xref: /linux/arch/sparc/kernel/ds.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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/delay.h>
13 #include <linux/mutex.h>
14 #include <linux/kthread.h>
15 #include <linux/reboot.h>
16 #include <linux/cpu.h>
17 
18 #include <asm/hypervisor.h>
19 #include <asm/ldc.h>
20 #include <asm/vio.h>
21 #include <asm/mdesc.h>
22 #include <asm/head.h>
23 #include <asm/irq.h>
24 
25 #include "kernel.h"
26 
27 #define DRV_MODULE_NAME		"ds"
28 #define PFX DRV_MODULE_NAME	": "
29 #define DRV_MODULE_VERSION	"1.0"
30 #define DRV_MODULE_RELDATE	"Jul 11, 2007"
31 
32 static char version[] =
33 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
34 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
35 MODULE_DESCRIPTION("Sun LDOM domain services driver");
36 MODULE_LICENSE("GPL");
37 MODULE_VERSION(DRV_MODULE_VERSION);
38 
39 struct ds_msg_tag {
40 	__u32			type;
41 #define DS_INIT_REQ		0x00
42 #define DS_INIT_ACK		0x01
43 #define DS_INIT_NACK		0x02
44 #define DS_REG_REQ		0x03
45 #define DS_REG_ACK		0x04
46 #define DS_REG_NACK		0x05
47 #define DS_UNREG_REQ		0x06
48 #define DS_UNREG_ACK		0x07
49 #define DS_UNREG_NACK		0x08
50 #define DS_DATA			0x09
51 #define DS_NACK			0x0a
52 
53 	__u32			len;
54 };
55 
56 /* Result codes */
57 #define DS_OK			0x00
58 #define DS_REG_VER_NACK		0x01
59 #define DS_REG_DUP		0x02
60 #define DS_INV_HDL		0x03
61 #define DS_TYPE_UNKNOWN		0x04
62 
63 struct ds_version {
64 	__u16			major;
65 	__u16			minor;
66 };
67 
68 struct ds_ver_req {
69 	struct ds_msg_tag	tag;
70 	struct ds_version	ver;
71 };
72 
73 struct ds_ver_ack {
74 	struct ds_msg_tag	tag;
75 	__u16			minor;
76 };
77 
78 struct ds_ver_nack {
79 	struct ds_msg_tag	tag;
80 	__u16			major;
81 };
82 
83 struct ds_reg_req {
84 	struct ds_msg_tag	tag;
85 	__u64			handle;
86 	__u16			major;
87 	__u16			minor;
88 	char			svc_id[0];
89 };
90 
91 struct ds_reg_ack {
92 	struct ds_msg_tag	tag;
93 	__u64			handle;
94 	__u16			minor;
95 };
96 
97 struct ds_reg_nack {
98 	struct ds_msg_tag	tag;
99 	__u64			handle;
100 	__u16			major;
101 };
102 
103 struct ds_unreg_req {
104 	struct ds_msg_tag	tag;
105 	__u64			handle;
106 };
107 
108 struct ds_unreg_ack {
109 	struct ds_msg_tag	tag;
110 	__u64			handle;
111 };
112 
113 struct ds_unreg_nack {
114 	struct ds_msg_tag	tag;
115 	__u64			handle;
116 };
117 
118 struct ds_data {
119 	struct ds_msg_tag	tag;
120 	__u64			handle;
121 };
122 
123 struct ds_data_nack {
124 	struct ds_msg_tag	tag;
125 	__u64			handle;
126 	__u64			result;
127 };
128 
129 struct ds_info;
130 struct ds_cap_state {
131 	__u64			handle;
132 
133 	void			(*data)(struct ds_info *dp,
134 					struct ds_cap_state *cp,
135 					void *buf, int len);
136 
137 	const char		*service_id;
138 
139 	u8			state;
140 #define CAP_STATE_UNKNOWN	0x00
141 #define CAP_STATE_REG_SENT	0x01
142 #define CAP_STATE_REGISTERED	0x02
143 };
144 
145 static void md_update_data(struct ds_info *dp, struct ds_cap_state *cp,
146 			   void *buf, int len);
147 static void domain_shutdown_data(struct ds_info *dp,
148 				 struct ds_cap_state *cp,
149 				 void *buf, int len);
150 static void domain_panic_data(struct ds_info *dp,
151 			      struct ds_cap_state *cp,
152 			      void *buf, int len);
153 #ifdef CONFIG_HOTPLUG_CPU
154 static void dr_cpu_data(struct ds_info *dp,
155 			struct ds_cap_state *cp,
156 			void *buf, int len);
157 #endif
158 static void ds_pri_data(struct ds_info *dp,
159 			struct ds_cap_state *cp,
160 			void *buf, int len);
161 static void ds_var_data(struct ds_info *dp,
162 			struct ds_cap_state *cp,
163 			void *buf, int len);
164 
165 static struct ds_cap_state ds_states_template[] = {
166 	{
167 		.service_id	= "md-update",
168 		.data		= md_update_data,
169 	},
170 	{
171 		.service_id	= "domain-shutdown",
172 		.data		= domain_shutdown_data,
173 	},
174 	{
175 		.service_id	= "domain-panic",
176 		.data		= domain_panic_data,
177 	},
178 #ifdef CONFIG_HOTPLUG_CPU
179 	{
180 		.service_id	= "dr-cpu",
181 		.data		= dr_cpu_data,
182 	},
183 #endif
184 	{
185 		.service_id	= "pri",
186 		.data		= ds_pri_data,
187 	},
188 	{
189 		.service_id	= "var-config",
190 		.data		= ds_var_data,
191 	},
192 	{
193 		.service_id	= "var-config-backup",
194 		.data		= ds_var_data,
195 	},
196 };
197 
198 static DEFINE_SPINLOCK(ds_lock);
199 
200 struct ds_info {
201 	struct ldc_channel	*lp;
202 	u8			hs_state;
203 #define DS_HS_START		0x01
204 #define DS_HS_DONE		0x02
205 
206 	u64			id;
207 
208 	void			*rcv_buf;
209 	int			rcv_buf_len;
210 
211 	struct ds_cap_state	*ds_states;
212 	int			num_ds_states;
213 
214 	struct ds_info		*next;
215 };
216 
217 static struct ds_info *ds_info_list;
218 
219 static struct ds_cap_state *find_cap(struct ds_info *dp, u64 handle)
220 {
221 	unsigned int index = handle >> 32;
222 
223 	if (index >= dp->num_ds_states)
224 		return NULL;
225 	return &dp->ds_states[index];
226 }
227 
228 static struct ds_cap_state *find_cap_by_string(struct ds_info *dp,
229 					       const char *name)
230 {
231 	int i;
232 
233 	for (i = 0; i < dp->num_ds_states; i++) {
234 		if (strcmp(dp->ds_states[i].service_id, name))
235 			continue;
236 
237 		return &dp->ds_states[i];
238 	}
239 	return NULL;
240 }
241 
242 static int __ds_send(struct ldc_channel *lp, void *data, int len)
243 {
244 	int err, limit = 1000;
245 
246 	err = -EINVAL;
247 	while (limit-- > 0) {
248 		err = ldc_write(lp, data, len);
249 		if (!err || (err != -EAGAIN))
250 			break;
251 		udelay(1);
252 	}
253 
254 	return err;
255 }
256 
257 static int ds_send(struct ldc_channel *lp, void *data, int len)
258 {
259 	unsigned long flags;
260 	int err;
261 
262 	spin_lock_irqsave(&ds_lock, flags);
263 	err = __ds_send(lp, data, len);
264 	spin_unlock_irqrestore(&ds_lock, flags);
265 
266 	return err;
267 }
268 
269 struct ds_md_update_req {
270 	__u64				req_num;
271 };
272 
273 struct ds_md_update_res {
274 	__u64				req_num;
275 	__u32				result;
276 };
277 
278 static void md_update_data(struct ds_info *dp,
279 			   struct ds_cap_state *cp,
280 			   void *buf, int len)
281 {
282 	struct ldc_channel *lp = dp->lp;
283 	struct ds_data *dpkt = buf;
284 	struct ds_md_update_req *rp;
285 	struct {
286 		struct ds_data		data;
287 		struct ds_md_update_res	res;
288 	} pkt;
289 
290 	rp = (struct ds_md_update_req *) (dpkt + 1);
291 
292 	printk(KERN_INFO "ds-%llu: Machine description update.\n", dp->id);
293 
294 	mdesc_update();
295 
296 	memset(&pkt, 0, sizeof(pkt));
297 	pkt.data.tag.type = DS_DATA;
298 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
299 	pkt.data.handle = cp->handle;
300 	pkt.res.req_num = rp->req_num;
301 	pkt.res.result = DS_OK;
302 
303 	ds_send(lp, &pkt, sizeof(pkt));
304 }
305 
306 struct ds_shutdown_req {
307 	__u64				req_num;
308 	__u32				ms_delay;
309 };
310 
311 struct ds_shutdown_res {
312 	__u64				req_num;
313 	__u32				result;
314 	char				reason[1];
315 };
316 
317 static void domain_shutdown_data(struct ds_info *dp,
318 				 struct ds_cap_state *cp,
319 				 void *buf, int len)
320 {
321 	struct ldc_channel *lp = dp->lp;
322 	struct ds_data *dpkt = buf;
323 	struct ds_shutdown_req *rp;
324 	struct {
325 		struct ds_data		data;
326 		struct ds_shutdown_res	res;
327 	} pkt;
328 
329 	rp = (struct ds_shutdown_req *) (dpkt + 1);
330 
331 	printk(KERN_ALERT "ds-%llu: Shutdown request from "
332 	       "LDOM manager received.\n", dp->id);
333 
334 	memset(&pkt, 0, sizeof(pkt));
335 	pkt.data.tag.type = DS_DATA;
336 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
337 	pkt.data.handle = cp->handle;
338 	pkt.res.req_num = rp->req_num;
339 	pkt.res.result = DS_OK;
340 	pkt.res.reason[0] = 0;
341 
342 	ds_send(lp, &pkt, sizeof(pkt));
343 
344 	orderly_poweroff(true);
345 }
346 
347 struct ds_panic_req {
348 	__u64				req_num;
349 };
350 
351 struct ds_panic_res {
352 	__u64				req_num;
353 	__u32				result;
354 	char				reason[1];
355 };
356 
357 static void domain_panic_data(struct ds_info *dp,
358 			      struct ds_cap_state *cp,
359 			      void *buf, int len)
360 {
361 	struct ldc_channel *lp = dp->lp;
362 	struct ds_data *dpkt = buf;
363 	struct ds_panic_req *rp;
364 	struct {
365 		struct ds_data		data;
366 		struct ds_panic_res	res;
367 	} pkt;
368 
369 	rp = (struct ds_panic_req *) (dpkt + 1);
370 
371 	printk(KERN_ALERT "ds-%llu: Panic request from "
372 	       "LDOM manager received.\n", dp->id);
373 
374 	memset(&pkt, 0, sizeof(pkt));
375 	pkt.data.tag.type = DS_DATA;
376 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
377 	pkt.data.handle = cp->handle;
378 	pkt.res.req_num = rp->req_num;
379 	pkt.res.result = DS_OK;
380 	pkt.res.reason[0] = 0;
381 
382 	ds_send(lp, &pkt, sizeof(pkt));
383 
384 	panic("PANIC requested by LDOM manager.");
385 }
386 
387 #ifdef CONFIG_HOTPLUG_CPU
388 struct dr_cpu_tag {
389 	__u64				req_num;
390 	__u32				type;
391 #define DR_CPU_CONFIGURE		0x43
392 #define DR_CPU_UNCONFIGURE		0x55
393 #define DR_CPU_FORCE_UNCONFIGURE	0x46
394 #define DR_CPU_STATUS			0x53
395 
396 /* Responses */
397 #define DR_CPU_OK			0x6f
398 #define DR_CPU_ERROR			0x65
399 
400 	__u32				num_records;
401 };
402 
403 struct dr_cpu_resp_entry {
404 	__u32				cpu;
405 	__u32				result;
406 #define DR_CPU_RES_OK			0x00
407 #define DR_CPU_RES_FAILURE		0x01
408 #define DR_CPU_RES_BLOCKED		0x02
409 #define DR_CPU_RES_CPU_NOT_RESPONDING	0x03
410 #define DR_CPU_RES_NOT_IN_MD		0x04
411 
412 	__u32				stat;
413 #define DR_CPU_STAT_NOT_PRESENT		0x00
414 #define DR_CPU_STAT_UNCONFIGURED	0x01
415 #define DR_CPU_STAT_CONFIGURED		0x02
416 
417 	__u32				str_off;
418 };
419 
420 static void __dr_cpu_send_error(struct ds_info *dp,
421 				struct ds_cap_state *cp,
422 				struct ds_data *data)
423 {
424 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
425 	struct {
426 		struct ds_data		data;
427 		struct dr_cpu_tag	tag;
428 	} pkt;
429 	int msg_len;
430 
431 	memset(&pkt, 0, sizeof(pkt));
432 	pkt.data.tag.type = DS_DATA;
433 	pkt.data.handle = cp->handle;
434 	pkt.tag.req_num = tag->req_num;
435 	pkt.tag.type = DR_CPU_ERROR;
436 	pkt.tag.num_records = 0;
437 
438 	msg_len = (sizeof(struct ds_data) +
439 		   sizeof(struct dr_cpu_tag));
440 
441 	pkt.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
442 
443 	__ds_send(dp->lp, &pkt, msg_len);
444 }
445 
446 static void dr_cpu_send_error(struct ds_info *dp,
447 			      struct ds_cap_state *cp,
448 			      struct ds_data *data)
449 {
450 	unsigned long flags;
451 
452 	spin_lock_irqsave(&ds_lock, flags);
453 	__dr_cpu_send_error(dp, cp, data);
454 	spin_unlock_irqrestore(&ds_lock, flags);
455 }
456 
457 #define CPU_SENTINEL	0xffffffff
458 
459 static void purge_dups(u32 *list, u32 num_ents)
460 {
461 	unsigned int i;
462 
463 	for (i = 0; i < num_ents; i++) {
464 		u32 cpu = list[i];
465 		unsigned int j;
466 
467 		if (cpu == CPU_SENTINEL)
468 			continue;
469 
470 		for (j = i + 1; j < num_ents; j++) {
471 			if (list[j] == cpu)
472 				list[j] = CPU_SENTINEL;
473 		}
474 	}
475 }
476 
477 static int dr_cpu_size_response(int ncpus)
478 {
479 	return (sizeof(struct ds_data) +
480 		sizeof(struct dr_cpu_tag) +
481 		(sizeof(struct dr_cpu_resp_entry) * ncpus));
482 }
483 
484 static void dr_cpu_init_response(struct ds_data *resp, u64 req_num,
485 				 u64 handle, int resp_len, int ncpus,
486 				 cpumask_t *mask, u32 default_stat)
487 {
488 	struct dr_cpu_resp_entry *ent;
489 	struct dr_cpu_tag *tag;
490 	int i, cpu;
491 
492 	tag = (struct dr_cpu_tag *) (resp + 1);
493 	ent = (struct dr_cpu_resp_entry *) (tag + 1);
494 
495 	resp->tag.type = DS_DATA;
496 	resp->tag.len = resp_len - sizeof(struct ds_msg_tag);
497 	resp->handle = handle;
498 	tag->req_num = req_num;
499 	tag->type = DR_CPU_OK;
500 	tag->num_records = ncpus;
501 
502 	i = 0;
503 	for_each_cpu(cpu, mask) {
504 		ent[i].cpu = cpu;
505 		ent[i].result = DR_CPU_RES_OK;
506 		ent[i].stat = default_stat;
507 		i++;
508 	}
509 	BUG_ON(i != ncpus);
510 }
511 
512 static void dr_cpu_mark(struct ds_data *resp, int cpu, int ncpus,
513 			u32 res, u32 stat)
514 {
515 	struct dr_cpu_resp_entry *ent;
516 	struct dr_cpu_tag *tag;
517 	int i;
518 
519 	tag = (struct dr_cpu_tag *) (resp + 1);
520 	ent = (struct dr_cpu_resp_entry *) (tag + 1);
521 
522 	for (i = 0; i < ncpus; i++) {
523 		if (ent[i].cpu != cpu)
524 			continue;
525 		ent[i].result = res;
526 		ent[i].stat = stat;
527 		break;
528 	}
529 }
530 
531 static int dr_cpu_configure(struct ds_info *dp, struct ds_cap_state *cp,
532 			    u64 req_num, cpumask_t *mask)
533 {
534 	struct ds_data *resp;
535 	int resp_len, ncpus, cpu;
536 	unsigned long flags;
537 
538 	ncpus = cpumask_weight(mask);
539 	resp_len = dr_cpu_size_response(ncpus);
540 	resp = kzalloc(resp_len, GFP_KERNEL);
541 	if (!resp)
542 		return -ENOMEM;
543 
544 	dr_cpu_init_response(resp, req_num, cp->handle,
545 			     resp_len, ncpus, mask,
546 			     DR_CPU_STAT_CONFIGURED);
547 
548 	mdesc_populate_present_mask(mask);
549 	mdesc_fill_in_cpu_data(mask);
550 
551 	for_each_cpu(cpu, mask) {
552 		int err;
553 
554 		printk(KERN_INFO "ds-%llu: Starting cpu %d...\n",
555 		       dp->id, cpu);
556 		err = cpu_up(cpu);
557 		if (err) {
558 			__u32 res = DR_CPU_RES_FAILURE;
559 			__u32 stat = DR_CPU_STAT_UNCONFIGURED;
560 
561 			if (!cpu_present(cpu)) {
562 				/* CPU not present in MD */
563 				res = DR_CPU_RES_NOT_IN_MD;
564 				stat = DR_CPU_STAT_NOT_PRESENT;
565 			} else if (err == -ENODEV) {
566 				/* CPU did not call in successfully */
567 				res = DR_CPU_RES_CPU_NOT_RESPONDING;
568 			}
569 
570 			printk(KERN_INFO "ds-%llu: CPU startup failed err=%d\n",
571 			       dp->id, err);
572 			dr_cpu_mark(resp, cpu, ncpus, res, stat);
573 		}
574 	}
575 
576 	spin_lock_irqsave(&ds_lock, flags);
577 	__ds_send(dp->lp, resp, resp_len);
578 	spin_unlock_irqrestore(&ds_lock, flags);
579 
580 	kfree(resp);
581 
582 	/* Redistribute IRQs, taking into account the new cpus.  */
583 	fixup_irqs();
584 
585 	return 0;
586 }
587 
588 static int dr_cpu_unconfigure(struct ds_info *dp,
589 			      struct ds_cap_state *cp,
590 			      u64 req_num,
591 			      cpumask_t *mask)
592 {
593 	struct ds_data *resp;
594 	int resp_len, ncpus, cpu;
595 	unsigned long flags;
596 
597 	ncpus = cpumask_weight(mask);
598 	resp_len = dr_cpu_size_response(ncpus);
599 	resp = kzalloc(resp_len, GFP_KERNEL);
600 	if (!resp)
601 		return -ENOMEM;
602 
603 	dr_cpu_init_response(resp, req_num, cp->handle,
604 			     resp_len, ncpus, mask,
605 			     DR_CPU_STAT_UNCONFIGURED);
606 
607 	for_each_cpu(cpu, mask) {
608 		int err;
609 
610 		printk(KERN_INFO "ds-%llu: Shutting down cpu %d...\n",
611 		       dp->id, cpu);
612 		err = cpu_down(cpu);
613 		if (err)
614 			dr_cpu_mark(resp, cpu, ncpus,
615 				    DR_CPU_RES_FAILURE,
616 				    DR_CPU_STAT_CONFIGURED);
617 	}
618 
619 	spin_lock_irqsave(&ds_lock, flags);
620 	__ds_send(dp->lp, resp, resp_len);
621 	spin_unlock_irqrestore(&ds_lock, flags);
622 
623 	kfree(resp);
624 
625 	return 0;
626 }
627 
628 static void dr_cpu_data(struct ds_info *dp, struct ds_cap_state *cp, void *buf,
629 			int len)
630 {
631 	struct ds_data *data = buf;
632 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
633 	u32 *cpu_list = (u32 *) (tag + 1);
634 	u64 req_num = tag->req_num;
635 	cpumask_t mask;
636 	unsigned int i;
637 	int err;
638 
639 	switch (tag->type) {
640 	case DR_CPU_CONFIGURE:
641 	case DR_CPU_UNCONFIGURE:
642 	case DR_CPU_FORCE_UNCONFIGURE:
643 		break;
644 
645 	default:
646 		dr_cpu_send_error(dp, cp, data);
647 		return;
648 	}
649 
650 	purge_dups(cpu_list, tag->num_records);
651 
652 	cpumask_clear(&mask);
653 	for (i = 0; i < tag->num_records; i++) {
654 		if (cpu_list[i] == CPU_SENTINEL)
655 			continue;
656 
657 		if (cpu_list[i] < nr_cpu_ids)
658 			cpumask_set_cpu(cpu_list[i], &mask);
659 	}
660 
661 	if (tag->type == DR_CPU_CONFIGURE)
662 		err = dr_cpu_configure(dp, cp, req_num, &mask);
663 	else
664 		err = dr_cpu_unconfigure(dp, cp, req_num, &mask);
665 
666 	if (err)
667 		dr_cpu_send_error(dp, cp, data);
668 }
669 #endif /* CONFIG_HOTPLUG_CPU */
670 
671 struct ds_pri_msg {
672 	__u64				req_num;
673 	__u64				type;
674 #define DS_PRI_REQUEST			0x00
675 #define DS_PRI_DATA			0x01
676 #define DS_PRI_UPDATE			0x02
677 };
678 
679 static void ds_pri_data(struct ds_info *dp,
680 			struct ds_cap_state *cp,
681 			void *buf, int len)
682 {
683 	struct ds_data *dpkt = buf;
684 	struct ds_pri_msg *rp;
685 
686 	rp = (struct ds_pri_msg *) (dpkt + 1);
687 
688 	printk(KERN_INFO "ds-%llu: PRI REQ [%llx:%llx], len=%d\n",
689 	       dp->id, rp->req_num, rp->type, len);
690 }
691 
692 struct ds_var_hdr {
693 	__u32				type;
694 #define DS_VAR_SET_REQ			0x00
695 #define DS_VAR_DELETE_REQ		0x01
696 #define DS_VAR_SET_RESP			0x02
697 #define DS_VAR_DELETE_RESP		0x03
698 };
699 
700 struct ds_var_set_msg {
701 	struct ds_var_hdr		hdr;
702 	char				name_and_value[0];
703 };
704 
705 struct ds_var_delete_msg {
706 	struct ds_var_hdr		hdr;
707 	char				name[0];
708 };
709 
710 struct ds_var_resp {
711 	struct ds_var_hdr		hdr;
712 	__u32				result;
713 #define DS_VAR_SUCCESS			0x00
714 #define DS_VAR_NO_SPACE			0x01
715 #define DS_VAR_INVALID_VAR		0x02
716 #define DS_VAR_INVALID_VAL		0x03
717 #define DS_VAR_NOT_PRESENT		0x04
718 };
719 
720 static DEFINE_MUTEX(ds_var_mutex);
721 static int ds_var_doorbell;
722 static int ds_var_response;
723 
724 static void ds_var_data(struct ds_info *dp,
725 			struct ds_cap_state *cp,
726 			void *buf, int len)
727 {
728 	struct ds_data *dpkt = buf;
729 	struct ds_var_resp *rp;
730 
731 	rp = (struct ds_var_resp *) (dpkt + 1);
732 
733 	if (rp->hdr.type != DS_VAR_SET_RESP &&
734 	    rp->hdr.type != DS_VAR_DELETE_RESP)
735 		return;
736 
737 	ds_var_response = rp->result;
738 	wmb();
739 	ds_var_doorbell = 1;
740 }
741 
742 void ldom_set_var(const char *var, const char *value)
743 {
744 	struct ds_cap_state *cp;
745 	struct ds_info *dp;
746 	unsigned long flags;
747 
748 	spin_lock_irqsave(&ds_lock, flags);
749 	cp = NULL;
750 	for (dp = ds_info_list; dp; dp = dp->next) {
751 		struct ds_cap_state *tmp;
752 
753 		tmp = find_cap_by_string(dp, "var-config");
754 		if (tmp && tmp->state == CAP_STATE_REGISTERED) {
755 			cp = tmp;
756 			break;
757 		}
758 	}
759 	if (!cp) {
760 		for (dp = ds_info_list; dp; dp = dp->next) {
761 			struct ds_cap_state *tmp;
762 
763 			tmp = find_cap_by_string(dp, "var-config-backup");
764 			if (tmp && tmp->state == CAP_STATE_REGISTERED) {
765 				cp = tmp;
766 				break;
767 			}
768 		}
769 	}
770 	spin_unlock_irqrestore(&ds_lock, flags);
771 
772 	if (cp) {
773 		union {
774 			struct {
775 				struct ds_data		data;
776 				struct ds_var_set_msg	msg;
777 			} header;
778 			char			all[512];
779 		} pkt;
780 		char  *base, *p;
781 		int msg_len, loops;
782 
783 		if (strlen(var) + strlen(value) + 2 >
784 		    sizeof(pkt) - sizeof(pkt.header)) {
785 			printk(KERN_ERR PFX
786 				"contents length: %zu, which more than max: %lu,"
787 				"so could not set (%s) variable to (%s).\n",
788 				strlen(var) + strlen(value) + 2,
789 				sizeof(pkt) - sizeof(pkt.header), var, value);
790 			return;
791 		}
792 
793 		memset(&pkt, 0, sizeof(pkt));
794 		pkt.header.data.tag.type = DS_DATA;
795 		pkt.header.data.handle = cp->handle;
796 		pkt.header.msg.hdr.type = DS_VAR_SET_REQ;
797 		base = p = &pkt.header.msg.name_and_value[0];
798 		strcpy(p, var);
799 		p += strlen(var) + 1;
800 		strcpy(p, value);
801 		p += strlen(value) + 1;
802 
803 		msg_len = (sizeof(struct ds_data) +
804 			   sizeof(struct ds_var_set_msg) +
805 			   (p - base));
806 		msg_len = (msg_len + 3) & ~3;
807 		pkt.header.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
808 
809 		mutex_lock(&ds_var_mutex);
810 
811 		spin_lock_irqsave(&ds_lock, flags);
812 		ds_var_doorbell = 0;
813 		ds_var_response = -1;
814 
815 		__ds_send(dp->lp, &pkt, msg_len);
816 		spin_unlock_irqrestore(&ds_lock, flags);
817 
818 		loops = 1000;
819 		while (ds_var_doorbell == 0) {
820 			if (loops-- < 0)
821 				break;
822 			barrier();
823 			udelay(100);
824 		}
825 
826 		mutex_unlock(&ds_var_mutex);
827 
828 		if (ds_var_doorbell == 0 ||
829 		    ds_var_response != DS_VAR_SUCCESS)
830 			printk(KERN_ERR "ds-%llu: var-config [%s:%s] "
831 			       "failed, response(%d).\n",
832 			       dp->id, var, value,
833 			       ds_var_response);
834 	} else {
835 		printk(KERN_ERR PFX "var-config not registered so "
836 		       "could not set (%s) variable to (%s).\n",
837 		       var, value);
838 	}
839 }
840 
841 static char full_boot_str[256] __attribute__((aligned(32)));
842 static int reboot_data_supported;
843 
844 void ldom_reboot(const char *boot_command)
845 {
846 	/* Don't bother with any of this if the boot_command
847 	 * is empty.
848 	 */
849 	if (boot_command && strlen(boot_command)) {
850 		unsigned long len;
851 
852 		snprintf(full_boot_str, sizeof(full_boot_str), "boot %s",
853 			 boot_command);
854 		len = strlen(full_boot_str);
855 
856 		if (reboot_data_supported) {
857 			unsigned long ra = kimage_addr_to_ra(full_boot_str);
858 			unsigned long hv_ret;
859 
860 			hv_ret = sun4v_reboot_data_set(ra, len);
861 			if (hv_ret != HV_EOK)
862 				pr_err("SUN4V: Unable to set reboot data "
863 				       "hv_ret=%lu\n", hv_ret);
864 		} else {
865 			ldom_set_var("reboot-command", full_boot_str);
866 		}
867 	}
868 	sun4v_mach_sir();
869 }
870 
871 void ldom_power_off(void)
872 {
873 	sun4v_mach_exit(0);
874 }
875 
876 static void ds_conn_reset(struct ds_info *dp)
877 {
878 	printk(KERN_ERR "ds-%llu: ds_conn_reset() from %pf\n",
879 	       dp->id, __builtin_return_address(0));
880 }
881 
882 static int register_services(struct ds_info *dp)
883 {
884 	struct ldc_channel *lp = dp->lp;
885 	int i;
886 
887 	for (i = 0; i < dp->num_ds_states; i++) {
888 		struct {
889 			struct ds_reg_req req;
890 			u8 id_buf[256];
891 		} pbuf;
892 		struct ds_cap_state *cp = &dp->ds_states[i];
893 		int err, msg_len;
894 		u64 new_count;
895 
896 		if (cp->state == CAP_STATE_REGISTERED)
897 			continue;
898 
899 		new_count = sched_clock() & 0xffffffff;
900 		cp->handle = ((u64) i << 32) | new_count;
901 
902 		msg_len = (sizeof(struct ds_reg_req) +
903 			   strlen(cp->service_id));
904 
905 		memset(&pbuf, 0, sizeof(pbuf));
906 		pbuf.req.tag.type = DS_REG_REQ;
907 		pbuf.req.tag.len = (msg_len - sizeof(struct ds_msg_tag));
908 		pbuf.req.handle = cp->handle;
909 		pbuf.req.major = 1;
910 		pbuf.req.minor = 0;
911 		strcpy(pbuf.req.svc_id, cp->service_id);
912 
913 		err = __ds_send(lp, &pbuf, msg_len);
914 		if (err > 0)
915 			cp->state = CAP_STATE_REG_SENT;
916 	}
917 	return 0;
918 }
919 
920 static int ds_handshake(struct ds_info *dp, struct ds_msg_tag *pkt)
921 {
922 
923 	if (dp->hs_state == DS_HS_START) {
924 		if (pkt->type != DS_INIT_ACK)
925 			goto conn_reset;
926 
927 		dp->hs_state = DS_HS_DONE;
928 
929 		return register_services(dp);
930 	}
931 
932 	if (dp->hs_state != DS_HS_DONE)
933 		goto conn_reset;
934 
935 	if (pkt->type == DS_REG_ACK) {
936 		struct ds_reg_ack *ap = (struct ds_reg_ack *) pkt;
937 		struct ds_cap_state *cp = find_cap(dp, ap->handle);
938 
939 		if (!cp) {
940 			printk(KERN_ERR "ds-%llu: REG ACK for unknown "
941 			       "handle %llx\n", dp->id, ap->handle);
942 			return 0;
943 		}
944 		printk(KERN_INFO "ds-%llu: Registered %s service.\n",
945 		       dp->id, cp->service_id);
946 		cp->state = CAP_STATE_REGISTERED;
947 	} else if (pkt->type == DS_REG_NACK) {
948 		struct ds_reg_nack *np = (struct ds_reg_nack *) pkt;
949 		struct ds_cap_state *cp = find_cap(dp, np->handle);
950 
951 		if (!cp) {
952 			printk(KERN_ERR "ds-%llu: REG NACK for "
953 			       "unknown handle %llx\n",
954 			       dp->id, np->handle);
955 			return 0;
956 		}
957 		cp->state = CAP_STATE_UNKNOWN;
958 	}
959 
960 	return 0;
961 
962 conn_reset:
963 	ds_conn_reset(dp);
964 	return -ECONNRESET;
965 }
966 
967 static void __send_ds_nack(struct ds_info *dp, u64 handle)
968 {
969 	struct ds_data_nack nack = {
970 		.tag = {
971 			.type = DS_NACK,
972 			.len = (sizeof(struct ds_data_nack) -
973 				sizeof(struct ds_msg_tag)),
974 		},
975 		.handle = handle,
976 		.result = DS_INV_HDL,
977 	};
978 
979 	__ds_send(dp->lp, &nack, sizeof(nack));
980 }
981 
982 static LIST_HEAD(ds_work_list);
983 static DECLARE_WAIT_QUEUE_HEAD(ds_wait);
984 
985 struct ds_queue_entry {
986 	struct list_head		list;
987 	struct ds_info			*dp;
988 	int				req_len;
989 	int				__pad;
990 	u64				req[0];
991 };
992 
993 static void process_ds_work(void)
994 {
995 	struct ds_queue_entry *qp, *tmp;
996 	unsigned long flags;
997 	LIST_HEAD(todo);
998 
999 	spin_lock_irqsave(&ds_lock, flags);
1000 	list_splice_init(&ds_work_list, &todo);
1001 	spin_unlock_irqrestore(&ds_lock, flags);
1002 
1003 	list_for_each_entry_safe(qp, tmp, &todo, list) {
1004 		struct ds_data *dpkt = (struct ds_data *) qp->req;
1005 		struct ds_info *dp = qp->dp;
1006 		struct ds_cap_state *cp = find_cap(dp, dpkt->handle);
1007 		int req_len = qp->req_len;
1008 
1009 		if (!cp) {
1010 			printk(KERN_ERR "ds-%llu: Data for unknown "
1011 			       "handle %llu\n",
1012 			       dp->id, dpkt->handle);
1013 
1014 			spin_lock_irqsave(&ds_lock, flags);
1015 			__send_ds_nack(dp, dpkt->handle);
1016 			spin_unlock_irqrestore(&ds_lock, flags);
1017 		} else {
1018 			cp->data(dp, cp, dpkt, req_len);
1019 		}
1020 
1021 		list_del(&qp->list);
1022 		kfree(qp);
1023 	}
1024 }
1025 
1026 static int ds_thread(void *__unused)
1027 {
1028 	DEFINE_WAIT(wait);
1029 
1030 	while (1) {
1031 		prepare_to_wait(&ds_wait, &wait, TASK_INTERRUPTIBLE);
1032 		if (list_empty(&ds_work_list))
1033 			schedule();
1034 		finish_wait(&ds_wait, &wait);
1035 
1036 		if (kthread_should_stop())
1037 			break;
1038 
1039 		process_ds_work();
1040 	}
1041 
1042 	return 0;
1043 }
1044 
1045 static int ds_data(struct ds_info *dp, struct ds_msg_tag *pkt, int len)
1046 {
1047 	struct ds_data *dpkt = (struct ds_data *) pkt;
1048 	struct ds_queue_entry *qp;
1049 
1050 	qp = kmalloc(sizeof(struct ds_queue_entry) + len, GFP_ATOMIC);
1051 	if (!qp) {
1052 		__send_ds_nack(dp, dpkt->handle);
1053 	} else {
1054 		qp->dp = dp;
1055 		memcpy(&qp->req, pkt, len);
1056 		list_add_tail(&qp->list, &ds_work_list);
1057 		wake_up(&ds_wait);
1058 	}
1059 	return 0;
1060 }
1061 
1062 static void ds_up(struct ds_info *dp)
1063 {
1064 	struct ldc_channel *lp = dp->lp;
1065 	struct ds_ver_req req;
1066 	int err;
1067 
1068 	req.tag.type = DS_INIT_REQ;
1069 	req.tag.len = sizeof(req) - sizeof(struct ds_msg_tag);
1070 	req.ver.major = 1;
1071 	req.ver.minor = 0;
1072 
1073 	err = __ds_send(lp, &req, sizeof(req));
1074 	if (err > 0)
1075 		dp->hs_state = DS_HS_START;
1076 }
1077 
1078 static void ds_reset(struct ds_info *dp)
1079 {
1080 	int i;
1081 
1082 	dp->hs_state = 0;
1083 
1084 	for (i = 0; i < dp->num_ds_states; i++) {
1085 		struct ds_cap_state *cp = &dp->ds_states[i];
1086 
1087 		cp->state = CAP_STATE_UNKNOWN;
1088 	}
1089 }
1090 
1091 static void ds_event(void *arg, int event)
1092 {
1093 	struct ds_info *dp = arg;
1094 	struct ldc_channel *lp = dp->lp;
1095 	unsigned long flags;
1096 	int err;
1097 
1098 	spin_lock_irqsave(&ds_lock, flags);
1099 
1100 	if (event == LDC_EVENT_UP) {
1101 		ds_up(dp);
1102 		spin_unlock_irqrestore(&ds_lock, flags);
1103 		return;
1104 	}
1105 
1106 	if (event == LDC_EVENT_RESET) {
1107 		ds_reset(dp);
1108 		spin_unlock_irqrestore(&ds_lock, flags);
1109 		return;
1110 	}
1111 
1112 	if (event != LDC_EVENT_DATA_READY) {
1113 		printk(KERN_WARNING "ds-%llu: Unexpected LDC event %d\n",
1114 		       dp->id, event);
1115 		spin_unlock_irqrestore(&ds_lock, flags);
1116 		return;
1117 	}
1118 
1119 	err = 0;
1120 	while (1) {
1121 		struct ds_msg_tag *tag;
1122 
1123 		err = ldc_read(lp, dp->rcv_buf, sizeof(*tag));
1124 
1125 		if (unlikely(err < 0)) {
1126 			if (err == -ECONNRESET)
1127 				ds_conn_reset(dp);
1128 			break;
1129 		}
1130 		if (err == 0)
1131 			break;
1132 
1133 		tag = dp->rcv_buf;
1134 		err = ldc_read(lp, tag + 1, tag->len);
1135 
1136 		if (unlikely(err < 0)) {
1137 			if (err == -ECONNRESET)
1138 				ds_conn_reset(dp);
1139 			break;
1140 		}
1141 		if (err < tag->len)
1142 			break;
1143 
1144 		if (tag->type < DS_DATA)
1145 			err = ds_handshake(dp, dp->rcv_buf);
1146 		else
1147 			err = ds_data(dp, dp->rcv_buf,
1148 				      sizeof(*tag) + err);
1149 		if (err == -ECONNRESET)
1150 			break;
1151 	}
1152 
1153 	spin_unlock_irqrestore(&ds_lock, flags);
1154 }
1155 
1156 static int ds_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1157 {
1158 	static int ds_version_printed;
1159 	struct ldc_channel_config ds_cfg = {
1160 		.event		= ds_event,
1161 		.mtu		= 4096,
1162 		.mode		= LDC_MODE_STREAM,
1163 	};
1164 	struct mdesc_handle *hp;
1165 	struct ldc_channel *lp;
1166 	struct ds_info *dp;
1167 	const u64 *val;
1168 	int err, i;
1169 
1170 	if (ds_version_printed++ == 0)
1171 		printk(KERN_INFO "%s", version);
1172 
1173 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1174 	err = -ENOMEM;
1175 	if (!dp)
1176 		goto out_err;
1177 
1178 	hp = mdesc_grab();
1179 	val = mdesc_get_property(hp, vdev->mp, "id", NULL);
1180 	if (val)
1181 		dp->id = *val;
1182 	mdesc_release(hp);
1183 
1184 	dp->rcv_buf = kzalloc(4096, GFP_KERNEL);
1185 	if (!dp->rcv_buf)
1186 		goto out_free_dp;
1187 
1188 	dp->rcv_buf_len = 4096;
1189 
1190 	dp->ds_states = kmemdup(ds_states_template,
1191 				sizeof(ds_states_template), GFP_KERNEL);
1192 	if (!dp->ds_states)
1193 		goto out_free_rcv_buf;
1194 
1195 	dp->num_ds_states = ARRAY_SIZE(ds_states_template);
1196 
1197 	for (i = 0; i < dp->num_ds_states; i++)
1198 		dp->ds_states[i].handle = ((u64)i << 32);
1199 
1200 	ds_cfg.tx_irq = vdev->tx_irq;
1201 	ds_cfg.rx_irq = vdev->rx_irq;
1202 
1203 	lp = ldc_alloc(vdev->channel_id, &ds_cfg, dp, "DS");
1204 	if (IS_ERR(lp)) {
1205 		err = PTR_ERR(lp);
1206 		goto out_free_ds_states;
1207 	}
1208 	dp->lp = lp;
1209 
1210 	err = ldc_bind(lp);
1211 	if (err)
1212 		goto out_free_ldc;
1213 
1214 	spin_lock_irq(&ds_lock);
1215 	dp->next = ds_info_list;
1216 	ds_info_list = dp;
1217 	spin_unlock_irq(&ds_lock);
1218 
1219 	return err;
1220 
1221 out_free_ldc:
1222 	ldc_free(dp->lp);
1223 
1224 out_free_ds_states:
1225 	kfree(dp->ds_states);
1226 
1227 out_free_rcv_buf:
1228 	kfree(dp->rcv_buf);
1229 
1230 out_free_dp:
1231 	kfree(dp);
1232 
1233 out_err:
1234 	return err;
1235 }
1236 
1237 static int ds_remove(struct vio_dev *vdev)
1238 {
1239 	return 0;
1240 }
1241 
1242 static const struct vio_device_id ds_match[] = {
1243 	{
1244 		.type = "domain-services-port",
1245 	},
1246 	{},
1247 };
1248 
1249 static struct vio_driver ds_driver = {
1250 	.id_table	= ds_match,
1251 	.probe		= ds_probe,
1252 	.remove		= ds_remove,
1253 	.name		= "ds",
1254 };
1255 
1256 static int __init ds_init(void)
1257 {
1258 	unsigned long hv_ret, major, minor;
1259 
1260 	if (tlb_type == hypervisor) {
1261 		hv_ret = sun4v_get_version(HV_GRP_REBOOT_DATA, &major, &minor);
1262 		if (hv_ret == HV_EOK) {
1263 			pr_info("SUN4V: Reboot data supported (maj=%lu,min=%lu).\n",
1264 				major, minor);
1265 			reboot_data_supported = 1;
1266 		}
1267 	}
1268 	kthread_run(ds_thread, NULL, "kldomd");
1269 
1270 	return vio_register_driver(&ds_driver);
1271 }
1272 
1273 fs_initcall(ds_init);
1274