Merge branch 'for-np' of git://git.wormnet.eu/alex/ts78xx into orion/master
[linux-2.6] / arch / sparc / kernel / ldc.c
1 /* ldc.c: Logical Domain Channel link-layer protocol driver.
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/slab.h>
9 #include <linux/spinlock.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/string.h>
13 #include <linux/scatterlist.h>
14 #include <linux/interrupt.h>
15 #include <linux/list.h>
16 #include <linux/init.h>
17
18 #include <asm/hypervisor.h>
19 #include <asm/iommu.h>
20 #include <asm/page.h>
21 #include <asm/ldc.h>
22 #include <asm/mdesc.h>
23
24 #define DRV_MODULE_NAME         "ldc"
25 #define PFX DRV_MODULE_NAME     ": "
26 #define DRV_MODULE_VERSION      "1.1"
27 #define DRV_MODULE_RELDATE      "July 22, 2008"
28
29 static char version[] __devinitdata =
30         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
31 #define LDC_PACKET_SIZE         64
32
33 /* Packet header layout for unreliable and reliable mode frames.
34  * When in RAW mode, packets are simply straight 64-byte payloads
35  * with no headers.
36  */
37 struct ldc_packet {
38         u8                      type;
39 #define LDC_CTRL                0x01
40 #define LDC_DATA                0x02
41 #define LDC_ERR                 0x10
42
43         u8                      stype;
44 #define LDC_INFO                0x01
45 #define LDC_ACK                 0x02
46 #define LDC_NACK                0x04
47
48         u8                      ctrl;
49 #define LDC_VERS                0x01 /* Link Version            */
50 #define LDC_RTS                 0x02 /* Request To Send         */
51 #define LDC_RTR                 0x03 /* Ready To Receive        */
52 #define LDC_RDX                 0x04 /* Ready for Data eXchange */
53 #define LDC_CTRL_MSK            0x0f
54
55         u8                      env;
56 #define LDC_LEN                 0x3f
57 #define LDC_FRAG_MASK           0xc0
58 #define LDC_START               0x40
59 #define LDC_STOP                0x80
60
61         u32                     seqid;
62
63         union {
64                 u8              u_data[LDC_PACKET_SIZE - 8];
65                 struct {
66                         u32     pad;
67                         u32     ackid;
68                         u8      r_data[LDC_PACKET_SIZE - 8 - 8];
69                 } r;
70         } u;
71 };
72
73 struct ldc_version {
74         u16 major;
75         u16 minor;
76 };
77
78 /* Ordered from largest major to lowest.  */
79 static struct ldc_version ver_arr[] = {
80         { .major = 1, .minor = 0 },
81 };
82
83 #define LDC_DEFAULT_MTU                 (4 * LDC_PACKET_SIZE)
84 #define LDC_DEFAULT_NUM_ENTRIES         (PAGE_SIZE / LDC_PACKET_SIZE)
85
86 struct ldc_channel;
87
88 struct ldc_mode_ops {
89         int (*write)(struct ldc_channel *, const void *, unsigned int);
90         int (*read)(struct ldc_channel *, void *, unsigned int);
91 };
92
93 static const struct ldc_mode_ops raw_ops;
94 static const struct ldc_mode_ops nonraw_ops;
95 static const struct ldc_mode_ops stream_ops;
96
97 int ldom_domaining_enabled;
98
99 struct ldc_iommu {
100         /* Protects arena alloc/free.  */
101         spinlock_t                      lock;
102         struct iommu_arena              arena;
103         struct ldc_mtable_entry         *page_table;
104 };
105
106 struct ldc_channel {
107         /* Protects all operations that depend upon channel state.  */
108         spinlock_t                      lock;
109
110         unsigned long                   id;
111
112         u8                              *mssbuf;
113         u32                             mssbuf_len;
114         u32                             mssbuf_off;
115
116         struct ldc_packet               *tx_base;
117         unsigned long                   tx_head;
118         unsigned long                   tx_tail;
119         unsigned long                   tx_num_entries;
120         unsigned long                   tx_ra;
121
122         unsigned long                   tx_acked;
123
124         struct ldc_packet               *rx_base;
125         unsigned long                   rx_head;
126         unsigned long                   rx_tail;
127         unsigned long                   rx_num_entries;
128         unsigned long                   rx_ra;
129
130         u32                             rcv_nxt;
131         u32                             snd_nxt;
132
133         unsigned long                   chan_state;
134
135         struct ldc_channel_config       cfg;
136         void                            *event_arg;
137
138         const struct ldc_mode_ops       *mops;
139
140         struct ldc_iommu                iommu;
141
142         struct ldc_version              ver;
143
144         u8                              hs_state;
145 #define LDC_HS_CLOSED                   0x00
146 #define LDC_HS_OPEN                     0x01
147 #define LDC_HS_GOTVERS                  0x02
148 #define LDC_HS_SENTRTR                  0x03
149 #define LDC_HS_GOTRTR                   0x04
150 #define LDC_HS_COMPLETE                 0x10
151
152         u8                              flags;
153 #define LDC_FLAG_ALLOCED_QUEUES         0x01
154 #define LDC_FLAG_REGISTERED_QUEUES      0x02
155 #define LDC_FLAG_REGISTERED_IRQS        0x04
156 #define LDC_FLAG_RESET                  0x10
157
158         u8                              mss;
159         u8                              state;
160
161 #define LDC_IRQ_NAME_MAX                32
162         char                            rx_irq_name[LDC_IRQ_NAME_MAX];
163         char                            tx_irq_name[LDC_IRQ_NAME_MAX];
164
165         struct hlist_head               mh_list;
166
167         struct hlist_node               list;
168 };
169
170 #define ldcdbg(TYPE, f, a...) \
171 do {    if (lp->cfg.debug & LDC_DEBUG_##TYPE) \
172                 printk(KERN_INFO PFX "ID[%lu] " f, lp->id, ## a); \
173 } while (0)
174
175 static const char *state_to_str(u8 state)
176 {
177         switch (state) {
178         case LDC_STATE_INVALID:
179                 return "INVALID";
180         case LDC_STATE_INIT:
181                 return "INIT";
182         case LDC_STATE_BOUND:
183                 return "BOUND";
184         case LDC_STATE_READY:
185                 return "READY";
186         case LDC_STATE_CONNECTED:
187                 return "CONNECTED";
188         default:
189                 return "<UNKNOWN>";
190         }
191 }
192
193 static void ldc_set_state(struct ldc_channel *lp, u8 state)
194 {
195         ldcdbg(STATE, "STATE (%s) --> (%s)\n",
196                state_to_str(lp->state),
197                state_to_str(state));
198
199         lp->state = state;
200 }
201
202 static unsigned long __advance(unsigned long off, unsigned long num_entries)
203 {
204         off += LDC_PACKET_SIZE;
205         if (off == (num_entries * LDC_PACKET_SIZE))
206                 off = 0;
207
208         return off;
209 }
210
211 static unsigned long rx_advance(struct ldc_channel *lp, unsigned long off)
212 {
213         return __advance(off, lp->rx_num_entries);
214 }
215
216 static unsigned long tx_advance(struct ldc_channel *lp, unsigned long off)
217 {
218         return __advance(off, lp->tx_num_entries);
219 }
220
221 static struct ldc_packet *handshake_get_tx_packet(struct ldc_channel *lp,
222                                                   unsigned long *new_tail)
223 {
224         struct ldc_packet *p;
225         unsigned long t;
226
227         t = tx_advance(lp, lp->tx_tail);
228         if (t == lp->tx_head)
229                 return NULL;
230
231         *new_tail = t;
232
233         p = lp->tx_base;
234         return p + (lp->tx_tail / LDC_PACKET_SIZE);
235 }
236
237 /* When we are in reliable or stream mode, have to track the next packet
238  * we haven't gotten an ACK for in the TX queue using tx_acked.  We have
239  * to be careful not to stomp over the queue past that point.  During
240  * the handshake, we don't have TX data packets pending in the queue
241  * and that's why handshake_get_tx_packet() need not be mindful of
242  * lp->tx_acked.
243  */
244 static unsigned long head_for_data(struct ldc_channel *lp)
245 {
246         if (lp->cfg.mode == LDC_MODE_STREAM)
247                 return lp->tx_acked;
248         return lp->tx_head;
249 }
250
251 static int tx_has_space_for(struct ldc_channel *lp, unsigned int size)
252 {
253         unsigned long limit, tail, new_tail, diff;
254         unsigned int mss;
255
256         limit = head_for_data(lp);
257         tail = lp->tx_tail;
258         new_tail = tx_advance(lp, tail);
259         if (new_tail == limit)
260                 return 0;
261
262         if (limit > new_tail)
263                 diff = limit - new_tail;
264         else
265                 diff = (limit +
266                         ((lp->tx_num_entries * LDC_PACKET_SIZE) - new_tail));
267         diff /= LDC_PACKET_SIZE;
268         mss = lp->mss;
269
270         if (diff * mss < size)
271                 return 0;
272
273         return 1;
274 }
275
276 static struct ldc_packet *data_get_tx_packet(struct ldc_channel *lp,
277                                              unsigned long *new_tail)
278 {
279         struct ldc_packet *p;
280         unsigned long h, t;
281
282         h = head_for_data(lp);
283         t = tx_advance(lp, lp->tx_tail);
284         if (t == h)
285                 return NULL;
286
287         *new_tail = t;
288
289         p = lp->tx_base;
290         return p + (lp->tx_tail / LDC_PACKET_SIZE);
291 }
292
293 static int set_tx_tail(struct ldc_channel *lp, unsigned long tail)
294 {
295         unsigned long orig_tail = lp->tx_tail;
296         int limit = 1000;
297
298         lp->tx_tail = tail;
299         while (limit-- > 0) {
300                 unsigned long err;
301
302                 err = sun4v_ldc_tx_set_qtail(lp->id, tail);
303                 if (!err)
304                         return 0;
305
306                 if (err != HV_EWOULDBLOCK) {
307                         lp->tx_tail = orig_tail;
308                         return -EINVAL;
309                 }
310                 udelay(1);
311         }
312
313         lp->tx_tail = orig_tail;
314         return -EBUSY;
315 }
316
317 /* This just updates the head value in the hypervisor using
318  * a polling loop with a timeout.  The caller takes care of
319  * upating software state representing the head change, if any.
320  */
321 static int __set_rx_head(struct ldc_channel *lp, unsigned long head)
322 {
323         int limit = 1000;
324
325         while (limit-- > 0) {
326                 unsigned long err;
327
328                 err = sun4v_ldc_rx_set_qhead(lp->id, head);
329                 if (!err)
330                         return 0;
331
332                 if (err != HV_EWOULDBLOCK)
333                         return -EINVAL;
334
335                 udelay(1);
336         }
337
338         return -EBUSY;
339 }
340
341 static int send_tx_packet(struct ldc_channel *lp,
342                           struct ldc_packet *p,
343                           unsigned long new_tail)
344 {
345         BUG_ON(p != (lp->tx_base + (lp->tx_tail / LDC_PACKET_SIZE)));
346
347         return set_tx_tail(lp, new_tail);
348 }
349
350 static struct ldc_packet *handshake_compose_ctrl(struct ldc_channel *lp,
351                                                  u8 stype, u8 ctrl,
352                                                  void *data, int dlen,
353                                                  unsigned long *new_tail)
354 {
355         struct ldc_packet *p = handshake_get_tx_packet(lp, new_tail);
356
357         if (p) {
358                 memset(p, 0, sizeof(*p));
359                 p->type = LDC_CTRL;
360                 p->stype = stype;
361                 p->ctrl = ctrl;
362                 if (data)
363                         memcpy(p->u.u_data, data, dlen);
364         }
365         return p;
366 }
367
368 static int start_handshake(struct ldc_channel *lp)
369 {
370         struct ldc_packet *p;
371         struct ldc_version *ver;
372         unsigned long new_tail;
373
374         ver = &ver_arr[0];
375
376         ldcdbg(HS, "SEND VER INFO maj[%u] min[%u]\n",
377                ver->major, ver->minor);
378
379         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
380                                    ver, sizeof(*ver), &new_tail);
381         if (p) {
382                 int err = send_tx_packet(lp, p, new_tail);
383                 if (!err)
384                         lp->flags &= ~LDC_FLAG_RESET;
385                 return err;
386         }
387         return -EBUSY;
388 }
389
390 static int send_version_nack(struct ldc_channel *lp,
391                              u16 major, u16 minor)
392 {
393         struct ldc_packet *p;
394         struct ldc_version ver;
395         unsigned long new_tail;
396
397         ver.major = major;
398         ver.minor = minor;
399
400         p = handshake_compose_ctrl(lp, LDC_NACK, LDC_VERS,
401                                    &ver, sizeof(ver), &new_tail);
402         if (p) {
403                 ldcdbg(HS, "SEND VER NACK maj[%u] min[%u]\n",
404                        ver.major, ver.minor);
405
406                 return send_tx_packet(lp, p, new_tail);
407         }
408         return -EBUSY;
409 }
410
411 static int send_version_ack(struct ldc_channel *lp,
412                             struct ldc_version *vp)
413 {
414         struct ldc_packet *p;
415         unsigned long new_tail;
416
417         p = handshake_compose_ctrl(lp, LDC_ACK, LDC_VERS,
418                                    vp, sizeof(*vp), &new_tail);
419         if (p) {
420                 ldcdbg(HS, "SEND VER ACK maj[%u] min[%u]\n",
421                        vp->major, vp->minor);
422
423                 return send_tx_packet(lp, p, new_tail);
424         }
425         return -EBUSY;
426 }
427
428 static int send_rts(struct ldc_channel *lp)
429 {
430         struct ldc_packet *p;
431         unsigned long new_tail;
432
433         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTS, NULL, 0,
434                                    &new_tail);
435         if (p) {
436                 p->env = lp->cfg.mode;
437                 p->seqid = 0;
438                 lp->rcv_nxt = 0;
439
440                 ldcdbg(HS, "SEND RTS env[0x%x] seqid[0x%x]\n",
441                        p->env, p->seqid);
442
443                 return send_tx_packet(lp, p, new_tail);
444         }
445         return -EBUSY;
446 }
447
448 static int send_rtr(struct ldc_channel *lp)
449 {
450         struct ldc_packet *p;
451         unsigned long new_tail;
452
453         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTR, NULL, 0,
454                                    &new_tail);
455         if (p) {
456                 p->env = lp->cfg.mode;
457                 p->seqid = 0;
458
459                 ldcdbg(HS, "SEND RTR env[0x%x] seqid[0x%x]\n",
460                        p->env, p->seqid);
461
462                 return send_tx_packet(lp, p, new_tail);
463         }
464         return -EBUSY;
465 }
466
467 static int send_rdx(struct ldc_channel *lp)
468 {
469         struct ldc_packet *p;
470         unsigned long new_tail;
471
472         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RDX, NULL, 0,
473                                    &new_tail);
474         if (p) {
475                 p->env = 0;
476                 p->seqid = ++lp->snd_nxt;
477                 p->u.r.ackid = lp->rcv_nxt;
478
479                 ldcdbg(HS, "SEND RDX env[0x%x] seqid[0x%x] ackid[0x%x]\n",
480                        p->env, p->seqid, p->u.r.ackid);
481
482                 return send_tx_packet(lp, p, new_tail);
483         }
484         return -EBUSY;
485 }
486
487 static int send_data_nack(struct ldc_channel *lp, struct ldc_packet *data_pkt)
488 {
489         struct ldc_packet *p;
490         unsigned long new_tail;
491         int err;
492
493         p = data_get_tx_packet(lp, &new_tail);
494         if (!p)
495                 return -EBUSY;
496         memset(p, 0, sizeof(*p));
497         p->type = data_pkt->type;
498         p->stype = LDC_NACK;
499         p->ctrl = data_pkt->ctrl & LDC_CTRL_MSK;
500         p->seqid = lp->snd_nxt + 1;
501         p->u.r.ackid = lp->rcv_nxt;
502
503         ldcdbg(HS, "SEND DATA NACK type[0x%x] ctl[0x%x] seq[0x%x] ack[0x%x]\n",
504                p->type, p->ctrl, p->seqid, p->u.r.ackid);
505
506         err = send_tx_packet(lp, p, new_tail);
507         if (!err)
508                 lp->snd_nxt++;
509
510         return err;
511 }
512
513 static int ldc_abort(struct ldc_channel *lp)
514 {
515         unsigned long hv_err;
516
517         ldcdbg(STATE, "ABORT\n");
518
519         /* We report but do not act upon the hypervisor errors because
520          * there really isn't much we can do if they fail at this point.
521          */
522         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
523         if (hv_err)
524                 printk(KERN_ERR PFX "ldc_abort: "
525                        "sun4v_ldc_tx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
526                        lp->id, lp->tx_ra, lp->tx_num_entries, hv_err);
527
528         hv_err = sun4v_ldc_tx_get_state(lp->id,
529                                         &lp->tx_head,
530                                         &lp->tx_tail,
531                                         &lp->chan_state);
532         if (hv_err)
533                 printk(KERN_ERR PFX "ldc_abort: "
534                        "sun4v_ldc_tx_get_state(%lx,...) failed, err=%lu\n",
535                        lp->id, hv_err);
536
537         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
538         if (hv_err)
539                 printk(KERN_ERR PFX "ldc_abort: "
540                        "sun4v_ldc_rx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
541                        lp->id, lp->rx_ra, lp->rx_num_entries, hv_err);
542
543         /* Refetch the RX queue state as well, because we could be invoked
544          * here in the queue processing context.
545          */
546         hv_err = sun4v_ldc_rx_get_state(lp->id,
547                                         &lp->rx_head,
548                                         &lp->rx_tail,
549                                         &lp->chan_state);
550         if (hv_err)
551                 printk(KERN_ERR PFX "ldc_abort: "
552                        "sun4v_ldc_rx_get_state(%lx,...) failed, err=%lu\n",
553                        lp->id, hv_err);
554
555         return -ECONNRESET;
556 }
557
558 static struct ldc_version *find_by_major(u16 major)
559 {
560         struct ldc_version *ret = NULL;
561         int i;
562
563         for (i = 0; i < ARRAY_SIZE(ver_arr); i++) {
564                 struct ldc_version *v = &ver_arr[i];
565                 if (v->major <= major) {
566                         ret = v;
567                         break;
568                 }
569         }
570         return ret;
571 }
572
573 static int process_ver_info(struct ldc_channel *lp, struct ldc_version *vp)
574 {
575         struct ldc_version *vap;
576         int err;
577
578         ldcdbg(HS, "GOT VERSION INFO major[%x] minor[%x]\n",
579                vp->major, vp->minor);
580
581         if (lp->hs_state == LDC_HS_GOTVERS) {
582                 lp->hs_state = LDC_HS_OPEN;
583                 memset(&lp->ver, 0, sizeof(lp->ver));
584         }
585
586         vap = find_by_major(vp->major);
587         if (!vap) {
588                 err = send_version_nack(lp, 0, 0);
589         } else if (vap->major != vp->major) {
590                 err = send_version_nack(lp, vap->major, vap->minor);
591         } else {
592                 struct ldc_version ver = *vp;
593                 if (ver.minor > vap->minor)
594                         ver.minor = vap->minor;
595                 err = send_version_ack(lp, &ver);
596                 if (!err) {
597                         lp->ver = ver;
598                         lp->hs_state = LDC_HS_GOTVERS;
599                 }
600         }
601         if (err)
602                 return ldc_abort(lp);
603
604         return 0;
605 }
606
607 static int process_ver_ack(struct ldc_channel *lp, struct ldc_version *vp)
608 {
609         ldcdbg(HS, "GOT VERSION ACK major[%x] minor[%x]\n",
610                vp->major, vp->minor);
611
612         if (lp->hs_state == LDC_HS_GOTVERS) {
613                 if (lp->ver.major != vp->major ||
614                     lp->ver.minor != vp->minor)
615                         return ldc_abort(lp);
616         } else {
617                 lp->ver = *vp;
618                 lp->hs_state = LDC_HS_GOTVERS;
619         }
620         if (send_rts(lp))
621                 return ldc_abort(lp);
622         return 0;
623 }
624
625 static int process_ver_nack(struct ldc_channel *lp, struct ldc_version *vp)
626 {
627         struct ldc_version *vap;
628         struct ldc_packet *p;
629         unsigned long new_tail;
630
631         if (vp->major == 0 && vp->minor == 0)
632                 return ldc_abort(lp);
633
634         vap = find_by_major(vp->major);
635         if (!vap)
636                 return ldc_abort(lp);
637
638         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
639                                            vap, sizeof(*vap),
640                                            &new_tail);
641         if (!p)
642                 return ldc_abort(lp);
643
644         return send_tx_packet(lp, p, new_tail);
645 }
646
647 static int process_version(struct ldc_channel *lp,
648                            struct ldc_packet *p)
649 {
650         struct ldc_version *vp;
651
652         vp = (struct ldc_version *) p->u.u_data;
653
654         switch (p->stype) {
655         case LDC_INFO:
656                 return process_ver_info(lp, vp);
657
658         case LDC_ACK:
659                 return process_ver_ack(lp, vp);
660
661         case LDC_NACK:
662                 return process_ver_nack(lp, vp);
663
664         default:
665                 return ldc_abort(lp);
666         }
667 }
668
669 static int process_rts(struct ldc_channel *lp,
670                        struct ldc_packet *p)
671 {
672         ldcdbg(HS, "GOT RTS stype[%x] seqid[%x] env[%x]\n",
673                p->stype, p->seqid, p->env);
674
675         if (p->stype     != LDC_INFO       ||
676             lp->hs_state != LDC_HS_GOTVERS ||
677             p->env       != lp->cfg.mode)
678                 return ldc_abort(lp);
679
680         lp->snd_nxt = p->seqid;
681         lp->rcv_nxt = p->seqid;
682         lp->hs_state = LDC_HS_SENTRTR;
683         if (send_rtr(lp))
684                 return ldc_abort(lp);
685
686         return 0;
687 }
688
689 static int process_rtr(struct ldc_channel *lp,
690                        struct ldc_packet *p)
691 {
692         ldcdbg(HS, "GOT RTR stype[%x] seqid[%x] env[%x]\n",
693                p->stype, p->seqid, p->env);
694
695         if (p->stype     != LDC_INFO ||
696             p->env       != lp->cfg.mode)
697                 return ldc_abort(lp);
698
699         lp->snd_nxt = p->seqid;
700         lp->hs_state = LDC_HS_COMPLETE;
701         ldc_set_state(lp, LDC_STATE_CONNECTED);
702         send_rdx(lp);
703
704         return LDC_EVENT_UP;
705 }
706
707 static int rx_seq_ok(struct ldc_channel *lp, u32 seqid)
708 {
709         return lp->rcv_nxt + 1 == seqid;
710 }
711
712 static int process_rdx(struct ldc_channel *lp,
713                        struct ldc_packet *p)
714 {
715         ldcdbg(HS, "GOT RDX stype[%x] seqid[%x] env[%x] ackid[%x]\n",
716                p->stype, p->seqid, p->env, p->u.r.ackid);
717
718         if (p->stype != LDC_INFO ||
719             !(rx_seq_ok(lp, p->seqid)))
720                 return ldc_abort(lp);
721
722         lp->rcv_nxt = p->seqid;
723
724         lp->hs_state = LDC_HS_COMPLETE;
725         ldc_set_state(lp, LDC_STATE_CONNECTED);
726
727         return LDC_EVENT_UP;
728 }
729
730 static int process_control_frame(struct ldc_channel *lp,
731                                  struct ldc_packet *p)
732 {
733         switch (p->ctrl) {
734         case LDC_VERS:
735                 return process_version(lp, p);
736
737         case LDC_RTS:
738                 return process_rts(lp, p);
739
740         case LDC_RTR:
741                 return process_rtr(lp, p);
742
743         case LDC_RDX:
744                 return process_rdx(lp, p);
745
746         default:
747                 return ldc_abort(lp);
748         }
749 }
750
751 static int process_error_frame(struct ldc_channel *lp,
752                                struct ldc_packet *p)
753 {
754         return ldc_abort(lp);
755 }
756
757 static int process_data_ack(struct ldc_channel *lp,
758                             struct ldc_packet *ack)
759 {
760         unsigned long head = lp->tx_acked;
761         u32 ackid = ack->u.r.ackid;
762
763         while (1) {
764                 struct ldc_packet *p = lp->tx_base + (head / LDC_PACKET_SIZE);
765
766                 head = tx_advance(lp, head);
767
768                 if (p->seqid == ackid) {
769                         lp->tx_acked = head;
770                         return 0;
771                 }
772                 if (head == lp->tx_tail)
773                         return ldc_abort(lp);
774         }
775
776         return 0;
777 }
778
779 static void send_events(struct ldc_channel *lp, unsigned int event_mask)
780 {
781         if (event_mask & LDC_EVENT_RESET)
782                 lp->cfg.event(lp->event_arg, LDC_EVENT_RESET);
783         if (event_mask & LDC_EVENT_UP)
784                 lp->cfg.event(lp->event_arg, LDC_EVENT_UP);
785         if (event_mask & LDC_EVENT_DATA_READY)
786                 lp->cfg.event(lp->event_arg, LDC_EVENT_DATA_READY);
787 }
788
789 static irqreturn_t ldc_rx(int irq, void *dev_id)
790 {
791         struct ldc_channel *lp = dev_id;
792         unsigned long orig_state, hv_err, flags;
793         unsigned int event_mask;
794
795         spin_lock_irqsave(&lp->lock, flags);
796
797         orig_state = lp->chan_state;
798         hv_err = sun4v_ldc_rx_get_state(lp->id,
799                                         &lp->rx_head,
800                                         &lp->rx_tail,
801                                         &lp->chan_state);
802
803         ldcdbg(RX, "RX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
804                orig_state, lp->chan_state, lp->rx_head, lp->rx_tail);
805
806         event_mask = 0;
807
808         if (lp->cfg.mode == LDC_MODE_RAW &&
809             lp->chan_state == LDC_CHANNEL_UP) {
810                 lp->hs_state = LDC_HS_COMPLETE;
811                 ldc_set_state(lp, LDC_STATE_CONNECTED);
812
813                 event_mask |= LDC_EVENT_UP;
814
815                 orig_state = lp->chan_state;
816         }
817
818         /* If we are in reset state, flush the RX queue and ignore
819          * everything.
820          */
821         if (lp->flags & LDC_FLAG_RESET) {
822                 (void) __set_rx_head(lp, lp->rx_tail);
823                 goto out;
824         }
825
826         /* Once we finish the handshake, we let the ldc_read()
827          * paths do all of the control frame and state management.
828          * Just trigger the callback.
829          */
830         if (lp->hs_state == LDC_HS_COMPLETE) {
831 handshake_complete:
832                 if (lp->chan_state != orig_state) {
833                         unsigned int event = LDC_EVENT_RESET;
834
835                         if (lp->chan_state == LDC_CHANNEL_UP)
836                                 event = LDC_EVENT_UP;
837
838                         event_mask |= event;
839                 }
840                 if (lp->rx_head != lp->rx_tail)
841                         event_mask |= LDC_EVENT_DATA_READY;
842
843                 goto out;
844         }
845
846         if (lp->chan_state != orig_state)
847                 goto out;
848
849         while (lp->rx_head != lp->rx_tail) {
850                 struct ldc_packet *p;
851                 unsigned long new;
852                 int err;
853
854                 p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
855
856                 switch (p->type) {
857                 case LDC_CTRL:
858                         err = process_control_frame(lp, p);
859                         if (err > 0)
860                                 event_mask |= err;
861                         break;
862
863                 case LDC_DATA:
864                         event_mask |= LDC_EVENT_DATA_READY;
865                         err = 0;
866                         break;
867
868                 case LDC_ERR:
869                         err = process_error_frame(lp, p);
870                         break;
871
872                 default:
873                         err = ldc_abort(lp);
874                         break;
875                 }
876
877                 if (err < 0)
878                         break;
879
880                 new = lp->rx_head;
881                 new += LDC_PACKET_SIZE;
882                 if (new == (lp->rx_num_entries * LDC_PACKET_SIZE))
883                         new = 0;
884                 lp->rx_head = new;
885
886                 err = __set_rx_head(lp, new);
887                 if (err < 0) {
888                         (void) ldc_abort(lp);
889                         break;
890                 }
891                 if (lp->hs_state == LDC_HS_COMPLETE)
892                         goto handshake_complete;
893         }
894
895 out:
896         spin_unlock_irqrestore(&lp->lock, flags);
897
898         send_events(lp, event_mask);
899
900         return IRQ_HANDLED;
901 }
902
903 static irqreturn_t ldc_tx(int irq, void *dev_id)
904 {
905         struct ldc_channel *lp = dev_id;
906         unsigned long flags, hv_err, orig_state;
907         unsigned int event_mask = 0;
908
909         spin_lock_irqsave(&lp->lock, flags);
910
911         orig_state = lp->chan_state;
912         hv_err = sun4v_ldc_tx_get_state(lp->id,
913                                         &lp->tx_head,
914                                         &lp->tx_tail,
915                                         &lp->chan_state);
916
917         ldcdbg(TX, " TX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
918                orig_state, lp->chan_state, lp->tx_head, lp->tx_tail);
919
920         if (lp->cfg.mode == LDC_MODE_RAW &&
921             lp->chan_state == LDC_CHANNEL_UP) {
922                 lp->hs_state = LDC_HS_COMPLETE;
923                 ldc_set_state(lp, LDC_STATE_CONNECTED);
924
925                 event_mask |= LDC_EVENT_UP;
926         }
927
928         spin_unlock_irqrestore(&lp->lock, flags);
929
930         send_events(lp, event_mask);
931
932         return IRQ_HANDLED;
933 }
934
935 /* XXX ldc_alloc() and ldc_free() needs to run under a mutex so
936  * XXX that addition and removal from the ldc_channel_list has
937  * XXX atomicity, otherwise the __ldc_channel_exists() check is
938  * XXX totally pointless as another thread can slip into ldc_alloc()
939  * XXX and add a channel with the same ID.  There also needs to be
940  * XXX a spinlock for ldc_channel_list.
941  */
942 static HLIST_HEAD(ldc_channel_list);
943
944 static int __ldc_channel_exists(unsigned long id)
945 {
946         struct ldc_channel *lp;
947         struct hlist_node *n;
948
949         hlist_for_each_entry(lp, n, &ldc_channel_list, list) {
950                 if (lp->id == id)
951                         return 1;
952         }
953         return 0;
954 }
955
956 static int alloc_queue(const char *name, unsigned long num_entries,
957                        struct ldc_packet **base, unsigned long *ra)
958 {
959         unsigned long size, order;
960         void *q;
961
962         size = num_entries * LDC_PACKET_SIZE;
963         order = get_order(size);
964
965         q = (void *) __get_free_pages(GFP_KERNEL, order);
966         if (!q) {
967                 printk(KERN_ERR PFX "Alloc of %s queue failed with "
968                        "size=%lu order=%lu\n", name, size, order);
969                 return -ENOMEM;
970         }
971
972         memset(q, 0, PAGE_SIZE << order);
973
974         *base = q;
975         *ra = __pa(q);
976
977         return 0;
978 }
979
980 static void free_queue(unsigned long num_entries, struct ldc_packet *q)
981 {
982         unsigned long size, order;
983
984         if (!q)
985                 return;
986
987         size = num_entries * LDC_PACKET_SIZE;
988         order = get_order(size);
989
990         free_pages((unsigned long)q, order);
991 }
992
993 /* XXX Make this configurable... XXX */
994 #define LDC_IOTABLE_SIZE        (8 * 1024)
995
996 static int ldc_iommu_init(struct ldc_channel *lp)
997 {
998         unsigned long sz, num_tsb_entries, tsbsize, order;
999         struct ldc_iommu *iommu = &lp->iommu;
1000         struct ldc_mtable_entry *table;
1001         unsigned long hv_err;
1002         int err;
1003
1004         num_tsb_entries = LDC_IOTABLE_SIZE;
1005         tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1006
1007         spin_lock_init(&iommu->lock);
1008
1009         sz = num_tsb_entries / 8;
1010         sz = (sz + 7UL) & ~7UL;
1011         iommu->arena.map = kzalloc(sz, GFP_KERNEL);
1012         if (!iommu->arena.map) {
1013                 printk(KERN_ERR PFX "Alloc of arena map failed, sz=%lu\n", sz);
1014                 return -ENOMEM;
1015         }
1016
1017         iommu->arena.limit = num_tsb_entries;
1018
1019         order = get_order(tsbsize);
1020
1021         table = (struct ldc_mtable_entry *)
1022                 __get_free_pages(GFP_KERNEL, order);
1023         err = -ENOMEM;
1024         if (!table) {
1025                 printk(KERN_ERR PFX "Alloc of MTE table failed, "
1026                        "size=%lu order=%lu\n", tsbsize, order);
1027                 goto out_free_map;
1028         }
1029
1030         memset(table, 0, PAGE_SIZE << order);
1031
1032         iommu->page_table = table;
1033
1034         hv_err = sun4v_ldc_set_map_table(lp->id, __pa(table),
1035                                          num_tsb_entries);
1036         err = -EINVAL;
1037         if (hv_err)
1038                 goto out_free_table;
1039
1040         return 0;
1041
1042 out_free_table:
1043         free_pages((unsigned long) table, order);
1044         iommu->page_table = NULL;
1045
1046 out_free_map:
1047         kfree(iommu->arena.map);
1048         iommu->arena.map = NULL;
1049
1050         return err;
1051 }
1052
1053 static void ldc_iommu_release(struct ldc_channel *lp)
1054 {
1055         struct ldc_iommu *iommu = &lp->iommu;
1056         unsigned long num_tsb_entries, tsbsize, order;
1057
1058         (void) sun4v_ldc_set_map_table(lp->id, 0, 0);
1059
1060         num_tsb_entries = iommu->arena.limit;
1061         tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1062         order = get_order(tsbsize);
1063
1064         free_pages((unsigned long) iommu->page_table, order);
1065         iommu->page_table = NULL;
1066
1067         kfree(iommu->arena.map);
1068         iommu->arena.map = NULL;
1069 }
1070
1071 struct ldc_channel *ldc_alloc(unsigned long id,
1072                               const struct ldc_channel_config *cfgp,
1073                               void *event_arg)
1074 {
1075         struct ldc_channel *lp;
1076         const struct ldc_mode_ops *mops;
1077         unsigned long dummy1, dummy2, hv_err;
1078         u8 mss, *mssbuf;
1079         int err;
1080
1081         err = -ENODEV;
1082         if (!ldom_domaining_enabled)
1083                 goto out_err;
1084
1085         err = -EINVAL;
1086         if (!cfgp)
1087                 goto out_err;
1088
1089         switch (cfgp->mode) {
1090         case LDC_MODE_RAW:
1091                 mops = &raw_ops;
1092                 mss = LDC_PACKET_SIZE;
1093                 break;
1094
1095         case LDC_MODE_UNRELIABLE:
1096                 mops = &nonraw_ops;
1097                 mss = LDC_PACKET_SIZE - 8;
1098                 break;
1099
1100         case LDC_MODE_STREAM:
1101                 mops = &stream_ops;
1102                 mss = LDC_PACKET_SIZE - 8 - 8;
1103                 break;
1104
1105         default:
1106                 goto out_err;
1107         }
1108
1109         if (!cfgp->event || !event_arg || !cfgp->rx_irq || !cfgp->tx_irq)
1110                 goto out_err;
1111
1112         hv_err = sun4v_ldc_tx_qinfo(id, &dummy1, &dummy2);
1113         err = -ENODEV;
1114         if (hv_err == HV_ECHANNEL)
1115                 goto out_err;
1116
1117         err = -EEXIST;
1118         if (__ldc_channel_exists(id))
1119                 goto out_err;
1120
1121         mssbuf = NULL;
1122
1123         lp = kzalloc(sizeof(*lp), GFP_KERNEL);
1124         err = -ENOMEM;
1125         if (!lp)
1126                 goto out_err;
1127
1128         spin_lock_init(&lp->lock);
1129
1130         lp->id = id;
1131
1132         err = ldc_iommu_init(lp);
1133         if (err)
1134                 goto out_free_ldc;
1135
1136         lp->mops = mops;
1137         lp->mss = mss;
1138
1139         lp->cfg = *cfgp;
1140         if (!lp->cfg.mtu)
1141                 lp->cfg.mtu = LDC_DEFAULT_MTU;
1142
1143         if (lp->cfg.mode == LDC_MODE_STREAM) {
1144                 mssbuf = kzalloc(lp->cfg.mtu, GFP_KERNEL);
1145                 if (!mssbuf) {
1146                         err = -ENOMEM;
1147                         goto out_free_iommu;
1148                 }
1149                 lp->mssbuf = mssbuf;
1150         }
1151
1152         lp->event_arg = event_arg;
1153
1154         /* XXX allow setting via ldc_channel_config to override defaults
1155          * XXX or use some formula based upon mtu
1156          */
1157         lp->tx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1158         lp->rx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1159
1160         err = alloc_queue("TX", lp->tx_num_entries,
1161                           &lp->tx_base, &lp->tx_ra);
1162         if (err)
1163                 goto out_free_mssbuf;
1164
1165         err = alloc_queue("RX", lp->rx_num_entries,
1166                           &lp->rx_base, &lp->rx_ra);
1167         if (err)
1168                 goto out_free_txq;
1169
1170         lp->flags |= LDC_FLAG_ALLOCED_QUEUES;
1171
1172         lp->hs_state = LDC_HS_CLOSED;
1173         ldc_set_state(lp, LDC_STATE_INIT);
1174
1175         INIT_HLIST_NODE(&lp->list);
1176         hlist_add_head(&lp->list, &ldc_channel_list);
1177
1178         INIT_HLIST_HEAD(&lp->mh_list);
1179
1180         return lp;
1181
1182 out_free_txq:
1183         free_queue(lp->tx_num_entries, lp->tx_base);
1184
1185 out_free_mssbuf:
1186         if (mssbuf)
1187                 kfree(mssbuf);
1188
1189 out_free_iommu:
1190         ldc_iommu_release(lp);
1191
1192 out_free_ldc:
1193         kfree(lp);
1194
1195 out_err:
1196         return ERR_PTR(err);
1197 }
1198 EXPORT_SYMBOL(ldc_alloc);
1199
1200 void ldc_free(struct ldc_channel *lp)
1201 {
1202         if (lp->flags & LDC_FLAG_REGISTERED_IRQS) {
1203                 free_irq(lp->cfg.rx_irq, lp);
1204                 free_irq(lp->cfg.tx_irq, lp);
1205         }
1206
1207         if (lp->flags & LDC_FLAG_REGISTERED_QUEUES) {
1208                 sun4v_ldc_tx_qconf(lp->id, 0, 0);
1209                 sun4v_ldc_rx_qconf(lp->id, 0, 0);
1210                 lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1211         }
1212         if (lp->flags & LDC_FLAG_ALLOCED_QUEUES) {
1213                 free_queue(lp->tx_num_entries, lp->tx_base);
1214                 free_queue(lp->rx_num_entries, lp->rx_base);
1215                 lp->flags &= ~LDC_FLAG_ALLOCED_QUEUES;
1216         }
1217
1218         hlist_del(&lp->list);
1219
1220         if (lp->mssbuf)
1221                 kfree(lp->mssbuf);
1222
1223         ldc_iommu_release(lp);
1224
1225         kfree(lp);
1226 }
1227 EXPORT_SYMBOL(ldc_free);
1228
1229 /* Bind the channel.  This registers the LDC queues with
1230  * the hypervisor and puts the channel into a pseudo-listening
1231  * state.  This does not initiate a handshake, ldc_connect() does
1232  * that.
1233  */
1234 int ldc_bind(struct ldc_channel *lp, const char *name)
1235 {
1236         unsigned long hv_err, flags;
1237         int err = -EINVAL;
1238
1239         if (!name ||
1240             (lp->state != LDC_STATE_INIT))
1241                 return -EINVAL;
1242
1243         snprintf(lp->rx_irq_name, LDC_IRQ_NAME_MAX, "%s RX", name);
1244         snprintf(lp->tx_irq_name, LDC_IRQ_NAME_MAX, "%s TX", name);
1245
1246         err = request_irq(lp->cfg.rx_irq, ldc_rx,
1247                           IRQF_SAMPLE_RANDOM | IRQF_SHARED,
1248                           lp->rx_irq_name, lp);
1249         if (err)
1250                 return err;
1251
1252         err = request_irq(lp->cfg.tx_irq, ldc_tx,
1253                           IRQF_SAMPLE_RANDOM | IRQF_SHARED,
1254                           lp->tx_irq_name, lp);
1255         if (err) {
1256                 free_irq(lp->cfg.rx_irq, lp);
1257                 return err;
1258         }
1259
1260
1261         spin_lock_irqsave(&lp->lock, flags);
1262
1263         enable_irq(lp->cfg.rx_irq);
1264         enable_irq(lp->cfg.tx_irq);
1265
1266         lp->flags |= LDC_FLAG_REGISTERED_IRQS;
1267
1268         err = -ENODEV;
1269         hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1270         if (hv_err)
1271                 goto out_free_irqs;
1272
1273         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1274         if (hv_err)
1275                 goto out_free_irqs;
1276
1277         hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1278         if (hv_err)
1279                 goto out_unmap_tx;
1280
1281         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1282         if (hv_err)
1283                 goto out_unmap_tx;
1284
1285         lp->flags |= LDC_FLAG_REGISTERED_QUEUES;
1286
1287         hv_err = sun4v_ldc_tx_get_state(lp->id,
1288                                         &lp->tx_head,
1289                                         &lp->tx_tail,
1290                                         &lp->chan_state);
1291         err = -EBUSY;
1292         if (hv_err)
1293                 goto out_unmap_rx;
1294
1295         lp->tx_acked = lp->tx_head;
1296
1297         lp->hs_state = LDC_HS_OPEN;
1298         ldc_set_state(lp, LDC_STATE_BOUND);
1299
1300         spin_unlock_irqrestore(&lp->lock, flags);
1301
1302         return 0;
1303
1304 out_unmap_rx:
1305         lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1306         sun4v_ldc_rx_qconf(lp->id, 0, 0);
1307
1308 out_unmap_tx:
1309         sun4v_ldc_tx_qconf(lp->id, 0, 0);
1310
1311 out_free_irqs:
1312         lp->flags &= ~LDC_FLAG_REGISTERED_IRQS;
1313         free_irq(lp->cfg.tx_irq, lp);
1314         free_irq(lp->cfg.rx_irq, lp);
1315
1316         spin_unlock_irqrestore(&lp->lock, flags);
1317
1318         return err;
1319 }
1320 EXPORT_SYMBOL(ldc_bind);
1321
1322 int ldc_connect(struct ldc_channel *lp)
1323 {
1324         unsigned long flags;
1325         int err;
1326
1327         if (lp->cfg.mode == LDC_MODE_RAW)
1328                 return -EINVAL;
1329
1330         spin_lock_irqsave(&lp->lock, flags);
1331
1332         if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1333             !(lp->flags & LDC_FLAG_REGISTERED_QUEUES) ||
1334             lp->hs_state != LDC_HS_OPEN)
1335                 err = -EINVAL;
1336         else
1337                 err = start_handshake(lp);
1338
1339         spin_unlock_irqrestore(&lp->lock, flags);
1340
1341         return err;
1342 }
1343 EXPORT_SYMBOL(ldc_connect);
1344
1345 int ldc_disconnect(struct ldc_channel *lp)
1346 {
1347         unsigned long hv_err, flags;
1348         int err;
1349
1350         if (lp->cfg.mode == LDC_MODE_RAW)
1351                 return -EINVAL;
1352
1353         if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1354             !(lp->flags & LDC_FLAG_REGISTERED_QUEUES))
1355                 return -EINVAL;
1356
1357         spin_lock_irqsave(&lp->lock, flags);
1358
1359         err = -ENODEV;
1360         hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1361         if (hv_err)
1362                 goto out_err;
1363
1364         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1365         if (hv_err)
1366                 goto out_err;
1367
1368         hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1369         if (hv_err)
1370                 goto out_err;
1371
1372         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1373         if (hv_err)
1374                 goto out_err;
1375
1376         ldc_set_state(lp, LDC_STATE_BOUND);
1377         lp->hs_state = LDC_HS_OPEN;
1378         lp->flags |= LDC_FLAG_RESET;
1379
1380         spin_unlock_irqrestore(&lp->lock, flags);
1381
1382         return 0;
1383
1384 out_err:
1385         sun4v_ldc_tx_qconf(lp->id, 0, 0);
1386         sun4v_ldc_rx_qconf(lp->id, 0, 0);
1387         free_irq(lp->cfg.tx_irq, lp);
1388         free_irq(lp->cfg.rx_irq, lp);
1389         lp->flags &= ~(LDC_FLAG_REGISTERED_IRQS |
1390                        LDC_FLAG_REGISTERED_QUEUES);
1391         ldc_set_state(lp, LDC_STATE_INIT);
1392
1393         spin_unlock_irqrestore(&lp->lock, flags);
1394
1395         return err;
1396 }
1397 EXPORT_SYMBOL(ldc_disconnect);
1398
1399 int ldc_state(struct ldc_channel *lp)
1400 {
1401         return lp->state;
1402 }
1403 EXPORT_SYMBOL(ldc_state);
1404
1405 static int write_raw(struct ldc_channel *lp, const void *buf, unsigned int size)
1406 {
1407         struct ldc_packet *p;
1408         unsigned long new_tail;
1409         int err;
1410
1411         if (size > LDC_PACKET_SIZE)
1412                 return -EMSGSIZE;
1413
1414         p = data_get_tx_packet(lp, &new_tail);
1415         if (!p)
1416                 return -EAGAIN;
1417
1418         memcpy(p, buf, size);
1419
1420         err = send_tx_packet(lp, p, new_tail);
1421         if (!err)
1422                 err = size;
1423
1424         return err;
1425 }
1426
1427 static int read_raw(struct ldc_channel *lp, void *buf, unsigned int size)
1428 {
1429         struct ldc_packet *p;
1430         unsigned long hv_err, new;
1431         int err;
1432
1433         if (size < LDC_PACKET_SIZE)
1434                 return -EINVAL;
1435
1436         hv_err = sun4v_ldc_rx_get_state(lp->id,
1437                                         &lp->rx_head,
1438                                         &lp->rx_tail,
1439                                         &lp->chan_state);
1440         if (hv_err)
1441                 return ldc_abort(lp);
1442
1443         if (lp->chan_state == LDC_CHANNEL_DOWN ||
1444             lp->chan_state == LDC_CHANNEL_RESETTING)
1445                 return -ECONNRESET;
1446
1447         if (lp->rx_head == lp->rx_tail)
1448                 return 0;
1449
1450         p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
1451         memcpy(buf, p, LDC_PACKET_SIZE);
1452
1453         new = rx_advance(lp, lp->rx_head);
1454         lp->rx_head = new;
1455
1456         err = __set_rx_head(lp, new);
1457         if (err < 0)
1458                 err = -ECONNRESET;
1459         else
1460                 err = LDC_PACKET_SIZE;
1461
1462         return err;
1463 }
1464
1465 static const struct ldc_mode_ops raw_ops = {
1466         .write          =       write_raw,
1467         .read           =       read_raw,
1468 };
1469
1470 static int write_nonraw(struct ldc_channel *lp, const void *buf,
1471                         unsigned int size)
1472 {
1473         unsigned long hv_err, tail;
1474         unsigned int copied;
1475         u32 seq;
1476         int err;
1477
1478         hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail,
1479                                         &lp->chan_state);
1480         if (unlikely(hv_err))
1481                 return -EBUSY;
1482
1483         if (unlikely(lp->chan_state != LDC_CHANNEL_UP))
1484                 return ldc_abort(lp);
1485
1486         if (!tx_has_space_for(lp, size))
1487                 return -EAGAIN;
1488
1489         seq = lp->snd_nxt;
1490         copied = 0;
1491         tail = lp->tx_tail;
1492         while (copied < size) {
1493                 struct ldc_packet *p = lp->tx_base + (tail / LDC_PACKET_SIZE);
1494                 u8 *data = ((lp->cfg.mode == LDC_MODE_UNRELIABLE) ?
1495                             p->u.u_data :
1496                             p->u.r.r_data);
1497                 int data_len;
1498
1499                 p->type = LDC_DATA;
1500                 p->stype = LDC_INFO;
1501                 p->ctrl = 0;
1502
1503                 data_len = size - copied;
1504                 if (data_len > lp->mss)
1505                         data_len = lp->mss;
1506
1507                 BUG_ON(data_len > LDC_LEN);
1508
1509                 p->env = (data_len |
1510                           (copied == 0 ? LDC_START : 0) |
1511                           (data_len == size - copied ? LDC_STOP : 0));
1512
1513                 p->seqid = ++seq;
1514
1515                 ldcdbg(DATA, "SENT DATA [%02x:%02x:%02x:%02x:%08x]\n",
1516                        p->type,
1517                        p->stype,
1518                        p->ctrl,
1519                        p->env,
1520                        p->seqid);
1521
1522                 memcpy(data, buf, data_len);
1523                 buf += data_len;
1524                 copied += data_len;
1525
1526                 tail = tx_advance(lp, tail);
1527         }
1528
1529         err = set_tx_tail(lp, tail);
1530         if (!err) {
1531                 lp->snd_nxt = seq;
1532                 err = size;
1533         }
1534
1535         return err;
1536 }
1537
1538 static int rx_bad_seq(struct ldc_channel *lp, struct ldc_packet *p,
1539                       struct ldc_packet *first_frag)
1540 {
1541         int err;
1542
1543         if (first_frag)
1544                 lp->rcv_nxt = first_frag->seqid - 1;
1545
1546         err = send_data_nack(lp, p);
1547         if (err)
1548                 return err;
1549
1550         err = __set_rx_head(lp, lp->rx_tail);
1551         if (err < 0)
1552                 return ldc_abort(lp);
1553
1554         return 0;
1555 }
1556
1557 static int data_ack_nack(struct ldc_channel *lp, struct ldc_packet *p)
1558 {
1559         if (p->stype & LDC_ACK) {
1560                 int err = process_data_ack(lp, p);
1561                 if (err)
1562                         return err;
1563         }
1564         if (p->stype & LDC_NACK)
1565                 return ldc_abort(lp);
1566
1567         return 0;
1568 }
1569
1570 static int rx_data_wait(struct ldc_channel *lp, unsigned long cur_head)
1571 {
1572         unsigned long dummy;
1573         int limit = 1000;
1574
1575         ldcdbg(DATA, "DATA WAIT cur_head[%lx] rx_head[%lx] rx_tail[%lx]\n",
1576                cur_head, lp->rx_head, lp->rx_tail);
1577         while (limit-- > 0) {
1578                 unsigned long hv_err;
1579
1580                 hv_err = sun4v_ldc_rx_get_state(lp->id,
1581                                                 &dummy,
1582                                                 &lp->rx_tail,
1583                                                 &lp->chan_state);
1584                 if (hv_err)
1585                         return ldc_abort(lp);
1586
1587                 if (lp->chan_state == LDC_CHANNEL_DOWN ||
1588                     lp->chan_state == LDC_CHANNEL_RESETTING)
1589                         return -ECONNRESET;
1590
1591                 if (cur_head != lp->rx_tail) {
1592                         ldcdbg(DATA, "DATA WAIT DONE "
1593                                "head[%lx] tail[%lx] chan_state[%lx]\n",
1594                                dummy, lp->rx_tail, lp->chan_state);
1595                         return 0;
1596                 }
1597
1598                 udelay(1);
1599         }
1600         return -EAGAIN;
1601 }
1602
1603 static int rx_set_head(struct ldc_channel *lp, unsigned long head)
1604 {
1605         int err = __set_rx_head(lp, head);
1606
1607         if (err < 0)
1608                 return ldc_abort(lp);
1609
1610         lp->rx_head = head;
1611         return 0;
1612 }
1613
1614 static void send_data_ack(struct ldc_channel *lp)
1615 {
1616         unsigned long new_tail;
1617         struct ldc_packet *p;
1618
1619         p = data_get_tx_packet(lp, &new_tail);
1620         if (likely(p)) {
1621                 int err;
1622
1623                 memset(p, 0, sizeof(*p));
1624                 p->type = LDC_DATA;
1625                 p->stype = LDC_ACK;
1626                 p->ctrl = 0;
1627                 p->seqid = lp->snd_nxt + 1;
1628                 p->u.r.ackid = lp->rcv_nxt;
1629
1630                 err = send_tx_packet(lp, p, new_tail);
1631                 if (!err)
1632                         lp->snd_nxt++;
1633         }
1634 }
1635
1636 static int read_nonraw(struct ldc_channel *lp, void *buf, unsigned int size)
1637 {
1638         struct ldc_packet *first_frag;
1639         unsigned long hv_err, new;
1640         int err, copied;
1641
1642         hv_err = sun4v_ldc_rx_get_state(lp->id,
1643                                         &lp->rx_head,
1644                                         &lp->rx_tail,
1645                                         &lp->chan_state);
1646         if (hv_err)
1647                 return ldc_abort(lp);
1648
1649         if (lp->chan_state == LDC_CHANNEL_DOWN ||
1650             lp->chan_state == LDC_CHANNEL_RESETTING)
1651                 return -ECONNRESET;
1652
1653         if (lp->rx_head == lp->rx_tail)
1654                 return 0;
1655
1656         first_frag = NULL;
1657         copied = err = 0;
1658         new = lp->rx_head;
1659         while (1) {
1660                 struct ldc_packet *p;
1661                 int pkt_len;
1662
1663                 BUG_ON(new == lp->rx_tail);
1664                 p = lp->rx_base + (new / LDC_PACKET_SIZE);
1665
1666                 ldcdbg(RX, "RX read pkt[%02x:%02x:%02x:%02x:%08x:%08x] "
1667                        "rcv_nxt[%08x]\n",
1668                        p->type,
1669                        p->stype,
1670                        p->ctrl,
1671                        p->env,
1672                        p->seqid,
1673                        p->u.r.ackid,
1674                        lp->rcv_nxt);
1675
1676                 if (unlikely(!rx_seq_ok(lp, p->seqid))) {
1677                         err = rx_bad_seq(lp, p, first_frag);
1678                         copied = 0;
1679                         break;
1680                 }
1681
1682                 if (p->type & LDC_CTRL) {
1683                         err = process_control_frame(lp, p);
1684                         if (err < 0)
1685                                 break;
1686                         err = 0;
1687                 }
1688
1689                 lp->rcv_nxt = p->seqid;
1690
1691                 if (!(p->type & LDC_DATA)) {
1692                         new = rx_advance(lp, new);
1693                         goto no_data;
1694                 }
1695                 if (p->stype & (LDC_ACK | LDC_NACK)) {
1696                         err = data_ack_nack(lp, p);
1697                         if (err)
1698                                 break;
1699                 }
1700                 if (!(p->stype & LDC_INFO)) {
1701                         new = rx_advance(lp, new);
1702                         err = rx_set_head(lp, new);
1703                         if (err)
1704                                 break;
1705                         goto no_data;
1706                 }
1707
1708                 pkt_len = p->env & LDC_LEN;
1709
1710                 /* Every initial packet starts with the START bit set.
1711                  *
1712                  * Singleton packets will have both START+STOP set.
1713                  *
1714                  * Fragments will have START set in the first frame, STOP
1715                  * set in the last frame, and neither bit set in middle
1716                  * frames of the packet.
1717                  *
1718                  * Therefore if we are at the beginning of a packet and
1719                  * we don't see START, or we are in the middle of a fragmented
1720                  * packet and do see START, we are unsynchronized and should
1721                  * flush the RX queue.
1722                  */
1723                 if ((first_frag == NULL && !(p->env & LDC_START)) ||
1724                     (first_frag != NULL &&  (p->env & LDC_START))) {
1725                         if (!first_frag)
1726                                 new = rx_advance(lp, new);
1727
1728                         err = rx_set_head(lp, new);
1729                         if (err)
1730                                 break;
1731
1732                         if (!first_frag)
1733                                 goto no_data;
1734                 }
1735                 if (!first_frag)
1736                         first_frag = p;
1737
1738                 if (pkt_len > size - copied) {
1739                         /* User didn't give us a big enough buffer,
1740                          * what to do?  This is a pretty serious error.
1741                          *
1742                          * Since we haven't updated the RX ring head to
1743                          * consume any of the packets, signal the error
1744                          * to the user and just leave the RX ring alone.
1745                          *
1746                          * This seems the best behavior because this allows
1747                          * a user of the LDC layer to start with a small
1748                          * RX buffer for ldc_read() calls and use -EMSGSIZE
1749                          * as a cue to enlarge it's read buffer.
1750                          */
1751                         err = -EMSGSIZE;
1752                         break;
1753                 }
1754
1755                 /* Ok, we are gonna eat this one.  */
1756                 new = rx_advance(lp, new);
1757
1758                 memcpy(buf,
1759                        (lp->cfg.mode == LDC_MODE_UNRELIABLE ?
1760                         p->u.u_data : p->u.r.r_data), pkt_len);
1761                 buf += pkt_len;
1762                 copied += pkt_len;
1763
1764                 if (p->env & LDC_STOP)
1765                         break;
1766
1767 no_data:
1768                 if (new == lp->rx_tail) {
1769                         err = rx_data_wait(lp, new);
1770                         if (err)
1771                                 break;
1772                 }
1773         }
1774
1775         if (!err)
1776                 err = rx_set_head(lp, new);
1777
1778         if (err && first_frag)
1779                 lp->rcv_nxt = first_frag->seqid - 1;
1780
1781         if (!err) {
1782                 err = copied;
1783                 if (err > 0 && lp->cfg.mode != LDC_MODE_UNRELIABLE)
1784                         send_data_ack(lp);
1785         }
1786
1787         return err;
1788 }
1789
1790 static const struct ldc_mode_ops nonraw_ops = {
1791         .write          =       write_nonraw,
1792         .read           =       read_nonraw,
1793 };
1794
1795 static int write_stream(struct ldc_channel *lp, const void *buf,
1796                         unsigned int size)
1797 {
1798         if (size > lp->cfg.mtu)
1799                 size = lp->cfg.mtu;
1800         return write_nonraw(lp, buf, size);
1801 }
1802
1803 static int read_stream(struct ldc_channel *lp, void *buf, unsigned int size)
1804 {
1805         if (!lp->mssbuf_len) {
1806                 int err = read_nonraw(lp, lp->mssbuf, lp->cfg.mtu);
1807                 if (err < 0)
1808                         return err;
1809
1810                 lp->mssbuf_len = err;
1811                 lp->mssbuf_off = 0;
1812         }
1813
1814         if (size > lp->mssbuf_len)
1815                 size = lp->mssbuf_len;
1816         memcpy(buf, lp->mssbuf + lp->mssbuf_off, size);
1817
1818         lp->mssbuf_off += size;
1819         lp->mssbuf_len -= size;
1820
1821         return size;
1822 }
1823
1824 static const struct ldc_mode_ops stream_ops = {
1825         .write          =       write_stream,
1826         .read           =       read_stream,
1827 };
1828
1829 int ldc_write(struct ldc_channel *lp, const void *buf, unsigned int size)
1830 {
1831         unsigned long flags;
1832         int err;
1833
1834         if (!buf)
1835                 return -EINVAL;
1836
1837         if (!size)
1838                 return 0;
1839
1840         spin_lock_irqsave(&lp->lock, flags);
1841
1842         if (lp->hs_state != LDC_HS_COMPLETE)
1843                 err = -ENOTCONN;
1844         else
1845                 err = lp->mops->write(lp, buf, size);
1846
1847         spin_unlock_irqrestore(&lp->lock, flags);
1848
1849         return err;
1850 }
1851 EXPORT_SYMBOL(ldc_write);
1852
1853 int ldc_read(struct ldc_channel *lp, void *buf, unsigned int size)
1854 {
1855         unsigned long flags;
1856         int err;
1857
1858         if (!buf)
1859                 return -EINVAL;
1860
1861         if (!size)
1862                 return 0;
1863
1864         spin_lock_irqsave(&lp->lock, flags);
1865
1866         if (lp->hs_state != LDC_HS_COMPLETE)
1867                 err = -ENOTCONN;
1868         else
1869                 err = lp->mops->read(lp, buf, size);
1870
1871         spin_unlock_irqrestore(&lp->lock, flags);
1872
1873         return err;
1874 }
1875 EXPORT_SYMBOL(ldc_read);
1876
1877 static long arena_alloc(struct ldc_iommu *iommu, unsigned long npages)
1878 {
1879         struct iommu_arena *arena = &iommu->arena;
1880         unsigned long n, i, start, end, limit;
1881         int pass;
1882
1883         limit = arena->limit;
1884         start = arena->hint;
1885         pass = 0;
1886
1887 again:
1888         n = find_next_zero_bit(arena->map, limit, start);
1889         end = n + npages;
1890         if (unlikely(end >= limit)) {
1891                 if (likely(pass < 1)) {
1892                         limit = start;
1893                         start = 0;
1894                         pass++;
1895                         goto again;
1896                 } else {
1897                         /* Scanned the whole thing, give up. */
1898                         return -1;
1899                 }
1900         }
1901
1902         for (i = n; i < end; i++) {
1903                 if (test_bit(i, arena->map)) {
1904                         start = i + 1;
1905                         goto again;
1906                 }
1907         }
1908
1909         for (i = n; i < end; i++)
1910                 __set_bit(i, arena->map);
1911
1912         arena->hint = end;
1913
1914         return n;
1915 }
1916
1917 #define COOKIE_PGSZ_CODE        0xf000000000000000ULL
1918 #define COOKIE_PGSZ_CODE_SHIFT  60ULL
1919
1920 static u64 pagesize_code(void)
1921 {
1922         switch (PAGE_SIZE) {
1923         default:
1924         case (8ULL * 1024ULL):
1925                 return 0;
1926         case (64ULL * 1024ULL):
1927                 return 1;
1928         case (512ULL * 1024ULL):
1929                 return 2;
1930         case (4ULL * 1024ULL * 1024ULL):
1931                 return 3;
1932         case (32ULL * 1024ULL * 1024ULL):
1933                 return 4;
1934         case (256ULL * 1024ULL * 1024ULL):
1935                 return 5;
1936         }
1937 }
1938
1939 static u64 make_cookie(u64 index, u64 pgsz_code, u64 page_offset)
1940 {
1941         return ((pgsz_code << COOKIE_PGSZ_CODE_SHIFT) |
1942                 (index << PAGE_SHIFT) |
1943                 page_offset);
1944 }
1945
1946 static u64 cookie_to_index(u64 cookie, unsigned long *shift)
1947 {
1948         u64 szcode = cookie >> COOKIE_PGSZ_CODE_SHIFT;
1949
1950         cookie &= ~COOKIE_PGSZ_CODE;
1951
1952         *shift = szcode * 3;
1953
1954         return (cookie >> (13ULL + (szcode * 3ULL)));
1955 }
1956
1957 static struct ldc_mtable_entry *alloc_npages(struct ldc_iommu *iommu,
1958                                              unsigned long npages)
1959 {
1960         long entry;
1961
1962         entry = arena_alloc(iommu, npages);
1963         if (unlikely(entry < 0))
1964                 return NULL;
1965
1966         return iommu->page_table + entry;
1967 }
1968
1969 static u64 perm_to_mte(unsigned int map_perm)
1970 {
1971         u64 mte_base;
1972
1973         mte_base = pagesize_code();
1974
1975         if (map_perm & LDC_MAP_SHADOW) {
1976                 if (map_perm & LDC_MAP_R)
1977                         mte_base |= LDC_MTE_COPY_R;
1978                 if (map_perm & LDC_MAP_W)
1979                         mte_base |= LDC_MTE_COPY_W;
1980         }
1981         if (map_perm & LDC_MAP_DIRECT) {
1982                 if (map_perm & LDC_MAP_R)
1983                         mte_base |= LDC_MTE_READ;
1984                 if (map_perm & LDC_MAP_W)
1985                         mte_base |= LDC_MTE_WRITE;
1986                 if (map_perm & LDC_MAP_X)
1987                         mte_base |= LDC_MTE_EXEC;
1988         }
1989         if (map_perm & LDC_MAP_IO) {
1990                 if (map_perm & LDC_MAP_R)
1991                         mte_base |= LDC_MTE_IOMMU_R;
1992                 if (map_perm & LDC_MAP_W)
1993                         mte_base |= LDC_MTE_IOMMU_W;
1994         }
1995
1996         return mte_base;
1997 }
1998
1999 static int pages_in_region(unsigned long base, long len)
2000 {
2001         int count = 0;
2002
2003         do {
2004                 unsigned long new = (base + PAGE_SIZE) & PAGE_MASK;
2005
2006                 len -= (new - base);
2007                 base = new;
2008                 count++;
2009         } while (len > 0);
2010
2011         return count;
2012 }
2013
2014 struct cookie_state {
2015         struct ldc_mtable_entry         *page_table;
2016         struct ldc_trans_cookie         *cookies;
2017         u64                             mte_base;
2018         u64                             prev_cookie;
2019         u32                             pte_idx;
2020         u32                             nc;
2021 };
2022
2023 static void fill_cookies(struct cookie_state *sp, unsigned long pa,
2024                          unsigned long off, unsigned long len)
2025 {
2026         do {
2027                 unsigned long tlen, new = pa + PAGE_SIZE;
2028                 u64 this_cookie;
2029
2030                 sp->page_table[sp->pte_idx].mte = sp->mte_base | pa;
2031
2032                 tlen = PAGE_SIZE;
2033                 if (off)
2034                         tlen = PAGE_SIZE - off;
2035                 if (tlen > len)
2036                         tlen = len;
2037
2038                 this_cookie = make_cookie(sp->pte_idx,
2039                                           pagesize_code(), off);
2040
2041                 off = 0;
2042
2043                 if (this_cookie == sp->prev_cookie) {
2044                         sp->cookies[sp->nc - 1].cookie_size += tlen;
2045                 } else {
2046                         sp->cookies[sp->nc].cookie_addr = this_cookie;
2047                         sp->cookies[sp->nc].cookie_size = tlen;
2048                         sp->nc++;
2049                 }
2050                 sp->prev_cookie = this_cookie + tlen;
2051
2052                 sp->pte_idx++;
2053
2054                 len -= tlen;
2055                 pa = new;
2056         } while (len > 0);
2057 }
2058
2059 static int sg_count_one(struct scatterlist *sg)
2060 {
2061         unsigned long base = page_to_pfn(sg_page(sg)) << PAGE_SHIFT;
2062         long len = sg->length;
2063
2064         if ((sg->offset | len) & (8UL - 1))
2065                 return -EFAULT;
2066
2067         return pages_in_region(base + sg->offset, len);
2068 }
2069
2070 static int sg_count_pages(struct scatterlist *sg, int num_sg)
2071 {
2072         int count;
2073         int i;
2074
2075         count = 0;
2076         for (i = 0; i < num_sg; i++) {
2077                 int err = sg_count_one(sg + i);
2078                 if (err < 0)
2079                         return err;
2080                 count += err;
2081         }
2082
2083         return count;
2084 }
2085
2086 int ldc_map_sg(struct ldc_channel *lp,
2087                struct scatterlist *sg, int num_sg,
2088                struct ldc_trans_cookie *cookies, int ncookies,
2089                unsigned int map_perm)
2090 {
2091         unsigned long i, npages, flags;
2092         struct ldc_mtable_entry *base;
2093         struct cookie_state state;
2094         struct ldc_iommu *iommu;
2095         int err;
2096
2097         if (map_perm & ~LDC_MAP_ALL)
2098                 return -EINVAL;
2099
2100         err = sg_count_pages(sg, num_sg);
2101         if (err < 0)
2102                 return err;
2103
2104         npages = err;
2105         if (err > ncookies)
2106                 return -EMSGSIZE;
2107
2108         iommu = &lp->iommu;
2109
2110         spin_lock_irqsave(&iommu->lock, flags);
2111         base = alloc_npages(iommu, npages);
2112         spin_unlock_irqrestore(&iommu->lock, flags);
2113
2114         if (!base)
2115                 return -ENOMEM;
2116
2117         state.page_table = iommu->page_table;
2118         state.cookies = cookies;
2119         state.mte_base = perm_to_mte(map_perm);
2120         state.prev_cookie = ~(u64)0;
2121         state.pte_idx = (base - iommu->page_table);
2122         state.nc = 0;
2123
2124         for (i = 0; i < num_sg; i++)
2125                 fill_cookies(&state, page_to_pfn(sg_page(&sg[i])) << PAGE_SHIFT,
2126                              sg[i].offset, sg[i].length);
2127
2128         return state.nc;
2129 }
2130 EXPORT_SYMBOL(ldc_map_sg);
2131
2132 int ldc_map_single(struct ldc_channel *lp,
2133                    void *buf, unsigned int len,
2134                    struct ldc_trans_cookie *cookies, int ncookies,
2135                    unsigned int map_perm)
2136 {
2137         unsigned long npages, pa, flags;
2138         struct ldc_mtable_entry *base;
2139         struct cookie_state state;
2140         struct ldc_iommu *iommu;
2141
2142         if ((map_perm & ~LDC_MAP_ALL) || (ncookies < 1))
2143                 return -EINVAL;
2144
2145         pa = __pa(buf);
2146         if ((pa | len) & (8UL - 1))
2147                 return -EFAULT;
2148
2149         npages = pages_in_region(pa, len);
2150
2151         iommu = &lp->iommu;
2152
2153         spin_lock_irqsave(&iommu->lock, flags);
2154         base = alloc_npages(iommu, npages);
2155         spin_unlock_irqrestore(&iommu->lock, flags);
2156
2157         if (!base)
2158                 return -ENOMEM;
2159
2160         state.page_table = iommu->page_table;
2161         state.cookies = cookies;
2162         state.mte_base = perm_to_mte(map_perm);
2163         state.prev_cookie = ~(u64)0;
2164         state.pte_idx = (base - iommu->page_table);
2165         state.nc = 0;
2166         fill_cookies(&state, (pa & PAGE_MASK), (pa & ~PAGE_MASK), len);
2167         BUG_ON(state.nc != 1);
2168
2169         return state.nc;
2170 }
2171 EXPORT_SYMBOL(ldc_map_single);
2172
2173 static void free_npages(unsigned long id, struct ldc_iommu *iommu,
2174                         u64 cookie, u64 size)
2175 {
2176         struct iommu_arena *arena = &iommu->arena;
2177         unsigned long i, shift, index, npages;
2178         struct ldc_mtable_entry *base;
2179
2180         npages = PAGE_ALIGN(((cookie & ~PAGE_MASK) + size)) >> PAGE_SHIFT;
2181         index = cookie_to_index(cookie, &shift);
2182         base = iommu->page_table + index;
2183
2184         BUG_ON(index > arena->limit ||
2185                (index + npages) > arena->limit);
2186
2187         for (i = 0; i < npages; i++) {
2188                 if (base->cookie)
2189                         sun4v_ldc_revoke(id, cookie + (i << shift),
2190                                          base->cookie);
2191                 base->mte = 0;
2192                 __clear_bit(index + i, arena->map);
2193         }
2194 }
2195
2196 void ldc_unmap(struct ldc_channel *lp, struct ldc_trans_cookie *cookies,
2197                int ncookies)
2198 {
2199         struct ldc_iommu *iommu = &lp->iommu;
2200         unsigned long flags;
2201         int i;
2202
2203         spin_lock_irqsave(&iommu->lock, flags);
2204         for (i = 0; i < ncookies; i++) {
2205                 u64 addr = cookies[i].cookie_addr;
2206                 u64 size = cookies[i].cookie_size;
2207
2208                 free_npages(lp->id, iommu, addr, size);
2209         }
2210         spin_unlock_irqrestore(&iommu->lock, flags);
2211 }
2212 EXPORT_SYMBOL(ldc_unmap);
2213
2214 int ldc_copy(struct ldc_channel *lp, int copy_dir,
2215              void *buf, unsigned int len, unsigned long offset,
2216              struct ldc_trans_cookie *cookies, int ncookies)
2217 {
2218         unsigned int orig_len;
2219         unsigned long ra;
2220         int i;
2221
2222         if (copy_dir != LDC_COPY_IN && copy_dir != LDC_COPY_OUT) {
2223                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Bad copy_dir[%d]\n",
2224                        lp->id, copy_dir);
2225                 return -EINVAL;
2226         }
2227
2228         ra = __pa(buf);
2229         if ((ra | len | offset) & (8UL - 1)) {
2230                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Unaligned buffer "
2231                        "ra[%lx] len[%x] offset[%lx]\n",
2232                        lp->id, ra, len, offset);
2233                 return -EFAULT;
2234         }
2235
2236         if (lp->hs_state != LDC_HS_COMPLETE ||
2237             (lp->flags & LDC_FLAG_RESET)) {
2238                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Link down hs_state[%x] "
2239                        "flags[%x]\n", lp->id, lp->hs_state, lp->flags);
2240                 return -ECONNRESET;
2241         }
2242
2243         orig_len = len;
2244         for (i = 0; i < ncookies; i++) {
2245                 unsigned long cookie_raddr = cookies[i].cookie_addr;
2246                 unsigned long this_len = cookies[i].cookie_size;
2247                 unsigned long actual_len;
2248
2249                 if (unlikely(offset)) {
2250                         unsigned long this_off = offset;
2251
2252                         if (this_off > this_len)
2253                                 this_off = this_len;
2254
2255                         offset -= this_off;
2256                         this_len -= this_off;
2257                         if (!this_len)
2258                                 continue;
2259                         cookie_raddr += this_off;
2260                 }
2261
2262                 if (this_len > len)
2263                         this_len = len;
2264
2265                 while (1) {
2266                         unsigned long hv_err;
2267
2268                         hv_err = sun4v_ldc_copy(lp->id, copy_dir,
2269                                                 cookie_raddr, ra,
2270                                                 this_len, &actual_len);
2271                         if (unlikely(hv_err)) {
2272                                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] "
2273                                        "HV error %lu\n",
2274                                        lp->id, hv_err);
2275                                 if (lp->hs_state != LDC_HS_COMPLETE ||
2276                                     (lp->flags & LDC_FLAG_RESET))
2277                                         return -ECONNRESET;
2278                                 else
2279                                         return -EFAULT;
2280                         }
2281
2282                         cookie_raddr += actual_len;
2283                         ra += actual_len;
2284                         len -= actual_len;
2285                         if (actual_len == this_len)
2286                                 break;
2287
2288                         this_len -= actual_len;
2289                 }
2290
2291                 if (!len)
2292                         break;
2293         }
2294
2295         /* It is caller policy what to do about short copies.
2296          * For example, a networking driver can declare the
2297          * packet a runt and drop it.
2298          */
2299
2300         return orig_len - len;
2301 }
2302 EXPORT_SYMBOL(ldc_copy);
2303
2304 void *ldc_alloc_exp_dring(struct ldc_channel *lp, unsigned int len,
2305                           struct ldc_trans_cookie *cookies, int *ncookies,
2306                           unsigned int map_perm)
2307 {
2308         void *buf;
2309         int err;
2310
2311         if (len & (8UL - 1))
2312                 return ERR_PTR(-EINVAL);
2313
2314         buf = kzalloc(len, GFP_KERNEL);
2315         if (!buf)
2316                 return ERR_PTR(-ENOMEM);
2317
2318         err = ldc_map_single(lp, buf, len, cookies, *ncookies, map_perm);
2319         if (err < 0) {
2320                 kfree(buf);
2321                 return ERR_PTR(err);
2322         }
2323         *ncookies = err;
2324
2325         return buf;
2326 }
2327 EXPORT_SYMBOL(ldc_alloc_exp_dring);
2328
2329 void ldc_free_exp_dring(struct ldc_channel *lp, void *buf, unsigned int len,
2330                         struct ldc_trans_cookie *cookies, int ncookies)
2331 {
2332         ldc_unmap(lp, cookies, ncookies);
2333         kfree(buf);
2334 }
2335 EXPORT_SYMBOL(ldc_free_exp_dring);
2336
2337 static int __init ldc_init(void)
2338 {
2339         unsigned long major, minor;
2340         struct mdesc_handle *hp;
2341         const u64 *v;
2342         int err;
2343         u64 mp;
2344
2345         hp = mdesc_grab();
2346         if (!hp)
2347                 return -ENODEV;
2348
2349         mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
2350         err = -ENODEV;
2351         if (mp == MDESC_NODE_NULL)
2352                 goto out;
2353
2354         v = mdesc_get_property(hp, mp, "domaining-enabled", NULL);
2355         if (!v)
2356                 goto out;
2357
2358         major = 1;
2359         minor = 0;
2360         if (sun4v_hvapi_register(HV_GRP_LDOM, major, &minor)) {
2361                 printk(KERN_INFO PFX "Could not register LDOM hvapi.\n");
2362                 goto out;
2363         }
2364
2365         printk(KERN_INFO "%s", version);
2366
2367         if (!*v) {
2368                 printk(KERN_INFO PFX "Domaining disabled.\n");
2369                 goto out;
2370         }
2371         ldom_domaining_enabled = 1;
2372         err = 0;
2373
2374 out:
2375         mdesc_release(hp);
2376         return err;
2377 }
2378
2379 core_initcall(ldc_init);