xref: /linux/arch/sparc/kernel/irq_64.c (revision 0be3ff0c)
1 // SPDX-License-Identifier: GPL-2.0
2 /* irq.c: UltraSparc IRQ handling/init/registry.
3  *
4  * Copyright (C) 1997, 2007, 2008 David S. Miller (davem@davemloft.net)
5  * Copyright (C) 1998  Eddie C. Dost    (ecd@skynet.be)
6  * Copyright (C) 1998  Jakub Jelinek    (jj@ultra.linux.cz)
7  */
8 
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 <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 <linux/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 #include <asm/softirq_stack.h>
46 
47 #include "entry.h"
48 #include "cpumap.h"
49 #include "kstack.h"
50 
51 struct ino_bucket *ivector_table;
52 unsigned long ivector_table_pa;
53 
54 /* On several sun4u processors, it is illegal to mix bypass and
55  * non-bypass accesses.  Therefore we access all INO buckets
56  * using bypass accesses only.
57  */
58 static unsigned long bucket_get_chain_pa(unsigned long bucket_pa)
59 {
60 	unsigned long ret;
61 
62 	__asm__ __volatile__("ldxa	[%1] %2, %0"
63 			     : "=&r" (ret)
64 			     : "r" (bucket_pa +
65 				    offsetof(struct ino_bucket,
66 					     __irq_chain_pa)),
67 			       "i" (ASI_PHYS_USE_EC));
68 
69 	return ret;
70 }
71 
72 static void bucket_clear_chain_pa(unsigned long bucket_pa)
73 {
74 	__asm__ __volatile__("stxa	%%g0, [%0] %1"
75 			     : /* no outputs */
76 			     : "r" (bucket_pa +
77 				    offsetof(struct ino_bucket,
78 					     __irq_chain_pa)),
79 			       "i" (ASI_PHYS_USE_EC));
80 }
81 
82 static unsigned int bucket_get_irq(unsigned long bucket_pa)
83 {
84 	unsigned int ret;
85 
86 	__asm__ __volatile__("lduwa	[%1] %2, %0"
87 			     : "=&r" (ret)
88 			     : "r" (bucket_pa +
89 				    offsetof(struct ino_bucket,
90 					     __irq)),
91 			       "i" (ASI_PHYS_USE_EC));
92 
93 	return ret;
94 }
95 
96 static void bucket_set_irq(unsigned long bucket_pa, unsigned int irq)
97 {
98 	__asm__ __volatile__("stwa	%0, [%1] %2"
99 			     : /* no outputs */
100 			     : "r" (irq),
101 			       "r" (bucket_pa +
102 				    offsetof(struct ino_bucket,
103 					     __irq)),
104 			       "i" (ASI_PHYS_USE_EC));
105 }
106 
107 #define irq_work_pa(__cpu)	&(trap_block[(__cpu)].irq_worklist_pa)
108 
109 static unsigned long hvirq_major __initdata;
110 static int __init early_hvirq_major(char *p)
111 {
112 	int rc = kstrtoul(p, 10, &hvirq_major);
113 
114 	return rc;
115 }
116 early_param("hvirq", early_hvirq_major);
117 
118 static int hv_irq_version;
119 
120 /* Major version 2.0 of HV_GRP_INTR added support for the VIRQ cookie
121  * based interfaces, but:
122  *
123  * 1) Several OSs, Solaris and Linux included, use them even when only
124  *    negotiating version 1.0 (or failing to negotiate at all).  So the
125  *    hypervisor has a workaround that provides the VIRQ interfaces even
126  *    when only verion 1.0 of the API is in use.
127  *
128  * 2) Second, and more importantly, with major version 2.0 these VIRQ
129  *    interfaces only were actually hooked up for LDC interrupts, even
130  *    though the Hypervisor specification clearly stated:
131  *
132  *	The new interrupt API functions will be available to a guest
133  *	when it negotiates version 2.0 in the interrupt API group 0x2. When
134  *	a guest negotiates version 2.0, all interrupt sources will only
135  *	support using the cookie interface, and any attempt to use the
136  *	version 1.0 interrupt APIs numbered 0xa0 to 0xa6 will result in the
137  *	ENOTSUPPORTED error being returned.
138  *
139  *   with an emphasis on "all interrupt sources".
140  *
141  * To correct this, major version 3.0 was created which does actually
142  * support VIRQs for all interrupt sources (not just LDC devices).  So
143  * if we want to move completely over the cookie based VIRQs we must
144  * negotiate major version 3.0 or later of HV_GRP_INTR.
145  */
146 static bool sun4v_cookie_only_virqs(void)
147 {
148 	if (hv_irq_version >= 3)
149 		return true;
150 	return false;
151 }
152 
153 static void __init irq_init_hv(void)
154 {
155 	unsigned long hv_error, major, minor = 0;
156 
157 	if (tlb_type != hypervisor)
158 		return;
159 
160 	if (hvirq_major)
161 		major = hvirq_major;
162 	else
163 		major = 3;
164 
165 	hv_error = sun4v_hvapi_register(HV_GRP_INTR, major, &minor);
166 	if (!hv_error)
167 		hv_irq_version = major;
168 	else
169 		hv_irq_version = 1;
170 
171 	pr_info("SUN4V: Using IRQ API major %d, cookie only virqs %s\n",
172 		hv_irq_version,
173 		sun4v_cookie_only_virqs() ? "enabled" : "disabled");
174 }
175 
176 /* This function is for the timer interrupt.*/
177 int __init arch_probe_nr_irqs(void)
178 {
179 	return 1;
180 }
181 
182 #define DEFAULT_NUM_IVECS	(0xfffU)
183 static unsigned int nr_ivec = DEFAULT_NUM_IVECS;
184 #define NUM_IVECS (nr_ivec)
185 
186 static unsigned int __init size_nr_ivec(void)
187 {
188 	if (tlb_type == hypervisor) {
189 		switch (sun4v_chip_type) {
190 		/* Athena's devhandle|devino is large.*/
191 		case SUN4V_CHIP_SPARC64X:
192 			nr_ivec = 0xffff;
193 			break;
194 		}
195 	}
196 	return nr_ivec;
197 }
198 
199 struct irq_handler_data {
200 	union {
201 		struct {
202 			unsigned int dev_handle;
203 			unsigned int dev_ino;
204 		};
205 		unsigned long sysino;
206 	};
207 	struct ino_bucket bucket;
208 	unsigned long	iclr;
209 	unsigned long	imap;
210 };
211 
212 static inline unsigned int irq_data_to_handle(struct irq_data *data)
213 {
214 	struct irq_handler_data *ihd = irq_data_get_irq_handler_data(data);
215 
216 	return ihd->dev_handle;
217 }
218 
219 static inline unsigned int irq_data_to_ino(struct irq_data *data)
220 {
221 	struct irq_handler_data *ihd = irq_data_get_irq_handler_data(data);
222 
223 	return ihd->dev_ino;
224 }
225 
226 static inline unsigned long irq_data_to_sysino(struct irq_data *data)
227 {
228 	struct irq_handler_data *ihd = irq_data_get_irq_handler_data(data);
229 
230 	return ihd->sysino;
231 }
232 
233 void irq_free(unsigned int irq)
234 {
235 	void *data = irq_get_handler_data(irq);
236 
237 	kfree(data);
238 	irq_set_handler_data(irq, NULL);
239 	irq_free_descs(irq, 1);
240 }
241 
242 unsigned int irq_alloc(unsigned int dev_handle, unsigned int dev_ino)
243 {
244 	int irq;
245 
246 	irq = __irq_alloc_descs(-1, 1, 1, numa_node_id(), NULL, NULL);
247 	if (irq <= 0)
248 		goto out;
249 
250 	return irq;
251 out:
252 	return 0;
253 }
254 
255 static unsigned int cookie_exists(u32 devhandle, unsigned int devino)
256 {
257 	unsigned long hv_err, cookie;
258 	struct ino_bucket *bucket;
259 	unsigned int irq = 0U;
260 
261 	hv_err = sun4v_vintr_get_cookie(devhandle, devino, &cookie);
262 	if (hv_err) {
263 		pr_err("HV get cookie failed hv_err = %ld\n", hv_err);
264 		goto out;
265 	}
266 
267 	if (cookie & ((1UL << 63UL))) {
268 		cookie = ~cookie;
269 		bucket = (struct ino_bucket *) __va(cookie);
270 		irq = bucket->__irq;
271 	}
272 out:
273 	return irq;
274 }
275 
276 static unsigned int sysino_exists(u32 devhandle, unsigned int devino)
277 {
278 	unsigned long sysino = sun4v_devino_to_sysino(devhandle, devino);
279 	struct ino_bucket *bucket;
280 	unsigned int irq;
281 
282 	bucket = &ivector_table[sysino];
283 	irq = bucket_get_irq(__pa(bucket));
284 
285 	return irq;
286 }
287 
288 void ack_bad_irq(unsigned int irq)
289 {
290 	pr_crit("BAD IRQ ack %d\n", irq);
291 }
292 
293 void irq_install_pre_handler(int irq,
294 			     void (*func)(unsigned int, void *, void *),
295 			     void *arg1, void *arg2)
296 {
297 	pr_warn("IRQ pre handler NOT supported.\n");
298 }
299 
300 /*
301  * /proc/interrupts printing:
302  */
303 int arch_show_interrupts(struct seq_file *p, int prec)
304 {
305 	int j;
306 
307 	seq_printf(p, "NMI: ");
308 	for_each_online_cpu(j)
309 		seq_printf(p, "%10u ", cpu_data(j).__nmi_count);
310 	seq_printf(p, "     Non-maskable interrupts\n");
311 	return 0;
312 }
313 
314 static unsigned int sun4u_compute_tid(unsigned long imap, unsigned long cpuid)
315 {
316 	unsigned int tid;
317 
318 	if (this_is_starfire) {
319 		tid = starfire_translate(imap, cpuid);
320 		tid <<= IMAP_TID_SHIFT;
321 		tid &= IMAP_TID_UPA;
322 	} else {
323 		if (tlb_type == cheetah || tlb_type == cheetah_plus) {
324 			unsigned long ver;
325 
326 			__asm__ ("rdpr %%ver, %0" : "=r" (ver));
327 			if ((ver >> 32UL) == __JALAPENO_ID ||
328 			    (ver >> 32UL) == __SERRANO_ID) {
329 				tid = cpuid << IMAP_TID_SHIFT;
330 				tid &= IMAP_TID_JBUS;
331 			} else {
332 				unsigned int a = cpuid & 0x1f;
333 				unsigned int n = (cpuid >> 5) & 0x1f;
334 
335 				tid = ((a << IMAP_AID_SHIFT) |
336 				       (n << IMAP_NID_SHIFT));
337 				tid &= (IMAP_AID_SAFARI |
338 					IMAP_NID_SAFARI);
339 			}
340 		} else {
341 			tid = cpuid << IMAP_TID_SHIFT;
342 			tid &= IMAP_TID_UPA;
343 		}
344 	}
345 
346 	return tid;
347 }
348 
349 #ifdef CONFIG_SMP
350 static int irq_choose_cpu(unsigned int irq, const struct cpumask *affinity)
351 {
352 	cpumask_t mask;
353 	int cpuid;
354 
355 	cpumask_copy(&mask, affinity);
356 	if (cpumask_equal(&mask, cpu_online_mask)) {
357 		cpuid = map_to_cpu(irq);
358 	} else {
359 		cpumask_t tmp;
360 
361 		cpumask_and(&tmp, cpu_online_mask, &mask);
362 		cpuid = cpumask_empty(&tmp) ? map_to_cpu(irq) : cpumask_first(&tmp);
363 	}
364 
365 	return cpuid;
366 }
367 #else
368 #define irq_choose_cpu(irq, affinity)	\
369 	real_hard_smp_processor_id()
370 #endif
371 
372 static void sun4u_irq_enable(struct irq_data *data)
373 {
374 	struct irq_handler_data *handler_data;
375 
376 	handler_data = irq_data_get_irq_handler_data(data);
377 	if (likely(handler_data)) {
378 		unsigned long cpuid, imap, val;
379 		unsigned int tid;
380 
381 		cpuid = irq_choose_cpu(data->irq,
382 				       irq_data_get_affinity_mask(data));
383 		imap = handler_data->imap;
384 
385 		tid = sun4u_compute_tid(imap, cpuid);
386 
387 		val = upa_readq(imap);
388 		val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
389 			 IMAP_AID_SAFARI | IMAP_NID_SAFARI);
390 		val |= tid | IMAP_VALID;
391 		upa_writeq(val, imap);
392 		upa_writeq(ICLR_IDLE, handler_data->iclr);
393 	}
394 }
395 
396 static int sun4u_set_affinity(struct irq_data *data,
397 			       const struct cpumask *mask, bool force)
398 {
399 	struct irq_handler_data *handler_data;
400 
401 	handler_data = irq_data_get_irq_handler_data(data);
402 	if (likely(handler_data)) {
403 		unsigned long cpuid, imap, val;
404 		unsigned int tid;
405 
406 		cpuid = irq_choose_cpu(data->irq, mask);
407 		imap = handler_data->imap;
408 
409 		tid = sun4u_compute_tid(imap, cpuid);
410 
411 		val = upa_readq(imap);
412 		val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
413 			 IMAP_AID_SAFARI | IMAP_NID_SAFARI);
414 		val |= tid | IMAP_VALID;
415 		upa_writeq(val, imap);
416 		upa_writeq(ICLR_IDLE, handler_data->iclr);
417 	}
418 
419 	return 0;
420 }
421 
422 /* Don't do anything.  The desc->status check for IRQ_DISABLED in
423  * handler_irq() will skip the handler call and that will leave the
424  * interrupt in the sent state.  The next ->enable() call will hit the
425  * ICLR register to reset the state machine.
426  *
427  * This scheme is necessary, instead of clearing the Valid bit in the
428  * IMAP register, to handle the case of IMAP registers being shared by
429  * multiple INOs (and thus ICLR registers).  Since we use a different
430  * virtual IRQ for each shared IMAP instance, the generic code thinks
431  * there is only one user so it prematurely calls ->disable() on
432  * free_irq().
433  *
434  * We have to provide an explicit ->disable() method instead of using
435  * NULL to get the default.  The reason is that if the generic code
436  * sees that, it also hooks up a default ->shutdown method which
437  * invokes ->mask() which we do not want.  See irq_chip_set_defaults().
438  */
439 static void sun4u_irq_disable(struct irq_data *data)
440 {
441 }
442 
443 static void sun4u_irq_eoi(struct irq_data *data)
444 {
445 	struct irq_handler_data *handler_data;
446 
447 	handler_data = irq_data_get_irq_handler_data(data);
448 	if (likely(handler_data))
449 		upa_writeq(ICLR_IDLE, handler_data->iclr);
450 }
451 
452 static void sun4v_irq_enable(struct irq_data *data)
453 {
454 	unsigned long cpuid = irq_choose_cpu(data->irq,
455 					     irq_data_get_affinity_mask(data));
456 	unsigned int ino = irq_data_to_sysino(data);
457 	int err;
458 
459 	err = sun4v_intr_settarget(ino, cpuid);
460 	if (err != HV_EOK)
461 		printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
462 		       "err(%d)\n", ino, cpuid, err);
463 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
464 	if (err != HV_EOK)
465 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
466 		       "err(%d)\n", ino, err);
467 	err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
468 	if (err != HV_EOK)
469 		printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
470 		       ino, err);
471 }
472 
473 static int sun4v_set_affinity(struct irq_data *data,
474 			       const struct cpumask *mask, bool force)
475 {
476 	unsigned long cpuid = irq_choose_cpu(data->irq, mask);
477 	unsigned int ino = irq_data_to_sysino(data);
478 	int err;
479 
480 	err = sun4v_intr_settarget(ino, cpuid);
481 	if (err != HV_EOK)
482 		printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
483 		       "err(%d)\n", ino, cpuid, err);
484 
485 	return 0;
486 }
487 
488 static void sun4v_irq_disable(struct irq_data *data)
489 {
490 	unsigned int ino = irq_data_to_sysino(data);
491 	int err;
492 
493 	err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
494 	if (err != HV_EOK)
495 		printk(KERN_ERR "sun4v_intr_setenabled(%x): "
496 		       "err(%d)\n", ino, err);
497 }
498 
499 static void sun4v_irq_eoi(struct irq_data *data)
500 {
501 	unsigned int ino = irq_data_to_sysino(data);
502 	int err;
503 
504 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
505 	if (err != HV_EOK)
506 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
507 		       "err(%d)\n", ino, err);
508 }
509 
510 static void sun4v_virq_enable(struct irq_data *data)
511 {
512 	unsigned long dev_handle = irq_data_to_handle(data);
513 	unsigned long dev_ino = irq_data_to_ino(data);
514 	unsigned long cpuid;
515 	int err;
516 
517 	cpuid = irq_choose_cpu(data->irq, irq_data_get_affinity_mask(data));
518 
519 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
520 	if (err != HV_EOK)
521 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
522 		       "err(%d)\n",
523 		       dev_handle, dev_ino, cpuid, err);
524 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
525 				    HV_INTR_STATE_IDLE);
526 	if (err != HV_EOK)
527 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
528 		       "HV_INTR_STATE_IDLE): err(%d)\n",
529 		       dev_handle, dev_ino, err);
530 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
531 				    HV_INTR_ENABLED);
532 	if (err != HV_EOK)
533 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
534 		       "HV_INTR_ENABLED): err(%d)\n",
535 		       dev_handle, dev_ino, err);
536 }
537 
538 static int sun4v_virt_set_affinity(struct irq_data *data,
539 				    const struct cpumask *mask, bool force)
540 {
541 	unsigned long dev_handle = irq_data_to_handle(data);
542 	unsigned long dev_ino = irq_data_to_ino(data);
543 	unsigned long cpuid;
544 	int err;
545 
546 	cpuid = irq_choose_cpu(data->irq, mask);
547 
548 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
549 	if (err != HV_EOK)
550 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
551 		       "err(%d)\n",
552 		       dev_handle, dev_ino, cpuid, err);
553 
554 	return 0;
555 }
556 
557 static void sun4v_virq_disable(struct irq_data *data)
558 {
559 	unsigned long dev_handle = irq_data_to_handle(data);
560 	unsigned long dev_ino = irq_data_to_ino(data);
561 	int err;
562 
563 
564 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
565 				    HV_INTR_DISABLED);
566 	if (err != HV_EOK)
567 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
568 		       "HV_INTR_DISABLED): err(%d)\n",
569 		       dev_handle, dev_ino, err);
570 }
571 
572 static void sun4v_virq_eoi(struct irq_data *data)
573 {
574 	unsigned long dev_handle = irq_data_to_handle(data);
575 	unsigned long dev_ino = irq_data_to_ino(data);
576 	int err;
577 
578 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
579 				    HV_INTR_STATE_IDLE);
580 	if (err != HV_EOK)
581 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
582 		       "HV_INTR_STATE_IDLE): err(%d)\n",
583 		       dev_handle, dev_ino, err);
584 }
585 
586 static struct irq_chip sun4u_irq = {
587 	.name			= "sun4u",
588 	.irq_enable		= sun4u_irq_enable,
589 	.irq_disable		= sun4u_irq_disable,
590 	.irq_eoi		= sun4u_irq_eoi,
591 	.irq_set_affinity	= sun4u_set_affinity,
592 	.flags			= IRQCHIP_EOI_IF_HANDLED,
593 };
594 
595 static struct irq_chip sun4v_irq = {
596 	.name			= "sun4v",
597 	.irq_enable		= sun4v_irq_enable,
598 	.irq_disable		= sun4v_irq_disable,
599 	.irq_eoi		= sun4v_irq_eoi,
600 	.irq_set_affinity	= sun4v_set_affinity,
601 	.flags			= IRQCHIP_EOI_IF_HANDLED,
602 };
603 
604 static struct irq_chip sun4v_virq = {
605 	.name			= "vsun4v",
606 	.irq_enable		= sun4v_virq_enable,
607 	.irq_disable		= sun4v_virq_disable,
608 	.irq_eoi		= sun4v_virq_eoi,
609 	.irq_set_affinity	= sun4v_virt_set_affinity,
610 	.flags			= IRQCHIP_EOI_IF_HANDLED,
611 };
612 
613 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
614 {
615 	struct irq_handler_data *handler_data;
616 	struct ino_bucket *bucket;
617 	unsigned int irq;
618 	int ino;
619 
620 	BUG_ON(tlb_type == hypervisor);
621 
622 	ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
623 	bucket = &ivector_table[ino];
624 	irq = bucket_get_irq(__pa(bucket));
625 	if (!irq) {
626 		irq = irq_alloc(0, ino);
627 		bucket_set_irq(__pa(bucket), irq);
628 		irq_set_chip_and_handler_name(irq, &sun4u_irq,
629 					      handle_fasteoi_irq, "IVEC");
630 	}
631 
632 	handler_data = irq_get_handler_data(irq);
633 	if (unlikely(handler_data))
634 		goto out;
635 
636 	handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
637 	if (unlikely(!handler_data)) {
638 		prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
639 		prom_halt();
640 	}
641 	irq_set_handler_data(irq, handler_data);
642 
643 	handler_data->imap  = imap;
644 	handler_data->iclr  = iclr;
645 
646 out:
647 	return irq;
648 }
649 
650 static unsigned int sun4v_build_common(u32 devhandle, unsigned int devino,
651 		void (*handler_data_init)(struct irq_handler_data *data,
652 		u32 devhandle, unsigned int devino),
653 		struct irq_chip *chip)
654 {
655 	struct irq_handler_data *data;
656 	unsigned int irq;
657 
658 	irq = irq_alloc(devhandle, devino);
659 	if (!irq)
660 		goto out;
661 
662 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
663 	if (unlikely(!data)) {
664 		pr_err("IRQ handler data allocation failed.\n");
665 		irq_free(irq);
666 		irq = 0;
667 		goto out;
668 	}
669 
670 	irq_set_handler_data(irq, data);
671 	handler_data_init(data, devhandle, devino);
672 	irq_set_chip_and_handler_name(irq, chip, handle_fasteoi_irq, "IVEC");
673 	data->imap = ~0UL;
674 	data->iclr = ~0UL;
675 out:
676 	return irq;
677 }
678 
679 static unsigned long cookie_assign(unsigned int irq, u32 devhandle,
680 		unsigned int devino)
681 {
682 	struct irq_handler_data *ihd = irq_get_handler_data(irq);
683 	unsigned long hv_error, cookie;
684 
685 	/* handler_irq needs to find the irq. cookie is seen signed in
686 	 * sun4v_dev_mondo and treated as a non ivector_table delivery.
687 	 */
688 	ihd->bucket.__irq = irq;
689 	cookie = ~__pa(&ihd->bucket);
690 
691 	hv_error = sun4v_vintr_set_cookie(devhandle, devino, cookie);
692 	if (hv_error)
693 		pr_err("HV vintr set cookie failed = %ld\n", hv_error);
694 
695 	return hv_error;
696 }
697 
698 static void cookie_handler_data(struct irq_handler_data *data,
699 				u32 devhandle, unsigned int devino)
700 {
701 	data->dev_handle = devhandle;
702 	data->dev_ino = devino;
703 }
704 
705 static unsigned int cookie_build_irq(u32 devhandle, unsigned int devino,
706 				     struct irq_chip *chip)
707 {
708 	unsigned long hv_error;
709 	unsigned int irq;
710 
711 	irq = sun4v_build_common(devhandle, devino, cookie_handler_data, chip);
712 
713 	hv_error = cookie_assign(irq, devhandle, devino);
714 	if (hv_error) {
715 		irq_free(irq);
716 		irq = 0;
717 	}
718 
719 	return irq;
720 }
721 
722 static unsigned int sun4v_build_cookie(u32 devhandle, unsigned int devino)
723 {
724 	unsigned int irq;
725 
726 	irq = cookie_exists(devhandle, devino);
727 	if (irq)
728 		goto out;
729 
730 	irq = cookie_build_irq(devhandle, devino, &sun4v_virq);
731 
732 out:
733 	return irq;
734 }
735 
736 static void sysino_set_bucket(unsigned int irq)
737 {
738 	struct irq_handler_data *ihd = irq_get_handler_data(irq);
739 	struct ino_bucket *bucket;
740 	unsigned long sysino;
741 
742 	sysino = sun4v_devino_to_sysino(ihd->dev_handle, ihd->dev_ino);
743 	BUG_ON(sysino >= nr_ivec);
744 	bucket = &ivector_table[sysino];
745 	bucket_set_irq(__pa(bucket), irq);
746 }
747 
748 static void sysino_handler_data(struct irq_handler_data *data,
749 				u32 devhandle, unsigned int devino)
750 {
751 	unsigned long sysino;
752 
753 	sysino = sun4v_devino_to_sysino(devhandle, devino);
754 	data->sysino = sysino;
755 }
756 
757 static unsigned int sysino_build_irq(u32 devhandle, unsigned int devino,
758 				     struct irq_chip *chip)
759 {
760 	unsigned int irq;
761 
762 	irq = sun4v_build_common(devhandle, devino, sysino_handler_data, chip);
763 	if (!irq)
764 		goto out;
765 
766 	sysino_set_bucket(irq);
767 out:
768 	return irq;
769 }
770 
771 static int sun4v_build_sysino(u32 devhandle, unsigned int devino)
772 {
773 	int irq;
774 
775 	irq = sysino_exists(devhandle, devino);
776 	if (irq)
777 		goto out;
778 
779 	irq = sysino_build_irq(devhandle, devino, &sun4v_irq);
780 out:
781 	return irq;
782 }
783 
784 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
785 {
786 	unsigned int irq;
787 
788 	if (sun4v_cookie_only_virqs())
789 		irq = sun4v_build_cookie(devhandle, devino);
790 	else
791 		irq = sun4v_build_sysino(devhandle, devino);
792 
793 	return irq;
794 }
795 
796 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
797 {
798 	int irq;
799 
800 	irq = cookie_build_irq(devhandle, devino, &sun4v_virq);
801 	if (!irq)
802 		goto out;
803 
804 	/* This is borrowed from the original function.
805 	 */
806 	irq_set_status_flags(irq, IRQ_NOAUTOEN);
807 
808 out:
809 	return irq;
810 }
811 
812 void *hardirq_stack[NR_CPUS];
813 void *softirq_stack[NR_CPUS];
814 
815 void __irq_entry handler_irq(int pil, struct pt_regs *regs)
816 {
817 	unsigned long pstate, bucket_pa;
818 	struct pt_regs *old_regs;
819 	void *orig_sp;
820 
821 	clear_softint(1 << pil);
822 
823 	old_regs = set_irq_regs(regs);
824 	irq_enter();
825 
826 	/* Grab an atomic snapshot of the pending IVECs.  */
827 	__asm__ __volatile__("rdpr	%%pstate, %0\n\t"
828 			     "wrpr	%0, %3, %%pstate\n\t"
829 			     "ldx	[%2], %1\n\t"
830 			     "stx	%%g0, [%2]\n\t"
831 			     "wrpr	%0, 0x0, %%pstate\n\t"
832 			     : "=&r" (pstate), "=&r" (bucket_pa)
833 			     : "r" (irq_work_pa(smp_processor_id())),
834 			       "i" (PSTATE_IE)
835 			     : "memory");
836 
837 	orig_sp = set_hardirq_stack();
838 
839 	while (bucket_pa) {
840 		unsigned long next_pa;
841 		unsigned int irq;
842 
843 		next_pa = bucket_get_chain_pa(bucket_pa);
844 		irq = bucket_get_irq(bucket_pa);
845 		bucket_clear_chain_pa(bucket_pa);
846 
847 		generic_handle_irq(irq);
848 
849 		bucket_pa = next_pa;
850 	}
851 
852 	restore_hardirq_stack(orig_sp);
853 
854 	irq_exit();
855 	set_irq_regs(old_regs);
856 }
857 
858 void do_softirq_own_stack(void)
859 {
860 	void *orig_sp, *sp = softirq_stack[smp_processor_id()];
861 
862 	sp += THREAD_SIZE - 192 - STACK_BIAS;
863 
864 	__asm__ __volatile__("mov %%sp, %0\n\t"
865 			     "mov %1, %%sp"
866 			     : "=&r" (orig_sp)
867 			     : "r" (sp));
868 	__do_softirq();
869 	__asm__ __volatile__("mov %0, %%sp"
870 			     : : "r" (orig_sp));
871 }
872 
873 #ifdef CONFIG_HOTPLUG_CPU
874 void fixup_irqs(void)
875 {
876 	unsigned int irq;
877 
878 	for (irq = 0; irq < NR_IRQS; irq++) {
879 		struct irq_desc *desc = irq_to_desc(irq);
880 		struct irq_data *data;
881 		unsigned long flags;
882 
883 		if (!desc)
884 			continue;
885 		data = irq_desc_get_irq_data(desc);
886 		raw_spin_lock_irqsave(&desc->lock, flags);
887 		if (desc->action && !irqd_is_per_cpu(data)) {
888 			if (data->chip->irq_set_affinity)
889 				data->chip->irq_set_affinity(data,
890 					irq_data_get_affinity_mask(data),
891 					false);
892 		}
893 		raw_spin_unlock_irqrestore(&desc->lock, flags);
894 	}
895 
896 	tick_ops->disable_irq();
897 }
898 #endif
899 
900 struct sun5_timer {
901 	u64	count0;
902 	u64	limit0;
903 	u64	count1;
904 	u64	limit1;
905 };
906 
907 static struct sun5_timer *prom_timers;
908 static u64 prom_limit0, prom_limit1;
909 
910 static void map_prom_timers(void)
911 {
912 	struct device_node *dp;
913 	const unsigned int *addr;
914 
915 	/* PROM timer node hangs out in the top level of device siblings... */
916 	dp = of_find_node_by_path("/");
917 	dp = dp->child;
918 	while (dp) {
919 		if (of_node_name_eq(dp, "counter-timer"))
920 			break;
921 		dp = dp->sibling;
922 	}
923 
924 	/* Assume if node is not present, PROM uses different tick mechanism
925 	 * which we should not care about.
926 	 */
927 	if (!dp) {
928 		prom_timers = (struct sun5_timer *) 0;
929 		return;
930 	}
931 
932 	/* If PROM is really using this, it must be mapped by him. */
933 	addr = of_get_property(dp, "address", NULL);
934 	if (!addr) {
935 		prom_printf("PROM does not have timer mapped, trying to continue.\n");
936 		prom_timers = (struct sun5_timer *) 0;
937 		return;
938 	}
939 	prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
940 }
941 
942 static void kill_prom_timer(void)
943 {
944 	if (!prom_timers)
945 		return;
946 
947 	/* Save them away for later. */
948 	prom_limit0 = prom_timers->limit0;
949 	prom_limit1 = prom_timers->limit1;
950 
951 	/* Just as in sun4c PROM uses timer which ticks at IRQ 14.
952 	 * We turn both off here just to be paranoid.
953 	 */
954 	prom_timers->limit0 = 0;
955 	prom_timers->limit1 = 0;
956 
957 	/* Wheee, eat the interrupt packet too... */
958 	__asm__ __volatile__(
959 "	mov	0x40, %%g2\n"
960 "	ldxa	[%%g0] %0, %%g1\n"
961 "	ldxa	[%%g2] %1, %%g1\n"
962 "	stxa	%%g0, [%%g0] %0\n"
963 "	membar	#Sync\n"
964 	: /* no outputs */
965 	: "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
966 	: "g1", "g2");
967 }
968 
969 void notrace init_irqwork_curcpu(void)
970 {
971 	int cpu = hard_smp_processor_id();
972 
973 	trap_block[cpu].irq_worklist_pa = 0UL;
974 }
975 
976 /* Please be very careful with register_one_mondo() and
977  * sun4v_register_mondo_queues().
978  *
979  * On SMP this gets invoked from the CPU trampoline before
980  * the cpu has fully taken over the trap table from OBP,
981  * and it's kernel stack + %g6 thread register state is
982  * not fully cooked yet.
983  *
984  * Therefore you cannot make any OBP calls, not even prom_printf,
985  * from these two routines.
986  */
987 static void notrace register_one_mondo(unsigned long paddr, unsigned long type,
988 				       unsigned long qmask)
989 {
990 	unsigned long num_entries = (qmask + 1) / 64;
991 	unsigned long status;
992 
993 	status = sun4v_cpu_qconf(type, paddr, num_entries);
994 	if (status != HV_EOK) {
995 		prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
996 			    "err %lu\n", type, paddr, num_entries, status);
997 		prom_halt();
998 	}
999 }
1000 
1001 void notrace sun4v_register_mondo_queues(int this_cpu)
1002 {
1003 	struct trap_per_cpu *tb = &trap_block[this_cpu];
1004 
1005 	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
1006 			   tb->cpu_mondo_qmask);
1007 	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
1008 			   tb->dev_mondo_qmask);
1009 	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
1010 			   tb->resum_qmask);
1011 	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
1012 			   tb->nonresum_qmask);
1013 }
1014 
1015 /* Each queue region must be a power of 2 multiple of 64 bytes in
1016  * size.  The base real address must be aligned to the size of the
1017  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
1018  */
1019 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
1020 {
1021 	unsigned long size = PAGE_ALIGN(qmask + 1);
1022 	unsigned long order = get_order(size);
1023 	unsigned long p;
1024 
1025 	p = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
1026 	if (!p) {
1027 		prom_printf("SUN4V: Error, cannot allocate queue.\n");
1028 		prom_halt();
1029 	}
1030 
1031 	*pa_ptr = __pa(p);
1032 }
1033 
1034 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
1035 {
1036 #ifdef CONFIG_SMP
1037 	unsigned long page;
1038 	void *mondo, *p;
1039 
1040 	BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > PAGE_SIZE);
1041 
1042 	/* Make sure mondo block is 64byte aligned */
1043 	p = kzalloc(127, GFP_KERNEL);
1044 	if (!p) {
1045 		prom_printf("SUN4V: Error, cannot allocate mondo block.\n");
1046 		prom_halt();
1047 	}
1048 	mondo = (void *)(((unsigned long)p + 63) & ~0x3f);
1049 	tb->cpu_mondo_block_pa = __pa(mondo);
1050 
1051 	page = get_zeroed_page(GFP_KERNEL);
1052 	if (!page) {
1053 		prom_printf("SUN4V: Error, cannot allocate cpu list page.\n");
1054 		prom_halt();
1055 	}
1056 
1057 	tb->cpu_list_pa = __pa(page);
1058 #endif
1059 }
1060 
1061 /* Allocate mondo and error queues for all possible cpus.  */
1062 static void __init sun4v_init_mondo_queues(void)
1063 {
1064 	int cpu;
1065 
1066 	for_each_possible_cpu(cpu) {
1067 		struct trap_per_cpu *tb = &trap_block[cpu];
1068 
1069 		alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
1070 		alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
1071 		alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
1072 		alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
1073 		alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
1074 		alloc_one_queue(&tb->nonresum_kernel_buf_pa,
1075 				tb->nonresum_qmask);
1076 	}
1077 }
1078 
1079 static void __init init_send_mondo_info(void)
1080 {
1081 	int cpu;
1082 
1083 	for_each_possible_cpu(cpu) {
1084 		struct trap_per_cpu *tb = &trap_block[cpu];
1085 
1086 		init_cpu_send_mondo_info(tb);
1087 	}
1088 }
1089 
1090 static struct irqaction timer_irq_action = {
1091 	.name = "timer",
1092 };
1093 
1094 static void __init irq_ivector_init(void)
1095 {
1096 	unsigned long size, order;
1097 	unsigned int ivecs;
1098 
1099 	/* If we are doing cookie only VIRQs then we do not need the ivector
1100 	 * table to process interrupts.
1101 	 */
1102 	if (sun4v_cookie_only_virqs())
1103 		return;
1104 
1105 	ivecs = size_nr_ivec();
1106 	size = sizeof(struct ino_bucket) * ivecs;
1107 	order = get_order(size);
1108 	ivector_table = (struct ino_bucket *)
1109 		__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
1110 	if (!ivector_table) {
1111 		prom_printf("Fatal error, cannot allocate ivector_table\n");
1112 		prom_halt();
1113 	}
1114 	__flush_dcache_range((unsigned long) ivector_table,
1115 			     ((unsigned long) ivector_table) + size);
1116 
1117 	ivector_table_pa = __pa(ivector_table);
1118 }
1119 
1120 /* Only invoked on boot processor.*/
1121 void __init init_IRQ(void)
1122 {
1123 	irq_init_hv();
1124 	irq_ivector_init();
1125 	map_prom_timers();
1126 	kill_prom_timer();
1127 
1128 	if (tlb_type == hypervisor)
1129 		sun4v_init_mondo_queues();
1130 
1131 	init_send_mondo_info();
1132 
1133 	if (tlb_type == hypervisor) {
1134 		/* Load up the boot cpu's entries.  */
1135 		sun4v_register_mondo_queues(hard_smp_processor_id());
1136 	}
1137 
1138 	/* We need to clear any IRQ's pending in the soft interrupt
1139 	 * registers, a spurious one could be left around from the
1140 	 * PROM timer which we just disabled.
1141 	 */
1142 	clear_softint(get_softint());
1143 
1144 	/* Now that ivector table is initialized, it is safe
1145 	 * to receive IRQ vector traps.  We will normally take
1146 	 * one or two right now, in case some device PROM used
1147 	 * to boot us wants to speak to us.  We just ignore them.
1148 	 */
1149 	__asm__ __volatile__("rdpr	%%pstate, %%g1\n\t"
1150 			     "or	%%g1, %0, %%g1\n\t"
1151 			     "wrpr	%%g1, 0x0, %%pstate"
1152 			     : /* No outputs */
1153 			     : "i" (PSTATE_IE)
1154 			     : "g1");
1155 
1156 	irq_to_desc(0)->action = &timer_irq_action;
1157 }
1158