xref: /openbmc/linux/arch/sparc/kernel/ldc.c (revision 82ced6fd)
1 /* ldc.c: Logical Domain Channel link-layer protocol driver.
2  *
3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/module.h>
8 #include <linux/slab.h>
9 #include <linux/spinlock.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/string.h>
13 #include <linux/scatterlist.h>
14 #include <linux/interrupt.h>
15 #include <linux/list.h>
16 #include <linux/init.h>
17 
18 #include <asm/hypervisor.h>
19 #include <asm/iommu.h>
20 #include <asm/page.h>
21 #include <asm/ldc.h>
22 #include <asm/mdesc.h>
23 
24 #define DRV_MODULE_NAME		"ldc"
25 #define PFX DRV_MODULE_NAME	": "
26 #define DRV_MODULE_VERSION	"1.1"
27 #define DRV_MODULE_RELDATE	"July 22, 2008"
28 
29 static char version[] __devinitdata =
30 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
31 #define LDC_PACKET_SIZE		64
32 
33 /* Packet header layout for unreliable and reliable mode frames.
34  * When in RAW mode, packets are simply straight 64-byte payloads
35  * with no headers.
36  */
37 struct ldc_packet {
38 	u8			type;
39 #define LDC_CTRL		0x01
40 #define LDC_DATA		0x02
41 #define LDC_ERR			0x10
42 
43 	u8			stype;
44 #define LDC_INFO		0x01
45 #define LDC_ACK			0x02
46 #define LDC_NACK		0x04
47 
48 	u8			ctrl;
49 #define LDC_VERS		0x01 /* Link Version		*/
50 #define LDC_RTS			0x02 /* Request To Send		*/
51 #define LDC_RTR			0x03 /* Ready To Receive	*/
52 #define LDC_RDX			0x04 /* Ready for Data eXchange	*/
53 #define LDC_CTRL_MSK		0x0f
54 
55 	u8			env;
56 #define LDC_LEN			0x3f
57 #define LDC_FRAG_MASK		0xc0
58 #define LDC_START		0x40
59 #define LDC_STOP		0x80
60 
61 	u32			seqid;
62 
63 	union {
64 		u8		u_data[LDC_PACKET_SIZE - 8];
65 		struct {
66 			u32	pad;
67 			u32	ackid;
68 			u8	r_data[LDC_PACKET_SIZE - 8 - 8];
69 		} r;
70 	} u;
71 };
72 
73 struct ldc_version {
74 	u16 major;
75 	u16 minor;
76 };
77 
78 /* Ordered from largest major to lowest.  */
79 static struct ldc_version ver_arr[] = {
80 	{ .major = 1, .minor = 0 },
81 };
82 
83 #define LDC_DEFAULT_MTU			(4 * LDC_PACKET_SIZE)
84 #define LDC_DEFAULT_NUM_ENTRIES		(PAGE_SIZE / LDC_PACKET_SIZE)
85 
86 struct ldc_channel;
87 
88 struct ldc_mode_ops {
89 	int (*write)(struct ldc_channel *, const void *, unsigned int);
90 	int (*read)(struct ldc_channel *, void *, unsigned int);
91 };
92 
93 static const struct ldc_mode_ops raw_ops;
94 static const struct ldc_mode_ops nonraw_ops;
95 static const struct ldc_mode_ops stream_ops;
96 
97 int ldom_domaining_enabled;
98 
99 struct ldc_iommu {
100 	/* Protects arena alloc/free.  */
101 	spinlock_t			lock;
102 	struct iommu_arena		arena;
103 	struct ldc_mtable_entry		*page_table;
104 };
105 
106 struct ldc_channel {
107 	/* Protects all operations that depend upon channel state.  */
108 	spinlock_t			lock;
109 
110 	unsigned long			id;
111 
112 	u8				*mssbuf;
113 	u32				mssbuf_len;
114 	u32				mssbuf_off;
115 
116 	struct ldc_packet		*tx_base;
117 	unsigned long			tx_head;
118 	unsigned long			tx_tail;
119 	unsigned long			tx_num_entries;
120 	unsigned long			tx_ra;
121 
122 	unsigned long			tx_acked;
123 
124 	struct ldc_packet		*rx_base;
125 	unsigned long			rx_head;
126 	unsigned long			rx_tail;
127 	unsigned long			rx_num_entries;
128 	unsigned long			rx_ra;
129 
130 	u32				rcv_nxt;
131 	u32				snd_nxt;
132 
133 	unsigned long			chan_state;
134 
135 	struct ldc_channel_config	cfg;
136 	void				*event_arg;
137 
138 	const struct ldc_mode_ops	*mops;
139 
140 	struct ldc_iommu		iommu;
141 
142 	struct ldc_version		ver;
143 
144 	u8				hs_state;
145 #define LDC_HS_CLOSED			0x00
146 #define LDC_HS_OPEN			0x01
147 #define LDC_HS_GOTVERS			0x02
148 #define LDC_HS_SENTRTR			0x03
149 #define LDC_HS_GOTRTR			0x04
150 #define LDC_HS_COMPLETE			0x10
151 
152 	u8				flags;
153 #define LDC_FLAG_ALLOCED_QUEUES		0x01
154 #define LDC_FLAG_REGISTERED_QUEUES	0x02
155 #define LDC_FLAG_REGISTERED_IRQS	0x04
156 #define LDC_FLAG_RESET			0x10
157 
158 	u8				mss;
159 	u8				state;
160 
161 #define LDC_IRQ_NAME_MAX		32
162 	char				rx_irq_name[LDC_IRQ_NAME_MAX];
163 	char				tx_irq_name[LDC_IRQ_NAME_MAX];
164 
165 	struct hlist_head		mh_list;
166 
167 	struct hlist_node		list;
168 };
169 
170 #define ldcdbg(TYPE, f, a...) \
171 do {	if (lp->cfg.debug & LDC_DEBUG_##TYPE) \
172 		printk(KERN_INFO PFX "ID[%lu] " f, lp->id, ## a); \
173 } while (0)
174 
175 static const char *state_to_str(u8 state)
176 {
177 	switch (state) {
178 	case LDC_STATE_INVALID:
179 		return "INVALID";
180 	case LDC_STATE_INIT:
181 		return "INIT";
182 	case LDC_STATE_BOUND:
183 		return "BOUND";
184 	case LDC_STATE_READY:
185 		return "READY";
186 	case LDC_STATE_CONNECTED:
187 		return "CONNECTED";
188 	default:
189 		return "<UNKNOWN>";
190 	}
191 }
192 
193 static void ldc_set_state(struct ldc_channel *lp, u8 state)
194 {
195 	ldcdbg(STATE, "STATE (%s) --> (%s)\n",
196 	       state_to_str(lp->state),
197 	       state_to_str(state));
198 
199 	lp->state = state;
200 }
201 
202 static unsigned long __advance(unsigned long off, unsigned long num_entries)
203 {
204 	off += LDC_PACKET_SIZE;
205 	if (off == (num_entries * LDC_PACKET_SIZE))
206 		off = 0;
207 
208 	return off;
209 }
210 
211 static unsigned long rx_advance(struct ldc_channel *lp, unsigned long off)
212 {
213 	return __advance(off, lp->rx_num_entries);
214 }
215 
216 static unsigned long tx_advance(struct ldc_channel *lp, unsigned long off)
217 {
218 	return __advance(off, lp->tx_num_entries);
219 }
220 
221 static struct ldc_packet *handshake_get_tx_packet(struct ldc_channel *lp,
222 						  unsigned long *new_tail)
223 {
224 	struct ldc_packet *p;
225 	unsigned long t;
226 
227 	t = tx_advance(lp, lp->tx_tail);
228 	if (t == lp->tx_head)
229 		return NULL;
230 
231 	*new_tail = t;
232 
233 	p = lp->tx_base;
234 	return p + (lp->tx_tail / LDC_PACKET_SIZE);
235 }
236 
237 /* When we are in reliable or stream mode, have to track the next packet
238  * we haven't gotten an ACK for in the TX queue using tx_acked.  We have
239  * to be careful not to stomp over the queue past that point.  During
240  * the handshake, we don't have TX data packets pending in the queue
241  * and that's why handshake_get_tx_packet() need not be mindful of
242  * lp->tx_acked.
243  */
244 static unsigned long head_for_data(struct ldc_channel *lp)
245 {
246 	if (lp->cfg.mode == LDC_MODE_STREAM)
247 		return lp->tx_acked;
248 	return lp->tx_head;
249 }
250 
251 static int tx_has_space_for(struct ldc_channel *lp, unsigned int size)
252 {
253 	unsigned long limit, tail, new_tail, diff;
254 	unsigned int mss;
255 
256 	limit = head_for_data(lp);
257 	tail = lp->tx_tail;
258 	new_tail = tx_advance(lp, tail);
259 	if (new_tail == limit)
260 		return 0;
261 
262 	if (limit > new_tail)
263 		diff = limit - new_tail;
264 	else
265 		diff = (limit +
266 			((lp->tx_num_entries * LDC_PACKET_SIZE) - new_tail));
267 	diff /= LDC_PACKET_SIZE;
268 	mss = lp->mss;
269 
270 	if (diff * mss < size)
271 		return 0;
272 
273 	return 1;
274 }
275 
276 static struct ldc_packet *data_get_tx_packet(struct ldc_channel *lp,
277 					     unsigned long *new_tail)
278 {
279 	struct ldc_packet *p;
280 	unsigned long h, t;
281 
282 	h = head_for_data(lp);
283 	t = tx_advance(lp, lp->tx_tail);
284 	if (t == h)
285 		return NULL;
286 
287 	*new_tail = t;
288 
289 	p = lp->tx_base;
290 	return p + (lp->tx_tail / LDC_PACKET_SIZE);
291 }
292 
293 static int set_tx_tail(struct ldc_channel *lp, unsigned long tail)
294 {
295 	unsigned long orig_tail = lp->tx_tail;
296 	int limit = 1000;
297 
298 	lp->tx_tail = tail;
299 	while (limit-- > 0) {
300 		unsigned long err;
301 
302 		err = sun4v_ldc_tx_set_qtail(lp->id, tail);
303 		if (!err)
304 			return 0;
305 
306 		if (err != HV_EWOULDBLOCK) {
307 			lp->tx_tail = orig_tail;
308 			return -EINVAL;
309 		}
310 		udelay(1);
311 	}
312 
313 	lp->tx_tail = orig_tail;
314 	return -EBUSY;
315 }
316 
317 /* This just updates the head value in the hypervisor using
318  * a polling loop with a timeout.  The caller takes care of
319  * upating software state representing the head change, if any.
320  */
321 static int __set_rx_head(struct ldc_channel *lp, unsigned long head)
322 {
323 	int limit = 1000;
324 
325 	while (limit-- > 0) {
326 		unsigned long err;
327 
328 		err = sun4v_ldc_rx_set_qhead(lp->id, head);
329 		if (!err)
330 			return 0;
331 
332 		if (err != HV_EWOULDBLOCK)
333 			return -EINVAL;
334 
335 		udelay(1);
336 	}
337 
338 	return -EBUSY;
339 }
340 
341 static int send_tx_packet(struct ldc_channel *lp,
342 			  struct ldc_packet *p,
343 			  unsigned long new_tail)
344 {
345 	BUG_ON(p != (lp->tx_base + (lp->tx_tail / LDC_PACKET_SIZE)));
346 
347 	return set_tx_tail(lp, new_tail);
348 }
349 
350 static struct ldc_packet *handshake_compose_ctrl(struct ldc_channel *lp,
351 						 u8 stype, u8 ctrl,
352 						 void *data, int dlen,
353 						 unsigned long *new_tail)
354 {
355 	struct ldc_packet *p = handshake_get_tx_packet(lp, new_tail);
356 
357 	if (p) {
358 		memset(p, 0, sizeof(*p));
359 		p->type = LDC_CTRL;
360 		p->stype = stype;
361 		p->ctrl = ctrl;
362 		if (data)
363 			memcpy(p->u.u_data, data, dlen);
364 	}
365 	return p;
366 }
367 
368 static int start_handshake(struct ldc_channel *lp)
369 {
370 	struct ldc_packet *p;
371 	struct ldc_version *ver;
372 	unsigned long new_tail;
373 
374 	ver = &ver_arr[0];
375 
376 	ldcdbg(HS, "SEND VER INFO maj[%u] min[%u]\n",
377 	       ver->major, ver->minor);
378 
379 	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
380 				   ver, sizeof(*ver), &new_tail);
381 	if (p) {
382 		int err = send_tx_packet(lp, p, new_tail);
383 		if (!err)
384 			lp->flags &= ~LDC_FLAG_RESET;
385 		return err;
386 	}
387 	return -EBUSY;
388 }
389 
390 static int send_version_nack(struct ldc_channel *lp,
391 			     u16 major, u16 minor)
392 {
393 	struct ldc_packet *p;
394 	struct ldc_version ver;
395 	unsigned long new_tail;
396 
397 	ver.major = major;
398 	ver.minor = minor;
399 
400 	p = handshake_compose_ctrl(lp, LDC_NACK, LDC_VERS,
401 				   &ver, sizeof(ver), &new_tail);
402 	if (p) {
403 		ldcdbg(HS, "SEND VER NACK maj[%u] min[%u]\n",
404 		       ver.major, ver.minor);
405 
406 		return send_tx_packet(lp, p, new_tail);
407 	}
408 	return -EBUSY;
409 }
410 
411 static int send_version_ack(struct ldc_channel *lp,
412 			    struct ldc_version *vp)
413 {
414 	struct ldc_packet *p;
415 	unsigned long new_tail;
416 
417 	p = handshake_compose_ctrl(lp, LDC_ACK, LDC_VERS,
418 				   vp, sizeof(*vp), &new_tail);
419 	if (p) {
420 		ldcdbg(HS, "SEND VER ACK maj[%u] min[%u]\n",
421 		       vp->major, vp->minor);
422 
423 		return send_tx_packet(lp, p, new_tail);
424 	}
425 	return -EBUSY;
426 }
427 
428 static int send_rts(struct ldc_channel *lp)
429 {
430 	struct ldc_packet *p;
431 	unsigned long new_tail;
432 
433 	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTS, NULL, 0,
434 				   &new_tail);
435 	if (p) {
436 		p->env = lp->cfg.mode;
437 		p->seqid = 0;
438 		lp->rcv_nxt = 0;
439 
440 		ldcdbg(HS, "SEND RTS env[0x%x] seqid[0x%x]\n",
441 		       p->env, p->seqid);
442 
443 		return send_tx_packet(lp, p, new_tail);
444 	}
445 	return -EBUSY;
446 }
447 
448 static int send_rtr(struct ldc_channel *lp)
449 {
450 	struct ldc_packet *p;
451 	unsigned long new_tail;
452 
453 	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTR, NULL, 0,
454 				   &new_tail);
455 	if (p) {
456 		p->env = lp->cfg.mode;
457 		p->seqid = 0;
458 
459 		ldcdbg(HS, "SEND RTR env[0x%x] seqid[0x%x]\n",
460 		       p->env, p->seqid);
461 
462 		return send_tx_packet(lp, p, new_tail);
463 	}
464 	return -EBUSY;
465 }
466 
467 static int send_rdx(struct ldc_channel *lp)
468 {
469 	struct ldc_packet *p;
470 	unsigned long new_tail;
471 
472 	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RDX, NULL, 0,
473 				   &new_tail);
474 	if (p) {
475 		p->env = 0;
476 		p->seqid = ++lp->snd_nxt;
477 		p->u.r.ackid = lp->rcv_nxt;
478 
479 		ldcdbg(HS, "SEND RDX env[0x%x] seqid[0x%x] ackid[0x%x]\n",
480 		       p->env, p->seqid, p->u.r.ackid);
481 
482 		return send_tx_packet(lp, p, new_tail);
483 	}
484 	return -EBUSY;
485 }
486 
487 static int send_data_nack(struct ldc_channel *lp, struct ldc_packet *data_pkt)
488 {
489 	struct ldc_packet *p;
490 	unsigned long new_tail;
491 	int err;
492 
493 	p = data_get_tx_packet(lp, &new_tail);
494 	if (!p)
495 		return -EBUSY;
496 	memset(p, 0, sizeof(*p));
497 	p->type = data_pkt->type;
498 	p->stype = LDC_NACK;
499 	p->ctrl = data_pkt->ctrl & LDC_CTRL_MSK;
500 	p->seqid = lp->snd_nxt + 1;
501 	p->u.r.ackid = lp->rcv_nxt;
502 
503 	ldcdbg(HS, "SEND DATA NACK type[0x%x] ctl[0x%x] seq[0x%x] ack[0x%x]\n",
504 	       p->type, p->ctrl, p->seqid, p->u.r.ackid);
505 
506 	err = send_tx_packet(lp, p, new_tail);
507 	if (!err)
508 		lp->snd_nxt++;
509 
510 	return err;
511 }
512 
513 static int ldc_abort(struct ldc_channel *lp)
514 {
515 	unsigned long hv_err;
516 
517 	ldcdbg(STATE, "ABORT\n");
518 
519 	/* We report but do not act upon the hypervisor errors because
520 	 * there really isn't much we can do if they fail at this point.
521 	 */
522 	hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
523 	if (hv_err)
524 		printk(KERN_ERR PFX "ldc_abort: "
525 		       "sun4v_ldc_tx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
526 		       lp->id, lp->tx_ra, lp->tx_num_entries, hv_err);
527 
528 	hv_err = sun4v_ldc_tx_get_state(lp->id,
529 					&lp->tx_head,
530 					&lp->tx_tail,
531 					&lp->chan_state);
532 	if (hv_err)
533 		printk(KERN_ERR PFX "ldc_abort: "
534 		       "sun4v_ldc_tx_get_state(%lx,...) failed, err=%lu\n",
535 		       lp->id, hv_err);
536 
537 	hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
538 	if (hv_err)
539 		printk(KERN_ERR PFX "ldc_abort: "
540 		       "sun4v_ldc_rx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
541 		       lp->id, lp->rx_ra, lp->rx_num_entries, hv_err);
542 
543 	/* Refetch the RX queue state as well, because we could be invoked
544 	 * here in the queue processing context.
545 	 */
546 	hv_err = sun4v_ldc_rx_get_state(lp->id,
547 					&lp->rx_head,
548 					&lp->rx_tail,
549 					&lp->chan_state);
550 	if (hv_err)
551 		printk(KERN_ERR PFX "ldc_abort: "
552 		       "sun4v_ldc_rx_get_state(%lx,...) failed, err=%lu\n",
553 		       lp->id, hv_err);
554 
555 	return -ECONNRESET;
556 }
557 
558 static struct ldc_version *find_by_major(u16 major)
559 {
560 	struct ldc_version *ret = NULL;
561 	int i;
562 
563 	for (i = 0; i < ARRAY_SIZE(ver_arr); i++) {
564 		struct ldc_version *v = &ver_arr[i];
565 		if (v->major <= major) {
566 			ret = v;
567 			break;
568 		}
569 	}
570 	return ret;
571 }
572 
573 static int process_ver_info(struct ldc_channel *lp, struct ldc_version *vp)
574 {
575 	struct ldc_version *vap;
576 	int err;
577 
578 	ldcdbg(HS, "GOT VERSION INFO major[%x] minor[%x]\n",
579 	       vp->major, vp->minor);
580 
581 	if (lp->hs_state == LDC_HS_GOTVERS) {
582 		lp->hs_state = LDC_HS_OPEN;
583 		memset(&lp->ver, 0, sizeof(lp->ver));
584 	}
585 
586 	vap = find_by_major(vp->major);
587 	if (!vap) {
588 		err = send_version_nack(lp, 0, 0);
589 	} else if (vap->major != vp->major) {
590 		err = send_version_nack(lp, vap->major, vap->minor);
591 	} else {
592 		struct ldc_version ver = *vp;
593 		if (ver.minor > vap->minor)
594 			ver.minor = vap->minor;
595 		err = send_version_ack(lp, &ver);
596 		if (!err) {
597 			lp->ver = ver;
598 			lp->hs_state = LDC_HS_GOTVERS;
599 		}
600 	}
601 	if (err)
602 		return ldc_abort(lp);
603 
604 	return 0;
605 }
606 
607 static int process_ver_ack(struct ldc_channel *lp, struct ldc_version *vp)
608 {
609 	ldcdbg(HS, "GOT VERSION ACK major[%x] minor[%x]\n",
610 	       vp->major, vp->minor);
611 
612 	if (lp->hs_state == LDC_HS_GOTVERS) {
613 		if (lp->ver.major != vp->major ||
614 		    lp->ver.minor != vp->minor)
615 			return ldc_abort(lp);
616 	} else {
617 		lp->ver = *vp;
618 		lp->hs_state = LDC_HS_GOTVERS;
619 	}
620 	if (send_rts(lp))
621 		return ldc_abort(lp);
622 	return 0;
623 }
624 
625 static int process_ver_nack(struct ldc_channel *lp, struct ldc_version *vp)
626 {
627 	struct ldc_version *vap;
628 	struct ldc_packet *p;
629 	unsigned long new_tail;
630 
631 	if (vp->major == 0 && vp->minor == 0)
632 		return ldc_abort(lp);
633 
634 	vap = find_by_major(vp->major);
635 	if (!vap)
636 		return ldc_abort(lp);
637 
638 	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
639 					   vap, sizeof(*vap),
640 					   &new_tail);
641 	if (!p)
642 		return ldc_abort(lp);
643 
644 	return send_tx_packet(lp, p, new_tail);
645 }
646 
647 static int process_version(struct ldc_channel *lp,
648 			   struct ldc_packet *p)
649 {
650 	struct ldc_version *vp;
651 
652 	vp = (struct ldc_version *) p->u.u_data;
653 
654 	switch (p->stype) {
655 	case LDC_INFO:
656 		return process_ver_info(lp, vp);
657 
658 	case LDC_ACK:
659 		return process_ver_ack(lp, vp);
660 
661 	case LDC_NACK:
662 		return process_ver_nack(lp, vp);
663 
664 	default:
665 		return ldc_abort(lp);
666 	}
667 }
668 
669 static int process_rts(struct ldc_channel *lp,
670 		       struct ldc_packet *p)
671 {
672 	ldcdbg(HS, "GOT RTS stype[%x] seqid[%x] env[%x]\n",
673 	       p->stype, p->seqid, p->env);
674 
675 	if (p->stype     != LDC_INFO	   ||
676 	    lp->hs_state != LDC_HS_GOTVERS ||
677 	    p->env       != lp->cfg.mode)
678 		return ldc_abort(lp);
679 
680 	lp->snd_nxt = p->seqid;
681 	lp->rcv_nxt = p->seqid;
682 	lp->hs_state = LDC_HS_SENTRTR;
683 	if (send_rtr(lp))
684 		return ldc_abort(lp);
685 
686 	return 0;
687 }
688 
689 static int process_rtr(struct ldc_channel *lp,
690 		       struct ldc_packet *p)
691 {
692 	ldcdbg(HS, "GOT RTR stype[%x] seqid[%x] env[%x]\n",
693 	       p->stype, p->seqid, p->env);
694 
695 	if (p->stype     != LDC_INFO ||
696 	    p->env       != lp->cfg.mode)
697 		return ldc_abort(lp);
698 
699 	lp->snd_nxt = p->seqid;
700 	lp->hs_state = LDC_HS_COMPLETE;
701 	ldc_set_state(lp, LDC_STATE_CONNECTED);
702 	send_rdx(lp);
703 
704 	return LDC_EVENT_UP;
705 }
706 
707 static int rx_seq_ok(struct ldc_channel *lp, u32 seqid)
708 {
709 	return lp->rcv_nxt + 1 == seqid;
710 }
711 
712 static int process_rdx(struct ldc_channel *lp,
713 		       struct ldc_packet *p)
714 {
715 	ldcdbg(HS, "GOT RDX stype[%x] seqid[%x] env[%x] ackid[%x]\n",
716 	       p->stype, p->seqid, p->env, p->u.r.ackid);
717 
718 	if (p->stype != LDC_INFO ||
719 	    !(rx_seq_ok(lp, p->seqid)))
720 		return ldc_abort(lp);
721 
722 	lp->rcv_nxt = p->seqid;
723 
724 	lp->hs_state = LDC_HS_COMPLETE;
725 	ldc_set_state(lp, LDC_STATE_CONNECTED);
726 
727 	return LDC_EVENT_UP;
728 }
729 
730 static int process_control_frame(struct ldc_channel *lp,
731 				 struct ldc_packet *p)
732 {
733 	switch (p->ctrl) {
734 	case LDC_VERS:
735 		return process_version(lp, p);
736 
737 	case LDC_RTS:
738 		return process_rts(lp, p);
739 
740 	case LDC_RTR:
741 		return process_rtr(lp, p);
742 
743 	case LDC_RDX:
744 		return process_rdx(lp, p);
745 
746 	default:
747 		return ldc_abort(lp);
748 	}
749 }
750 
751 static int process_error_frame(struct ldc_channel *lp,
752 			       struct ldc_packet *p)
753 {
754 	return ldc_abort(lp);
755 }
756 
757 static int process_data_ack(struct ldc_channel *lp,
758 			    struct ldc_packet *ack)
759 {
760 	unsigned long head = lp->tx_acked;
761 	u32 ackid = ack->u.r.ackid;
762 
763 	while (1) {
764 		struct ldc_packet *p = lp->tx_base + (head / LDC_PACKET_SIZE);
765 
766 		head = tx_advance(lp, head);
767 
768 		if (p->seqid == ackid) {
769 			lp->tx_acked = head;
770 			return 0;
771 		}
772 		if (head == lp->tx_tail)
773 			return ldc_abort(lp);
774 	}
775 
776 	return 0;
777 }
778 
779 static void send_events(struct ldc_channel *lp, unsigned int event_mask)
780 {
781 	if (event_mask & LDC_EVENT_RESET)
782 		lp->cfg.event(lp->event_arg, LDC_EVENT_RESET);
783 	if (event_mask & LDC_EVENT_UP)
784 		lp->cfg.event(lp->event_arg, LDC_EVENT_UP);
785 	if (event_mask & LDC_EVENT_DATA_READY)
786 		lp->cfg.event(lp->event_arg, LDC_EVENT_DATA_READY);
787 }
788 
789 static irqreturn_t ldc_rx(int irq, void *dev_id)
790 {
791 	struct ldc_channel *lp = dev_id;
792 	unsigned long orig_state, hv_err, flags;
793 	unsigned int event_mask;
794 
795 	spin_lock_irqsave(&lp->lock, flags);
796 
797 	orig_state = lp->chan_state;
798 	hv_err = sun4v_ldc_rx_get_state(lp->id,
799 					&lp->rx_head,
800 					&lp->rx_tail,
801 					&lp->chan_state);
802 
803 	ldcdbg(RX, "RX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
804 	       orig_state, lp->chan_state, lp->rx_head, lp->rx_tail);
805 
806 	event_mask = 0;
807 
808 	if (lp->cfg.mode == LDC_MODE_RAW &&
809 	    lp->chan_state == LDC_CHANNEL_UP) {
810 		lp->hs_state = LDC_HS_COMPLETE;
811 		ldc_set_state(lp, LDC_STATE_CONNECTED);
812 
813 		event_mask |= LDC_EVENT_UP;
814 
815 		orig_state = lp->chan_state;
816 	}
817 
818 	/* If we are in reset state, flush the RX queue and ignore
819 	 * everything.
820 	 */
821 	if (lp->flags & LDC_FLAG_RESET) {
822 		(void) __set_rx_head(lp, lp->rx_tail);
823 		goto out;
824 	}
825 
826 	/* Once we finish the handshake, we let the ldc_read()
827 	 * paths do all of the control frame and state management.
828 	 * Just trigger the callback.
829 	 */
830 	if (lp->hs_state == LDC_HS_COMPLETE) {
831 handshake_complete:
832 		if (lp->chan_state != orig_state) {
833 			unsigned int event = LDC_EVENT_RESET;
834 
835 			if (lp->chan_state == LDC_CHANNEL_UP)
836 				event = LDC_EVENT_UP;
837 
838 			event_mask |= event;
839 		}
840 		if (lp->rx_head != lp->rx_tail)
841 			event_mask |= LDC_EVENT_DATA_READY;
842 
843 		goto out;
844 	}
845 
846 	if (lp->chan_state != orig_state)
847 		goto out;
848 
849 	while (lp->rx_head != lp->rx_tail) {
850 		struct ldc_packet *p;
851 		unsigned long new;
852 		int err;
853 
854 		p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
855 
856 		switch (p->type) {
857 		case LDC_CTRL:
858 			err = process_control_frame(lp, p);
859 			if (err > 0)
860 				event_mask |= err;
861 			break;
862 
863 		case LDC_DATA:
864 			event_mask |= LDC_EVENT_DATA_READY;
865 			err = 0;
866 			break;
867 
868 		case LDC_ERR:
869 			err = process_error_frame(lp, p);
870 			break;
871 
872 		default:
873 			err = ldc_abort(lp);
874 			break;
875 		}
876 
877 		if (err < 0)
878 			break;
879 
880 		new = lp->rx_head;
881 		new += LDC_PACKET_SIZE;
882 		if (new == (lp->rx_num_entries * LDC_PACKET_SIZE))
883 			new = 0;
884 		lp->rx_head = new;
885 
886 		err = __set_rx_head(lp, new);
887 		if (err < 0) {
888 			(void) ldc_abort(lp);
889 			break;
890 		}
891 		if (lp->hs_state == LDC_HS_COMPLETE)
892 			goto handshake_complete;
893 	}
894 
895 out:
896 	spin_unlock_irqrestore(&lp->lock, flags);
897 
898 	send_events(lp, event_mask);
899 
900 	return IRQ_HANDLED;
901 }
902 
903 static irqreturn_t ldc_tx(int irq, void *dev_id)
904 {
905 	struct ldc_channel *lp = dev_id;
906 	unsigned long flags, hv_err, orig_state;
907 	unsigned int event_mask = 0;
908 
909 	spin_lock_irqsave(&lp->lock, flags);
910 
911 	orig_state = lp->chan_state;
912 	hv_err = sun4v_ldc_tx_get_state(lp->id,
913 					&lp->tx_head,
914 					&lp->tx_tail,
915 					&lp->chan_state);
916 
917 	ldcdbg(TX, " TX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
918 	       orig_state, lp->chan_state, lp->tx_head, lp->tx_tail);
919 
920 	if (lp->cfg.mode == LDC_MODE_RAW &&
921 	    lp->chan_state == LDC_CHANNEL_UP) {
922 		lp->hs_state = LDC_HS_COMPLETE;
923 		ldc_set_state(lp, LDC_STATE_CONNECTED);
924 
925 		event_mask |= LDC_EVENT_UP;
926 	}
927 
928 	spin_unlock_irqrestore(&lp->lock, flags);
929 
930 	send_events(lp, event_mask);
931 
932 	return IRQ_HANDLED;
933 }
934 
935 /* XXX ldc_alloc() and ldc_free() needs to run under a mutex so
936  * XXX that addition and removal from the ldc_channel_list has
937  * XXX atomicity, otherwise the __ldc_channel_exists() check is
938  * XXX totally pointless as another thread can slip into ldc_alloc()
939  * XXX and add a channel with the same ID.  There also needs to be
940  * XXX a spinlock for ldc_channel_list.
941  */
942 static HLIST_HEAD(ldc_channel_list);
943 
944 static int __ldc_channel_exists(unsigned long id)
945 {
946 	struct ldc_channel *lp;
947 	struct hlist_node *n;
948 
949 	hlist_for_each_entry(lp, n, &ldc_channel_list, list) {
950 		if (lp->id == id)
951 			return 1;
952 	}
953 	return 0;
954 }
955 
956 static int alloc_queue(const char *name, unsigned long num_entries,
957 		       struct ldc_packet **base, unsigned long *ra)
958 {
959 	unsigned long size, order;
960 	void *q;
961 
962 	size = num_entries * LDC_PACKET_SIZE;
963 	order = get_order(size);
964 
965 	q = (void *) __get_free_pages(GFP_KERNEL, order);
966 	if (!q) {
967 		printk(KERN_ERR PFX "Alloc of %s queue failed with "
968 		       "size=%lu order=%lu\n", name, size, order);
969 		return -ENOMEM;
970 	}
971 
972 	memset(q, 0, PAGE_SIZE << order);
973 
974 	*base = q;
975 	*ra = __pa(q);
976 
977 	return 0;
978 }
979 
980 static void free_queue(unsigned long num_entries, struct ldc_packet *q)
981 {
982 	unsigned long size, order;
983 
984 	if (!q)
985 		return;
986 
987 	size = num_entries * LDC_PACKET_SIZE;
988 	order = get_order(size);
989 
990 	free_pages((unsigned long)q, order);
991 }
992 
993 /* XXX Make this configurable... XXX */
994 #define LDC_IOTABLE_SIZE	(8 * 1024)
995 
996 static int ldc_iommu_init(struct ldc_channel *lp)
997 {
998 	unsigned long sz, num_tsb_entries, tsbsize, order;
999 	struct ldc_iommu *iommu = &lp->iommu;
1000 	struct ldc_mtable_entry *table;
1001 	unsigned long hv_err;
1002 	int err;
1003 
1004 	num_tsb_entries = LDC_IOTABLE_SIZE;
1005 	tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1006 
1007 	spin_lock_init(&iommu->lock);
1008 
1009 	sz = num_tsb_entries / 8;
1010 	sz = (sz + 7UL) & ~7UL;
1011 	iommu->arena.map = kzalloc(sz, GFP_KERNEL);
1012 	if (!iommu->arena.map) {
1013 		printk(KERN_ERR PFX "Alloc of arena map failed, sz=%lu\n", sz);
1014 		return -ENOMEM;
1015 	}
1016 
1017 	iommu->arena.limit = num_tsb_entries;
1018 
1019 	order = get_order(tsbsize);
1020 
1021 	table = (struct ldc_mtable_entry *)
1022 		__get_free_pages(GFP_KERNEL, order);
1023 	err = -ENOMEM;
1024 	if (!table) {
1025 		printk(KERN_ERR PFX "Alloc of MTE table failed, "
1026 		       "size=%lu order=%lu\n", tsbsize, order);
1027 		goto out_free_map;
1028 	}
1029 
1030 	memset(table, 0, PAGE_SIZE << order);
1031 
1032 	iommu->page_table = table;
1033 
1034 	hv_err = sun4v_ldc_set_map_table(lp->id, __pa(table),
1035 					 num_tsb_entries);
1036 	err = -EINVAL;
1037 	if (hv_err)
1038 		goto out_free_table;
1039 
1040 	return 0;
1041 
1042 out_free_table:
1043 	free_pages((unsigned long) table, order);
1044 	iommu->page_table = NULL;
1045 
1046 out_free_map:
1047 	kfree(iommu->arena.map);
1048 	iommu->arena.map = NULL;
1049 
1050 	return err;
1051 }
1052 
1053 static void ldc_iommu_release(struct ldc_channel *lp)
1054 {
1055 	struct ldc_iommu *iommu = &lp->iommu;
1056 	unsigned long num_tsb_entries, tsbsize, order;
1057 
1058 	(void) sun4v_ldc_set_map_table(lp->id, 0, 0);
1059 
1060 	num_tsb_entries = iommu->arena.limit;
1061 	tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1062 	order = get_order(tsbsize);
1063 
1064 	free_pages((unsigned long) iommu->page_table, order);
1065 	iommu->page_table = NULL;
1066 
1067 	kfree(iommu->arena.map);
1068 	iommu->arena.map = NULL;
1069 }
1070 
1071 struct ldc_channel *ldc_alloc(unsigned long id,
1072 			      const struct ldc_channel_config *cfgp,
1073 			      void *event_arg)
1074 {
1075 	struct ldc_channel *lp;
1076 	const struct ldc_mode_ops *mops;
1077 	unsigned long dummy1, dummy2, hv_err;
1078 	u8 mss, *mssbuf;
1079 	int err;
1080 
1081 	err = -ENODEV;
1082 	if (!ldom_domaining_enabled)
1083 		goto out_err;
1084 
1085 	err = -EINVAL;
1086 	if (!cfgp)
1087 		goto out_err;
1088 
1089 	switch (cfgp->mode) {
1090 	case LDC_MODE_RAW:
1091 		mops = &raw_ops;
1092 		mss = LDC_PACKET_SIZE;
1093 		break;
1094 
1095 	case LDC_MODE_UNRELIABLE:
1096 		mops = &nonraw_ops;
1097 		mss = LDC_PACKET_SIZE - 8;
1098 		break;
1099 
1100 	case LDC_MODE_STREAM:
1101 		mops = &stream_ops;
1102 		mss = LDC_PACKET_SIZE - 8 - 8;
1103 		break;
1104 
1105 	default:
1106 		goto out_err;
1107 	}
1108 
1109 	if (!cfgp->event || !event_arg || !cfgp->rx_irq || !cfgp->tx_irq)
1110 		goto out_err;
1111 
1112 	hv_err = sun4v_ldc_tx_qinfo(id, &dummy1, &dummy2);
1113 	err = -ENODEV;
1114 	if (hv_err == HV_ECHANNEL)
1115 		goto out_err;
1116 
1117 	err = -EEXIST;
1118 	if (__ldc_channel_exists(id))
1119 		goto out_err;
1120 
1121 	mssbuf = NULL;
1122 
1123 	lp = kzalloc(sizeof(*lp), GFP_KERNEL);
1124 	err = -ENOMEM;
1125 	if (!lp)
1126 		goto out_err;
1127 
1128 	spin_lock_init(&lp->lock);
1129 
1130 	lp->id = id;
1131 
1132 	err = ldc_iommu_init(lp);
1133 	if (err)
1134 		goto out_free_ldc;
1135 
1136 	lp->mops = mops;
1137 	lp->mss = mss;
1138 
1139 	lp->cfg = *cfgp;
1140 	if (!lp->cfg.mtu)
1141 		lp->cfg.mtu = LDC_DEFAULT_MTU;
1142 
1143 	if (lp->cfg.mode == LDC_MODE_STREAM) {
1144 		mssbuf = kzalloc(lp->cfg.mtu, GFP_KERNEL);
1145 		if (!mssbuf) {
1146 			err = -ENOMEM;
1147 			goto out_free_iommu;
1148 		}
1149 		lp->mssbuf = mssbuf;
1150 	}
1151 
1152 	lp->event_arg = event_arg;
1153 
1154 	/* XXX allow setting via ldc_channel_config to override defaults
1155 	 * XXX or use some formula based upon mtu
1156 	 */
1157 	lp->tx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1158 	lp->rx_num_entries = LDC_DEFAULT_NUM_ENTRIES;
1159 
1160 	err = alloc_queue("TX", lp->tx_num_entries,
1161 			  &lp->tx_base, &lp->tx_ra);
1162 	if (err)
1163 		goto out_free_mssbuf;
1164 
1165 	err = alloc_queue("RX", lp->rx_num_entries,
1166 			  &lp->rx_base, &lp->rx_ra);
1167 	if (err)
1168 		goto out_free_txq;
1169 
1170 	lp->flags |= LDC_FLAG_ALLOCED_QUEUES;
1171 
1172 	lp->hs_state = LDC_HS_CLOSED;
1173 	ldc_set_state(lp, LDC_STATE_INIT);
1174 
1175 	INIT_HLIST_NODE(&lp->list);
1176 	hlist_add_head(&lp->list, &ldc_channel_list);
1177 
1178 	INIT_HLIST_HEAD(&lp->mh_list);
1179 
1180 	return lp;
1181 
1182 out_free_txq:
1183 	free_queue(lp->tx_num_entries, lp->tx_base);
1184 
1185 out_free_mssbuf:
1186 	kfree(mssbuf);
1187 
1188 out_free_iommu:
1189 	ldc_iommu_release(lp);
1190 
1191 out_free_ldc:
1192 	kfree(lp);
1193 
1194 out_err:
1195 	return ERR_PTR(err);
1196 }
1197 EXPORT_SYMBOL(ldc_alloc);
1198 
1199 void ldc_free(struct ldc_channel *lp)
1200 {
1201 	if (lp->flags & LDC_FLAG_REGISTERED_IRQS) {
1202 		free_irq(lp->cfg.rx_irq, lp);
1203 		free_irq(lp->cfg.tx_irq, lp);
1204 	}
1205 
1206 	if (lp->flags & LDC_FLAG_REGISTERED_QUEUES) {
1207 		sun4v_ldc_tx_qconf(lp->id, 0, 0);
1208 		sun4v_ldc_rx_qconf(lp->id, 0, 0);
1209 		lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1210 	}
1211 	if (lp->flags & LDC_FLAG_ALLOCED_QUEUES) {
1212 		free_queue(lp->tx_num_entries, lp->tx_base);
1213 		free_queue(lp->rx_num_entries, lp->rx_base);
1214 		lp->flags &= ~LDC_FLAG_ALLOCED_QUEUES;
1215 	}
1216 
1217 	hlist_del(&lp->list);
1218 
1219 	kfree(lp->mssbuf);
1220 
1221 	ldc_iommu_release(lp);
1222 
1223 	kfree(lp);
1224 }
1225 EXPORT_SYMBOL(ldc_free);
1226 
1227 /* Bind the channel.  This registers the LDC queues with
1228  * the hypervisor and puts the channel into a pseudo-listening
1229  * state.  This does not initiate a handshake, ldc_connect() does
1230  * that.
1231  */
1232 int ldc_bind(struct ldc_channel *lp, const char *name)
1233 {
1234 	unsigned long hv_err, flags;
1235 	int err = -EINVAL;
1236 
1237 	if (!name ||
1238 	    (lp->state != LDC_STATE_INIT))
1239 		return -EINVAL;
1240 
1241 	snprintf(lp->rx_irq_name, LDC_IRQ_NAME_MAX, "%s RX", name);
1242 	snprintf(lp->tx_irq_name, LDC_IRQ_NAME_MAX, "%s TX", name);
1243 
1244 	err = request_irq(lp->cfg.rx_irq, ldc_rx,
1245 			  IRQF_SAMPLE_RANDOM | IRQF_SHARED,
1246 			  lp->rx_irq_name, lp);
1247 	if (err)
1248 		return err;
1249 
1250 	err = request_irq(lp->cfg.tx_irq, ldc_tx,
1251 			  IRQF_SAMPLE_RANDOM | IRQF_SHARED,
1252 			  lp->tx_irq_name, lp);
1253 	if (err) {
1254 		free_irq(lp->cfg.rx_irq, lp);
1255 		return err;
1256 	}
1257 
1258 
1259 	spin_lock_irqsave(&lp->lock, flags);
1260 
1261 	enable_irq(lp->cfg.rx_irq);
1262 	enable_irq(lp->cfg.tx_irq);
1263 
1264 	lp->flags |= LDC_FLAG_REGISTERED_IRQS;
1265 
1266 	err = -ENODEV;
1267 	hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1268 	if (hv_err)
1269 		goto out_free_irqs;
1270 
1271 	hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1272 	if (hv_err)
1273 		goto out_free_irqs;
1274 
1275 	hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1276 	if (hv_err)
1277 		goto out_unmap_tx;
1278 
1279 	hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1280 	if (hv_err)
1281 		goto out_unmap_tx;
1282 
1283 	lp->flags |= LDC_FLAG_REGISTERED_QUEUES;
1284 
1285 	hv_err = sun4v_ldc_tx_get_state(lp->id,
1286 					&lp->tx_head,
1287 					&lp->tx_tail,
1288 					&lp->chan_state);
1289 	err = -EBUSY;
1290 	if (hv_err)
1291 		goto out_unmap_rx;
1292 
1293 	lp->tx_acked = lp->tx_head;
1294 
1295 	lp->hs_state = LDC_HS_OPEN;
1296 	ldc_set_state(lp, LDC_STATE_BOUND);
1297 
1298 	spin_unlock_irqrestore(&lp->lock, flags);
1299 
1300 	return 0;
1301 
1302 out_unmap_rx:
1303 	lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1304 	sun4v_ldc_rx_qconf(lp->id, 0, 0);
1305 
1306 out_unmap_tx:
1307 	sun4v_ldc_tx_qconf(lp->id, 0, 0);
1308 
1309 out_free_irqs:
1310 	lp->flags &= ~LDC_FLAG_REGISTERED_IRQS;
1311 	free_irq(lp->cfg.tx_irq, lp);
1312 	free_irq(lp->cfg.rx_irq, lp);
1313 
1314 	spin_unlock_irqrestore(&lp->lock, flags);
1315 
1316 	return err;
1317 }
1318 EXPORT_SYMBOL(ldc_bind);
1319 
1320 int ldc_connect(struct ldc_channel *lp)
1321 {
1322 	unsigned long flags;
1323 	int err;
1324 
1325 	if (lp->cfg.mode == LDC_MODE_RAW)
1326 		return -EINVAL;
1327 
1328 	spin_lock_irqsave(&lp->lock, flags);
1329 
1330 	if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1331 	    !(lp->flags & LDC_FLAG_REGISTERED_QUEUES) ||
1332 	    lp->hs_state != LDC_HS_OPEN)
1333 		err = -EINVAL;
1334 	else
1335 		err = start_handshake(lp);
1336 
1337 	spin_unlock_irqrestore(&lp->lock, flags);
1338 
1339 	return err;
1340 }
1341 EXPORT_SYMBOL(ldc_connect);
1342 
1343 int ldc_disconnect(struct ldc_channel *lp)
1344 {
1345 	unsigned long hv_err, flags;
1346 	int err;
1347 
1348 	if (lp->cfg.mode == LDC_MODE_RAW)
1349 		return -EINVAL;
1350 
1351 	if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1352 	    !(lp->flags & LDC_FLAG_REGISTERED_QUEUES))
1353 		return -EINVAL;
1354 
1355 	spin_lock_irqsave(&lp->lock, flags);
1356 
1357 	err = -ENODEV;
1358 	hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1359 	if (hv_err)
1360 		goto out_err;
1361 
1362 	hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1363 	if (hv_err)
1364 		goto out_err;
1365 
1366 	hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1367 	if (hv_err)
1368 		goto out_err;
1369 
1370 	hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1371 	if (hv_err)
1372 		goto out_err;
1373 
1374 	ldc_set_state(lp, LDC_STATE_BOUND);
1375 	lp->hs_state = LDC_HS_OPEN;
1376 	lp->flags |= LDC_FLAG_RESET;
1377 
1378 	spin_unlock_irqrestore(&lp->lock, flags);
1379 
1380 	return 0;
1381 
1382 out_err:
1383 	sun4v_ldc_tx_qconf(lp->id, 0, 0);
1384 	sun4v_ldc_rx_qconf(lp->id, 0, 0);
1385 	free_irq(lp->cfg.tx_irq, lp);
1386 	free_irq(lp->cfg.rx_irq, lp);
1387 	lp->flags &= ~(LDC_FLAG_REGISTERED_IRQS |
1388 		       LDC_FLAG_REGISTERED_QUEUES);
1389 	ldc_set_state(lp, LDC_STATE_INIT);
1390 
1391 	spin_unlock_irqrestore(&lp->lock, flags);
1392 
1393 	return err;
1394 }
1395 EXPORT_SYMBOL(ldc_disconnect);
1396 
1397 int ldc_state(struct ldc_channel *lp)
1398 {
1399 	return lp->state;
1400 }
1401 EXPORT_SYMBOL(ldc_state);
1402 
1403 static int write_raw(struct ldc_channel *lp, const void *buf, unsigned int size)
1404 {
1405 	struct ldc_packet *p;
1406 	unsigned long new_tail;
1407 	int err;
1408 
1409 	if (size > LDC_PACKET_SIZE)
1410 		return -EMSGSIZE;
1411 
1412 	p = data_get_tx_packet(lp, &new_tail);
1413 	if (!p)
1414 		return -EAGAIN;
1415 
1416 	memcpy(p, buf, size);
1417 
1418 	err = send_tx_packet(lp, p, new_tail);
1419 	if (!err)
1420 		err = size;
1421 
1422 	return err;
1423 }
1424 
1425 static int read_raw(struct ldc_channel *lp, void *buf, unsigned int size)
1426 {
1427 	struct ldc_packet *p;
1428 	unsigned long hv_err, new;
1429 	int err;
1430 
1431 	if (size < LDC_PACKET_SIZE)
1432 		return -EINVAL;
1433 
1434 	hv_err = sun4v_ldc_rx_get_state(lp->id,
1435 					&lp->rx_head,
1436 					&lp->rx_tail,
1437 					&lp->chan_state);
1438 	if (hv_err)
1439 		return ldc_abort(lp);
1440 
1441 	if (lp->chan_state == LDC_CHANNEL_DOWN ||
1442 	    lp->chan_state == LDC_CHANNEL_RESETTING)
1443 		return -ECONNRESET;
1444 
1445 	if (lp->rx_head == lp->rx_tail)
1446 		return 0;
1447 
1448 	p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
1449 	memcpy(buf, p, LDC_PACKET_SIZE);
1450 
1451 	new = rx_advance(lp, lp->rx_head);
1452 	lp->rx_head = new;
1453 
1454 	err = __set_rx_head(lp, new);
1455 	if (err < 0)
1456 		err = -ECONNRESET;
1457 	else
1458 		err = LDC_PACKET_SIZE;
1459 
1460 	return err;
1461 }
1462 
1463 static const struct ldc_mode_ops raw_ops = {
1464 	.write		=	write_raw,
1465 	.read		=	read_raw,
1466 };
1467 
1468 static int write_nonraw(struct ldc_channel *lp, const void *buf,
1469 			unsigned int size)
1470 {
1471 	unsigned long hv_err, tail;
1472 	unsigned int copied;
1473 	u32 seq;
1474 	int err;
1475 
1476 	hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail,
1477 					&lp->chan_state);
1478 	if (unlikely(hv_err))
1479 		return -EBUSY;
1480 
1481 	if (unlikely(lp->chan_state != LDC_CHANNEL_UP))
1482 		return ldc_abort(lp);
1483 
1484 	if (!tx_has_space_for(lp, size))
1485 		return -EAGAIN;
1486 
1487 	seq = lp->snd_nxt;
1488 	copied = 0;
1489 	tail = lp->tx_tail;
1490 	while (copied < size) {
1491 		struct ldc_packet *p = lp->tx_base + (tail / LDC_PACKET_SIZE);
1492 		u8 *data = ((lp->cfg.mode == LDC_MODE_UNRELIABLE) ?
1493 			    p->u.u_data :
1494 			    p->u.r.r_data);
1495 		int data_len;
1496 
1497 		p->type = LDC_DATA;
1498 		p->stype = LDC_INFO;
1499 		p->ctrl = 0;
1500 
1501 		data_len = size - copied;
1502 		if (data_len > lp->mss)
1503 			data_len = lp->mss;
1504 
1505 		BUG_ON(data_len > LDC_LEN);
1506 
1507 		p->env = (data_len |
1508 			  (copied == 0 ? LDC_START : 0) |
1509 			  (data_len == size - copied ? LDC_STOP : 0));
1510 
1511 		p->seqid = ++seq;
1512 
1513 		ldcdbg(DATA, "SENT DATA [%02x:%02x:%02x:%02x:%08x]\n",
1514 		       p->type,
1515 		       p->stype,
1516 		       p->ctrl,
1517 		       p->env,
1518 		       p->seqid);
1519 
1520 		memcpy(data, buf, data_len);
1521 		buf += data_len;
1522 		copied += data_len;
1523 
1524 		tail = tx_advance(lp, tail);
1525 	}
1526 
1527 	err = set_tx_tail(lp, tail);
1528 	if (!err) {
1529 		lp->snd_nxt = seq;
1530 		err = size;
1531 	}
1532 
1533 	return err;
1534 }
1535 
1536 static int rx_bad_seq(struct ldc_channel *lp, struct ldc_packet *p,
1537 		      struct ldc_packet *first_frag)
1538 {
1539 	int err;
1540 
1541 	if (first_frag)
1542 		lp->rcv_nxt = first_frag->seqid - 1;
1543 
1544 	err = send_data_nack(lp, p);
1545 	if (err)
1546 		return err;
1547 
1548 	err = __set_rx_head(lp, lp->rx_tail);
1549 	if (err < 0)
1550 		return ldc_abort(lp);
1551 
1552 	return 0;
1553 }
1554 
1555 static int data_ack_nack(struct ldc_channel *lp, struct ldc_packet *p)
1556 {
1557 	if (p->stype & LDC_ACK) {
1558 		int err = process_data_ack(lp, p);
1559 		if (err)
1560 			return err;
1561 	}
1562 	if (p->stype & LDC_NACK)
1563 		return ldc_abort(lp);
1564 
1565 	return 0;
1566 }
1567 
1568 static int rx_data_wait(struct ldc_channel *lp, unsigned long cur_head)
1569 {
1570 	unsigned long dummy;
1571 	int limit = 1000;
1572 
1573 	ldcdbg(DATA, "DATA WAIT cur_head[%lx] rx_head[%lx] rx_tail[%lx]\n",
1574 	       cur_head, lp->rx_head, lp->rx_tail);
1575 	while (limit-- > 0) {
1576 		unsigned long hv_err;
1577 
1578 		hv_err = sun4v_ldc_rx_get_state(lp->id,
1579 						&dummy,
1580 						&lp->rx_tail,
1581 						&lp->chan_state);
1582 		if (hv_err)
1583 			return ldc_abort(lp);
1584 
1585 		if (lp->chan_state == LDC_CHANNEL_DOWN ||
1586 		    lp->chan_state == LDC_CHANNEL_RESETTING)
1587 			return -ECONNRESET;
1588 
1589 		if (cur_head != lp->rx_tail) {
1590 			ldcdbg(DATA, "DATA WAIT DONE "
1591 			       "head[%lx] tail[%lx] chan_state[%lx]\n",
1592 			       dummy, lp->rx_tail, lp->chan_state);
1593 			return 0;
1594 		}
1595 
1596 		udelay(1);
1597 	}
1598 	return -EAGAIN;
1599 }
1600 
1601 static int rx_set_head(struct ldc_channel *lp, unsigned long head)
1602 {
1603 	int err = __set_rx_head(lp, head);
1604 
1605 	if (err < 0)
1606 		return ldc_abort(lp);
1607 
1608 	lp->rx_head = head;
1609 	return 0;
1610 }
1611 
1612 static void send_data_ack(struct ldc_channel *lp)
1613 {
1614 	unsigned long new_tail;
1615 	struct ldc_packet *p;
1616 
1617 	p = data_get_tx_packet(lp, &new_tail);
1618 	if (likely(p)) {
1619 		int err;
1620 
1621 		memset(p, 0, sizeof(*p));
1622 		p->type = LDC_DATA;
1623 		p->stype = LDC_ACK;
1624 		p->ctrl = 0;
1625 		p->seqid = lp->snd_nxt + 1;
1626 		p->u.r.ackid = lp->rcv_nxt;
1627 
1628 		err = send_tx_packet(lp, p, new_tail);
1629 		if (!err)
1630 			lp->snd_nxt++;
1631 	}
1632 }
1633 
1634 static int read_nonraw(struct ldc_channel *lp, void *buf, unsigned int size)
1635 {
1636 	struct ldc_packet *first_frag;
1637 	unsigned long hv_err, new;
1638 	int err, copied;
1639 
1640 	hv_err = sun4v_ldc_rx_get_state(lp->id,
1641 					&lp->rx_head,
1642 					&lp->rx_tail,
1643 					&lp->chan_state);
1644 	if (hv_err)
1645 		return ldc_abort(lp);
1646 
1647 	if (lp->chan_state == LDC_CHANNEL_DOWN ||
1648 	    lp->chan_state == LDC_CHANNEL_RESETTING)
1649 		return -ECONNRESET;
1650 
1651 	if (lp->rx_head == lp->rx_tail)
1652 		return 0;
1653 
1654 	first_frag = NULL;
1655 	copied = err = 0;
1656 	new = lp->rx_head;
1657 	while (1) {
1658 		struct ldc_packet *p;
1659 		int pkt_len;
1660 
1661 		BUG_ON(new == lp->rx_tail);
1662 		p = lp->rx_base + (new / LDC_PACKET_SIZE);
1663 
1664 		ldcdbg(RX, "RX read pkt[%02x:%02x:%02x:%02x:%08x:%08x] "
1665 		       "rcv_nxt[%08x]\n",
1666 		       p->type,
1667 		       p->stype,
1668 		       p->ctrl,
1669 		       p->env,
1670 		       p->seqid,
1671 		       p->u.r.ackid,
1672 		       lp->rcv_nxt);
1673 
1674 		if (unlikely(!rx_seq_ok(lp, p->seqid))) {
1675 			err = rx_bad_seq(lp, p, first_frag);
1676 			copied = 0;
1677 			break;
1678 		}
1679 
1680 		if (p->type & LDC_CTRL) {
1681 			err = process_control_frame(lp, p);
1682 			if (err < 0)
1683 				break;
1684 			err = 0;
1685 		}
1686 
1687 		lp->rcv_nxt = p->seqid;
1688 
1689 		if (!(p->type & LDC_DATA)) {
1690 			new = rx_advance(lp, new);
1691 			goto no_data;
1692 		}
1693 		if (p->stype & (LDC_ACK | LDC_NACK)) {
1694 			err = data_ack_nack(lp, p);
1695 			if (err)
1696 				break;
1697 		}
1698 		if (!(p->stype & LDC_INFO)) {
1699 			new = rx_advance(lp, new);
1700 			err = rx_set_head(lp, new);
1701 			if (err)
1702 				break;
1703 			goto no_data;
1704 		}
1705 
1706 		pkt_len = p->env & LDC_LEN;
1707 
1708 		/* Every initial packet starts with the START bit set.
1709 		 *
1710 		 * Singleton packets will have both START+STOP set.
1711 		 *
1712 		 * Fragments will have START set in the first frame, STOP
1713 		 * set in the last frame, and neither bit set in middle
1714 		 * frames of the packet.
1715 		 *
1716 		 * Therefore if we are at the beginning of a packet and
1717 		 * we don't see START, or we are in the middle of a fragmented
1718 		 * packet and do see START, we are unsynchronized and should
1719 		 * flush the RX queue.
1720 		 */
1721 		if ((first_frag == NULL && !(p->env & LDC_START)) ||
1722 		    (first_frag != NULL &&  (p->env & LDC_START))) {
1723 			if (!first_frag)
1724 				new = rx_advance(lp, new);
1725 
1726 			err = rx_set_head(lp, new);
1727 			if (err)
1728 				break;
1729 
1730 			if (!first_frag)
1731 				goto no_data;
1732 		}
1733 		if (!first_frag)
1734 			first_frag = p;
1735 
1736 		if (pkt_len > size - copied) {
1737 			/* User didn't give us a big enough buffer,
1738 			 * what to do?  This is a pretty serious error.
1739 			 *
1740 			 * Since we haven't updated the RX ring head to
1741 			 * consume any of the packets, signal the error
1742 			 * to the user and just leave the RX ring alone.
1743 			 *
1744 			 * This seems the best behavior because this allows
1745 			 * a user of the LDC layer to start with a small
1746 			 * RX buffer for ldc_read() calls and use -EMSGSIZE
1747 			 * as a cue to enlarge it's read buffer.
1748 			 */
1749 			err = -EMSGSIZE;
1750 			break;
1751 		}
1752 
1753 		/* Ok, we are gonna eat this one.  */
1754 		new = rx_advance(lp, new);
1755 
1756 		memcpy(buf,
1757 		       (lp->cfg.mode == LDC_MODE_UNRELIABLE ?
1758 			p->u.u_data : p->u.r.r_data), pkt_len);
1759 		buf += pkt_len;
1760 		copied += pkt_len;
1761 
1762 		if (p->env & LDC_STOP)
1763 			break;
1764 
1765 no_data:
1766 		if (new == lp->rx_tail) {
1767 			err = rx_data_wait(lp, new);
1768 			if (err)
1769 				break;
1770 		}
1771 	}
1772 
1773 	if (!err)
1774 		err = rx_set_head(lp, new);
1775 
1776 	if (err && first_frag)
1777 		lp->rcv_nxt = first_frag->seqid - 1;
1778 
1779 	if (!err) {
1780 		err = copied;
1781 		if (err > 0 && lp->cfg.mode != LDC_MODE_UNRELIABLE)
1782 			send_data_ack(lp);
1783 	}
1784 
1785 	return err;
1786 }
1787 
1788 static const struct ldc_mode_ops nonraw_ops = {
1789 	.write		=	write_nonraw,
1790 	.read		=	read_nonraw,
1791 };
1792 
1793 static int write_stream(struct ldc_channel *lp, const void *buf,
1794 			unsigned int size)
1795 {
1796 	if (size > lp->cfg.mtu)
1797 		size = lp->cfg.mtu;
1798 	return write_nonraw(lp, buf, size);
1799 }
1800 
1801 static int read_stream(struct ldc_channel *lp, void *buf, unsigned int size)
1802 {
1803 	if (!lp->mssbuf_len) {
1804 		int err = read_nonraw(lp, lp->mssbuf, lp->cfg.mtu);
1805 		if (err < 0)
1806 			return err;
1807 
1808 		lp->mssbuf_len = err;
1809 		lp->mssbuf_off = 0;
1810 	}
1811 
1812 	if (size > lp->mssbuf_len)
1813 		size = lp->mssbuf_len;
1814 	memcpy(buf, lp->mssbuf + lp->mssbuf_off, size);
1815 
1816 	lp->mssbuf_off += size;
1817 	lp->mssbuf_len -= size;
1818 
1819 	return size;
1820 }
1821 
1822 static const struct ldc_mode_ops stream_ops = {
1823 	.write		=	write_stream,
1824 	.read		=	read_stream,
1825 };
1826 
1827 int ldc_write(struct ldc_channel *lp, const void *buf, unsigned int size)
1828 {
1829 	unsigned long flags;
1830 	int err;
1831 
1832 	if (!buf)
1833 		return -EINVAL;
1834 
1835 	if (!size)
1836 		return 0;
1837 
1838 	spin_lock_irqsave(&lp->lock, flags);
1839 
1840 	if (lp->hs_state != LDC_HS_COMPLETE)
1841 		err = -ENOTCONN;
1842 	else
1843 		err = lp->mops->write(lp, buf, size);
1844 
1845 	spin_unlock_irqrestore(&lp->lock, flags);
1846 
1847 	return err;
1848 }
1849 EXPORT_SYMBOL(ldc_write);
1850 
1851 int ldc_read(struct ldc_channel *lp, void *buf, unsigned int size)
1852 {
1853 	unsigned long flags;
1854 	int err;
1855 
1856 	if (!buf)
1857 		return -EINVAL;
1858 
1859 	if (!size)
1860 		return 0;
1861 
1862 	spin_lock_irqsave(&lp->lock, flags);
1863 
1864 	if (lp->hs_state != LDC_HS_COMPLETE)
1865 		err = -ENOTCONN;
1866 	else
1867 		err = lp->mops->read(lp, buf, size);
1868 
1869 	spin_unlock_irqrestore(&lp->lock, flags);
1870 
1871 	return err;
1872 }
1873 EXPORT_SYMBOL(ldc_read);
1874 
1875 static long arena_alloc(struct ldc_iommu *iommu, unsigned long npages)
1876 {
1877 	struct iommu_arena *arena = &iommu->arena;
1878 	unsigned long n, i, start, end, limit;
1879 	int pass;
1880 
1881 	limit = arena->limit;
1882 	start = arena->hint;
1883 	pass = 0;
1884 
1885 again:
1886 	n = find_next_zero_bit(arena->map, limit, start);
1887 	end = n + npages;
1888 	if (unlikely(end >= limit)) {
1889 		if (likely(pass < 1)) {
1890 			limit = start;
1891 			start = 0;
1892 			pass++;
1893 			goto again;
1894 		} else {
1895 			/* Scanned the whole thing, give up. */
1896 			return -1;
1897 		}
1898 	}
1899 
1900 	for (i = n; i < end; i++) {
1901 		if (test_bit(i, arena->map)) {
1902 			start = i + 1;
1903 			goto again;
1904 		}
1905 	}
1906 
1907 	for (i = n; i < end; i++)
1908 		__set_bit(i, arena->map);
1909 
1910 	arena->hint = end;
1911 
1912 	return n;
1913 }
1914 
1915 #define COOKIE_PGSZ_CODE	0xf000000000000000ULL
1916 #define COOKIE_PGSZ_CODE_SHIFT	60ULL
1917 
1918 static u64 pagesize_code(void)
1919 {
1920 	switch (PAGE_SIZE) {
1921 	default:
1922 	case (8ULL * 1024ULL):
1923 		return 0;
1924 	case (64ULL * 1024ULL):
1925 		return 1;
1926 	case (512ULL * 1024ULL):
1927 		return 2;
1928 	case (4ULL * 1024ULL * 1024ULL):
1929 		return 3;
1930 	case (32ULL * 1024ULL * 1024ULL):
1931 		return 4;
1932 	case (256ULL * 1024ULL * 1024ULL):
1933 		return 5;
1934 	}
1935 }
1936 
1937 static u64 make_cookie(u64 index, u64 pgsz_code, u64 page_offset)
1938 {
1939 	return ((pgsz_code << COOKIE_PGSZ_CODE_SHIFT) |
1940 		(index << PAGE_SHIFT) |
1941 		page_offset);
1942 }
1943 
1944 static u64 cookie_to_index(u64 cookie, unsigned long *shift)
1945 {
1946 	u64 szcode = cookie >> COOKIE_PGSZ_CODE_SHIFT;
1947 
1948 	cookie &= ~COOKIE_PGSZ_CODE;
1949 
1950 	*shift = szcode * 3;
1951 
1952 	return (cookie >> (13ULL + (szcode * 3ULL)));
1953 }
1954 
1955 static struct ldc_mtable_entry *alloc_npages(struct ldc_iommu *iommu,
1956 					     unsigned long npages)
1957 {
1958 	long entry;
1959 
1960 	entry = arena_alloc(iommu, npages);
1961 	if (unlikely(entry < 0))
1962 		return NULL;
1963 
1964 	return iommu->page_table + entry;
1965 }
1966 
1967 static u64 perm_to_mte(unsigned int map_perm)
1968 {
1969 	u64 mte_base;
1970 
1971 	mte_base = pagesize_code();
1972 
1973 	if (map_perm & LDC_MAP_SHADOW) {
1974 		if (map_perm & LDC_MAP_R)
1975 			mte_base |= LDC_MTE_COPY_R;
1976 		if (map_perm & LDC_MAP_W)
1977 			mte_base |= LDC_MTE_COPY_W;
1978 	}
1979 	if (map_perm & LDC_MAP_DIRECT) {
1980 		if (map_perm & LDC_MAP_R)
1981 			mte_base |= LDC_MTE_READ;
1982 		if (map_perm & LDC_MAP_W)
1983 			mte_base |= LDC_MTE_WRITE;
1984 		if (map_perm & LDC_MAP_X)
1985 			mte_base |= LDC_MTE_EXEC;
1986 	}
1987 	if (map_perm & LDC_MAP_IO) {
1988 		if (map_perm & LDC_MAP_R)
1989 			mte_base |= LDC_MTE_IOMMU_R;
1990 		if (map_perm & LDC_MAP_W)
1991 			mte_base |= LDC_MTE_IOMMU_W;
1992 	}
1993 
1994 	return mte_base;
1995 }
1996 
1997 static int pages_in_region(unsigned long base, long len)
1998 {
1999 	int count = 0;
2000 
2001 	do {
2002 		unsigned long new = (base + PAGE_SIZE) & PAGE_MASK;
2003 
2004 		len -= (new - base);
2005 		base = new;
2006 		count++;
2007 	} while (len > 0);
2008 
2009 	return count;
2010 }
2011 
2012 struct cookie_state {
2013 	struct ldc_mtable_entry		*page_table;
2014 	struct ldc_trans_cookie		*cookies;
2015 	u64				mte_base;
2016 	u64				prev_cookie;
2017 	u32				pte_idx;
2018 	u32				nc;
2019 };
2020 
2021 static void fill_cookies(struct cookie_state *sp, unsigned long pa,
2022 			 unsigned long off, unsigned long len)
2023 {
2024 	do {
2025 		unsigned long tlen, new = pa + PAGE_SIZE;
2026 		u64 this_cookie;
2027 
2028 		sp->page_table[sp->pte_idx].mte = sp->mte_base | pa;
2029 
2030 		tlen = PAGE_SIZE;
2031 		if (off)
2032 			tlen = PAGE_SIZE - off;
2033 		if (tlen > len)
2034 			tlen = len;
2035 
2036 		this_cookie = make_cookie(sp->pte_idx,
2037 					  pagesize_code(), off);
2038 
2039 		off = 0;
2040 
2041 		if (this_cookie == sp->prev_cookie) {
2042 			sp->cookies[sp->nc - 1].cookie_size += tlen;
2043 		} else {
2044 			sp->cookies[sp->nc].cookie_addr = this_cookie;
2045 			sp->cookies[sp->nc].cookie_size = tlen;
2046 			sp->nc++;
2047 		}
2048 		sp->prev_cookie = this_cookie + tlen;
2049 
2050 		sp->pte_idx++;
2051 
2052 		len -= tlen;
2053 		pa = new;
2054 	} while (len > 0);
2055 }
2056 
2057 static int sg_count_one(struct scatterlist *sg)
2058 {
2059 	unsigned long base = page_to_pfn(sg_page(sg)) << PAGE_SHIFT;
2060 	long len = sg->length;
2061 
2062 	if ((sg->offset | len) & (8UL - 1))
2063 		return -EFAULT;
2064 
2065 	return pages_in_region(base + sg->offset, len);
2066 }
2067 
2068 static int sg_count_pages(struct scatterlist *sg, int num_sg)
2069 {
2070 	int count;
2071 	int i;
2072 
2073 	count = 0;
2074 	for (i = 0; i < num_sg; i++) {
2075 		int err = sg_count_one(sg + i);
2076 		if (err < 0)
2077 			return err;
2078 		count += err;
2079 	}
2080 
2081 	return count;
2082 }
2083 
2084 int ldc_map_sg(struct ldc_channel *lp,
2085 	       struct scatterlist *sg, int num_sg,
2086 	       struct ldc_trans_cookie *cookies, int ncookies,
2087 	       unsigned int map_perm)
2088 {
2089 	unsigned long i, npages, flags;
2090 	struct ldc_mtable_entry *base;
2091 	struct cookie_state state;
2092 	struct ldc_iommu *iommu;
2093 	int err;
2094 
2095 	if (map_perm & ~LDC_MAP_ALL)
2096 		return -EINVAL;
2097 
2098 	err = sg_count_pages(sg, num_sg);
2099 	if (err < 0)
2100 		return err;
2101 
2102 	npages = err;
2103 	if (err > ncookies)
2104 		return -EMSGSIZE;
2105 
2106 	iommu = &lp->iommu;
2107 
2108 	spin_lock_irqsave(&iommu->lock, flags);
2109 	base = alloc_npages(iommu, npages);
2110 	spin_unlock_irqrestore(&iommu->lock, flags);
2111 
2112 	if (!base)
2113 		return -ENOMEM;
2114 
2115 	state.page_table = iommu->page_table;
2116 	state.cookies = cookies;
2117 	state.mte_base = perm_to_mte(map_perm);
2118 	state.prev_cookie = ~(u64)0;
2119 	state.pte_idx = (base - iommu->page_table);
2120 	state.nc = 0;
2121 
2122 	for (i = 0; i < num_sg; i++)
2123 		fill_cookies(&state, page_to_pfn(sg_page(&sg[i])) << PAGE_SHIFT,
2124 			     sg[i].offset, sg[i].length);
2125 
2126 	return state.nc;
2127 }
2128 EXPORT_SYMBOL(ldc_map_sg);
2129 
2130 int ldc_map_single(struct ldc_channel *lp,
2131 		   void *buf, unsigned int len,
2132 		   struct ldc_trans_cookie *cookies, int ncookies,
2133 		   unsigned int map_perm)
2134 {
2135 	unsigned long npages, pa, flags;
2136 	struct ldc_mtable_entry *base;
2137 	struct cookie_state state;
2138 	struct ldc_iommu *iommu;
2139 
2140 	if ((map_perm & ~LDC_MAP_ALL) || (ncookies < 1))
2141 		return -EINVAL;
2142 
2143 	pa = __pa(buf);
2144 	if ((pa | len) & (8UL - 1))
2145 		return -EFAULT;
2146 
2147 	npages = pages_in_region(pa, len);
2148 
2149 	iommu = &lp->iommu;
2150 
2151 	spin_lock_irqsave(&iommu->lock, flags);
2152 	base = alloc_npages(iommu, npages);
2153 	spin_unlock_irqrestore(&iommu->lock, flags);
2154 
2155 	if (!base)
2156 		return -ENOMEM;
2157 
2158 	state.page_table = iommu->page_table;
2159 	state.cookies = cookies;
2160 	state.mte_base = perm_to_mte(map_perm);
2161 	state.prev_cookie = ~(u64)0;
2162 	state.pte_idx = (base - iommu->page_table);
2163 	state.nc = 0;
2164 	fill_cookies(&state, (pa & PAGE_MASK), (pa & ~PAGE_MASK), len);
2165 	BUG_ON(state.nc != 1);
2166 
2167 	return state.nc;
2168 }
2169 EXPORT_SYMBOL(ldc_map_single);
2170 
2171 static void free_npages(unsigned long id, struct ldc_iommu *iommu,
2172 			u64 cookie, u64 size)
2173 {
2174 	struct iommu_arena *arena = &iommu->arena;
2175 	unsigned long i, shift, index, npages;
2176 	struct ldc_mtable_entry *base;
2177 
2178 	npages = PAGE_ALIGN(((cookie & ~PAGE_MASK) + size)) >> PAGE_SHIFT;
2179 	index = cookie_to_index(cookie, &shift);
2180 	base = iommu->page_table + index;
2181 
2182 	BUG_ON(index > arena->limit ||
2183 	       (index + npages) > arena->limit);
2184 
2185 	for (i = 0; i < npages; i++) {
2186 		if (base->cookie)
2187 			sun4v_ldc_revoke(id, cookie + (i << shift),
2188 					 base->cookie);
2189 		base->mte = 0;
2190 		__clear_bit(index + i, arena->map);
2191 	}
2192 }
2193 
2194 void ldc_unmap(struct ldc_channel *lp, struct ldc_trans_cookie *cookies,
2195 	       int ncookies)
2196 {
2197 	struct ldc_iommu *iommu = &lp->iommu;
2198 	unsigned long flags;
2199 	int i;
2200 
2201 	spin_lock_irqsave(&iommu->lock, flags);
2202 	for (i = 0; i < ncookies; i++) {
2203 		u64 addr = cookies[i].cookie_addr;
2204 		u64 size = cookies[i].cookie_size;
2205 
2206 		free_npages(lp->id, iommu, addr, size);
2207 	}
2208 	spin_unlock_irqrestore(&iommu->lock, flags);
2209 }
2210 EXPORT_SYMBOL(ldc_unmap);
2211 
2212 int ldc_copy(struct ldc_channel *lp, int copy_dir,
2213 	     void *buf, unsigned int len, unsigned long offset,
2214 	     struct ldc_trans_cookie *cookies, int ncookies)
2215 {
2216 	unsigned int orig_len;
2217 	unsigned long ra;
2218 	int i;
2219 
2220 	if (copy_dir != LDC_COPY_IN && copy_dir != LDC_COPY_OUT) {
2221 		printk(KERN_ERR PFX "ldc_copy: ID[%lu] Bad copy_dir[%d]\n",
2222 		       lp->id, copy_dir);
2223 		return -EINVAL;
2224 	}
2225 
2226 	ra = __pa(buf);
2227 	if ((ra | len | offset) & (8UL - 1)) {
2228 		printk(KERN_ERR PFX "ldc_copy: ID[%lu] Unaligned buffer "
2229 		       "ra[%lx] len[%x] offset[%lx]\n",
2230 		       lp->id, ra, len, offset);
2231 		return -EFAULT;
2232 	}
2233 
2234 	if (lp->hs_state != LDC_HS_COMPLETE ||
2235 	    (lp->flags & LDC_FLAG_RESET)) {
2236 		printk(KERN_ERR PFX "ldc_copy: ID[%lu] Link down hs_state[%x] "
2237 		       "flags[%x]\n", lp->id, lp->hs_state, lp->flags);
2238 		return -ECONNRESET;
2239 	}
2240 
2241 	orig_len = len;
2242 	for (i = 0; i < ncookies; i++) {
2243 		unsigned long cookie_raddr = cookies[i].cookie_addr;
2244 		unsigned long this_len = cookies[i].cookie_size;
2245 		unsigned long actual_len;
2246 
2247 		if (unlikely(offset)) {
2248 			unsigned long this_off = offset;
2249 
2250 			if (this_off > this_len)
2251 				this_off = this_len;
2252 
2253 			offset -= this_off;
2254 			this_len -= this_off;
2255 			if (!this_len)
2256 				continue;
2257 			cookie_raddr += this_off;
2258 		}
2259 
2260 		if (this_len > len)
2261 			this_len = len;
2262 
2263 		while (1) {
2264 			unsigned long hv_err;
2265 
2266 			hv_err = sun4v_ldc_copy(lp->id, copy_dir,
2267 						cookie_raddr, ra,
2268 						this_len, &actual_len);
2269 			if (unlikely(hv_err)) {
2270 				printk(KERN_ERR PFX "ldc_copy: ID[%lu] "
2271 				       "HV error %lu\n",
2272 				       lp->id, hv_err);
2273 				if (lp->hs_state != LDC_HS_COMPLETE ||
2274 				    (lp->flags & LDC_FLAG_RESET))
2275 					return -ECONNRESET;
2276 				else
2277 					return -EFAULT;
2278 			}
2279 
2280 			cookie_raddr += actual_len;
2281 			ra += actual_len;
2282 			len -= actual_len;
2283 			if (actual_len == this_len)
2284 				break;
2285 
2286 			this_len -= actual_len;
2287 		}
2288 
2289 		if (!len)
2290 			break;
2291 	}
2292 
2293 	/* It is caller policy what to do about short copies.
2294 	 * For example, a networking driver can declare the
2295 	 * packet a runt and drop it.
2296 	 */
2297 
2298 	return orig_len - len;
2299 }
2300 EXPORT_SYMBOL(ldc_copy);
2301 
2302 void *ldc_alloc_exp_dring(struct ldc_channel *lp, unsigned int len,
2303 			  struct ldc_trans_cookie *cookies, int *ncookies,
2304 			  unsigned int map_perm)
2305 {
2306 	void *buf;
2307 	int err;
2308 
2309 	if (len & (8UL - 1))
2310 		return ERR_PTR(-EINVAL);
2311 
2312 	buf = kzalloc(len, GFP_KERNEL);
2313 	if (!buf)
2314 		return ERR_PTR(-ENOMEM);
2315 
2316 	err = ldc_map_single(lp, buf, len, cookies, *ncookies, map_perm);
2317 	if (err < 0) {
2318 		kfree(buf);
2319 		return ERR_PTR(err);
2320 	}
2321 	*ncookies = err;
2322 
2323 	return buf;
2324 }
2325 EXPORT_SYMBOL(ldc_alloc_exp_dring);
2326 
2327 void ldc_free_exp_dring(struct ldc_channel *lp, void *buf, unsigned int len,
2328 			struct ldc_trans_cookie *cookies, int ncookies)
2329 {
2330 	ldc_unmap(lp, cookies, ncookies);
2331 	kfree(buf);
2332 }
2333 EXPORT_SYMBOL(ldc_free_exp_dring);
2334 
2335 static int __init ldc_init(void)
2336 {
2337 	unsigned long major, minor;
2338 	struct mdesc_handle *hp;
2339 	const u64 *v;
2340 	int err;
2341 	u64 mp;
2342 
2343 	hp = mdesc_grab();
2344 	if (!hp)
2345 		return -ENODEV;
2346 
2347 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
2348 	err = -ENODEV;
2349 	if (mp == MDESC_NODE_NULL)
2350 		goto out;
2351 
2352 	v = mdesc_get_property(hp, mp, "domaining-enabled", NULL);
2353 	if (!v)
2354 		goto out;
2355 
2356 	major = 1;
2357 	minor = 0;
2358 	if (sun4v_hvapi_register(HV_GRP_LDOM, major, &minor)) {
2359 		printk(KERN_INFO PFX "Could not register LDOM hvapi.\n");
2360 		goto out;
2361 	}
2362 
2363 	printk(KERN_INFO "%s", version);
2364 
2365 	if (!*v) {
2366 		printk(KERN_INFO PFX "Domaining disabled.\n");
2367 		goto out;
2368 	}
2369 	ldom_domaining_enabled = 1;
2370 	err = 0;
2371 
2372 out:
2373 	mdesc_release(hp);
2374 	return err;
2375 }
2376 
2377 core_initcall(ldc_init);
2378