Merge tag 'iwlwifi-next-for-kalle-2014-12-30' of https://git.kernel.org/pub/scm/linux...
[cascardo/linux.git] / 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/export.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 #include <linux/bitmap.h>
18
19 #include <asm/hypervisor.h>
20 #include <asm/iommu.h>
21 #include <asm/page.h>
22 #include <asm/ldc.h>
23 #include <asm/mdesc.h>
24
25 #define DRV_MODULE_NAME         "ldc"
26 #define PFX DRV_MODULE_NAME     ": "
27 #define DRV_MODULE_VERSION      "1.1"
28 #define DRV_MODULE_RELDATE      "July 22, 2008"
29
30 static char version[] =
31         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
32 #define LDC_PACKET_SIZE         64
33
34 /* Packet header layout for unreliable and reliable mode frames.
35  * When in RAW mode, packets are simply straight 64-byte payloads
36  * with no headers.
37  */
38 struct ldc_packet {
39         u8                      type;
40 #define LDC_CTRL                0x01
41 #define LDC_DATA                0x02
42 #define LDC_ERR                 0x10
43
44         u8                      stype;
45 #define LDC_INFO                0x01
46 #define LDC_ACK                 0x02
47 #define LDC_NACK                0x04
48
49         u8                      ctrl;
50 #define LDC_VERS                0x01 /* Link Version            */
51 #define LDC_RTS                 0x02 /* Request To Send         */
52 #define LDC_RTR                 0x03 /* Ready To Receive        */
53 #define LDC_RDX                 0x04 /* Ready for Data eXchange */
54 #define LDC_CTRL_MSK            0x0f
55
56         u8                      env;
57 #define LDC_LEN                 0x3f
58 #define LDC_FRAG_MASK           0xc0
59 #define LDC_START               0x40
60 #define LDC_STOP                0x80
61
62         u32                     seqid;
63
64         union {
65                 u8              u_data[LDC_PACKET_SIZE - 8];
66                 struct {
67                         u32     pad;
68                         u32     ackid;
69                         u8      r_data[LDC_PACKET_SIZE - 8 - 8];
70                 } r;
71         } u;
72 };
73
74 struct ldc_version {
75         u16 major;
76         u16 minor;
77 };
78
79 /* Ordered from largest major to lowest.  */
80 static struct ldc_version ver_arr[] = {
81         { .major = 1, .minor = 0 },
82 };
83
84 #define LDC_DEFAULT_MTU                 (4 * LDC_PACKET_SIZE)
85 #define LDC_DEFAULT_NUM_ENTRIES         (PAGE_SIZE / LDC_PACKET_SIZE)
86
87 struct ldc_channel;
88
89 struct ldc_mode_ops {
90         int (*write)(struct ldc_channel *, const void *, unsigned int);
91         int (*read)(struct ldc_channel *, void *, unsigned int);
92 };
93
94 static const struct ldc_mode_ops raw_ops;
95 static const struct ldc_mode_ops nonraw_ops;
96 static const struct ldc_mode_ops stream_ops;
97
98 int ldom_domaining_enabled;
99
100 struct ldc_iommu {
101         /* Protects arena alloc/free.  */
102         spinlock_t                      lock;
103         struct iommu_arena              arena;
104         struct ldc_mtable_entry         *page_table;
105 };
106
107 struct ldc_channel {
108         /* Protects all operations that depend upon channel state.  */
109         spinlock_t                      lock;
110
111         unsigned long                   id;
112
113         u8                              *mssbuf;
114         u32                             mssbuf_len;
115         u32                             mssbuf_off;
116
117         struct ldc_packet               *tx_base;
118         unsigned long                   tx_head;
119         unsigned long                   tx_tail;
120         unsigned long                   tx_num_entries;
121         unsigned long                   tx_ra;
122
123         unsigned long                   tx_acked;
124
125         struct ldc_packet               *rx_base;
126         unsigned long                   rx_head;
127         unsigned long                   rx_tail;
128         unsigned long                   rx_num_entries;
129         unsigned long                   rx_ra;
130
131         u32                             rcv_nxt;
132         u32                             snd_nxt;
133
134         unsigned long                   chan_state;
135
136         struct ldc_channel_config       cfg;
137         void                            *event_arg;
138
139         const struct ldc_mode_ops       *mops;
140
141         struct ldc_iommu                iommu;
142
143         struct ldc_version              ver;
144
145         u8                              hs_state;
146 #define LDC_HS_CLOSED                   0x00
147 #define LDC_HS_OPEN                     0x01
148 #define LDC_HS_GOTVERS                  0x02
149 #define LDC_HS_SENTRTR                  0x03
150 #define LDC_HS_GOTRTR                   0x04
151 #define LDC_HS_COMPLETE                 0x10
152
153         u8                              flags;
154 #define LDC_FLAG_ALLOCED_QUEUES         0x01
155 #define LDC_FLAG_REGISTERED_QUEUES      0x02
156 #define LDC_FLAG_REGISTERED_IRQS        0x04
157 #define LDC_FLAG_RESET                  0x10
158
159         u8                              mss;
160         u8                              state;
161
162 #define LDC_IRQ_NAME_MAX                32
163         char                            rx_irq_name[LDC_IRQ_NAME_MAX];
164         char                            tx_irq_name[LDC_IRQ_NAME_MAX];
165
166         struct hlist_head               mh_list;
167
168         struct hlist_node               list;
169 };
170
171 #define ldcdbg(TYPE, f, a...) \
172 do {    if (lp->cfg.debug & LDC_DEBUG_##TYPE) \
173                 printk(KERN_INFO PFX "ID[%lu] " f, lp->id, ## a); \
174 } while (0)
175
176 static const char *state_to_str(u8 state)
177 {
178         switch (state) {
179         case LDC_STATE_INVALID:
180                 return "INVALID";
181         case LDC_STATE_INIT:
182                 return "INIT";
183         case LDC_STATE_BOUND:
184                 return "BOUND";
185         case LDC_STATE_READY:
186                 return "READY";
187         case LDC_STATE_CONNECTED:
188                 return "CONNECTED";
189         default:
190                 return "<UNKNOWN>";
191         }
192 }
193
194 static void ldc_set_state(struct ldc_channel *lp, u8 state)
195 {
196         ldcdbg(STATE, "STATE (%s) --> (%s)\n",
197                state_to_str(lp->state),
198                state_to_str(state));
199
200         lp->state = state;
201 }
202
203 static unsigned long __advance(unsigned long off, unsigned long num_entries)
204 {
205         off += LDC_PACKET_SIZE;
206         if (off == (num_entries * LDC_PACKET_SIZE))
207                 off = 0;
208
209         return off;
210 }
211
212 static unsigned long rx_advance(struct ldc_channel *lp, unsigned long off)
213 {
214         return __advance(off, lp->rx_num_entries);
215 }
216
217 static unsigned long tx_advance(struct ldc_channel *lp, unsigned long off)
218 {
219         return __advance(off, lp->tx_num_entries);
220 }
221
222 static struct ldc_packet *handshake_get_tx_packet(struct ldc_channel *lp,
223                                                   unsigned long *new_tail)
224 {
225         struct ldc_packet *p;
226         unsigned long t;
227
228         t = tx_advance(lp, lp->tx_tail);
229         if (t == lp->tx_head)
230                 return NULL;
231
232         *new_tail = t;
233
234         p = lp->tx_base;
235         return p + (lp->tx_tail / LDC_PACKET_SIZE);
236 }
237
238 /* When we are in reliable or stream mode, have to track the next packet
239  * we haven't gotten an ACK for in the TX queue using tx_acked.  We have
240  * to be careful not to stomp over the queue past that point.  During
241  * the handshake, we don't have TX data packets pending in the queue
242  * and that's why handshake_get_tx_packet() need not be mindful of
243  * lp->tx_acked.
244  */
245 static unsigned long head_for_data(struct ldc_channel *lp)
246 {
247         if (lp->cfg.mode == LDC_MODE_STREAM)
248                 return lp->tx_acked;
249         return lp->tx_head;
250 }
251
252 static int tx_has_space_for(struct ldc_channel *lp, unsigned int size)
253 {
254         unsigned long limit, tail, new_tail, diff;
255         unsigned int mss;
256
257         limit = head_for_data(lp);
258         tail = lp->tx_tail;
259         new_tail = tx_advance(lp, tail);
260         if (new_tail == limit)
261                 return 0;
262
263         if (limit > new_tail)
264                 diff = limit - new_tail;
265         else
266                 diff = (limit +
267                         ((lp->tx_num_entries * LDC_PACKET_SIZE) - new_tail));
268         diff /= LDC_PACKET_SIZE;
269         mss = lp->mss;
270
271         if (diff * mss < size)
272                 return 0;
273
274         return 1;
275 }
276
277 static struct ldc_packet *data_get_tx_packet(struct ldc_channel *lp,
278                                              unsigned long *new_tail)
279 {
280         struct ldc_packet *p;
281         unsigned long h, t;
282
283         h = head_for_data(lp);
284         t = tx_advance(lp, lp->tx_tail);
285         if (t == h)
286                 return NULL;
287
288         *new_tail = t;
289
290         p = lp->tx_base;
291         return p + (lp->tx_tail / LDC_PACKET_SIZE);
292 }
293
294 static int set_tx_tail(struct ldc_channel *lp, unsigned long tail)
295 {
296         unsigned long orig_tail = lp->tx_tail;
297         int limit = 1000;
298
299         lp->tx_tail = tail;
300         while (limit-- > 0) {
301                 unsigned long err;
302
303                 err = sun4v_ldc_tx_set_qtail(lp->id, tail);
304                 if (!err)
305                         return 0;
306
307                 if (err != HV_EWOULDBLOCK) {
308                         lp->tx_tail = orig_tail;
309                         return -EINVAL;
310                 }
311                 udelay(1);
312         }
313
314         lp->tx_tail = orig_tail;
315         return -EBUSY;
316 }
317
318 /* This just updates the head value in the hypervisor using
319  * a polling loop with a timeout.  The caller takes care of
320  * upating software state representing the head change, if any.
321  */
322 static int __set_rx_head(struct ldc_channel *lp, unsigned long head)
323 {
324         int limit = 1000;
325
326         while (limit-- > 0) {
327                 unsigned long err;
328
329                 err = sun4v_ldc_rx_set_qhead(lp->id, head);
330                 if (!err)
331                         return 0;
332
333                 if (err != HV_EWOULDBLOCK)
334                         return -EINVAL;
335
336                 udelay(1);
337         }
338
339         return -EBUSY;
340 }
341
342 static int send_tx_packet(struct ldc_channel *lp,
343                           struct ldc_packet *p,
344                           unsigned long new_tail)
345 {
346         BUG_ON(p != (lp->tx_base + (lp->tx_tail / LDC_PACKET_SIZE)));
347
348         return set_tx_tail(lp, new_tail);
349 }
350
351 static struct ldc_packet *handshake_compose_ctrl(struct ldc_channel *lp,
352                                                  u8 stype, u8 ctrl,
353                                                  void *data, int dlen,
354                                                  unsigned long *new_tail)
355 {
356         struct ldc_packet *p = handshake_get_tx_packet(lp, new_tail);
357
358         if (p) {
359                 memset(p, 0, sizeof(*p));
360                 p->type = LDC_CTRL;
361                 p->stype = stype;
362                 p->ctrl = ctrl;
363                 if (data)
364                         memcpy(p->u.u_data, data, dlen);
365         }
366         return p;
367 }
368
369 static int start_handshake(struct ldc_channel *lp)
370 {
371         struct ldc_packet *p;
372         struct ldc_version *ver;
373         unsigned long new_tail;
374
375         ver = &ver_arr[0];
376
377         ldcdbg(HS, "SEND VER INFO maj[%u] min[%u]\n",
378                ver->major, ver->minor);
379
380         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
381                                    ver, sizeof(*ver), &new_tail);
382         if (p) {
383                 int err = send_tx_packet(lp, p, new_tail);
384                 if (!err)
385                         lp->flags &= ~LDC_FLAG_RESET;
386                 return err;
387         }
388         return -EBUSY;
389 }
390
391 static int send_version_nack(struct ldc_channel *lp,
392                              u16 major, u16 minor)
393 {
394         struct ldc_packet *p;
395         struct ldc_version ver;
396         unsigned long new_tail;
397
398         ver.major = major;
399         ver.minor = minor;
400
401         p = handshake_compose_ctrl(lp, LDC_NACK, LDC_VERS,
402                                    &ver, sizeof(ver), &new_tail);
403         if (p) {
404                 ldcdbg(HS, "SEND VER NACK maj[%u] min[%u]\n",
405                        ver.major, ver.minor);
406
407                 return send_tx_packet(lp, p, new_tail);
408         }
409         return -EBUSY;
410 }
411
412 static int send_version_ack(struct ldc_channel *lp,
413                             struct ldc_version *vp)
414 {
415         struct ldc_packet *p;
416         unsigned long new_tail;
417
418         p = handshake_compose_ctrl(lp, LDC_ACK, LDC_VERS,
419                                    vp, sizeof(*vp), &new_tail);
420         if (p) {
421                 ldcdbg(HS, "SEND VER ACK maj[%u] min[%u]\n",
422                        vp->major, vp->minor);
423
424                 return send_tx_packet(lp, p, new_tail);
425         }
426         return -EBUSY;
427 }
428
429 static int send_rts(struct ldc_channel *lp)
430 {
431         struct ldc_packet *p;
432         unsigned long new_tail;
433
434         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTS, NULL, 0,
435                                    &new_tail);
436         if (p) {
437                 p->env = lp->cfg.mode;
438                 p->seqid = 0;
439                 lp->rcv_nxt = 0;
440
441                 ldcdbg(HS, "SEND RTS env[0x%x] seqid[0x%x]\n",
442                        p->env, p->seqid);
443
444                 return send_tx_packet(lp, p, new_tail);
445         }
446         return -EBUSY;
447 }
448
449 static int send_rtr(struct ldc_channel *lp)
450 {
451         struct ldc_packet *p;
452         unsigned long new_tail;
453
454         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTR, NULL, 0,
455                                    &new_tail);
456         if (p) {
457                 p->env = lp->cfg.mode;
458                 p->seqid = 0;
459
460                 ldcdbg(HS, "SEND RTR env[0x%x] seqid[0x%x]\n",
461                        p->env, p->seqid);
462
463                 return send_tx_packet(lp, p, new_tail);
464         }
465         return -EBUSY;
466 }
467
468 static int send_rdx(struct ldc_channel *lp)
469 {
470         struct ldc_packet *p;
471         unsigned long new_tail;
472
473         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RDX, NULL, 0,
474                                    &new_tail);
475         if (p) {
476                 p->env = 0;
477                 p->seqid = ++lp->snd_nxt;
478                 p->u.r.ackid = lp->rcv_nxt;
479
480                 ldcdbg(HS, "SEND RDX env[0x%x] seqid[0x%x] ackid[0x%x]\n",
481                        p->env, p->seqid, p->u.r.ackid);
482
483                 return send_tx_packet(lp, p, new_tail);
484         }
485         return -EBUSY;
486 }
487
488 static int send_data_nack(struct ldc_channel *lp, struct ldc_packet *data_pkt)
489 {
490         struct ldc_packet *p;
491         unsigned long new_tail;
492         int err;
493
494         p = data_get_tx_packet(lp, &new_tail);
495         if (!p)
496                 return -EBUSY;
497         memset(p, 0, sizeof(*p));
498         p->type = data_pkt->type;
499         p->stype = LDC_NACK;
500         p->ctrl = data_pkt->ctrl & LDC_CTRL_MSK;
501         p->seqid = lp->snd_nxt + 1;
502         p->u.r.ackid = lp->rcv_nxt;
503
504         ldcdbg(HS, "SEND DATA NACK type[0x%x] ctl[0x%x] seq[0x%x] ack[0x%x]\n",
505                p->type, p->ctrl, p->seqid, p->u.r.ackid);
506
507         err = send_tx_packet(lp, p, new_tail);
508         if (!err)
509                 lp->snd_nxt++;
510
511         return err;
512 }
513
514 static int ldc_abort(struct ldc_channel *lp)
515 {
516         unsigned long hv_err;
517
518         ldcdbg(STATE, "ABORT\n");
519
520         /* We report but do not act upon the hypervisor errors because
521          * there really isn't much we can do if they fail at this point.
522          */
523         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
524         if (hv_err)
525                 printk(KERN_ERR PFX "ldc_abort: "
526                        "sun4v_ldc_tx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
527                        lp->id, lp->tx_ra, lp->tx_num_entries, hv_err);
528
529         hv_err = sun4v_ldc_tx_get_state(lp->id,
530                                         &lp->tx_head,
531                                         &lp->tx_tail,
532                                         &lp->chan_state);
533         if (hv_err)
534                 printk(KERN_ERR PFX "ldc_abort: "
535                        "sun4v_ldc_tx_get_state(%lx,...) failed, err=%lu\n",
536                        lp->id, hv_err);
537
538         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
539         if (hv_err)
540                 printk(KERN_ERR PFX "ldc_abort: "
541                        "sun4v_ldc_rx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
542                        lp->id, lp->rx_ra, lp->rx_num_entries, hv_err);
543
544         /* Refetch the RX queue state as well, because we could be invoked
545          * here in the queue processing context.
546          */
547         hv_err = sun4v_ldc_rx_get_state(lp->id,
548                                         &lp->rx_head,
549                                         &lp->rx_tail,
550                                         &lp->chan_state);
551         if (hv_err)
552                 printk(KERN_ERR PFX "ldc_abort: "
553                        "sun4v_ldc_rx_get_state(%lx,...) failed, err=%lu\n",
554                        lp->id, hv_err);
555
556         return -ECONNRESET;
557 }
558
559 static struct ldc_version *find_by_major(u16 major)
560 {
561         struct ldc_version *ret = NULL;
562         int i;
563
564         for (i = 0; i < ARRAY_SIZE(ver_arr); i++) {
565                 struct ldc_version *v = &ver_arr[i];
566                 if (v->major <= major) {
567                         ret = v;
568                         break;
569                 }
570         }
571         return ret;
572 }
573
574 static int process_ver_info(struct ldc_channel *lp, struct ldc_version *vp)
575 {
576         struct ldc_version *vap;
577         int err;
578
579         ldcdbg(HS, "GOT VERSION INFO major[%x] minor[%x]\n",
580                vp->major, vp->minor);
581
582         if (lp->hs_state == LDC_HS_GOTVERS) {
583                 lp->hs_state = LDC_HS_OPEN;
584                 memset(&lp->ver, 0, sizeof(lp->ver));
585         }
586
587         vap = find_by_major(vp->major);
588         if (!vap) {
589                 err = send_version_nack(lp, 0, 0);
590         } else if (vap->major != vp->major) {
591                 err = send_version_nack(lp, vap->major, vap->minor);
592         } else {
593                 struct ldc_version ver = *vp;
594                 if (ver.minor > vap->minor)
595                         ver.minor = vap->minor;
596                 err = send_version_ack(lp, &ver);
597                 if (!err) {
598                         lp->ver = ver;
599                         lp->hs_state = LDC_HS_GOTVERS;
600                 }
601         }
602         if (err)
603                 return ldc_abort(lp);
604
605         return 0;
606 }
607
608 static int process_ver_ack(struct ldc_channel *lp, struct ldc_version *vp)
609 {
610         ldcdbg(HS, "GOT VERSION ACK major[%x] minor[%x]\n",
611                vp->major, vp->minor);
612
613         if (lp->hs_state == LDC_HS_GOTVERS) {
614                 if (lp->ver.major != vp->major ||
615                     lp->ver.minor != vp->minor)
616                         return ldc_abort(lp);
617         } else {
618                 lp->ver = *vp;
619                 lp->hs_state = LDC_HS_GOTVERS;
620         }
621         if (send_rts(lp))
622                 return ldc_abort(lp);
623         return 0;
624 }
625
626 static int process_ver_nack(struct ldc_channel *lp, struct ldc_version *vp)
627 {
628         struct ldc_version *vap;
629         struct ldc_packet *p;
630         unsigned long new_tail;
631
632         if (vp->major == 0 && vp->minor == 0)
633                 return ldc_abort(lp);
634
635         vap = find_by_major(vp->major);
636         if (!vap)
637                 return ldc_abort(lp);
638
639         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
640                                            vap, sizeof(*vap),
641                                            &new_tail);
642         if (!p)
643                 return ldc_abort(lp);
644
645         return send_tx_packet(lp, p, new_tail);
646 }
647
648 static int process_version(struct ldc_channel *lp,
649                            struct ldc_packet *p)
650 {
651         struct ldc_version *vp;
652
653         vp = (struct ldc_version *) p->u.u_data;
654
655         switch (p->stype) {
656         case LDC_INFO:
657                 return process_ver_info(lp, vp);
658
659         case LDC_ACK:
660                 return process_ver_ack(lp, vp);
661
662         case LDC_NACK:
663                 return process_ver_nack(lp, vp);
664
665         default:
666                 return ldc_abort(lp);
667         }
668 }
669
670 static int process_rts(struct ldc_channel *lp,
671                        struct ldc_packet *p)
672 {
673         ldcdbg(HS, "GOT RTS stype[%x] seqid[%x] env[%x]\n",
674                p->stype, p->seqid, p->env);
675
676         if (p->stype     != LDC_INFO       ||
677             lp->hs_state != LDC_HS_GOTVERS ||
678             p->env       != lp->cfg.mode)
679                 return ldc_abort(lp);
680
681         lp->snd_nxt = p->seqid;
682         lp->rcv_nxt = p->seqid;
683         lp->hs_state = LDC_HS_SENTRTR;
684         if (send_rtr(lp))
685                 return ldc_abort(lp);
686
687         return 0;
688 }
689
690 static int process_rtr(struct ldc_channel *lp,
691                        struct ldc_packet *p)
692 {
693         ldcdbg(HS, "GOT RTR stype[%x] seqid[%x] env[%x]\n",
694                p->stype, p->seqid, p->env);
695
696         if (p->stype     != LDC_INFO ||
697             p->env       != lp->cfg.mode)
698                 return ldc_abort(lp);
699
700         lp->snd_nxt = p->seqid;
701         lp->hs_state = LDC_HS_COMPLETE;
702         ldc_set_state(lp, LDC_STATE_CONNECTED);
703         send_rdx(lp);
704
705         return LDC_EVENT_UP;
706 }
707
708 static int rx_seq_ok(struct ldc_channel *lp, u32 seqid)
709 {
710         return lp->rcv_nxt + 1 == seqid;
711 }
712
713 static int process_rdx(struct ldc_channel *lp,
714                        struct ldc_packet *p)
715 {
716         ldcdbg(HS, "GOT RDX stype[%x] seqid[%x] env[%x] ackid[%x]\n",
717                p->stype, p->seqid, p->env, p->u.r.ackid);
718
719         if (p->stype != LDC_INFO ||
720             !(rx_seq_ok(lp, p->seqid)))
721                 return ldc_abort(lp);
722
723         lp->rcv_nxt = p->seqid;
724
725         lp->hs_state = LDC_HS_COMPLETE;
726         ldc_set_state(lp, LDC_STATE_CONNECTED);
727
728         return LDC_EVENT_UP;
729 }
730
731 static int process_control_frame(struct ldc_channel *lp,
732                                  struct ldc_packet *p)
733 {
734         switch (p->ctrl) {
735         case LDC_VERS:
736                 return process_version(lp, p);
737
738         case LDC_RTS:
739                 return process_rts(lp, p);
740
741         case LDC_RTR:
742                 return process_rtr(lp, p);
743
744         case LDC_RDX:
745                 return process_rdx(lp, p);
746
747         default:
748                 return ldc_abort(lp);
749         }
750 }
751
752 static int process_error_frame(struct ldc_channel *lp,
753                                struct ldc_packet *p)
754 {
755         return ldc_abort(lp);
756 }
757
758 static int process_data_ack(struct ldc_channel *lp,
759                             struct ldc_packet *ack)
760 {
761         unsigned long head = lp->tx_acked;
762         u32 ackid = ack->u.r.ackid;
763
764         while (1) {
765                 struct ldc_packet *p = lp->tx_base + (head / LDC_PACKET_SIZE);
766
767                 head = tx_advance(lp, head);
768
769                 if (p->seqid == ackid) {
770                         lp->tx_acked = head;
771                         return 0;
772                 }
773                 if (head == lp->tx_tail)
774                         return ldc_abort(lp);
775         }
776
777         return 0;
778 }
779
780 static void send_events(struct ldc_channel *lp, unsigned int event_mask)
781 {
782         if (event_mask & LDC_EVENT_RESET)
783                 lp->cfg.event(lp->event_arg, LDC_EVENT_RESET);
784         if (event_mask & LDC_EVENT_UP)
785                 lp->cfg.event(lp->event_arg, LDC_EVENT_UP);
786         if (event_mask & LDC_EVENT_DATA_READY)
787                 lp->cfg.event(lp->event_arg, LDC_EVENT_DATA_READY);
788 }
789
790 static irqreturn_t ldc_rx(int irq, void *dev_id)
791 {
792         struct ldc_channel *lp = dev_id;
793         unsigned long orig_state, flags;
794         unsigned int event_mask;
795
796         spin_lock_irqsave(&lp->lock, flags);
797
798         orig_state = lp->chan_state;
799
800         /* We should probably check for hypervisor errors here and
801          * reset the LDC channel if we get one.
802          */
803         sun4v_ldc_rx_get_state(lp->id,
804                                &lp->rx_head,
805                                &lp->rx_tail,
806                                &lp->chan_state);
807
808         ldcdbg(RX, "RX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
809                orig_state, lp->chan_state, lp->rx_head, lp->rx_tail);
810
811         event_mask = 0;
812
813         if (lp->cfg.mode == LDC_MODE_RAW &&
814             lp->chan_state == LDC_CHANNEL_UP) {
815                 lp->hs_state = LDC_HS_COMPLETE;
816                 ldc_set_state(lp, LDC_STATE_CONNECTED);
817
818                 event_mask |= LDC_EVENT_UP;
819
820                 orig_state = lp->chan_state;
821         }
822
823         /* If we are in reset state, flush the RX queue and ignore
824          * everything.
825          */
826         if (lp->flags & LDC_FLAG_RESET) {
827                 (void) __set_rx_head(lp, lp->rx_tail);
828                 goto out;
829         }
830
831         /* Once we finish the handshake, we let the ldc_read()
832          * paths do all of the control frame and state management.
833          * Just trigger the callback.
834          */
835         if (lp->hs_state == LDC_HS_COMPLETE) {
836 handshake_complete:
837                 if (lp->chan_state != orig_state) {
838                         unsigned int event = LDC_EVENT_RESET;
839
840                         if (lp->chan_state == LDC_CHANNEL_UP)
841                                 event = LDC_EVENT_UP;
842
843                         event_mask |= event;
844                 }
845                 if (lp->rx_head != lp->rx_tail)
846                         event_mask |= LDC_EVENT_DATA_READY;
847
848                 goto out;
849         }
850
851         if (lp->chan_state != orig_state)
852                 goto out;
853
854         while (lp->rx_head != lp->rx_tail) {
855                 struct ldc_packet *p;
856                 unsigned long new;
857                 int err;
858
859                 p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
860
861                 switch (p->type) {
862                 case LDC_CTRL:
863                         err = process_control_frame(lp, p);
864                         if (err > 0)
865                                 event_mask |= err;
866                         break;
867
868                 case LDC_DATA:
869                         event_mask |= LDC_EVENT_DATA_READY;
870                         err = 0;
871                         break;
872
873                 case LDC_ERR:
874                         err = process_error_frame(lp, p);
875                         break;
876
877                 default:
878                         err = ldc_abort(lp);
879                         break;
880                 }
881
882                 if (err < 0)
883                         break;
884
885                 new = lp->rx_head;
886                 new += LDC_PACKET_SIZE;
887                 if (new == (lp->rx_num_entries * LDC_PACKET_SIZE))
888                         new = 0;
889                 lp->rx_head = new;
890
891                 err = __set_rx_head(lp, new);
892                 if (err < 0) {
893                         (void) ldc_abort(lp);
894                         break;
895                 }
896                 if (lp->hs_state == LDC_HS_COMPLETE)
897                         goto handshake_complete;
898         }
899
900 out:
901         spin_unlock_irqrestore(&lp->lock, flags);
902
903         send_events(lp, event_mask);
904
905         return IRQ_HANDLED;
906 }
907
908 static irqreturn_t ldc_tx(int irq, void *dev_id)
909 {
910         struct ldc_channel *lp = dev_id;
911         unsigned long flags, orig_state;
912         unsigned int event_mask = 0;
913
914         spin_lock_irqsave(&lp->lock, flags);
915
916         orig_state = lp->chan_state;
917
918         /* We should probably check for hypervisor errors here and
919          * reset the LDC channel if we get one.
920          */
921         sun4v_ldc_tx_get_state(lp->id,
922                                &lp->tx_head,
923                                &lp->tx_tail,
924                                &lp->chan_state);
925
926         ldcdbg(TX, " TX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
927                orig_state, lp->chan_state, lp->tx_head, lp->tx_tail);
928
929         if (lp->cfg.mode == LDC_MODE_RAW &&
930             lp->chan_state == LDC_CHANNEL_UP) {
931                 lp->hs_state = LDC_HS_COMPLETE;
932                 ldc_set_state(lp, LDC_STATE_CONNECTED);
933
934                 event_mask |= LDC_EVENT_UP;
935         }
936
937         spin_unlock_irqrestore(&lp->lock, flags);
938
939         send_events(lp, event_mask);
940
941         return IRQ_HANDLED;
942 }
943
944 /* XXX ldc_alloc() and ldc_free() needs to run under a mutex so
945  * XXX that addition and removal from the ldc_channel_list has
946  * XXX atomicity, otherwise the __ldc_channel_exists() check is
947  * XXX totally pointless as another thread can slip into ldc_alloc()
948  * XXX and add a channel with the same ID.  There also needs to be
949  * XXX a spinlock for ldc_channel_list.
950  */
951 static HLIST_HEAD(ldc_channel_list);
952
953 static int __ldc_channel_exists(unsigned long id)
954 {
955         struct ldc_channel *lp;
956
957         hlist_for_each_entry(lp, &ldc_channel_list, list) {
958                 if (lp->id == id)
959                         return 1;
960         }
961         return 0;
962 }
963
964 static int alloc_queue(const char *name, unsigned long num_entries,
965                        struct ldc_packet **base, unsigned long *ra)
966 {
967         unsigned long size, order;
968         void *q;
969
970         size = num_entries * LDC_PACKET_SIZE;
971         order = get_order(size);
972
973         q = (void *) __get_free_pages(GFP_KERNEL, order);
974         if (!q) {
975                 printk(KERN_ERR PFX "Alloc of %s queue failed with "
976                        "size=%lu order=%lu\n", name, size, order);
977                 return -ENOMEM;
978         }
979
980         memset(q, 0, PAGE_SIZE << order);
981
982         *base = q;
983         *ra = __pa(q);
984
985         return 0;
986 }
987
988 static void free_queue(unsigned long num_entries, struct ldc_packet *q)
989 {
990         unsigned long size, order;
991
992         if (!q)
993                 return;
994
995         size = num_entries * LDC_PACKET_SIZE;
996         order = get_order(size);
997
998         free_pages((unsigned long)q, order);
999 }
1000
1001 /* XXX Make this configurable... XXX */
1002 #define LDC_IOTABLE_SIZE        (8 * 1024)
1003
1004 static int ldc_iommu_init(struct ldc_channel *lp)
1005 {
1006         unsigned long sz, num_tsb_entries, tsbsize, order;
1007         struct ldc_iommu *iommu = &lp->iommu;
1008         struct ldc_mtable_entry *table;
1009         unsigned long hv_err;
1010         int err;
1011
1012         num_tsb_entries = LDC_IOTABLE_SIZE;
1013         tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1014
1015         spin_lock_init(&iommu->lock);
1016
1017         sz = num_tsb_entries / 8;
1018         sz = (sz + 7UL) & ~7UL;
1019         iommu->arena.map = kzalloc(sz, GFP_KERNEL);
1020         if (!iommu->arena.map) {
1021                 printk(KERN_ERR PFX "Alloc of arena map failed, sz=%lu\n", sz);
1022                 return -ENOMEM;
1023         }
1024
1025         iommu->arena.limit = num_tsb_entries;
1026
1027         order = get_order(tsbsize);
1028
1029         table = (struct ldc_mtable_entry *)
1030                 __get_free_pages(GFP_KERNEL, order);
1031         err = -ENOMEM;
1032         if (!table) {
1033                 printk(KERN_ERR PFX "Alloc of MTE table failed, "
1034                        "size=%lu order=%lu\n", tsbsize, order);
1035                 goto out_free_map;
1036         }
1037
1038         memset(table, 0, PAGE_SIZE << order);
1039
1040         iommu->page_table = table;
1041
1042         hv_err = sun4v_ldc_set_map_table(lp->id, __pa(table),
1043                                          num_tsb_entries);
1044         err = -EINVAL;
1045         if (hv_err)
1046                 goto out_free_table;
1047
1048         return 0;
1049
1050 out_free_table:
1051         free_pages((unsigned long) table, order);
1052         iommu->page_table = NULL;
1053
1054 out_free_map:
1055         kfree(iommu->arena.map);
1056         iommu->arena.map = NULL;
1057
1058         return err;
1059 }
1060
1061 static void ldc_iommu_release(struct ldc_channel *lp)
1062 {
1063         struct ldc_iommu *iommu = &lp->iommu;
1064         unsigned long num_tsb_entries, tsbsize, order;
1065
1066         (void) sun4v_ldc_set_map_table(lp->id, 0, 0);
1067
1068         num_tsb_entries = iommu->arena.limit;
1069         tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1070         order = get_order(tsbsize);
1071
1072         free_pages((unsigned long) iommu->page_table, order);
1073         iommu->page_table = NULL;
1074
1075         kfree(iommu->arena.map);
1076         iommu->arena.map = NULL;
1077 }
1078
1079 struct ldc_channel *ldc_alloc(unsigned long id,
1080                               const struct ldc_channel_config *cfgp,
1081                               void *event_arg,
1082                               const char *name)
1083 {
1084         struct ldc_channel *lp;
1085         const struct ldc_mode_ops *mops;
1086         unsigned long dummy1, dummy2, hv_err;
1087         u8 mss, *mssbuf;
1088         int err;
1089
1090         err = -ENODEV;
1091         if (!ldom_domaining_enabled)
1092                 goto out_err;
1093
1094         err = -EINVAL;
1095         if (!cfgp)
1096                 goto out_err;
1097         if (!name)
1098                 goto out_err;
1099
1100         switch (cfgp->mode) {
1101         case LDC_MODE_RAW:
1102                 mops = &raw_ops;
1103                 mss = LDC_PACKET_SIZE;
1104                 break;
1105
1106         case LDC_MODE_UNRELIABLE:
1107                 mops = &nonraw_ops;
1108                 mss = LDC_PACKET_SIZE - 8;
1109                 break;
1110
1111         case LDC_MODE_STREAM:
1112                 mops = &stream_ops;
1113                 mss = LDC_PACKET_SIZE - 8 - 8;
1114                 break;
1115
1116         default:
1117                 goto out_err;
1118         }
1119
1120         if (!cfgp->event || !event_arg || !cfgp->rx_irq || !cfgp->tx_irq)
1121                 goto out_err;
1122
1123         hv_err = sun4v_ldc_tx_qinfo(id, &dummy1, &dummy2);
1124         err = -ENODEV;
1125         if (hv_err == HV_ECHANNEL)
1126                 goto out_err;
1127
1128         err = -EEXIST;
1129         if (__ldc_channel_exists(id))
1130                 goto out_err;
1131
1132         mssbuf = NULL;
1133
1134         lp = kzalloc(sizeof(*lp), GFP_KERNEL);
1135         err = -ENOMEM;
1136         if (!lp)
1137                 goto out_err;
1138
1139         spin_lock_init(&lp->lock);
1140
1141         lp->id = id;
1142
1143         err = ldc_iommu_init(lp);
1144         if (err)
1145                 goto out_free_ldc;
1146
1147         lp->mops = mops;
1148         lp->mss = mss;
1149
1150         lp->cfg = *cfgp;
1151         if (!lp->cfg.mtu)
1152                 lp->cfg.mtu = LDC_DEFAULT_MTU;
1153
1154         if (lp->cfg.mode == LDC_MODE_STREAM) {
1155                 mssbuf = kzalloc(lp->cfg.mtu, GFP_KERNEL);
1156                 if (!mssbuf) {
1157                         err = -ENOMEM;
1158                         goto out_free_iommu;
1159                 }
1160                 lp->mssbuf = mssbuf;
1161         }
1162
1163         lp->event_arg = event_arg;
1164
1165         /* XXX allow setting via ldc_channel_config to override defaults
1166          * XXX or use some formula based upon mtu
1167          */
1168         lp->tx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1169         lp->rx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1170
1171         err = alloc_queue("TX", lp->tx_num_entries,
1172                           &lp->tx_base, &lp->tx_ra);
1173         if (err)
1174                 goto out_free_mssbuf;
1175
1176         err = alloc_queue("RX", lp->rx_num_entries,
1177                           &lp->rx_base, &lp->rx_ra);
1178         if (err)
1179                 goto out_free_txq;
1180
1181         lp->flags |= LDC_FLAG_ALLOCED_QUEUES;
1182
1183         lp->hs_state = LDC_HS_CLOSED;
1184         ldc_set_state(lp, LDC_STATE_INIT);
1185
1186         INIT_HLIST_NODE(&lp->list);
1187         hlist_add_head(&lp->list, &ldc_channel_list);
1188
1189         INIT_HLIST_HEAD(&lp->mh_list);
1190
1191         snprintf(lp->rx_irq_name, LDC_IRQ_NAME_MAX, "%s RX", name);
1192         snprintf(lp->tx_irq_name, LDC_IRQ_NAME_MAX, "%s TX", name);
1193
1194         err = request_irq(lp->cfg.rx_irq, ldc_rx, 0,
1195                           lp->rx_irq_name, lp);
1196         if (err)
1197                 goto out_free_txq;
1198
1199         err = request_irq(lp->cfg.tx_irq, ldc_tx, 0,
1200                           lp->tx_irq_name, lp);
1201         if (err) {
1202                 free_irq(lp->cfg.rx_irq, lp);
1203                 goto out_free_txq;
1204         }
1205
1206         return lp;
1207
1208 out_free_txq:
1209         free_queue(lp->tx_num_entries, lp->tx_base);
1210
1211 out_free_mssbuf:
1212         kfree(mssbuf);
1213
1214 out_free_iommu:
1215         ldc_iommu_release(lp);
1216
1217 out_free_ldc:
1218         kfree(lp);
1219
1220 out_err:
1221         return ERR_PTR(err);
1222 }
1223 EXPORT_SYMBOL(ldc_alloc);
1224
1225 void ldc_unbind(struct ldc_channel *lp)
1226 {
1227         if (lp->flags & LDC_FLAG_REGISTERED_IRQS) {
1228                 free_irq(lp->cfg.rx_irq, lp);
1229                 free_irq(lp->cfg.tx_irq, lp);
1230                 lp->flags &= ~LDC_FLAG_REGISTERED_IRQS;
1231         }
1232
1233         if (lp->flags & LDC_FLAG_REGISTERED_QUEUES) {
1234                 sun4v_ldc_tx_qconf(lp->id, 0, 0);
1235                 sun4v_ldc_rx_qconf(lp->id, 0, 0);
1236                 lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1237         }
1238         if (lp->flags & LDC_FLAG_ALLOCED_QUEUES) {
1239                 free_queue(lp->tx_num_entries, lp->tx_base);
1240                 free_queue(lp->rx_num_entries, lp->rx_base);
1241                 lp->flags &= ~LDC_FLAG_ALLOCED_QUEUES;
1242         }
1243
1244         ldc_set_state(lp, LDC_STATE_INIT);
1245 }
1246 EXPORT_SYMBOL(ldc_unbind);
1247
1248 void ldc_free(struct ldc_channel *lp)
1249 {
1250         ldc_unbind(lp);
1251         hlist_del(&lp->list);
1252         kfree(lp->mssbuf);
1253         ldc_iommu_release(lp);
1254
1255         kfree(lp);
1256 }
1257 EXPORT_SYMBOL(ldc_free);
1258
1259 /* Bind the channel.  This registers the LDC queues with
1260  * the hypervisor and puts the channel into a pseudo-listening
1261  * state.  This does not initiate a handshake, ldc_connect() does
1262  * that.
1263  */
1264 int ldc_bind(struct ldc_channel *lp)
1265 {
1266         unsigned long hv_err, flags;
1267         int err = -EINVAL;
1268
1269         if (lp->state != LDC_STATE_INIT)
1270                 return -EINVAL;
1271
1272         spin_lock_irqsave(&lp->lock, flags);
1273
1274         enable_irq(lp->cfg.rx_irq);
1275         enable_irq(lp->cfg.tx_irq);
1276
1277         lp->flags |= LDC_FLAG_REGISTERED_IRQS;
1278
1279         err = -ENODEV;
1280         hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1281         if (hv_err)
1282                 goto out_free_irqs;
1283
1284         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1285         if (hv_err)
1286                 goto out_free_irqs;
1287
1288         hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1289         if (hv_err)
1290                 goto out_unmap_tx;
1291
1292         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1293         if (hv_err)
1294                 goto out_unmap_tx;
1295
1296         lp->flags |= LDC_FLAG_REGISTERED_QUEUES;
1297
1298         hv_err = sun4v_ldc_tx_get_state(lp->id,
1299                                         &lp->tx_head,
1300                                         &lp->tx_tail,
1301                                         &lp->chan_state);
1302         err = -EBUSY;
1303         if (hv_err)
1304                 goto out_unmap_rx;
1305
1306         lp->tx_acked = lp->tx_head;
1307
1308         lp->hs_state = LDC_HS_OPEN;
1309         ldc_set_state(lp, LDC_STATE_BOUND);
1310
1311         spin_unlock_irqrestore(&lp->lock, flags);
1312
1313         return 0;
1314
1315 out_unmap_rx:
1316         lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1317         sun4v_ldc_rx_qconf(lp->id, 0, 0);
1318
1319 out_unmap_tx:
1320         sun4v_ldc_tx_qconf(lp->id, 0, 0);
1321
1322 out_free_irqs:
1323         lp->flags &= ~LDC_FLAG_REGISTERED_IRQS;
1324         free_irq(lp->cfg.tx_irq, lp);
1325         free_irq(lp->cfg.rx_irq, lp);
1326
1327         spin_unlock_irqrestore(&lp->lock, flags);
1328
1329         return err;
1330 }
1331 EXPORT_SYMBOL(ldc_bind);
1332
1333 int ldc_connect(struct ldc_channel *lp)
1334 {
1335         unsigned long flags;
1336         int err;
1337
1338         if (lp->cfg.mode == LDC_MODE_RAW)
1339                 return -EINVAL;
1340
1341         spin_lock_irqsave(&lp->lock, flags);
1342
1343         if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1344             !(lp->flags & LDC_FLAG_REGISTERED_QUEUES) ||
1345             lp->hs_state != LDC_HS_OPEN)
1346                 err = ((lp->hs_state > LDC_HS_OPEN) ? 0 : -EINVAL);
1347         else
1348                 err = start_handshake(lp);
1349
1350         spin_unlock_irqrestore(&lp->lock, flags);
1351
1352         return err;
1353 }
1354 EXPORT_SYMBOL(ldc_connect);
1355
1356 int ldc_disconnect(struct ldc_channel *lp)
1357 {
1358         unsigned long hv_err, flags;
1359         int err;
1360
1361         if (lp->cfg.mode == LDC_MODE_RAW)
1362                 return -EINVAL;
1363
1364         if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1365             !(lp->flags & LDC_FLAG_REGISTERED_QUEUES))
1366                 return -EINVAL;
1367
1368         spin_lock_irqsave(&lp->lock, flags);
1369
1370         err = -ENODEV;
1371         hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1372         if (hv_err)
1373                 goto out_err;
1374
1375         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1376         if (hv_err)
1377                 goto out_err;
1378
1379         hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1380         if (hv_err)
1381                 goto out_err;
1382
1383         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1384         if (hv_err)
1385                 goto out_err;
1386
1387         ldc_set_state(lp, LDC_STATE_BOUND);
1388         lp->hs_state = LDC_HS_OPEN;
1389         lp->flags |= LDC_FLAG_RESET;
1390
1391         spin_unlock_irqrestore(&lp->lock, flags);
1392
1393         return 0;
1394
1395 out_err:
1396         sun4v_ldc_tx_qconf(lp->id, 0, 0);
1397         sun4v_ldc_rx_qconf(lp->id, 0, 0);
1398         free_irq(lp->cfg.tx_irq, lp);
1399         free_irq(lp->cfg.rx_irq, lp);
1400         lp->flags &= ~(LDC_FLAG_REGISTERED_IRQS |
1401                        LDC_FLAG_REGISTERED_QUEUES);
1402         ldc_set_state(lp, LDC_STATE_INIT);
1403
1404         spin_unlock_irqrestore(&lp->lock, flags);
1405
1406         return err;
1407 }
1408 EXPORT_SYMBOL(ldc_disconnect);
1409
1410 int ldc_state(struct ldc_channel *lp)
1411 {
1412         return lp->state;
1413 }
1414 EXPORT_SYMBOL(ldc_state);
1415
1416 static int write_raw(struct ldc_channel *lp, const void *buf, unsigned int size)
1417 {
1418         struct ldc_packet *p;
1419         unsigned long new_tail;
1420         int err;
1421
1422         if (size > LDC_PACKET_SIZE)
1423                 return -EMSGSIZE;
1424
1425         p = data_get_tx_packet(lp, &new_tail);
1426         if (!p)
1427                 return -EAGAIN;
1428
1429         memcpy(p, buf, size);
1430
1431         err = send_tx_packet(lp, p, new_tail);
1432         if (!err)
1433                 err = size;
1434
1435         return err;
1436 }
1437
1438 static int read_raw(struct ldc_channel *lp, void *buf, unsigned int size)
1439 {
1440         struct ldc_packet *p;
1441         unsigned long hv_err, new;
1442         int err;
1443
1444         if (size < LDC_PACKET_SIZE)
1445                 return -EINVAL;
1446
1447         hv_err = sun4v_ldc_rx_get_state(lp->id,
1448                                         &lp->rx_head,
1449                                         &lp->rx_tail,
1450                                         &lp->chan_state);
1451         if (hv_err)
1452                 return ldc_abort(lp);
1453
1454         if (lp->chan_state == LDC_CHANNEL_DOWN ||
1455             lp->chan_state == LDC_CHANNEL_RESETTING)
1456                 return -ECONNRESET;
1457
1458         if (lp->rx_head == lp->rx_tail)
1459                 return 0;
1460
1461         p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
1462         memcpy(buf, p, LDC_PACKET_SIZE);
1463
1464         new = rx_advance(lp, lp->rx_head);
1465         lp->rx_head = new;
1466
1467         err = __set_rx_head(lp, new);
1468         if (err < 0)
1469                 err = -ECONNRESET;
1470         else
1471                 err = LDC_PACKET_SIZE;
1472
1473         return err;
1474 }
1475
1476 static const struct ldc_mode_ops raw_ops = {
1477         .write          =       write_raw,
1478         .read           =       read_raw,
1479 };
1480
1481 static int write_nonraw(struct ldc_channel *lp, const void *buf,
1482                         unsigned int size)
1483 {
1484         unsigned long hv_err, tail;
1485         unsigned int copied;
1486         u32 seq;
1487         int err;
1488
1489         hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail,
1490                                         &lp->chan_state);
1491         if (unlikely(hv_err))
1492                 return -EBUSY;
1493
1494         if (unlikely(lp->chan_state != LDC_CHANNEL_UP))
1495                 return ldc_abort(lp);
1496
1497         if (!tx_has_space_for(lp, size))
1498                 return -EAGAIN;
1499
1500         seq = lp->snd_nxt;
1501         copied = 0;
1502         tail = lp->tx_tail;
1503         while (copied < size) {
1504                 struct ldc_packet *p = lp->tx_base + (tail / LDC_PACKET_SIZE);
1505                 u8 *data = ((lp->cfg.mode == LDC_MODE_UNRELIABLE) ?
1506                             p->u.u_data :
1507                             p->u.r.r_data);
1508                 int data_len;
1509
1510                 p->type = LDC_DATA;
1511                 p->stype = LDC_INFO;
1512                 p->ctrl = 0;
1513
1514                 data_len = size - copied;
1515                 if (data_len > lp->mss)
1516                         data_len = lp->mss;
1517
1518                 BUG_ON(data_len > LDC_LEN);
1519
1520                 p->env = (data_len |
1521                           (copied == 0 ? LDC_START : 0) |
1522                           (data_len == size - copied ? LDC_STOP : 0));
1523
1524                 p->seqid = ++seq;
1525
1526                 ldcdbg(DATA, "SENT DATA [%02x:%02x:%02x:%02x:%08x]\n",
1527                        p->type,
1528                        p->stype,
1529                        p->ctrl,
1530                        p->env,
1531                        p->seqid);
1532
1533                 memcpy(data, buf, data_len);
1534                 buf += data_len;
1535                 copied += data_len;
1536
1537                 tail = tx_advance(lp, tail);
1538         }
1539
1540         err = set_tx_tail(lp, tail);
1541         if (!err) {
1542                 lp->snd_nxt = seq;
1543                 err = size;
1544         }
1545
1546         return err;
1547 }
1548
1549 static int rx_bad_seq(struct ldc_channel *lp, struct ldc_packet *p,
1550                       struct ldc_packet *first_frag)
1551 {
1552         int err;
1553
1554         if (first_frag)
1555                 lp->rcv_nxt = first_frag->seqid - 1;
1556
1557         err = send_data_nack(lp, p);
1558         if (err)
1559                 return err;
1560
1561         err = __set_rx_head(lp, lp->rx_tail);
1562         if (err < 0)
1563                 return ldc_abort(lp);
1564
1565         return 0;
1566 }
1567
1568 static int data_ack_nack(struct ldc_channel *lp, struct ldc_packet *p)
1569 {
1570         if (p->stype & LDC_ACK) {
1571                 int err = process_data_ack(lp, p);
1572                 if (err)
1573                         return err;
1574         }
1575         if (p->stype & LDC_NACK)
1576                 return ldc_abort(lp);
1577
1578         return 0;
1579 }
1580
1581 static int rx_data_wait(struct ldc_channel *lp, unsigned long cur_head)
1582 {
1583         unsigned long dummy;
1584         int limit = 1000;
1585
1586         ldcdbg(DATA, "DATA WAIT cur_head[%lx] rx_head[%lx] rx_tail[%lx]\n",
1587                cur_head, lp->rx_head, lp->rx_tail);
1588         while (limit-- > 0) {
1589                 unsigned long hv_err;
1590
1591                 hv_err = sun4v_ldc_rx_get_state(lp->id,
1592                                                 &dummy,
1593                                                 &lp->rx_tail,
1594                                                 &lp->chan_state);
1595                 if (hv_err)
1596                         return ldc_abort(lp);
1597
1598                 if (lp->chan_state == LDC_CHANNEL_DOWN ||
1599                     lp->chan_state == LDC_CHANNEL_RESETTING)
1600                         return -ECONNRESET;
1601
1602                 if (cur_head != lp->rx_tail) {
1603                         ldcdbg(DATA, "DATA WAIT DONE "
1604                                "head[%lx] tail[%lx] chan_state[%lx]\n",
1605                                dummy, lp->rx_tail, lp->chan_state);
1606                         return 0;
1607                 }
1608
1609                 udelay(1);
1610         }
1611         return -EAGAIN;
1612 }
1613
1614 static int rx_set_head(struct ldc_channel *lp, unsigned long head)
1615 {
1616         int err = __set_rx_head(lp, head);
1617
1618         if (err < 0)
1619                 return ldc_abort(lp);
1620
1621         lp->rx_head = head;
1622         return 0;
1623 }
1624
1625 static void send_data_ack(struct ldc_channel *lp)
1626 {
1627         unsigned long new_tail;
1628         struct ldc_packet *p;
1629
1630         p = data_get_tx_packet(lp, &new_tail);
1631         if (likely(p)) {
1632                 int err;
1633
1634                 memset(p, 0, sizeof(*p));
1635                 p->type = LDC_DATA;
1636                 p->stype = LDC_ACK;
1637                 p->ctrl = 0;
1638                 p->seqid = lp->snd_nxt + 1;
1639                 p->u.r.ackid = lp->rcv_nxt;
1640
1641                 err = send_tx_packet(lp, p, new_tail);
1642                 if (!err)
1643                         lp->snd_nxt++;
1644         }
1645 }
1646
1647 static int read_nonraw(struct ldc_channel *lp, void *buf, unsigned int size)
1648 {
1649         struct ldc_packet *first_frag;
1650         unsigned long hv_err, new;
1651         int err, copied;
1652
1653         hv_err = sun4v_ldc_rx_get_state(lp->id,
1654                                         &lp->rx_head,
1655                                         &lp->rx_tail,
1656                                         &lp->chan_state);
1657         if (hv_err)
1658                 return ldc_abort(lp);
1659
1660         if (lp->chan_state == LDC_CHANNEL_DOWN ||
1661             lp->chan_state == LDC_CHANNEL_RESETTING)
1662                 return -ECONNRESET;
1663
1664         if (lp->rx_head == lp->rx_tail)
1665                 return 0;
1666
1667         first_frag = NULL;
1668         copied = err = 0;
1669         new = lp->rx_head;
1670         while (1) {
1671                 struct ldc_packet *p;
1672                 int pkt_len;
1673
1674                 BUG_ON(new == lp->rx_tail);
1675                 p = lp->rx_base + (new / LDC_PACKET_SIZE);
1676
1677                 ldcdbg(RX, "RX read pkt[%02x:%02x:%02x:%02x:%08x:%08x] "
1678                        "rcv_nxt[%08x]\n",
1679                        p->type,
1680                        p->stype,
1681                        p->ctrl,
1682                        p->env,
1683                        p->seqid,
1684                        p->u.r.ackid,
1685                        lp->rcv_nxt);
1686
1687                 if (unlikely(!rx_seq_ok(lp, p->seqid))) {
1688                         err = rx_bad_seq(lp, p, first_frag);
1689                         copied = 0;
1690                         break;
1691                 }
1692
1693                 if (p->type & LDC_CTRL) {
1694                         err = process_control_frame(lp, p);
1695                         if (err < 0)
1696                                 break;
1697                         err = 0;
1698                 }
1699
1700                 lp->rcv_nxt = p->seqid;
1701
1702                 if (!(p->type & LDC_DATA)) {
1703                         new = rx_advance(lp, new);
1704                         goto no_data;
1705                 }
1706                 if (p->stype & (LDC_ACK | LDC_NACK)) {
1707                         err = data_ack_nack(lp, p);
1708                         if (err)
1709                                 break;
1710                 }
1711                 if (!(p->stype & LDC_INFO)) {
1712                         new = rx_advance(lp, new);
1713                         err = rx_set_head(lp, new);
1714                         if (err)
1715                                 break;
1716                         goto no_data;
1717                 }
1718
1719                 pkt_len = p->env & LDC_LEN;
1720
1721                 /* Every initial packet starts with the START bit set.
1722                  *
1723                  * Singleton packets will have both START+STOP set.
1724                  *
1725                  * Fragments will have START set in the first frame, STOP
1726                  * set in the last frame, and neither bit set in middle
1727                  * frames of the packet.
1728                  *
1729                  * Therefore if we are at the beginning of a packet and
1730                  * we don't see START, or we are in the middle of a fragmented
1731                  * packet and do see START, we are unsynchronized and should
1732                  * flush the RX queue.
1733                  */
1734                 if ((first_frag == NULL && !(p->env & LDC_START)) ||
1735                     (first_frag != NULL &&  (p->env & LDC_START))) {
1736                         if (!first_frag)
1737                                 new = rx_advance(lp, new);
1738
1739                         err = rx_set_head(lp, new);
1740                         if (err)
1741                                 break;
1742
1743                         if (!first_frag)
1744                                 goto no_data;
1745                 }
1746                 if (!first_frag)
1747                         first_frag = p;
1748
1749                 if (pkt_len > size - copied) {
1750                         /* User didn't give us a big enough buffer,
1751                          * what to do?  This is a pretty serious error.
1752                          *
1753                          * Since we haven't updated the RX ring head to
1754                          * consume any of the packets, signal the error
1755                          * to the user and just leave the RX ring alone.
1756                          *
1757                          * This seems the best behavior because this allows
1758                          * a user of the LDC layer to start with a small
1759                          * RX buffer for ldc_read() calls and use -EMSGSIZE
1760                          * as a cue to enlarge it's read buffer.
1761                          */
1762                         err = -EMSGSIZE;
1763                         break;
1764                 }
1765
1766                 /* Ok, we are gonna eat this one.  */
1767                 new = rx_advance(lp, new);
1768
1769                 memcpy(buf,
1770                        (lp->cfg.mode == LDC_MODE_UNRELIABLE ?
1771                         p->u.u_data : p->u.r.r_data), pkt_len);
1772                 buf += pkt_len;
1773                 copied += pkt_len;
1774
1775                 if (p->env & LDC_STOP)
1776                         break;
1777
1778 no_data:
1779                 if (new == lp->rx_tail) {
1780                         err = rx_data_wait(lp, new);
1781                         if (err)
1782                                 break;
1783                 }
1784         }
1785
1786         if (!err)
1787                 err = rx_set_head(lp, new);
1788
1789         if (err && first_frag)
1790                 lp->rcv_nxt = first_frag->seqid - 1;
1791
1792         if (!err) {
1793                 err = copied;
1794                 if (err > 0 && lp->cfg.mode != LDC_MODE_UNRELIABLE)
1795                         send_data_ack(lp);
1796         }
1797
1798         return err;
1799 }
1800
1801 static const struct ldc_mode_ops nonraw_ops = {
1802         .write          =       write_nonraw,
1803         .read           =       read_nonraw,
1804 };
1805
1806 static int write_stream(struct ldc_channel *lp, const void *buf,
1807                         unsigned int size)
1808 {
1809         if (size > lp->cfg.mtu)
1810                 size = lp->cfg.mtu;
1811         return write_nonraw(lp, buf, size);
1812 }
1813
1814 static int read_stream(struct ldc_channel *lp, void *buf, unsigned int size)
1815 {
1816         if (!lp->mssbuf_len) {
1817                 int err = read_nonraw(lp, lp->mssbuf, lp->cfg.mtu);
1818                 if (err < 0)
1819                         return err;
1820
1821                 lp->mssbuf_len = err;
1822                 lp->mssbuf_off = 0;
1823         }
1824
1825         if (size > lp->mssbuf_len)
1826                 size = lp->mssbuf_len;
1827         memcpy(buf, lp->mssbuf + lp->mssbuf_off, size);
1828
1829         lp->mssbuf_off += size;
1830         lp->mssbuf_len -= size;
1831
1832         return size;
1833 }
1834
1835 static const struct ldc_mode_ops stream_ops = {
1836         .write          =       write_stream,
1837         .read           =       read_stream,
1838 };
1839
1840 int ldc_write(struct ldc_channel *lp, const void *buf, unsigned int size)
1841 {
1842         unsigned long flags;
1843         int err;
1844
1845         if (!buf)
1846                 return -EINVAL;
1847
1848         if (!size)
1849                 return 0;
1850
1851         spin_lock_irqsave(&lp->lock, flags);
1852
1853         if (lp->hs_state != LDC_HS_COMPLETE)
1854                 err = -ENOTCONN;
1855         else
1856                 err = lp->mops->write(lp, buf, size);
1857
1858         spin_unlock_irqrestore(&lp->lock, flags);
1859
1860         return err;
1861 }
1862 EXPORT_SYMBOL(ldc_write);
1863
1864 int ldc_read(struct ldc_channel *lp, void *buf, unsigned int size)
1865 {
1866         unsigned long flags;
1867         int err;
1868
1869         if (!buf)
1870                 return -EINVAL;
1871
1872         if (!size)
1873                 return 0;
1874
1875         spin_lock_irqsave(&lp->lock, flags);
1876
1877         if (lp->hs_state != LDC_HS_COMPLETE)
1878                 err = -ENOTCONN;
1879         else
1880                 err = lp->mops->read(lp, buf, size);
1881
1882         spin_unlock_irqrestore(&lp->lock, flags);
1883
1884         return err;
1885 }
1886 EXPORT_SYMBOL(ldc_read);
1887
1888 static long arena_alloc(struct ldc_iommu *iommu, unsigned long npages)
1889 {
1890         struct iommu_arena *arena = &iommu->arena;
1891         unsigned long n, start, end, limit;
1892         int pass;
1893
1894         limit = arena->limit;
1895         start = arena->hint;
1896         pass = 0;
1897
1898 again:
1899         n = bitmap_find_next_zero_area(arena->map, limit, start, npages, 0);
1900         end = n + npages;
1901         if (unlikely(end >= limit)) {
1902                 if (likely(pass < 1)) {
1903                         limit = start;
1904                         start = 0;
1905                         pass++;
1906                         goto again;
1907                 } else {
1908                         /* Scanned the whole thing, give up. */
1909                         return -1;
1910                 }
1911         }
1912         bitmap_set(arena->map, n, npages);
1913
1914         arena->hint = end;
1915
1916         return n;
1917 }
1918
1919 #define COOKIE_PGSZ_CODE        0xf000000000000000ULL
1920 #define COOKIE_PGSZ_CODE_SHIFT  60ULL
1921
1922 static u64 pagesize_code(void)
1923 {
1924         switch (PAGE_SIZE) {
1925         default:
1926         case (8ULL * 1024ULL):
1927                 return 0;
1928         case (64ULL * 1024ULL):
1929                 return 1;
1930         case (512ULL * 1024ULL):
1931                 return 2;
1932         case (4ULL * 1024ULL * 1024ULL):
1933                 return 3;
1934         case (32ULL * 1024ULL * 1024ULL):
1935                 return 4;
1936         case (256ULL * 1024ULL * 1024ULL):
1937                 return 5;
1938         }
1939 }
1940
1941 static u64 make_cookie(u64 index, u64 pgsz_code, u64 page_offset)
1942 {
1943         return ((pgsz_code << COOKIE_PGSZ_CODE_SHIFT) |
1944                 (index << PAGE_SHIFT) |
1945                 page_offset);
1946 }
1947
1948 static u64 cookie_to_index(u64 cookie, unsigned long *shift)
1949 {
1950         u64 szcode = cookie >> COOKIE_PGSZ_CODE_SHIFT;
1951
1952         cookie &= ~COOKIE_PGSZ_CODE;
1953
1954         *shift = szcode * 3;
1955
1956         return (cookie >> (13ULL + (szcode * 3ULL)));
1957 }
1958
1959 static struct ldc_mtable_entry *alloc_npages(struct ldc_iommu *iommu,
1960                                              unsigned long npages)
1961 {
1962         long entry;
1963
1964         entry = arena_alloc(iommu, npages);
1965         if (unlikely(entry < 0))
1966                 return NULL;
1967
1968         return iommu->page_table + entry;
1969 }
1970
1971 static u64 perm_to_mte(unsigned int map_perm)
1972 {
1973         u64 mte_base;
1974
1975         mte_base = pagesize_code();
1976
1977         if (map_perm & LDC_MAP_SHADOW) {
1978                 if (map_perm & LDC_MAP_R)
1979                         mte_base |= LDC_MTE_COPY_R;
1980                 if (map_perm & LDC_MAP_W)
1981                         mte_base |= LDC_MTE_COPY_W;
1982         }
1983         if (map_perm & LDC_MAP_DIRECT) {
1984                 if (map_perm & LDC_MAP_R)
1985                         mte_base |= LDC_MTE_READ;
1986                 if (map_perm & LDC_MAP_W)
1987                         mte_base |= LDC_MTE_WRITE;
1988                 if (map_perm & LDC_MAP_X)
1989                         mte_base |= LDC_MTE_EXEC;
1990         }
1991         if (map_perm & LDC_MAP_IO) {
1992                 if (map_perm & LDC_MAP_R)
1993                         mte_base |= LDC_MTE_IOMMU_R;
1994                 if (map_perm & LDC_MAP_W)
1995                         mte_base |= LDC_MTE_IOMMU_W;
1996         }
1997
1998         return mte_base;
1999 }
2000
2001 static int pages_in_region(unsigned long base, long len)
2002 {
2003         int count = 0;
2004
2005         do {
2006                 unsigned long new = (base + PAGE_SIZE) & PAGE_MASK;
2007
2008                 len -= (new - base);
2009                 base = new;
2010                 count++;
2011         } while (len > 0);
2012
2013         return count;
2014 }
2015
2016 struct cookie_state {
2017         struct ldc_mtable_entry         *page_table;
2018         struct ldc_trans_cookie         *cookies;
2019         u64                             mte_base;
2020         u64                             prev_cookie;
2021         u32                             pte_idx;
2022         u32                             nc;
2023 };
2024
2025 static void fill_cookies(struct cookie_state *sp, unsigned long pa,
2026                          unsigned long off, unsigned long len)
2027 {
2028         do {
2029                 unsigned long tlen, new = pa + PAGE_SIZE;
2030                 u64 this_cookie;
2031
2032                 sp->page_table[sp->pte_idx].mte = sp->mte_base | pa;
2033
2034                 tlen = PAGE_SIZE;
2035                 if (off)
2036                         tlen = PAGE_SIZE - off;
2037                 if (tlen > len)
2038                         tlen = len;
2039
2040                 this_cookie = make_cookie(sp->pte_idx,
2041                                           pagesize_code(), off);
2042
2043                 off = 0;
2044
2045                 if (this_cookie == sp->prev_cookie) {
2046                         sp->cookies[sp->nc - 1].cookie_size += tlen;
2047                 } else {
2048                         sp->cookies[sp->nc].cookie_addr = this_cookie;
2049                         sp->cookies[sp->nc].cookie_size = tlen;
2050                         sp->nc++;
2051                 }
2052                 sp->prev_cookie = this_cookie + tlen;
2053
2054                 sp->pte_idx++;
2055
2056                 len -= tlen;
2057                 pa = new;
2058         } while (len > 0);
2059 }
2060
2061 static int sg_count_one(struct scatterlist *sg)
2062 {
2063         unsigned long base = page_to_pfn(sg_page(sg)) << PAGE_SHIFT;
2064         long len = sg->length;
2065
2066         if ((sg->offset | len) & (8UL - 1))
2067                 return -EFAULT;
2068
2069         return pages_in_region(base + sg->offset, len);
2070 }
2071
2072 static int sg_count_pages(struct scatterlist *sg, int num_sg)
2073 {
2074         int count;
2075         int i;
2076
2077         count = 0;
2078         for (i = 0; i < num_sg; i++) {
2079                 int err = sg_count_one(sg + i);
2080                 if (err < 0)
2081                         return err;
2082                 count += err;
2083         }
2084
2085         return count;
2086 }
2087
2088 int ldc_map_sg(struct ldc_channel *lp,
2089                struct scatterlist *sg, int num_sg,
2090                struct ldc_trans_cookie *cookies, int ncookies,
2091                unsigned int map_perm)
2092 {
2093         unsigned long i, npages, flags;
2094         struct ldc_mtable_entry *base;
2095         struct cookie_state state;
2096         struct ldc_iommu *iommu;
2097         int err;
2098
2099         if (map_perm & ~LDC_MAP_ALL)
2100                 return -EINVAL;
2101
2102         err = sg_count_pages(sg, num_sg);
2103         if (err < 0)
2104                 return err;
2105
2106         npages = err;
2107         if (err > ncookies)
2108                 return -EMSGSIZE;
2109
2110         iommu = &lp->iommu;
2111
2112         spin_lock_irqsave(&iommu->lock, flags);
2113         base = alloc_npages(iommu, npages);
2114         spin_unlock_irqrestore(&iommu->lock, flags);
2115
2116         if (!base)
2117                 return -ENOMEM;
2118
2119         state.page_table = iommu->page_table;
2120         state.cookies = cookies;
2121         state.mte_base = perm_to_mte(map_perm);
2122         state.prev_cookie = ~(u64)0;
2123         state.pte_idx = (base - iommu->page_table);
2124         state.nc = 0;
2125
2126         for (i = 0; i < num_sg; i++)
2127                 fill_cookies(&state, page_to_pfn(sg_page(&sg[i])) << PAGE_SHIFT,
2128                              sg[i].offset, sg[i].length);
2129
2130         return state.nc;
2131 }
2132 EXPORT_SYMBOL(ldc_map_sg);
2133
2134 int ldc_map_single(struct ldc_channel *lp,
2135                    void *buf, unsigned int len,
2136                    struct ldc_trans_cookie *cookies, int ncookies,
2137                    unsigned int map_perm)
2138 {
2139         unsigned long npages, pa, flags;
2140         struct ldc_mtable_entry *base;
2141         struct cookie_state state;
2142         struct ldc_iommu *iommu;
2143
2144         if ((map_perm & ~LDC_MAP_ALL) || (ncookies < 1))
2145                 return -EINVAL;
2146
2147         pa = __pa(buf);
2148         if ((pa | len) & (8UL - 1))
2149                 return -EFAULT;
2150
2151         npages = pages_in_region(pa, len);
2152
2153         iommu = &lp->iommu;
2154
2155         spin_lock_irqsave(&iommu->lock, flags);
2156         base = alloc_npages(iommu, npages);
2157         spin_unlock_irqrestore(&iommu->lock, flags);
2158
2159         if (!base)
2160                 return -ENOMEM;
2161
2162         state.page_table = iommu->page_table;
2163         state.cookies = cookies;
2164         state.mte_base = perm_to_mte(map_perm);
2165         state.prev_cookie = ~(u64)0;
2166         state.pte_idx = (base - iommu->page_table);
2167         state.nc = 0;
2168         fill_cookies(&state, (pa & PAGE_MASK), (pa & ~PAGE_MASK), len);
2169         BUG_ON(state.nc > ncookies);
2170
2171         return state.nc;
2172 }
2173 EXPORT_SYMBOL(ldc_map_single);
2174
2175 static void free_npages(unsigned long id, struct ldc_iommu *iommu,
2176                         u64 cookie, u64 size)
2177 {
2178         struct iommu_arena *arena = &iommu->arena;
2179         unsigned long i, shift, index, npages;
2180         struct ldc_mtable_entry *base;
2181
2182         npages = PAGE_ALIGN(((cookie & ~PAGE_MASK) + size)) >> PAGE_SHIFT;
2183         index = cookie_to_index(cookie, &shift);
2184         base = iommu->page_table + index;
2185
2186         BUG_ON(index > arena->limit ||
2187                (index + npages) > arena->limit);
2188
2189         for (i = 0; i < npages; i++) {
2190                 if (base->cookie)
2191                         sun4v_ldc_revoke(id, cookie + (i << shift),
2192                                          base->cookie);
2193                 base->mte = 0;
2194                 __clear_bit(index + i, arena->map);
2195         }
2196 }
2197
2198 void ldc_unmap(struct ldc_channel *lp, struct ldc_trans_cookie *cookies,
2199                int ncookies)
2200 {
2201         struct ldc_iommu *iommu = &lp->iommu;
2202         unsigned long flags;
2203         int i;
2204
2205         spin_lock_irqsave(&iommu->lock, flags);
2206         for (i = 0; i < ncookies; i++) {
2207                 u64 addr = cookies[i].cookie_addr;
2208                 u64 size = cookies[i].cookie_size;
2209
2210                 free_npages(lp->id, iommu, addr, size);
2211         }
2212         spin_unlock_irqrestore(&iommu->lock, flags);
2213 }
2214 EXPORT_SYMBOL(ldc_unmap);
2215
2216 int ldc_copy(struct ldc_channel *lp, int copy_dir,
2217              void *buf, unsigned int len, unsigned long offset,
2218              struct ldc_trans_cookie *cookies, int ncookies)
2219 {
2220         unsigned int orig_len;
2221         unsigned long ra;
2222         int i;
2223
2224         if (copy_dir != LDC_COPY_IN && copy_dir != LDC_COPY_OUT) {
2225                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Bad copy_dir[%d]\n",
2226                        lp->id, copy_dir);
2227                 return -EINVAL;
2228         }
2229
2230         ra = __pa(buf);
2231         if ((ra | len | offset) & (8UL - 1)) {
2232                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Unaligned buffer "
2233                        "ra[%lx] len[%x] offset[%lx]\n",
2234                        lp->id, ra, len, offset);
2235                 return -EFAULT;
2236         }
2237
2238         if (lp->hs_state != LDC_HS_COMPLETE ||
2239             (lp->flags & LDC_FLAG_RESET)) {
2240                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Link down hs_state[%x] "
2241                        "flags[%x]\n", lp->id, lp->hs_state, lp->flags);
2242                 return -ECONNRESET;
2243         }
2244
2245         orig_len = len;
2246         for (i = 0; i < ncookies; i++) {
2247                 unsigned long cookie_raddr = cookies[i].cookie_addr;
2248                 unsigned long this_len = cookies[i].cookie_size;
2249                 unsigned long actual_len;
2250
2251                 if (unlikely(offset)) {
2252                         unsigned long this_off = offset;
2253
2254                         if (this_off > this_len)
2255                                 this_off = this_len;
2256
2257                         offset -= this_off;
2258                         this_len -= this_off;
2259                         if (!this_len)
2260                                 continue;
2261                         cookie_raddr += this_off;
2262                 }
2263
2264                 if (this_len > len)
2265                         this_len = len;
2266
2267                 while (1) {
2268                         unsigned long hv_err;
2269
2270                         hv_err = sun4v_ldc_copy(lp->id, copy_dir,
2271                                                 cookie_raddr, ra,
2272                                                 this_len, &actual_len);
2273                         if (unlikely(hv_err)) {
2274                                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] "
2275                                        "HV error %lu\n",
2276                                        lp->id, hv_err);
2277                                 if (lp->hs_state != LDC_HS_COMPLETE ||
2278                                     (lp->flags & LDC_FLAG_RESET))
2279                                         return -ECONNRESET;
2280                                 else
2281                                         return -EFAULT;
2282                         }
2283
2284                         cookie_raddr += actual_len;
2285                         ra += actual_len;
2286                         len -= actual_len;
2287                         if (actual_len == this_len)
2288                                 break;
2289
2290                         this_len -= actual_len;
2291                 }
2292
2293                 if (!len)
2294                         break;
2295         }
2296
2297         /* It is caller policy what to do about short copies.
2298          * For example, a networking driver can declare the
2299          * packet a runt and drop it.
2300          */
2301
2302         return orig_len - len;
2303 }
2304 EXPORT_SYMBOL(ldc_copy);
2305
2306 void *ldc_alloc_exp_dring(struct ldc_channel *lp, unsigned int len,
2307                           struct ldc_trans_cookie *cookies, int *ncookies,
2308                           unsigned int map_perm)
2309 {
2310         void *buf;
2311         int err;
2312
2313         if (len & (8UL - 1))
2314                 return ERR_PTR(-EINVAL);
2315
2316         buf = kzalloc(len, GFP_KERNEL);
2317         if (!buf)
2318                 return ERR_PTR(-ENOMEM);
2319
2320         err = ldc_map_single(lp, buf, len, cookies, *ncookies, map_perm);
2321         if (err < 0) {
2322                 kfree(buf);
2323                 return ERR_PTR(err);
2324         }
2325         *ncookies = err;
2326
2327         return buf;
2328 }
2329 EXPORT_SYMBOL(ldc_alloc_exp_dring);
2330
2331 void ldc_free_exp_dring(struct ldc_channel *lp, void *buf, unsigned int len,
2332                         struct ldc_trans_cookie *cookies, int ncookies)
2333 {
2334         ldc_unmap(lp, cookies, ncookies);
2335         kfree(buf);
2336 }
2337 EXPORT_SYMBOL(ldc_free_exp_dring);
2338
2339 static int __init ldc_init(void)
2340 {
2341         unsigned long major, minor;
2342         struct mdesc_handle *hp;
2343         const u64 *v;
2344         int err;
2345         u64 mp;
2346
2347         hp = mdesc_grab();
2348         if (!hp)
2349                 return -ENODEV;
2350
2351         mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
2352         err = -ENODEV;
2353         if (mp == MDESC_NODE_NULL)
2354                 goto out;
2355
2356         v = mdesc_get_property(hp, mp, "domaining-enabled", NULL);
2357         if (!v)
2358                 goto out;
2359
2360         major = 1;
2361         minor = 0;
2362         if (sun4v_hvapi_register(HV_GRP_LDOM, major, &minor)) {
2363                 printk(KERN_INFO PFX "Could not register LDOM hvapi.\n");
2364                 goto out;
2365         }
2366
2367         printk(KERN_INFO "%s", version);
2368
2369         if (!*v) {
2370                 printk(KERN_INFO PFX "Domaining disabled.\n");
2371                 goto out;
2372         }
2373         ldom_domaining_enabled = 1;
2374         err = 0;
2375
2376 out:
2377         mdesc_release(hp);
2378         return err;
2379 }
2380
2381 core_initcall(ldc_init);