xref: /linux/arch/sparc/kernel/ds.c (revision 39f75da7bcc829ddc4d40bb60d0e95520de7898b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* ds.c: Domain Services driver for Logical Domains
3  *
4  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
5  */
6 
7 #include <linux/kernel.h>
8 #include <linux/module.h>
9 #include <linux/types.h>
10 #include <linux/string.h>
11 #include <linux/slab.h>
12 #include <linux/sched.h>
13 #include <linux/sched/clock.h>
14 #include <linux/delay.h>
15 #include <linux/mutex.h>
16 #include <linux/kthread.h>
17 #include <linux/reboot.h>
18 #include <linux/cpu.h>
19 
20 #include <asm/hypervisor.h>
21 #include <asm/ldc.h>
22 #include <asm/vio.h>
23 #include <asm/mdesc.h>
24 #include <asm/head.h>
25 #include <asm/irq.h>
26 
27 #include "kernel.h"
28 
29 #define DRV_MODULE_NAME		"ds"
30 #define PFX DRV_MODULE_NAME	": "
31 #define DRV_MODULE_VERSION	"1.0"
32 #define DRV_MODULE_RELDATE	"Jul 11, 2007"
33 
34 static char version[] =
35 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
36 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
37 MODULE_DESCRIPTION("Sun LDOM domain services driver");
38 MODULE_LICENSE("GPL");
39 MODULE_VERSION(DRV_MODULE_VERSION);
40 
41 struct ds_msg_tag {
42 	__u32			type;
43 #define DS_INIT_REQ		0x00
44 #define DS_INIT_ACK		0x01
45 #define DS_INIT_NACK		0x02
46 #define DS_REG_REQ		0x03
47 #define DS_REG_ACK		0x04
48 #define DS_REG_NACK		0x05
49 #define DS_UNREG_REQ		0x06
50 #define DS_UNREG_ACK		0x07
51 #define DS_UNREG_NACK		0x08
52 #define DS_DATA			0x09
53 #define DS_NACK			0x0a
54 
55 	__u32			len;
56 };
57 
58 /* Result codes */
59 #define DS_OK			0x00
60 #define DS_REG_VER_NACK		0x01
61 #define DS_REG_DUP		0x02
62 #define DS_INV_HDL		0x03
63 #define DS_TYPE_UNKNOWN		0x04
64 
65 struct ds_version {
66 	__u16			major;
67 	__u16			minor;
68 };
69 
70 struct ds_ver_req {
71 	struct ds_msg_tag	tag;
72 	struct ds_version	ver;
73 };
74 
75 struct ds_ver_ack {
76 	struct ds_msg_tag	tag;
77 	__u16			minor;
78 };
79 
80 struct ds_ver_nack {
81 	struct ds_msg_tag	tag;
82 	__u16			major;
83 };
84 
85 struct ds_reg_req {
86 	struct ds_msg_tag	tag;
87 	__u64			handle;
88 	__u16			major;
89 	__u16			minor;
90 	char			svc_id[];
91 };
92 
93 struct ds_reg_ack {
94 	struct ds_msg_tag	tag;
95 	__u64			handle;
96 	__u16			minor;
97 };
98 
99 struct ds_reg_nack {
100 	struct ds_msg_tag	tag;
101 	__u64			handle;
102 	__u16			major;
103 };
104 
105 struct ds_unreg_req {
106 	struct ds_msg_tag	tag;
107 	__u64			handle;
108 };
109 
110 struct ds_unreg_ack {
111 	struct ds_msg_tag	tag;
112 	__u64			handle;
113 };
114 
115 struct ds_unreg_nack {
116 	struct ds_msg_tag	tag;
117 	__u64			handle;
118 };
119 
120 struct ds_data {
121 	struct ds_msg_tag	tag;
122 	__u64			handle;
123 };
124 
125 struct ds_data_nack {
126 	struct ds_msg_tag	tag;
127 	__u64			handle;
128 	__u64			result;
129 };
130 
131 struct ds_info;
132 struct ds_cap_state {
133 	__u64			handle;
134 
135 	void			(*data)(struct ds_info *dp,
136 					struct ds_cap_state *cp,
137 					void *buf, int len);
138 
139 	const char		*service_id;
140 
141 	u8			state;
142 #define CAP_STATE_UNKNOWN	0x00
143 #define CAP_STATE_REG_SENT	0x01
144 #define CAP_STATE_REGISTERED	0x02
145 };
146 
147 static void md_update_data(struct ds_info *dp, struct ds_cap_state *cp,
148 			   void *buf, int len);
149 static void domain_shutdown_data(struct ds_info *dp,
150 				 struct ds_cap_state *cp,
151 				 void *buf, int len);
152 static void domain_panic_data(struct ds_info *dp,
153 			      struct ds_cap_state *cp,
154 			      void *buf, int len);
155 #ifdef CONFIG_HOTPLUG_CPU
156 static void dr_cpu_data(struct ds_info *dp,
157 			struct ds_cap_state *cp,
158 			void *buf, int len);
159 #endif
160 static void ds_pri_data(struct ds_info *dp,
161 			struct ds_cap_state *cp,
162 			void *buf, int len);
163 static void ds_var_data(struct ds_info *dp,
164 			struct ds_cap_state *cp,
165 			void *buf, int len);
166 
167 static struct ds_cap_state ds_states_template[] = {
168 	{
169 		.service_id	= "md-update",
170 		.data		= md_update_data,
171 	},
172 	{
173 		.service_id	= "domain-shutdown",
174 		.data		= domain_shutdown_data,
175 	},
176 	{
177 		.service_id	= "domain-panic",
178 		.data		= domain_panic_data,
179 	},
180 #ifdef CONFIG_HOTPLUG_CPU
181 	{
182 		.service_id	= "dr-cpu",
183 		.data		= dr_cpu_data,
184 	},
185 #endif
186 	{
187 		.service_id	= "pri",
188 		.data		= ds_pri_data,
189 	},
190 	{
191 		.service_id	= "var-config",
192 		.data		= ds_var_data,
193 	},
194 	{
195 		.service_id	= "var-config-backup",
196 		.data		= ds_var_data,
197 	},
198 };
199 
200 static DEFINE_SPINLOCK(ds_lock);
201 
202 struct ds_info {
203 	struct ldc_channel	*lp;
204 	u8			hs_state;
205 #define DS_HS_START		0x01
206 #define DS_HS_DONE		0x02
207 
208 	u64			id;
209 
210 	void			*rcv_buf;
211 	int			rcv_buf_len;
212 
213 	struct ds_cap_state	*ds_states;
214 	int			num_ds_states;
215 
216 	struct ds_info		*next;
217 };
218 
219 static struct ds_info *ds_info_list;
220 
221 static struct ds_cap_state *find_cap(struct ds_info *dp, u64 handle)
222 {
223 	unsigned int index = handle >> 32;
224 
225 	if (index >= dp->num_ds_states)
226 		return NULL;
227 	return &dp->ds_states[index];
228 }
229 
230 static struct ds_cap_state *find_cap_by_string(struct ds_info *dp,
231 					       const char *name)
232 {
233 	int i;
234 
235 	for (i = 0; i < dp->num_ds_states; i++) {
236 		if (strcmp(dp->ds_states[i].service_id, name))
237 			continue;
238 
239 		return &dp->ds_states[i];
240 	}
241 	return NULL;
242 }
243 
244 static int __ds_send(struct ldc_channel *lp, void *data, int len)
245 {
246 	int err, limit = 1000;
247 
248 	err = -EINVAL;
249 	while (limit-- > 0) {
250 		err = ldc_write(lp, data, len);
251 		if (!err || (err != -EAGAIN))
252 			break;
253 		udelay(1);
254 	}
255 
256 	return err;
257 }
258 
259 static int ds_send(struct ldc_channel *lp, void *data, int len)
260 {
261 	unsigned long flags;
262 	int err;
263 
264 	spin_lock_irqsave(&ds_lock, flags);
265 	err = __ds_send(lp, data, len);
266 	spin_unlock_irqrestore(&ds_lock, flags);
267 
268 	return err;
269 }
270 
271 struct ds_md_update_req {
272 	__u64				req_num;
273 };
274 
275 struct ds_md_update_res {
276 	__u64				req_num;
277 	__u32				result;
278 };
279 
280 static void md_update_data(struct ds_info *dp,
281 			   struct ds_cap_state *cp,
282 			   void *buf, int len)
283 {
284 	struct ldc_channel *lp = dp->lp;
285 	struct ds_data *dpkt = buf;
286 	struct ds_md_update_req *rp;
287 	struct {
288 		struct ds_data		data;
289 		struct ds_md_update_res	res;
290 	} pkt;
291 
292 	rp = (struct ds_md_update_req *) (dpkt + 1);
293 
294 	printk(KERN_INFO "ds-%llu: Machine description update.\n", dp->id);
295 
296 	mdesc_update();
297 
298 	memset(&pkt, 0, sizeof(pkt));
299 	pkt.data.tag.type = DS_DATA;
300 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
301 	pkt.data.handle = cp->handle;
302 	pkt.res.req_num = rp->req_num;
303 	pkt.res.result = DS_OK;
304 
305 	ds_send(lp, &pkt, sizeof(pkt));
306 }
307 
308 struct ds_shutdown_req {
309 	__u64				req_num;
310 	__u32				ms_delay;
311 };
312 
313 struct ds_shutdown_res {
314 	__u64				req_num;
315 	__u32				result;
316 	char				reason[1];
317 };
318 
319 static void domain_shutdown_data(struct ds_info *dp,
320 				 struct ds_cap_state *cp,
321 				 void *buf, int len)
322 {
323 	struct ldc_channel *lp = dp->lp;
324 	struct ds_data *dpkt = buf;
325 	struct ds_shutdown_req *rp;
326 	struct {
327 		struct ds_data		data;
328 		struct ds_shutdown_res	res;
329 	} pkt;
330 
331 	rp = (struct ds_shutdown_req *) (dpkt + 1);
332 
333 	printk(KERN_ALERT "ds-%llu: Shutdown request from "
334 	       "LDOM manager received.\n", dp->id);
335 
336 	memset(&pkt, 0, sizeof(pkt));
337 	pkt.data.tag.type = DS_DATA;
338 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
339 	pkt.data.handle = cp->handle;
340 	pkt.res.req_num = rp->req_num;
341 	pkt.res.result = DS_OK;
342 	pkt.res.reason[0] = 0;
343 
344 	ds_send(lp, &pkt, sizeof(pkt));
345 
346 	orderly_poweroff(true);
347 }
348 
349 struct ds_panic_req {
350 	__u64				req_num;
351 };
352 
353 struct ds_panic_res {
354 	__u64				req_num;
355 	__u32				result;
356 	char				reason[1];
357 };
358 
359 static void domain_panic_data(struct ds_info *dp,
360 			      struct ds_cap_state *cp,
361 			      void *buf, int len)
362 {
363 	struct ldc_channel *lp = dp->lp;
364 	struct ds_data *dpkt = buf;
365 	struct ds_panic_req *rp;
366 	struct {
367 		struct ds_data		data;
368 		struct ds_panic_res	res;
369 	} pkt;
370 
371 	rp = (struct ds_panic_req *) (dpkt + 1);
372 
373 	printk(KERN_ALERT "ds-%llu: Panic request from "
374 	       "LDOM manager received.\n", dp->id);
375 
376 	memset(&pkt, 0, sizeof(pkt));
377 	pkt.data.tag.type = DS_DATA;
378 	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
379 	pkt.data.handle = cp->handle;
380 	pkt.res.req_num = rp->req_num;
381 	pkt.res.result = DS_OK;
382 	pkt.res.reason[0] = 0;
383 
384 	ds_send(lp, &pkt, sizeof(pkt));
385 
386 	panic("PANIC requested by LDOM manager.");
387 }
388 
389 #ifdef CONFIG_HOTPLUG_CPU
390 struct dr_cpu_tag {
391 	__u64				req_num;
392 	__u32				type;
393 #define DR_CPU_CONFIGURE		0x43
394 #define DR_CPU_UNCONFIGURE		0x55
395 #define DR_CPU_FORCE_UNCONFIGURE	0x46
396 #define DR_CPU_STATUS			0x53
397 
398 /* Responses */
399 #define DR_CPU_OK			0x6f
400 #define DR_CPU_ERROR			0x65
401 
402 	__u32				num_records;
403 };
404 
405 struct dr_cpu_resp_entry {
406 	__u32				cpu;
407 	__u32				result;
408 #define DR_CPU_RES_OK			0x00
409 #define DR_CPU_RES_FAILURE		0x01
410 #define DR_CPU_RES_BLOCKED		0x02
411 #define DR_CPU_RES_CPU_NOT_RESPONDING	0x03
412 #define DR_CPU_RES_NOT_IN_MD		0x04
413 
414 	__u32				stat;
415 #define DR_CPU_STAT_NOT_PRESENT		0x00
416 #define DR_CPU_STAT_UNCONFIGURED	0x01
417 #define DR_CPU_STAT_CONFIGURED		0x02
418 
419 	__u32				str_off;
420 };
421 
422 static void __dr_cpu_send_error(struct ds_info *dp,
423 				struct ds_cap_state *cp,
424 				struct ds_data *data)
425 {
426 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
427 	struct {
428 		struct ds_data		data;
429 		struct dr_cpu_tag	tag;
430 	} pkt;
431 	int msg_len;
432 
433 	memset(&pkt, 0, sizeof(pkt));
434 	pkt.data.tag.type = DS_DATA;
435 	pkt.data.handle = cp->handle;
436 	pkt.tag.req_num = tag->req_num;
437 	pkt.tag.type = DR_CPU_ERROR;
438 	pkt.tag.num_records = 0;
439 
440 	msg_len = (sizeof(struct ds_data) +
441 		   sizeof(struct dr_cpu_tag));
442 
443 	pkt.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
444 
445 	__ds_send(dp->lp, &pkt, msg_len);
446 }
447 
448 static void dr_cpu_send_error(struct ds_info *dp,
449 			      struct ds_cap_state *cp,
450 			      struct ds_data *data)
451 {
452 	unsigned long flags;
453 
454 	spin_lock_irqsave(&ds_lock, flags);
455 	__dr_cpu_send_error(dp, cp, data);
456 	spin_unlock_irqrestore(&ds_lock, flags);
457 }
458 
459 #define CPU_SENTINEL	0xffffffff
460 
461 static void purge_dups(u32 *list, u32 num_ents)
462 {
463 	unsigned int i;
464 
465 	for (i = 0; i < num_ents; i++) {
466 		u32 cpu = list[i];
467 		unsigned int j;
468 
469 		if (cpu == CPU_SENTINEL)
470 			continue;
471 
472 		for (j = i + 1; j < num_ents; j++) {
473 			if (list[j] == cpu)
474 				list[j] = CPU_SENTINEL;
475 		}
476 	}
477 }
478 
479 static int dr_cpu_size_response(int ncpus)
480 {
481 	return (sizeof(struct ds_data) +
482 		sizeof(struct dr_cpu_tag) +
483 		(sizeof(struct dr_cpu_resp_entry) * ncpus));
484 }
485 
486 static void dr_cpu_init_response(struct ds_data *resp, u64 req_num,
487 				 u64 handle, int resp_len, int ncpus,
488 				 cpumask_t *mask, u32 default_stat)
489 {
490 	struct dr_cpu_resp_entry *ent;
491 	struct dr_cpu_tag *tag;
492 	int i, cpu;
493 
494 	tag = (struct dr_cpu_tag *) (resp + 1);
495 	ent = (struct dr_cpu_resp_entry *) (tag + 1);
496 
497 	resp->tag.type = DS_DATA;
498 	resp->tag.len = resp_len - sizeof(struct ds_msg_tag);
499 	resp->handle = handle;
500 	tag->req_num = req_num;
501 	tag->type = DR_CPU_OK;
502 	tag->num_records = ncpus;
503 
504 	i = 0;
505 	for_each_cpu(cpu, mask) {
506 		ent[i].cpu = cpu;
507 		ent[i].result = DR_CPU_RES_OK;
508 		ent[i].stat = default_stat;
509 		i++;
510 	}
511 	BUG_ON(i != ncpus);
512 }
513 
514 static void dr_cpu_mark(struct ds_data *resp, int cpu, int ncpus,
515 			u32 res, u32 stat)
516 {
517 	struct dr_cpu_resp_entry *ent;
518 	struct dr_cpu_tag *tag;
519 	int i;
520 
521 	tag = (struct dr_cpu_tag *) (resp + 1);
522 	ent = (struct dr_cpu_resp_entry *) (tag + 1);
523 
524 	for (i = 0; i < ncpus; i++) {
525 		if (ent[i].cpu != cpu)
526 			continue;
527 		ent[i].result = res;
528 		ent[i].stat = stat;
529 		break;
530 	}
531 }
532 
533 static int dr_cpu_configure(struct ds_info *dp, struct ds_cap_state *cp,
534 			    u64 req_num, 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 = add_cpu(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 = remove_cpu(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 dr_cpu_data(struct ds_info *dp, struct ds_cap_state *cp, void *buf,
631 			int len)
632 {
633 	struct ds_data *data = buf;
634 	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
635 	u32 *cpu_list = (u32 *) (tag + 1);
636 	u64 req_num = tag->req_num;
637 	cpumask_t mask;
638 	unsigned int i;
639 	int err;
640 
641 	switch (tag->type) {
642 	case DR_CPU_CONFIGURE:
643 	case DR_CPU_UNCONFIGURE:
644 	case DR_CPU_FORCE_UNCONFIGURE:
645 		break;
646 
647 	default:
648 		dr_cpu_send_error(dp, cp, data);
649 		return;
650 	}
651 
652 	purge_dups(cpu_list, tag->num_records);
653 
654 	cpumask_clear(&mask);
655 	for (i = 0; i < tag->num_records; i++) {
656 		if (cpu_list[i] == CPU_SENTINEL)
657 			continue;
658 
659 		if (cpu_list[i] < nr_cpu_ids)
660 			cpumask_set_cpu(cpu_list[i], &mask);
661 	}
662 
663 	if (tag->type == DR_CPU_CONFIGURE)
664 		err = dr_cpu_configure(dp, cp, req_num, &mask);
665 	else
666 		err = dr_cpu_unconfigure(dp, cp, req_num, &mask);
667 
668 	if (err)
669 		dr_cpu_send_error(dp, cp, data);
670 }
671 #endif /* CONFIG_HOTPLUG_CPU */
672 
673 struct ds_pri_msg {
674 	__u64				req_num;
675 	__u64				type;
676 #define DS_PRI_REQUEST			0x00
677 #define DS_PRI_DATA			0x01
678 #define DS_PRI_UPDATE			0x02
679 };
680 
681 static void ds_pri_data(struct ds_info *dp,
682 			struct ds_cap_state *cp,
683 			void *buf, int len)
684 {
685 	struct ds_data *dpkt = buf;
686 	struct ds_pri_msg *rp;
687 
688 	rp = (struct ds_pri_msg *) (dpkt + 1);
689 
690 	printk(KERN_INFO "ds-%llu: PRI REQ [%llx:%llx], len=%d\n",
691 	       dp->id, rp->req_num, rp->type, len);
692 }
693 
694 struct ds_var_hdr {
695 	__u32				type;
696 #define DS_VAR_SET_REQ			0x00
697 #define DS_VAR_DELETE_REQ		0x01
698 #define DS_VAR_SET_RESP			0x02
699 #define DS_VAR_DELETE_RESP		0x03
700 };
701 
702 struct ds_var_set_msg {
703 	struct ds_var_hdr		hdr;
704 	char				name_and_value[];
705 };
706 
707 struct ds_var_delete_msg {
708 	struct ds_var_hdr		hdr;
709 	char				name[];
710 };
711 
712 struct ds_var_resp {
713 	struct ds_var_hdr		hdr;
714 	__u32				result;
715 #define DS_VAR_SUCCESS			0x00
716 #define DS_VAR_NO_SPACE			0x01
717 #define DS_VAR_INVALID_VAR		0x02
718 #define DS_VAR_INVALID_VAL		0x03
719 #define DS_VAR_NOT_PRESENT		0x04
720 };
721 
722 static DEFINE_MUTEX(ds_var_mutex);
723 static int ds_var_doorbell;
724 static int ds_var_response;
725 
726 static void ds_var_data(struct ds_info *dp,
727 			struct ds_cap_state *cp,
728 			void *buf, int len)
729 {
730 	struct ds_data *dpkt = buf;
731 	struct ds_var_resp *rp;
732 
733 	rp = (struct ds_var_resp *) (dpkt + 1);
734 
735 	if (rp->hdr.type != DS_VAR_SET_RESP &&
736 	    rp->hdr.type != DS_VAR_DELETE_RESP)
737 		return;
738 
739 	ds_var_response = rp->result;
740 	wmb();
741 	ds_var_doorbell = 1;
742 }
743 
744 void ldom_set_var(const char *var, const char *value)
745 {
746 	struct ds_cap_state *cp;
747 	struct ds_info *dp;
748 	unsigned long flags;
749 
750 	spin_lock_irqsave(&ds_lock, flags);
751 	cp = NULL;
752 	for (dp = ds_info_list; dp; dp = dp->next) {
753 		struct ds_cap_state *tmp;
754 
755 		tmp = find_cap_by_string(dp, "var-config");
756 		if (tmp && tmp->state == CAP_STATE_REGISTERED) {
757 			cp = tmp;
758 			break;
759 		}
760 	}
761 	if (!cp) {
762 		for (dp = ds_info_list; dp; dp = dp->next) {
763 			struct ds_cap_state *tmp;
764 
765 			tmp = find_cap_by_string(dp, "var-config-backup");
766 			if (tmp && tmp->state == CAP_STATE_REGISTERED) {
767 				cp = tmp;
768 				break;
769 			}
770 		}
771 	}
772 	spin_unlock_irqrestore(&ds_lock, flags);
773 
774 	if (cp) {
775 		union {
776 			struct {
777 				struct ds_data		data;
778 				struct ds_var_set_msg	msg;
779 			} header;
780 			char			all[512];
781 		} pkt;
782 		char  *base, *p;
783 		int msg_len, loops;
784 
785 		if (strlen(var) + strlen(value) + 2 >
786 		    sizeof(pkt) - sizeof(pkt.header)) {
787 			printk(KERN_ERR PFX
788 				"contents length: %zu, which more than max: %lu,"
789 				"so could not set (%s) variable to (%s).\n",
790 				strlen(var) + strlen(value) + 2,
791 				sizeof(pkt) - sizeof(pkt.header), var, value);
792 			return;
793 		}
794 
795 		memset(&pkt, 0, sizeof(pkt));
796 		pkt.header.data.tag.type = DS_DATA;
797 		pkt.header.data.handle = cp->handle;
798 		pkt.header.msg.hdr.type = DS_VAR_SET_REQ;
799 		base = p = &pkt.header.msg.name_and_value[0];
800 		strcpy(p, var);
801 		p += strlen(var) + 1;
802 		strcpy(p, value);
803 		p += strlen(value) + 1;
804 
805 		msg_len = (sizeof(struct ds_data) +
806 			   sizeof(struct ds_var_set_msg) +
807 			   (p - base));
808 		msg_len = (msg_len + 3) & ~3;
809 		pkt.header.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
810 
811 		mutex_lock(&ds_var_mutex);
812 
813 		spin_lock_irqsave(&ds_lock, flags);
814 		ds_var_doorbell = 0;
815 		ds_var_response = -1;
816 
817 		__ds_send(dp->lp, &pkt, msg_len);
818 		spin_unlock_irqrestore(&ds_lock, flags);
819 
820 		loops = 1000;
821 		while (ds_var_doorbell == 0) {
822 			if (loops-- < 0)
823 				break;
824 			barrier();
825 			udelay(100);
826 		}
827 
828 		mutex_unlock(&ds_var_mutex);
829 
830 		if (ds_var_doorbell == 0 ||
831 		    ds_var_response != DS_VAR_SUCCESS)
832 			printk(KERN_ERR "ds-%llu: var-config [%s:%s] "
833 			       "failed, response(%d).\n",
834 			       dp->id, var, value,
835 			       ds_var_response);
836 	} else {
837 		printk(KERN_ERR PFX "var-config not registered so "
838 		       "could not set (%s) variable to (%s).\n",
839 		       var, value);
840 	}
841 }
842 
843 static char full_boot_str[256] __attribute__((aligned(32)));
844 static int reboot_data_supported;
845 
846 void ldom_reboot(const char *boot_command)
847 {
848 	/* Don't bother with any of this if the boot_command
849 	 * is empty.
850 	 */
851 	if (boot_command && strlen(boot_command)) {
852 		unsigned long len;
853 
854 		snprintf(full_boot_str, sizeof(full_boot_str), "boot %s",
855 			 boot_command);
856 		len = strlen(full_boot_str);
857 
858 		if (reboot_data_supported) {
859 			unsigned long ra = kimage_addr_to_ra(full_boot_str);
860 			unsigned long hv_ret;
861 
862 			hv_ret = sun4v_reboot_data_set(ra, len);
863 			if (hv_ret != HV_EOK)
864 				pr_err("SUN4V: Unable to set reboot data "
865 				       "hv_ret=%lu\n", hv_ret);
866 		} else {
867 			ldom_set_var("reboot-command", full_boot_str);
868 		}
869 	}
870 	sun4v_mach_sir();
871 }
872 
873 void ldom_power_off(void)
874 {
875 	sun4v_mach_exit(0);
876 }
877 
878 static void ds_conn_reset(struct ds_info *dp)
879 {
880 	printk(KERN_ERR "ds-%llu: ds_conn_reset() from %ps\n",
881 	       dp->id, __builtin_return_address(0));
882 }
883 
884 static int register_services(struct ds_info *dp)
885 {
886 	struct ldc_channel *lp = dp->lp;
887 	int i;
888 
889 	for (i = 0; i < dp->num_ds_states; i++) {
890 		struct {
891 			struct ds_reg_req req;
892 			u8 id_buf[256];
893 		} pbuf;
894 		struct ds_cap_state *cp = &dp->ds_states[i];
895 		int err, msg_len;
896 		u64 new_count;
897 
898 		if (cp->state == CAP_STATE_REGISTERED)
899 			continue;
900 
901 		new_count = sched_clock() & 0xffffffff;
902 		cp->handle = ((u64) i << 32) | new_count;
903 
904 		msg_len = (sizeof(struct ds_reg_req) +
905 			   strlen(cp->service_id));
906 
907 		memset(&pbuf, 0, sizeof(pbuf));
908 		pbuf.req.tag.type = DS_REG_REQ;
909 		pbuf.req.tag.len = (msg_len - sizeof(struct ds_msg_tag));
910 		pbuf.req.handle = cp->handle;
911 		pbuf.req.major = 1;
912 		pbuf.req.minor = 0;
913 		strcpy(pbuf.id_buf, cp->service_id);
914 
915 		err = __ds_send(lp, &pbuf, msg_len);
916 		if (err > 0)
917 			cp->state = CAP_STATE_REG_SENT;
918 	}
919 	return 0;
920 }
921 
922 static int ds_handshake(struct ds_info *dp, struct ds_msg_tag *pkt)
923 {
924 
925 	if (dp->hs_state == DS_HS_START) {
926 		if (pkt->type != DS_INIT_ACK)
927 			goto conn_reset;
928 
929 		dp->hs_state = DS_HS_DONE;
930 
931 		return register_services(dp);
932 	}
933 
934 	if (dp->hs_state != DS_HS_DONE)
935 		goto conn_reset;
936 
937 	if (pkt->type == DS_REG_ACK) {
938 		struct ds_reg_ack *ap = (struct ds_reg_ack *) pkt;
939 		struct ds_cap_state *cp = find_cap(dp, ap->handle);
940 
941 		if (!cp) {
942 			printk(KERN_ERR "ds-%llu: REG ACK for unknown "
943 			       "handle %llx\n", dp->id, ap->handle);
944 			return 0;
945 		}
946 		printk(KERN_INFO "ds-%llu: Registered %s service.\n",
947 		       dp->id, cp->service_id);
948 		cp->state = CAP_STATE_REGISTERED;
949 	} else if (pkt->type == DS_REG_NACK) {
950 		struct ds_reg_nack *np = (struct ds_reg_nack *) pkt;
951 		struct ds_cap_state *cp = find_cap(dp, np->handle);
952 
953 		if (!cp) {
954 			printk(KERN_ERR "ds-%llu: REG NACK for "
955 			       "unknown handle %llx\n",
956 			       dp->id, np->handle);
957 			return 0;
958 		}
959 		cp->state = CAP_STATE_UNKNOWN;
960 	}
961 
962 	return 0;
963 
964 conn_reset:
965 	ds_conn_reset(dp);
966 	return -ECONNRESET;
967 }
968 
969 static void __send_ds_nack(struct ds_info *dp, u64 handle)
970 {
971 	struct ds_data_nack nack = {
972 		.tag = {
973 			.type = DS_NACK,
974 			.len = (sizeof(struct ds_data_nack) -
975 				sizeof(struct ds_msg_tag)),
976 		},
977 		.handle = handle,
978 		.result = DS_INV_HDL,
979 	};
980 
981 	__ds_send(dp->lp, &nack, sizeof(nack));
982 }
983 
984 static LIST_HEAD(ds_work_list);
985 static DECLARE_WAIT_QUEUE_HEAD(ds_wait);
986 
987 struct ds_queue_entry {
988 	struct list_head		list;
989 	struct ds_info			*dp;
990 	int				req_len;
991 	int				__pad;
992 	u64				req[];
993 };
994 
995 static void process_ds_work(void)
996 {
997 	struct ds_queue_entry *qp, *tmp;
998 	unsigned long flags;
999 	LIST_HEAD(todo);
1000 
1001 	spin_lock_irqsave(&ds_lock, flags);
1002 	list_splice_init(&ds_work_list, &todo);
1003 	spin_unlock_irqrestore(&ds_lock, flags);
1004 
1005 	list_for_each_entry_safe(qp, tmp, &todo, list) {
1006 		struct ds_data *dpkt = (struct ds_data *) qp->req;
1007 		struct ds_info *dp = qp->dp;
1008 		struct ds_cap_state *cp = find_cap(dp, dpkt->handle);
1009 		int req_len = qp->req_len;
1010 
1011 		if (!cp) {
1012 			printk(KERN_ERR "ds-%llu: Data for unknown "
1013 			       "handle %llu\n",
1014 			       dp->id, dpkt->handle);
1015 
1016 			spin_lock_irqsave(&ds_lock, flags);
1017 			__send_ds_nack(dp, dpkt->handle);
1018 			spin_unlock_irqrestore(&ds_lock, flags);
1019 		} else {
1020 			cp->data(dp, cp, dpkt, req_len);
1021 		}
1022 
1023 		list_del(&qp->list);
1024 		kfree(qp);
1025 	}
1026 }
1027 
1028 static int ds_thread(void *__unused)
1029 {
1030 	DEFINE_WAIT(wait);
1031 
1032 	while (1) {
1033 		prepare_to_wait(&ds_wait, &wait, TASK_INTERRUPTIBLE);
1034 		if (list_empty(&ds_work_list))
1035 			schedule();
1036 		finish_wait(&ds_wait, &wait);
1037 
1038 		if (kthread_should_stop())
1039 			break;
1040 
1041 		process_ds_work();
1042 	}
1043 
1044 	return 0;
1045 }
1046 
1047 static int ds_data(struct ds_info *dp, struct ds_msg_tag *pkt, int len)
1048 {
1049 	struct ds_data *dpkt = (struct ds_data *) pkt;
1050 	struct ds_queue_entry *qp;
1051 
1052 	qp = kmalloc(sizeof(struct ds_queue_entry) + len, GFP_ATOMIC);
1053 	if (!qp) {
1054 		__send_ds_nack(dp, dpkt->handle);
1055 	} else {
1056 		qp->dp = dp;
1057 		memcpy(&qp->req, pkt, len);
1058 		list_add_tail(&qp->list, &ds_work_list);
1059 		wake_up(&ds_wait);
1060 	}
1061 	return 0;
1062 }
1063 
1064 static void ds_up(struct ds_info *dp)
1065 {
1066 	struct ldc_channel *lp = dp->lp;
1067 	struct ds_ver_req req;
1068 	int err;
1069 
1070 	req.tag.type = DS_INIT_REQ;
1071 	req.tag.len = sizeof(req) - sizeof(struct ds_msg_tag);
1072 	req.ver.major = 1;
1073 	req.ver.minor = 0;
1074 
1075 	err = __ds_send(lp, &req, sizeof(req));
1076 	if (err > 0)
1077 		dp->hs_state = DS_HS_START;
1078 }
1079 
1080 static void ds_reset(struct ds_info *dp)
1081 {
1082 	int i;
1083 
1084 	dp->hs_state = 0;
1085 
1086 	for (i = 0; i < dp->num_ds_states; i++) {
1087 		struct ds_cap_state *cp = &dp->ds_states[i];
1088 
1089 		cp->state = CAP_STATE_UNKNOWN;
1090 	}
1091 }
1092 
1093 static void ds_event(void *arg, int event)
1094 {
1095 	struct ds_info *dp = arg;
1096 	struct ldc_channel *lp = dp->lp;
1097 	unsigned long flags;
1098 	int err;
1099 
1100 	spin_lock_irqsave(&ds_lock, flags);
1101 
1102 	if (event == LDC_EVENT_UP) {
1103 		ds_up(dp);
1104 		spin_unlock_irqrestore(&ds_lock, flags);
1105 		return;
1106 	}
1107 
1108 	if (event == LDC_EVENT_RESET) {
1109 		ds_reset(dp);
1110 		spin_unlock_irqrestore(&ds_lock, flags);
1111 		return;
1112 	}
1113 
1114 	if (event != LDC_EVENT_DATA_READY) {
1115 		printk(KERN_WARNING "ds-%llu: Unexpected LDC event %d\n",
1116 		       dp->id, event);
1117 		spin_unlock_irqrestore(&ds_lock, flags);
1118 		return;
1119 	}
1120 
1121 	err = 0;
1122 	while (1) {
1123 		struct ds_msg_tag *tag;
1124 
1125 		err = ldc_read(lp, dp->rcv_buf, sizeof(*tag));
1126 
1127 		if (unlikely(err < 0)) {
1128 			if (err == -ECONNRESET)
1129 				ds_conn_reset(dp);
1130 			break;
1131 		}
1132 		if (err == 0)
1133 			break;
1134 
1135 		tag = dp->rcv_buf;
1136 		err = ldc_read(lp, tag + 1, tag->len);
1137 
1138 		if (unlikely(err < 0)) {
1139 			if (err == -ECONNRESET)
1140 				ds_conn_reset(dp);
1141 			break;
1142 		}
1143 		if (err < tag->len)
1144 			break;
1145 
1146 		if (tag->type < DS_DATA)
1147 			err = ds_handshake(dp, dp->rcv_buf);
1148 		else
1149 			err = ds_data(dp, dp->rcv_buf,
1150 				      sizeof(*tag) + err);
1151 		if (err == -ECONNRESET)
1152 			break;
1153 	}
1154 
1155 	spin_unlock_irqrestore(&ds_lock, flags);
1156 }
1157 
1158 static int ds_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1159 {
1160 	static int ds_version_printed;
1161 	struct ldc_channel_config ds_cfg = {
1162 		.event		= ds_event,
1163 		.mtu		= 4096,
1164 		.mode		= LDC_MODE_STREAM,
1165 	};
1166 	struct mdesc_handle *hp;
1167 	struct ldc_channel *lp;
1168 	struct ds_info *dp;
1169 	const u64 *val;
1170 	int err, i;
1171 
1172 	if (ds_version_printed++ == 0)
1173 		printk(KERN_INFO "%s", version);
1174 
1175 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1176 	err = -ENOMEM;
1177 	if (!dp)
1178 		goto out_err;
1179 
1180 	hp = mdesc_grab();
1181 	val = mdesc_get_property(hp, vdev->mp, "id", NULL);
1182 	if (val)
1183 		dp->id = *val;
1184 	mdesc_release(hp);
1185 
1186 	dp->rcv_buf = kzalloc(4096, GFP_KERNEL);
1187 	if (!dp->rcv_buf)
1188 		goto out_free_dp;
1189 
1190 	dp->rcv_buf_len = 4096;
1191 
1192 	dp->ds_states = kmemdup(ds_states_template,
1193 				sizeof(ds_states_template), GFP_KERNEL);
1194 	if (!dp->ds_states)
1195 		goto out_free_rcv_buf;
1196 
1197 	dp->num_ds_states = ARRAY_SIZE(ds_states_template);
1198 
1199 	for (i = 0; i < dp->num_ds_states; i++)
1200 		dp->ds_states[i].handle = ((u64)i << 32);
1201 
1202 	ds_cfg.tx_irq = vdev->tx_irq;
1203 	ds_cfg.rx_irq = vdev->rx_irq;
1204 
1205 	lp = ldc_alloc(vdev->channel_id, &ds_cfg, dp, "DS");
1206 	if (IS_ERR(lp)) {
1207 		err = PTR_ERR(lp);
1208 		goto out_free_ds_states;
1209 	}
1210 	dp->lp = lp;
1211 
1212 	err = ldc_bind(lp);
1213 	if (err)
1214 		goto out_free_ldc;
1215 
1216 	spin_lock_irq(&ds_lock);
1217 	dp->next = ds_info_list;
1218 	ds_info_list = dp;
1219 	spin_unlock_irq(&ds_lock);
1220 
1221 	return err;
1222 
1223 out_free_ldc:
1224 	ldc_free(dp->lp);
1225 
1226 out_free_ds_states:
1227 	kfree(dp->ds_states);
1228 
1229 out_free_rcv_buf:
1230 	kfree(dp->rcv_buf);
1231 
1232 out_free_dp:
1233 	kfree(dp);
1234 
1235 out_err:
1236 	return err;
1237 }
1238 
1239 static const struct vio_device_id ds_match[] = {
1240 	{
1241 		.type = "domain-services-port",
1242 	},
1243 	{},
1244 };
1245 
1246 static struct vio_driver ds_driver = {
1247 	.id_table	= ds_match,
1248 	.probe		= ds_probe,
1249 	.name		= "ds",
1250 };
1251 
1252 static int __init ds_init(void)
1253 {
1254 	unsigned long hv_ret, major, minor;
1255 
1256 	if (tlb_type == hypervisor) {
1257 		hv_ret = sun4v_get_version(HV_GRP_REBOOT_DATA, &major, &minor);
1258 		if (hv_ret == HV_EOK) {
1259 			pr_info("SUN4V: Reboot data supported (maj=%lu,min=%lu).\n",
1260 				major, minor);
1261 			reboot_data_supported = 1;
1262 		}
1263 	}
1264 	kthread_run(ds_thread, NULL, "kldomd");
1265 
1266 	return vio_register_driver(&ds_driver);
1267 }
1268 
1269 fs_initcall(ds_init);
1270