xref: /openbmc/linux/arch/sparc/kernel/ldc.c (revision b6bec26c)
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 	struct hlist_node *n;
957 
958 	hlist_for_each_entry(lp, n, &ldc_channel_list, list) {
959 		if (lp->id == id)
960 			return 1;
961 	}
962 	return 0;
963 }
964 
965 static int alloc_queue(const char *name, unsigned long num_entries,
966 		       struct ldc_packet **base, unsigned long *ra)
967 {
968 	unsigned long size, order;
969 	void *q;
970 
971 	size = num_entries * LDC_PACKET_SIZE;
972 	order = get_order(size);
973 
974 	q = (void *) __get_free_pages(GFP_KERNEL, order);
975 	if (!q) {
976 		printk(KERN_ERR PFX "Alloc of %s queue failed with "
977 		       "size=%lu order=%lu\n", name, size, order);
978 		return -ENOMEM;
979 	}
980 
981 	memset(q, 0, PAGE_SIZE << order);
982 
983 	*base = q;
984 	*ra = __pa(q);
985 
986 	return 0;
987 }
988 
989 static void free_queue(unsigned long num_entries, struct ldc_packet *q)
990 {
991 	unsigned long size, order;
992 
993 	if (!q)
994 		return;
995 
996 	size = num_entries * LDC_PACKET_SIZE;
997 	order = get_order(size);
998 
999 	free_pages((unsigned long)q, order);
1000 }
1001 
1002 /* XXX Make this configurable... XXX */
1003 #define LDC_IOTABLE_SIZE	(8 * 1024)
1004 
1005 static int ldc_iommu_init(struct ldc_channel *lp)
1006 {
1007 	unsigned long sz, num_tsb_entries, tsbsize, order;
1008 	struct ldc_iommu *iommu = &lp->iommu;
1009 	struct ldc_mtable_entry *table;
1010 	unsigned long hv_err;
1011 	int err;
1012 
1013 	num_tsb_entries = LDC_IOTABLE_SIZE;
1014 	tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1015 
1016 	spin_lock_init(&iommu->lock);
1017 
1018 	sz = num_tsb_entries / 8;
1019 	sz = (sz + 7UL) & ~7UL;
1020 	iommu->arena.map = kzalloc(sz, GFP_KERNEL);
1021 	if (!iommu->arena.map) {
1022 		printk(KERN_ERR PFX "Alloc of arena map failed, sz=%lu\n", sz);
1023 		return -ENOMEM;
1024 	}
1025 
1026 	iommu->arena.limit = num_tsb_entries;
1027 
1028 	order = get_order(tsbsize);
1029 
1030 	table = (struct ldc_mtable_entry *)
1031 		__get_free_pages(GFP_KERNEL, order);
1032 	err = -ENOMEM;
1033 	if (!table) {
1034 		printk(KERN_ERR PFX "Alloc of MTE table failed, "
1035 		       "size=%lu order=%lu\n", tsbsize, order);
1036 		goto out_free_map;
1037 	}
1038 
1039 	memset(table, 0, PAGE_SIZE << order);
1040 
1041 	iommu->page_table = table;
1042 
1043 	hv_err = sun4v_ldc_set_map_table(lp->id, __pa(table),
1044 					 num_tsb_entries);
1045 	err = -EINVAL;
1046 	if (hv_err)
1047 		goto out_free_table;
1048 
1049 	return 0;
1050 
1051 out_free_table:
1052 	free_pages((unsigned long) table, order);
1053 	iommu->page_table = NULL;
1054 
1055 out_free_map:
1056 	kfree(iommu->arena.map);
1057 	iommu->arena.map = NULL;
1058 
1059 	return err;
1060 }
1061 
1062 static void ldc_iommu_release(struct ldc_channel *lp)
1063 {
1064 	struct ldc_iommu *iommu = &lp->iommu;
1065 	unsigned long num_tsb_entries, tsbsize, order;
1066 
1067 	(void) sun4v_ldc_set_map_table(lp->id, 0, 0);
1068 
1069 	num_tsb_entries = iommu->arena.limit;
1070 	tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1071 	order = get_order(tsbsize);
1072 
1073 	free_pages((unsigned long) iommu->page_table, order);
1074 	iommu->page_table = NULL;
1075 
1076 	kfree(iommu->arena.map);
1077 	iommu->arena.map = NULL;
1078 }
1079 
1080 struct ldc_channel *ldc_alloc(unsigned long id,
1081 			      const struct ldc_channel_config *cfgp,
1082 			      void *event_arg)
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 
1098 	switch (cfgp->mode) {
1099 	case LDC_MODE_RAW:
1100 		mops = &raw_ops;
1101 		mss = LDC_PACKET_SIZE;
1102 		break;
1103 
1104 	case LDC_MODE_UNRELIABLE:
1105 		mops = &nonraw_ops;
1106 		mss = LDC_PACKET_SIZE - 8;
1107 		break;
1108 
1109 	case LDC_MODE_STREAM:
1110 		mops = &stream_ops;
1111 		mss = LDC_PACKET_SIZE - 8 - 8;
1112 		break;
1113 
1114 	default:
1115 		goto out_err;
1116 	}
1117 
1118 	if (!cfgp->event || !event_arg || !cfgp->rx_irq || !cfgp->tx_irq)
1119 		goto out_err;
1120 
1121 	hv_err = sun4v_ldc_tx_qinfo(id, &dummy1, &dummy2);
1122 	err = -ENODEV;
1123 	if (hv_err == HV_ECHANNEL)
1124 		goto out_err;
1125 
1126 	err = -EEXIST;
1127 	if (__ldc_channel_exists(id))
1128 		goto out_err;
1129 
1130 	mssbuf = NULL;
1131 
1132 	lp = kzalloc(sizeof(*lp), GFP_KERNEL);
1133 	err = -ENOMEM;
1134 	if (!lp)
1135 		goto out_err;
1136 
1137 	spin_lock_init(&lp->lock);
1138 
1139 	lp->id = id;
1140 
1141 	err = ldc_iommu_init(lp);
1142 	if (err)
1143 		goto out_free_ldc;
1144 
1145 	lp->mops = mops;
1146 	lp->mss = mss;
1147 
1148 	lp->cfg = *cfgp;
1149 	if (!lp->cfg.mtu)
1150 		lp->cfg.mtu = LDC_DEFAULT_MTU;
1151 
1152 	if (lp->cfg.mode == LDC_MODE_STREAM) {
1153 		mssbuf = kzalloc(lp->cfg.mtu, GFP_KERNEL);
1154 		if (!mssbuf) {
1155 			err = -ENOMEM;
1156 			goto out_free_iommu;
1157 		}
1158 		lp->mssbuf = mssbuf;
1159 	}
1160 
1161 	lp->event_arg = event_arg;
1162 
1163 	/* XXX allow setting via ldc_channel_config to override defaults
1164 	 * XXX or use some formula based upon mtu
1165 	 */
1166 	lp->tx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1167 	lp->rx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1168 
1169 	err = alloc_queue("TX", lp->tx_num_entries,
1170 			  &lp->tx_base, &lp->tx_ra);
1171 	if (err)
1172 		goto out_free_mssbuf;
1173 
1174 	err = alloc_queue("RX", lp->rx_num_entries,
1175 			  &lp->rx_base, &lp->rx_ra);
1176 	if (err)
1177 		goto out_free_txq;
1178 
1179 	lp->flags |= LDC_FLAG_ALLOCED_QUEUES;
1180 
1181 	lp->hs_state = LDC_HS_CLOSED;
1182 	ldc_set_state(lp, LDC_STATE_INIT);
1183 
1184 	INIT_HLIST_NODE(&lp->list);
1185 	hlist_add_head(&lp->list, &ldc_channel_list);
1186 
1187 	INIT_HLIST_HEAD(&lp->mh_list);
1188 
1189 	return lp;
1190 
1191 out_free_txq:
1192 	free_queue(lp->tx_num_entries, lp->tx_base);
1193 
1194 out_free_mssbuf:
1195 	kfree(mssbuf);
1196 
1197 out_free_iommu:
1198 	ldc_iommu_release(lp);
1199 
1200 out_free_ldc:
1201 	kfree(lp);
1202 
1203 out_err:
1204 	return ERR_PTR(err);
1205 }
1206 EXPORT_SYMBOL(ldc_alloc);
1207 
1208 void ldc_free(struct ldc_channel *lp)
1209 {
1210 	if (lp->flags & LDC_FLAG_REGISTERED_IRQS) {
1211 		free_irq(lp->cfg.rx_irq, lp);
1212 		free_irq(lp->cfg.tx_irq, lp);
1213 	}
1214 
1215 	if (lp->flags & LDC_FLAG_REGISTERED_QUEUES) {
1216 		sun4v_ldc_tx_qconf(lp->id, 0, 0);
1217 		sun4v_ldc_rx_qconf(lp->id, 0, 0);
1218 		lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1219 	}
1220 	if (lp->flags & LDC_FLAG_ALLOCED_QUEUES) {
1221 		free_queue(lp->tx_num_entries, lp->tx_base);
1222 		free_queue(lp->rx_num_entries, lp->rx_base);
1223 		lp->flags &= ~LDC_FLAG_ALLOCED_QUEUES;
1224 	}
1225 
1226 	hlist_del(&lp->list);
1227 
1228 	kfree(lp->mssbuf);
1229 
1230 	ldc_iommu_release(lp);
1231 
1232 	kfree(lp);
1233 }
1234 EXPORT_SYMBOL(ldc_free);
1235 
1236 /* Bind the channel.  This registers the LDC queues with
1237  * the hypervisor and puts the channel into a pseudo-listening
1238  * state.  This does not initiate a handshake, ldc_connect() does
1239  * that.
1240  */
1241 int ldc_bind(struct ldc_channel *lp, const char *name)
1242 {
1243 	unsigned long hv_err, flags;
1244 	int err = -EINVAL;
1245 
1246 	if (!name ||
1247 	    (lp->state != LDC_STATE_INIT))
1248 		return -EINVAL;
1249 
1250 	snprintf(lp->rx_irq_name, LDC_IRQ_NAME_MAX, "%s RX", name);
1251 	snprintf(lp->tx_irq_name, LDC_IRQ_NAME_MAX, "%s TX", name);
1252 
1253 	err = request_irq(lp->cfg.rx_irq, ldc_rx, IRQF_DISABLED,
1254 			  lp->rx_irq_name, lp);
1255 	if (err)
1256 		return err;
1257 
1258 	err = request_irq(lp->cfg.tx_irq, ldc_tx, IRQF_DISABLED,
1259 			  lp->tx_irq_name, lp);
1260 	if (err) {
1261 		free_irq(lp->cfg.rx_irq, lp);
1262 		return err;
1263 	}
1264 
1265 
1266 	spin_lock_irqsave(&lp->lock, flags);
1267 
1268 	enable_irq(lp->cfg.rx_irq);
1269 	enable_irq(lp->cfg.tx_irq);
1270 
1271 	lp->flags |= LDC_FLAG_REGISTERED_IRQS;
1272 
1273 	err = -ENODEV;
1274 	hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1275 	if (hv_err)
1276 		goto out_free_irqs;
1277 
1278 	hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1279 	if (hv_err)
1280 		goto out_free_irqs;
1281 
1282 	hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1283 	if (hv_err)
1284 		goto out_unmap_tx;
1285 
1286 	hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1287 	if (hv_err)
1288 		goto out_unmap_tx;
1289 
1290 	lp->flags |= LDC_FLAG_REGISTERED_QUEUES;
1291 
1292 	hv_err = sun4v_ldc_tx_get_state(lp->id,
1293 					&lp->tx_head,
1294 					&lp->tx_tail,
1295 					&lp->chan_state);
1296 	err = -EBUSY;
1297 	if (hv_err)
1298 		goto out_unmap_rx;
1299 
1300 	lp->tx_acked = lp->tx_head;
1301 
1302 	lp->hs_state = LDC_HS_OPEN;
1303 	ldc_set_state(lp, LDC_STATE_BOUND);
1304 
1305 	spin_unlock_irqrestore(&lp->lock, flags);
1306 
1307 	return 0;
1308 
1309 out_unmap_rx:
1310 	lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1311 	sun4v_ldc_rx_qconf(lp->id, 0, 0);
1312 
1313 out_unmap_tx:
1314 	sun4v_ldc_tx_qconf(lp->id, 0, 0);
1315 
1316 out_free_irqs:
1317 	lp->flags &= ~LDC_FLAG_REGISTERED_IRQS;
1318 	free_irq(lp->cfg.tx_irq, lp);
1319 	free_irq(lp->cfg.rx_irq, lp);
1320 
1321 	spin_unlock_irqrestore(&lp->lock, flags);
1322 
1323 	return err;
1324 }
1325 EXPORT_SYMBOL(ldc_bind);
1326 
1327 int ldc_connect(struct ldc_channel *lp)
1328 {
1329 	unsigned long flags;
1330 	int err;
1331 
1332 	if (lp->cfg.mode == LDC_MODE_RAW)
1333 		return -EINVAL;
1334 
1335 	spin_lock_irqsave(&lp->lock, flags);
1336 
1337 	if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1338 	    !(lp->flags & LDC_FLAG_REGISTERED_QUEUES) ||
1339 	    lp->hs_state != LDC_HS_OPEN)
1340 		err = -EINVAL;
1341 	else
1342 		err = start_handshake(lp);
1343 
1344 	spin_unlock_irqrestore(&lp->lock, flags);
1345 
1346 	return err;
1347 }
1348 EXPORT_SYMBOL(ldc_connect);
1349 
1350 int ldc_disconnect(struct ldc_channel *lp)
1351 {
1352 	unsigned long hv_err, flags;
1353 	int err;
1354 
1355 	if (lp->cfg.mode == LDC_MODE_RAW)
1356 		return -EINVAL;
1357 
1358 	if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1359 	    !(lp->flags & LDC_FLAG_REGISTERED_QUEUES))
1360 		return -EINVAL;
1361 
1362 	spin_lock_irqsave(&lp->lock, flags);
1363 
1364 	err = -ENODEV;
1365 	hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1366 	if (hv_err)
1367 		goto out_err;
1368 
1369 	hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1370 	if (hv_err)
1371 		goto out_err;
1372 
1373 	hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1374 	if (hv_err)
1375 		goto out_err;
1376 
1377 	hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1378 	if (hv_err)
1379 		goto out_err;
1380 
1381 	ldc_set_state(lp, LDC_STATE_BOUND);
1382 	lp->hs_state = LDC_HS_OPEN;
1383 	lp->flags |= LDC_FLAG_RESET;
1384 
1385 	spin_unlock_irqrestore(&lp->lock, flags);
1386 
1387 	return 0;
1388 
1389 out_err:
1390 	sun4v_ldc_tx_qconf(lp->id, 0, 0);
1391 	sun4v_ldc_rx_qconf(lp->id, 0, 0);
1392 	free_irq(lp->cfg.tx_irq, lp);
1393 	free_irq(lp->cfg.rx_irq, lp);
1394 	lp->flags &= ~(LDC_FLAG_REGISTERED_IRQS |
1395 		       LDC_FLAG_REGISTERED_QUEUES);
1396 	ldc_set_state(lp, LDC_STATE_INIT);
1397 
1398 	spin_unlock_irqrestore(&lp->lock, flags);
1399 
1400 	return err;
1401 }
1402 EXPORT_SYMBOL(ldc_disconnect);
1403 
1404 int ldc_state(struct ldc_channel *lp)
1405 {
1406 	return lp->state;
1407 }
1408 EXPORT_SYMBOL(ldc_state);
1409 
1410 static int write_raw(struct ldc_channel *lp, const void *buf, unsigned int size)
1411 {
1412 	struct ldc_packet *p;
1413 	unsigned long new_tail;
1414 	int err;
1415 
1416 	if (size > LDC_PACKET_SIZE)
1417 		return -EMSGSIZE;
1418 
1419 	p = data_get_tx_packet(lp, &new_tail);
1420 	if (!p)
1421 		return -EAGAIN;
1422 
1423 	memcpy(p, buf, size);
1424 
1425 	err = send_tx_packet(lp, p, new_tail);
1426 	if (!err)
1427 		err = size;
1428 
1429 	return err;
1430 }
1431 
1432 static int read_raw(struct ldc_channel *lp, void *buf, unsigned int size)
1433 {
1434 	struct ldc_packet *p;
1435 	unsigned long hv_err, new;
1436 	int err;
1437 
1438 	if (size < LDC_PACKET_SIZE)
1439 		return -EINVAL;
1440 
1441 	hv_err = sun4v_ldc_rx_get_state(lp->id,
1442 					&lp->rx_head,
1443 					&lp->rx_tail,
1444 					&lp->chan_state);
1445 	if (hv_err)
1446 		return ldc_abort(lp);
1447 
1448 	if (lp->chan_state == LDC_CHANNEL_DOWN ||
1449 	    lp->chan_state == LDC_CHANNEL_RESETTING)
1450 		return -ECONNRESET;
1451 
1452 	if (lp->rx_head == lp->rx_tail)
1453 		return 0;
1454 
1455 	p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
1456 	memcpy(buf, p, LDC_PACKET_SIZE);
1457 
1458 	new = rx_advance(lp, lp->rx_head);
1459 	lp->rx_head = new;
1460 
1461 	err = __set_rx_head(lp, new);
1462 	if (err < 0)
1463 		err = -ECONNRESET;
1464 	else
1465 		err = LDC_PACKET_SIZE;
1466 
1467 	return err;
1468 }
1469 
1470 static const struct ldc_mode_ops raw_ops = {
1471 	.write		=	write_raw,
1472 	.read		=	read_raw,
1473 };
1474 
1475 static int write_nonraw(struct ldc_channel *lp, const void *buf,
1476 			unsigned int size)
1477 {
1478 	unsigned long hv_err, tail;
1479 	unsigned int copied;
1480 	u32 seq;
1481 	int err;
1482 
1483 	hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail,
1484 					&lp->chan_state);
1485 	if (unlikely(hv_err))
1486 		return -EBUSY;
1487 
1488 	if (unlikely(lp->chan_state != LDC_CHANNEL_UP))
1489 		return ldc_abort(lp);
1490 
1491 	if (!tx_has_space_for(lp, size))
1492 		return -EAGAIN;
1493 
1494 	seq = lp->snd_nxt;
1495 	copied = 0;
1496 	tail = lp->tx_tail;
1497 	while (copied < size) {
1498 		struct ldc_packet *p = lp->tx_base + (tail / LDC_PACKET_SIZE);
1499 		u8 *data = ((lp->cfg.mode == LDC_MODE_UNRELIABLE) ?
1500 			    p->u.u_data :
1501 			    p->u.r.r_data);
1502 		int data_len;
1503 
1504 		p->type = LDC_DATA;
1505 		p->stype = LDC_INFO;
1506 		p->ctrl = 0;
1507 
1508 		data_len = size - copied;
1509 		if (data_len > lp->mss)
1510 			data_len = lp->mss;
1511 
1512 		BUG_ON(data_len > LDC_LEN);
1513 
1514 		p->env = (data_len |
1515 			  (copied == 0 ? LDC_START : 0) |
1516 			  (data_len == size - copied ? LDC_STOP : 0));
1517 
1518 		p->seqid = ++seq;
1519 
1520 		ldcdbg(DATA, "SENT DATA [%02x:%02x:%02x:%02x:%08x]\n",
1521 		       p->type,
1522 		       p->stype,
1523 		       p->ctrl,
1524 		       p->env,
1525 		       p->seqid);
1526 
1527 		memcpy(data, buf, data_len);
1528 		buf += data_len;
1529 		copied += data_len;
1530 
1531 		tail = tx_advance(lp, tail);
1532 	}
1533 
1534 	err = set_tx_tail(lp, tail);
1535 	if (!err) {
1536 		lp->snd_nxt = seq;
1537 		err = size;
1538 	}
1539 
1540 	return err;
1541 }
1542 
1543 static int rx_bad_seq(struct ldc_channel *lp, struct ldc_packet *p,
1544 		      struct ldc_packet *first_frag)
1545 {
1546 	int err;
1547 
1548 	if (first_frag)
1549 		lp->rcv_nxt = first_frag->seqid - 1;
1550 
1551 	err = send_data_nack(lp, p);
1552 	if (err)
1553 		return err;
1554 
1555 	err = __set_rx_head(lp, lp->rx_tail);
1556 	if (err < 0)
1557 		return ldc_abort(lp);
1558 
1559 	return 0;
1560 }
1561 
1562 static int data_ack_nack(struct ldc_channel *lp, struct ldc_packet *p)
1563 {
1564 	if (p->stype & LDC_ACK) {
1565 		int err = process_data_ack(lp, p);
1566 		if (err)
1567 			return err;
1568 	}
1569 	if (p->stype & LDC_NACK)
1570 		return ldc_abort(lp);
1571 
1572 	return 0;
1573 }
1574 
1575 static int rx_data_wait(struct ldc_channel *lp, unsigned long cur_head)
1576 {
1577 	unsigned long dummy;
1578 	int limit = 1000;
1579 
1580 	ldcdbg(DATA, "DATA WAIT cur_head[%lx] rx_head[%lx] rx_tail[%lx]\n",
1581 	       cur_head, lp->rx_head, lp->rx_tail);
1582 	while (limit-- > 0) {
1583 		unsigned long hv_err;
1584 
1585 		hv_err = sun4v_ldc_rx_get_state(lp->id,
1586 						&dummy,
1587 						&lp->rx_tail,
1588 						&lp->chan_state);
1589 		if (hv_err)
1590 			return ldc_abort(lp);
1591 
1592 		if (lp->chan_state == LDC_CHANNEL_DOWN ||
1593 		    lp->chan_state == LDC_CHANNEL_RESETTING)
1594 			return -ECONNRESET;
1595 
1596 		if (cur_head != lp->rx_tail) {
1597 			ldcdbg(DATA, "DATA WAIT DONE "
1598 			       "head[%lx] tail[%lx] chan_state[%lx]\n",
1599 			       dummy, lp->rx_tail, lp->chan_state);
1600 			return 0;
1601 		}
1602 
1603 		udelay(1);
1604 	}
1605 	return -EAGAIN;
1606 }
1607 
1608 static int rx_set_head(struct ldc_channel *lp, unsigned long head)
1609 {
1610 	int err = __set_rx_head(lp, head);
1611 
1612 	if (err < 0)
1613 		return ldc_abort(lp);
1614 
1615 	lp->rx_head = head;
1616 	return 0;
1617 }
1618 
1619 static void send_data_ack(struct ldc_channel *lp)
1620 {
1621 	unsigned long new_tail;
1622 	struct ldc_packet *p;
1623 
1624 	p = data_get_tx_packet(lp, &new_tail);
1625 	if (likely(p)) {
1626 		int err;
1627 
1628 		memset(p, 0, sizeof(*p));
1629 		p->type = LDC_DATA;
1630 		p->stype = LDC_ACK;
1631 		p->ctrl = 0;
1632 		p->seqid = lp->snd_nxt + 1;
1633 		p->u.r.ackid = lp->rcv_nxt;
1634 
1635 		err = send_tx_packet(lp, p, new_tail);
1636 		if (!err)
1637 			lp->snd_nxt++;
1638 	}
1639 }
1640 
1641 static int read_nonraw(struct ldc_channel *lp, void *buf, unsigned int size)
1642 {
1643 	struct ldc_packet *first_frag;
1644 	unsigned long hv_err, new;
1645 	int err, copied;
1646 
1647 	hv_err = sun4v_ldc_rx_get_state(lp->id,
1648 					&lp->rx_head,
1649 					&lp->rx_tail,
1650 					&lp->chan_state);
1651 	if (hv_err)
1652 		return ldc_abort(lp);
1653 
1654 	if (lp->chan_state == LDC_CHANNEL_DOWN ||
1655 	    lp->chan_state == LDC_CHANNEL_RESETTING)
1656 		return -ECONNRESET;
1657 
1658 	if (lp->rx_head == lp->rx_tail)
1659 		return 0;
1660 
1661 	first_frag = NULL;
1662 	copied = err = 0;
1663 	new = lp->rx_head;
1664 	while (1) {
1665 		struct ldc_packet *p;
1666 		int pkt_len;
1667 
1668 		BUG_ON(new == lp->rx_tail);
1669 		p = lp->rx_base + (new / LDC_PACKET_SIZE);
1670 
1671 		ldcdbg(RX, "RX read pkt[%02x:%02x:%02x:%02x:%08x:%08x] "
1672 		       "rcv_nxt[%08x]\n",
1673 		       p->type,
1674 		       p->stype,
1675 		       p->ctrl,
1676 		       p->env,
1677 		       p->seqid,
1678 		       p->u.r.ackid,
1679 		       lp->rcv_nxt);
1680 
1681 		if (unlikely(!rx_seq_ok(lp, p->seqid))) {
1682 			err = rx_bad_seq(lp, p, first_frag);
1683 			copied = 0;
1684 			break;
1685 		}
1686 
1687 		if (p->type & LDC_CTRL) {
1688 			err = process_control_frame(lp, p);
1689 			if (err < 0)
1690 				break;
1691 			err = 0;
1692 		}
1693 
1694 		lp->rcv_nxt = p->seqid;
1695 
1696 		if (!(p->type & LDC_DATA)) {
1697 			new = rx_advance(lp, new);
1698 			goto no_data;
1699 		}
1700 		if (p->stype & (LDC_ACK | LDC_NACK)) {
1701 			err = data_ack_nack(lp, p);
1702 			if (err)
1703 				break;
1704 		}
1705 		if (!(p->stype & LDC_INFO)) {
1706 			new = rx_advance(lp, new);
1707 			err = rx_set_head(lp, new);
1708 			if (err)
1709 				break;
1710 			goto no_data;
1711 		}
1712 
1713 		pkt_len = p->env & LDC_LEN;
1714 
1715 		/* Every initial packet starts with the START bit set.
1716 		 *
1717 		 * Singleton packets will have both START+STOP set.
1718 		 *
1719 		 * Fragments will have START set in the first frame, STOP
1720 		 * set in the last frame, and neither bit set in middle
1721 		 * frames of the packet.
1722 		 *
1723 		 * Therefore if we are at the beginning of a packet and
1724 		 * we don't see START, or we are in the middle of a fragmented
1725 		 * packet and do see START, we are unsynchronized and should
1726 		 * flush the RX queue.
1727 		 */
1728 		if ((first_frag == NULL && !(p->env & LDC_START)) ||
1729 		    (first_frag != NULL &&  (p->env & LDC_START))) {
1730 			if (!first_frag)
1731 				new = rx_advance(lp, new);
1732 
1733 			err = rx_set_head(lp, new);
1734 			if (err)
1735 				break;
1736 
1737 			if (!first_frag)
1738 				goto no_data;
1739 		}
1740 		if (!first_frag)
1741 			first_frag = p;
1742 
1743 		if (pkt_len > size - copied) {
1744 			/* User didn't give us a big enough buffer,
1745 			 * what to do?  This is a pretty serious error.
1746 			 *
1747 			 * Since we haven't updated the RX ring head to
1748 			 * consume any of the packets, signal the error
1749 			 * to the user and just leave the RX ring alone.
1750 			 *
1751 			 * This seems the best behavior because this allows
1752 			 * a user of the LDC layer to start with a small
1753 			 * RX buffer for ldc_read() calls and use -EMSGSIZE
1754 			 * as a cue to enlarge it's read buffer.
1755 			 */
1756 			err = -EMSGSIZE;
1757 			break;
1758 		}
1759 
1760 		/* Ok, we are gonna eat this one.  */
1761 		new = rx_advance(lp, new);
1762 
1763 		memcpy(buf,
1764 		       (lp->cfg.mode == LDC_MODE_UNRELIABLE ?
1765 			p->u.u_data : p->u.r.r_data), pkt_len);
1766 		buf += pkt_len;
1767 		copied += pkt_len;
1768 
1769 		if (p->env & LDC_STOP)
1770 			break;
1771 
1772 no_data:
1773 		if (new == lp->rx_tail) {
1774 			err = rx_data_wait(lp, new);
1775 			if (err)
1776 				break;
1777 		}
1778 	}
1779 
1780 	if (!err)
1781 		err = rx_set_head(lp, new);
1782 
1783 	if (err && first_frag)
1784 		lp->rcv_nxt = first_frag->seqid - 1;
1785 
1786 	if (!err) {
1787 		err = copied;
1788 		if (err > 0 && lp->cfg.mode != LDC_MODE_UNRELIABLE)
1789 			send_data_ack(lp);
1790 	}
1791 
1792 	return err;
1793 }
1794 
1795 static const struct ldc_mode_ops nonraw_ops = {
1796 	.write		=	write_nonraw,
1797 	.read		=	read_nonraw,
1798 };
1799 
1800 static int write_stream(struct ldc_channel *lp, const void *buf,
1801 			unsigned int size)
1802 {
1803 	if (size > lp->cfg.mtu)
1804 		size = lp->cfg.mtu;
1805 	return write_nonraw(lp, buf, size);
1806 }
1807 
1808 static int read_stream(struct ldc_channel *lp, void *buf, unsigned int size)
1809 {
1810 	if (!lp->mssbuf_len) {
1811 		int err = read_nonraw(lp, lp->mssbuf, lp->cfg.mtu);
1812 		if (err < 0)
1813 			return err;
1814 
1815 		lp->mssbuf_len = err;
1816 		lp->mssbuf_off = 0;
1817 	}
1818 
1819 	if (size > lp->mssbuf_len)
1820 		size = lp->mssbuf_len;
1821 	memcpy(buf, lp->mssbuf + lp->mssbuf_off, size);
1822 
1823 	lp->mssbuf_off += size;
1824 	lp->mssbuf_len -= size;
1825 
1826 	return size;
1827 }
1828 
1829 static const struct ldc_mode_ops stream_ops = {
1830 	.write		=	write_stream,
1831 	.read		=	read_stream,
1832 };
1833 
1834 int ldc_write(struct ldc_channel *lp, const void *buf, unsigned int size)
1835 {
1836 	unsigned long flags;
1837 	int err;
1838 
1839 	if (!buf)
1840 		return -EINVAL;
1841 
1842 	if (!size)
1843 		return 0;
1844 
1845 	spin_lock_irqsave(&lp->lock, flags);
1846 
1847 	if (lp->hs_state != LDC_HS_COMPLETE)
1848 		err = -ENOTCONN;
1849 	else
1850 		err = lp->mops->write(lp, buf, size);
1851 
1852 	spin_unlock_irqrestore(&lp->lock, flags);
1853 
1854 	return err;
1855 }
1856 EXPORT_SYMBOL(ldc_write);
1857 
1858 int ldc_read(struct ldc_channel *lp, void *buf, unsigned int size)
1859 {
1860 	unsigned long flags;
1861 	int err;
1862 
1863 	if (!buf)
1864 		return -EINVAL;
1865 
1866 	if (!size)
1867 		return 0;
1868 
1869 	spin_lock_irqsave(&lp->lock, flags);
1870 
1871 	if (lp->hs_state != LDC_HS_COMPLETE)
1872 		err = -ENOTCONN;
1873 	else
1874 		err = lp->mops->read(lp, buf, size);
1875 
1876 	spin_unlock_irqrestore(&lp->lock, flags);
1877 
1878 	return err;
1879 }
1880 EXPORT_SYMBOL(ldc_read);
1881 
1882 static long arena_alloc(struct ldc_iommu *iommu, unsigned long npages)
1883 {
1884 	struct iommu_arena *arena = &iommu->arena;
1885 	unsigned long n, start, end, limit;
1886 	int pass;
1887 
1888 	limit = arena->limit;
1889 	start = arena->hint;
1890 	pass = 0;
1891 
1892 again:
1893 	n = bitmap_find_next_zero_area(arena->map, limit, start, npages, 0);
1894 	end = n + npages;
1895 	if (unlikely(end >= limit)) {
1896 		if (likely(pass < 1)) {
1897 			limit = start;
1898 			start = 0;
1899 			pass++;
1900 			goto again;
1901 		} else {
1902 			/* Scanned the whole thing, give up. */
1903 			return -1;
1904 		}
1905 	}
1906 	bitmap_set(arena->map, n, npages);
1907 
1908 	arena->hint = end;
1909 
1910 	return n;
1911 }
1912 
1913 #define COOKIE_PGSZ_CODE	0xf000000000000000ULL
1914 #define COOKIE_PGSZ_CODE_SHIFT	60ULL
1915 
1916 static u64 pagesize_code(void)
1917 {
1918 	switch (PAGE_SIZE) {
1919 	default:
1920 	case (8ULL * 1024ULL):
1921 		return 0;
1922 	case (64ULL * 1024ULL):
1923 		return 1;
1924 	case (512ULL * 1024ULL):
1925 		return 2;
1926 	case (4ULL * 1024ULL * 1024ULL):
1927 		return 3;
1928 	case (32ULL * 1024ULL * 1024ULL):
1929 		return 4;
1930 	case (256ULL * 1024ULL * 1024ULL):
1931 		return 5;
1932 	}
1933 }
1934 
1935 static u64 make_cookie(u64 index, u64 pgsz_code, u64 page_offset)
1936 {
1937 	return ((pgsz_code << COOKIE_PGSZ_CODE_SHIFT) |
1938 		(index << PAGE_SHIFT) |
1939 		page_offset);
1940 }
1941 
1942 static u64 cookie_to_index(u64 cookie, unsigned long *shift)
1943 {
1944 	u64 szcode = cookie >> COOKIE_PGSZ_CODE_SHIFT;
1945 
1946 	cookie &= ~COOKIE_PGSZ_CODE;
1947 
1948 	*shift = szcode * 3;
1949 
1950 	return (cookie >> (13ULL + (szcode * 3ULL)));
1951 }
1952 
1953 static struct ldc_mtable_entry *alloc_npages(struct ldc_iommu *iommu,
1954 					     unsigned long npages)
1955 {
1956 	long entry;
1957 
1958 	entry = arena_alloc(iommu, npages);
1959 	if (unlikely(entry < 0))
1960 		return NULL;
1961 
1962 	return iommu->page_table + entry;
1963 }
1964 
1965 static u64 perm_to_mte(unsigned int map_perm)
1966 {
1967 	u64 mte_base;
1968 
1969 	mte_base = pagesize_code();
1970 
1971 	if (map_perm & LDC_MAP_SHADOW) {
1972 		if (map_perm & LDC_MAP_R)
1973 			mte_base |= LDC_MTE_COPY_R;
1974 		if (map_perm & LDC_MAP_W)
1975 			mte_base |= LDC_MTE_COPY_W;
1976 	}
1977 	if (map_perm & LDC_MAP_DIRECT) {
1978 		if (map_perm & LDC_MAP_R)
1979 			mte_base |= LDC_MTE_READ;
1980 		if (map_perm & LDC_MAP_W)
1981 			mte_base |= LDC_MTE_WRITE;
1982 		if (map_perm & LDC_MAP_X)
1983 			mte_base |= LDC_MTE_EXEC;
1984 	}
1985 	if (map_perm & LDC_MAP_IO) {
1986 		if (map_perm & LDC_MAP_R)
1987 			mte_base |= LDC_MTE_IOMMU_R;
1988 		if (map_perm & LDC_MAP_W)
1989 			mte_base |= LDC_MTE_IOMMU_W;
1990 	}
1991 
1992 	return mte_base;
1993 }
1994 
1995 static int pages_in_region(unsigned long base, long len)
1996 {
1997 	int count = 0;
1998 
1999 	do {
2000 		unsigned long new = (base + PAGE_SIZE) & PAGE_MASK;
2001 
2002 		len -= (new - base);
2003 		base = new;
2004 		count++;
2005 	} while (len > 0);
2006 
2007 	return count;
2008 }
2009 
2010 struct cookie_state {
2011 	struct ldc_mtable_entry		*page_table;
2012 	struct ldc_trans_cookie		*cookies;
2013 	u64				mte_base;
2014 	u64				prev_cookie;
2015 	u32				pte_idx;
2016 	u32				nc;
2017 };
2018 
2019 static void fill_cookies(struct cookie_state *sp, unsigned long pa,
2020 			 unsigned long off, unsigned long len)
2021 {
2022 	do {
2023 		unsigned long tlen, new = pa + PAGE_SIZE;
2024 		u64 this_cookie;
2025 
2026 		sp->page_table[sp->pte_idx].mte = sp->mte_base | pa;
2027 
2028 		tlen = PAGE_SIZE;
2029 		if (off)
2030 			tlen = PAGE_SIZE - off;
2031 		if (tlen > len)
2032 			tlen = len;
2033 
2034 		this_cookie = make_cookie(sp->pte_idx,
2035 					  pagesize_code(), off);
2036 
2037 		off = 0;
2038 
2039 		if (this_cookie == sp->prev_cookie) {
2040 			sp->cookies[sp->nc - 1].cookie_size += tlen;
2041 		} else {
2042 			sp->cookies[sp->nc].cookie_addr = this_cookie;
2043 			sp->cookies[sp->nc].cookie_size = tlen;
2044 			sp->nc++;
2045 		}
2046 		sp->prev_cookie = this_cookie + tlen;
2047 
2048 		sp->pte_idx++;
2049 
2050 		len -= tlen;
2051 		pa = new;
2052 	} while (len > 0);
2053 }
2054 
2055 static int sg_count_one(struct scatterlist *sg)
2056 {
2057 	unsigned long base = page_to_pfn(sg_page(sg)) << PAGE_SHIFT;
2058 	long len = sg->length;
2059 
2060 	if ((sg->offset | len) & (8UL - 1))
2061 		return -EFAULT;
2062 
2063 	return pages_in_region(base + sg->offset, len);
2064 }
2065 
2066 static int sg_count_pages(struct scatterlist *sg, int num_sg)
2067 {
2068 	int count;
2069 	int i;
2070 
2071 	count = 0;
2072 	for (i = 0; i < num_sg; i++) {
2073 		int err = sg_count_one(sg + i);
2074 		if (err < 0)
2075 			return err;
2076 		count += err;
2077 	}
2078 
2079 	return count;
2080 }
2081 
2082 int ldc_map_sg(struct ldc_channel *lp,
2083 	       struct scatterlist *sg, int num_sg,
2084 	       struct ldc_trans_cookie *cookies, int ncookies,
2085 	       unsigned int map_perm)
2086 {
2087 	unsigned long i, npages, flags;
2088 	struct ldc_mtable_entry *base;
2089 	struct cookie_state state;
2090 	struct ldc_iommu *iommu;
2091 	int err;
2092 
2093 	if (map_perm & ~LDC_MAP_ALL)
2094 		return -EINVAL;
2095 
2096 	err = sg_count_pages(sg, num_sg);
2097 	if (err < 0)
2098 		return err;
2099 
2100 	npages = err;
2101 	if (err > ncookies)
2102 		return -EMSGSIZE;
2103 
2104 	iommu = &lp->iommu;
2105 
2106 	spin_lock_irqsave(&iommu->lock, flags);
2107 	base = alloc_npages(iommu, npages);
2108 	spin_unlock_irqrestore(&iommu->lock, flags);
2109 
2110 	if (!base)
2111 		return -ENOMEM;
2112 
2113 	state.page_table = iommu->page_table;
2114 	state.cookies = cookies;
2115 	state.mte_base = perm_to_mte(map_perm);
2116 	state.prev_cookie = ~(u64)0;
2117 	state.pte_idx = (base - iommu->page_table);
2118 	state.nc = 0;
2119 
2120 	for (i = 0; i < num_sg; i++)
2121 		fill_cookies(&state, page_to_pfn(sg_page(&sg[i])) << PAGE_SHIFT,
2122 			     sg[i].offset, sg[i].length);
2123 
2124 	return state.nc;
2125 }
2126 EXPORT_SYMBOL(ldc_map_sg);
2127 
2128 int ldc_map_single(struct ldc_channel *lp,
2129 		   void *buf, unsigned int len,
2130 		   struct ldc_trans_cookie *cookies, int ncookies,
2131 		   unsigned int map_perm)
2132 {
2133 	unsigned long npages, pa, flags;
2134 	struct ldc_mtable_entry *base;
2135 	struct cookie_state state;
2136 	struct ldc_iommu *iommu;
2137 
2138 	if ((map_perm & ~LDC_MAP_ALL) || (ncookies < 1))
2139 		return -EINVAL;
2140 
2141 	pa = __pa(buf);
2142 	if ((pa | len) & (8UL - 1))
2143 		return -EFAULT;
2144 
2145 	npages = pages_in_region(pa, len);
2146 
2147 	iommu = &lp->iommu;
2148 
2149 	spin_lock_irqsave(&iommu->lock, flags);
2150 	base = alloc_npages(iommu, npages);
2151 	spin_unlock_irqrestore(&iommu->lock, flags);
2152 
2153 	if (!base)
2154 		return -ENOMEM;
2155 
2156 	state.page_table = iommu->page_table;
2157 	state.cookies = cookies;
2158 	state.mte_base = perm_to_mte(map_perm);
2159 	state.prev_cookie = ~(u64)0;
2160 	state.pte_idx = (base - iommu->page_table);
2161 	state.nc = 0;
2162 	fill_cookies(&state, (pa & PAGE_MASK), (pa & ~PAGE_MASK), len);
2163 	BUG_ON(state.nc != 1);
2164 
2165 	return state.nc;
2166 }
2167 EXPORT_SYMBOL(ldc_map_single);
2168 
2169 static void free_npages(unsigned long id, struct ldc_iommu *iommu,
2170 			u64 cookie, u64 size)
2171 {
2172 	struct iommu_arena *arena = &iommu->arena;
2173 	unsigned long i, shift, index, npages;
2174 	struct ldc_mtable_entry *base;
2175 
2176 	npages = PAGE_ALIGN(((cookie & ~PAGE_MASK) + size)) >> PAGE_SHIFT;
2177 	index = cookie_to_index(cookie, &shift);
2178 	base = iommu->page_table + index;
2179 
2180 	BUG_ON(index > arena->limit ||
2181 	       (index + npages) > arena->limit);
2182 
2183 	for (i = 0; i < npages; i++) {
2184 		if (base->cookie)
2185 			sun4v_ldc_revoke(id, cookie + (i << shift),
2186 					 base->cookie);
2187 		base->mte = 0;
2188 		__clear_bit(index + i, arena->map);
2189 	}
2190 }
2191 
2192 void ldc_unmap(struct ldc_channel *lp, struct ldc_trans_cookie *cookies,
2193 	       int ncookies)
2194 {
2195 	struct ldc_iommu *iommu = &lp->iommu;
2196 	unsigned long flags;
2197 	int i;
2198 
2199 	spin_lock_irqsave(&iommu->lock, flags);
2200 	for (i = 0; i < ncookies; i++) {
2201 		u64 addr = cookies[i].cookie_addr;
2202 		u64 size = cookies[i].cookie_size;
2203 
2204 		free_npages(lp->id, iommu, addr, size);
2205 	}
2206 	spin_unlock_irqrestore(&iommu->lock, flags);
2207 }
2208 EXPORT_SYMBOL(ldc_unmap);
2209 
2210 int ldc_copy(struct ldc_channel *lp, int copy_dir,
2211 	     void *buf, unsigned int len, unsigned long offset,
2212 	     struct ldc_trans_cookie *cookies, int ncookies)
2213 {
2214 	unsigned int orig_len;
2215 	unsigned long ra;
2216 	int i;
2217 
2218 	if (copy_dir != LDC_COPY_IN && copy_dir != LDC_COPY_OUT) {
2219 		printk(KERN_ERR PFX "ldc_copy: ID[%lu] Bad copy_dir[%d]\n",
2220 		       lp->id, copy_dir);
2221 		return -EINVAL;
2222 	}
2223 
2224 	ra = __pa(buf);
2225 	if ((ra | len | offset) & (8UL - 1)) {
2226 		printk(KERN_ERR PFX "ldc_copy: ID[%lu] Unaligned buffer "
2227 		       "ra[%lx] len[%x] offset[%lx]\n",
2228 		       lp->id, ra, len, offset);
2229 		return -EFAULT;
2230 	}
2231 
2232 	if (lp->hs_state != LDC_HS_COMPLETE ||
2233 	    (lp->flags & LDC_FLAG_RESET)) {
2234 		printk(KERN_ERR PFX "ldc_copy: ID[%lu] Link down hs_state[%x] "
2235 		       "flags[%x]\n", lp->id, lp->hs_state, lp->flags);
2236 		return -ECONNRESET;
2237 	}
2238 
2239 	orig_len = len;
2240 	for (i = 0; i < ncookies; i++) {
2241 		unsigned long cookie_raddr = cookies[i].cookie_addr;
2242 		unsigned long this_len = cookies[i].cookie_size;
2243 		unsigned long actual_len;
2244 
2245 		if (unlikely(offset)) {
2246 			unsigned long this_off = offset;
2247 
2248 			if (this_off > this_len)
2249 				this_off = this_len;
2250 
2251 			offset -= this_off;
2252 			this_len -= this_off;
2253 			if (!this_len)
2254 				continue;
2255 			cookie_raddr += this_off;
2256 		}
2257 
2258 		if (this_len > len)
2259 			this_len = len;
2260 
2261 		while (1) {
2262 			unsigned long hv_err;
2263 
2264 			hv_err = sun4v_ldc_copy(lp->id, copy_dir,
2265 						cookie_raddr, ra,
2266 						this_len, &actual_len);
2267 			if (unlikely(hv_err)) {
2268 				printk(KERN_ERR PFX "ldc_copy: ID[%lu] "
2269 				       "HV error %lu\n",
2270 				       lp->id, hv_err);
2271 				if (lp->hs_state != LDC_HS_COMPLETE ||
2272 				    (lp->flags & LDC_FLAG_RESET))
2273 					return -ECONNRESET;
2274 				else
2275 					return -EFAULT;
2276 			}
2277 
2278 			cookie_raddr += actual_len;
2279 			ra += actual_len;
2280 			len -= actual_len;
2281 			if (actual_len == this_len)
2282 				break;
2283 
2284 			this_len -= actual_len;
2285 		}
2286 
2287 		if (!len)
2288 			break;
2289 	}
2290 
2291 	/* It is caller policy what to do about short copies.
2292 	 * For example, a networking driver can declare the
2293 	 * packet a runt and drop it.
2294 	 */
2295 
2296 	return orig_len - len;
2297 }
2298 EXPORT_SYMBOL(ldc_copy);
2299 
2300 void *ldc_alloc_exp_dring(struct ldc_channel *lp, unsigned int len,
2301 			  struct ldc_trans_cookie *cookies, int *ncookies,
2302 			  unsigned int map_perm)
2303 {
2304 	void *buf;
2305 	int err;
2306 
2307 	if (len & (8UL - 1))
2308 		return ERR_PTR(-EINVAL);
2309 
2310 	buf = kzalloc(len, GFP_KERNEL);
2311 	if (!buf)
2312 		return ERR_PTR(-ENOMEM);
2313 
2314 	err = ldc_map_single(lp, buf, len, cookies, *ncookies, map_perm);
2315 	if (err < 0) {
2316 		kfree(buf);
2317 		return ERR_PTR(err);
2318 	}
2319 	*ncookies = err;
2320 
2321 	return buf;
2322 }
2323 EXPORT_SYMBOL(ldc_alloc_exp_dring);
2324 
2325 void ldc_free_exp_dring(struct ldc_channel *lp, void *buf, unsigned int len,
2326 			struct ldc_trans_cookie *cookies, int ncookies)
2327 {
2328 	ldc_unmap(lp, cookies, ncookies);
2329 	kfree(buf);
2330 }
2331 EXPORT_SYMBOL(ldc_free_exp_dring);
2332 
2333 static int __init ldc_init(void)
2334 {
2335 	unsigned long major, minor;
2336 	struct mdesc_handle *hp;
2337 	const u64 *v;
2338 	int err;
2339 	u64 mp;
2340 
2341 	hp = mdesc_grab();
2342 	if (!hp)
2343 		return -ENODEV;
2344 
2345 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
2346 	err = -ENODEV;
2347 	if (mp == MDESC_NODE_NULL)
2348 		goto out;
2349 
2350 	v = mdesc_get_property(hp, mp, "domaining-enabled", NULL);
2351 	if (!v)
2352 		goto out;
2353 
2354 	major = 1;
2355 	minor = 0;
2356 	if (sun4v_hvapi_register(HV_GRP_LDOM, major, &minor)) {
2357 		printk(KERN_INFO PFX "Could not register LDOM hvapi.\n");
2358 		goto out;
2359 	}
2360 
2361 	printk(KERN_INFO "%s", version);
2362 
2363 	if (!*v) {
2364 		printk(KERN_INFO PFX "Domaining disabled.\n");
2365 		goto out;
2366 	}
2367 	ldom_domaining_enabled = 1;
2368 	err = 0;
2369 
2370 out:
2371 	mdesc_release(hp);
2372 	return err;
2373 }
2374 
2375 core_initcall(ldc_init);
2376