Merge tag 'trace-3.18-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[sfrench/cifs-2.6.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/sched.h>
9 #include <linux/linkage.h>
10 #include <linux/ptrace.h>
11 #include <linux/errno.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/signal.h>
14 #include <linux/mm.h>
15 #include <linux/interrupt.h>
16 #include <linux/slab.h>
17 #include <linux/random.h>
18 #include <linux/init.h>
19 #include <linux/delay.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/ftrace.h>
23 #include <linux/irq.h>
24 #include <linux/kmemleak.h>
25
26 #include <asm/ptrace.h>
27 #include <asm/processor.h>
28 #include <linux/atomic.h>
29 #include <asm/irq.h>
30 #include <asm/io.h>
31 #include <asm/iommu.h>
32 #include <asm/upa.h>
33 #include <asm/oplib.h>
34 #include <asm/prom.h>
35 #include <asm/timer.h>
36 #include <asm/smp.h>
37 #include <asm/starfire.h>
38 #include <asm/uaccess.h>
39 #include <asm/cache.h>
40 #include <asm/cpudata.h>
41 #include <asm/auxio.h>
42 #include <asm/head.h>
43 #include <asm/hypervisor.h>
44 #include <asm/cacheflush.h>
45
46 #include "entry.h"
47 #include "cpumap.h"
48 #include "kstack.h"
49
50 struct ino_bucket *ivector_table;
51 unsigned long ivector_table_pa;
52
53 /* On several sun4u processors, it is illegal to mix bypass and
54  * non-bypass accesses.  Therefore we access all INO buckets
55  * using bypass accesses only.
56  */
57 static unsigned long bucket_get_chain_pa(unsigned long bucket_pa)
58 {
59         unsigned long ret;
60
61         __asm__ __volatile__("ldxa      [%1] %2, %0"
62                              : "=&r" (ret)
63                              : "r" (bucket_pa +
64                                     offsetof(struct ino_bucket,
65                                              __irq_chain_pa)),
66                                "i" (ASI_PHYS_USE_EC));
67
68         return ret;
69 }
70
71 static void bucket_clear_chain_pa(unsigned long bucket_pa)
72 {
73         __asm__ __volatile__("stxa      %%g0, [%0] %1"
74                              : /* no outputs */
75                              : "r" (bucket_pa +
76                                     offsetof(struct ino_bucket,
77                                              __irq_chain_pa)),
78                                "i" (ASI_PHYS_USE_EC));
79 }
80
81 static unsigned int bucket_get_irq(unsigned long bucket_pa)
82 {
83         unsigned int ret;
84
85         __asm__ __volatile__("lduwa     [%1] %2, %0"
86                              : "=&r" (ret)
87                              : "r" (bucket_pa +
88                                     offsetof(struct ino_bucket,
89                                              __irq)),
90                                "i" (ASI_PHYS_USE_EC));
91
92         return ret;
93 }
94
95 static void bucket_set_irq(unsigned long bucket_pa, unsigned int irq)
96 {
97         __asm__ __volatile__("stwa      %0, [%1] %2"
98                              : /* no outputs */
99                              : "r" (irq),
100                                "r" (bucket_pa +
101                                     offsetof(struct ino_bucket,
102                                              __irq)),
103                                "i" (ASI_PHYS_USE_EC));
104 }
105
106 #define irq_work_pa(__cpu)      &(trap_block[(__cpu)].irq_worklist_pa)
107
108 static unsigned long hvirq_major __initdata;
109 static int __init early_hvirq_major(char *p)
110 {
111         int rc = kstrtoul(p, 10, &hvirq_major);
112
113         return rc;
114 }
115 early_param("hvirq", early_hvirq_major);
116
117 static int hv_irq_version;
118
119 /* Major version 2.0 of HV_GRP_INTR added support for the VIRQ cookie
120  * based interfaces, but:
121  *
122  * 1) Several OSs, Solaris and Linux included, use them even when only
123  *    negotiating version 1.0 (or failing to negotiate at all).  So the
124  *    hypervisor has a workaround that provides the VIRQ interfaces even
125  *    when only verion 1.0 of the API is in use.
126  *
127  * 2) Second, and more importantly, with major version 2.0 these VIRQ
128  *    interfaces only were actually hooked up for LDC interrupts, even
129  *    though the Hypervisor specification clearly stated:
130  *
131  *      The new interrupt API functions will be available to a guest
132  *      when it negotiates version 2.0 in the interrupt API group 0x2. When
133  *      a guest negotiates version 2.0, all interrupt sources will only
134  *      support using the cookie interface, and any attempt to use the
135  *      version 1.0 interrupt APIs numbered 0xa0 to 0xa6 will result in the
136  *      ENOTSUPPORTED error being returned.
137  *
138  *   with an emphasis on "all interrupt sources".
139  *
140  * To correct this, major version 3.0 was created which does actually
141  * support VIRQs for all interrupt sources (not just LDC devices).  So
142  * if we want to move completely over the cookie based VIRQs we must
143  * negotiate major version 3.0 or later of HV_GRP_INTR.
144  */
145 static bool sun4v_cookie_only_virqs(void)
146 {
147         if (hv_irq_version >= 3)
148                 return true;
149         return false;
150 }
151
152 static void __init irq_init_hv(void)
153 {
154         unsigned long hv_error, major, minor = 0;
155
156         if (tlb_type != hypervisor)
157                 return;
158
159         if (hvirq_major)
160                 major = hvirq_major;
161         else
162                 major = 3;
163
164         hv_error = sun4v_hvapi_register(HV_GRP_INTR, major, &minor);
165         if (!hv_error)
166                 hv_irq_version = major;
167         else
168                 hv_irq_version = 1;
169
170         pr_info("SUN4V: Using IRQ API major %d, cookie only virqs %s\n",
171                 hv_irq_version,
172                 sun4v_cookie_only_virqs() ? "enabled" : "disabled");
173 }
174
175 /* This function is for the timer interrupt.*/
176 int __init arch_probe_nr_irqs(void)
177 {
178         return 1;
179 }
180
181 #define DEFAULT_NUM_IVECS       (0xfffU)
182 static unsigned int nr_ivec = DEFAULT_NUM_IVECS;
183 #define NUM_IVECS (nr_ivec)
184
185 static unsigned int __init size_nr_ivec(void)
186 {
187         if (tlb_type == hypervisor) {
188                 switch (sun4v_chip_type) {
189                 /* Athena's devhandle|devino is large.*/
190                 case SUN4V_CHIP_SPARC64X:
191                         nr_ivec = 0xffff;
192                         break;
193                 }
194         }
195         return nr_ivec;
196 }
197
198 struct irq_handler_data {
199         union {
200                 struct {
201                         unsigned int dev_handle;
202                         unsigned int dev_ino;
203                 };
204                 unsigned long sysino;
205         };
206         struct ino_bucket bucket;
207         unsigned long   iclr;
208         unsigned long   imap;
209 };
210
211 static inline unsigned int irq_data_to_handle(struct irq_data *data)
212 {
213         struct irq_handler_data *ihd = data->handler_data;
214
215         return ihd->dev_handle;
216 }
217
218 static inline unsigned int irq_data_to_ino(struct irq_data *data)
219 {
220         struct irq_handler_data *ihd = data->handler_data;
221
222         return ihd->dev_ino;
223 }
224
225 static inline unsigned long irq_data_to_sysino(struct irq_data *data)
226 {
227         struct irq_handler_data *ihd = data->handler_data;
228
229         return ihd->sysino;
230 }
231
232 void irq_free(unsigned int irq)
233 {
234         void *data = irq_get_handler_data(irq);
235
236         kfree(data);
237         irq_set_handler_data(irq, NULL);
238         irq_free_descs(irq, 1);
239 }
240
241 unsigned int irq_alloc(unsigned int dev_handle, unsigned int dev_ino)
242 {
243         int irq;
244
245         irq = __irq_alloc_descs(-1, 1, 1, numa_node_id(), NULL);
246         if (irq <= 0)
247                 goto out;
248
249         return irq;
250 out:
251         return 0;
252 }
253
254 static unsigned int cookie_exists(u32 devhandle, unsigned int devino)
255 {
256         unsigned long hv_err, cookie;
257         struct ino_bucket *bucket;
258         unsigned int irq = 0U;
259
260         hv_err = sun4v_vintr_get_cookie(devhandle, devino, &cookie);
261         if (hv_err) {
262                 pr_err("HV get cookie failed hv_err = %ld\n", hv_err);
263                 goto out;
264         }
265
266         if (cookie & ((1UL << 63UL))) {
267                 cookie = ~cookie;
268                 bucket = (struct ino_bucket *) __va(cookie);
269                 irq = bucket->__irq;
270         }
271 out:
272         return irq;
273 }
274
275 static unsigned int sysino_exists(u32 devhandle, unsigned int devino)
276 {
277         unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
278         struct ino_bucket *bucket;
279         unsigned int irq;
280
281         bucket = &ivector_table[sysino];
282         irq = bucket_get_irq(__pa(bucket));
283
284         return irq;
285 }
286
287 void ack_bad_irq(unsigned int irq)
288 {
289         pr_crit("BAD IRQ ack %d\n", irq);
290 }
291
292 void irq_install_pre_handler(int irq,
293                              void (*func)(unsigned int, void *, void *),
294                              void *arg1, void *arg2)
295 {
296         pr_warn("IRQ pre handler NOT supported.\n");
297 }
298
299 /*
300  * /proc/interrupts printing:
301  */
302 int arch_show_interrupts(struct seq_file *p, int prec)
303 {
304         int j;
305
306         seq_printf(p, "NMI: ");
307         for_each_online_cpu(j)
308                 seq_printf(p, "%10u ", cpu_data(j).__nmi_count);
309         seq_printf(p, "     Non-maskable interrupts\n");
310         return 0;
311 }
312
313 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
314 {
315         unsigned int tid;
316
317         if (this_is_starfire) {
318                 tid = starfire_translate(imap, cpuid);
319                 tid <<= IMAP_TID_SHIFT;
320                 tid &= IMAP_TID_UPA;
321         } else {
322                 if (tlb_type == cheetah || tlb_type == cheetah_plus) {
323                         unsigned long ver;
324
325                         __asm__ ("rdpr %%ver, %0" : "=r" (ver));
326                         if ((ver >> 32UL) == __JALAPENO_ID ||
327                             (ver >> 32UL) == __SERRANO_ID) {
328                                 tid = cpuid << IMAP_TID_SHIFT;
329                                 tid &= IMAP_TID_JBUS;
330                         } else {
331                                 unsigned int a = cpuid & 0x1f;
332                                 unsigned int n = (cpuid >> 5) & 0x1f;
333
334                                 tid = ((a << IMAP_AID_SHIFT) |
335                                        (n << IMAP_NID_SHIFT));
336                                 tid &= (IMAP_AID_SAFARI |
337                                         IMAP_NID_SAFARI);
338                         }
339                 } else {
340                         tid = cpuid << IMAP_TID_SHIFT;
341                         tid &= IMAP_TID_UPA;
342                 }
343         }
344
345         return tid;
346 }
347
348 #ifdef CONFIG_SMP
349 static int irq_choose_cpu(unsigned int irq, const struct cpumask *affinity)
350 {
351         cpumask_t mask;
352         int cpuid;
353
354         cpumask_copy(&mask, affinity);
355         if (cpumask_equal(&mask, cpu_online_mask)) {
356                 cpuid = map_to_cpu(irq);
357         } else {
358                 cpumask_t tmp;
359
360                 cpumask_and(&tmp, cpu_online_mask, &mask);
361                 cpuid = cpumask_empty(&tmp) ? map_to_cpu(irq) : cpumask_first(&tmp);
362         }
363
364         return cpuid;
365 }
366 #else
367 #define irq_choose_cpu(irq, affinity)   \
368         real_hard_smp_processor_id()
369 #endif
370
371 static void sun4u_irq_enable(struct irq_data *data)
372 {
373         struct irq_handler_data *handler_data = data->handler_data;
374
375         if (likely(handler_data)) {
376                 unsigned long cpuid, imap, val;
377                 unsigned int tid;
378
379                 cpuid = irq_choose_cpu(data->irq, data->affinity);
380                 imap = handler_data->imap;
381
382                 tid = sun4u_compute_tid(imap, cpuid);
383
384                 val = upa_readq(imap);
385                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
386                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
387                 val |= tid | IMAP_VALID;
388                 upa_writeq(val, imap);
389                 upa_writeq(ICLR_IDLE, handler_data->iclr);
390         }
391 }
392
393 static int sun4u_set_affinity(struct irq_data *data,
394                                const struct cpumask *mask, bool force)
395 {
396         struct irq_handler_data *handler_data = data->handler_data;
397
398         if (likely(handler_data)) {
399                 unsigned long cpuid, imap, val;
400                 unsigned int tid;
401
402                 cpuid = irq_choose_cpu(data->irq, mask);
403                 imap = handler_data->imap;
404
405                 tid = sun4u_compute_tid(imap, cpuid);
406
407                 val = upa_readq(imap);
408                 val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
409                          IMAP_AID_SAFARI | IMAP_NID_SAFARI);
410                 val |= tid | IMAP_VALID;
411                 upa_writeq(val, imap);
412                 upa_writeq(ICLR_IDLE, handler_data->iclr);
413         }
414
415         return 0;
416 }
417
418 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
419  * handler_irq() will skip the handler call and that will leave the
420  * interrupt in the sent state.  The next ->enable() call will hit the
421  * ICLR register to reset the state machine.
422  *
423  * This scheme is necessary, instead of clearing the Valid bit in the
424  * IMAP register, to handle the case of IMAP registers being shared by
425  * multiple INOs (and thus ICLR registers).  Since we use a different
426  * virtual IRQ for each shared IMAP instance, the generic code thinks
427  * there is only one user so it prematurely calls ->disable() on
428  * free_irq().
429  *
430  * We have to provide an explicit ->disable() method instead of using
431  * NULL to get the default.  The reason is that if the generic code
432  * sees that, it also hooks up a default ->shutdown method which
433  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
434  */
435 static void sun4u_irq_disable(struct irq_data *data)
436 {
437 }
438
439 static void sun4u_irq_eoi(struct irq_data *data)
440 {
441         struct irq_handler_data *handler_data = data->handler_data;
442
443         if (likely(handler_data))
444                 upa_writeq(ICLR_IDLE, handler_data->iclr);
445 }
446
447 static void sun4v_irq_enable(struct irq_data *data)
448 {
449         unsigned long cpuid = irq_choose_cpu(data->irq, data->affinity);
450         unsigned int ino = irq_data_to_sysino(data);
451         int err;
452
453         err = sun4v_intr_settarget(ino, cpuid);
454         if (err != HV_EOK)
455                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
456                        "err(%d)\n", ino, cpuid, err);
457         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
458         if (err != HV_EOK)
459                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
460                        "err(%d)\n", ino, err);
461         err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
462         if (err != HV_EOK)
463                 printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
464                        ino, err);
465 }
466
467 static int sun4v_set_affinity(struct irq_data *data,
468                                const struct cpumask *mask, bool force)
469 {
470         unsigned long cpuid = irq_choose_cpu(data->irq, mask);
471         unsigned int ino = irq_data_to_sysino(data);
472         int err;
473
474         err = sun4v_intr_settarget(ino, cpuid);
475         if (err != HV_EOK)
476                 printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
477                        "err(%d)\n", ino, cpuid, err);
478
479         return 0;
480 }
481
482 static void sun4v_irq_disable(struct irq_data *data)
483 {
484         unsigned int ino = irq_data_to_sysino(data);
485         int err;
486
487         err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
488         if (err != HV_EOK)
489                 printk(KERN_ERR "sun4v_intr_setenabled(%x): "
490                        "err(%d)\n", ino, err);
491 }
492
493 static void sun4v_irq_eoi(struct irq_data *data)
494 {
495         unsigned int ino = irq_data_to_sysino(data);
496         int err;
497
498         err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
499         if (err != HV_EOK)
500                 printk(KERN_ERR "sun4v_intr_setstate(%x): "
501                        "err(%d)\n", ino, err);
502 }
503
504 static void sun4v_virq_enable(struct irq_data *data)
505 {
506         unsigned long dev_handle = irq_data_to_handle(data);
507         unsigned long dev_ino = irq_data_to_ino(data);
508         unsigned long cpuid;
509         int err;
510
511         cpuid = irq_choose_cpu(data->irq, data->affinity);
512
513         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
514         if (err != HV_EOK)
515                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
516                        "err(%d)\n",
517                        dev_handle, dev_ino, cpuid, err);
518         err = sun4v_vintr_set_state(dev_handle, dev_ino,
519                                     HV_INTR_STATE_IDLE);
520         if (err != HV_EOK)
521                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
522                        "HV_INTR_STATE_IDLE): err(%d)\n",
523                        dev_handle, dev_ino, err);
524         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
525                                     HV_INTR_ENABLED);
526         if (err != HV_EOK)
527                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
528                        "HV_INTR_ENABLED): err(%d)\n",
529                        dev_handle, dev_ino, err);
530 }
531
532 static int sun4v_virt_set_affinity(struct irq_data *data,
533                                     const struct cpumask *mask, bool force)
534 {
535         unsigned long dev_handle = irq_data_to_handle(data);
536         unsigned long dev_ino = irq_data_to_ino(data);
537         unsigned long cpuid;
538         int err;
539
540         cpuid = irq_choose_cpu(data->irq, mask);
541
542         err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
543         if (err != HV_EOK)
544                 printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
545                        "err(%d)\n",
546                        dev_handle, dev_ino, cpuid, err);
547
548         return 0;
549 }
550
551 static void sun4v_virq_disable(struct irq_data *data)
552 {
553         unsigned long dev_handle = irq_data_to_handle(data);
554         unsigned long dev_ino = irq_data_to_ino(data);
555         int err;
556
557
558         err = sun4v_vintr_set_valid(dev_handle, dev_ino,
559                                     HV_INTR_DISABLED);
560         if (err != HV_EOK)
561                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
562                        "HV_INTR_DISABLED): err(%d)\n",
563                        dev_handle, dev_ino, err);
564 }
565
566 static void sun4v_virq_eoi(struct irq_data *data)
567 {
568         unsigned long dev_handle = irq_data_to_handle(data);
569         unsigned long dev_ino = irq_data_to_ino(data);
570         int err;
571
572         err = sun4v_vintr_set_state(dev_handle, dev_ino,
573                                     HV_INTR_STATE_IDLE);
574         if (err != HV_EOK)
575                 printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
576                        "HV_INTR_STATE_IDLE): err(%d)\n",
577                        dev_handle, dev_ino, err);
578 }
579
580 static struct irq_chip sun4u_irq = {
581         .name                   = "sun4u",
582         .irq_enable             = sun4u_irq_enable,
583         .irq_disable            = sun4u_irq_disable,
584         .irq_eoi                = sun4u_irq_eoi,
585         .irq_set_affinity       = sun4u_set_affinity,
586         .flags                  = IRQCHIP_EOI_IF_HANDLED,
587 };
588
589 static struct irq_chip sun4v_irq = {
590         .name                   = "sun4v",
591         .irq_enable             = sun4v_irq_enable,
592         .irq_disable            = sun4v_irq_disable,
593         .irq_eoi                = sun4v_irq_eoi,
594         .irq_set_affinity       = sun4v_set_affinity,
595         .flags                  = IRQCHIP_EOI_IF_HANDLED,
596 };
597
598 static struct irq_chip sun4v_virq = {
599         .name                   = "vsun4v",
600         .irq_enable             = sun4v_virq_enable,
601         .irq_disable            = sun4v_virq_disable,
602         .irq_eoi                = sun4v_virq_eoi,
603         .irq_set_affinity       = sun4v_virt_set_affinity,
604         .flags                  = IRQCHIP_EOI_IF_HANDLED,
605 };
606
607 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
608 {
609         struct irq_handler_data *handler_data;
610         struct ino_bucket *bucket;
611         unsigned int irq;
612         int ino;
613
614         BUG_ON(tlb_type == hypervisor);
615
616         ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
617         bucket = &ivector_table[ino];
618         irq = bucket_get_irq(__pa(bucket));
619         if (!irq) {
620                 irq = irq_alloc(0, ino);
621                 bucket_set_irq(__pa(bucket), irq);
622                 irq_set_chip_and_handler_name(irq, &sun4u_irq,
623                                               handle_fasteoi_irq, "IVEC");
624         }
625
626         handler_data = irq_get_handler_data(irq);
627         if (unlikely(handler_data))
628                 goto out;
629
630         handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
631         if (unlikely(!handler_data)) {
632                 prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
633                 prom_halt();
634         }
635         irq_set_handler_data(irq, handler_data);
636
637         handler_data->imap  = imap;
638         handler_data->iclr  = iclr;
639
640 out:
641         return irq;
642 }
643
644 static unsigned int sun4v_build_common(u32 devhandle, unsigned int devino,
645                 void (*handler_data_init)(struct irq_handler_data *data,
646                 u32 devhandle, unsigned int devino),
647                 struct irq_chip *chip)
648 {
649         struct irq_handler_data *data;
650         unsigned int irq;
651
652         irq = irq_alloc(devhandle, devino);
653         if (!irq)
654                 goto out;
655
656         data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
657         if (unlikely(!data)) {
658                 pr_err("IRQ handler data allocation failed.\n");
659                 irq_free(irq);
660                 irq = 0;
661                 goto out;
662         }
663
664         irq_set_handler_data(irq, data);
665         handler_data_init(data, devhandle, devino);
666         irq_set_chip_and_handler_name(irq, chip, handle_fasteoi_irq, "IVEC");
667         data->imap = ~0UL;
668         data->iclr = ~0UL;
669 out:
670         return irq;
671 }
672
673 static unsigned long cookie_assign(unsigned int irq, u32 devhandle,
674                 unsigned int devino)
675 {
676         struct irq_handler_data *ihd = irq_get_handler_data(irq);
677         unsigned long hv_error, cookie;
678
679         /* handler_irq needs to find the irq. cookie is seen signed in
680          * sun4v_dev_mondo and treated as a non ivector_table delivery.
681          */
682         ihd->bucket.__irq = irq;
683         cookie = ~__pa(&ihd->bucket);
684
685         hv_error = sun4v_vintr_set_cookie(devhandle, devino, cookie);
686         if (hv_error)
687                 pr_err("HV vintr set cookie failed = %ld\n", hv_error);
688
689         return hv_error;
690 }
691
692 static void cookie_handler_data(struct irq_handler_data *data,
693                                 u32 devhandle, unsigned int devino)
694 {
695         data->dev_handle = devhandle;
696         data->dev_ino = devino;
697 }
698
699 static unsigned int cookie_build_irq(u32 devhandle, unsigned int devino,
700                                      struct irq_chip *chip)
701 {
702         unsigned long hv_error;
703         unsigned int irq;
704
705         irq = sun4v_build_common(devhandle, devino, cookie_handler_data, chip);
706
707         hv_error = cookie_assign(irq, devhandle, devino);
708         if (hv_error) {
709                 irq_free(irq);
710                 irq = 0;
711         }
712
713         return irq;
714 }
715
716 static unsigned int sun4v_build_cookie(u32 devhandle, unsigned int devino)
717 {
718         unsigned int irq;
719
720         irq = cookie_exists(devhandle, devino);
721         if (irq)
722                 goto out;
723
724         irq = cookie_build_irq(devhandle, devino, &sun4v_virq);
725
726 out:
727         return irq;
728 }
729
730 static void sysino_set_bucket(unsigned int irq)
731 {
732         struct irq_handler_data *ihd = irq_get_handler_data(irq);
733         struct ino_bucket *bucket;
734         unsigned long sysino;
735
736         sysino = sun4v_devino_to_sysino(ihd->dev_handle, ihd->dev_ino);
737         BUG_ON(sysino >= nr_ivec);
738         bucket = &ivector_table[sysino];
739         bucket_set_irq(__pa(bucket), irq);
740 }
741
742 static void sysino_handler_data(struct irq_handler_data *data,
743                                 u32 devhandle, unsigned int devino)
744 {
745         unsigned long sysino;
746
747         sysino = sun4v_devino_to_sysino(devhandle, devino);
748         data->sysino = sysino;
749 }
750
751 static unsigned int sysino_build_irq(u32 devhandle, unsigned int devino,
752                                      struct irq_chip *chip)
753 {
754         unsigned int irq;
755
756         irq = sun4v_build_common(devhandle, devino, sysino_handler_data, chip);
757         if (!irq)
758                 goto out;
759
760         sysino_set_bucket(irq);
761 out:
762         return irq;
763 }
764
765 static int sun4v_build_sysino(u32 devhandle, unsigned int devino)
766 {
767         int irq;
768
769         irq = sysino_exists(devhandle, devino);
770         if (irq)
771                 goto out;
772
773         irq = sysino_build_irq(devhandle, devino, &sun4v_irq);
774 out:
775         return irq;
776 }
777
778 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
779 {
780         unsigned int irq;
781
782         if (sun4v_cookie_only_virqs())
783                 irq = sun4v_build_cookie(devhandle, devino);
784         else
785                 irq = sun4v_build_sysino(devhandle, devino);
786
787         return irq;
788 }
789
790 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
791 {
792         int irq;
793
794         irq = cookie_build_irq(devhandle, devino, &sun4v_virq);
795         if (!irq)
796                 goto out;
797
798         /* This is borrowed from the original function.
799          */
800         irq_set_status_flags(irq, IRQ_NOAUTOEN);
801
802 out:
803         return irq;
804 }
805
806 void *hardirq_stack[NR_CPUS];
807 void *softirq_stack[NR_CPUS];
808
809 void __irq_entry handler_irq(int pil, struct pt_regs *regs)
810 {
811         unsigned long pstate, bucket_pa;
812         struct pt_regs *old_regs;
813         void *orig_sp;
814
815         clear_softint(1 << pil);
816
817         old_regs = set_irq_regs(regs);
818         irq_enter();
819
820         /* Grab an atomic snapshot of the pending IVECs.  */
821         __asm__ __volatile__("rdpr      %%pstate, %0\n\t"
822                              "wrpr      %0, %3, %%pstate\n\t"
823                              "ldx       [%2], %1\n\t"
824                              "stx       %%g0, [%2]\n\t"
825                              "wrpr      %0, 0x0, %%pstate\n\t"
826                              : "=&r" (pstate), "=&r" (bucket_pa)
827                              : "r" (irq_work_pa(smp_processor_id())),
828                                "i" (PSTATE_IE)
829                              : "memory");
830
831         orig_sp = set_hardirq_stack();
832
833         while (bucket_pa) {
834                 unsigned long next_pa;
835                 unsigned int irq;
836
837                 next_pa = bucket_get_chain_pa(bucket_pa);
838                 irq = bucket_get_irq(bucket_pa);
839                 bucket_clear_chain_pa(bucket_pa);
840
841                 generic_handle_irq(irq);
842
843                 bucket_pa = next_pa;
844         }
845
846         restore_hardirq_stack(orig_sp);
847
848         irq_exit();
849         set_irq_regs(old_regs);
850 }
851
852 void do_softirq_own_stack(void)
853 {
854         void *orig_sp, *sp = softirq_stack[smp_processor_id()];
855
856         sp += THREAD_SIZE - 192 - STACK_BIAS;
857
858         __asm__ __volatile__("mov %%sp, %0\n\t"
859                              "mov %1, %%sp"
860                              : "=&r" (orig_sp)
861                              : "r" (sp));
862         __do_softirq();
863         __asm__ __volatile__("mov %0, %%sp"
864                              : : "r" (orig_sp));
865 }
866
867 #ifdef CONFIG_HOTPLUG_CPU
868 void fixup_irqs(void)
869 {
870         unsigned int irq;
871
872         for (irq = 0; irq < NR_IRQS; irq++) {
873                 struct irq_desc *desc = irq_to_desc(irq);
874                 struct irq_data *data;
875                 unsigned long flags;
876
877                 if (!desc)
878                         continue;
879                 data = irq_desc_get_irq_data(desc);
880                 raw_spin_lock_irqsave(&desc->lock, flags);
881                 if (desc->action && !irqd_is_per_cpu(data)) {
882                         if (data->chip->irq_set_affinity)
883                                 data->chip->irq_set_affinity(data,
884                                                              data->affinity,
885                                                              false);
886                 }
887                 raw_spin_unlock_irqrestore(&desc->lock, flags);
888         }
889
890         tick_ops->disable_irq();
891 }
892 #endif
893
894 struct sun5_timer {
895         u64     count0;
896         u64     limit0;
897         u64     count1;
898         u64     limit1;
899 };
900
901 static struct sun5_timer *prom_timers;
902 static u64 prom_limit0, prom_limit1;
903
904 static void map_prom_timers(void)
905 {
906         struct device_node *dp;
907         const unsigned int *addr;
908
909         /* PROM timer node hangs out in the top level of device siblings... */
910         dp = of_find_node_by_path("/");
911         dp = dp->child;
912         while (dp) {
913                 if (!strcmp(dp->name, "counter-timer"))
914                         break;
915                 dp = dp->sibling;
916         }
917
918         /* Assume if node is not present, PROM uses different tick mechanism
919          * which we should not care about.
920          */
921         if (!dp) {
922                 prom_timers = (struct sun5_timer *) 0;
923                 return;
924         }
925
926         /* If PROM is really using this, it must be mapped by him. */
927         addr = of_get_property(dp, "address", NULL);
928         if (!addr) {
929                 prom_printf("PROM does not have timer mapped, trying to continue.\n");
930                 prom_timers = (struct sun5_timer *) 0;
931                 return;
932         }
933         prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
934 }
935
936 static void kill_prom_timer(void)
937 {
938         if (!prom_timers)
939                 return;
940
941         /* Save them away for later. */
942         prom_limit0 = prom_timers->limit0;
943         prom_limit1 = prom_timers->limit1;
944
945         /* Just as in sun4c PROM uses timer which ticks at IRQ 14.
946          * We turn both off here just to be paranoid.
947          */
948         prom_timers->limit0 = 0;
949         prom_timers->limit1 = 0;
950
951         /* Wheee, eat the interrupt packet too... */
952         __asm__ __volatile__(
953 "       mov     0x40, %%g2\n"
954 "       ldxa    [%%g0] %0, %%g1\n"
955 "       ldxa    [%%g2] %1, %%g1\n"
956 "       stxa    %%g0, [%%g0] %0\n"
957 "       membar  #Sync\n"
958         : /* no outputs */
959         : "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
960         : "g1", "g2");
961 }
962
963 void notrace init_irqwork_curcpu(void)
964 {
965         int cpu = hard_smp_processor_id();
966
967         trap_block[cpu].irq_worklist_pa = 0UL;
968 }
969
970 /* Please be very careful with register_one_mondo() and
971  * sun4v_register_mondo_queues().
972  *
973  * On SMP this gets invoked from the CPU trampoline before
974  * the cpu has fully taken over the trap table from OBP,
975  * and it's kernel stack + %g6 thread register state is
976  * not fully cooked yet.
977  *
978  * Therefore you cannot make any OBP calls, not even prom_printf,
979  * from these two routines.
980  */
981 static void notrace register_one_mondo(unsigned long paddr, unsigned long type,
982                                        unsigned long qmask)
983 {
984         unsigned long num_entries = (qmask + 1) / 64;
985         unsigned long status;
986
987         status = sun4v_cpu_qconf(type, paddr, num_entries);
988         if (status != HV_EOK) {
989                 prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
990                             "err %lu\n", type, paddr, num_entries, status);
991                 prom_halt();
992         }
993 }
994
995 void notrace sun4v_register_mondo_queues(int this_cpu)
996 {
997         struct trap_per_cpu *tb = &trap_block[this_cpu];
998
999         register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
1000                            tb->cpu_mondo_qmask);
1001         register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
1002                            tb->dev_mondo_qmask);
1003         register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
1004                            tb->resum_qmask);
1005         register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
1006                            tb->nonresum_qmask);
1007 }
1008
1009 /* Each queue region must be a power of 2 multiple of 64 bytes in
1010  * size.  The base real address must be aligned to the size of the
1011  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
1012  */
1013 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
1014 {
1015         unsigned long size = PAGE_ALIGN(qmask + 1);
1016         unsigned long order = get_order(size);
1017         unsigned long p;
1018
1019         p = __get_free_pages(GFP_KERNEL, order);
1020         if (!p) {
1021                 prom_printf("SUN4V: Error, cannot allocate queue.\n");
1022                 prom_halt();
1023         }
1024
1025         *pa_ptr = __pa(p);
1026 }
1027
1028 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
1029 {
1030 #ifdef CONFIG_SMP
1031         unsigned long page;
1032
1033         BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
1034
1035         page = get_zeroed_page(GFP_KERNEL);
1036         if (!page) {
1037                 prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
1038                 prom_halt();
1039         }
1040
1041         tb->cpu_mondo_block_pa = __pa(page);
1042         tb->cpu_list_pa = __pa(page + 64);
1043 #endif
1044 }
1045
1046 /* Allocate mondo and error queues for all possible cpus.  */
1047 static void __init sun4v_init_mondo_queues(void)
1048 {
1049         int cpu;
1050
1051         for_each_possible_cpu(cpu) {
1052                 struct trap_per_cpu *tb = &trap_block[cpu];
1053
1054                 alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
1055                 alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
1056                 alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
1057                 alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
1058                 alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
1059                 alloc_one_queue(&tb->nonresum_kernel_buf_pa,
1060                                 tb->nonresum_qmask);
1061         }
1062 }
1063
1064 static void __init init_send_mondo_info(void)
1065 {
1066         int cpu;
1067
1068         for_each_possible_cpu(cpu) {
1069                 struct trap_per_cpu *tb = &trap_block[cpu];
1070
1071                 init_cpu_send_mondo_info(tb);
1072         }
1073 }
1074
1075 static struct irqaction timer_irq_action = {
1076         .name = "timer",
1077 };
1078
1079 static void __init irq_ivector_init(void)
1080 {
1081         unsigned long size, order;
1082         unsigned int ivecs;
1083
1084         /* If we are doing cookie only VIRQs then we do not need the ivector
1085          * table to process interrupts.
1086          */
1087         if (sun4v_cookie_only_virqs())
1088                 return;
1089
1090         ivecs = size_nr_ivec();
1091         size = sizeof(struct ino_bucket) * ivecs;
1092         order = get_order(size);
1093         ivector_table = (struct ino_bucket *)
1094                 __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
1095         if (!ivector_table) {
1096                 prom_printf("Fatal error, cannot allocate ivector_table\n");
1097                 prom_halt();
1098         }
1099         __flush_dcache_range((unsigned long) ivector_table,
1100                              ((unsigned long) ivector_table) + size);
1101
1102         ivector_table_pa = __pa(ivector_table);
1103 }
1104
1105 /* Only invoked on boot processor.*/
1106 void __init init_IRQ(void)
1107 {
1108         irq_init_hv();
1109         irq_ivector_init();
1110         map_prom_timers();
1111         kill_prom_timer();
1112
1113         if (tlb_type == hypervisor)
1114                 sun4v_init_mondo_queues();
1115
1116         init_send_mondo_info();
1117
1118         if (tlb_type == hypervisor) {
1119                 /* Load up the boot cpu's entries.  */
1120                 sun4v_register_mondo_queues(hard_smp_processor_id());
1121         }
1122
1123         /* We need to clear any IRQ's pending in the soft interrupt
1124          * registers, a spurious one could be left around from the
1125          * PROM timer which we just disabled.
1126          */
1127         clear_softint(get_softint());
1128
1129         /* Now that ivector table is initialized, it is safe
1130          * to receive IRQ vector traps.  We will normally take
1131          * one or two right now, in case some device PROM used
1132          * to boot us wants to speak to us.  We just ignore them.
1133          */
1134         __asm__ __volatile__("rdpr      %%pstate, %%g1\n\t"
1135                              "or        %%g1, %0, %%g1\n\t"
1136                              "wrpr      %%g1, 0x0, %%pstate"
1137                              : /* No outputs */
1138                              : "i" (PSTATE_IE)
1139                              : "g1");
1140
1141         irq_to_desc(0)->action = &timer_irq_action;
1142 }