mondo_info(void)
980 {
981         int cpu;
982
983         for_each_possible_cpu(cpu) {
984                 struct trap_per_cpu *tb = &trap_block[cpu];
985
986                 init_cpu_send_mondo_info(tb);
987         }
988 }
989
990 static struct irqaction timer_irq_action = {
991         .name = "timer",
992 };
993
994 /* Only invoked on boot processor. */
995 void __init init_IRQ(void)
996 {
997         unsigned long size;
998
999         map_prom_timers();
1000         kill_prom_timer();
1001
1002         size = sizeof(struct ino_bucket) * NUM_IVECS;
1003         ivector_table = kzalloc(size, GFP_KERNEL);
1004         if (!ivector_table) {
1005                 prom_printf("Fatal error, cannot allocate ivector_table\n");
1006                 prom_halt();
1007         }
1008         __flush_dcache_range((unsigned long) ivector_table,
1009                              ((unsigned long) ivector_table) + size);
1010
1011         ivector_table_pa = __pa(ivector_table);
1012
1013         if (tlb_type == hypervisor)
1014                 sun4v_init_mondo_queues();
1015
1016         init_send_mondo_info();
1017
1018         if (tlb_type == hypervisor) {
1019                 /* Load up the boot cpu's entries.  */
1020                 sun4v_register_mondo_queues(hard_smp_processor_id());
1021         }
1022
1023         /* We need to clear any IRQ's pending in the soft interrupt
1024          * registers, a spurious one could be left around from the
1025          * PROM timer which we just disabled.
1026          */
1027         clear_softint(get_softint());
1028
1029         /* Now that ivector table is initialized, it is safe
1030          * to receive IRQ vector traps.  We will normally take
1031          * one or two right now, in case some device PROM used
1032          * to boot us wants to speak to us.  We just ignore them.
1033          */
1034         __asm__ __volatile__("rdpr      %%pstate, %%g1\n\t"
1035                              "or        %%g1, %0, %%g1\n\t"
1036                              "wrpr      %%g1, 0x0, %%pstate"
1037                              : /* No outputs */
1038                              : "i" (PSTATE_IE)
1039                              : "g1");
1040
1041         irq_desc[0].action = &timer_irq_action;
1042 }