xref: /openbmc/linux/drivers/s390/cio/qdio_main.c (revision afc98d90)
1 /*
2  * Linux for s390 qdio support, buffer handling, qdio API and module support.
3  *
4  * Copyright IBM Corp. 2000, 2008
5  * Author(s): Utz Bacher <utz.bacher@de.ibm.com>
6  *	      Jan Glauber <jang@linux.vnet.ibm.com>
7  * 2.6 cio integration by Cornelia Huck <cornelia.huck@de.ibm.com>
8  */
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/kernel.h>
12 #include <linux/timer.h>
13 #include <linux/delay.h>
14 #include <linux/gfp.h>
15 #include <linux/io.h>
16 #include <linux/atomic.h>
17 #include <asm/debug.h>
18 #include <asm/qdio.h>
19 #include <asm/ipl.h>
20 
21 #include "cio.h"
22 #include "css.h"
23 #include "device.h"
24 #include "qdio.h"
25 #include "qdio_debug.h"
26 
27 MODULE_AUTHOR("Utz Bacher <utz.bacher@de.ibm.com>,"\
28 	"Jan Glauber <jang@linux.vnet.ibm.com>");
29 MODULE_DESCRIPTION("QDIO base support");
30 MODULE_LICENSE("GPL");
31 
32 static inline int do_siga_sync(unsigned long schid,
33 			       unsigned int out_mask, unsigned int in_mask,
34 			       unsigned int fc)
35 {
36 	register unsigned long __fc asm ("0") = fc;
37 	register unsigned long __schid asm ("1") = schid;
38 	register unsigned long out asm ("2") = out_mask;
39 	register unsigned long in asm ("3") = in_mask;
40 	int cc;
41 
42 	asm volatile(
43 		"	siga	0\n"
44 		"	ipm	%0\n"
45 		"	srl	%0,28\n"
46 		: "=d" (cc)
47 		: "d" (__fc), "d" (__schid), "d" (out), "d" (in) : "cc");
48 	return cc;
49 }
50 
51 static inline int do_siga_input(unsigned long schid, unsigned int mask,
52 				unsigned int fc)
53 {
54 	register unsigned long __fc asm ("0") = fc;
55 	register unsigned long __schid asm ("1") = schid;
56 	register unsigned long __mask asm ("2") = mask;
57 	int cc;
58 
59 	asm volatile(
60 		"	siga	0\n"
61 		"	ipm	%0\n"
62 		"	srl	%0,28\n"
63 		: "=d" (cc)
64 		: "d" (__fc), "d" (__schid), "d" (__mask) : "cc");
65 	return cc;
66 }
67 
68 /**
69  * do_siga_output - perform SIGA-w/wt function
70  * @schid: subchannel id or in case of QEBSM the subchannel token
71  * @mask: which output queues to process
72  * @bb: busy bit indicator, set only if SIGA-w/wt could not access a buffer
73  * @fc: function code to perform
74  *
75  * Returns condition code.
76  * Note: For IQDC unicast queues only the highest priority queue is processed.
77  */
78 static inline int do_siga_output(unsigned long schid, unsigned long mask,
79 				 unsigned int *bb, unsigned int fc,
80 				 unsigned long aob)
81 {
82 	register unsigned long __fc asm("0") = fc;
83 	register unsigned long __schid asm("1") = schid;
84 	register unsigned long __mask asm("2") = mask;
85 	register unsigned long __aob asm("3") = aob;
86 	int cc;
87 
88 	asm volatile(
89 		"	siga	0\n"
90 		"	ipm	%0\n"
91 		"	srl	%0,28\n"
92 		: "=d" (cc), "+d" (__fc), "+d" (__aob)
93 		: "d" (__schid), "d" (__mask)
94 		: "cc");
95 	*bb = __fc >> 31;
96 	return cc;
97 }
98 
99 static inline int qdio_check_ccq(struct qdio_q *q, unsigned int ccq)
100 {
101 	/* all done or next buffer state different */
102 	if (ccq == 0 || ccq == 32)
103 		return 0;
104 	/* no buffer processed */
105 	if (ccq == 97)
106 		return 1;
107 	/* not all buffers processed */
108 	if (ccq == 96)
109 		return 2;
110 	/* notify devices immediately */
111 	DBF_ERROR("%4x ccq:%3d", SCH_NO(q), ccq);
112 	return -EIO;
113 }
114 
115 /**
116  * qdio_do_eqbs - extract buffer states for QEBSM
117  * @q: queue to manipulate
118  * @state: state of the extracted buffers
119  * @start: buffer number to start at
120  * @count: count of buffers to examine
121  * @auto_ack: automatically acknowledge buffers
122  *
123  * Returns the number of successfully extracted equal buffer states.
124  * Stops processing if a state is different from the last buffers state.
125  */
126 static int qdio_do_eqbs(struct qdio_q *q, unsigned char *state,
127 			int start, int count, int auto_ack)
128 {
129 	int rc, tmp_count = count, tmp_start = start, nr = q->nr, retried = 0;
130 	unsigned int ccq = 0;
131 
132 	qperf_inc(q, eqbs);
133 
134 	if (!q->is_input_q)
135 		nr += q->irq_ptr->nr_input_qs;
136 again:
137 	ccq = do_eqbs(q->irq_ptr->sch_token, state, nr, &tmp_start, &tmp_count,
138 		      auto_ack);
139 	rc = qdio_check_ccq(q, ccq);
140 	if (!rc)
141 		return count - tmp_count;
142 
143 	if (rc == 1) {
144 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "EQBS again:%2d", ccq);
145 		goto again;
146 	}
147 
148 	if (rc == 2) {
149 		qperf_inc(q, eqbs_partial);
150 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "EQBS part:%02x",
151 			tmp_count);
152 		/*
153 		 * Retry once, if that fails bail out and process the
154 		 * extracted buffers before trying again.
155 		 */
156 		if (!retried++)
157 			goto again;
158 		else
159 			return count - tmp_count;
160 	}
161 
162 	DBF_ERROR("%4x EQBS ERROR", SCH_NO(q));
163 	DBF_ERROR("%3d%3d%2d", count, tmp_count, nr);
164 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_GET_BUF_STATE,
165 		   q->nr, q->first_to_kick, count, q->irq_ptr->int_parm);
166 	return 0;
167 }
168 
169 /**
170  * qdio_do_sqbs - set buffer states for QEBSM
171  * @q: queue to manipulate
172  * @state: new state of the buffers
173  * @start: first buffer number to change
174  * @count: how many buffers to change
175  *
176  * Returns the number of successfully changed buffers.
177  * Does retrying until the specified count of buffer states is set or an
178  * error occurs.
179  */
180 static int qdio_do_sqbs(struct qdio_q *q, unsigned char state, int start,
181 			int count)
182 {
183 	unsigned int ccq = 0;
184 	int tmp_count = count, tmp_start = start;
185 	int nr = q->nr;
186 	int rc;
187 
188 	if (!count)
189 		return 0;
190 	qperf_inc(q, sqbs);
191 
192 	if (!q->is_input_q)
193 		nr += q->irq_ptr->nr_input_qs;
194 again:
195 	ccq = do_sqbs(q->irq_ptr->sch_token, state, nr, &tmp_start, &tmp_count);
196 	rc = qdio_check_ccq(q, ccq);
197 	if (!rc) {
198 		WARN_ON_ONCE(tmp_count);
199 		return count - tmp_count;
200 	}
201 
202 	if (rc == 1 || rc == 2) {
203 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "SQBS again:%2d", ccq);
204 		qperf_inc(q, sqbs_partial);
205 		goto again;
206 	}
207 
208 	DBF_ERROR("%4x SQBS ERROR", SCH_NO(q));
209 	DBF_ERROR("%3d%3d%2d", count, tmp_count, nr);
210 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_SET_BUF_STATE,
211 		   q->nr, q->first_to_kick, count, q->irq_ptr->int_parm);
212 	return 0;
213 }
214 
215 /* returns number of examined buffers and their common state in *state */
216 static inline int get_buf_states(struct qdio_q *q, unsigned int bufnr,
217 				 unsigned char *state, unsigned int count,
218 				 int auto_ack, int merge_pending)
219 {
220 	unsigned char __state = 0;
221 	int i;
222 
223 	if (is_qebsm(q))
224 		return qdio_do_eqbs(q, state, bufnr, count, auto_ack);
225 
226 	for (i = 0; i < count; i++) {
227 		if (!__state) {
228 			__state = q->slsb.val[bufnr];
229 			if (merge_pending && __state == SLSB_P_OUTPUT_PENDING)
230 				__state = SLSB_P_OUTPUT_EMPTY;
231 		} else if (merge_pending) {
232 			if ((q->slsb.val[bufnr] & __state) != __state)
233 				break;
234 		} else if (q->slsb.val[bufnr] != __state)
235 			break;
236 		bufnr = next_buf(bufnr);
237 	}
238 	*state = __state;
239 	return i;
240 }
241 
242 static inline int get_buf_state(struct qdio_q *q, unsigned int bufnr,
243 				unsigned char *state, int auto_ack)
244 {
245 	return get_buf_states(q, bufnr, state, 1, auto_ack, 0);
246 }
247 
248 /* wrap-around safe setting of slsb states, returns number of changed buffers */
249 static inline int set_buf_states(struct qdio_q *q, int bufnr,
250 				 unsigned char state, int count)
251 {
252 	int i;
253 
254 	if (is_qebsm(q))
255 		return qdio_do_sqbs(q, state, bufnr, count);
256 
257 	for (i = 0; i < count; i++) {
258 		xchg(&q->slsb.val[bufnr], state);
259 		bufnr = next_buf(bufnr);
260 	}
261 	return count;
262 }
263 
264 static inline int set_buf_state(struct qdio_q *q, int bufnr,
265 				unsigned char state)
266 {
267 	return set_buf_states(q, bufnr, state, 1);
268 }
269 
270 /* set slsb states to initial state */
271 static void qdio_init_buf_states(struct qdio_irq *irq_ptr)
272 {
273 	struct qdio_q *q;
274 	int i;
275 
276 	for_each_input_queue(irq_ptr, q, i)
277 		set_buf_states(q, 0, SLSB_P_INPUT_NOT_INIT,
278 			       QDIO_MAX_BUFFERS_PER_Q);
279 	for_each_output_queue(irq_ptr, q, i)
280 		set_buf_states(q, 0, SLSB_P_OUTPUT_NOT_INIT,
281 			       QDIO_MAX_BUFFERS_PER_Q);
282 }
283 
284 static inline int qdio_siga_sync(struct qdio_q *q, unsigned int output,
285 			  unsigned int input)
286 {
287 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
288 	unsigned int fc = QDIO_SIGA_SYNC;
289 	int cc;
290 
291 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-s:%1d", q->nr);
292 	qperf_inc(q, siga_sync);
293 
294 	if (is_qebsm(q)) {
295 		schid = q->irq_ptr->sch_token;
296 		fc |= QDIO_SIGA_QEBSM_FLAG;
297 	}
298 
299 	cc = do_siga_sync(schid, output, input, fc);
300 	if (unlikely(cc))
301 		DBF_ERROR("%4x SIGA-S:%2d", SCH_NO(q), cc);
302 	return (cc) ? -EIO : 0;
303 }
304 
305 static inline int qdio_siga_sync_q(struct qdio_q *q)
306 {
307 	if (q->is_input_q)
308 		return qdio_siga_sync(q, 0, q->mask);
309 	else
310 		return qdio_siga_sync(q, q->mask, 0);
311 }
312 
313 static int qdio_siga_output(struct qdio_q *q, unsigned int *busy_bit,
314 	unsigned long aob)
315 {
316 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
317 	unsigned int fc = QDIO_SIGA_WRITE;
318 	u64 start_time = 0;
319 	int retries = 0, cc;
320 	unsigned long laob = 0;
321 
322 	if (q->u.out.use_cq && aob != 0) {
323 		fc = QDIO_SIGA_WRITEQ;
324 		laob = aob;
325 	}
326 
327 	if (is_qebsm(q)) {
328 		schid = q->irq_ptr->sch_token;
329 		fc |= QDIO_SIGA_QEBSM_FLAG;
330 	}
331 again:
332 	WARN_ON_ONCE((aob && queue_type(q) != QDIO_IQDIO_QFMT) ||
333 		(aob && fc != QDIO_SIGA_WRITEQ));
334 	cc = do_siga_output(schid, q->mask, busy_bit, fc, laob);
335 
336 	/* hipersocket busy condition */
337 	if (unlikely(*busy_bit)) {
338 		retries++;
339 
340 		if (!start_time) {
341 			start_time = get_tod_clock_fast();
342 			goto again;
343 		}
344 		if (get_tod_clock_fast() - start_time < QDIO_BUSY_BIT_PATIENCE)
345 			goto again;
346 	}
347 	if (retries) {
348 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr,
349 			      "%4x cc2 BB1:%1d", SCH_NO(q), q->nr);
350 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "count:%u", retries);
351 	}
352 	return cc;
353 }
354 
355 static inline int qdio_siga_input(struct qdio_q *q)
356 {
357 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
358 	unsigned int fc = QDIO_SIGA_READ;
359 	int cc;
360 
361 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-r:%1d", q->nr);
362 	qperf_inc(q, siga_read);
363 
364 	if (is_qebsm(q)) {
365 		schid = q->irq_ptr->sch_token;
366 		fc |= QDIO_SIGA_QEBSM_FLAG;
367 	}
368 
369 	cc = do_siga_input(schid, q->mask, fc);
370 	if (unlikely(cc))
371 		DBF_ERROR("%4x SIGA-R:%2d", SCH_NO(q), cc);
372 	return (cc) ? -EIO : 0;
373 }
374 
375 #define qdio_siga_sync_out(q) qdio_siga_sync(q, ~0U, 0)
376 #define qdio_siga_sync_all(q) qdio_siga_sync(q, ~0U, ~0U)
377 
378 static inline void qdio_sync_queues(struct qdio_q *q)
379 {
380 	/* PCI capable outbound queues will also be scanned so sync them too */
381 	if (pci_out_supported(q))
382 		qdio_siga_sync_all(q);
383 	else
384 		qdio_siga_sync_q(q);
385 }
386 
387 int debug_get_buf_state(struct qdio_q *q, unsigned int bufnr,
388 			unsigned char *state)
389 {
390 	if (need_siga_sync(q))
391 		qdio_siga_sync_q(q);
392 	return get_buf_states(q, bufnr, state, 1, 0, 0);
393 }
394 
395 static inline void qdio_stop_polling(struct qdio_q *q)
396 {
397 	if (!q->u.in.polling)
398 		return;
399 
400 	q->u.in.polling = 0;
401 	qperf_inc(q, stop_polling);
402 
403 	/* show the card that we are not polling anymore */
404 	if (is_qebsm(q)) {
405 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
406 			       q->u.in.ack_count);
407 		q->u.in.ack_count = 0;
408 	} else
409 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
410 }
411 
412 static inline void account_sbals(struct qdio_q *q, int count)
413 {
414 	int pos = 0;
415 
416 	q->q_stats.nr_sbal_total += count;
417 	if (count == QDIO_MAX_BUFFERS_MASK) {
418 		q->q_stats.nr_sbals[7]++;
419 		return;
420 	}
421 	while (count >>= 1)
422 		pos++;
423 	q->q_stats.nr_sbals[pos]++;
424 }
425 
426 static void process_buffer_error(struct qdio_q *q, int count)
427 {
428 	unsigned char state = (q->is_input_q) ? SLSB_P_INPUT_NOT_INIT :
429 					SLSB_P_OUTPUT_NOT_INIT;
430 
431 	q->qdio_error = QDIO_ERROR_SLSB_STATE;
432 
433 	/* special handling for no target buffer empty */
434 	if ((!q->is_input_q &&
435 	    (q->sbal[q->first_to_check]->element[15].sflags) == 0x10)) {
436 		qperf_inc(q, target_full);
437 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "OUTFULL FTC:%02x",
438 			      q->first_to_check);
439 		goto set;
440 	}
441 
442 	DBF_ERROR("%4x BUF ERROR", SCH_NO(q));
443 	DBF_ERROR((q->is_input_q) ? "IN:%2d" : "OUT:%2d", q->nr);
444 	DBF_ERROR("FTC:%3d C:%3d", q->first_to_check, count);
445 	DBF_ERROR("F14:%2x F15:%2x",
446 		  q->sbal[q->first_to_check]->element[14].sflags,
447 		  q->sbal[q->first_to_check]->element[15].sflags);
448 
449 set:
450 	/*
451 	 * Interrupts may be avoided as long as the error is present
452 	 * so change the buffer state immediately to avoid starvation.
453 	 */
454 	set_buf_states(q, q->first_to_check, state, count);
455 }
456 
457 static inline void inbound_primed(struct qdio_q *q, int count)
458 {
459 	int new;
460 
461 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in prim: %02x", count);
462 
463 	/* for QEBSM the ACK was already set by EQBS */
464 	if (is_qebsm(q)) {
465 		if (!q->u.in.polling) {
466 			q->u.in.polling = 1;
467 			q->u.in.ack_count = count;
468 			q->u.in.ack_start = q->first_to_check;
469 			return;
470 		}
471 
472 		/* delete the previous ACK's */
473 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
474 			       q->u.in.ack_count);
475 		q->u.in.ack_count = count;
476 		q->u.in.ack_start = q->first_to_check;
477 		return;
478 	}
479 
480 	/*
481 	 * ACK the newest buffer. The ACK will be removed in qdio_stop_polling
482 	 * or by the next inbound run.
483 	 */
484 	new = add_buf(q->first_to_check, count - 1);
485 	if (q->u.in.polling) {
486 		/* reset the previous ACK but first set the new one */
487 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
488 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
489 	} else {
490 		q->u.in.polling = 1;
491 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
492 	}
493 
494 	q->u.in.ack_start = new;
495 	count--;
496 	if (!count)
497 		return;
498 	/* need to change ALL buffers to get more interrupts */
499 	set_buf_states(q, q->first_to_check, SLSB_P_INPUT_NOT_INIT, count);
500 }
501 
502 static int get_inbound_buffer_frontier(struct qdio_q *q)
503 {
504 	int count, stop;
505 	unsigned char state = 0;
506 
507 	q->timestamp = get_tod_clock_fast();
508 
509 	/*
510 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
511 	 * would return 0.
512 	 */
513 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
514 	stop = add_buf(q->first_to_check, count);
515 
516 	if (q->first_to_check == stop)
517 		goto out;
518 
519 	/*
520 	 * No siga sync here, as a PCI or we after a thin interrupt
521 	 * already sync'ed the queues.
522 	 */
523 	count = get_buf_states(q, q->first_to_check, &state, count, 1, 0);
524 	if (!count)
525 		goto out;
526 
527 	switch (state) {
528 	case SLSB_P_INPUT_PRIMED:
529 		inbound_primed(q, count);
530 		q->first_to_check = add_buf(q->first_to_check, count);
531 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
532 			qperf_inc(q, inbound_queue_full);
533 		if (q->irq_ptr->perf_stat_enabled)
534 			account_sbals(q, count);
535 		break;
536 	case SLSB_P_INPUT_ERROR:
537 		process_buffer_error(q, count);
538 		q->first_to_check = add_buf(q->first_to_check, count);
539 		atomic_sub(count, &q->nr_buf_used);
540 		if (q->irq_ptr->perf_stat_enabled)
541 			account_sbals_error(q, count);
542 		break;
543 	case SLSB_CU_INPUT_EMPTY:
544 	case SLSB_P_INPUT_NOT_INIT:
545 	case SLSB_P_INPUT_ACK:
546 		if (q->irq_ptr->perf_stat_enabled)
547 			q->q_stats.nr_sbal_nop++;
548 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in nop");
549 		break;
550 	default:
551 		WARN_ON_ONCE(1);
552 	}
553 out:
554 	return q->first_to_check;
555 }
556 
557 static int qdio_inbound_q_moved(struct qdio_q *q)
558 {
559 	int bufnr;
560 
561 	bufnr = get_inbound_buffer_frontier(q);
562 
563 	if (bufnr != q->last_move) {
564 		q->last_move = bufnr;
565 		if (!is_thinint_irq(q->irq_ptr) && MACHINE_IS_LPAR)
566 			q->u.in.timestamp = get_tod_clock();
567 		return 1;
568 	} else
569 		return 0;
570 }
571 
572 static inline int qdio_inbound_q_done(struct qdio_q *q)
573 {
574 	unsigned char state = 0;
575 
576 	if (!atomic_read(&q->nr_buf_used))
577 		return 1;
578 
579 	if (need_siga_sync(q))
580 		qdio_siga_sync_q(q);
581 	get_buf_state(q, q->first_to_check, &state, 0);
582 
583 	if (state == SLSB_P_INPUT_PRIMED || state == SLSB_P_INPUT_ERROR)
584 		/* more work coming */
585 		return 0;
586 
587 	if (is_thinint_irq(q->irq_ptr))
588 		return 1;
589 
590 	/* don't poll under z/VM */
591 	if (MACHINE_IS_VM)
592 		return 1;
593 
594 	/*
595 	 * At this point we know, that inbound first_to_check
596 	 * has (probably) not moved (see qdio_inbound_processing).
597 	 */
598 	if (get_tod_clock_fast() > q->u.in.timestamp + QDIO_INPUT_THRESHOLD) {
599 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in done:%02x",
600 			      q->first_to_check);
601 		return 1;
602 	} else
603 		return 0;
604 }
605 
606 static inline int contains_aobs(struct qdio_q *q)
607 {
608 	return !q->is_input_q && q->u.out.use_cq;
609 }
610 
611 static inline void qdio_handle_aobs(struct qdio_q *q, int start, int count)
612 {
613 	unsigned char state = 0;
614 	int j, b = start;
615 
616 	if (!contains_aobs(q))
617 		return;
618 
619 	for (j = 0; j < count; ++j) {
620 		get_buf_state(q, b, &state, 0);
621 		if (state == SLSB_P_OUTPUT_PENDING) {
622 			struct qaob *aob = q->u.out.aobs[b];
623 			if (aob == NULL)
624 				continue;
625 
626 			q->u.out.sbal_state[b].flags |=
627 				QDIO_OUTBUF_STATE_FLAG_PENDING;
628 			q->u.out.aobs[b] = NULL;
629 		} else if (state == SLSB_P_OUTPUT_EMPTY) {
630 			q->u.out.sbal_state[b].aob = NULL;
631 		}
632 		b = next_buf(b);
633 	}
634 }
635 
636 static inline unsigned long qdio_aob_for_buffer(struct qdio_output_q *q,
637 					int bufnr)
638 {
639 	unsigned long phys_aob = 0;
640 
641 	if (!q->use_cq)
642 		goto out;
643 
644 	if (!q->aobs[bufnr]) {
645 		struct qaob *aob = qdio_allocate_aob();
646 		q->aobs[bufnr] = aob;
647 	}
648 	if (q->aobs[bufnr]) {
649 		q->sbal_state[bufnr].flags = QDIO_OUTBUF_STATE_FLAG_NONE;
650 		q->sbal_state[bufnr].aob = q->aobs[bufnr];
651 		q->aobs[bufnr]->user1 = (u64) q->sbal_state[bufnr].user;
652 		phys_aob = virt_to_phys(q->aobs[bufnr]);
653 		WARN_ON_ONCE(phys_aob & 0xFF);
654 	}
655 
656 out:
657 	return phys_aob;
658 }
659 
660 static void qdio_kick_handler(struct qdio_q *q)
661 {
662 	int start = q->first_to_kick;
663 	int end = q->first_to_check;
664 	int count;
665 
666 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
667 		return;
668 
669 	count = sub_buf(end, start);
670 
671 	if (q->is_input_q) {
672 		qperf_inc(q, inbound_handler);
673 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "kih s:%02x c:%02x", start, count);
674 	} else {
675 		qperf_inc(q, outbound_handler);
676 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "koh: s:%02x c:%02x",
677 			      start, count);
678 	}
679 
680 	qdio_handle_aobs(q, start, count);
681 
682 	q->handler(q->irq_ptr->cdev, q->qdio_error, q->nr, start, count,
683 		   q->irq_ptr->int_parm);
684 
685 	/* for the next time */
686 	q->first_to_kick = end;
687 	q->qdio_error = 0;
688 }
689 
690 static void __qdio_inbound_processing(struct qdio_q *q)
691 {
692 	qperf_inc(q, tasklet_inbound);
693 
694 	if (!qdio_inbound_q_moved(q))
695 		return;
696 
697 	qdio_kick_handler(q);
698 
699 	if (!qdio_inbound_q_done(q)) {
700 		/* means poll time is not yet over */
701 		qperf_inc(q, tasklet_inbound_resched);
702 		if (likely(q->irq_ptr->state != QDIO_IRQ_STATE_STOPPED)) {
703 			tasklet_schedule(&q->tasklet);
704 			return;
705 		}
706 	}
707 
708 	qdio_stop_polling(q);
709 	/*
710 	 * We need to check again to not lose initiative after
711 	 * resetting the ACK state.
712 	 */
713 	if (!qdio_inbound_q_done(q)) {
714 		qperf_inc(q, tasklet_inbound_resched2);
715 		if (likely(q->irq_ptr->state != QDIO_IRQ_STATE_STOPPED))
716 			tasklet_schedule(&q->tasklet);
717 	}
718 }
719 
720 void qdio_inbound_processing(unsigned long data)
721 {
722 	struct qdio_q *q = (struct qdio_q *)data;
723 	__qdio_inbound_processing(q);
724 }
725 
726 static int get_outbound_buffer_frontier(struct qdio_q *q)
727 {
728 	int count, stop;
729 	unsigned char state = 0;
730 
731 	q->timestamp = get_tod_clock_fast();
732 
733 	if (need_siga_sync(q))
734 		if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
735 		    !pci_out_supported(q)) ||
736 		    (queue_type(q) == QDIO_IQDIO_QFMT &&
737 		    multicast_outbound(q)))
738 			qdio_siga_sync_q(q);
739 
740 	/*
741 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
742 	 * would return 0.
743 	 */
744 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
745 	stop = add_buf(q->first_to_check, count);
746 	if (q->first_to_check == stop)
747 		goto out;
748 
749 	count = get_buf_states(q, q->first_to_check, &state, count, 0, 1);
750 	if (!count)
751 		goto out;
752 
753 	switch (state) {
754 	case SLSB_P_OUTPUT_EMPTY:
755 		/* the adapter got it */
756 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr,
757 			"out empty:%1d %02x", q->nr, count);
758 
759 		atomic_sub(count, &q->nr_buf_used);
760 		q->first_to_check = add_buf(q->first_to_check, count);
761 		if (q->irq_ptr->perf_stat_enabled)
762 			account_sbals(q, count);
763 
764 		break;
765 	case SLSB_P_OUTPUT_ERROR:
766 		process_buffer_error(q, count);
767 		q->first_to_check = add_buf(q->first_to_check, count);
768 		atomic_sub(count, &q->nr_buf_used);
769 		if (q->irq_ptr->perf_stat_enabled)
770 			account_sbals_error(q, count);
771 		break;
772 	case SLSB_CU_OUTPUT_PRIMED:
773 		/* the adapter has not fetched the output yet */
774 		if (q->irq_ptr->perf_stat_enabled)
775 			q->q_stats.nr_sbal_nop++;
776 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out primed:%1d",
777 			      q->nr);
778 		break;
779 	case SLSB_P_OUTPUT_NOT_INIT:
780 	case SLSB_P_OUTPUT_HALTED:
781 		break;
782 	default:
783 		WARN_ON_ONCE(1);
784 	}
785 
786 out:
787 	return q->first_to_check;
788 }
789 
790 /* all buffers processed? */
791 static inline int qdio_outbound_q_done(struct qdio_q *q)
792 {
793 	return atomic_read(&q->nr_buf_used) == 0;
794 }
795 
796 static inline int qdio_outbound_q_moved(struct qdio_q *q)
797 {
798 	int bufnr;
799 
800 	bufnr = get_outbound_buffer_frontier(q);
801 
802 	if (bufnr != q->last_move) {
803 		q->last_move = bufnr;
804 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out moved:%1d", q->nr);
805 		return 1;
806 	} else
807 		return 0;
808 }
809 
810 static int qdio_kick_outbound_q(struct qdio_q *q, unsigned long aob)
811 {
812 	int retries = 0, cc;
813 	unsigned int busy_bit;
814 
815 	if (!need_siga_out(q))
816 		return 0;
817 
818 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w:%1d", q->nr);
819 retry:
820 	qperf_inc(q, siga_write);
821 
822 	cc = qdio_siga_output(q, &busy_bit, aob);
823 	switch (cc) {
824 	case 0:
825 		break;
826 	case 2:
827 		if (busy_bit) {
828 			while (++retries < QDIO_BUSY_BIT_RETRIES) {
829 				mdelay(QDIO_BUSY_BIT_RETRY_DELAY);
830 				goto retry;
831 			}
832 			DBF_ERROR("%4x cc2 BBC:%1d", SCH_NO(q), q->nr);
833 			cc = -EBUSY;
834 		} else {
835 			DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w cc2:%1d", q->nr);
836 			cc = -ENOBUFS;
837 		}
838 		break;
839 	case 1:
840 	case 3:
841 		DBF_ERROR("%4x SIGA-W:%1d", SCH_NO(q), cc);
842 		cc = -EIO;
843 		break;
844 	}
845 	if (retries) {
846 		DBF_ERROR("%4x cc2 BB2:%1d", SCH_NO(q), q->nr);
847 		DBF_ERROR("count:%u", retries);
848 	}
849 	return cc;
850 }
851 
852 static void __qdio_outbound_processing(struct qdio_q *q)
853 {
854 	qperf_inc(q, tasklet_outbound);
855 	WARN_ON_ONCE(atomic_read(&q->nr_buf_used) < 0);
856 
857 	if (qdio_outbound_q_moved(q))
858 		qdio_kick_handler(q);
859 
860 	if (queue_type(q) == QDIO_ZFCP_QFMT)
861 		if (!pci_out_supported(q) && !qdio_outbound_q_done(q))
862 			goto sched;
863 
864 	if (q->u.out.pci_out_enabled)
865 		return;
866 
867 	/*
868 	 * Now we know that queue type is either qeth without pci enabled
869 	 * or HiperSockets. Make sure buffer switch from PRIMED to EMPTY
870 	 * is noticed and outbound_handler is called after some time.
871 	 */
872 	if (qdio_outbound_q_done(q))
873 		del_timer(&q->u.out.timer);
874 	else
875 		if (!timer_pending(&q->u.out.timer))
876 			mod_timer(&q->u.out.timer, jiffies + 10 * HZ);
877 	return;
878 
879 sched:
880 	if (unlikely(q->irq_ptr->state == QDIO_IRQ_STATE_STOPPED))
881 		return;
882 	tasklet_schedule(&q->tasklet);
883 }
884 
885 /* outbound tasklet */
886 void qdio_outbound_processing(unsigned long data)
887 {
888 	struct qdio_q *q = (struct qdio_q *)data;
889 	__qdio_outbound_processing(q);
890 }
891 
892 void qdio_outbound_timer(unsigned long data)
893 {
894 	struct qdio_q *q = (struct qdio_q *)data;
895 
896 	if (unlikely(q->irq_ptr->state == QDIO_IRQ_STATE_STOPPED))
897 		return;
898 	tasklet_schedule(&q->tasklet);
899 }
900 
901 static inline void qdio_check_outbound_after_thinint(struct qdio_q *q)
902 {
903 	struct qdio_q *out;
904 	int i;
905 
906 	if (!pci_out_supported(q))
907 		return;
908 
909 	for_each_output_queue(q->irq_ptr, out, i)
910 		if (!qdio_outbound_q_done(out))
911 			tasklet_schedule(&out->tasklet);
912 }
913 
914 static void __tiqdio_inbound_processing(struct qdio_q *q)
915 {
916 	qperf_inc(q, tasklet_inbound);
917 	if (need_siga_sync(q) && need_siga_sync_after_ai(q))
918 		qdio_sync_queues(q);
919 
920 	/*
921 	 * The interrupt could be caused by a PCI request. Check the
922 	 * PCI capable outbound queues.
923 	 */
924 	qdio_check_outbound_after_thinint(q);
925 
926 	if (!qdio_inbound_q_moved(q))
927 		return;
928 
929 	qdio_kick_handler(q);
930 
931 	if (!qdio_inbound_q_done(q)) {
932 		qperf_inc(q, tasklet_inbound_resched);
933 		if (likely(q->irq_ptr->state != QDIO_IRQ_STATE_STOPPED)) {
934 			tasklet_schedule(&q->tasklet);
935 			return;
936 		}
937 	}
938 
939 	qdio_stop_polling(q);
940 	/*
941 	 * We need to check again to not lose initiative after
942 	 * resetting the ACK state.
943 	 */
944 	if (!qdio_inbound_q_done(q)) {
945 		qperf_inc(q, tasklet_inbound_resched2);
946 		if (likely(q->irq_ptr->state != QDIO_IRQ_STATE_STOPPED))
947 			tasklet_schedule(&q->tasklet);
948 	}
949 }
950 
951 void tiqdio_inbound_processing(unsigned long data)
952 {
953 	struct qdio_q *q = (struct qdio_q *)data;
954 	__tiqdio_inbound_processing(q);
955 }
956 
957 static inline void qdio_set_state(struct qdio_irq *irq_ptr,
958 				  enum qdio_irq_states state)
959 {
960 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "newstate: %1d", state);
961 
962 	irq_ptr->state = state;
963 	mb();
964 }
965 
966 static void qdio_irq_check_sense(struct qdio_irq *irq_ptr, struct irb *irb)
967 {
968 	if (irb->esw.esw0.erw.cons) {
969 		DBF_ERROR("%4x sense:", irq_ptr->schid.sch_no);
970 		DBF_ERROR_HEX(irb, 64);
971 		DBF_ERROR_HEX(irb->ecw, 64);
972 	}
973 }
974 
975 /* PCI interrupt handler */
976 static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
977 {
978 	int i;
979 	struct qdio_q *q;
980 
981 	if (unlikely(irq_ptr->state == QDIO_IRQ_STATE_STOPPED))
982 		return;
983 
984 	for_each_input_queue(irq_ptr, q, i) {
985 		if (q->u.in.queue_start_poll) {
986 			/* skip if polling is enabled or already in work */
987 			if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
988 				     &q->u.in.queue_irq_state)) {
989 				qperf_inc(q, int_discarded);
990 				continue;
991 			}
992 			q->u.in.queue_start_poll(q->irq_ptr->cdev, q->nr,
993 						 q->irq_ptr->int_parm);
994 		} else {
995 			tasklet_schedule(&q->tasklet);
996 		}
997 	}
998 
999 	if (!(irq_ptr->qib.ac & QIB_AC_OUTBOUND_PCI_SUPPORTED))
1000 		return;
1001 
1002 	for_each_output_queue(irq_ptr, q, i) {
1003 		if (qdio_outbound_q_done(q))
1004 			continue;
1005 		if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
1006 			qdio_siga_sync_q(q);
1007 		tasklet_schedule(&q->tasklet);
1008 	}
1009 }
1010 
1011 static void qdio_handle_activate_check(struct ccw_device *cdev,
1012 				unsigned long intparm, int cstat, int dstat)
1013 {
1014 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1015 	struct qdio_q *q;
1016 	int count;
1017 
1018 	DBF_ERROR("%4x ACT CHECK", irq_ptr->schid.sch_no);
1019 	DBF_ERROR("intp :%lx", intparm);
1020 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1021 
1022 	if (irq_ptr->nr_input_qs) {
1023 		q = irq_ptr->input_qs[0];
1024 	} else if (irq_ptr->nr_output_qs) {
1025 		q = irq_ptr->output_qs[0];
1026 	} else {
1027 		dump_stack();
1028 		goto no_handler;
1029 	}
1030 
1031 	count = sub_buf(q->first_to_check, q->first_to_kick);
1032 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_ACTIVATE,
1033 		   q->nr, q->first_to_kick, count, irq_ptr->int_parm);
1034 no_handler:
1035 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1036 	/*
1037 	 * In case of z/VM LGR (Live Guest Migration) QDIO recovery will happen.
1038 	 * Therefore we call the LGR detection function here.
1039 	 */
1040 	lgr_info_log();
1041 }
1042 
1043 static void qdio_establish_handle_irq(struct ccw_device *cdev, int cstat,
1044 				      int dstat)
1045 {
1046 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1047 
1048 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "qest irq");
1049 
1050 	if (cstat)
1051 		goto error;
1052 	if (dstat & ~(DEV_STAT_DEV_END | DEV_STAT_CHN_END))
1053 		goto error;
1054 	if (!(dstat & DEV_STAT_DEV_END))
1055 		goto error;
1056 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ESTABLISHED);
1057 	return;
1058 
1059 error:
1060 	DBF_ERROR("%4x EQ:error", irq_ptr->schid.sch_no);
1061 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1062 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1063 }
1064 
1065 /* qdio interrupt handler */
1066 void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
1067 		      struct irb *irb)
1068 {
1069 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1070 	int cstat, dstat;
1071 
1072 	if (!intparm || !irq_ptr) {
1073 		DBF_ERROR("qint:%4x", cdev->private->schid.sch_no);
1074 		return;
1075 	}
1076 
1077 	if (irq_ptr->perf_stat_enabled)
1078 		irq_ptr->perf_stat.qdio_int++;
1079 
1080 	if (IS_ERR(irb)) {
1081 		DBF_ERROR("%4x IO error", irq_ptr->schid.sch_no);
1082 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1083 		wake_up(&cdev->private->wait_q);
1084 		return;
1085 	}
1086 	qdio_irq_check_sense(irq_ptr, irb);
1087 	cstat = irb->scsw.cmd.cstat;
1088 	dstat = irb->scsw.cmd.dstat;
1089 
1090 	switch (irq_ptr->state) {
1091 	case QDIO_IRQ_STATE_INACTIVE:
1092 		qdio_establish_handle_irq(cdev, cstat, dstat);
1093 		break;
1094 	case QDIO_IRQ_STATE_CLEANUP:
1095 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1096 		break;
1097 	case QDIO_IRQ_STATE_ESTABLISHED:
1098 	case QDIO_IRQ_STATE_ACTIVE:
1099 		if (cstat & SCHN_STAT_PCI) {
1100 			qdio_int_handler_pci(irq_ptr);
1101 			return;
1102 		}
1103 		if (cstat || dstat)
1104 			qdio_handle_activate_check(cdev, intparm, cstat,
1105 						   dstat);
1106 		break;
1107 	case QDIO_IRQ_STATE_STOPPED:
1108 		break;
1109 	default:
1110 		WARN_ON_ONCE(1);
1111 	}
1112 	wake_up(&cdev->private->wait_q);
1113 }
1114 
1115 /**
1116  * qdio_get_ssqd_desc - get qdio subchannel description
1117  * @cdev: ccw device to get description for
1118  * @data: where to store the ssqd
1119  *
1120  * Returns 0 or an error code. The results of the chsc are stored in the
1121  * specified structure.
1122  */
1123 int qdio_get_ssqd_desc(struct ccw_device *cdev,
1124 		       struct qdio_ssqd_desc *data)
1125 {
1126 
1127 	if (!cdev || !cdev->private)
1128 		return -EINVAL;
1129 
1130 	DBF_EVENT("get ssqd:%4x", cdev->private->schid.sch_no);
1131 	return qdio_setup_get_ssqd(NULL, &cdev->private->schid, data);
1132 }
1133 EXPORT_SYMBOL_GPL(qdio_get_ssqd_desc);
1134 
1135 static void qdio_shutdown_queues(struct ccw_device *cdev)
1136 {
1137 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1138 	struct qdio_q *q;
1139 	int i;
1140 
1141 	for_each_input_queue(irq_ptr, q, i)
1142 		tasklet_kill(&q->tasklet);
1143 
1144 	for_each_output_queue(irq_ptr, q, i) {
1145 		del_timer(&q->u.out.timer);
1146 		tasklet_kill(&q->tasklet);
1147 	}
1148 }
1149 
1150 /**
1151  * qdio_shutdown - shut down a qdio subchannel
1152  * @cdev: associated ccw device
1153  * @how: use halt or clear to shutdown
1154  */
1155 int qdio_shutdown(struct ccw_device *cdev, int how)
1156 {
1157 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1158 	int rc;
1159 	unsigned long flags;
1160 
1161 	if (!irq_ptr)
1162 		return -ENODEV;
1163 
1164 	WARN_ON_ONCE(irqs_disabled());
1165 	DBF_EVENT("qshutdown:%4x", cdev->private->schid.sch_no);
1166 
1167 	mutex_lock(&irq_ptr->setup_mutex);
1168 	/*
1169 	 * Subchannel was already shot down. We cannot prevent being called
1170 	 * twice since cio may trigger a shutdown asynchronously.
1171 	 */
1172 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1173 		mutex_unlock(&irq_ptr->setup_mutex);
1174 		return 0;
1175 	}
1176 
1177 	/*
1178 	 * Indicate that the device is going down. Scheduling the queue
1179 	 * tasklets is forbidden from here on.
1180 	 */
1181 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1182 
1183 	tiqdio_remove_input_queues(irq_ptr);
1184 	qdio_shutdown_queues(cdev);
1185 	qdio_shutdown_debug_entries(irq_ptr);
1186 
1187 	/* cleanup subchannel */
1188 	spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
1189 
1190 	if (how & QDIO_FLAG_CLEANUP_USING_CLEAR)
1191 		rc = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
1192 	else
1193 		/* default behaviour is halt */
1194 		rc = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
1195 	if (rc) {
1196 		DBF_ERROR("%4x SHUTD ERR", irq_ptr->schid.sch_no);
1197 		DBF_ERROR("rc:%4d", rc);
1198 		goto no_cleanup;
1199 	}
1200 
1201 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
1202 	spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
1203 	wait_event_interruptible_timeout(cdev->private->wait_q,
1204 		irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
1205 		irq_ptr->state == QDIO_IRQ_STATE_ERR,
1206 		10 * HZ);
1207 	spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
1208 
1209 no_cleanup:
1210 	qdio_shutdown_thinint(irq_ptr);
1211 
1212 	/* restore interrupt handler */
1213 	if ((void *)cdev->handler == (void *)qdio_int_handler)
1214 		cdev->handler = irq_ptr->orig_handler;
1215 	spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
1216 
1217 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1218 	mutex_unlock(&irq_ptr->setup_mutex);
1219 	if (rc)
1220 		return rc;
1221 	return 0;
1222 }
1223 EXPORT_SYMBOL_GPL(qdio_shutdown);
1224 
1225 /**
1226  * qdio_free - free data structures for a qdio subchannel
1227  * @cdev: associated ccw device
1228  */
1229 int qdio_free(struct ccw_device *cdev)
1230 {
1231 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1232 
1233 	if (!irq_ptr)
1234 		return -ENODEV;
1235 
1236 	DBF_EVENT("qfree:%4x", cdev->private->schid.sch_no);
1237 	mutex_lock(&irq_ptr->setup_mutex);
1238 
1239 	if (irq_ptr->debug_area != NULL) {
1240 		debug_unregister(irq_ptr->debug_area);
1241 		irq_ptr->debug_area = NULL;
1242 	}
1243 	cdev->private->qdio_data = NULL;
1244 	mutex_unlock(&irq_ptr->setup_mutex);
1245 
1246 	qdio_release_memory(irq_ptr);
1247 	return 0;
1248 }
1249 EXPORT_SYMBOL_GPL(qdio_free);
1250 
1251 /**
1252  * qdio_allocate - allocate qdio queues and associated data
1253  * @init_data: initialization data
1254  */
1255 int qdio_allocate(struct qdio_initialize *init_data)
1256 {
1257 	struct qdio_irq *irq_ptr;
1258 
1259 	DBF_EVENT("qallocate:%4x", init_data->cdev->private->schid.sch_no);
1260 
1261 	if ((init_data->no_input_qs && !init_data->input_handler) ||
1262 	    (init_data->no_output_qs && !init_data->output_handler))
1263 		return -EINVAL;
1264 
1265 	if ((init_data->no_input_qs > QDIO_MAX_QUEUES_PER_IRQ) ||
1266 	    (init_data->no_output_qs > QDIO_MAX_QUEUES_PER_IRQ))
1267 		return -EINVAL;
1268 
1269 	if ((!init_data->input_sbal_addr_array) ||
1270 	    (!init_data->output_sbal_addr_array))
1271 		return -EINVAL;
1272 
1273 	/* irq_ptr must be in GFP_DMA since it contains ccw1.cda */
1274 	irq_ptr = (void *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1275 	if (!irq_ptr)
1276 		goto out_err;
1277 
1278 	mutex_init(&irq_ptr->setup_mutex);
1279 	qdio_allocate_dbf(init_data, irq_ptr);
1280 
1281 	/*
1282 	 * Allocate a page for the chsc calls in qdio_establish.
1283 	 * Must be pre-allocated since a zfcp recovery will call
1284 	 * qdio_establish. In case of low memory and swap on a zfcp disk
1285 	 * we may not be able to allocate memory otherwise.
1286 	 */
1287 	irq_ptr->chsc_page = get_zeroed_page(GFP_KERNEL);
1288 	if (!irq_ptr->chsc_page)
1289 		goto out_rel;
1290 
1291 	/* qdr is used in ccw1.cda which is u32 */
1292 	irq_ptr->qdr = (struct qdr *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1293 	if (!irq_ptr->qdr)
1294 		goto out_rel;
1295 
1296 	if (qdio_allocate_qs(irq_ptr, init_data->no_input_qs,
1297 			     init_data->no_output_qs))
1298 		goto out_rel;
1299 
1300 	init_data->cdev->private->qdio_data = irq_ptr;
1301 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1302 	return 0;
1303 out_rel:
1304 	qdio_release_memory(irq_ptr);
1305 out_err:
1306 	return -ENOMEM;
1307 }
1308 EXPORT_SYMBOL_GPL(qdio_allocate);
1309 
1310 static void qdio_detect_hsicq(struct qdio_irq *irq_ptr)
1311 {
1312 	struct qdio_q *q = irq_ptr->input_qs[0];
1313 	int i, use_cq = 0;
1314 
1315 	if (irq_ptr->nr_input_qs > 1 && queue_type(q) == QDIO_IQDIO_QFMT)
1316 		use_cq = 1;
1317 
1318 	for_each_output_queue(irq_ptr, q, i) {
1319 		if (use_cq) {
1320 			if (qdio_enable_async_operation(&q->u.out) < 0) {
1321 				use_cq = 0;
1322 				continue;
1323 			}
1324 		} else
1325 			qdio_disable_async_operation(&q->u.out);
1326 	}
1327 	DBF_EVENT("use_cq:%d", use_cq);
1328 }
1329 
1330 /**
1331  * qdio_establish - establish queues on a qdio subchannel
1332  * @init_data: initialization data
1333  */
1334 int qdio_establish(struct qdio_initialize *init_data)
1335 {
1336 	struct qdio_irq *irq_ptr;
1337 	struct ccw_device *cdev = init_data->cdev;
1338 	unsigned long saveflags;
1339 	int rc;
1340 
1341 	DBF_EVENT("qestablish:%4x", cdev->private->schid.sch_no);
1342 
1343 	irq_ptr = cdev->private->qdio_data;
1344 	if (!irq_ptr)
1345 		return -ENODEV;
1346 
1347 	if (cdev->private->state != DEV_STATE_ONLINE)
1348 		return -EINVAL;
1349 
1350 	mutex_lock(&irq_ptr->setup_mutex);
1351 	qdio_setup_irq(init_data);
1352 
1353 	rc = qdio_establish_thinint(irq_ptr);
1354 	if (rc) {
1355 		mutex_unlock(&irq_ptr->setup_mutex);
1356 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1357 		return rc;
1358 	}
1359 
1360 	/* establish q */
1361 	irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
1362 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1363 	irq_ptr->ccw.count = irq_ptr->equeue.count;
1364 	irq_ptr->ccw.cda = (u32)((addr_t)irq_ptr->qdr);
1365 
1366 	spin_lock_irqsave(get_ccwdev_lock(cdev), saveflags);
1367 	ccw_device_set_options_mask(cdev, 0);
1368 
1369 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ESTABLISH, 0, 0);
1370 	if (rc) {
1371 		DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
1372 		DBF_ERROR("rc:%4x", rc);
1373 	}
1374 	spin_unlock_irqrestore(get_ccwdev_lock(cdev), saveflags);
1375 
1376 	if (rc) {
1377 		mutex_unlock(&irq_ptr->setup_mutex);
1378 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1379 		return rc;
1380 	}
1381 
1382 	wait_event_interruptible_timeout(cdev->private->wait_q,
1383 		irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
1384 		irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
1385 
1386 	if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
1387 		mutex_unlock(&irq_ptr->setup_mutex);
1388 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1389 		return -EIO;
1390 	}
1391 
1392 	qdio_setup_ssqd_info(irq_ptr);
1393 
1394 	qdio_detect_hsicq(irq_ptr);
1395 
1396 	/* qebsm is now setup if available, initialize buffer states */
1397 	qdio_init_buf_states(irq_ptr);
1398 
1399 	mutex_unlock(&irq_ptr->setup_mutex);
1400 	qdio_print_subchannel_info(irq_ptr, cdev);
1401 	qdio_setup_debug_entries(irq_ptr, cdev);
1402 	return 0;
1403 }
1404 EXPORT_SYMBOL_GPL(qdio_establish);
1405 
1406 /**
1407  * qdio_activate - activate queues on a qdio subchannel
1408  * @cdev: associated cdev
1409  */
1410 int qdio_activate(struct ccw_device *cdev)
1411 {
1412 	struct qdio_irq *irq_ptr;
1413 	int rc;
1414 	unsigned long saveflags;
1415 
1416 	DBF_EVENT("qactivate:%4x", cdev->private->schid.sch_no);
1417 
1418 	irq_ptr = cdev->private->qdio_data;
1419 	if (!irq_ptr)
1420 		return -ENODEV;
1421 
1422 	if (cdev->private->state != DEV_STATE_ONLINE)
1423 		return -EINVAL;
1424 
1425 	mutex_lock(&irq_ptr->setup_mutex);
1426 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1427 		rc = -EBUSY;
1428 		goto out;
1429 	}
1430 
1431 	irq_ptr->ccw.cmd_code = irq_ptr->aqueue.cmd;
1432 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1433 	irq_ptr->ccw.count = irq_ptr->aqueue.count;
1434 	irq_ptr->ccw.cda = 0;
1435 
1436 	spin_lock_irqsave(get_ccwdev_lock(cdev), saveflags);
1437 	ccw_device_set_options(cdev, CCWDEV_REPORT_ALL);
1438 
1439 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ACTIVATE,
1440 			      0, DOIO_DENY_PREFETCH);
1441 	if (rc) {
1442 		DBF_ERROR("%4x act IO ERR", irq_ptr->schid.sch_no);
1443 		DBF_ERROR("rc:%4x", rc);
1444 	}
1445 	spin_unlock_irqrestore(get_ccwdev_lock(cdev), saveflags);
1446 
1447 	if (rc)
1448 		goto out;
1449 
1450 	if (is_thinint_irq(irq_ptr))
1451 		tiqdio_add_input_queues(irq_ptr);
1452 
1453 	/* wait for subchannel to become active */
1454 	msleep(5);
1455 
1456 	switch (irq_ptr->state) {
1457 	case QDIO_IRQ_STATE_STOPPED:
1458 	case QDIO_IRQ_STATE_ERR:
1459 		rc = -EIO;
1460 		break;
1461 	default:
1462 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ACTIVE);
1463 		rc = 0;
1464 	}
1465 out:
1466 	mutex_unlock(&irq_ptr->setup_mutex);
1467 	return rc;
1468 }
1469 EXPORT_SYMBOL_GPL(qdio_activate);
1470 
1471 static inline int buf_in_between(int bufnr, int start, int count)
1472 {
1473 	int end = add_buf(start, count);
1474 
1475 	if (end > start) {
1476 		if (bufnr >= start && bufnr < end)
1477 			return 1;
1478 		else
1479 			return 0;
1480 	}
1481 
1482 	/* wrap-around case */
1483 	if ((bufnr >= start && bufnr <= QDIO_MAX_BUFFERS_PER_Q) ||
1484 	    (bufnr < end))
1485 		return 1;
1486 	else
1487 		return 0;
1488 }
1489 
1490 /**
1491  * handle_inbound - reset processed input buffers
1492  * @q: queue containing the buffers
1493  * @callflags: flags
1494  * @bufnr: first buffer to process
1495  * @count: how many buffers are emptied
1496  */
1497 static int handle_inbound(struct qdio_q *q, unsigned int callflags,
1498 			  int bufnr, int count)
1499 {
1500 	int diff;
1501 
1502 	qperf_inc(q, inbound_call);
1503 
1504 	if (!q->u.in.polling)
1505 		goto set;
1506 
1507 	/* protect against stop polling setting an ACK for an emptied slsb */
1508 	if (count == QDIO_MAX_BUFFERS_PER_Q) {
1509 		/* overwriting everything, just delete polling status */
1510 		q->u.in.polling = 0;
1511 		q->u.in.ack_count = 0;
1512 		goto set;
1513 	} else if (buf_in_between(q->u.in.ack_start, bufnr, count)) {
1514 		if (is_qebsm(q)) {
1515 			/* partial overwrite, just update ack_start */
1516 			diff = add_buf(bufnr, count);
1517 			diff = sub_buf(diff, q->u.in.ack_start);
1518 			q->u.in.ack_count -= diff;
1519 			if (q->u.in.ack_count <= 0) {
1520 				q->u.in.polling = 0;
1521 				q->u.in.ack_count = 0;
1522 				goto set;
1523 			}
1524 			q->u.in.ack_start = add_buf(q->u.in.ack_start, diff);
1525 		}
1526 		else
1527 			/* the only ACK will be deleted, so stop polling */
1528 			q->u.in.polling = 0;
1529 	}
1530 
1531 set:
1532 	count = set_buf_states(q, bufnr, SLSB_CU_INPUT_EMPTY, count);
1533 	atomic_add(count, &q->nr_buf_used);
1534 
1535 	if (need_siga_in(q))
1536 		return qdio_siga_input(q);
1537 
1538 	return 0;
1539 }
1540 
1541 /**
1542  * handle_outbound - process filled outbound buffers
1543  * @q: queue containing the buffers
1544  * @callflags: flags
1545  * @bufnr: first buffer to process
1546  * @count: how many buffers are filled
1547  */
1548 static int handle_outbound(struct qdio_q *q, unsigned int callflags,
1549 			   int bufnr, int count)
1550 {
1551 	unsigned char state = 0;
1552 	int used, rc = 0;
1553 
1554 	qperf_inc(q, outbound_call);
1555 
1556 	count = set_buf_states(q, bufnr, SLSB_CU_OUTPUT_PRIMED, count);
1557 	used = atomic_add_return(count, &q->nr_buf_used);
1558 
1559 	if (used == QDIO_MAX_BUFFERS_PER_Q)
1560 		qperf_inc(q, outbound_queue_full);
1561 
1562 	if (callflags & QDIO_FLAG_PCI_OUT) {
1563 		q->u.out.pci_out_enabled = 1;
1564 		qperf_inc(q, pci_request_int);
1565 	} else
1566 		q->u.out.pci_out_enabled = 0;
1567 
1568 	if (queue_type(q) == QDIO_IQDIO_QFMT) {
1569 		unsigned long phys_aob = 0;
1570 
1571 		/* One SIGA-W per buffer required for unicast HSI */
1572 		WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
1573 
1574 		phys_aob = qdio_aob_for_buffer(&q->u.out, bufnr);
1575 
1576 		rc = qdio_kick_outbound_q(q, phys_aob);
1577 	} else if (need_siga_sync(q)) {
1578 		rc = qdio_siga_sync_q(q);
1579 	} else {
1580 		/* try to fast requeue buffers */
1581 		get_buf_state(q, prev_buf(bufnr), &state, 0);
1582 		if (state != SLSB_CU_OUTPUT_PRIMED)
1583 			rc = qdio_kick_outbound_q(q, 0);
1584 		else
1585 			qperf_inc(q, fast_requeue);
1586 	}
1587 
1588 	/* in case of SIGA errors we must process the error immediately */
1589 	if (used >= q->u.out.scan_threshold || rc)
1590 		tasklet_schedule(&q->tasklet);
1591 	else
1592 		/* free the SBALs in case of no further traffic */
1593 		if (!timer_pending(&q->u.out.timer))
1594 			mod_timer(&q->u.out.timer, jiffies + HZ);
1595 	return rc;
1596 }
1597 
1598 /**
1599  * do_QDIO - process input or output buffers
1600  * @cdev: associated ccw_device for the qdio subchannel
1601  * @callflags: input or output and special flags from the program
1602  * @q_nr: queue number
1603  * @bufnr: buffer number
1604  * @count: how many buffers to process
1605  */
1606 int do_QDIO(struct ccw_device *cdev, unsigned int callflags,
1607 	    int q_nr, unsigned int bufnr, unsigned int count)
1608 {
1609 	struct qdio_irq *irq_ptr;
1610 
1611 	if (bufnr >= QDIO_MAX_BUFFERS_PER_Q || count > QDIO_MAX_BUFFERS_PER_Q)
1612 		return -EINVAL;
1613 
1614 	irq_ptr = cdev->private->qdio_data;
1615 	if (!irq_ptr)
1616 		return -ENODEV;
1617 
1618 	DBF_DEV_EVENT(DBF_INFO, irq_ptr,
1619 		      "do%02x b:%02x c:%02x", callflags, bufnr, count);
1620 
1621 	if (irq_ptr->state != QDIO_IRQ_STATE_ACTIVE)
1622 		return -EIO;
1623 	if (!count)
1624 		return 0;
1625 	if (callflags & QDIO_FLAG_SYNC_INPUT)
1626 		return handle_inbound(irq_ptr->input_qs[q_nr],
1627 				      callflags, bufnr, count);
1628 	else if (callflags & QDIO_FLAG_SYNC_OUTPUT)
1629 		return handle_outbound(irq_ptr->output_qs[q_nr],
1630 				       callflags, bufnr, count);
1631 	return -EINVAL;
1632 }
1633 EXPORT_SYMBOL_GPL(do_QDIO);
1634 
1635 /**
1636  * qdio_start_irq - process input buffers
1637  * @cdev: associated ccw_device for the qdio subchannel
1638  * @nr: input queue number
1639  *
1640  * Return codes
1641  *   0 - success
1642  *   1 - irqs not started since new data is available
1643  */
1644 int qdio_start_irq(struct ccw_device *cdev, int nr)
1645 {
1646 	struct qdio_q *q;
1647 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1648 
1649 	if (!irq_ptr)
1650 		return -ENODEV;
1651 	q = irq_ptr->input_qs[nr];
1652 
1653 	clear_nonshared_ind(irq_ptr);
1654 	qdio_stop_polling(q);
1655 	clear_bit(QDIO_QUEUE_IRQS_DISABLED, &q->u.in.queue_irq_state);
1656 
1657 	/*
1658 	 * We need to check again to not lose initiative after
1659 	 * resetting the ACK state.
1660 	 */
1661 	if (test_nonshared_ind(irq_ptr))
1662 		goto rescan;
1663 	if (!qdio_inbound_q_done(q))
1664 		goto rescan;
1665 	return 0;
1666 
1667 rescan:
1668 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1669 			     &q->u.in.queue_irq_state))
1670 		return 0;
1671 	else
1672 		return 1;
1673 
1674 }
1675 EXPORT_SYMBOL(qdio_start_irq);
1676 
1677 /**
1678  * qdio_get_next_buffers - process input buffers
1679  * @cdev: associated ccw_device for the qdio subchannel
1680  * @nr: input queue number
1681  * @bufnr: first filled buffer number
1682  * @error: buffers are in error state
1683  *
1684  * Return codes
1685  *   < 0 - error
1686  *   = 0 - no new buffers found
1687  *   > 0 - number of processed buffers
1688  */
1689 int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
1690 			  int *error)
1691 {
1692 	struct qdio_q *q;
1693 	int start, end;
1694 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1695 
1696 	if (!irq_ptr)
1697 		return -ENODEV;
1698 	q = irq_ptr->input_qs[nr];
1699 
1700 	/*
1701 	 * Cannot rely on automatic sync after interrupt since queues may
1702 	 * also be examined without interrupt.
1703 	 */
1704 	if (need_siga_sync(q))
1705 		qdio_sync_queues(q);
1706 
1707 	/* check the PCI capable outbound queues. */
1708 	qdio_check_outbound_after_thinint(q);
1709 
1710 	if (!qdio_inbound_q_moved(q))
1711 		return 0;
1712 
1713 	/* Note: upper-layer MUST stop processing immediately here ... */
1714 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
1715 		return -EIO;
1716 
1717 	start = q->first_to_kick;
1718 	end = q->first_to_check;
1719 	*bufnr = start;
1720 	*error = q->qdio_error;
1721 
1722 	/* for the next time */
1723 	q->first_to_kick = end;
1724 	q->qdio_error = 0;
1725 	return sub_buf(end, start);
1726 }
1727 EXPORT_SYMBOL(qdio_get_next_buffers);
1728 
1729 /**
1730  * qdio_stop_irq - disable interrupt processing for the device
1731  * @cdev: associated ccw_device for the qdio subchannel
1732  * @nr: input queue number
1733  *
1734  * Return codes
1735  *   0 - interrupts were already disabled
1736  *   1 - interrupts successfully disabled
1737  */
1738 int qdio_stop_irq(struct ccw_device *cdev, int nr)
1739 {
1740 	struct qdio_q *q;
1741 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1742 
1743 	if (!irq_ptr)
1744 		return -ENODEV;
1745 	q = irq_ptr->input_qs[nr];
1746 
1747 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1748 			     &q->u.in.queue_irq_state))
1749 		return 0;
1750 	else
1751 		return 1;
1752 }
1753 EXPORT_SYMBOL(qdio_stop_irq);
1754 
1755 /**
1756  * qdio_pnso_brinfo() - perform network subchannel op #0 - bridge info.
1757  * @schid:		Subchannel ID.
1758  * @cnc:		Boolean Change-Notification Control
1759  * @response:		Response code will be stored at this address
1760  * @cb: 		Callback function will be executed for each element
1761  *			of the address list
1762  * @priv:		Pointer passed from the caller to qdio_pnso_brinfo()
1763  * @type:		Type of the address entry passed to the callback
1764  * @entry:		Entry containg the address of the specified type
1765  * @priv:		Pointer to pass to the callback function.
1766  *
1767  * Performs "Store-network-bridging-information list" operation and calls
1768  * the callback function for every entry in the list. If "change-
1769  * notification-control" is set, further changes in the address list
1770  * will be reported via the IPA command.
1771  */
1772 int qdio_pnso_brinfo(struct subchannel_id schid,
1773 		int cnc, u16 *response,
1774 		void (*cb)(void *priv, enum qdio_brinfo_entry_type type,
1775 				void *entry),
1776 		void *priv)
1777 {
1778 	struct chsc_pnso_area *rr;
1779 	int rc;
1780 	u32 prev_instance = 0;
1781 	int isfirstblock = 1;
1782 	int i, size, elems;
1783 
1784 	rr = (struct chsc_pnso_area *)get_zeroed_page(GFP_KERNEL);
1785 	if (rr == NULL)
1786 		return -ENOMEM;
1787 	do {
1788 		/* on the first iteration, naihdr.resume_token will be zero */
1789 		rc = chsc_pnso_brinfo(schid, rr, rr->naihdr.resume_token, cnc);
1790 		if (rc != 0 && rc != -EBUSY)
1791 			goto out;
1792 		if (rr->response.code != 1) {
1793 			rc = -EIO;
1794 			continue;
1795 		} else
1796 			rc = 0;
1797 
1798 		if (cb == NULL)
1799 			continue;
1800 
1801 		size = rr->naihdr.naids;
1802 		elems = (rr->response.length -
1803 				sizeof(struct chsc_header) -
1804 				sizeof(struct chsc_brinfo_naihdr)) /
1805 				size;
1806 
1807 		if (!isfirstblock && (rr->naihdr.instance != prev_instance)) {
1808 			/* Inform the caller that they need to scrap */
1809 			/* the data that was already reported via cb */
1810 				rc = -EAGAIN;
1811 				break;
1812 		}
1813 		isfirstblock = 0;
1814 		prev_instance = rr->naihdr.instance;
1815 		for (i = 0; i < elems; i++)
1816 			switch (size) {
1817 			case sizeof(struct qdio_brinfo_entry_l3_ipv6):
1818 				(*cb)(priv, l3_ipv6_addr,
1819 						&rr->entries.l3_ipv6[i]);
1820 				break;
1821 			case sizeof(struct qdio_brinfo_entry_l3_ipv4):
1822 				(*cb)(priv, l3_ipv4_addr,
1823 						&rr->entries.l3_ipv4[i]);
1824 				break;
1825 			case sizeof(struct qdio_brinfo_entry_l2):
1826 				(*cb)(priv, l2_addr_lnid,
1827 						&rr->entries.l2[i]);
1828 				break;
1829 			default:
1830 				WARN_ON_ONCE(1);
1831 				rc = -EIO;
1832 				goto out;
1833 			}
1834 	} while (rr->response.code == 0x0107 ||  /* channel busy */
1835 		  (rr->response.code == 1 && /* list stored */
1836 		   /* resume token is non-zero => list incomplete */
1837 		   (rr->naihdr.resume_token.t1 || rr->naihdr.resume_token.t2)));
1838 	(*response) = rr->response.code;
1839 
1840 out:
1841 	free_page((unsigned long)rr);
1842 	return rc;
1843 }
1844 EXPORT_SYMBOL_GPL(qdio_pnso_brinfo);
1845 
1846 static int __init init_QDIO(void)
1847 {
1848 	int rc;
1849 
1850 	rc = qdio_debug_init();
1851 	if (rc)
1852 		return rc;
1853 	rc = qdio_setup_init();
1854 	if (rc)
1855 		goto out_debug;
1856 	rc = tiqdio_allocate_memory();
1857 	if (rc)
1858 		goto out_cache;
1859 	rc = tiqdio_register_thinints();
1860 	if (rc)
1861 		goto out_ti;
1862 	return 0;
1863 
1864 out_ti:
1865 	tiqdio_free_memory();
1866 out_cache:
1867 	qdio_setup_exit();
1868 out_debug:
1869 	qdio_debug_exit();
1870 	return rc;
1871 }
1872 
1873 static void __exit exit_QDIO(void)
1874 {
1875 	tiqdio_unregister_thinints();
1876 	tiqdio_free_memory();
1877 	qdio_setup_exit();
1878 	qdio_debug_exit();
1879 }
1880 
1881 module_init(init_QDIO);
1882 module_exit(exit_QDIO);
1883