xref: /openbmc/linux/arch/sparc/kernel/ds.c (revision 95e9fd10)
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[] __devinitdata =
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 __cpuinit dr_cpu_configure(struct ds_info *dp,
532 				      struct ds_cap_state *cp,
533 				      u64 req_num,
534 				      cpumask_t *mask)
535 {
536 	struct ds_data *resp;
537 	int resp_len, ncpus, cpu;
538 	unsigned long flags;
539 
540 	ncpus = cpumask_weight(mask);
541 	resp_len = dr_cpu_size_response(ncpus);
542 	resp = kzalloc(resp_len, GFP_KERNEL);
543 	if (!resp)
544 		return -ENOMEM;
545 
546 	dr_cpu_init_response(resp, req_num, cp->handle,
547 			     resp_len, ncpus, mask,
548 			     DR_CPU_STAT_CONFIGURED);
549 
550 	mdesc_populate_present_mask(mask);
551 	mdesc_fill_in_cpu_data(mask);
552 
553 	for_each_cpu(cpu, mask) {
554 		int err;
555 
556 		printk(KERN_INFO "ds-%llu: Starting cpu %d...\n",
557 		       dp->id, cpu);
558 		err = cpu_up(cpu);
559 		if (err) {
560 			__u32 res = DR_CPU_RES_FAILURE;
561 			__u32 stat = DR_CPU_STAT_UNCONFIGURED;
562 
563 			if (!cpu_present(cpu)) {
564 				/* CPU not present in MD */
565 				res = DR_CPU_RES_NOT_IN_MD;
566 				stat = DR_CPU_STAT_NOT_PRESENT;
567 			} else if (err == -ENODEV) {
568 				/* CPU did not call in successfully */
569 				res = DR_CPU_RES_CPU_NOT_RESPONDING;
570 			}
571 
572 			printk(KERN_INFO "ds-%llu: CPU startup failed err=%d\n",
573 			       dp->id, err);
574 			dr_cpu_mark(resp, cpu, ncpus, res, stat);
575 		}
576 	}
577 
578 	spin_lock_irqsave(&ds_lock, flags);
579 	__ds_send(dp->lp, resp, resp_len);
580 	spin_unlock_irqrestore(&ds_lock, flags);
581 
582 	kfree(resp);
583 
584 	/* Redistribute IRQs, taking into account the new cpus.  */
585 	fixup_irqs();
586 
587 	return 0;
588 }
589 
590 static int dr_cpu_unconfigure(struct ds_info *dp,
591 			      struct ds_cap_state *cp,
592 			      u64 req_num,
593 			      cpumask_t *mask)
594 {
595 	struct ds_data *resp;
596 	int resp_len, ncpus, cpu;
597 	unsigned long flags;
598 
599 	ncpus = cpumask_weight(mask);
600 	resp_len = dr_cpu_size_response(ncpus);
601 	resp = kzalloc(resp_len, GFP_KERNEL);
602 	if (!resp)
603 		return -ENOMEM;
604 
605 	dr_cpu_init_response(resp, req_num, cp->handle,
606 			     resp_len, ncpus, mask,
607 			     DR_CPU_STAT_UNCONFIGURED);
608 
609 	for_each_cpu(cpu, mask) {
610 		int err;
611 
612 		printk(KERN_INFO "ds-%llu: Shutting down cpu %d...\n",
613 		       dp->id, cpu);
614 		err = cpu_down(cpu);
615 		if (err)
616 			dr_cpu_mark(resp, cpu, ncpus,
617 				    DR_CPU_RES_FAILURE,
618 				    DR_CPU_STAT_CONFIGURED);
619 	}
620 
621 	spin_lock_irqsave(&ds_lock, flags);
622 	__ds_send(dp->lp, resp, resp_len);
623 	spin_unlock_irqrestore(&ds_lock, flags);
624 
625 	kfree(resp);
626 
627 	return 0;
628 }
629 
630 static void __cpuinit dr_cpu_data(struct ds_info *dp,
631 				  struct ds_cap_state *cp,
632 				  void *buf, int len)
633 {
634 	struct ds_data *data = buf;
635 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
636 	u32 *cpu_list = (u32 *) (tag + 1);
637 	u64 req_num = tag->req_num;
638 	cpumask_t mask;
639 	unsigned int i;
640 	int err;
641 
642 	switch (tag->type) {
643 	case DR_CPU_CONFIGURE:
644 	case DR_CPU_UNCONFIGURE:
645 	case DR_CPU_FORCE_UNCONFIGURE:
646 		break;
647 
648 	default:
649 		dr_cpu_send_error(dp, cp, data);
650 		return;
651 	}
652 
653 	purge_dups(cpu_list, tag->num_records);
654 
655 	cpumask_clear(&mask);
656 	for (i = 0; i < tag->num_records; i++) {
657 		if (cpu_list[i] == CPU_SENTINEL)
658 			continue;
659 
660 		if (cpu_list[i] < nr_cpu_ids)
661 			cpumask_set_cpu(cpu_list[i], &mask);
662 	}
663 
664 	if (tag->type == DR_CPU_CONFIGURE)
665 		err = dr_cpu_configure(dp, cp, req_num, &mask);
666 	else
667 		err = dr_cpu_unconfigure(dp, cp, req_num, &mask);
668 
669 	if (err)
670 		dr_cpu_send_error(dp, cp, data);
671 }
672 #endif /* CONFIG_HOTPLUG_CPU */
673 
674 struct ds_pri_msg {
675 	__u64				req_num;
676 	__u64				type;
677 #define DS_PRI_REQUEST			0x00
678 #define DS_PRI_DATA			0x01
679 #define DS_PRI_UPDATE			0x02
680 };
681 
682 static void ds_pri_data(struct ds_info *dp,
683 			struct ds_cap_state *cp,
684 			void *buf, int len)
685 {
686 	struct ds_data *dpkt = buf;
687 	struct ds_pri_msg *rp;
688 
689 	rp = (struct ds_pri_msg *) (dpkt + 1);
690 
691 	printk(KERN_INFO "ds-%llu: PRI REQ [%llx:%llx], len=%d\n",
692 	       dp->id, rp->req_num, rp->type, len);
693 }
694 
695 struct ds_var_hdr {
696 	__u32				type;
697 #define DS_VAR_SET_REQ			0x00
698 #define DS_VAR_DELETE_REQ		0x01
699 #define DS_VAR_SET_RESP			0x02
700 #define DS_VAR_DELETE_RESP		0x03
701 };
702 
703 struct ds_var_set_msg {
704 	struct ds_var_hdr		hdr;
705 	char				name_and_value[0];
706 };
707 
708 struct ds_var_delete_msg {
709 	struct ds_var_hdr		hdr;
710 	char				name[0];
711 };
712 
713 struct ds_var_resp {
714 	struct ds_var_hdr		hdr;
715 	__u32				result;
716 #define DS_VAR_SUCCESS			0x00
717 #define DS_VAR_NO_SPACE			0x01
718 #define DS_VAR_INVALID_VAR		0x02
719 #define DS_VAR_INVALID_VAL		0x03
720 #define DS_VAR_NOT_PRESENT		0x04
721 };
722 
723 static DEFINE_MUTEX(ds_var_mutex);
724 static int ds_var_doorbell;
725 static int ds_var_response;
726 
727 static void ds_var_data(struct ds_info *dp,
728 			struct ds_cap_state *cp,
729 			void *buf, int len)
730 {
731 	struct ds_data *dpkt = buf;
732 	struct ds_var_resp *rp;
733 
734 	rp = (struct ds_var_resp *) (dpkt + 1);
735 
736 	if (rp->hdr.type != DS_VAR_SET_RESP &&
737 	    rp->hdr.type != DS_VAR_DELETE_RESP)
738 		return;
739 
740 	ds_var_response = rp->result;
741 	wmb();
742 	ds_var_doorbell = 1;
743 }
744 
745 void ldom_set_var(const char *var, const char *value)
746 {
747 	struct ds_cap_state *cp;
748 	struct ds_info *dp;
749 	unsigned long flags;
750 
751 	spin_lock_irqsave(&ds_lock, flags);
752 	cp = NULL;
753 	for (dp = ds_info_list; dp; dp = dp->next) {
754 		struct ds_cap_state *tmp;
755 
756 		tmp = find_cap_by_string(dp, "var-config");
757 		if (tmp && tmp->state == CAP_STATE_REGISTERED) {
758 			cp = tmp;
759 			break;
760 		}
761 	}
762 	if (!cp) {
763 		for (dp = ds_info_list; dp; dp = dp->next) {
764 			struct ds_cap_state *tmp;
765 
766 			tmp = find_cap_by_string(dp, "var-config-backup");
767 			if (tmp && tmp->state == CAP_STATE_REGISTERED) {
768 				cp = tmp;
769 				break;
770 			}
771 		}
772 	}
773 	spin_unlock_irqrestore(&ds_lock, flags);
774 
775 	if (cp) {
776 		union {
777 			struct {
778 				struct ds_data		data;
779 				struct ds_var_set_msg	msg;
780 			} header;
781 			char			all[512];
782 		} pkt;
783 		char  *base, *p;
784 		int msg_len, loops;
785 
786 		memset(&pkt, 0, sizeof(pkt));
787 		pkt.header.data.tag.type = DS_DATA;
788 		pkt.header.data.handle = cp->handle;
789 		pkt.header.msg.hdr.type = DS_VAR_SET_REQ;
790 		base = p = &pkt.header.msg.name_and_value[0];
791 		strcpy(p, var);
792 		p += strlen(var) + 1;
793 		strcpy(p, value);
794 		p += strlen(value) + 1;
795 
796 		msg_len = (sizeof(struct ds_data) +
797 			   sizeof(struct ds_var_set_msg) +
798 			   (p - base));
799 		msg_len = (msg_len + 3) & ~3;
800 		pkt.header.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
801 
802 		mutex_lock(&ds_var_mutex);
803 
804 		spin_lock_irqsave(&ds_lock, flags);
805 		ds_var_doorbell = 0;
806 		ds_var_response = -1;
807 
808 		__ds_send(dp->lp, &pkt, msg_len);
809 		spin_unlock_irqrestore(&ds_lock, flags);
810 
811 		loops = 1000;
812 		while (ds_var_doorbell == 0) {
813 			if (loops-- < 0)
814 				break;
815 			barrier();
816 			udelay(100);
817 		}
818 
819 		mutex_unlock(&ds_var_mutex);
820 
821 		if (ds_var_doorbell == 0 ||
822 		    ds_var_response != DS_VAR_SUCCESS)
823 			printk(KERN_ERR "ds-%llu: var-config [%s:%s] "
824 			       "failed, response(%d).\n",
825 			       dp->id, var, value,
826 			       ds_var_response);
827 	} else {
828 		printk(KERN_ERR PFX "var-config not registered so "
829 		       "could not set (%s) variable to (%s).\n",
830 		       var, value);
831 	}
832 }
833 
834 static char full_boot_str[256] __attribute__((aligned(32)));
835 static int reboot_data_supported;
836 
837 void ldom_reboot(const char *boot_command)
838 {
839 	/* Don't bother with any of this if the boot_command
840 	 * is empty.
841 	 */
842 	if (boot_command && strlen(boot_command)) {
843 		unsigned long len;
844 
845 		strcpy(full_boot_str, "boot ");
846 		strcpy(full_boot_str + strlen("boot "), boot_command);
847 		len = strlen(full_boot_str);
848 
849 		if (reboot_data_supported) {
850 			unsigned long ra = kimage_addr_to_ra(full_boot_str);
851 			unsigned long hv_ret;
852 
853 			hv_ret = sun4v_reboot_data_set(ra, len);
854 			if (hv_ret != HV_EOK)
855 				pr_err("SUN4V: Unable to set reboot data "
856 				       "hv_ret=%lu\n", hv_ret);
857 		} else {
858 			ldom_set_var("reboot-command", full_boot_str);
859 		}
860 	}
861 	sun4v_mach_sir();
862 }
863 
864 void ldom_power_off(void)
865 {
866 	sun4v_mach_exit(0);
867 }
868 
869 static void ds_conn_reset(struct ds_info *dp)
870 {
871 	printk(KERN_ERR "ds-%llu: ds_conn_reset() from %pf\n",
872 	       dp->id, __builtin_return_address(0));
873 }
874 
875 static int register_services(struct ds_info *dp)
876 {
877 	struct ldc_channel *lp = dp->lp;
878 	int i;
879 
880 	for (i = 0; i < dp->num_ds_states; i++) {
881 		struct {
882 			struct ds_reg_req req;
883 			u8 id_buf[256];
884 		} pbuf;
885 		struct ds_cap_state *cp = &dp->ds_states[i];
886 		int err, msg_len;
887 		u64 new_count;
888 
889 		if (cp->state == CAP_STATE_REGISTERED)
890 			continue;
891 
892 		new_count = sched_clock() & 0xffffffff;
893 		cp->handle = ((u64) i << 32) | new_count;
894 
895 		msg_len = (sizeof(struct ds_reg_req) +
896 			   strlen(cp->service_id));
897 
898 		memset(&pbuf, 0, sizeof(pbuf));
899 		pbuf.req.tag.type = DS_REG_REQ;
900 		pbuf.req.tag.len = (msg_len - sizeof(struct ds_msg_tag));
901 		pbuf.req.handle = cp->handle;
902 		pbuf.req.major = 1;
903 		pbuf.req.minor = 0;
904 		strcpy(pbuf.req.svc_id, cp->service_id);
905 
906 		err = __ds_send(lp, &pbuf, msg_len);
907 		if (err > 0)
908 			cp->state = CAP_STATE_REG_SENT;
909 	}
910 	return 0;
911 }
912 
913 static int ds_handshake(struct ds_info *dp, struct ds_msg_tag *pkt)
914 {
915 
916 	if (dp->hs_state == DS_HS_START) {
917 		if (pkt->type != DS_INIT_ACK)
918 			goto conn_reset;
919 
920 		dp->hs_state = DS_HS_DONE;
921 
922 		return register_services(dp);
923 	}
924 
925 	if (dp->hs_state != DS_HS_DONE)
926 		goto conn_reset;
927 
928 	if (pkt->type == DS_REG_ACK) {
929 		struct ds_reg_ack *ap = (struct ds_reg_ack *) pkt;
930 		struct ds_cap_state *cp = find_cap(dp, ap->handle);
931 
932 		if (!cp) {
933 			printk(KERN_ERR "ds-%llu: REG ACK for unknown "
934 			       "handle %llx\n", dp->id, ap->handle);
935 			return 0;
936 		}
937 		printk(KERN_INFO "ds-%llu: Registered %s service.\n",
938 		       dp->id, cp->service_id);
939 		cp->state = CAP_STATE_REGISTERED;
940 	} else if (pkt->type == DS_REG_NACK) {
941 		struct ds_reg_nack *np = (struct ds_reg_nack *) pkt;
942 		struct ds_cap_state *cp = find_cap(dp, np->handle);
943 
944 		if (!cp) {
945 			printk(KERN_ERR "ds-%llu: REG NACK for "
946 			       "unknown handle %llx\n",
947 			       dp->id, np->handle);
948 			return 0;
949 		}
950 		cp->state = CAP_STATE_UNKNOWN;
951 	}
952 
953 	return 0;
954 
955 conn_reset:
956 	ds_conn_reset(dp);
957 	return -ECONNRESET;
958 }
959 
960 static void __send_ds_nack(struct ds_info *dp, u64 handle)
961 {
962 	struct ds_data_nack nack = {
963 		.tag = {
964 			.type = DS_NACK,
965 			.len = (sizeof(struct ds_data_nack) -
966 				sizeof(struct ds_msg_tag)),
967 		},
968 		.handle = handle,
969 		.result = DS_INV_HDL,
970 	};
971 
972 	__ds_send(dp->lp, &nack, sizeof(nack));
973 }
974 
975 static LIST_HEAD(ds_work_list);
976 static DECLARE_WAIT_QUEUE_HEAD(ds_wait);
977 
978 struct ds_queue_entry {
979 	struct list_head		list;
980 	struct ds_info			*dp;
981 	int				req_len;
982 	int				__pad;
983 	u64				req[0];
984 };
985 
986 static void process_ds_work(void)
987 {
988 	struct ds_queue_entry *qp, *tmp;
989 	unsigned long flags;
990 	LIST_HEAD(todo);
991 
992 	spin_lock_irqsave(&ds_lock, flags);
993 	list_splice_init(&ds_work_list, &todo);
994 	spin_unlock_irqrestore(&ds_lock, flags);
995 
996 	list_for_each_entry_safe(qp, tmp, &todo, list) {
997 		struct ds_data *dpkt = (struct ds_data *) qp->req;
998 		struct ds_info *dp = qp->dp;
999 		struct ds_cap_state *cp = find_cap(dp, dpkt->handle);
1000 		int req_len = qp->req_len;
1001 
1002 		if (!cp) {
1003 			printk(KERN_ERR "ds-%llu: Data for unknown "
1004 			       "handle %llu\n",
1005 			       dp->id, dpkt->handle);
1006 
1007 			spin_lock_irqsave(&ds_lock, flags);
1008 			__send_ds_nack(dp, dpkt->handle);
1009 			spin_unlock_irqrestore(&ds_lock, flags);
1010 		} else {
1011 			cp->data(dp, cp, dpkt, req_len);
1012 		}
1013 
1014 		list_del(&qp->list);
1015 		kfree(qp);
1016 	}
1017 }
1018 
1019 static int ds_thread(void *__unused)
1020 {
1021 	DEFINE_WAIT(wait);
1022 
1023 	while (1) {
1024 		prepare_to_wait(&ds_wait, &wait, TASK_INTERRUPTIBLE);
1025 		if (list_empty(&ds_work_list))
1026 			schedule();
1027 		finish_wait(&ds_wait, &wait);
1028 
1029 		if (kthread_should_stop())
1030 			break;
1031 
1032 		process_ds_work();
1033 	}
1034 
1035 	return 0;
1036 }
1037 
1038 static int ds_data(struct ds_info *dp, struct ds_msg_tag *pkt, int len)
1039 {
1040 	struct ds_data *dpkt = (struct ds_data *) pkt;
1041 	struct ds_queue_entry *qp;
1042 
1043 	qp = kmalloc(sizeof(struct ds_queue_entry) + len, GFP_ATOMIC);
1044 	if (!qp) {
1045 		__send_ds_nack(dp, dpkt->handle);
1046 	} else {
1047 		qp->dp = dp;
1048 		memcpy(&qp->req, pkt, len);
1049 		list_add_tail(&qp->list, &ds_work_list);
1050 		wake_up(&ds_wait);
1051 	}
1052 	return 0;
1053 }
1054 
1055 static void ds_up(struct ds_info *dp)
1056 {
1057 	struct ldc_channel *lp = dp->lp;
1058 	struct ds_ver_req req;
1059 	int err;
1060 
1061 	req.tag.type = DS_INIT_REQ;
1062 	req.tag.len = sizeof(req) - sizeof(struct ds_msg_tag);
1063 	req.ver.major = 1;
1064 	req.ver.minor = 0;
1065 
1066 	err = __ds_send(lp, &req, sizeof(req));
1067 	if (err > 0)
1068 		dp->hs_state = DS_HS_START;
1069 }
1070 
1071 static void ds_reset(struct ds_info *dp)
1072 {
1073 	int i;
1074 
1075 	dp->hs_state = 0;
1076 
1077 	for (i = 0; i < dp->num_ds_states; i++) {
1078 		struct ds_cap_state *cp = &dp->ds_states[i];
1079 
1080 		cp->state = CAP_STATE_UNKNOWN;
1081 	}
1082 }
1083 
1084 static void ds_event(void *arg, int event)
1085 {
1086 	struct ds_info *dp = arg;
1087 	struct ldc_channel *lp = dp->lp;
1088 	unsigned long flags;
1089 	int err;
1090 
1091 	spin_lock_irqsave(&ds_lock, flags);
1092 
1093 	if (event == LDC_EVENT_UP) {
1094 		ds_up(dp);
1095 		spin_unlock_irqrestore(&ds_lock, flags);
1096 		return;
1097 	}
1098 
1099 	if (event == LDC_EVENT_RESET) {
1100 		ds_reset(dp);
1101 		spin_unlock_irqrestore(&ds_lock, flags);
1102 		return;
1103 	}
1104 
1105 	if (event != LDC_EVENT_DATA_READY) {
1106 		printk(KERN_WARNING "ds-%llu: Unexpected LDC event %d\n",
1107 		       dp->id, event);
1108 		spin_unlock_irqrestore(&ds_lock, flags);
1109 		return;
1110 	}
1111 
1112 	err = 0;
1113 	while (1) {
1114 		struct ds_msg_tag *tag;
1115 
1116 		err = ldc_read(lp, dp->rcv_buf, sizeof(*tag));
1117 
1118 		if (unlikely(err < 0)) {
1119 			if (err == -ECONNRESET)
1120 				ds_conn_reset(dp);
1121 			break;
1122 		}
1123 		if (err == 0)
1124 			break;
1125 
1126 		tag = dp->rcv_buf;
1127 		err = ldc_read(lp, tag + 1, tag->len);
1128 
1129 		if (unlikely(err < 0)) {
1130 			if (err == -ECONNRESET)
1131 				ds_conn_reset(dp);
1132 			break;
1133 		}
1134 		if (err < tag->len)
1135 			break;
1136 
1137 		if (tag->type < DS_DATA)
1138 			err = ds_handshake(dp, dp->rcv_buf);
1139 		else
1140 			err = ds_data(dp, dp->rcv_buf,
1141 				      sizeof(*tag) + err);
1142 		if (err == -ECONNRESET)
1143 			break;
1144 	}
1145 
1146 	spin_unlock_irqrestore(&ds_lock, flags);
1147 }
1148 
1149 static int __devinit ds_probe(struct vio_dev *vdev,
1150 			      const struct vio_device_id *id)
1151 {
1152 	static int ds_version_printed;
1153 	struct ldc_channel_config ds_cfg = {
1154 		.event		= ds_event,
1155 		.mtu		= 4096,
1156 		.mode		= LDC_MODE_STREAM,
1157 	};
1158 	struct mdesc_handle *hp;
1159 	struct ldc_channel *lp;
1160 	struct ds_info *dp;
1161 	const u64 *val;
1162 	int err, i;
1163 
1164 	if (ds_version_printed++ == 0)
1165 		printk(KERN_INFO "%s", version);
1166 
1167 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1168 	err = -ENOMEM;
1169 	if (!dp)
1170 		goto out_err;
1171 
1172 	hp = mdesc_grab();
1173 	val = mdesc_get_property(hp, vdev->mp, "id", NULL);
1174 	if (val)
1175 		dp->id = *val;
1176 	mdesc_release(hp);
1177 
1178 	dp->rcv_buf = kzalloc(4096, GFP_KERNEL);
1179 	if (!dp->rcv_buf)
1180 		goto out_free_dp;
1181 
1182 	dp->rcv_buf_len = 4096;
1183 
1184 	dp->ds_states = kmemdup(ds_states_template,
1185 				sizeof(ds_states_template), GFP_KERNEL);
1186 	if (!dp->ds_states)
1187 		goto out_free_rcv_buf;
1188 
1189 	dp->num_ds_states = ARRAY_SIZE(ds_states_template);
1190 
1191 	for (i = 0; i < dp->num_ds_states; i++)
1192 		dp->ds_states[i].handle = ((u64)i << 32);
1193 
1194 	ds_cfg.tx_irq = vdev->tx_irq;
1195 	ds_cfg.rx_irq = vdev->rx_irq;
1196 
1197 	lp = ldc_alloc(vdev->channel_id, &ds_cfg, dp);
1198 	if (IS_ERR(lp)) {
1199 		err = PTR_ERR(lp);
1200 		goto out_free_ds_states;
1201 	}
1202 	dp->lp = lp;
1203 
1204 	err = ldc_bind(lp, "DS");
1205 	if (err)
1206 		goto out_free_ldc;
1207 
1208 	spin_lock_irq(&ds_lock);
1209 	dp->next = ds_info_list;
1210 	ds_info_list = dp;
1211 	spin_unlock_irq(&ds_lock);
1212 
1213 	return err;
1214 
1215 out_free_ldc:
1216 	ldc_free(dp->lp);
1217 
1218 out_free_ds_states:
1219 	kfree(dp->ds_states);
1220 
1221 out_free_rcv_buf:
1222 	kfree(dp->rcv_buf);
1223 
1224 out_free_dp:
1225 	kfree(dp);
1226 
1227 out_err:
1228 	return err;
1229 }
1230 
1231 static int ds_remove(struct vio_dev *vdev)
1232 {
1233 	return 0;
1234 }
1235 
1236 static const struct vio_device_id ds_match[] = {
1237 	{
1238 		.type = "domain-services-port",
1239 	},
1240 	{},
1241 };
1242 
1243 static struct vio_driver ds_driver = {
1244 	.id_table	= ds_match,
1245 	.probe		= ds_probe,
1246 	.remove		= ds_remove,
1247 	.name		= "ds",
1248 };
1249 
1250 static int __init ds_init(void)
1251 {
1252 	unsigned long hv_ret, major, minor;
1253 
1254 	if (tlb_type == hypervisor) {
1255 		hv_ret = sun4v_get_version(HV_GRP_REBOOT_DATA, &major, &minor);
1256 		if (hv_ret == HV_EOK) {
1257 			pr_info("SUN4V: Reboot data supported (maj=%lu,min=%lu).\n",
1258 				major, minor);
1259 			reboot_data_supported = 1;
1260 		}
1261 	}
1262 	kthread_run(ds_thread, NULL, "kldomd");
1263 
1264 	return vio_register_driver(&ds_driver);
1265 }
1266 
1267 fs_initcall(ds_init);
1268