Merge master.kernel.org:/pub/scm/linux/kernel/git/davej/cpufreq
[linux-2.6] / arch / sparc64 / kernel / ds.c
1 /* ds.c: Domain Services driver for Logical Domains
2  *
3  * Copyright (C) 2007 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/ldc.h>
19 #include <asm/vio.h>
20 #include <asm/mdesc.h>
21 #include <asm/head.h>
22 #include <asm/irq.h>
23
24 #define DRV_MODULE_NAME         "ds"
25 #define PFX DRV_MODULE_NAME     ": "
26 #define DRV_MODULE_VERSION      "1.0"
27 #define DRV_MODULE_RELDATE      "Jul 11, 2007"
28
29 static char version[] __devinitdata =
30         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
31 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
32 MODULE_DESCRIPTION("Sun LDOM domain services driver");
33 MODULE_LICENSE("GPL");
34 MODULE_VERSION(DRV_MODULE_VERSION);
35
36 struct ds_msg_tag {
37         __u32                   type;
38 #define DS_INIT_REQ             0x00
39 #define DS_INIT_ACK             0x01
40 #define DS_INIT_NACK            0x02
41 #define DS_REG_REQ              0x03
42 #define DS_REG_ACK              0x04
43 #define DS_REG_NACK             0x05
44 #define DS_UNREG_REQ            0x06
45 #define DS_UNREG_ACK            0x07
46 #define DS_UNREG_NACK           0x08
47 #define DS_DATA                 0x09
48 #define DS_NACK                 0x0a
49
50         __u32                   len;
51 };
52
53 /* Result codes */
54 #define DS_OK                   0x00
55 #define DS_REG_VER_NACK         0x01
56 #define DS_REG_DUP              0x02
57 #define DS_INV_HDL              0x03
58 #define DS_TYPE_UNKNOWN         0x04
59
60 struct ds_version {
61         __u16                   major;
62         __u16                   minor;
63 };
64
65 struct ds_ver_req {
66         struct ds_msg_tag       tag;
67         struct ds_version       ver;
68 };
69
70 struct ds_ver_ack {
71         struct ds_msg_tag       tag;
72         __u16                   minor;
73 };
74
75 struct ds_ver_nack {
76         struct ds_msg_tag       tag;
77         __u16                   major;
78 };
79
80 struct ds_reg_req {
81         struct ds_msg_tag       tag;
82         __u64                   handle;
83         __u16                   major;
84         __u16                   minor;
85         char                    svc_id[0];
86 };
87
88 struct ds_reg_ack {
89         struct ds_msg_tag       tag;
90         __u64                   handle;
91         __u16                   minor;
92 };
93
94 struct ds_reg_nack {
95         struct ds_msg_tag       tag;
96         __u64                   handle;
97         __u16                   major;
98 };
99
100 struct ds_unreg_req {
101         struct ds_msg_tag       tag;
102         __u64                   handle;
103 };
104
105 struct ds_unreg_ack {
106         struct ds_msg_tag       tag;
107         __u64                   handle;
108 };
109
110 struct ds_unreg_nack {
111         struct ds_msg_tag       tag;
112         __u64                   handle;
113 };
114
115 struct ds_data {
116         struct ds_msg_tag       tag;
117         __u64                   handle;
118 };
119
120 struct ds_data_nack {
121         struct ds_msg_tag       tag;
122         __u64                   handle;
123         __u64                   result;
124 };
125
126 struct ds_info;
127 struct ds_cap_state {
128         __u64                   handle;
129
130         void                    (*data)(struct ds_info *dp,
131                                         struct ds_cap_state *cp,
132                                         void *buf, int len);
133
134         const char              *service_id;
135
136         u8                      state;
137 #define CAP_STATE_UNKNOWN       0x00
138 #define CAP_STATE_REG_SENT      0x01
139 #define CAP_STATE_REGISTERED    0x02
140 };
141
142 static void md_update_data(struct ds_info *dp, struct ds_cap_state *cp,
143                            void *buf, int len);
144 static void domain_shutdown_data(struct ds_info *dp,
145                                  struct ds_cap_state *cp,
146                                  void *buf, int len);
147 static void domain_panic_data(struct ds_info *dp,
148                               struct ds_cap_state *cp,
149                               void *buf, int len);
150 #ifdef CONFIG_HOTPLUG_CPU
151 static void dr_cpu_data(struct ds_info *dp,
152                         struct ds_cap_state *cp,
153                         void *buf, int len);
154 #endif
155 static void ds_pri_data(struct ds_info *dp,
156                         struct ds_cap_state *cp,
157                         void *buf, int len);
158 static void ds_var_data(struct ds_info *dp,
159                         struct ds_cap_state *cp,
160                         void *buf, int len);
161
162 struct ds_cap_state ds_states_template[] = {
163         {
164                 .service_id     = "md-update",
165                 .data           = md_update_data,
166         },
167         {
168                 .service_id     = "domain-shutdown",
169                 .data           = domain_shutdown_data,
170         },
171         {
172                 .service_id     = "domain-panic",
173                 .data           = domain_panic_data,
174         },
175 #ifdef CONFIG_HOTPLUG_CPU
176         {
177                 .service_id     = "dr-cpu",
178                 .data           = dr_cpu_data,
179         },
180 #endif
181         {
182                 .service_id     = "pri",
183                 .data           = ds_pri_data,
184         },
185         {
186                 .service_id     = "var-config",
187                 .data           = ds_var_data,
188         },
189         {
190                 .service_id     = "var-config-backup",
191                 .data           = ds_var_data,
192         },
193 };
194
195 static DEFINE_SPINLOCK(ds_lock);
196
197 struct ds_info {
198         struct ldc_channel      *lp;
199         u8                      hs_state;
200 #define DS_HS_START             0x01
201 #define DS_HS_DONE              0x02
202
203         u64                     id;
204
205         void                    *rcv_buf;
206         int                     rcv_buf_len;
207
208         struct ds_cap_state     *ds_states;
209         int                     num_ds_states;
210
211         struct ds_info          *next;
212 };
213
214 static struct ds_info *ds_info_list;
215
216 static struct ds_cap_state *find_cap(struct ds_info *dp, u64 handle)
217 {
218         unsigned int index = handle >> 32;
219
220         if (index >= dp->num_ds_states)
221                 return NULL;
222         return &dp->ds_states[index];
223 }
224
225 static struct ds_cap_state *find_cap_by_string(struct ds_info *dp,
226                                                const char *name)
227 {
228         int i;
229
230         for (i = 0; i < dp->num_ds_states; i++) {
231                 if (strcmp(dp->ds_states[i].service_id, name))
232                         continue;
233
234                 return &dp->ds_states[i];
235         }
236         return NULL;
237 }
238
239 static int __ds_send(struct ldc_channel *lp, void *data, int len)
240 {
241         int err, limit = 1000;
242
243         err = -EINVAL;
244         while (limit-- > 0) {
245                 err = ldc_write(lp, data, len);
246                 if (!err || (err != -EAGAIN))
247                         break;
248                 udelay(1);
249         }
250
251         return err;
252 }
253
254 static int ds_send(struct ldc_channel *lp, void *data, int len)
255 {
256         unsigned long flags;
257         int err;
258
259         spin_lock_irqsave(&ds_lock, flags);
260         err = __ds_send(lp, data, len);
261         spin_unlock_irqrestore(&ds_lock, flags);
262
263         return err;
264 }
265
266 struct ds_md_update_req {
267         __u64                           req_num;
268 };
269
270 struct ds_md_update_res {
271         __u64                           req_num;
272         __u32                           result;
273 };
274
275 static void md_update_data(struct ds_info *dp,
276                            struct ds_cap_state *cp,
277                            void *buf, int len)
278 {
279         struct ldc_channel *lp = dp->lp;
280         struct ds_data *dpkt = buf;
281         struct ds_md_update_req *rp;
282         struct {
283                 struct ds_data          data;
284                 struct ds_md_update_res res;
285         } pkt;
286
287         rp = (struct ds_md_update_req *) (dpkt + 1);
288
289         printk(KERN_INFO "ds-%lu: Machine description update.\n", dp->id);
290
291         mdesc_update();
292
293         memset(&pkt, 0, sizeof(pkt));
294         pkt.data.tag.type = DS_DATA;
295         pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
296         pkt.data.handle = cp->handle;
297         pkt.res.req_num = rp->req_num;
298         pkt.res.result = DS_OK;
299
300         ds_send(lp, &pkt, sizeof(pkt));
301 }
302
303 struct ds_shutdown_req {
304         __u64                           req_num;
305         __u32                           ms_delay;
306 };
307
308 struct ds_shutdown_res {
309         __u64                           req_num;
310         __u32                           result;
311         char                            reason[1];
312 };
313
314 static void domain_shutdown_data(struct ds_info *dp,
315                                  struct ds_cap_state *cp,
316                                  void *buf, int len)
317 {
318         struct ldc_channel *lp = dp->lp;
319         struct ds_data *dpkt = buf;
320         struct ds_shutdown_req *rp;
321         struct {
322                 struct ds_data          data;
323                 struct ds_shutdown_res  res;
324         } pkt;
325
326         rp = (struct ds_shutdown_req *) (dpkt + 1);
327
328         printk(KERN_ALERT "ds-%lu: Shutdown request from "
329                "LDOM manager received.\n", dp->id);
330
331         memset(&pkt, 0, sizeof(pkt));
332         pkt.data.tag.type = DS_DATA;
333         pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
334         pkt.data.handle = cp->handle;
335         pkt.res.req_num = rp->req_num;
336         pkt.res.result = DS_OK;
337         pkt.res.reason[0] = 0;
338
339         ds_send(lp, &pkt, sizeof(pkt));
340
341         orderly_poweroff(true);
342 }
343
344 struct ds_panic_req {
345         __u64                           req_num;
346 };
347
348 struct ds_panic_res {
349         __u64                           req_num;
350         __u32                           result;
351         char                            reason[1];
352 };
353
354 static void domain_panic_data(struct ds_info *dp,
355                               struct ds_cap_state *cp,
356                               void *buf, int len)
357 {
358         struct ldc_channel *lp = dp->lp;
359         struct ds_data *dpkt = buf;
360         struct ds_panic_req *rp;
361         struct {
362                 struct ds_data          data;
363                 struct ds_panic_res     res;
364         } pkt;
365
366         rp = (struct ds_panic_req *) (dpkt + 1);
367
368         printk(KERN_ALERT "ds-%lu: Panic request from "
369                "LDOM manager received.\n", dp->id);
370
371         memset(&pkt, 0, sizeof(pkt));
372         pkt.data.tag.type = DS_DATA;
373         pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
374         pkt.data.handle = cp->handle;
375         pkt.res.req_num = rp->req_num;
376         pkt.res.result = DS_OK;
377         pkt.res.reason[0] = 0;
378
379         ds_send(lp, &pkt, sizeof(pkt));
380
381         panic("PANIC requested by LDOM manager.");
382 }
383
384 #ifdef CONFIG_HOTPLUG_CPU
385 struct dr_cpu_tag {
386         __u64                           req_num;
387         __u32                           type;
388 #define DR_CPU_CONFIGURE                0x43
389 #define DR_CPU_UNCONFIGURE              0x55
390 #define DR_CPU_FORCE_UNCONFIGURE        0x46
391 #define DR_CPU_STATUS                   0x53
392
393 /* Responses */
394 #define DR_CPU_OK                       0x6f
395 #define DR_CPU_ERROR                    0x65
396
397         __u32                           num_records;
398 };
399
400 struct dr_cpu_resp_entry {
401         __u32                           cpu;
402         __u32                           result;
403 #define DR_CPU_RES_OK                   0x00
404 #define DR_CPU_RES_FAILURE              0x01
405 #define DR_CPU_RES_BLOCKED              0x02
406 #define DR_CPU_RES_CPU_NOT_RESPONDING   0x03
407 #define DR_CPU_RES_NOT_IN_MD            0x04
408
409         __u32                           stat;
410 #define DR_CPU_STAT_NOT_PRESENT         0x00
411 #define DR_CPU_STAT_UNCONFIGURED        0x01
412 #define DR_CPU_STAT_CONFIGURED          0x02
413
414         __u32                           str_off;
415 };
416
417 static void __dr_cpu_send_error(struct ds_info *dp,
418                                 struct ds_cap_state *cp,
419                                 struct ds_data *data)
420 {
421         struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
422         struct {
423                 struct ds_data          data;
424                 struct dr_cpu_tag       tag;
425         } pkt;
426         int msg_len;
427
428         memset(&pkt, 0, sizeof(pkt));
429         pkt.data.tag.type = DS_DATA;
430         pkt.data.handle = cp->handle;
431         pkt.tag.req_num = tag->req_num;
432         pkt.tag.type = DR_CPU_ERROR;
433         pkt.tag.num_records = 0;
434
435         msg_len = (sizeof(struct ds_data) +
436                    sizeof(struct dr_cpu_tag));
437
438         pkt.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
439
440         __ds_send(dp->lp, &pkt, msg_len);
441 }
442
443 static void dr_cpu_send_error(struct ds_info *dp,
444                               struct ds_cap_state *cp,
445                               struct ds_data *data)
446 {
447         unsigned long flags;
448
449         spin_lock_irqsave(&ds_lock, flags);
450         __dr_cpu_send_error(dp, cp, data);
451         spin_unlock_irqrestore(&ds_lock, flags);
452 }
453
454 #define CPU_SENTINEL    0xffffffff
455
456 static void purge_dups(u32 *list, u32 num_ents)
457 {
458         unsigned int i;
459
460         for (i = 0; i < num_ents; i++) {
461                 u32 cpu = list[i];
462                 unsigned int j;
463
464                 if (cpu == CPU_SENTINEL)
465                         continue;
466
467                 for (j = i + 1; j < num_ents; j++) {
468                         if (list[j] == cpu)
469                                 list[j] = CPU_SENTINEL;
470                 }
471         }
472 }
473
474 static int dr_cpu_size_response(int ncpus)
475 {
476         return (sizeof(struct ds_data) +
477                 sizeof(struct dr_cpu_tag) +
478                 (sizeof(struct dr_cpu_resp_entry) * ncpus));
479 }
480
481 static void dr_cpu_init_response(struct ds_data *resp, u64 req_num,
482                                  u64 handle, int resp_len, int ncpus,
483                                  cpumask_t *mask, u32 default_stat)
484 {
485         struct dr_cpu_resp_entry *ent;
486         struct dr_cpu_tag *tag;
487         int i, cpu;
488
489         tag = (struct dr_cpu_tag *) (resp + 1);
490         ent = (struct dr_cpu_resp_entry *) (tag + 1);
491
492         resp->tag.type = DS_DATA;
493         resp->tag.len = resp_len - sizeof(struct ds_msg_tag);
494         resp->handle = handle;
495         tag->req_num = req_num;
496         tag->type = DR_CPU_OK;
497         tag->num_records = ncpus;
498
499         i = 0;
500         for_each_cpu_mask(cpu, *mask) {
501                 ent[i].cpu = cpu;
502                 ent[i].result = DR_CPU_RES_OK;
503                 ent[i].stat = default_stat;
504                 i++;
505         }
506         BUG_ON(i != ncpus);
507 }
508
509 static void dr_cpu_mark(struct ds_data *resp, int cpu, int ncpus,
510                         u32 res, u32 stat)
511 {
512         struct dr_cpu_resp_entry *ent;
513         struct dr_cpu_tag *tag;
514         int i;
515
516         tag = (struct dr_cpu_tag *) (resp + 1);
517         ent = (struct dr_cpu_resp_entry *) (tag + 1);
518
519         for (i = 0; i < ncpus; i++) {
520                 if (ent[i].cpu != cpu)
521                         continue;
522                 ent[i].result = res;
523                 ent[i].stat = stat;
524                 break;
525         }
526 }
527
528 static int dr_cpu_configure(struct ds_info *dp,
529                             struct ds_cap_state *cp,
530                             u64 req_num,
531                             cpumask_t *mask)
532 {
533         struct ds_data *resp;
534         int resp_len, ncpus, cpu;
535         unsigned long flags;
536
537         ncpus = cpus_weight(*mask);
538         resp_len = dr_cpu_size_response(ncpus);
539         resp = kzalloc(resp_len, GFP_KERNEL);
540         if (!resp)
541                 return -ENOMEM;
542
543         dr_cpu_init_response(resp, req_num, cp->handle,
544                              resp_len, ncpus, mask,
545                              DR_CPU_STAT_CONFIGURED);
546
547         mdesc_fill_in_cpu_data(*mask);
548
549         for_each_cpu_mask(cpu, *mask) {
550                 int err;
551
552                 printk(KERN_INFO "ds-%lu: Starting cpu %d...\n",
553                        dp->id, cpu);
554                 err = cpu_up(cpu);
555                 if (err) {
556                         __u32 res = DR_CPU_RES_FAILURE;
557                         __u32 stat = DR_CPU_STAT_UNCONFIGURED;
558
559                         if (!cpu_present(cpu)) {
560                                 /* CPU not present in MD */
561                                 res = DR_CPU_RES_NOT_IN_MD;
562                                 stat = DR_CPU_STAT_NOT_PRESENT;
563                         } else if (err == -ENODEV) {
564                                 /* CPU did not call in successfully */
565                                 res = DR_CPU_RES_CPU_NOT_RESPONDING;
566                         }
567
568                         printk(KERN_INFO "ds-%lu: CPU startup failed err=%d\n",
569                                dp->id, err);
570                         dr_cpu_mark(resp, cpu, ncpus, res, stat);
571                 }
572         }
573
574         spin_lock_irqsave(&ds_lock, flags);
575         __ds_send(dp->lp, resp, resp_len);
576         spin_unlock_irqrestore(&ds_lock, flags);
577
578         kfree(resp);
579
580         /* Redistribute IRQs, taking into account the new cpus.  */
581         fixup_irqs();
582
583         return 0;
584 }
585
586 static int dr_cpu_unconfigure(struct ds_info *dp,
587                               struct ds_cap_state *cp,
588                               u64 req_num,
589                               cpumask_t *mask)
590 {
591         struct ds_data *resp;
592         int resp_len, ncpus, cpu;
593         unsigned long flags;
594
595         ncpus = cpus_weight(*mask);
596         resp_len = dr_cpu_size_response(ncpus);
597         resp = kzalloc(resp_len, GFP_KERNEL);
598         if (!resp)
599                 return -ENOMEM;
600
601         dr_cpu_init_response(resp, req_num, cp->handle,
602                              resp_len, ncpus, mask,
603                              DR_CPU_STAT_UNCONFIGURED);
604
605         for_each_cpu_mask(cpu, *mask) {
606                 int err;
607
608                 printk(KERN_INFO "ds-%lu: Shutting down cpu %d...\n",
609                        dp->id, cpu);
610                 err = cpu_down(cpu);
611                 if (err)
612                         dr_cpu_mark(resp, cpu, ncpus,
613                                     DR_CPU_RES_FAILURE,
614                                     DR_CPU_STAT_CONFIGURED);
615         }
616
617         spin_lock_irqsave(&ds_lock, flags);
618         __ds_send(dp->lp, resp, resp_len);
619         spin_unlock_irqrestore(&ds_lock, flags);
620
621         kfree(resp);
622
623         return 0;
624 }
625
626 static void dr_cpu_data(struct ds_info *dp,
627                         struct ds_cap_state *cp,
628                         void *buf, int len)
629 {
630         struct ds_data *data = buf;
631         struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
632         u32 *cpu_list = (u32 *) (tag + 1);
633         u64 req_num = tag->req_num;
634         cpumask_t mask;
635         unsigned int i;
636         int err;
637
638         switch (tag->type) {
639         case DR_CPU_CONFIGURE:
640         case DR_CPU_UNCONFIGURE:
641         case DR_CPU_FORCE_UNCONFIGURE:
642                 break;
643
644         default:
645                 dr_cpu_send_error(dp, cp, data);
646                 return;
647         }
648
649         purge_dups(cpu_list, tag->num_records);
650
651         cpus_clear(mask);
652         for (i = 0; i < tag->num_records; i++) {
653                 if (cpu_list[i] == CPU_SENTINEL)
654                         continue;
655
656                 if (cpu_list[i] < NR_CPUS)
657                         cpu_set(cpu_list[i], mask);
658         }
659
660         if (tag->type == DR_CPU_CONFIGURE)
661                 err = dr_cpu_configure(dp, cp, req_num, &mask);
662         else
663                 err = dr_cpu_unconfigure(dp, cp, req_num, &mask);
664
665         if (err)
666                 dr_cpu_send_error(dp, cp, data);
667 }
668 #endif /* CONFIG_HOTPLUG_CPU */
669
670 struct ds_pri_msg {
671         __u64                           req_num;
672         __u64                           type;
673 #define DS_PRI_REQUEST                  0x00
674 #define DS_PRI_DATA                     0x01
675 #define DS_PRI_UPDATE                   0x02
676 };
677
678 static void ds_pri_data(struct ds_info *dp,
679                         struct ds_cap_state *cp,
680                         void *buf, int len)
681 {
682         struct ds_data *dpkt = buf;
683         struct ds_pri_msg *rp;
684
685         rp = (struct ds_pri_msg *) (dpkt + 1);
686
687         printk(KERN_INFO "ds-%lu: PRI REQ [%lx:%lx], len=%d\n",
688                dp->id, rp->req_num, rp->type, len);
689 }
690
691 struct ds_var_hdr {
692         __u32                           type;
693 #define DS_VAR_SET_REQ                  0x00
694 #define DS_VAR_DELETE_REQ               0x01
695 #define DS_VAR_SET_RESP                 0x02
696 #define DS_VAR_DELETE_RESP              0x03
697 };
698
699 struct ds_var_set_msg {
700         struct ds_var_hdr               hdr;
701         char                            name_and_value[0];
702 };
703
704 struct ds_var_delete_msg {
705         struct ds_var_hdr               hdr;
706         char                            name[0];
707 };
708
709 struct ds_var_resp {
710         struct ds_var_hdr               hdr;
711         __u32                           result;
712 #define DS_VAR_SUCCESS                  0x00
713 #define DS_VAR_NO_SPACE                 0x01
714 #define DS_VAR_INVALID_VAR              0x02
715 #define DS_VAR_INVALID_VAL              0x03
716 #define DS_VAR_NOT_PRESENT              0x04
717 };
718
719 static DEFINE_MUTEX(ds_var_mutex);
720 static int ds_var_doorbell;
721 static int ds_var_response;
722
723 static void ds_var_data(struct ds_info *dp,
724                         struct ds_cap_state *cp,
725                         void *buf, int len)
726 {
727         struct ds_data *dpkt = buf;
728         struct ds_var_resp *rp;
729
730         rp = (struct ds_var_resp *) (dpkt + 1);
731
732         if (rp->hdr.type != DS_VAR_SET_RESP &&
733             rp->hdr.type != DS_VAR_DELETE_RESP)
734                 return;
735
736         ds_var_response = rp->result;
737         wmb();
738         ds_var_doorbell = 1;
739 }
740
741 void ldom_set_var(const char *var, const char *value)
742 {
743         struct ds_cap_state *cp;
744         struct ds_info *dp;
745         unsigned long flags;
746
747         spin_lock_irqsave(&ds_lock, flags);
748         cp = NULL;
749         for (dp = ds_info_list; dp; dp = dp->next) {
750                 struct ds_cap_state *tmp;
751
752                 tmp = find_cap_by_string(dp, "var-config");
753                 if (tmp && tmp->state == CAP_STATE_REGISTERED) {
754                         cp = tmp;
755                         break;
756                 }
757         }
758         if (!cp) {
759                 for (dp = ds_info_list; dp; dp = dp->next) {
760                         struct ds_cap_state *tmp;
761
762                         tmp = find_cap_by_string(dp, "var-config-backup");
763                         if (tmp && tmp->state == CAP_STATE_REGISTERED) {
764                                 cp = tmp;
765                                 break;
766                         }
767                 }
768         }
769         spin_unlock_irqrestore(&ds_lock, flags);
770
771         if (cp) {
772                 union {
773                         struct {
774                                 struct ds_data          data;
775                                 struct ds_var_set_msg   msg;
776                         } header;
777                         char                    all[512];
778                 } pkt;
779                 char  *base, *p;
780                 int msg_len, loops;
781
782                 memset(&pkt, 0, sizeof(pkt));
783                 pkt.header.data.tag.type = DS_DATA;
784                 pkt.header.data.handle = cp->handle;
785                 pkt.header.msg.hdr.type = DS_VAR_SET_REQ;
786                 base = p = &pkt.header.msg.name_and_value[0];
787                 strcpy(p, var);
788                 p += strlen(var) + 1;
789                 strcpy(p, value);
790                 p += strlen(value) + 1;
791
792                 msg_len = (sizeof(struct ds_data) +
793                            sizeof(struct ds_var_set_msg) +
794                            (p - base));
795                 msg_len = (msg_len + 3) & ~3;
796                 pkt.header.data.tag.len = msg_len - sizeof(struct ds_msg_tag);
797
798                 mutex_lock(&ds_var_mutex);
799
800                 spin_lock_irqsave(&ds_lock, flags);
801                 ds_var_doorbell = 0;
802                 ds_var_response = -1;
803
804                 __ds_send(dp->lp, &pkt, msg_len);
805                 spin_unlock_irqrestore(&ds_lock, flags);
806
807                 loops = 1000;
808                 while (ds_var_doorbell == 0) {
809                         if (loops-- < 0)
810                                 break;
811                         barrier();
812                         udelay(100);
813                 }
814
815                 mutex_unlock(&ds_var_mutex);
816
817                 if (ds_var_doorbell == 0 ||
818                     ds_var_response != DS_VAR_SUCCESS)
819                         printk(KERN_ERR "ds-%lu: var-config [%s:%s] "
820                                "failed, response(%d).\n",
821                                dp->id, var, value,
822                                ds_var_response);
823         } else {
824                 printk(KERN_ERR PFX "var-config not registered so "
825                        "could not set (%s) variable to (%s).\n",
826                        var, value);
827         }
828 }
829
830 void ldom_reboot(const char *boot_command)
831 {
832         /* Don't bother with any of this if the boot_command
833          * is empty.
834          */
835         if (boot_command && strlen(boot_command)) {
836                 char full_boot_str[256];
837
838                 strcpy(full_boot_str, "boot ");
839                 strcpy(full_boot_str + strlen("boot "), boot_command);
840
841                 ldom_set_var("reboot-command", full_boot_str);
842         }
843         sun4v_mach_sir();
844 }
845
846 void ldom_power_off(void)
847 {
848         sun4v_mach_exit(0);
849 }
850
851 static void ds_conn_reset(struct ds_info *dp)
852 {
853         printk(KERN_ERR "ds-%lu: ds_conn_reset() from %p\n",
854                dp->id, __builtin_return_address(0));
855 }
856
857 static int register_services(struct ds_info *dp)
858 {
859         struct ldc_channel *lp = dp->lp;
860         int i;
861
862         for (i = 0; i < dp->num_ds_states; i++) {
863                 struct {
864                         struct ds_reg_req req;
865                         u8 id_buf[256];
866                 } pbuf;
867                 struct ds_cap_state *cp = &dp->ds_states[i];
868                 int err, msg_len;
869                 u64 new_count;
870
871                 if (cp->state == CAP_STATE_REGISTERED)
872                         continue;
873
874                 new_count = sched_clock() & 0xffffffff;
875                 cp->handle = ((u64) i << 32) | new_count;
876
877                 msg_len = (sizeof(struct ds_reg_req) +
878                            strlen(cp->service_id));
879
880                 memset(&pbuf, 0, sizeof(pbuf));
881                 pbuf.req.tag.type = DS_REG_REQ;
882                 pbuf.req.tag.len = (msg_len - sizeof(struct ds_msg_tag));
883                 pbuf.req.handle = cp->handle;
884                 pbuf.req.major = 1;
885                 pbuf.req.minor = 0;
886                 strcpy(pbuf.req.svc_id, cp->service_id);
887
888                 err = __ds_send(lp, &pbuf, msg_len);
889                 if (err > 0)
890                         cp->state = CAP_STATE_REG_SENT;
891         }
892         return 0;
893 }
894
895 static int ds_handshake(struct ds_info *dp, struct ds_msg_tag *pkt)
896 {
897
898         if (dp->hs_state == DS_HS_START) {
899                 if (pkt->type != DS_INIT_ACK)
900                         goto conn_reset;
901
902                 dp->hs_state = DS_HS_DONE;
903
904                 return register_services(dp);
905         }
906
907         if (dp->hs_state != DS_HS_DONE)
908                 goto conn_reset;
909
910         if (pkt->type == DS_REG_ACK) {
911                 struct ds_reg_ack *ap = (struct ds_reg_ack *) pkt;
912                 struct ds_cap_state *cp = find_cap(dp, ap->handle);
913
914                 if (!cp) {
915                         printk(KERN_ERR "ds-%lu: REG ACK for unknown "
916                                "handle %lx\n", dp->id, ap->handle);
917                         return 0;
918                 }
919                 printk(KERN_INFO "ds-%lu: Registered %s service.\n",
920                        dp->id, cp->service_id);
921                 cp->state = CAP_STATE_REGISTERED;
922         } else if (pkt->type == DS_REG_NACK) {
923                 struct ds_reg_nack *np = (struct ds_reg_nack *) pkt;
924                 struct ds_cap_state *cp = find_cap(dp, np->handle);
925
926                 if (!cp) {
927                         printk(KERN_ERR "ds-%lu: REG NACK for "
928                                "unknown handle %lx\n",
929                                dp->id, np->handle);
930                         return 0;
931                 }
932                 cp->state = CAP_STATE_UNKNOWN;
933         }
934
935         return 0;
936
937 conn_reset:
938         ds_conn_reset(dp);
939         return -ECONNRESET;
940 }
941
942 static void __send_ds_nack(struct ds_info *dp, u64 handle)
943 {
944         struct ds_data_nack nack = {
945                 .tag = {
946                         .type = DS_NACK,
947                         .len = (sizeof(struct ds_data_nack) -
948                                 sizeof(struct ds_msg_tag)),
949                 },
950                 .handle = handle,
951                 .result = DS_INV_HDL,
952         };
953
954         __ds_send(dp->lp, &nack, sizeof(nack));
955 }
956
957 static LIST_HEAD(ds_work_list);
958 static DECLARE_WAIT_QUEUE_HEAD(ds_wait);
959
960 struct ds_queue_entry {
961         struct list_head                list;
962         struct ds_info                  *dp;
963         int                             req_len;
964         int                             __pad;
965         u64                             req[0];
966 };
967
968 static void process_ds_work(void)
969 {
970         struct ds_queue_entry *qp, *tmp;
971         unsigned long flags;
972         LIST_HEAD(todo);
973
974         spin_lock_irqsave(&ds_lock, flags);
975         list_splice(&ds_work_list, &todo);
976         INIT_LIST_HEAD(&ds_work_list);
977         spin_unlock_irqrestore(&ds_lock, flags);
978
979         list_for_each_entry_safe(qp, tmp, &todo, list) {
980                 struct ds_data *dpkt = (struct ds_data *) qp->req;
981                 struct ds_info *dp = qp->dp;
982                 struct ds_cap_state *cp = find_cap(dp, dpkt->handle);
983                 int req_len = qp->req_len;
984
985                 if (!cp) {
986                         printk(KERN_ERR "ds-%lu: Data for unknown "
987                                "handle %lu\n",
988                                dp->id, dpkt->handle);
989
990                         spin_lock_irqsave(&ds_lock, flags);
991                         __send_ds_nack(dp, dpkt->handle);
992                         spin_unlock_irqrestore(&ds_lock, flags);
993                 } else {
994                         cp->data(dp, cp, dpkt, req_len);
995                 }
996
997                 list_del(&qp->list);
998                 kfree(qp);
999         }
1000 }
1001
1002 static int ds_thread(void *__unused)
1003 {
1004         DEFINE_WAIT(wait);
1005
1006         while (1) {
1007                 prepare_to_wait(&ds_wait, &wait, TASK_INTERRUPTIBLE);
1008                 if (list_empty(&ds_work_list))
1009                         schedule();
1010                 finish_wait(&ds_wait, &wait);
1011
1012                 if (kthread_should_stop())
1013                         break;
1014
1015                 process_ds_work();
1016         }
1017
1018         return 0;
1019 }
1020
1021 static int ds_data(struct ds_info *dp, struct ds_msg_tag *pkt, int len)
1022 {
1023         struct ds_data *dpkt = (struct ds_data *) pkt;
1024         struct ds_queue_entry *qp;
1025
1026         qp = kmalloc(sizeof(struct ds_queue_entry) + len, GFP_ATOMIC);
1027         if (!qp) {
1028                 __send_ds_nack(dp, dpkt->handle);
1029         } else {
1030                 qp->dp = dp;
1031                 memcpy(&qp->req, pkt, len);
1032                 list_add_tail(&qp->list, &ds_work_list);
1033                 wake_up(&ds_wait);
1034         }
1035         return 0;
1036 }
1037
1038 static void ds_up(struct ds_info *dp)
1039 {
1040         struct ldc_channel *lp = dp->lp;
1041         struct ds_ver_req req;
1042         int err;
1043
1044         req.tag.type = DS_INIT_REQ;
1045         req.tag.len = sizeof(req) - sizeof(struct ds_msg_tag);
1046         req.ver.major = 1;
1047         req.ver.minor = 0;
1048
1049         err = __ds_send(lp, &req, sizeof(req));
1050         if (err > 0)
1051                 dp->hs_state = DS_HS_START;
1052 }
1053
1054 static void ds_reset(struct ds_info *dp)
1055 {
1056         int i;
1057
1058         dp->hs_state = 0;
1059
1060         for (i = 0; i < dp->num_ds_states; i++) {
1061                 struct ds_cap_state *cp = &dp->ds_states[i];
1062
1063                 cp->state = CAP_STATE_UNKNOWN;
1064         }
1065 }
1066
1067 static void ds_event(void *arg, int event)
1068 {
1069         struct ds_info *dp = arg;
1070         struct ldc_channel *lp = dp->lp;
1071         unsigned long flags;
1072         int err;
1073
1074         spin_lock_irqsave(&ds_lock, flags);
1075
1076         if (event == LDC_EVENT_UP) {
1077                 ds_up(dp);
1078                 spin_unlock_irqrestore(&ds_lock, flags);
1079                 return;
1080         }
1081
1082         if (event == LDC_EVENT_RESET) {
1083                 ds_reset(dp);
1084                 spin_unlock_irqrestore(&ds_lock, flags);
1085                 return;
1086         }
1087
1088         if (event != LDC_EVENT_DATA_READY) {
1089                 printk(KERN_WARNING "ds-%lu: Unexpected LDC event %d\n",
1090                        dp->id, event);
1091                 spin_unlock_irqrestore(&ds_lock, flags);
1092                 return;
1093         }
1094
1095         err = 0;
1096         while (1) {
1097                 struct ds_msg_tag *tag;
1098
1099                 err = ldc_read(lp, dp->rcv_buf, sizeof(*tag));
1100
1101                 if (unlikely(err < 0)) {
1102                         if (err == -ECONNRESET)
1103                                 ds_conn_reset(dp);
1104                         break;
1105                 }
1106                 if (err == 0)
1107                         break;
1108
1109                 tag = dp->rcv_buf;
1110                 err = ldc_read(lp, tag + 1, tag->len);
1111
1112                 if (unlikely(err < 0)) {
1113                         if (err == -ECONNRESET)
1114                                 ds_conn_reset(dp);
1115                         break;
1116                 }
1117                 if (err < tag->len)
1118                         break;
1119
1120                 if (tag->type < DS_DATA)
1121                         err = ds_handshake(dp, dp->rcv_buf);
1122                 else
1123                         err = ds_data(dp, dp->rcv_buf,
1124                                       sizeof(*tag) + err);
1125                 if (err == -ECONNRESET)
1126                         break;
1127         }
1128
1129         spin_unlock_irqrestore(&ds_lock, flags);
1130 }
1131
1132 static int __devinit ds_probe(struct vio_dev *vdev,
1133                               const struct vio_device_id *id)
1134 {
1135         static int ds_version_printed;
1136         struct ldc_channel_config ds_cfg = {
1137                 .event          = ds_event,
1138                 .mtu            = 4096,
1139                 .mode           = LDC_MODE_STREAM,
1140         };
1141         struct mdesc_handle *hp;
1142         struct ldc_channel *lp;
1143         struct ds_info *dp;
1144         const u64 *val;
1145         int err, i;
1146
1147         if (ds_version_printed++ == 0)
1148                 printk(KERN_INFO "%s", version);
1149
1150         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1151         err = -ENOMEM;
1152         if (!dp)
1153                 goto out_err;
1154
1155         hp = mdesc_grab();
1156         val = mdesc_get_property(hp, vdev->mp, "id", NULL);
1157         if (val)
1158                 dp->id = *val;
1159         mdesc_release(hp);
1160
1161         dp->rcv_buf = kzalloc(4096, GFP_KERNEL);
1162         if (!dp->rcv_buf)
1163                 goto out_free_dp;
1164
1165         dp->rcv_buf_len = 4096;
1166
1167         dp->ds_states = kzalloc(sizeof(ds_states_template),
1168                                 GFP_KERNEL);
1169         if (!dp->ds_states)
1170                 goto out_free_rcv_buf;
1171
1172         memcpy(dp->ds_states, ds_states_template,
1173                sizeof(ds_states_template));
1174         dp->num_ds_states = ARRAY_SIZE(ds_states_template);
1175
1176         for (i = 0; i < dp->num_ds_states; i++)
1177                 dp->ds_states[i].handle = ((u64)i << 32);
1178
1179         ds_cfg.tx_irq = vdev->tx_irq;
1180         ds_cfg.rx_irq = vdev->rx_irq;
1181
1182         lp = ldc_alloc(vdev->channel_id, &ds_cfg, dp);
1183         if (IS_ERR(lp)) {
1184                 err = PTR_ERR(lp);
1185                 goto out_free_ds_states;
1186         }
1187         dp->lp = lp;
1188
1189         err = ldc_bind(lp, "DS");
1190         if (err)
1191                 goto out_free_ldc;
1192
1193         spin_lock_irq(&ds_lock);
1194         dp->next = ds_info_list;
1195         ds_info_list = dp;
1196         spin_unlock_irq(&ds_lock);
1197
1198         return err;
1199
1200 out_free_ldc:
1201         ldc_free(dp->lp);
1202
1203 out_free_ds_states:
1204         kfree(dp->ds_states);
1205
1206 out_free_rcv_buf:
1207         kfree(dp->rcv_buf);
1208
1209 out_free_dp:
1210         kfree(dp);
1211
1212 out_err:
1213         return err;
1214 }
1215
1216 static int ds_remove(struct vio_dev *vdev)
1217 {
1218         return 0;
1219 }
1220
1221 static struct vio_device_id ds_match[] = {
1222         {
1223                 .type = "domain-services-port",
1224         },
1225         {},
1226 };
1227
1228 static struct vio_driver ds_driver = {
1229         .id_table       = ds_match,
1230         .probe          = ds_probe,
1231         .remove         = ds_remove,
1232         .driver         = {
1233                 .name   = "ds",
1234                 .owner  = THIS_MODULE,
1235         }
1236 };
1237
1238 static int __init ds_init(void)
1239 {
1240         kthread_run(ds_thread, NULL, "kldomd");
1241
1242         return vio_register_driver(&ds_driver);
1243 }
1244
1245 subsys_initcall(ds_init);