xref: /linux/arch/sparc/kernel/ds.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1 /* ds.c: Domain Services driver for Logical Domains
2  *
3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/module.h>
8 #include <linux/types.h>
9 #include <linux/string.h>
10 #include <linux/slab.h>
11 #include <linux/sched.h>
12 #include <linux/sched/clock.h>
13 #include <linux/delay.h>
14 #include <linux/mutex.h>
15 #include <linux/kthread.h>
16 #include <linux/reboot.h>
17 #include <linux/cpu.h>
18 
19 #include <asm/hypervisor.h>
20 #include <asm/ldc.h>
21 #include <asm/vio.h>
22 #include <asm/mdesc.h>
23 #include <asm/head.h>
24 #include <asm/irq.h>
25 
26 #include "kernel.h"
27 
28 #define DRV_MODULE_NAME		"ds"
29 #define PFX DRV_MODULE_NAME	": "
30 #define DRV_MODULE_VERSION	"1.0"
31 #define DRV_MODULE_RELDATE	"Jul 11, 2007"
32 
33 static char version[] =
34 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
35 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
36 MODULE_DESCRIPTION("Sun LDOM domain services driver");
37 MODULE_LICENSE("GPL");
38 MODULE_VERSION(DRV_MODULE_VERSION);
39 
40 struct ds_msg_tag {
41 	__u32			type;
42 #define DS_INIT_REQ		0x00
43 #define DS_INIT_ACK		0x01
44 #define DS_INIT_NACK		0x02
45 #define DS_REG_REQ		0x03
46 #define DS_REG_ACK		0x04
47 #define DS_REG_NACK		0x05
48 #define DS_UNREG_REQ		0x06
49 #define DS_UNREG_ACK		0x07
50 #define DS_UNREG_NACK		0x08
51 #define DS_DATA			0x09
52 #define DS_NACK			0x0a
53 
54 	__u32			len;
55 };
56 
57 /* Result codes */
58 #define DS_OK			0x00
59 #define DS_REG_VER_NACK		0x01
60 #define DS_REG_DUP		0x02
61 #define DS_INV_HDL		0x03
62 #define DS_TYPE_UNKNOWN		0x04
63 
64 struct ds_version {
65 	__u16			major;
66 	__u16			minor;
67 };
68 
69 struct ds_ver_req {
70 	struct ds_msg_tag	tag;
71 	struct ds_version	ver;
72 };
73 
74 struct ds_ver_ack {
75 	struct ds_msg_tag	tag;
76 	__u16			minor;
77 };
78 
79 struct ds_ver_nack {
80 	struct ds_msg_tag	tag;
81 	__u16			major;
82 };
83 
84 struct ds_reg_req {
85 	struct ds_msg_tag	tag;
86 	__u64			handle;
87 	__u16			major;
88 	__u16			minor;
89 	char			svc_id[0];
90 };
91 
92 struct ds_reg_ack {
93 	struct ds_msg_tag	tag;
94 	__u64			handle;
95 	__u16			minor;
96 };
97 
98 struct ds_reg_nack {
99 	struct ds_msg_tag	tag;
100 	__u64			handle;
101 	__u16			major;
102 };
103 
104 struct ds_unreg_req {
105 	struct ds_msg_tag	tag;
106 	__u64			handle;
107 };
108 
109 struct ds_unreg_ack {
110 	struct ds_msg_tag	tag;
111 	__u64			handle;
112 };
113 
114 struct ds_unreg_nack {
115 	struct ds_msg_tag	tag;
116 	__u64			handle;
117 };
118 
119 struct ds_data {
120 	struct ds_msg_tag	tag;
121 	__u64			handle;
122 };
123 
124 struct ds_data_nack {
125 	struct ds_msg_tag	tag;
126 	__u64			handle;
127 	__u64			result;
128 };
129 
130 struct ds_info;
131 struct ds_cap_state {
132 	__u64			handle;
133 
134 	void			(*data)(struct ds_info *dp,
135 					struct ds_cap_state *cp,
136 					void *buf, int len);
137 
138 	const char		*service_id;
139 
140 	u8			state;
141 #define CAP_STATE_UNKNOWN	0x00
142 #define CAP_STATE_REG_SENT	0x01
143 #define CAP_STATE_REGISTERED	0x02
144 };
145 
146 static void md_update_data(struct ds_info *dp, struct ds_cap_state *cp,
147 			   void *buf, int len);
148 static void domain_shutdown_data(struct ds_info *dp,
149 				 struct ds_cap_state *cp,
150 				 void *buf, int len);
151 static void domain_panic_data(struct ds_info *dp,
152 			      struct ds_cap_state *cp,
153 			      void *buf, int len);
154 #ifdef CONFIG_HOTPLUG_CPU
155 static void dr_cpu_data(struct ds_info *dp,
156 			struct ds_cap_state *cp,
157 			void *buf, int len);
158 #endif
159 static void ds_pri_data(struct ds_info *dp,
160 			struct ds_cap_state *cp,
161 			void *buf, int len);
162 static void ds_var_data(struct ds_info *dp,
163 			struct ds_cap_state *cp,
164 			void *buf, int len);
165 
166 static struct ds_cap_state ds_states_template[] = {
167 	{
168 		.service_id	= "md-update",
169 		.data		= md_update_data,
170 	},
171 	{
172 		.service_id	= "domain-shutdown",
173 		.data		= domain_shutdown_data,
174 	},
175 	{
176 		.service_id	= "domain-panic",
177 		.data		= domain_panic_data,
178 	},
179 #ifdef CONFIG_HOTPLUG_CPU
180 	{
181 		.service_id	= "dr-cpu",
182 		.data		= dr_cpu_data,
183 	},
184 #endif
185 	{
186 		.service_id	= "pri",
187 		.data		= ds_pri_data,
188 	},
189 	{
190 		.service_id	= "var-config",
191 		.data		= ds_var_data,
192 	},
193 	{
194 		.service_id	= "var-config-backup",
195 		.data		= ds_var_data,
196 	},
197 };
198 
199 static DEFINE_SPINLOCK(ds_lock);
200 
201 struct ds_info {
202 	struct ldc_channel	*lp;
203 	u8			hs_state;
204 #define DS_HS_START		0x01
205 #define DS_HS_DONE		0x02
206 
207 	u64			id;
208 
209 	void			*rcv_buf;
210 	int			rcv_buf_len;
211 
212 	struct ds_cap_state	*ds_states;
213 	int			num_ds_states;
214 
215 	struct ds_info		*next;
216 };
217 
218 static struct ds_info *ds_info_list;
219 
220 static struct ds_cap_state *find_cap(struct ds_info *dp, u64 handle)
221 {
222 	unsigned int index = handle >> 32;
223 
224 	if (index >= dp->num_ds_states)
225 		return NULL;
226 	return &dp->ds_states[index];
227 }
228 
229 static struct ds_cap_state *find_cap_by_string(struct ds_info *dp,
230 					       const char *name)
231 {
232 	int i;
233 
234 	for (i = 0; i < dp->num_ds_states; i++) {
235 		if (strcmp(dp->ds_states[i].service_id, name))
236 			continue;
237 
238 		return &dp->ds_states[i];
239 	}
240 	return NULL;
241 }
242 
243 static int __ds_send(struct ldc_channel *lp, void *data, int len)
244 {
245 	int err, limit = 1000;
246 
247 	err = -EINVAL;
248 	while (limit-- > 0) {
249 		err = ldc_write(lp, data, len);
250 		if (!err || (err != -EAGAIN))
251 			break;
252 		udelay(1);
253 	}
254 
255 	return err;
256 }
257 
258 static int ds_send(struct ldc_channel *lp, void *data, int len)
259 {
260 	unsigned long flags;
261 	int err;
262 
263 	spin_lock_irqsave(&ds_lock, flags);
264 	err = __ds_send(lp, data, len);
265 	spin_unlock_irqrestore(&ds_lock, flags);
266 
267 	return err;
268 }
269 
270 struct ds_md_update_req {
271 	__u64				req_num;
272 };
273 
274 struct ds_md_update_res {
275 	__u64				req_num;
276 	__u32				result;
277 };
278 
279 static void md_update_data(struct ds_info *dp,
280 			   struct ds_cap_state *cp,
281 			   void *buf, int len)
282 {
283 	struct ldc_channel *lp = dp->lp;
284 	struct ds_data *dpkt = buf;
285 	struct ds_md_update_req *rp;
286 	struct {
287 		struct ds_data		data;
288 		struct ds_md_update_res	res;
289 	} pkt;
290 
291 	rp = (struct ds_md_update_req *) (dpkt + 1);
292 
293 	printk(KERN_INFO "ds-%llu: Machine description update.\n", dp->id);
294 
295 	mdesc_update();
296 
297 	memset(&pkt, 0, sizeof(pkt));
298 	pkt.data.tag.type = DS_DATA;
299 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
300 	pkt.data.handle = cp->handle;
301 	pkt.res.req_num = rp->req_num;
302 	pkt.res.result = DS_OK;
303 
304 	ds_send(lp, &pkt, sizeof(pkt));
305 }
306 
307 struct ds_shutdown_req {
308 	__u64				req_num;
309 	__u32				ms_delay;
310 };
311 
312 struct ds_shutdown_res {
313 	__u64				req_num;
314 	__u32				result;
315 	char				reason[1];
316 };
317 
318 static void domain_shutdown_data(struct ds_info *dp,
319 				 struct ds_cap_state *cp,
320 				 void *buf, int len)
321 {
322 	struct ldc_channel *lp = dp->lp;
323 	struct ds_data *dpkt = buf;
324 	struct ds_shutdown_req *rp;
325 	struct {
326 		struct ds_data		data;
327 		struct ds_shutdown_res	res;
328 	} pkt;
329 
330 	rp = (struct ds_shutdown_req *) (dpkt + 1);
331 
332 	printk(KERN_ALERT "ds-%llu: Shutdown request from "
333 	       "LDOM manager received.\n", dp->id);
334 
335 	memset(&pkt, 0, sizeof(pkt));
336 	pkt.data.tag.type = DS_DATA;
337 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
338 	pkt.data.handle = cp->handle;
339 	pkt.res.req_num = rp->req_num;
340 	pkt.res.result = DS_OK;
341 	pkt.res.reason[0] = 0;
342 
343 	ds_send(lp, &pkt, sizeof(pkt));
344 
345 	orderly_poweroff(true);
346 }
347 
348 struct ds_panic_req {
349 	__u64				req_num;
350 };
351 
352 struct ds_panic_res {
353 	__u64				req_num;
354 	__u32				result;
355 	char				reason[1];
356 };
357 
358 static void domain_panic_data(struct ds_info *dp,
359 			      struct ds_cap_state *cp,
360 			      void *buf, int len)
361 {
362 	struct ldc_channel *lp = dp->lp;
363 	struct ds_data *dpkt = buf;
364 	struct ds_panic_req *rp;
365 	struct {
366 		struct ds_data		data;
367 		struct ds_panic_res	res;
368 	} pkt;
369 
370 	rp = (struct ds_panic_req *) (dpkt + 1);
371 
372 	printk(KERN_ALERT "ds-%llu: Panic request from "
373 	       "LDOM manager received.\n", dp->id);
374 
375 	memset(&pkt, 0, sizeof(pkt));
376 	pkt.data.tag.type = DS_DATA;
377 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
378 	pkt.data.handle = cp->handle;
379 	pkt.res.req_num = rp->req_num;
380 	pkt.res.result = DS_OK;
381 	pkt.res.reason[0] = 0;
382 
383 	ds_send(lp, &pkt, sizeof(pkt));
384 
385 	panic("PANIC requested by LDOM manager.");
386 }
387 
388 #ifdef CONFIG_HOTPLUG_CPU
389 struct dr_cpu_tag {
390 	__u64				req_num;
391 	__u32				type;
392 #define DR_CPU_CONFIGURE		0x43
393 #define DR_CPU_UNCONFIGURE		0x55
394 #define DR_CPU_FORCE_UNCONFIGURE	0x46
395 #define DR_CPU_STATUS			0x53
396 
397 /* Responses */
398 #define DR_CPU_OK			0x6f
399 #define DR_CPU_ERROR			0x65
400 
401 	__u32				num_records;
402 };
403 
404 struct dr_cpu_resp_entry {
405 	__u32				cpu;
406 	__u32				result;
407 #define DR_CPU_RES_OK			0x00
408 #define DR_CPU_RES_FAILURE		0x01
409 #define DR_CPU_RES_BLOCKED		0x02
410 #define DR_CPU_RES_CPU_NOT_RESPONDING	0x03
411 #define DR_CPU_RES_NOT_IN_MD		0x04
412 
413 	__u32				stat;
414 #define DR_CPU_STAT_NOT_PRESENT		0x00
415 #define DR_CPU_STAT_UNCONFIGURED	0x01
416 #define DR_CPU_STAT_CONFIGURED		0x02
417 
418 	__u32				str_off;
419 };
420 
421 static void __dr_cpu_send_error(struct ds_info *dp,
422 				struct ds_cap_state *cp,
423 				struct ds_data *data)
424 {
425 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
426 	struct {
427 		struct ds_data		data;
428 		struct dr_cpu_tag	tag;
429 	} pkt;
430 	int msg_len;
431 
432 	memset(&pkt, 0, sizeof(pkt));
433 	pkt.data.tag.type = DS_DATA;
434 	pkt.data.handle = cp->handle;
435 	pkt.tag.req_num = tag->req_num;
436 	pkt.tag.type = DR_CPU_ERROR;
437 	pkt.tag.num_records = 0;
438 
439 	msg_len = (sizeof(struct ds_data) +
440 		   sizeof(struct dr_cpu_tag));
441 
442 	pkt.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
443 
444 	__ds_send(dp->lp, &pkt, msg_len);
445 }
446 
447 static void dr_cpu_send_error(struct ds_info *dp,
448 			      struct ds_cap_state *cp,
449 			      struct ds_data *data)
450 {
451 	unsigned long flags;
452 
453 	spin_lock_irqsave(&ds_lock, flags);
454 	__dr_cpu_send_error(dp, cp, data);
455 	spin_unlock_irqrestore(&ds_lock, flags);
456 }
457 
458 #define CPU_SENTINEL	0xffffffff
459 
460 static void purge_dups(u32 *list, u32 num_ents)
461 {
462 	unsigned int i;
463 
464 	for (i = 0; i < num_ents; i++) {
465 		u32 cpu = list[i];
466 		unsigned int j;
467 
468 		if (cpu == CPU_SENTINEL)
469 			continue;
470 
471 		for (j = i + 1; j < num_ents; j++) {
472 			if (list[j] == cpu)
473 				list[j] = CPU_SENTINEL;
474 		}
475 	}
476 }
477 
478 static int dr_cpu_size_response(int ncpus)
479 {
480 	return (sizeof(struct ds_data) +
481 		sizeof(struct dr_cpu_tag) +
482 		(sizeof(struct dr_cpu_resp_entry) * ncpus));
483 }
484 
485 static void dr_cpu_init_response(struct ds_data *resp, u64 req_num,
486 				 u64 handle, int resp_len, int ncpus,
487 				 cpumask_t *mask, u32 default_stat)
488 {
489 	struct dr_cpu_resp_entry *ent;
490 	struct dr_cpu_tag *tag;
491 	int i, cpu;
492 
493 	tag = (struct dr_cpu_tag *) (resp + 1);
494 	ent = (struct dr_cpu_resp_entry *) (tag + 1);
495 
496 	resp->tag.type = DS_DATA;
497 	resp->tag.len = resp_len - sizeof(struct ds_msg_tag);
498 	resp->handle = handle;
499 	tag->req_num = req_num;
500 	tag->type = DR_CPU_OK;
501 	tag->num_records = ncpus;
502 
503 	i = 0;
504 	for_each_cpu(cpu, mask) {
505 		ent[i].cpu = cpu;
506 		ent[i].result = DR_CPU_RES_OK;
507 		ent[i].stat = default_stat;
508 		i++;
509 	}
510 	BUG_ON(i != ncpus);
511 }
512 
513 static void dr_cpu_mark(struct ds_data *resp, int cpu, int ncpus,
514 			u32 res, u32 stat)
515 {
516 	struct dr_cpu_resp_entry *ent;
517 	struct dr_cpu_tag *tag;
518 	int i;
519 
520 	tag = (struct dr_cpu_tag *) (resp + 1);
521 	ent = (struct dr_cpu_resp_entry *) (tag + 1);
522 
523 	for (i = 0; i < ncpus; i++) {
524 		if (ent[i].cpu != cpu)
525 			continue;
526 		ent[i].result = res;
527 		ent[i].stat = stat;
528 		break;
529 	}
530 }
531 
532 static int dr_cpu_configure(struct ds_info *dp, struct ds_cap_state *cp,
533 			    u64 req_num, cpumask_t *mask)
534 {
535 	struct ds_data *resp;
536 	int resp_len, ncpus, cpu;
537 	unsigned long flags;
538 
539 	ncpus = cpumask_weight(mask);
540 	resp_len = dr_cpu_size_response(ncpus);
541 	resp = kzalloc(resp_len, GFP_KERNEL);
542 	if (!resp)
543 		return -ENOMEM;
544 
545 	dr_cpu_init_response(resp, req_num, cp->handle,
546 			     resp_len, ncpus, mask,
547 			     DR_CPU_STAT_CONFIGURED);
548 
549 	mdesc_populate_present_mask(mask);
550 	mdesc_fill_in_cpu_data(mask);
551 
552 	for_each_cpu(cpu, mask) {
553 		int err;
554 
555 		printk(KERN_INFO "ds-%llu: Starting cpu %d...\n",
556 		       dp->id, cpu);
557 		err = cpu_up(cpu);
558 		if (err) {
559 			__u32 res = DR_CPU_RES_FAILURE;
560 			__u32 stat = DR_CPU_STAT_UNCONFIGURED;
561 
562 			if (!cpu_present(cpu)) {
563 				/* CPU not present in MD */
564 				res = DR_CPU_RES_NOT_IN_MD;
565 				stat = DR_CPU_STAT_NOT_PRESENT;
566 			} else if (err == -ENODEV) {
567 				/* CPU did not call in successfully */
568 				res = DR_CPU_RES_CPU_NOT_RESPONDING;
569 			}
570 
571 			printk(KERN_INFO "ds-%llu: CPU startup failed err=%d\n",
572 			       dp->id, err);
573 			dr_cpu_mark(resp, cpu, ncpus, res, stat);
574 		}
575 	}
576 
577 	spin_lock_irqsave(&ds_lock, flags);
578 	__ds_send(dp->lp, resp, resp_len);
579 	spin_unlock_irqrestore(&ds_lock, flags);
580 
581 	kfree(resp);
582 
583 	/* Redistribute IRQs, taking into account the new cpus.  */
584 	fixup_irqs();
585 
586 	return 0;
587 }
588 
589 static int dr_cpu_unconfigure(struct ds_info *dp,
590 			      struct ds_cap_state *cp,
591 			      u64 req_num,
592 			      cpumask_t *mask)
593 {
594 	struct ds_data *resp;
595 	int resp_len, ncpus, cpu;
596 	unsigned long flags;
597 
598 	ncpus = cpumask_weight(mask);
599 	resp_len = dr_cpu_size_response(ncpus);
600 	resp = kzalloc(resp_len, GFP_KERNEL);
601 	if (!resp)
602 		return -ENOMEM;
603 
604 	dr_cpu_init_response(resp, req_num, cp->handle,
605 			     resp_len, ncpus, mask,
606 			     DR_CPU_STAT_UNCONFIGURED);
607 
608 	for_each_cpu(cpu, mask) {
609 		int err;
610 
611 		printk(KERN_INFO "ds-%llu: Shutting down cpu %d...\n",
612 		       dp->id, cpu);
613 		err = cpu_down(cpu);
614 		if (err)
615 			dr_cpu_mark(resp, cpu, ncpus,
616 				    DR_CPU_RES_FAILURE,
617 				    DR_CPU_STAT_CONFIGURED);
618 	}
619 
620 	spin_lock_irqsave(&ds_lock, flags);
621 	__ds_send(dp->lp, resp, resp_len);
622 	spin_unlock_irqrestore(&ds_lock, flags);
623 
624 	kfree(resp);
625 
626 	return 0;
627 }
628 
629 static void dr_cpu_data(struct ds_info *dp, struct ds_cap_state *cp, void *buf,
630 			int len)
631 {
632 	struct ds_data *data = buf;
633 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
634 	u32 *cpu_list = (u32 *) (tag + 1);
635 	u64 req_num = tag->req_num;
636 	cpumask_t mask;
637 	unsigned int i;
638 	int err;
639 
640 	switch (tag->type) {
641 	case DR_CPU_CONFIGURE:
642 	case DR_CPU_UNCONFIGURE:
643 	case DR_CPU_FORCE_UNCONFIGURE:
644 		break;
645 
646 	default:
647 		dr_cpu_send_error(dp, cp, data);
648 		return;
649 	}
650 
651 	purge_dups(cpu_list, tag->num_records);
652 
653 	cpumask_clear(&mask);
654 	for (i = 0; i < tag->num_records; i++) {
655 		if (cpu_list[i] == CPU_SENTINEL)
656 			continue;
657 
658 		if (cpu_list[i] < nr_cpu_ids)
659 			cpumask_set_cpu(cpu_list[i], &mask);
660 	}
661 
662 	if (tag->type == DR_CPU_CONFIGURE)
663 		err = dr_cpu_configure(dp, cp, req_num, &mask);
664 	else
665 		err = dr_cpu_unconfigure(dp, cp, req_num, &mask);
666 
667 	if (err)
668 		dr_cpu_send_error(dp, cp, data);
669 }
670 #endif /* CONFIG_HOTPLUG_CPU */
671 
672 struct ds_pri_msg {
673 	__u64				req_num;
674 	__u64				type;
675 #define DS_PRI_REQUEST			0x00
676 #define DS_PRI_DATA			0x01
677 #define DS_PRI_UPDATE			0x02
678 };
679 
680 static void ds_pri_data(struct ds_info *dp,
681 			struct ds_cap_state *cp,
682 			void *buf, int len)
683 {
684 	struct ds_data *dpkt = buf;
685 	struct ds_pri_msg *rp;
686 
687 	rp = (struct ds_pri_msg *) (dpkt + 1);
688 
689 	printk(KERN_INFO "ds-%llu: PRI REQ [%llx:%llx], len=%d\n",
690 	       dp->id, rp->req_num, rp->type, len);
691 }
692 
693 struct ds_var_hdr {
694 	__u32				type;
695 #define DS_VAR_SET_REQ			0x00
696 #define DS_VAR_DELETE_REQ		0x01
697 #define DS_VAR_SET_RESP			0x02
698 #define DS_VAR_DELETE_RESP		0x03
699 };
700 
701 struct ds_var_set_msg {
702 	struct ds_var_hdr		hdr;
703 	char				name_and_value[0];
704 };
705 
706 struct ds_var_delete_msg {
707 	struct ds_var_hdr		hdr;
708 	char				name[0];
709 };
710 
711 struct ds_var_resp {
712 	struct ds_var_hdr		hdr;
713 	__u32				result;
714 #define DS_VAR_SUCCESS			0x00
715 #define DS_VAR_NO_SPACE			0x01
716 #define DS_VAR_INVALID_VAR		0x02
717 #define DS_VAR_INVALID_VAL		0x03
718 #define DS_VAR_NOT_PRESENT		0x04
719 };
720 
721 static DEFINE_MUTEX(ds_var_mutex);
722 static int ds_var_doorbell;
723 static int ds_var_response;
724 
725 static void ds_var_data(struct ds_info *dp,
726 			struct ds_cap_state *cp,
727 			void *buf, int len)
728 {
729 	struct ds_data *dpkt = buf;
730 	struct ds_var_resp *rp;
731 
732 	rp = (struct ds_var_resp *) (dpkt + 1);
733 
734 	if (rp->hdr.type != DS_VAR_SET_RESP &&
735 	    rp->hdr.type != DS_VAR_DELETE_RESP)
736 		return;
737 
738 	ds_var_response = rp->result;
739 	wmb();
740 	ds_var_doorbell = 1;
741 }
742 
743 void ldom_set_var(const char *var, const char *value)
744 {
745 	struct ds_cap_state *cp;
746 	struct ds_info *dp;
747 	unsigned long flags;
748 
749 	spin_lock_irqsave(&ds_lock, flags);
750 	cp = NULL;
751 	for (dp = ds_info_list; dp; dp = dp->next) {
752 		struct ds_cap_state *tmp;
753 
754 		tmp = find_cap_by_string(dp, "var-config");
755 		if (tmp && tmp->state == CAP_STATE_REGISTERED) {
756 			cp = tmp;
757 			break;
758 		}
759 	}
760 	if (!cp) {
761 		for (dp = ds_info_list; dp; dp = dp->next) {
762 			struct ds_cap_state *tmp;
763 
764 			tmp = find_cap_by_string(dp, "var-config-backup");
765 			if (tmp && tmp->state == CAP_STATE_REGISTERED) {
766 				cp = tmp;
767 				break;
768 			}
769 		}
770 	}
771 	spin_unlock_irqrestore(&ds_lock, flags);
772 
773 	if (cp) {
774 		union {
775 			struct {
776 				struct ds_data		data;
777 				struct ds_var_set_msg	msg;
778 			} header;
779 			char			all[512];
780 		} pkt;
781 		char  *base, *p;
782 		int msg_len, loops;
783 
784 		if (strlen(var) + strlen(value) + 2 >
785 		    sizeof(pkt) - sizeof(pkt.header)) {
786 			printk(KERN_ERR PFX
787 				"contents length: %zu, which more than max: %lu,"
788 				"so could not set (%s) variable to (%s).\n",
789 				strlen(var) + strlen(value) + 2,
790 				sizeof(pkt) - sizeof(pkt.header), var, value);
791 			return;
792 		}
793 
794 		memset(&pkt, 0, sizeof(pkt));
795 		pkt.header.data.tag.type = DS_DATA;
796 		pkt.header.data.handle = cp->handle;
797 		pkt.header.msg.hdr.type = DS_VAR_SET_REQ;
798 		base = p = &pkt.header.msg.name_and_value[0];
799 		strcpy(p, var);
800 		p += strlen(var) + 1;
801 		strcpy(p, value);
802 		p += strlen(value) + 1;
803 
804 		msg_len = (sizeof(struct ds_data) +
805 			   sizeof(struct ds_var_set_msg) +
806 			   (p - base));
807 		msg_len = (msg_len + 3) & ~3;
808 		pkt.header.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
809 
810 		mutex_lock(&ds_var_mutex);
811 
812 		spin_lock_irqsave(&ds_lock, flags);
813 		ds_var_doorbell = 0;
814 		ds_var_response = -1;
815 
816 		__ds_send(dp->lp, &pkt, msg_len);
817 		spin_unlock_irqrestore(&ds_lock, flags);
818 
819 		loops = 1000;
820 		while (ds_var_doorbell == 0) {
821 			if (loops-- < 0)
822 				break;
823 			barrier();
824 			udelay(100);
825 		}
826 
827 		mutex_unlock(&ds_var_mutex);
828 
829 		if (ds_var_doorbell == 0 ||
830 		    ds_var_response != DS_VAR_SUCCESS)
831 			printk(KERN_ERR "ds-%llu: var-config [%s:%s] "
832 			       "failed, response(%d).\n",
833 			       dp->id, var, value,
834 			       ds_var_response);
835 	} else {
836 		printk(KERN_ERR PFX "var-config not registered so "
837 		       "could not set (%s) variable to (%s).\n",
838 		       var, value);
839 	}
840 }
841 
842 static char full_boot_str[256] __attribute__((aligned(32)));
843 static int reboot_data_supported;
844 
845 void ldom_reboot(const char *boot_command)
846 {
847 	/* Don't bother with any of this if the boot_command
848 	 * is empty.
849 	 */
850 	if (boot_command && strlen(boot_command)) {
851 		unsigned long len;
852 
853 		snprintf(full_boot_str, sizeof(full_boot_str), "boot %s",
854 			 boot_command);
855 		len = strlen(full_boot_str);
856 
857 		if (reboot_data_supported) {
858 			unsigned long ra = kimage_addr_to_ra(full_boot_str);
859 			unsigned long hv_ret;
860 
861 			hv_ret = sun4v_reboot_data_set(ra, len);
862 			if (hv_ret != HV_EOK)
863 				pr_err("SUN4V: Unable to set reboot data "
864 				       "hv_ret=%lu\n", hv_ret);
865 		} else {
866 			ldom_set_var("reboot-command", full_boot_str);
867 		}
868 	}
869 	sun4v_mach_sir();
870 }
871 
872 void ldom_power_off(void)
873 {
874 	sun4v_mach_exit(0);
875 }
876 
877 static void ds_conn_reset(struct ds_info *dp)
878 {
879 	printk(KERN_ERR "ds-%llu: ds_conn_reset() from %pf\n",
880 	       dp->id, __builtin_return_address(0));
881 }
882 
883 static int register_services(struct ds_info *dp)
884 {
885 	struct ldc_channel *lp = dp->lp;
886 	int i;
887 
888 	for (i = 0; i < dp->num_ds_states; i++) {
889 		struct {
890 			struct ds_reg_req req;
891 			u8 id_buf[256];
892 		} pbuf;
893 		struct ds_cap_state *cp = &dp->ds_states[i];
894 		int err, msg_len;
895 		u64 new_count;
896 
897 		if (cp->state == CAP_STATE_REGISTERED)
898 			continue;
899 
900 		new_count = sched_clock() & 0xffffffff;
901 		cp->handle = ((u64) i << 32) | new_count;
902 
903 		msg_len = (sizeof(struct ds_reg_req) +
904 			   strlen(cp->service_id));
905 
906 		memset(&pbuf, 0, sizeof(pbuf));
907 		pbuf.req.tag.type = DS_REG_REQ;
908 		pbuf.req.tag.len = (msg_len - sizeof(struct ds_msg_tag));
909 		pbuf.req.handle = cp->handle;
910 		pbuf.req.major = 1;
911 		pbuf.req.minor = 0;
912 		strcpy(pbuf.id_buf, cp->service_id);
913 
914 		err = __ds_send(lp, &pbuf, msg_len);
915 		if (err > 0)
916 			cp->state = CAP_STATE_REG_SENT;
917 	}
918 	return 0;
919 }
920 
921 static int ds_handshake(struct ds_info *dp, struct ds_msg_tag *pkt)
922 {
923 
924 	if (dp->hs_state == DS_HS_START) {
925 		if (pkt->type != DS_INIT_ACK)
926 			goto conn_reset;
927 
928 		dp->hs_state = DS_HS_DONE;
929 
930 		return register_services(dp);
931 	}
932 
933 	if (dp->hs_state != DS_HS_DONE)
934 		goto conn_reset;
935 
936 	if (pkt->type == DS_REG_ACK) {
937 		struct ds_reg_ack *ap = (struct ds_reg_ack *) pkt;
938 		struct ds_cap_state *cp = find_cap(dp, ap->handle);
939 
940 		if (!cp) {
941 			printk(KERN_ERR "ds-%llu: REG ACK for unknown "
942 			       "handle %llx\n", dp->id, ap->handle);
943 			return 0;
944 		}
945 		printk(KERN_INFO "ds-%llu: Registered %s service.\n",
946 		       dp->id, cp->service_id);
947 		cp->state = CAP_STATE_REGISTERED;
948 	} else if (pkt->type == DS_REG_NACK) {
949 		struct ds_reg_nack *np = (struct ds_reg_nack *) pkt;
950 		struct ds_cap_state *cp = find_cap(dp, np->handle);
951 
952 		if (!cp) {
953 			printk(KERN_ERR "ds-%llu: REG NACK for "
954 			       "unknown handle %llx\n",
955 			       dp->id, np->handle);
956 			return 0;
957 		}
958 		cp->state = CAP_STATE_UNKNOWN;
959 	}
960 
961 	return 0;
962 
963 conn_reset:
964 	ds_conn_reset(dp);
965 	return -ECONNRESET;
966 }
967 
968 static void __send_ds_nack(struct ds_info *dp, u64 handle)
969 {
970 	struct ds_data_nack nack = {
971 		.tag = {
972 			.type = DS_NACK,
973 			.len = (sizeof(struct ds_data_nack) -
974 				sizeof(struct ds_msg_tag)),
975 		},
976 		.handle = handle,
977 		.result = DS_INV_HDL,
978 	};
979 
980 	__ds_send(dp->lp, &nack, sizeof(nack));
981 }
982 
983 static LIST_HEAD(ds_work_list);
984 static DECLARE_WAIT_QUEUE_HEAD(ds_wait);
985 
986 struct ds_queue_entry {
987 	struct list_head		list;
988 	struct ds_info			*dp;
989 	int				req_len;
990 	int				__pad;
991 	u64				req[0];
992 };
993 
994 static void process_ds_work(void)
995 {
996 	struct ds_queue_entry *qp, *tmp;
997 	unsigned long flags;
998 	LIST_HEAD(todo);
999 
1000 	spin_lock_irqsave(&ds_lock, flags);
1001 	list_splice_init(&ds_work_list, &todo);
1002 	spin_unlock_irqrestore(&ds_lock, flags);
1003 
1004 	list_for_each_entry_safe(qp, tmp, &todo, list) {
1005 		struct ds_data *dpkt = (struct ds_data *) qp->req;
1006 		struct ds_info *dp = qp->dp;
1007 		struct ds_cap_state *cp = find_cap(dp, dpkt->handle);
1008 		int req_len = qp->req_len;
1009 
1010 		if (!cp) {
1011 			printk(KERN_ERR "ds-%llu: Data for unknown "
1012 			       "handle %llu\n",
1013 			       dp->id, dpkt->handle);
1014 
1015 			spin_lock_irqsave(&ds_lock, flags);
1016 			__send_ds_nack(dp, dpkt->handle);
1017 			spin_unlock_irqrestore(&ds_lock, flags);
1018 		} else {
1019 			cp->data(dp, cp, dpkt, req_len);
1020 		}
1021 
1022 		list_del(&qp->list);
1023 		kfree(qp);
1024 	}
1025 }
1026 
1027 static int ds_thread(void *__unused)
1028 {
1029 	DEFINE_WAIT(wait);
1030 
1031 	while (1) {
1032 		prepare_to_wait(&ds_wait, &wait, TASK_INTERRUPTIBLE);
1033 		if (list_empty(&ds_work_list))
1034 			schedule();
1035 		finish_wait(&ds_wait, &wait);
1036 
1037 		if (kthread_should_stop())
1038 			break;
1039 
1040 		process_ds_work();
1041 	}
1042 
1043 	return 0;
1044 }
1045 
1046 static int ds_data(struct ds_info *dp, struct ds_msg_tag *pkt, int len)
1047 {
1048 	struct ds_data *dpkt = (struct ds_data *) pkt;
1049 	struct ds_queue_entry *qp;
1050 
1051 	qp = kmalloc(sizeof(struct ds_queue_entry) + len, GFP_ATOMIC);
1052 	if (!qp) {
1053 		__send_ds_nack(dp, dpkt->handle);
1054 	} else {
1055 		qp->dp = dp;
1056 		memcpy(&qp->req, pkt, len);
1057 		list_add_tail(&qp->list, &ds_work_list);
1058 		wake_up(&ds_wait);
1059 	}
1060 	return 0;
1061 }
1062 
1063 static void ds_up(struct ds_info *dp)
1064 {
1065 	struct ldc_channel *lp = dp->lp;
1066 	struct ds_ver_req req;
1067 	int err;
1068 
1069 	req.tag.type = DS_INIT_REQ;
1070 	req.tag.len = sizeof(req) - sizeof(struct ds_msg_tag);
1071 	req.ver.major = 1;
1072 	req.ver.minor = 0;
1073 
1074 	err = __ds_send(lp, &req, sizeof(req));
1075 	if (err > 0)
1076 		dp->hs_state = DS_HS_START;
1077 }
1078 
1079 static void ds_reset(struct ds_info *dp)
1080 {
1081 	int i;
1082 
1083 	dp->hs_state = 0;
1084 
1085 	for (i = 0; i < dp->num_ds_states; i++) {
1086 		struct ds_cap_state *cp = &dp->ds_states[i];
1087 
1088 		cp->state = CAP_STATE_UNKNOWN;
1089 	}
1090 }
1091 
1092 static void ds_event(void *arg, int event)
1093 {
1094 	struct ds_info *dp = arg;
1095 	struct ldc_channel *lp = dp->lp;
1096 	unsigned long flags;
1097 	int err;
1098 
1099 	spin_lock_irqsave(&ds_lock, flags);
1100 
1101 	if (event == LDC_EVENT_UP) {
1102 		ds_up(dp);
1103 		spin_unlock_irqrestore(&ds_lock, flags);
1104 		return;
1105 	}
1106 
1107 	if (event == LDC_EVENT_RESET) {
1108 		ds_reset(dp);
1109 		spin_unlock_irqrestore(&ds_lock, flags);
1110 		return;
1111 	}
1112 
1113 	if (event != LDC_EVENT_DATA_READY) {
1114 		printk(KERN_WARNING "ds-%llu: Unexpected LDC event %d\n",
1115 		       dp->id, event);
1116 		spin_unlock_irqrestore(&ds_lock, flags);
1117 		return;
1118 	}
1119 
1120 	err = 0;
1121 	while (1) {
1122 		struct ds_msg_tag *tag;
1123 
1124 		err = ldc_read(lp, dp->rcv_buf, sizeof(*tag));
1125 
1126 		if (unlikely(err < 0)) {
1127 			if (err == -ECONNRESET)
1128 				ds_conn_reset(dp);
1129 			break;
1130 		}
1131 		if (err == 0)
1132 			break;
1133 
1134 		tag = dp->rcv_buf;
1135 		err = ldc_read(lp, tag + 1, tag->len);
1136 
1137 		if (unlikely(err < 0)) {
1138 			if (err == -ECONNRESET)
1139 				ds_conn_reset(dp);
1140 			break;
1141 		}
1142 		if (err < tag->len)
1143 			break;
1144 
1145 		if (tag->type < DS_DATA)
1146 			err = ds_handshake(dp, dp->rcv_buf);
1147 		else
1148 			err = ds_data(dp, dp->rcv_buf,
1149 				      sizeof(*tag) + err);
1150 		if (err == -ECONNRESET)
1151 			break;
1152 	}
1153 
1154 	spin_unlock_irqrestore(&ds_lock, flags);
1155 }
1156 
1157 static int ds_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1158 {
1159 	static int ds_version_printed;
1160 	struct ldc_channel_config ds_cfg = {
1161 		.event		= ds_event,
1162 		.mtu		= 4096,
1163 		.mode		= LDC_MODE_STREAM,
1164 	};
1165 	struct mdesc_handle *hp;
1166 	struct ldc_channel *lp;
1167 	struct ds_info *dp;
1168 	const u64 *val;
1169 	int err, i;
1170 
1171 	if (ds_version_printed++ == 0)
1172 		printk(KERN_INFO "%s", version);
1173 
1174 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1175 	err = -ENOMEM;
1176 	if (!dp)
1177 		goto out_err;
1178 
1179 	hp = mdesc_grab();
1180 	val = mdesc_get_property(hp, vdev->mp, "id", NULL);
1181 	if (val)
1182 		dp->id = *val;
1183 	mdesc_release(hp);
1184 
1185 	dp->rcv_buf = kzalloc(4096, GFP_KERNEL);
1186 	if (!dp->rcv_buf)
1187 		goto out_free_dp;
1188 
1189 	dp->rcv_buf_len = 4096;
1190 
1191 	dp->ds_states = kmemdup(ds_states_template,
1192 				sizeof(ds_states_template), GFP_KERNEL);
1193 	if (!dp->ds_states)
1194 		goto out_free_rcv_buf;
1195 
1196 	dp->num_ds_states = ARRAY_SIZE(ds_states_template);
1197 
1198 	for (i = 0; i < dp->num_ds_states; i++)
1199 		dp->ds_states[i].handle = ((u64)i << 32);
1200 
1201 	ds_cfg.tx_irq = vdev->tx_irq;
1202 	ds_cfg.rx_irq = vdev->rx_irq;
1203 
1204 	lp = ldc_alloc(vdev->channel_id, &ds_cfg, dp, "DS");
1205 	if (IS_ERR(lp)) {
1206 		err = PTR_ERR(lp);
1207 		goto out_free_ds_states;
1208 	}
1209 	dp->lp = lp;
1210 
1211 	err = ldc_bind(lp);
1212 	if (err)
1213 		goto out_free_ldc;
1214 
1215 	spin_lock_irq(&ds_lock);
1216 	dp->next = ds_info_list;
1217 	ds_info_list = dp;
1218 	spin_unlock_irq(&ds_lock);
1219 
1220 	return err;
1221 
1222 out_free_ldc:
1223 	ldc_free(dp->lp);
1224 
1225 out_free_ds_states:
1226 	kfree(dp->ds_states);
1227 
1228 out_free_rcv_buf:
1229 	kfree(dp->rcv_buf);
1230 
1231 out_free_dp:
1232 	kfree(dp);
1233 
1234 out_err:
1235 	return err;
1236 }
1237 
1238 static int ds_remove(struct vio_dev *vdev)
1239 {
1240 	return 0;
1241 }
1242 
1243 static const struct vio_device_id ds_match[] = {
1244 	{
1245 		.type = "domain-services-port",
1246 	},
1247 	{},
1248 };
1249 
1250 static struct vio_driver ds_driver = {
1251 	.id_table	= ds_match,
1252 	.probe		= ds_probe,
1253 	.remove		= ds_remove,
1254 	.name		= "ds",
1255 };
1256 
1257 static int __init ds_init(void)
1258 {
1259 	unsigned long hv_ret, major, minor;
1260 
1261 	if (tlb_type == hypervisor) {
1262 		hv_ret = sun4v_get_version(HV_GRP_REBOOT_DATA, &major, &minor);
1263 		if (hv_ret == HV_EOK) {
1264 			pr_info("SUN4V: Reboot data supported (maj=%lu,min=%lu).\n",
1265 				major, minor);
1266 			reboot_data_supported = 1;
1267 		}
1268 	}
1269 	kthread_run(ds_thread, NULL, "kldomd");
1270 
1271 	return vio_register_driver(&ds_driver);
1272 }
1273 
1274 fs_initcall(ds_init);
1275