xref: /linux/arch/sparc/kernel/irq_64.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
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 	int cpuid;
353 
354 	if (cpumask_equal(affinity, cpu_online_mask)) {
355 		cpuid = map_to_cpu(irq);
356 	} else {
357 		cpuid = cpumask_first_and(affinity, cpu_online_mask);
358 		cpuid = cpuid < nr_cpu_ids ? cpuid : map_to_cpu(irq);
359 	}
360 
361 	return cpuid;
362 }
363 #else
364 #define irq_choose_cpu(irq, affinity)	\
365 	real_hard_smp_processor_id()
366 #endif
367 
368 static void sun4u_irq_enable(struct irq_data *data)
369 {
370 	struct irq_handler_data *handler_data;
371 
372 	handler_data = irq_data_get_irq_handler_data(data);
373 	if (likely(handler_data)) {
374 		unsigned long cpuid, imap, val;
375 		unsigned int tid;
376 
377 		cpuid = irq_choose_cpu(data->irq,
378 				       irq_data_get_affinity_mask(data));
379 		imap = handler_data->imap;
380 
381 		tid = sun4u_compute_tid(imap, cpuid);
382 
383 		val = upa_readq(imap);
384 		val &= ~(IMAP_TID_UPA | IMAP_TID_JBUS |
385 			 IMAP_AID_SAFARI | IMAP_NID_SAFARI);
386 		val |= tid | IMAP_VALID;
387 		upa_writeq(val, imap);
388 		upa_writeq(ICLR_IDLE, handler_data->iclr);
389 	}
390 }
391 
392 static int sun4u_set_affinity(struct irq_data *data,
393 			       const struct cpumask *mask, bool force)
394 {
395 	struct irq_handler_data *handler_data;
396 
397 	handler_data = irq_data_get_irq_handler_data(data);
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;
442 
443 	handler_data = irq_data_get_irq_handler_data(data);
444 	if (likely(handler_data))
445 		upa_writeq(ICLR_IDLE, handler_data->iclr);
446 }
447 
448 static void sun4v_irq_enable(struct irq_data *data)
449 {
450 	unsigned long cpuid = irq_choose_cpu(data->irq,
451 					     irq_data_get_affinity_mask(data));
452 	unsigned int ino = irq_data_to_sysino(data);
453 	int err;
454 
455 	err = sun4v_intr_settarget(ino, cpuid);
456 	if (err != HV_EOK)
457 		printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
458 		       "err(%d)\n", ino, cpuid, err);
459 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
460 	if (err != HV_EOK)
461 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
462 		       "err(%d)\n", ino, err);
463 	err = sun4v_intr_setenabled(ino, HV_INTR_ENABLED);
464 	if (err != HV_EOK)
465 		printk(KERN_ERR "sun4v_intr_setenabled(%x): err(%d)\n",
466 		       ino, err);
467 }
468 
469 static int sun4v_set_affinity(struct irq_data *data,
470 			       const struct cpumask *mask, bool force)
471 {
472 	unsigned long cpuid = irq_choose_cpu(data->irq, mask);
473 	unsigned int ino = irq_data_to_sysino(data);
474 	int err;
475 
476 	err = sun4v_intr_settarget(ino, cpuid);
477 	if (err != HV_EOK)
478 		printk(KERN_ERR "sun4v_intr_settarget(%x,%lu): "
479 		       "err(%d)\n", ino, cpuid, err);
480 
481 	return 0;
482 }
483 
484 static void sun4v_irq_disable(struct irq_data *data)
485 {
486 	unsigned int ino = irq_data_to_sysino(data);
487 	int err;
488 
489 	err = sun4v_intr_setenabled(ino, HV_INTR_DISABLED);
490 	if (err != HV_EOK)
491 		printk(KERN_ERR "sun4v_intr_setenabled(%x): "
492 		       "err(%d)\n", ino, err);
493 }
494 
495 static void sun4v_irq_eoi(struct irq_data *data)
496 {
497 	unsigned int ino = irq_data_to_sysino(data);
498 	int err;
499 
500 	err = sun4v_intr_setstate(ino, HV_INTR_STATE_IDLE);
501 	if (err != HV_EOK)
502 		printk(KERN_ERR "sun4v_intr_setstate(%x): "
503 		       "err(%d)\n", ino, err);
504 }
505 
506 static void sun4v_virq_enable(struct irq_data *data)
507 {
508 	unsigned long dev_handle = irq_data_to_handle(data);
509 	unsigned long dev_ino = irq_data_to_ino(data);
510 	unsigned long cpuid;
511 	int err;
512 
513 	cpuid = irq_choose_cpu(data->irq, irq_data_get_affinity_mask(data));
514 
515 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
516 	if (err != HV_EOK)
517 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
518 		       "err(%d)\n",
519 		       dev_handle, dev_ino, cpuid, err);
520 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
521 				    HV_INTR_STATE_IDLE);
522 	if (err != HV_EOK)
523 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
524 		       "HV_INTR_STATE_IDLE): err(%d)\n",
525 		       dev_handle, dev_ino, err);
526 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
527 				    HV_INTR_ENABLED);
528 	if (err != HV_EOK)
529 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
530 		       "HV_INTR_ENABLED): err(%d)\n",
531 		       dev_handle, dev_ino, err);
532 }
533 
534 static int sun4v_virt_set_affinity(struct irq_data *data,
535 				    const struct cpumask *mask, bool force)
536 {
537 	unsigned long dev_handle = irq_data_to_handle(data);
538 	unsigned long dev_ino = irq_data_to_ino(data);
539 	unsigned long cpuid;
540 	int err;
541 
542 	cpuid = irq_choose_cpu(data->irq, mask);
543 
544 	err = sun4v_vintr_set_target(dev_handle, dev_ino, cpuid);
545 	if (err != HV_EOK)
546 		printk(KERN_ERR "sun4v_vintr_set_target(%lx,%lx,%lu): "
547 		       "err(%d)\n",
548 		       dev_handle, dev_ino, cpuid, err);
549 
550 	return 0;
551 }
552 
553 static void sun4v_virq_disable(struct irq_data *data)
554 {
555 	unsigned long dev_handle = irq_data_to_handle(data);
556 	unsigned long dev_ino = irq_data_to_ino(data);
557 	int err;
558 
559 
560 	err = sun4v_vintr_set_valid(dev_handle, dev_ino,
561 				    HV_INTR_DISABLED);
562 	if (err != HV_EOK)
563 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
564 		       "HV_INTR_DISABLED): err(%d)\n",
565 		       dev_handle, dev_ino, err);
566 }
567 
568 static void sun4v_virq_eoi(struct irq_data *data)
569 {
570 	unsigned long dev_handle = irq_data_to_handle(data);
571 	unsigned long dev_ino = irq_data_to_ino(data);
572 	int err;
573 
574 	err = sun4v_vintr_set_state(dev_handle, dev_ino,
575 				    HV_INTR_STATE_IDLE);
576 	if (err != HV_EOK)
577 		printk(KERN_ERR "sun4v_vintr_set_state(%lx,%lx,"
578 		       "HV_INTR_STATE_IDLE): err(%d)\n",
579 		       dev_handle, dev_ino, err);
580 }
581 
582 static struct irq_chip sun4u_irq = {
583 	.name			= "sun4u",
584 	.irq_enable		= sun4u_irq_enable,
585 	.irq_disable		= sun4u_irq_disable,
586 	.irq_eoi		= sun4u_irq_eoi,
587 	.irq_set_affinity	= sun4u_set_affinity,
588 	.flags			= IRQCHIP_EOI_IF_HANDLED,
589 };
590 
591 static struct irq_chip sun4v_irq = {
592 	.name			= "sun4v",
593 	.irq_enable		= sun4v_irq_enable,
594 	.irq_disable		= sun4v_irq_disable,
595 	.irq_eoi		= sun4v_irq_eoi,
596 	.irq_set_affinity	= sun4v_set_affinity,
597 	.flags			= IRQCHIP_EOI_IF_HANDLED,
598 };
599 
600 static struct irq_chip sun4v_virq = {
601 	.name			= "vsun4v",
602 	.irq_enable		= sun4v_virq_enable,
603 	.irq_disable		= sun4v_virq_disable,
604 	.irq_eoi		= sun4v_virq_eoi,
605 	.irq_set_affinity	= sun4v_virt_set_affinity,
606 	.flags			= IRQCHIP_EOI_IF_HANDLED,
607 };
608 
609 unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
610 {
611 	struct irq_handler_data *handler_data;
612 	struct ino_bucket *bucket;
613 	unsigned int irq;
614 	int ino;
615 
616 	BUG_ON(tlb_type == hypervisor);
617 
618 	ino = (upa_readq(imap) & (IMAP_IGN | IMAP_INO)) + inofixup;
619 	bucket = &ivector_table[ino];
620 	irq = bucket_get_irq(__pa(bucket));
621 	if (!irq) {
622 		irq = irq_alloc(0, ino);
623 		bucket_set_irq(__pa(bucket), irq);
624 		irq_set_chip_and_handler_name(irq, &sun4u_irq,
625 					      handle_fasteoi_irq, "IVEC");
626 	}
627 
628 	handler_data = irq_get_handler_data(irq);
629 	if (unlikely(handler_data))
630 		goto out;
631 
632 	handler_data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
633 	if (unlikely(!handler_data)) {
634 		prom_printf("IRQ: kzalloc(irq_handler_data) failed.\n");
635 		prom_halt();
636 	}
637 	irq_set_handler_data(irq, handler_data);
638 
639 	handler_data->imap  = imap;
640 	handler_data->iclr  = iclr;
641 
642 out:
643 	return irq;
644 }
645 
646 static unsigned int sun4v_build_common(u32 devhandle, unsigned int devino,
647 		void (*handler_data_init)(struct irq_handler_data *data,
648 		u32 devhandle, unsigned int devino),
649 		struct irq_chip *chip)
650 {
651 	struct irq_handler_data *data;
652 	unsigned int irq;
653 
654 	irq = irq_alloc(devhandle, devino);
655 	if (!irq)
656 		goto out;
657 
658 	data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
659 	if (unlikely(!data)) {
660 		pr_err("IRQ handler data allocation failed.\n");
661 		irq_free(irq);
662 		irq = 0;
663 		goto out;
664 	}
665 
666 	irq_set_handler_data(irq, data);
667 	handler_data_init(data, devhandle, devino);
668 	irq_set_chip_and_handler_name(irq, chip, handle_fasteoi_irq, "IVEC");
669 	data->imap = ~0UL;
670 	data->iclr = ~0UL;
671 out:
672 	return irq;
673 }
674 
675 static unsigned long cookie_assign(unsigned int irq, u32 devhandle,
676 		unsigned int devino)
677 {
678 	struct irq_handler_data *ihd = irq_get_handler_data(irq);
679 	unsigned long hv_error, cookie;
680 
681 	/* handler_irq needs to find the irq. cookie is seen signed in
682 	 * sun4v_dev_mondo and treated as a non ivector_table delivery.
683 	 */
684 	ihd->bucket.__irq = irq;
685 	cookie = ~__pa(&ihd->bucket);
686 
687 	hv_error = sun4v_vintr_set_cookie(devhandle, devino, cookie);
688 	if (hv_error)
689 		pr_err("HV vintr set cookie failed = %ld\n", hv_error);
690 
691 	return hv_error;
692 }
693 
694 static void cookie_handler_data(struct irq_handler_data *data,
695 				u32 devhandle, unsigned int devino)
696 {
697 	data->dev_handle = devhandle;
698 	data->dev_ino = devino;
699 }
700 
701 static unsigned int cookie_build_irq(u32 devhandle, unsigned int devino,
702 				     struct irq_chip *chip)
703 {
704 	unsigned long hv_error;
705 	unsigned int irq;
706 
707 	irq = sun4v_build_common(devhandle, devino, cookie_handler_data, chip);
708 
709 	hv_error = cookie_assign(irq, devhandle, devino);
710 	if (hv_error) {
711 		irq_free(irq);
712 		irq = 0;
713 	}
714 
715 	return irq;
716 }
717 
718 static unsigned int sun4v_build_cookie(u32 devhandle, unsigned int devino)
719 {
720 	unsigned int irq;
721 
722 	irq = cookie_exists(devhandle, devino);
723 	if (irq)
724 		goto out;
725 
726 	irq = cookie_build_irq(devhandle, devino, &sun4v_virq);
727 
728 out:
729 	return irq;
730 }
731 
732 static void sysino_set_bucket(unsigned int irq)
733 {
734 	struct irq_handler_data *ihd = irq_get_handler_data(irq);
735 	struct ino_bucket *bucket;
736 	unsigned long sysino;
737 
738 	sysino = sun4v_devino_to_sysino(ihd->dev_handle, ihd->dev_ino);
739 	BUG_ON(sysino >= nr_ivec);
740 	bucket = &ivector_table[sysino];
741 	bucket_set_irq(__pa(bucket), irq);
742 }
743 
744 static void sysino_handler_data(struct irq_handler_data *data,
745 				u32 devhandle, unsigned int devino)
746 {
747 	unsigned long sysino;
748 
749 	sysino = sun4v_devino_to_sysino(devhandle, devino);
750 	data->sysino = sysino;
751 }
752 
753 static unsigned int sysino_build_irq(u32 devhandle, unsigned int devino,
754 				     struct irq_chip *chip)
755 {
756 	unsigned int irq;
757 
758 	irq = sun4v_build_common(devhandle, devino, sysino_handler_data, chip);
759 	if (!irq)
760 		goto out;
761 
762 	sysino_set_bucket(irq);
763 out:
764 	return irq;
765 }
766 
767 static int sun4v_build_sysino(u32 devhandle, unsigned int devino)
768 {
769 	int irq;
770 
771 	irq = sysino_exists(devhandle, devino);
772 	if (irq)
773 		goto out;
774 
775 	irq = sysino_build_irq(devhandle, devino, &sun4v_irq);
776 out:
777 	return irq;
778 }
779 
780 unsigned int sun4v_build_irq(u32 devhandle, unsigned int devino)
781 {
782 	unsigned int irq;
783 
784 	if (sun4v_cookie_only_virqs())
785 		irq = sun4v_build_cookie(devhandle, devino);
786 	else
787 		irq = sun4v_build_sysino(devhandle, devino);
788 
789 	return irq;
790 }
791 
792 unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
793 {
794 	int irq;
795 
796 	irq = cookie_build_irq(devhandle, devino, &sun4v_virq);
797 	if (!irq)
798 		goto out;
799 
800 	/* This is borrowed from the original function.
801 	 */
802 	irq_set_status_flags(irq, IRQ_NOAUTOEN);
803 
804 out:
805 	return irq;
806 }
807 
808 void *hardirq_stack[NR_CPUS];
809 void *softirq_stack[NR_CPUS];
810 
811 void __irq_entry handler_irq(int pil, struct pt_regs *regs)
812 {
813 	unsigned long pstate, bucket_pa;
814 	struct pt_regs *old_regs;
815 	void *orig_sp;
816 
817 	clear_softint(1 << pil);
818 
819 	old_regs = set_irq_regs(regs);
820 	irq_enter();
821 
822 	/* Grab an atomic snapshot of the pending IVECs.  */
823 	__asm__ __volatile__("rdpr	%%pstate, %0\n\t"
824 			     "wrpr	%0, %3, %%pstate\n\t"
825 			     "ldx	[%2], %1\n\t"
826 			     "stx	%%g0, [%2]\n\t"
827 			     "wrpr	%0, 0x0, %%pstate\n\t"
828 			     : "=&r" (pstate), "=&r" (bucket_pa)
829 			     : "r" (irq_work_pa(smp_processor_id())),
830 			       "i" (PSTATE_IE)
831 			     : "memory");
832 
833 	orig_sp = set_hardirq_stack();
834 
835 	while (bucket_pa) {
836 		unsigned long next_pa;
837 		unsigned int irq;
838 
839 		next_pa = bucket_get_chain_pa(bucket_pa);
840 		irq = bucket_get_irq(bucket_pa);
841 		bucket_clear_chain_pa(bucket_pa);
842 
843 		generic_handle_irq(irq);
844 
845 		bucket_pa = next_pa;
846 	}
847 
848 	restore_hardirq_stack(orig_sp);
849 
850 	irq_exit();
851 	set_irq_regs(old_regs);
852 }
853 
854 #ifdef CONFIG_SOFTIRQ_ON_OWN_STACK
855 void do_softirq_own_stack(void)
856 {
857 	void *orig_sp, *sp = softirq_stack[smp_processor_id()];
858 
859 	sp += THREAD_SIZE - 192 - STACK_BIAS;
860 
861 	__asm__ __volatile__("mov %%sp, %0\n\t"
862 			     "mov %1, %%sp"
863 			     : "=&r" (orig_sp)
864 			     : "r" (sp));
865 	__do_softirq();
866 	__asm__ __volatile__("mov %0, %%sp"
867 			     : : "r" (orig_sp));
868 }
869 #endif
870 
871 #ifdef CONFIG_HOTPLUG_CPU
872 void fixup_irqs(void)
873 {
874 	unsigned int irq;
875 
876 	for (irq = 0; irq < NR_IRQS; irq++) {
877 		struct irq_desc *desc = irq_to_desc(irq);
878 		struct irq_data *data;
879 		unsigned long flags;
880 
881 		if (!desc)
882 			continue;
883 		data = irq_desc_get_irq_data(desc);
884 		raw_spin_lock_irqsave(&desc->lock, flags);
885 		if (desc->action && !irqd_is_per_cpu(data)) {
886 			if (data->chip->irq_set_affinity)
887 				data->chip->irq_set_affinity(data,
888 					irq_data_get_affinity_mask(data),
889 					false);
890 		}
891 		raw_spin_unlock_irqrestore(&desc->lock, flags);
892 	}
893 
894 	tick_ops->disable_irq();
895 }
896 #endif
897 
898 struct sun5_timer {
899 	u64	count0;
900 	u64	limit0;
901 	u64	count1;
902 	u64	limit1;
903 };
904 
905 static struct sun5_timer *prom_timers;
906 static u64 prom_limit0, prom_limit1;
907 
908 static void map_prom_timers(void)
909 {
910 	struct device_node *dp;
911 	const unsigned int *addr;
912 
913 	/* PROM timer node hangs out in the top level of device siblings... */
914 	dp = of_find_node_by_path("/");
915 	dp = dp->child;
916 	while (dp) {
917 		if (of_node_name_eq(dp, "counter-timer"))
918 			break;
919 		dp = dp->sibling;
920 	}
921 
922 	/* Assume if node is not present, PROM uses different tick mechanism
923 	 * which we should not care about.
924 	 */
925 	if (!dp) {
926 		prom_timers = (struct sun5_timer *) 0;
927 		return;
928 	}
929 
930 	/* If PROM is really using this, it must be mapped by him. */
931 	addr = of_get_property(dp, "address", NULL);
932 	if (!addr) {
933 		prom_printf("PROM does not have timer mapped, trying to continue.\n");
934 		prom_timers = (struct sun5_timer *) 0;
935 		return;
936 	}
937 	prom_timers = (struct sun5_timer *) ((unsigned long)addr[0]);
938 }
939 
940 static void kill_prom_timer(void)
941 {
942 	if (!prom_timers)
943 		return;
944 
945 	/* Save them away for later. */
946 	prom_limit0 = prom_timers->limit0;
947 	prom_limit1 = prom_timers->limit1;
948 
949 	/* Just as in sun4c PROM uses timer which ticks at IRQ 14.
950 	 * We turn both off here just to be paranoid.
951 	 */
952 	prom_timers->limit0 = 0;
953 	prom_timers->limit1 = 0;
954 
955 	/* Wheee, eat the interrupt packet too... */
956 	__asm__ __volatile__(
957 "	mov	0x40, %%g2\n"
958 "	ldxa	[%%g0] %0, %%g1\n"
959 "	ldxa	[%%g2] %1, %%g1\n"
960 "	stxa	%%g0, [%%g0] %0\n"
961 "	membar	#Sync\n"
962 	: /* no outputs */
963 	: "i" (ASI_INTR_RECEIVE), "i" (ASI_INTR_R)
964 	: "g1", "g2");
965 }
966 
967 void notrace init_irqwork_curcpu(void)
968 {
969 	int cpu = hard_smp_processor_id();
970 
971 	trap_block[cpu].irq_worklist_pa = 0UL;
972 }
973 
974 /* Please be very careful with register_one_mondo() and
975  * sun4v_register_mondo_queues().
976  *
977  * On SMP this gets invoked from the CPU trampoline before
978  * the cpu has fully taken over the trap table from OBP,
979  * and its kernel stack + %g6 thread register state is
980  * not fully cooked yet.
981  *
982  * Therefore you cannot make any OBP calls, not even prom_printf,
983  * from these two routines.
984  */
985 static void notrace register_one_mondo(unsigned long paddr, unsigned long type,
986 				       unsigned long qmask)
987 {
988 	unsigned long num_entries = (qmask + 1) / 64;
989 	unsigned long status;
990 
991 	status = sun4v_cpu_qconf(type, paddr, num_entries);
992 	if (status != HV_EOK) {
993 		prom_printf("SUN4V: sun4v_cpu_qconf(%lu:%lx:%lu) failed, "
994 			    "err %lu\n", type, paddr, num_entries, status);
995 		prom_halt();
996 	}
997 }
998 
999 void notrace sun4v_register_mondo_queues(int this_cpu)
1000 {
1001 	struct trap_per_cpu *tb = &trap_block[this_cpu];
1002 
1003 	register_one_mondo(tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO,
1004 			   tb->cpu_mondo_qmask);
1005 	register_one_mondo(tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO,
1006 			   tb->dev_mondo_qmask);
1007 	register_one_mondo(tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR,
1008 			   tb->resum_qmask);
1009 	register_one_mondo(tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR,
1010 			   tb->nonresum_qmask);
1011 }
1012 
1013 /* Each queue region must be a power of 2 multiple of 64 bytes in
1014  * size.  The base real address must be aligned to the size of the
1015  * region.  Thus, an 8KB queue must be 8KB aligned, for example.
1016  */
1017 static void __init alloc_one_queue(unsigned long *pa_ptr, unsigned long qmask)
1018 {
1019 	unsigned long size = PAGE_ALIGN(qmask + 1);
1020 	unsigned long order = get_order(size);
1021 	unsigned long p;
1022 
1023 	p = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
1024 	if (!p) {
1025 		prom_printf("SUN4V: Error, cannot allocate queue.\n");
1026 		prom_halt();
1027 	}
1028 
1029 	*pa_ptr = __pa(p);
1030 }
1031 
1032 static void __init init_cpu_send_mondo_info(struct trap_per_cpu *tb)
1033 {
1034 #ifdef CONFIG_SMP
1035 	unsigned long page;
1036 	void *mondo, *p;
1037 
1038 	BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > PAGE_SIZE);
1039 
1040 	/* Make sure mondo block is 64byte aligned */
1041 	p = kzalloc(127, GFP_KERNEL);
1042 	if (!p) {
1043 		prom_printf("SUN4V: Error, cannot allocate mondo block.\n");
1044 		prom_halt();
1045 	}
1046 	mondo = (void *)(((unsigned long)p + 63) & ~0x3f);
1047 	tb->cpu_mondo_block_pa = __pa(mondo);
1048 
1049 	page = get_zeroed_page(GFP_KERNEL);
1050 	if (!page) {
1051 		prom_printf("SUN4V: Error, cannot allocate cpu list page.\n");
1052 		prom_halt();
1053 	}
1054 
1055 	tb->cpu_list_pa = __pa(page);
1056 #endif
1057 }
1058 
1059 /* Allocate mondo and error queues for all possible cpus.  */
1060 static void __init sun4v_init_mondo_queues(void)
1061 {
1062 	int cpu;
1063 
1064 	for_each_possible_cpu(cpu) {
1065 		struct trap_per_cpu *tb = &trap_block[cpu];
1066 
1067 		alloc_one_queue(&tb->cpu_mondo_pa, tb->cpu_mondo_qmask);
1068 		alloc_one_queue(&tb->dev_mondo_pa, tb->dev_mondo_qmask);
1069 		alloc_one_queue(&tb->resum_mondo_pa, tb->resum_qmask);
1070 		alloc_one_queue(&tb->resum_kernel_buf_pa, tb->resum_qmask);
1071 		alloc_one_queue(&tb->nonresum_mondo_pa, tb->nonresum_qmask);
1072 		alloc_one_queue(&tb->nonresum_kernel_buf_pa,
1073 				tb->nonresum_qmask);
1074 	}
1075 }
1076 
1077 static void __init init_send_mondo_info(void)
1078 {
1079 	int cpu;
1080 
1081 	for_each_possible_cpu(cpu) {
1082 		struct trap_per_cpu *tb = &trap_block[cpu];
1083 
1084 		init_cpu_send_mondo_info(tb);
1085 	}
1086 }
1087 
1088 static struct irqaction timer_irq_action = {
1089 	.name = "timer",
1090 };
1091 
1092 static void __init irq_ivector_init(void)
1093 {
1094 	unsigned long size, order;
1095 	unsigned int ivecs;
1096 
1097 	/* If we are doing cookie only VIRQs then we do not need the ivector
1098 	 * table to process interrupts.
1099 	 */
1100 	if (sun4v_cookie_only_virqs())
1101 		return;
1102 
1103 	ivecs = size_nr_ivec();
1104 	size = sizeof(struct ino_bucket) * ivecs;
1105 	order = get_order(size);
1106 	ivector_table = (struct ino_bucket *)
1107 		__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
1108 	if (!ivector_table) {
1109 		prom_printf("Fatal error, cannot allocate ivector_table\n");
1110 		prom_halt();
1111 	}
1112 	__flush_dcache_range((unsigned long) ivector_table,
1113 			     ((unsigned long) ivector_table) + size);
1114 
1115 	ivector_table_pa = __pa(ivector_table);
1116 }
1117 
1118 /* Only invoked on boot processor.*/
1119 void __init init_IRQ(void)
1120 {
1121 	irq_init_hv();
1122 	irq_ivector_init();
1123 	map_prom_timers();
1124 	kill_prom_timer();
1125 
1126 	if (tlb_type == hypervisor)
1127 		sun4v_init_mondo_queues();
1128 
1129 	init_send_mondo_info();
1130 
1131 	if (tlb_type == hypervisor) {
1132 		/* Load up the boot cpu's entries.  */
1133 		sun4v_register_mondo_queues(hard_smp_processor_id());
1134 	}
1135 
1136 	/* We need to clear any IRQ's pending in the soft interrupt
1137 	 * registers, a spurious one could be left around from the
1138 	 * PROM timer which we just disabled.
1139 	 */
1140 	clear_softint(get_softint());
1141 
1142 	/* Now that ivector table is initialized, it is safe
1143 	 * to receive IRQ vector traps.  We will normally take
1144 	 * one or two right now, in case some device PROM used
1145 	 * to boot us wants to speak to us.  We just ignore them.
1146 	 */
1147 	__asm__ __volatile__("rdpr	%%pstate, %%g1\n\t"
1148 			     "or	%%g1, %0, %%g1\n\t"
1149 			     "wrpr	%%g1, 0x0, %%pstate"
1150 			     : /* No outputs */
1151 			     : "i" (PSTATE_IE)
1152 			     : "g1");
1153 
1154 	irq_to_desc(0)->action = &timer_irq_action;
1155 }
1156