xref: /openbmc/linux/drivers/s390/cio/qdio_main.c (revision bc5aa3a0)
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 	WARN_ON_ONCE(aob && ((queue_type(q) != QDIO_IQDIO_QFMT) ||
323 			     !q->u.out.use_cq));
324 	if (q->u.out.use_cq && aob != 0) {
325 		fc = QDIO_SIGA_WRITEQ;
326 		laob = aob;
327 	}
328 
329 	if (is_qebsm(q)) {
330 		schid = q->irq_ptr->sch_token;
331 		fc |= QDIO_SIGA_QEBSM_FLAG;
332 	}
333 again:
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, unsigned int count)
413 {
414 	int pos;
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 	pos = ilog2(count);
422 	q->q_stats.nr_sbals[pos]++;
423 }
424 
425 static void process_buffer_error(struct qdio_q *q, int count)
426 {
427 	unsigned char state = (q->is_input_q) ? SLSB_P_INPUT_NOT_INIT :
428 					SLSB_P_OUTPUT_NOT_INIT;
429 
430 	q->qdio_error = QDIO_ERROR_SLSB_STATE;
431 
432 	/* special handling for no target buffer empty */
433 	if ((!q->is_input_q &&
434 	    (q->sbal[q->first_to_check]->element[15].sflags) == 0x10)) {
435 		qperf_inc(q, target_full);
436 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "OUTFULL FTC:%02x",
437 			      q->first_to_check);
438 		goto set;
439 	}
440 
441 	DBF_ERROR("%4x BUF ERROR", SCH_NO(q));
442 	DBF_ERROR((q->is_input_q) ? "IN:%2d" : "OUT:%2d", q->nr);
443 	DBF_ERROR("FTC:%3d C:%3d", q->first_to_check, count);
444 	DBF_ERROR("F14:%2x F15:%2x",
445 		  q->sbal[q->first_to_check]->element[14].sflags,
446 		  q->sbal[q->first_to_check]->element[15].sflags);
447 
448 set:
449 	/*
450 	 * Interrupts may be avoided as long as the error is present
451 	 * so change the buffer state immediately to avoid starvation.
452 	 */
453 	set_buf_states(q, q->first_to_check, state, count);
454 }
455 
456 static inline void inbound_primed(struct qdio_q *q, int count)
457 {
458 	int new;
459 
460 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in prim: %02x", count);
461 
462 	/* for QEBSM the ACK was already set by EQBS */
463 	if (is_qebsm(q)) {
464 		if (!q->u.in.polling) {
465 			q->u.in.polling = 1;
466 			q->u.in.ack_count = count;
467 			q->u.in.ack_start = q->first_to_check;
468 			return;
469 		}
470 
471 		/* delete the previous ACK's */
472 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
473 			       q->u.in.ack_count);
474 		q->u.in.ack_count = count;
475 		q->u.in.ack_start = q->first_to_check;
476 		return;
477 	}
478 
479 	/*
480 	 * ACK the newest buffer. The ACK will be removed in qdio_stop_polling
481 	 * or by the next inbound run.
482 	 */
483 	new = add_buf(q->first_to_check, count - 1);
484 	if (q->u.in.polling) {
485 		/* reset the previous ACK but first set the new one */
486 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
487 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
488 	} else {
489 		q->u.in.polling = 1;
490 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
491 	}
492 
493 	q->u.in.ack_start = new;
494 	count--;
495 	if (!count)
496 		return;
497 	/* need to change ALL buffers to get more interrupts */
498 	set_buf_states(q, q->first_to_check, SLSB_P_INPUT_NOT_INIT, count);
499 }
500 
501 static int get_inbound_buffer_frontier(struct qdio_q *q)
502 {
503 	int count, stop;
504 	unsigned char state = 0;
505 
506 	q->timestamp = get_tod_clock_fast();
507 
508 	/*
509 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
510 	 * would return 0.
511 	 */
512 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
513 	stop = add_buf(q->first_to_check, count);
514 
515 	if (q->first_to_check == stop)
516 		goto out;
517 
518 	/*
519 	 * No siga sync here, as a PCI or we after a thin interrupt
520 	 * already sync'ed the queues.
521 	 */
522 	count = get_buf_states(q, q->first_to_check, &state, count, 1, 0);
523 	if (!count)
524 		goto out;
525 
526 	switch (state) {
527 	case SLSB_P_INPUT_PRIMED:
528 		inbound_primed(q, count);
529 		q->first_to_check = add_buf(q->first_to_check, count);
530 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
531 			qperf_inc(q, inbound_queue_full);
532 		if (q->irq_ptr->perf_stat_enabled)
533 			account_sbals(q, count);
534 		break;
535 	case SLSB_P_INPUT_ERROR:
536 		process_buffer_error(q, count);
537 		q->first_to_check = add_buf(q->first_to_check, count);
538 		atomic_sub(count, &q->nr_buf_used);
539 		if (q->irq_ptr->perf_stat_enabled)
540 			account_sbals_error(q, count);
541 		break;
542 	case SLSB_CU_INPUT_EMPTY:
543 	case SLSB_P_INPUT_NOT_INIT:
544 	case SLSB_P_INPUT_ACK:
545 		if (q->irq_ptr->perf_stat_enabled)
546 			q->q_stats.nr_sbal_nop++;
547 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in nop");
548 		break;
549 	default:
550 		WARN_ON_ONCE(1);
551 	}
552 out:
553 	return q->first_to_check;
554 }
555 
556 static int qdio_inbound_q_moved(struct qdio_q *q)
557 {
558 	int bufnr;
559 
560 	bufnr = get_inbound_buffer_frontier(q);
561 
562 	if (bufnr != q->last_move) {
563 		q->last_move = bufnr;
564 		if (!is_thinint_irq(q->irq_ptr) && MACHINE_IS_LPAR)
565 			q->u.in.timestamp = get_tod_clock();
566 		return 1;
567 	} else
568 		return 0;
569 }
570 
571 static inline int qdio_inbound_q_done(struct qdio_q *q)
572 {
573 	unsigned char state = 0;
574 
575 	if (!atomic_read(&q->nr_buf_used))
576 		return 1;
577 
578 	if (need_siga_sync(q))
579 		qdio_siga_sync_q(q);
580 	get_buf_state(q, q->first_to_check, &state, 0);
581 
582 	if (state == SLSB_P_INPUT_PRIMED || state == SLSB_P_INPUT_ERROR)
583 		/* more work coming */
584 		return 0;
585 
586 	if (is_thinint_irq(q->irq_ptr))
587 		return 1;
588 
589 	/* don't poll under z/VM */
590 	if (MACHINE_IS_VM)
591 		return 1;
592 
593 	/*
594 	 * At this point we know, that inbound first_to_check
595 	 * has (probably) not moved (see qdio_inbound_processing).
596 	 */
597 	if (get_tod_clock_fast() > q->u.in.timestamp + QDIO_INPUT_THRESHOLD) {
598 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in done:%02x",
599 			      q->first_to_check);
600 		return 1;
601 	} else
602 		return 0;
603 }
604 
605 static inline int contains_aobs(struct qdio_q *q)
606 {
607 	return !q->is_input_q && q->u.out.use_cq;
608 }
609 
610 static inline void qdio_handle_aobs(struct qdio_q *q, int start, int count)
611 {
612 	unsigned char state = 0;
613 	int j, b = start;
614 
615 	if (!contains_aobs(q))
616 		return;
617 
618 	for (j = 0; j < count; ++j) {
619 		get_buf_state(q, b, &state, 0);
620 		if (state == SLSB_P_OUTPUT_PENDING) {
621 			struct qaob *aob = q->u.out.aobs[b];
622 			if (aob == NULL)
623 				continue;
624 
625 			q->u.out.sbal_state[b].flags |=
626 				QDIO_OUTBUF_STATE_FLAG_PENDING;
627 			q->u.out.aobs[b] = NULL;
628 		} else if (state == SLSB_P_OUTPUT_EMPTY) {
629 			q->u.out.sbal_state[b].aob = NULL;
630 		}
631 		b = next_buf(b);
632 	}
633 }
634 
635 static inline unsigned long qdio_aob_for_buffer(struct qdio_output_q *q,
636 					int bufnr)
637 {
638 	unsigned long phys_aob = 0;
639 
640 	if (!q->use_cq)
641 		goto out;
642 
643 	if (!q->aobs[bufnr]) {
644 		struct qaob *aob = qdio_allocate_aob();
645 		q->aobs[bufnr] = aob;
646 	}
647 	if (q->aobs[bufnr]) {
648 		q->sbal_state[bufnr].flags = QDIO_OUTBUF_STATE_FLAG_NONE;
649 		q->sbal_state[bufnr].aob = q->aobs[bufnr];
650 		q->aobs[bufnr]->user1 = (u64) q->sbal_state[bufnr].user;
651 		phys_aob = virt_to_phys(q->aobs[bufnr]);
652 		WARN_ON_ONCE(phys_aob & 0xFF);
653 	}
654 
655 out:
656 	return phys_aob;
657 }
658 
659 static void qdio_kick_handler(struct qdio_q *q)
660 {
661 	int start = q->first_to_kick;
662 	int end = q->first_to_check;
663 	int count;
664 
665 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
666 		return;
667 
668 	count = sub_buf(end, start);
669 
670 	if (q->is_input_q) {
671 		qperf_inc(q, inbound_handler);
672 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "kih s:%02x c:%02x", start, count);
673 	} else {
674 		qperf_inc(q, outbound_handler);
675 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "koh: s:%02x c:%02x",
676 			      start, count);
677 	}
678 
679 	qdio_handle_aobs(q, start, count);
680 
681 	q->handler(q->irq_ptr->cdev, q->qdio_error, q->nr, start, count,
682 		   q->irq_ptr->int_parm);
683 
684 	/* for the next time */
685 	q->first_to_kick = end;
686 	q->qdio_error = 0;
687 }
688 
689 static inline int qdio_tasklet_schedule(struct qdio_q *q)
690 {
691 	if (likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE)) {
692 		tasklet_schedule(&q->tasklet);
693 		return 0;
694 	}
695 	return -EPERM;
696 }
697 
698 static void __qdio_inbound_processing(struct qdio_q *q)
699 {
700 	qperf_inc(q, tasklet_inbound);
701 
702 	if (!qdio_inbound_q_moved(q))
703 		return;
704 
705 	qdio_kick_handler(q);
706 
707 	if (!qdio_inbound_q_done(q)) {
708 		/* means poll time is not yet over */
709 		qperf_inc(q, tasklet_inbound_resched);
710 		if (!qdio_tasklet_schedule(q))
711 			return;
712 	}
713 
714 	qdio_stop_polling(q);
715 	/*
716 	 * We need to check again to not lose initiative after
717 	 * resetting the ACK state.
718 	 */
719 	if (!qdio_inbound_q_done(q)) {
720 		qperf_inc(q, tasklet_inbound_resched2);
721 		qdio_tasklet_schedule(q);
722 	}
723 }
724 
725 void qdio_inbound_processing(unsigned long data)
726 {
727 	struct qdio_q *q = (struct qdio_q *)data;
728 	__qdio_inbound_processing(q);
729 }
730 
731 static int get_outbound_buffer_frontier(struct qdio_q *q)
732 {
733 	int count, stop;
734 	unsigned char state = 0;
735 
736 	q->timestamp = get_tod_clock_fast();
737 
738 	if (need_siga_sync(q))
739 		if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
740 		    !pci_out_supported(q)) ||
741 		    (queue_type(q) == QDIO_IQDIO_QFMT &&
742 		    multicast_outbound(q)))
743 			qdio_siga_sync_q(q);
744 
745 	/*
746 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
747 	 * would return 0.
748 	 */
749 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
750 	stop = add_buf(q->first_to_check, count);
751 	if (q->first_to_check == stop)
752 		goto out;
753 
754 	count = get_buf_states(q, q->first_to_check, &state, count, 0, 1);
755 	if (!count)
756 		goto out;
757 
758 	switch (state) {
759 	case SLSB_P_OUTPUT_EMPTY:
760 		/* the adapter got it */
761 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr,
762 			"out empty:%1d %02x", q->nr, count);
763 
764 		atomic_sub(count, &q->nr_buf_used);
765 		q->first_to_check = add_buf(q->first_to_check, count);
766 		if (q->irq_ptr->perf_stat_enabled)
767 			account_sbals(q, count);
768 
769 		break;
770 	case SLSB_P_OUTPUT_ERROR:
771 		process_buffer_error(q, count);
772 		q->first_to_check = add_buf(q->first_to_check, count);
773 		atomic_sub(count, &q->nr_buf_used);
774 		if (q->irq_ptr->perf_stat_enabled)
775 			account_sbals_error(q, count);
776 		break;
777 	case SLSB_CU_OUTPUT_PRIMED:
778 		/* the adapter has not fetched the output yet */
779 		if (q->irq_ptr->perf_stat_enabled)
780 			q->q_stats.nr_sbal_nop++;
781 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out primed:%1d",
782 			      q->nr);
783 		break;
784 	case SLSB_P_OUTPUT_NOT_INIT:
785 	case SLSB_P_OUTPUT_HALTED:
786 		break;
787 	default:
788 		WARN_ON_ONCE(1);
789 	}
790 
791 out:
792 	return q->first_to_check;
793 }
794 
795 /* all buffers processed? */
796 static inline int qdio_outbound_q_done(struct qdio_q *q)
797 {
798 	return atomic_read(&q->nr_buf_used) == 0;
799 }
800 
801 static inline int qdio_outbound_q_moved(struct qdio_q *q)
802 {
803 	int bufnr;
804 
805 	bufnr = get_outbound_buffer_frontier(q);
806 
807 	if (bufnr != q->last_move) {
808 		q->last_move = bufnr;
809 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out moved:%1d", q->nr);
810 		return 1;
811 	} else
812 		return 0;
813 }
814 
815 static int qdio_kick_outbound_q(struct qdio_q *q, unsigned long aob)
816 {
817 	int retries = 0, cc;
818 	unsigned int busy_bit;
819 
820 	if (!need_siga_out(q))
821 		return 0;
822 
823 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w:%1d", q->nr);
824 retry:
825 	qperf_inc(q, siga_write);
826 
827 	cc = qdio_siga_output(q, &busy_bit, aob);
828 	switch (cc) {
829 	case 0:
830 		break;
831 	case 2:
832 		if (busy_bit) {
833 			while (++retries < QDIO_BUSY_BIT_RETRIES) {
834 				mdelay(QDIO_BUSY_BIT_RETRY_DELAY);
835 				goto retry;
836 			}
837 			DBF_ERROR("%4x cc2 BBC:%1d", SCH_NO(q), q->nr);
838 			cc = -EBUSY;
839 		} else {
840 			DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w cc2:%1d", q->nr);
841 			cc = -ENOBUFS;
842 		}
843 		break;
844 	case 1:
845 	case 3:
846 		DBF_ERROR("%4x SIGA-W:%1d", SCH_NO(q), cc);
847 		cc = -EIO;
848 		break;
849 	}
850 	if (retries) {
851 		DBF_ERROR("%4x cc2 BB2:%1d", SCH_NO(q), q->nr);
852 		DBF_ERROR("count:%u", retries);
853 	}
854 	return cc;
855 }
856 
857 static void __qdio_outbound_processing(struct qdio_q *q)
858 {
859 	qperf_inc(q, tasklet_outbound);
860 	WARN_ON_ONCE(atomic_read(&q->nr_buf_used) < 0);
861 
862 	if (qdio_outbound_q_moved(q))
863 		qdio_kick_handler(q);
864 
865 	if (queue_type(q) == QDIO_ZFCP_QFMT)
866 		if (!pci_out_supported(q) && !qdio_outbound_q_done(q))
867 			goto sched;
868 
869 	if (q->u.out.pci_out_enabled)
870 		return;
871 
872 	/*
873 	 * Now we know that queue type is either qeth without pci enabled
874 	 * or HiperSockets. Make sure buffer switch from PRIMED to EMPTY
875 	 * is noticed and outbound_handler is called after some time.
876 	 */
877 	if (qdio_outbound_q_done(q))
878 		del_timer_sync(&q->u.out.timer);
879 	else
880 		if (!timer_pending(&q->u.out.timer) &&
881 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
882 			mod_timer(&q->u.out.timer, jiffies + 10 * HZ);
883 	return;
884 
885 sched:
886 	qdio_tasklet_schedule(q);
887 }
888 
889 /* outbound tasklet */
890 void qdio_outbound_processing(unsigned long data)
891 {
892 	struct qdio_q *q = (struct qdio_q *)data;
893 	__qdio_outbound_processing(q);
894 }
895 
896 void qdio_outbound_timer(unsigned long data)
897 {
898 	struct qdio_q *q = (struct qdio_q *)data;
899 
900 	qdio_tasklet_schedule(q);
901 }
902 
903 static inline void qdio_check_outbound_after_thinint(struct qdio_q *q)
904 {
905 	struct qdio_q *out;
906 	int i;
907 
908 	if (!pci_out_supported(q))
909 		return;
910 
911 	for_each_output_queue(q->irq_ptr, out, i)
912 		if (!qdio_outbound_q_done(out))
913 			qdio_tasklet_schedule(out);
914 }
915 
916 static void __tiqdio_inbound_processing(struct qdio_q *q)
917 {
918 	qperf_inc(q, tasklet_inbound);
919 	if (need_siga_sync(q) && need_siga_sync_after_ai(q))
920 		qdio_sync_queues(q);
921 
922 	/*
923 	 * The interrupt could be caused by a PCI request. Check the
924 	 * PCI capable outbound queues.
925 	 */
926 	qdio_check_outbound_after_thinint(q);
927 
928 	if (!qdio_inbound_q_moved(q))
929 		return;
930 
931 	qdio_kick_handler(q);
932 
933 	if (!qdio_inbound_q_done(q)) {
934 		qperf_inc(q, tasklet_inbound_resched);
935 		if (!qdio_tasklet_schedule(q))
936 			return;
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 		qdio_tasklet_schedule(q);
947 	}
948 }
949 
950 void tiqdio_inbound_processing(unsigned long data)
951 {
952 	struct qdio_q *q = (struct qdio_q *)data;
953 	__tiqdio_inbound_processing(q);
954 }
955 
956 static inline void qdio_set_state(struct qdio_irq *irq_ptr,
957 				  enum qdio_irq_states state)
958 {
959 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "newstate: %1d", state);
960 
961 	irq_ptr->state = state;
962 	mb();
963 }
964 
965 static void qdio_irq_check_sense(struct qdio_irq *irq_ptr, struct irb *irb)
966 {
967 	if (irb->esw.esw0.erw.cons) {
968 		DBF_ERROR("%4x sense:", irq_ptr->schid.sch_no);
969 		DBF_ERROR_HEX(irb, 64);
970 		DBF_ERROR_HEX(irb->ecw, 64);
971 	}
972 }
973 
974 /* PCI interrupt handler */
975 static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
976 {
977 	int i;
978 	struct qdio_q *q;
979 
980 	if (unlikely(irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
981 		return;
982 
983 	for_each_input_queue(irq_ptr, q, i) {
984 		if (q->u.in.queue_start_poll) {
985 			/* skip if polling is enabled or already in work */
986 			if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
987 				     &q->u.in.queue_irq_state)) {
988 				qperf_inc(q, int_discarded);
989 				continue;
990 			}
991 			q->u.in.queue_start_poll(q->irq_ptr->cdev, q->nr,
992 						 q->irq_ptr->int_parm);
993 		} else {
994 			tasklet_schedule(&q->tasklet);
995 		}
996 	}
997 
998 	if (!(irq_ptr->qib.ac & QIB_AC_OUTBOUND_PCI_SUPPORTED))
999 		return;
1000 
1001 	for_each_output_queue(irq_ptr, q, i) {
1002 		if (qdio_outbound_q_done(q))
1003 			continue;
1004 		if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
1005 			qdio_siga_sync_q(q);
1006 		qdio_tasklet_schedule(q);
1007 	}
1008 }
1009 
1010 static void qdio_handle_activate_check(struct ccw_device *cdev,
1011 				unsigned long intparm, int cstat, int dstat)
1012 {
1013 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1014 	struct qdio_q *q;
1015 	int count;
1016 
1017 	DBF_ERROR("%4x ACT CHECK", irq_ptr->schid.sch_no);
1018 	DBF_ERROR("intp :%lx", intparm);
1019 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1020 
1021 	if (irq_ptr->nr_input_qs) {
1022 		q = irq_ptr->input_qs[0];
1023 	} else if (irq_ptr->nr_output_qs) {
1024 		q = irq_ptr->output_qs[0];
1025 	} else {
1026 		dump_stack();
1027 		goto no_handler;
1028 	}
1029 
1030 	count = sub_buf(q->first_to_check, q->first_to_kick);
1031 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_ACTIVATE,
1032 		   q->nr, q->first_to_kick, count, irq_ptr->int_parm);
1033 no_handler:
1034 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1035 	/*
1036 	 * In case of z/VM LGR (Live Guest Migration) QDIO recovery will happen.
1037 	 * Therefore we call the LGR detection function here.
1038 	 */
1039 	lgr_info_log();
1040 }
1041 
1042 static void qdio_establish_handle_irq(struct ccw_device *cdev, int cstat,
1043 				      int dstat)
1044 {
1045 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1046 
1047 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "qest irq");
1048 
1049 	if (cstat)
1050 		goto error;
1051 	if (dstat & ~(DEV_STAT_DEV_END | DEV_STAT_CHN_END))
1052 		goto error;
1053 	if (!(dstat & DEV_STAT_DEV_END))
1054 		goto error;
1055 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ESTABLISHED);
1056 	return;
1057 
1058 error:
1059 	DBF_ERROR("%4x EQ:error", irq_ptr->schid.sch_no);
1060 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1061 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1062 }
1063 
1064 /* qdio interrupt handler */
1065 void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
1066 		      struct irb *irb)
1067 {
1068 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1069 	struct subchannel_id schid;
1070 	int cstat, dstat;
1071 
1072 	if (!intparm || !irq_ptr) {
1073 		ccw_device_get_schid(cdev, &schid);
1074 		DBF_ERROR("qint:%4x", schid.sch_no);
1075 		return;
1076 	}
1077 
1078 	if (irq_ptr->perf_stat_enabled)
1079 		irq_ptr->perf_stat.qdio_int++;
1080 
1081 	if (IS_ERR(irb)) {
1082 		DBF_ERROR("%4x IO error", irq_ptr->schid.sch_no);
1083 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1084 		wake_up(&cdev->private->wait_q);
1085 		return;
1086 	}
1087 	qdio_irq_check_sense(irq_ptr, irb);
1088 	cstat = irb->scsw.cmd.cstat;
1089 	dstat = irb->scsw.cmd.dstat;
1090 
1091 	switch (irq_ptr->state) {
1092 	case QDIO_IRQ_STATE_INACTIVE:
1093 		qdio_establish_handle_irq(cdev, cstat, dstat);
1094 		break;
1095 	case QDIO_IRQ_STATE_CLEANUP:
1096 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1097 		break;
1098 	case QDIO_IRQ_STATE_ESTABLISHED:
1099 	case QDIO_IRQ_STATE_ACTIVE:
1100 		if (cstat & SCHN_STAT_PCI) {
1101 			qdio_int_handler_pci(irq_ptr);
1102 			return;
1103 		}
1104 		if (cstat || dstat)
1105 			qdio_handle_activate_check(cdev, intparm, cstat,
1106 						   dstat);
1107 		break;
1108 	case QDIO_IRQ_STATE_STOPPED:
1109 		break;
1110 	default:
1111 		WARN_ON_ONCE(1);
1112 	}
1113 	wake_up(&cdev->private->wait_q);
1114 }
1115 
1116 /**
1117  * qdio_get_ssqd_desc - get qdio subchannel description
1118  * @cdev: ccw device to get description for
1119  * @data: where to store the ssqd
1120  *
1121  * Returns 0 or an error code. The results of the chsc are stored in the
1122  * specified structure.
1123  */
1124 int qdio_get_ssqd_desc(struct ccw_device *cdev,
1125 		       struct qdio_ssqd_desc *data)
1126 {
1127 	struct subchannel_id schid;
1128 
1129 	if (!cdev || !cdev->private)
1130 		return -EINVAL;
1131 
1132 	ccw_device_get_schid(cdev, &schid);
1133 	DBF_EVENT("get ssqd:%4x", schid.sch_no);
1134 	return qdio_setup_get_ssqd(NULL, &schid, data);
1135 }
1136 EXPORT_SYMBOL_GPL(qdio_get_ssqd_desc);
1137 
1138 static void qdio_shutdown_queues(struct ccw_device *cdev)
1139 {
1140 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1141 	struct qdio_q *q;
1142 	int i;
1143 
1144 	for_each_input_queue(irq_ptr, q, i)
1145 		tasklet_kill(&q->tasklet);
1146 
1147 	for_each_output_queue(irq_ptr, q, i) {
1148 		del_timer_sync(&q->u.out.timer);
1149 		tasklet_kill(&q->tasklet);
1150 	}
1151 }
1152 
1153 /**
1154  * qdio_shutdown - shut down a qdio subchannel
1155  * @cdev: associated ccw device
1156  * @how: use halt or clear to shutdown
1157  */
1158 int qdio_shutdown(struct ccw_device *cdev, int how)
1159 {
1160 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1161 	struct subchannel_id schid;
1162 	int rc;
1163 
1164 	if (!irq_ptr)
1165 		return -ENODEV;
1166 
1167 	WARN_ON_ONCE(irqs_disabled());
1168 	ccw_device_get_schid(cdev, &schid);
1169 	DBF_EVENT("qshutdown:%4x", schid.sch_no);
1170 
1171 	mutex_lock(&irq_ptr->setup_mutex);
1172 	/*
1173 	 * Subchannel was already shot down. We cannot prevent being called
1174 	 * twice since cio may trigger a shutdown asynchronously.
1175 	 */
1176 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1177 		mutex_unlock(&irq_ptr->setup_mutex);
1178 		return 0;
1179 	}
1180 
1181 	/*
1182 	 * Indicate that the device is going down. Scheduling the queue
1183 	 * tasklets is forbidden from here on.
1184 	 */
1185 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1186 
1187 	tiqdio_remove_input_queues(irq_ptr);
1188 	qdio_shutdown_queues(cdev);
1189 	qdio_shutdown_debug_entries(irq_ptr);
1190 
1191 	/* cleanup subchannel */
1192 	spin_lock_irq(get_ccwdev_lock(cdev));
1193 
1194 	if (how & QDIO_FLAG_CLEANUP_USING_CLEAR)
1195 		rc = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
1196 	else
1197 		/* default behaviour is halt */
1198 		rc = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
1199 	if (rc) {
1200 		DBF_ERROR("%4x SHUTD ERR", irq_ptr->schid.sch_no);
1201 		DBF_ERROR("rc:%4d", rc);
1202 		goto no_cleanup;
1203 	}
1204 
1205 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
1206 	spin_unlock_irq(get_ccwdev_lock(cdev));
1207 	wait_event_interruptible_timeout(cdev->private->wait_q,
1208 		irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
1209 		irq_ptr->state == QDIO_IRQ_STATE_ERR,
1210 		10 * HZ);
1211 	spin_lock_irq(get_ccwdev_lock(cdev));
1212 
1213 no_cleanup:
1214 	qdio_shutdown_thinint(irq_ptr);
1215 
1216 	/* restore interrupt handler */
1217 	if ((void *)cdev->handler == (void *)qdio_int_handler)
1218 		cdev->handler = irq_ptr->orig_handler;
1219 	spin_unlock_irq(get_ccwdev_lock(cdev));
1220 
1221 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1222 	mutex_unlock(&irq_ptr->setup_mutex);
1223 	if (rc)
1224 		return rc;
1225 	return 0;
1226 }
1227 EXPORT_SYMBOL_GPL(qdio_shutdown);
1228 
1229 /**
1230  * qdio_free - free data structures for a qdio subchannel
1231  * @cdev: associated ccw device
1232  */
1233 int qdio_free(struct ccw_device *cdev)
1234 {
1235 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1236 	struct subchannel_id schid;
1237 
1238 	if (!irq_ptr)
1239 		return -ENODEV;
1240 
1241 	ccw_device_get_schid(cdev, &schid);
1242 	DBF_EVENT("qfree:%4x", schid.sch_no);
1243 	DBF_DEV_EVENT(DBF_ERR, irq_ptr, "dbf abandoned");
1244 	mutex_lock(&irq_ptr->setup_mutex);
1245 
1246 	irq_ptr->debug_area = NULL;
1247 	cdev->private->qdio_data = NULL;
1248 	mutex_unlock(&irq_ptr->setup_mutex);
1249 
1250 	qdio_release_memory(irq_ptr);
1251 	return 0;
1252 }
1253 EXPORT_SYMBOL_GPL(qdio_free);
1254 
1255 /**
1256  * qdio_allocate - allocate qdio queues and associated data
1257  * @init_data: initialization data
1258  */
1259 int qdio_allocate(struct qdio_initialize *init_data)
1260 {
1261 	struct subchannel_id schid;
1262 	struct qdio_irq *irq_ptr;
1263 
1264 	ccw_device_get_schid(init_data->cdev, &schid);
1265 	DBF_EVENT("qallocate:%4x", schid.sch_no);
1266 
1267 	if ((init_data->no_input_qs && !init_data->input_handler) ||
1268 	    (init_data->no_output_qs && !init_data->output_handler))
1269 		return -EINVAL;
1270 
1271 	if ((init_data->no_input_qs > QDIO_MAX_QUEUES_PER_IRQ) ||
1272 	    (init_data->no_output_qs > QDIO_MAX_QUEUES_PER_IRQ))
1273 		return -EINVAL;
1274 
1275 	if ((!init_data->input_sbal_addr_array) ||
1276 	    (!init_data->output_sbal_addr_array))
1277 		return -EINVAL;
1278 
1279 	/* irq_ptr must be in GFP_DMA since it contains ccw1.cda */
1280 	irq_ptr = (void *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1281 	if (!irq_ptr)
1282 		goto out_err;
1283 
1284 	mutex_init(&irq_ptr->setup_mutex);
1285 	if (qdio_allocate_dbf(init_data, irq_ptr))
1286 		goto out_rel;
1287 
1288 	/*
1289 	 * Allocate a page for the chsc calls in qdio_establish.
1290 	 * Must be pre-allocated since a zfcp recovery will call
1291 	 * qdio_establish. In case of low memory and swap on a zfcp disk
1292 	 * we may not be able to allocate memory otherwise.
1293 	 */
1294 	irq_ptr->chsc_page = get_zeroed_page(GFP_KERNEL);
1295 	if (!irq_ptr->chsc_page)
1296 		goto out_rel;
1297 
1298 	/* qdr is used in ccw1.cda which is u32 */
1299 	irq_ptr->qdr = (struct qdr *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1300 	if (!irq_ptr->qdr)
1301 		goto out_rel;
1302 
1303 	if (qdio_allocate_qs(irq_ptr, init_data->no_input_qs,
1304 			     init_data->no_output_qs))
1305 		goto out_rel;
1306 
1307 	init_data->cdev->private->qdio_data = irq_ptr;
1308 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1309 	return 0;
1310 out_rel:
1311 	qdio_release_memory(irq_ptr);
1312 out_err:
1313 	return -ENOMEM;
1314 }
1315 EXPORT_SYMBOL_GPL(qdio_allocate);
1316 
1317 static void qdio_detect_hsicq(struct qdio_irq *irq_ptr)
1318 {
1319 	struct qdio_q *q = irq_ptr->input_qs[0];
1320 	int i, use_cq = 0;
1321 
1322 	if (irq_ptr->nr_input_qs > 1 && queue_type(q) == QDIO_IQDIO_QFMT)
1323 		use_cq = 1;
1324 
1325 	for_each_output_queue(irq_ptr, q, i) {
1326 		if (use_cq) {
1327 			if (qdio_enable_async_operation(&q->u.out) < 0) {
1328 				use_cq = 0;
1329 				continue;
1330 			}
1331 		} else
1332 			qdio_disable_async_operation(&q->u.out);
1333 	}
1334 	DBF_EVENT("use_cq:%d", use_cq);
1335 }
1336 
1337 /**
1338  * qdio_establish - establish queues on a qdio subchannel
1339  * @init_data: initialization data
1340  */
1341 int qdio_establish(struct qdio_initialize *init_data)
1342 {
1343 	struct ccw_device *cdev = init_data->cdev;
1344 	struct subchannel_id schid;
1345 	struct qdio_irq *irq_ptr;
1346 	int rc;
1347 
1348 	ccw_device_get_schid(cdev, &schid);
1349 	DBF_EVENT("qestablish:%4x", schid.sch_no);
1350 
1351 	irq_ptr = cdev->private->qdio_data;
1352 	if (!irq_ptr)
1353 		return -ENODEV;
1354 
1355 	mutex_lock(&irq_ptr->setup_mutex);
1356 	qdio_setup_irq(init_data);
1357 
1358 	rc = qdio_establish_thinint(irq_ptr);
1359 	if (rc) {
1360 		mutex_unlock(&irq_ptr->setup_mutex);
1361 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1362 		return rc;
1363 	}
1364 
1365 	/* establish q */
1366 	irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
1367 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1368 	irq_ptr->ccw.count = irq_ptr->equeue.count;
1369 	irq_ptr->ccw.cda = (u32)((addr_t)irq_ptr->qdr);
1370 
1371 	spin_lock_irq(get_ccwdev_lock(cdev));
1372 	ccw_device_set_options_mask(cdev, 0);
1373 
1374 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ESTABLISH, 0, 0);
1375 	spin_unlock_irq(get_ccwdev_lock(cdev));
1376 	if (rc) {
1377 		DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
1378 		DBF_ERROR("rc:%4x", rc);
1379 		mutex_unlock(&irq_ptr->setup_mutex);
1380 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1381 		return rc;
1382 	}
1383 
1384 	wait_event_interruptible_timeout(cdev->private->wait_q,
1385 		irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
1386 		irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
1387 
1388 	if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
1389 		mutex_unlock(&irq_ptr->setup_mutex);
1390 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1391 		return -EIO;
1392 	}
1393 
1394 	qdio_setup_ssqd_info(irq_ptr);
1395 
1396 	qdio_detect_hsicq(irq_ptr);
1397 
1398 	/* qebsm is now setup if available, initialize buffer states */
1399 	qdio_init_buf_states(irq_ptr);
1400 
1401 	mutex_unlock(&irq_ptr->setup_mutex);
1402 	qdio_print_subchannel_info(irq_ptr, cdev);
1403 	qdio_setup_debug_entries(irq_ptr, cdev);
1404 	return 0;
1405 }
1406 EXPORT_SYMBOL_GPL(qdio_establish);
1407 
1408 /**
1409  * qdio_activate - activate queues on a qdio subchannel
1410  * @cdev: associated cdev
1411  */
1412 int qdio_activate(struct ccw_device *cdev)
1413 {
1414 	struct subchannel_id schid;
1415 	struct qdio_irq *irq_ptr;
1416 	int rc;
1417 
1418 	ccw_device_get_schid(cdev, &schid);
1419 	DBF_EVENT("qactivate:%4x", schid.sch_no);
1420 
1421 	irq_ptr = cdev->private->qdio_data;
1422 	if (!irq_ptr)
1423 		return -ENODEV;
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_irq(get_ccwdev_lock(cdev));
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 	spin_unlock_irq(get_ccwdev_lock(cdev));
1442 	if (rc) {
1443 		DBF_ERROR("%4x act IO ERR", irq_ptr->schid.sch_no);
1444 		DBF_ERROR("rc:%4x", rc);
1445 		goto out;
1446 	}
1447 
1448 	if (is_thinint_irq(irq_ptr))
1449 		tiqdio_add_input_queues(irq_ptr);
1450 
1451 	/* wait for subchannel to become active */
1452 	msleep(5);
1453 
1454 	switch (irq_ptr->state) {
1455 	case QDIO_IRQ_STATE_STOPPED:
1456 	case QDIO_IRQ_STATE_ERR:
1457 		rc = -EIO;
1458 		break;
1459 	default:
1460 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ACTIVE);
1461 		rc = 0;
1462 	}
1463 out:
1464 	mutex_unlock(&irq_ptr->setup_mutex);
1465 	return rc;
1466 }
1467 EXPORT_SYMBOL_GPL(qdio_activate);
1468 
1469 static inline int buf_in_between(int bufnr, int start, int count)
1470 {
1471 	int end = add_buf(start, count);
1472 
1473 	if (end > start) {
1474 		if (bufnr >= start && bufnr < end)
1475 			return 1;
1476 		else
1477 			return 0;
1478 	}
1479 
1480 	/* wrap-around case */
1481 	if ((bufnr >= start && bufnr <= QDIO_MAX_BUFFERS_PER_Q) ||
1482 	    (bufnr < end))
1483 		return 1;
1484 	else
1485 		return 0;
1486 }
1487 
1488 /**
1489  * handle_inbound - reset processed input buffers
1490  * @q: queue containing the buffers
1491  * @callflags: flags
1492  * @bufnr: first buffer to process
1493  * @count: how many buffers are emptied
1494  */
1495 static int handle_inbound(struct qdio_q *q, unsigned int callflags,
1496 			  int bufnr, int count)
1497 {
1498 	int diff;
1499 
1500 	qperf_inc(q, inbound_call);
1501 
1502 	if (!q->u.in.polling)
1503 		goto set;
1504 
1505 	/* protect against stop polling setting an ACK for an emptied slsb */
1506 	if (count == QDIO_MAX_BUFFERS_PER_Q) {
1507 		/* overwriting everything, just delete polling status */
1508 		q->u.in.polling = 0;
1509 		q->u.in.ack_count = 0;
1510 		goto set;
1511 	} else if (buf_in_between(q->u.in.ack_start, bufnr, count)) {
1512 		if (is_qebsm(q)) {
1513 			/* partial overwrite, just update ack_start */
1514 			diff = add_buf(bufnr, count);
1515 			diff = sub_buf(diff, q->u.in.ack_start);
1516 			q->u.in.ack_count -= diff;
1517 			if (q->u.in.ack_count <= 0) {
1518 				q->u.in.polling = 0;
1519 				q->u.in.ack_count = 0;
1520 				goto set;
1521 			}
1522 			q->u.in.ack_start = add_buf(q->u.in.ack_start, diff);
1523 		}
1524 		else
1525 			/* the only ACK will be deleted, so stop polling */
1526 			q->u.in.polling = 0;
1527 	}
1528 
1529 set:
1530 	count = set_buf_states(q, bufnr, SLSB_CU_INPUT_EMPTY, count);
1531 	atomic_add(count, &q->nr_buf_used);
1532 
1533 	if (need_siga_in(q))
1534 		return qdio_siga_input(q);
1535 
1536 	return 0;
1537 }
1538 
1539 /**
1540  * handle_outbound - process filled outbound buffers
1541  * @q: queue containing the buffers
1542  * @callflags: flags
1543  * @bufnr: first buffer to process
1544  * @count: how many buffers are filled
1545  */
1546 static int handle_outbound(struct qdio_q *q, unsigned int callflags,
1547 			   int bufnr, int count)
1548 {
1549 	unsigned char state = 0;
1550 	int used, rc = 0;
1551 
1552 	qperf_inc(q, outbound_call);
1553 
1554 	count = set_buf_states(q, bufnr, SLSB_CU_OUTPUT_PRIMED, count);
1555 	used = atomic_add_return(count, &q->nr_buf_used);
1556 
1557 	if (used == QDIO_MAX_BUFFERS_PER_Q)
1558 		qperf_inc(q, outbound_queue_full);
1559 
1560 	if (callflags & QDIO_FLAG_PCI_OUT) {
1561 		q->u.out.pci_out_enabled = 1;
1562 		qperf_inc(q, pci_request_int);
1563 	} else
1564 		q->u.out.pci_out_enabled = 0;
1565 
1566 	if (queue_type(q) == QDIO_IQDIO_QFMT) {
1567 		unsigned long phys_aob = 0;
1568 
1569 		/* One SIGA-W per buffer required for unicast HSI */
1570 		WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
1571 
1572 		phys_aob = qdio_aob_for_buffer(&q->u.out, bufnr);
1573 
1574 		rc = qdio_kick_outbound_q(q, phys_aob);
1575 	} else if (need_siga_sync(q)) {
1576 		rc = qdio_siga_sync_q(q);
1577 	} else {
1578 		/* try to fast requeue buffers */
1579 		get_buf_state(q, prev_buf(bufnr), &state, 0);
1580 		if (state != SLSB_CU_OUTPUT_PRIMED)
1581 			rc = qdio_kick_outbound_q(q, 0);
1582 		else
1583 			qperf_inc(q, fast_requeue);
1584 	}
1585 
1586 	/* in case of SIGA errors we must process the error immediately */
1587 	if (used >= q->u.out.scan_threshold || rc)
1588 		qdio_tasklet_schedule(q);
1589 	else
1590 		/* free the SBALs in case of no further traffic */
1591 		if (!timer_pending(&q->u.out.timer) &&
1592 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
1593 			mod_timer(&q->u.out.timer, jiffies + HZ);
1594 	return rc;
1595 }
1596 
1597 /**
1598  * do_QDIO - process input or output buffers
1599  * @cdev: associated ccw_device for the qdio subchannel
1600  * @callflags: input or output and special flags from the program
1601  * @q_nr: queue number
1602  * @bufnr: buffer number
1603  * @count: how many buffers to process
1604  */
1605 int do_QDIO(struct ccw_device *cdev, unsigned int callflags,
1606 	    int q_nr, unsigned int bufnr, unsigned int count)
1607 {
1608 	struct qdio_irq *irq_ptr;
1609 
1610 	if (bufnr >= QDIO_MAX_BUFFERS_PER_Q || count > QDIO_MAX_BUFFERS_PER_Q)
1611 		return -EINVAL;
1612 
1613 	irq_ptr = cdev->private->qdio_data;
1614 	if (!irq_ptr)
1615 		return -ENODEV;
1616 
1617 	DBF_DEV_EVENT(DBF_INFO, irq_ptr,
1618 		      "do%02x b:%02x c:%02x", callflags, bufnr, count);
1619 
1620 	if (irq_ptr->state != QDIO_IRQ_STATE_ACTIVE)
1621 		return -EIO;
1622 	if (!count)
1623 		return 0;
1624 	if (callflags & QDIO_FLAG_SYNC_INPUT)
1625 		return handle_inbound(irq_ptr->input_qs[q_nr],
1626 				      callflags, bufnr, count);
1627 	else if (callflags & QDIO_FLAG_SYNC_OUTPUT)
1628 		return handle_outbound(irq_ptr->output_qs[q_nr],
1629 				       callflags, bufnr, count);
1630 	return -EINVAL;
1631 }
1632 EXPORT_SYMBOL_GPL(do_QDIO);
1633 
1634 /**
1635  * qdio_start_irq - process input buffers
1636  * @cdev: associated ccw_device for the qdio subchannel
1637  * @nr: input queue number
1638  *
1639  * Return codes
1640  *   0 - success
1641  *   1 - irqs not started since new data is available
1642  */
1643 int qdio_start_irq(struct ccw_device *cdev, int nr)
1644 {
1645 	struct qdio_q *q;
1646 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1647 
1648 	if (!irq_ptr)
1649 		return -ENODEV;
1650 	q = irq_ptr->input_qs[nr];
1651 
1652 	clear_nonshared_ind(irq_ptr);
1653 	qdio_stop_polling(q);
1654 	clear_bit(QDIO_QUEUE_IRQS_DISABLED, &q->u.in.queue_irq_state);
1655 
1656 	/*
1657 	 * We need to check again to not lose initiative after
1658 	 * resetting the ACK state.
1659 	 */
1660 	if (test_nonshared_ind(irq_ptr))
1661 		goto rescan;
1662 	if (!qdio_inbound_q_done(q))
1663 		goto rescan;
1664 	return 0;
1665 
1666 rescan:
1667 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1668 			     &q->u.in.queue_irq_state))
1669 		return 0;
1670 	else
1671 		return 1;
1672 
1673 }
1674 EXPORT_SYMBOL(qdio_start_irq);
1675 
1676 /**
1677  * qdio_get_next_buffers - process input buffers
1678  * @cdev: associated ccw_device for the qdio subchannel
1679  * @nr: input queue number
1680  * @bufnr: first filled buffer number
1681  * @error: buffers are in error state
1682  *
1683  * Return codes
1684  *   < 0 - error
1685  *   = 0 - no new buffers found
1686  *   > 0 - number of processed buffers
1687  */
1688 int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
1689 			  int *error)
1690 {
1691 	struct qdio_q *q;
1692 	int start, end;
1693 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1694 
1695 	if (!irq_ptr)
1696 		return -ENODEV;
1697 	q = irq_ptr->input_qs[nr];
1698 
1699 	/*
1700 	 * Cannot rely on automatic sync after interrupt since queues may
1701 	 * also be examined without interrupt.
1702 	 */
1703 	if (need_siga_sync(q))
1704 		qdio_sync_queues(q);
1705 
1706 	/* check the PCI capable outbound queues. */
1707 	qdio_check_outbound_after_thinint(q);
1708 
1709 	if (!qdio_inbound_q_moved(q))
1710 		return 0;
1711 
1712 	/* Note: upper-layer MUST stop processing immediately here ... */
1713 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
1714 		return -EIO;
1715 
1716 	start = q->first_to_kick;
1717 	end = q->first_to_check;
1718 	*bufnr = start;
1719 	*error = q->qdio_error;
1720 
1721 	/* for the next time */
1722 	q->first_to_kick = end;
1723 	q->qdio_error = 0;
1724 	return sub_buf(end, start);
1725 }
1726 EXPORT_SYMBOL(qdio_get_next_buffers);
1727 
1728 /**
1729  * qdio_stop_irq - disable interrupt processing for the device
1730  * @cdev: associated ccw_device for the qdio subchannel
1731  * @nr: input queue number
1732  *
1733  * Return codes
1734  *   0 - interrupts were already disabled
1735  *   1 - interrupts successfully disabled
1736  */
1737 int qdio_stop_irq(struct ccw_device *cdev, int nr)
1738 {
1739 	struct qdio_q *q;
1740 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1741 
1742 	if (!irq_ptr)
1743 		return -ENODEV;
1744 	q = irq_ptr->input_qs[nr];
1745 
1746 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1747 			     &q->u.in.queue_irq_state))
1748 		return 0;
1749 	else
1750 		return 1;
1751 }
1752 EXPORT_SYMBOL(qdio_stop_irq);
1753 
1754 /**
1755  * qdio_pnso_brinfo() - perform network subchannel op #0 - bridge info.
1756  * @schid:		Subchannel ID.
1757  * @cnc:		Boolean Change-Notification Control
1758  * @response:		Response code will be stored at this address
1759  * @cb: 		Callback function will be executed for each element
1760  *			of the address list
1761  * @priv:		Pointer passed from the caller to qdio_pnso_brinfo()
1762  * @type:		Type of the address entry passed to the callback
1763  * @entry:		Entry containg the address of the specified type
1764  * @priv:		Pointer to pass to the callback function.
1765  *
1766  * Performs "Store-network-bridging-information list" operation and calls
1767  * the callback function for every entry in the list. If "change-
1768  * notification-control" is set, further changes in the address list
1769  * will be reported via the IPA command.
1770  */
1771 int qdio_pnso_brinfo(struct subchannel_id schid,
1772 		int cnc, u16 *response,
1773 		void (*cb)(void *priv, enum qdio_brinfo_entry_type type,
1774 				void *entry),
1775 		void *priv)
1776 {
1777 	struct chsc_pnso_area *rr;
1778 	int rc;
1779 	u32 prev_instance = 0;
1780 	int isfirstblock = 1;
1781 	int i, size, elems;
1782 
1783 	rr = (struct chsc_pnso_area *)get_zeroed_page(GFP_KERNEL);
1784 	if (rr == NULL)
1785 		return -ENOMEM;
1786 	do {
1787 		/* on the first iteration, naihdr.resume_token will be zero */
1788 		rc = chsc_pnso_brinfo(schid, rr, rr->naihdr.resume_token, cnc);
1789 		if (rc != 0 && rc != -EBUSY)
1790 			goto out;
1791 		if (rr->response.code != 1) {
1792 			rc = -EIO;
1793 			continue;
1794 		} else
1795 			rc = 0;
1796 
1797 		if (cb == NULL)
1798 			continue;
1799 
1800 		size = rr->naihdr.naids;
1801 		elems = (rr->response.length -
1802 				sizeof(struct chsc_header) -
1803 				sizeof(struct chsc_brinfo_naihdr)) /
1804 				size;
1805 
1806 		if (!isfirstblock && (rr->naihdr.instance != prev_instance)) {
1807 			/* Inform the caller that they need to scrap */
1808 			/* the data that was already reported via cb */
1809 				rc = -EAGAIN;
1810 				break;
1811 		}
1812 		isfirstblock = 0;
1813 		prev_instance = rr->naihdr.instance;
1814 		for (i = 0; i < elems; i++)
1815 			switch (size) {
1816 			case sizeof(struct qdio_brinfo_entry_l3_ipv6):
1817 				(*cb)(priv, l3_ipv6_addr,
1818 						&rr->entries.l3_ipv6[i]);
1819 				break;
1820 			case sizeof(struct qdio_brinfo_entry_l3_ipv4):
1821 				(*cb)(priv, l3_ipv4_addr,
1822 						&rr->entries.l3_ipv4[i]);
1823 				break;
1824 			case sizeof(struct qdio_brinfo_entry_l2):
1825 				(*cb)(priv, l2_addr_lnid,
1826 						&rr->entries.l2[i]);
1827 				break;
1828 			default:
1829 				WARN_ON_ONCE(1);
1830 				rc = -EIO;
1831 				goto out;
1832 			}
1833 	} while (rr->response.code == 0x0107 ||  /* channel busy */
1834 		  (rr->response.code == 1 && /* list stored */
1835 		   /* resume token is non-zero => list incomplete */
1836 		   (rr->naihdr.resume_token.t1 || rr->naihdr.resume_token.t2)));
1837 	(*response) = rr->response.code;
1838 
1839 out:
1840 	free_page((unsigned long)rr);
1841 	return rc;
1842 }
1843 EXPORT_SYMBOL_GPL(qdio_pnso_brinfo);
1844 
1845 static int __init init_QDIO(void)
1846 {
1847 	int rc;
1848 
1849 	rc = qdio_debug_init();
1850 	if (rc)
1851 		return rc;
1852 	rc = qdio_setup_init();
1853 	if (rc)
1854 		goto out_debug;
1855 	rc = tiqdio_allocate_memory();
1856 	if (rc)
1857 		goto out_cache;
1858 	rc = tiqdio_register_thinints();
1859 	if (rc)
1860 		goto out_ti;
1861 	return 0;
1862 
1863 out_ti:
1864 	tiqdio_free_memory();
1865 out_cache:
1866 	qdio_setup_exit();
1867 out_debug:
1868 	qdio_debug_exit();
1869 	return rc;
1870 }
1871 
1872 static void __exit exit_QDIO(void)
1873 {
1874 	tiqdio_unregister_thinints();
1875 	tiqdio_free_memory();
1876 	qdio_setup_exit();
1877 	qdio_debug_exit();
1878 }
1879 
1880 module_init(init_QDIO);
1881 module_exit(exit_QDIO);
1882