]> git.karo-electronics.de Git - linux-beck.git/blob - arch/sparc/kernel/irq_64.c
sparc64: Add function graph tracer support.
[linux-beck.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
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 #include "cpumap.h"
49
50 #define NUM_IVECS       (IMAP_INR + 1)
51
52 struct ino_bucket *ivector_table;
53 unsigned long ivector_table_pa;
54
55 /* On several sun4u processors, it is illegal to mix bypass and
56  * non-bypass accesses.  Therefore we access all INO buckets
57  * using bypass accesses only.
58  */
59 static unsigned long bucket_get_chain_pa(unsigned long bucket_pa)
60 {
61         unsigned long ret;
62
63         __asm__ __volatile__("ldxa      [%1] %2, %0"
64                              : "=&r" (ret)
65                              : "r" (bucket_pa +
66                                     offsetof(struct ino_bucket,
67                                              __irq_chain_pa)),
68                                "i" (ASI_PHYS_USE_EC));
69
70         return ret;
71 }
72
73 static void bucket_clear_chain_pa(unsigned long bucket_pa)
74 {
75         __asm__ __volatile__("stxa      %%g0, [%0] %1"
76                              : /* no outputs */
77                              : "r" (bucket_pa +
78                                     offsetof(struct ino_bucket,
79                                              __irq_chain_pa)),
80                                "i" (ASI_PHYS_USE_EC));
81 }
82
83 static unsigned int bucket_get_virt_irq(unsigned long bucket_pa)
84 {
85         unsigned int ret;
86
87         __asm__ __volatile__("lduwa     [%1] %2, %0"
88                              : "=&r" (ret)
89                              : "r" (bucket_pa +
90                                     offsetof(struct ino_bucket,
91                                              __virt_irq)),
92                                "i" (ASI_PHYS_USE_EC));
93
94         return ret;
95 }
96
97 static void bucket_set_virt_irq(unsigned long bucket_pa,
98                                 unsigned int virt_irq)
99 {
100         __asm__ __volatile__("stwa      %0, [%1] %2"
101                              : /* no outputs */
102                              : "r" (virt_irq),
103                                "r" (bucket_pa +
104                                     offsetof(struct ino_bucket,
105                                              __virt_irq)),
106                                "i" (ASI_PHYS_USE_EC));
107 }
108
109 #define irq_work_pa(__cpu)      &(trap_block[(__cpu)].irq_worklist_pa)
110
111 static struct {
112         unsigned int dev_handle;
113         unsigned int dev_ino;
114         unsigned int in_use;
115 } virt_irq_table[NR_IRQS];
116 static DEFINE_SPINLOCK(virt_irq_alloc_lock);
117
118 unsigned char virt_irq_alloc(unsigned int dev_handle,
119                              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(&virt_irq_alloc_lock, flags);
127
128         for (ent = 1; ent < NR_IRQS; ent++) {
129                 if (!virt_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                 virt_irq_table[ent].dev_handle = dev_handle;
137                 virt_irq_table[ent].dev_ino = dev_ino;
138                 virt_irq_table[ent].in_use = 1;
139         }
140
141         spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
142
143         return ent;
144 }
145
146 #ifdef CONFIG_PCI_MSI
147 void virt_irq_free(unsigned int virt_irq)
148 {
149         unsigned long flags;
150
151         if (virt_irq >= NR_IRQS)
152                 return;
153
154         spin_lock_irqsave(&virt_irq_alloc_lock, flags);
155
156         virt_irq_table[virt_irq].in_use = 0;
157
158         spin_unlock_irqrestore(&virt_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].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 virt_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(virt_irq);
262         } else {
263                 cpumask_t tmp;
264
265                 cpus_and(tmp, cpu_online_map, mask);
266                 cpuid = cpus_empty(tmp) ? map_to_cpu(virt_irq) : first_cpu(tmp);
267         }
268
269         return cpuid;
270 }
271 #else
272 #define irq_choose_cpu(virt_irq, affinity)      \
273         real_hard_smp_processor_id()
274 #endif
275
276 static void sun4u_irq_enable(unsigned int virt_irq)
277 {
278         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
279
280         if (likely(data)) {
281                 unsigned long cpuid, imap, val;
282                 unsigned int tid;
283
284                 cpuid = irq_choose_cpu(virt_irq,
285                                        irq_desc[virt_irq].affinity);
286                 imap = data->imap;
287
288                 tid = sun4u_compute_tid(imap, cpuid);
289
290                 val = upa_readq(imap);
291                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
292                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
293                 val |= tid | IMAP_VALID;
294                 upa_writeq(val, imap);
295                 upa_writeq(ICLR_IDLE, data->iclr);
296         }
297 }
298
299 static int sun4u_set_affinity(unsigned int virt_irq,
300                                const struct cpumask *mask)
301 {
302         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
303
304         if (likely(data)) {
305                 unsigned long cpuid, imap, val;
306                 unsigned int tid;
307
308                 cpuid = irq_choose_cpu(virt_irq, mask);
309                 imap = data->imap;
310
311                 tid = sun4u_compute_tid(imap, cpuid);
312
313                 val = upa_readq(imap);
314                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
315                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
316                 val |= tid | IMAP_VALID;
317                 upa_writeq(val, imap);
318                 upa_writeq(ICLR_IDLE, data->iclr);
319         }
320
321         return 0;
322 }
323
324 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
325  * handler_irq() will skip the handler call and that will leave the
326  * interrupt in the sent state.  The next ->enable() call will hit the
327  * ICLR register to reset the state machine.
328  *
329  * This scheme is necessary, instead of clearing the Valid bit in the
330  * IMAP register, to handle the case of IMAP registers being shared by
331  * multiple INOs (and thus ICLR registers).  Since we use a different
332  * virtual IRQ for each shared IMAP instance, the generic code thinks
333  * there is only one user so it prematurely calls ->disable() on
334  * free_irq().
335  *
336  * We have to provide an explicit ->disable() method instead of using
337  * NULL to get the default.  The reason is that if the generic code
338  * sees that, it also hooks up a default ->shutdown method which
339  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
340  */
341 static void sun4u_irq_disable(unsigned int virt_irq)
342 {
343 }
344
345 static void sun4u_irq_eoi(unsigned int virt_irq)
346 {
347         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
348         struct irq_desc *desc = irq_desc + virt_irq;
349
350         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
351                 return;
352
353         if (likely(data))
354                 upa_writeq(ICLR_IDLE, data->iclr);
355 }
356
357 static void sun4v_irq_enable(unsigned int virt_irq)
358 {
359         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
360         unsigned long cpuid = irq_choose_cpu(virt_irq,
361                                              irq_desc[virt_irq].affinity);
362         int err;
363
364         err = sun4v_intr_settarget(ino, cpuid);
365         if (err != HV_EOK)
366                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
367                        "err(%d)\n", ino, cpuid, err);
368         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
369         if (err != HV_EOK)
370                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
371                        "err(%d)\n", ino, err);
372         err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
373         if (err != HV_EOK)
374                 printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
375                        ino, err);
376 }
377
378 static int sun4v_set_affinity(unsigned int virt_irq,
379                                const struct cpumask *mask)
380 {
381         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
382         unsigned long cpuid = irq_choose_cpu(virt_irq, mask);
383         int err;
384
385         err = sun4v_intr_settarget(ino, cpuid);
386         if (err != HV_EOK)
387                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
388                        "err(%d)\n", ino, cpuid, err);
389
390         return 0;
391 }
392
393 static void sun4v_irq_disable(unsigned int virt_irq)
394 {
395         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
396         int err;
397
398         err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
399         if (err != HV_EOK)
400                 printk(KERN_ERR "sun4v_intr_setenabled(%x): "
401                        "err(%d)\n", ino, err);
402 }
403
404 static void sun4v_irq_eoi(unsigned int virt_irq)
405 {
406         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
407         struct irq_desc *desc = irq_desc + virt_irq;
408         int err;
409
410         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
411                 return;
412
413         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
414         if (err != HV_EOK)
415                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
416                        "err(%d)\n", ino, err);
417 }
418
419 static void sun4v_virq_enable(unsigned int virt_irq)
420 {
421         unsigned long cpuid, dev_handle, dev_ino;
422         int err;
423
424         cpuid = irq_choose_cpu(virt_irq, irq_desc[virt_irq].affinity);
425
426         dev_handle = virt_irq_table[virt_irq].dev_handle;
427         dev_ino = virt_irq_table[virt_irq].dev_ino;
428
429         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
430         if (err != HV_EOK)
431                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
432                        "err(%d)\n",
433                        dev_handle, dev_ino, cpuid, err);
434         err = sun4v_vintr_set_state(dev_handle, dev_ino,
435                                     HV_INTR_STATE_IDLE);
436         if (err != HV_EOK)
437                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
438                        "HV_INTR_STATE_IDLE): err(%d)\n",
439                        dev_handle, dev_ino, err);
440         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
441                                     HV_INTR_ENABLED);
442         if (err != HV_EOK)
443                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
444                        "HV_INTR_ENABLED): err(%d)\n",
445                        dev_handle, dev_ino, err);
446 }
447
448 static int sun4v_virt_set_affinity(unsigned int virt_irq,
449                                     const struct cpumask *mask)
450 {
451         unsigned long cpuid, dev_handle, dev_ino;
452         int err;
453
454         cpuid = irq_choose_cpu(virt_irq, mask);
455
456         dev_handle = virt_irq_table[virt_irq].dev_handle;
457         dev_ino = virt_irq_table[virt_irq].dev_ino;
458
459         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
460         if (err != HV_EOK)
461                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
462                        "err(%d)\n",
463                        dev_handle, dev_ino, cpuid, err);
464
465         return 0;
466 }
467
468 static void sun4v_virq_disable(unsigned int virt_irq)
469 {
470         unsigned long dev_handle, dev_ino;
471         int err;
472
473         dev_handle = virt_irq_table[virt_irq].dev_handle;
474         dev_ino = virt_irq_table[virt_irq].dev_ino;
475
476         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
477                                     HV_INTR_DISABLED);
478         if (err != HV_EOK)
479                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
480                        "HV_INTR_DISABLED): err(%d)\n",
481                        dev_handle, dev_ino, err);
482 }
483
484 static void sun4v_virq_eoi(unsigned int virt_irq)
485 {
486         struct irq_desc *desc = irq_desc + virt_irq;
487         unsigned long dev_handle, dev_ino;
488         int err;
489
490         if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
491                 return;
492
493         dev_handle = virt_irq_table[virt_irq].dev_handle;
494         dev_ino = virt_irq_table[virt_irq].dev_ino;
495
496         err = sun4v_vintr_set_state(dev_handle, dev_ino,
497                                     HV_INTR_STATE_IDLE);
498         if (err != HV_EOK)
499                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
500                        "HV_INTR_STATE_IDLE): err(%d)\n",
501                        dev_handle, dev_ino, err);
502 }
503
504 static struct irq_chip sun4u_irq = {
505         .name           = "sun4u",
506         .enable         = sun4u_irq_enable,
507         .disable        = sun4u_irq_disable,
508         .eoi            = sun4u_irq_eoi,
509         .set_affinity   = sun4u_set_affinity,
510 };
511
512 static struct irq_chip sun4v_irq = {
513         .name           = "sun4v",
514         .enable         = sun4v_irq_enable,
515         .disable        = sun4v_irq_disable,
516         .eoi            = sun4v_irq_eoi,
517         .set_affinity   = sun4v_set_affinity,
518 };
519
520 static struct irq_chip sun4v_virq = {
521         .name           = "vsun4v",
522         .enable         = sun4v_virq_enable,
523         .disable        = sun4v_virq_disable,
524         .eoi            = sun4v_virq_eoi,
525         .set_affinity   = sun4v_virt_set_affinity,
526 };
527
528 static void pre_flow_handler(unsigned int virt_irq,
529                                       struct irq_desc *desc)
530 {
531         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
532         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
533
534         data->pre_handler(ino, data->arg1, data->arg2);
535
536         handle_fasteoi_irq(virt_irq, desc);
537 }
538
539 void irq_install_pre_handler(int virt_irq,
540                              void (*func)(unsigned int, void *, void *),
541                              void *arg1, void *arg2)
542 {
543         struct irq_handler_data *data = get_irq_chip_data(virt_irq);
544         struct irq_desc *desc = irq_desc + virt_irq;
545
546         data->pre_handler = func;
547         data->arg1 = arg1;
548         data->arg2 = arg2;
549
550         desc->handle_irq = pre_flow_handler;
551 }
552
553 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
554 {
555         struct ino_bucket *bucket;
556         struct irq_handler_data *data;
557         unsigned int virt_irq;
558         int ino;
559
560         BUG_ON(tlb_type == hypervisor);
561
562         ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
563         bucket = &ivector_table[ino];
564         virt_irq = bucket_get_virt_irq(__pa(bucket));
565         if (!virt_irq) {
566                 virt_irq = virt_irq_alloc(0, ino);
567                 bucket_set_virt_irq(__pa(bucket), virt_irq);
568                 set_irq_chip_and_handler_name(virt_irq,
569                                               &sun4u_irq,
570                                               handle_fasteoi_irq,
571                                               "IVEC");
572         }
573
574         data = get_irq_chip_data(virt_irq);
575         if (unlikely(data))
576                 goto out;
577
578         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
579         if (unlikely(!data)) {
580                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
581                 prom_halt();
582         }
583         set_irq_chip_data(virt_irq, data);
584
585         data->imap  = imap;
586         data->iclr  = iclr;
587
588 out:
589         return virt_irq;
590 }
591
592 static unsigned int sun4v_build_common(unsigned long sysino,
593                                        struct irq_chip *chip)
594 {
595         struct ino_bucket *bucket;
596         struct irq_handler_data *data;
597         unsigned int virt_irq;
598
599         BUG_ON(tlb_type != hypervisor);
600
601         bucket = &ivector_table[sysino];
602         virt_irq = bucket_get_virt_irq(__pa(bucket));
603         if (!virt_irq) {
604                 virt_irq = virt_irq_alloc(0, sysino);
605                 bucket_set_virt_irq(__pa(bucket), virt_irq);
606                 set_irq_chip_and_handler_name(virt_irq, chip,
607                                               handle_fasteoi_irq,
608                                               "IVEC");
609         }
610
611         data = get_irq_chip_data(virt_irq);
612         if (unlikely(data))
613                 goto out;
614
615         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
616         if (unlikely(!data)) {
617                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
618                 prom_halt();
619         }
620         set_irq_chip_data(virt_irq, data);
621
622         /* Catch accidental accesses to these things.  IMAP/ICLR handling
623          * is done by hypervisor calls on sun4v platforms, not by direct
624          * register accesses.
625          */
626         data->imap = ~0UL;
627         data->iclr = ~0UL;
628
629 out:
630         return virt_irq;
631 }
632
633 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
634 {
635         unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
636
637         return sun4v_build_common(sysino, &sun4v_irq);
638 }
639
640 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
641 {
642         struct irq_handler_data *data;
643         unsigned long hv_err, cookie;
644         struct ino_bucket *bucket;
645         struct irq_desc *desc;
646         unsigned int virt_irq;
647
648         bucket = kzalloc(sizeof(struct ino_bucket), GFP_ATOMIC);
649         if (unlikely(!bucket))
650                 return 0;
651         __flush_dcache_range((unsigned long) bucket,
652                              ((unsigned long) bucket +
653                               sizeof(struct ino_bucket)));
654
655         virt_irq = virt_irq_alloc(devhandle, devino);
656         bucket_set_virt_irq(__pa(bucket), virt_irq);
657
658         set_irq_chip_and_handler_name(virt_irq, &sun4v_virq,
659                                       handle_fasteoi_irq,
660                                       "IVEC");
661
662         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
663         if (unlikely(!data))
664                 return 0;
665
666         /* In order to make the LDC channel startup sequence easier,
667          * especially wrt. locking, we do not let request_irq() enable
668          * the interrupt.
669          */
670         desc = irq_desc + virt_irq;
671         desc->status |= IRQ_NOAUTOEN;
672
673         set_irq_chip_data(virt_irq, data);
674
675         /* Catch accidental accesses to these things.  IMAP/ICLR handling
676          * is done by hypervisor calls on sun4v platforms, not by direct
677          * register accesses.
678          */
679         data->imap = ~0UL;
680         data->iclr = ~0UL;
681
682         cookie = ~__pa(bucket);
683         hv_err = sun4v_vintr_set_cookie(devhandle, devino, cookie);
684         if (hv_err) {
685                 prom_printf("IRQ: Fatal, cannot set cookie for [%x:%x] "
686                             "err=%lu\n", devhandle, devino, hv_err);
687                 prom_halt();
688         }
689
690         return virt_irq;
691 }
692
693 void ack_bad_irq(unsigned int virt_irq)
694 {
695         unsigned int ino = virt_irq_table[virt_irq].dev_ino;
696
697         if (!ino)
698                 ino = 0xdeadbeef;
699
700         printk(KERN_CRIT "Unexpected IRQ from ino[%x] virt_irq[%u]\n",
701                ino, virt_irq);
702 }
703
704 void *hardirq_stack[NR_CPUS];
705 void *softirq_stack[NR_CPUS];
706
707 static __attribute__((always_inline)) void *set_hardirq_stack(void)
708 {
709         void *orig_sp, *sp = hardirq_stack[smp_processor_id()];
710
711         __asm__ __volatile__("mov %%sp, %0" : "=r" (orig_sp));
712         if (orig_sp < sp ||
713             orig_sp > (sp + THREAD_SIZE)) {
714                 sp += THREAD_SIZE - 192 - STACK_BIAS;
715                 __asm__ __volatile__("mov %0, %%sp" : : "r" (sp));
716         }
717
718         return orig_sp;
719 }
720 static __attribute__((always_inline)) void restore_hardirq_stack(void *orig_sp)
721 {
722         __asm__ __volatile__("mov %0, %%sp" : : "r" (orig_sp));
723 }
724
725 void __irq_entry handler_irq(int irq, struct pt_regs *regs)
726 {
727         unsigned long pstate, bucket_pa;
728         struct pt_regs *old_regs;
729         void *orig_sp;
730
731         clear_softint(1 << irq);
732
733         old_regs = set_irq_regs(regs);
734         irq_enter();
735
736         /* Grab an atomic snapshot of the pending IVECs.  */
737         __asm__ __volatile__("rdpr      %%pstate, %0\n\t"
738                              "wrpr      %0, %3, %%pstate\n\t"
739                              "ldx       [%2], %1\n\t"
740                              "stx       %%g0, [%2]\n\t"
741                              "wrpr      %0, 0x0, %%pstate\n\t"
742                              : "=&r" (pstate), "=&r" (bucket_pa)
743                              : "r" (irq_work_pa(smp_processor_id())),
744                                "i" (PSTATE_IE)
745                              : "memory");
746
747         orig_sp = set_hardirq_stack();
748
749         while (bucket_pa) {
750                 struct irq_desc *desc;
751                 unsigned long next_pa;
752                 unsigned int virt_irq;
753
754                 next_pa = bucket_get_chain_pa(bucket_pa);
755                 virt_irq = bucket_get_virt_irq(bucket_pa);
756                 bucket_clear_chain_pa(bucket_pa);
757
758                 desc = irq_desc + virt_irq;
759
760                 if (!(desc->status & IRQ_DISABLED))
761                         desc->handle_irq(virt_irq, desc);
762
763                 bucket_pa = next_pa;
764         }
765
766         restore_hardirq_stack(orig_sp);
767
768         irq_exit();
769         set_irq_regs(old_regs);
770 }
771
772 void do_softirq(void)
773 {
774         unsigned long flags;
775
776         if (in_interrupt())
777                 return;
778
779         local_irq_save(flags);
780
781         if (local_softirq_pending()) {
782                 void *orig_sp, *sp = softirq_stack[smp_processor_id()];
783
784                 sp += THREAD_SIZE - 192 - STACK_BIAS;
785
786                 __asm__ __volatile__("mov %%sp, %0\n\t"
787                                      "mov %1, %%sp"
788                                      : "=&r" (orig_sp)
789                                      : "r" (sp));
790                 __do_softirq();
791                 __asm__ __volatile__("mov %0, %%sp"
792                                      : : "r" (orig_sp));
793         }
794
795         local_irq_restore(flags);
796 }
797
798 #ifdef CONFIG_HOTPLUG_CPU
799 void fixup_irqs(void)
800 {
801         unsigned int irq;
802
803         for (irq = 0; irq < NR_IRQS; irq++) {
804                 unsigned long flags;
805
806                 raw_spin_lock_irqsave(&irq_desc[irq].lock, flags);
807                 if (irq_desc[irq].action &&
808                     !(irq_desc[irq].status & IRQ_PER_CPU)) {
809                         if (irq_desc[irq].chip->set_affinity)
810                                 irq_desc[irq].chip->set_affinity(irq,
811                                         irq_desc[irq].affinity);
812                 }
813                 raw_spin_unlock_irqrestore(&irq_desc[irq].lock, flags);
814         }
815
816         tick_ops->disable_irq();
817 }
818 #endif
819
820 struct sun5_timer {
821         u64     count0;
822         u64     limit0;
823         u64     count1;
824         u64     limit1;
825 };
826
827 static struct sun5_timer *prom_timers;
828 static u64 prom_limit0, prom_limit1;
829
830 static void map_prom_timers(void)
831 {
832         struct device_node *dp;
833         const unsigned int *addr;
834
835         /* PROM timer node hangs out in the top level of device siblings... */
836         dp = of_find_node_by_path("/");
837         dp = dp->child;
838         while (dp) {
839                 if (!strcmp(dp->name, "counter-timer"))
840                         break;
841                 dp = dp->sibling;
842         }
843
844         /* Assume if node is not present, PROM uses different tick mechanism
845          * which we should not care about.
846          */
847         if (!dp) {
848                 prom_timers = (struct sun5_timer *) 0;
849                 return;
850         }
851
852         /* If PROM is really using this, it must be mapped by him. */
853         addr = of_get_property(dp, "address", NULL);
854         if (!addr) {
855                 prom_printf("PROM does not have timer mapped, trying to continue.\n");
856                 prom_timers = (struct sun5_timer *) 0;
857                 return;
858         }
859         prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
860 }
861
862 static void kill_prom_timer(void)
863 {
864         if (!prom_timers)
865                 return;
866
867         /* Save them away for later. */
868         prom_limit0 = prom_timers->limit0;
869         prom_limit1 = prom_timers->limit1;
870
871         /* Just as in sun4c/sun4m PROM uses timer which ticks at IRQ 14.
872          * We turn both off here just to be paranoid.
873          */
874         prom_timers->limit0 = 0;
875         prom_timers->limit1 = 0;
876
877         /* Wheee, eat the interrupt packet too... */
878         __asm__ __volatile__(
879 "       mov     0x40, %%g2\n"
880 "       ldxa    [%%g0] %0, %%g1\n"
881 "       ldxa    [%%g2] %1, %%g1\n"
882 "       stxa    %%g0, [%%g0] %0\n"
883 "       membar  #Sync\n"
884         : /* no outputs */
885         : "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
886         : "g1", "g2");
887 }
888
889 void notrace init_irqwork_curcpu(void)
890 {
891         int cpu = hard_smp_processor_id();
892
893         trap_block[cpu].irq_worklist_pa = 0UL;
894 }
895
896 /* Please be very careful with register_one_mondo() and
897  * sun4v_register_mondo_queues().
898  *
899  * On SMP this gets invoked from the CPU trampoline before
900  * the cpu has fully taken over the trap table from OBP,
901  * and it's kernel stack + %g6 thread register state is
902  * not fully cooked yet.
903  *
904  * Therefore you cannot make any OBP calls, not even prom_printf,
905  * from these two routines.
906  */
907 static void __cpuinit notrace register_one_mondo(unsigned long paddr, unsigned long type, unsigned long qmask)
908 {
909         unsigned long num_entries = (qmask + 1) / 64;
910         unsigned long status;
911
912         status = sun4v_cpu_qconf(type, paddr, num_entries);
913         if (status != HV_EOK) {
914                 prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
915                             "err %lu\n", type, paddr, num_entries, status);
916                 prom_halt();
917         }
918 }
919
920 void __cpuinit notrace sun4v_register_mondo_queues(int this_cpu)
921 {
922         struct trap_per_cpu *tb = &trap_block[this_cpu];
923
924         register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
925                            tb->cpu_mondo_qmask);
926         register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
927                            tb->dev_mondo_qmask);
928         register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
929                            tb->resum_qmask);
930         register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
931                            tb->nonresum_qmask);
932 }
933
934 /* Each queue region must be a power of 2 multiple of 64 bytes in
935  * size.  The base real address must be aligned to the size of the
936  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
937  */
938 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
939 {
940         unsigned long size = PAGE_ALIGN(qmask + 1);
941         unsigned long order = get_order(size);
942         unsigned long p;
943
944         p = __get_free_pages(GFP_KERNEL, order);
945         if (!p) {
946                 prom_printf("SUN4V: Error, cannot allocate queue.\n");
947                 prom_halt();
948         }
949
950         *pa_ptr = __pa(p);
951 }
952
953 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
954 {
955 #ifdef CONFIG_SMP
956         unsigned long page;
957
958         BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
959
960         page = get_zeroed_page(GFP_KERNEL);
961         if (!page) {
962                 prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
963                 prom_halt();
964         }
965
966         tb->cpu_mondo_block_pa = __pa(page);
967         tb->cpu_list_pa = __pa(page + 64);
968 #endif
969 }
970
971 /* Allocate mondo and error queues for all possible cpus.  */
972 static void __init sun4v_init_mondo_queues(void)
973 {
974         int cpu;
975
976         for_each_possible_cpu(cpu) {
977                 struct trap_per_cpu *tb = &trap_block[cpu];
978
979                 alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
980                 alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
981                 alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
982                 alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
983                 alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
984                 alloc_one_queue(&tb->nonresum_kernel_buf_pa,
985                                 tb->nonresum_qmask);
986         }
987 }
988
989 static void __init init_send_mondo_info(void)
990 {
991         int cpu;
992
993         for_each_possible_cpu(cpu) {
994                 struct trap_per_cpu *tb = &trap_block[cpu];
995
996                 init_cpu_send_mondo_info(tb);
997         }
998 }
999
1000 static struct irqaction timer_irq_action = {
1001         .name = "timer",
1002 };
1003
1004 /* Only invoked on boot processor. */
1005 void __init init_IRQ(void)
1006 {
1007         unsigned long size;
1008
1009         map_prom_timers();
1010         kill_prom_timer();
1011
1012         size = sizeof(struct ino_bucket) * NUM_IVECS;
1013         ivector_table = kzalloc(size, GFP_KERNEL);
1014         if (!ivector_table) {
1015                 prom_printf("Fatal error, cannot allocate ivector_table\n");
1016                 prom_halt();
1017         }
1018         __flush_dcache_range((unsigned long) ivector_table,
1019                              ((unsigned long) ivector_table) + size);
1020
1021         ivector_table_pa = __pa(ivector_table);
1022
1023         if (tlb_type == hypervisor)
1024                 sun4v_init_mondo_queues();
1025
1026         init_send_mondo_info();
1027
1028         if (tlb_type == hypervisor) {
1029                 /* Load up the boot cpu's entries.  */
1030                 sun4v_register_mondo_queues(hard_smp_processor_id());
1031         }
1032
1033         /* We need to clear any IRQ's pending in the soft interrupt
1034          * registers, a spurious one could be left around from the
1035          * PROM timer which we just disabled.
1036          */
1037         clear_softint(get_softint());
1038
1039         /* Now that ivector table is initialized, it is safe
1040          * to receive IRQ vector traps.  We will normally take
1041          * one or two right now, in case some device PROM used
1042          * to boot us wants to speak to us.  We just ignore them.
1043          */
1044         __asm__ __volatile__("rdpr      %%pstate, %%g1\n\t"
1045                              "or        %%g1, %0, %%g1\n\t"
1046                              "wrpr      %%g1, 0x0, %%pstate"
1047                              : /* No outputs */
1048                              : "i" (PSTATE_IE)
1049                              : "g1");
1050
1051         irq_desc[0].action = &timer_irq_action;
1052 }