xref: /linux/drivers/s390/cio/qdio_main.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Linux for s390 qdio support, buffer handling, qdio API and module support.
4  *
5  * Copyright IBM Corp. 2000, 2008
6  * Author(s): Utz Bacher <utz.bacher@de.ibm.com>
7  *	      Jan Glauber <jang@linux.vnet.ibm.com>
8  * 2.6 cio integration by Cornelia Huck <cornelia.huck@de.ibm.com>
9  */
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/timer.h>
14 #include <linux/delay.h>
15 #include <linux/gfp.h>
16 #include <linux/io.h>
17 #include <linux/atomic.h>
18 #include <asm/debug.h>
19 #include <asm/qdio.h>
20 #include <asm/ipl.h>
21 
22 #include "cio.h"
23 #include "css.h"
24 #include "device.h"
25 #include "qdio.h"
26 #include "qdio_debug.h"
27 
28 MODULE_AUTHOR("Utz Bacher <utz.bacher@de.ibm.com>,"\
29 	"Jan Glauber <jang@linux.vnet.ibm.com>");
30 MODULE_DESCRIPTION("QDIO base support");
31 MODULE_LICENSE("GPL");
32 
33 static inline int do_siga_sync(unsigned long schid,
34 			       unsigned int out_mask, unsigned int in_mask,
35 			       unsigned int fc)
36 {
37 	register unsigned long __fc asm ("0") = fc;
38 	register unsigned long __schid asm ("1") = schid;
39 	register unsigned long out asm ("2") = out_mask;
40 	register unsigned long in asm ("3") = in_mask;
41 	int cc;
42 
43 	asm volatile(
44 		"	siga	0\n"
45 		"	ipm	%0\n"
46 		"	srl	%0,28\n"
47 		: "=d" (cc)
48 		: "d" (__fc), "d" (__schid), "d" (out), "d" (in) : "cc");
49 	return cc;
50 }
51 
52 static inline int do_siga_input(unsigned long schid, unsigned int mask,
53 				unsigned int fc)
54 {
55 	register unsigned long __fc asm ("0") = fc;
56 	register unsigned long __schid asm ("1") = schid;
57 	register unsigned long __mask asm ("2") = mask;
58 	int cc;
59 
60 	asm volatile(
61 		"	siga	0\n"
62 		"	ipm	%0\n"
63 		"	srl	%0,28\n"
64 		: "=d" (cc)
65 		: "d" (__fc), "d" (__schid), "d" (__mask) : "cc");
66 	return cc;
67 }
68 
69 /**
70  * do_siga_output - perform SIGA-w/wt function
71  * @schid: subchannel id or in case of QEBSM the subchannel token
72  * @mask: which output queues to process
73  * @bb: busy bit indicator, set only if SIGA-w/wt could not access a buffer
74  * @fc: function code to perform
75  * @aob: asynchronous operation block
76  *
77  * Returns condition code.
78  * Note: For IQDC unicast queues only the highest priority queue is processed.
79  */
80 static inline int do_siga_output(unsigned long schid, unsigned long mask,
81 				 unsigned int *bb, unsigned int fc,
82 				 unsigned long aob)
83 {
84 	register unsigned long __fc asm("0") = fc;
85 	register unsigned long __schid asm("1") = schid;
86 	register unsigned long __mask asm("2") = mask;
87 	register unsigned long __aob asm("3") = aob;
88 	int cc;
89 
90 	asm volatile(
91 		"	siga	0\n"
92 		"	ipm	%0\n"
93 		"	srl	%0,28\n"
94 		: "=d" (cc), "+d" (__fc), "+d" (__aob)
95 		: "d" (__schid), "d" (__mask)
96 		: "cc");
97 	*bb = __fc >> 31;
98 	return cc;
99 }
100 
101 /**
102  * qdio_do_eqbs - extract buffer states for QEBSM
103  * @q: queue to manipulate
104  * @state: state of the extracted buffers
105  * @start: buffer number to start at
106  * @count: count of buffers to examine
107  * @auto_ack: automatically acknowledge buffers
108  *
109  * Returns the number of successfully extracted equal buffer states.
110  * Stops processing if a state is different from the last buffers state.
111  */
112 static int qdio_do_eqbs(struct qdio_q *q, unsigned char *state,
113 			int start, int count, int auto_ack)
114 {
115 	int tmp_count = count, tmp_start = start, nr = q->nr;
116 	unsigned int ccq = 0;
117 
118 	qperf_inc(q, eqbs);
119 
120 	if (!q->is_input_q)
121 		nr += q->irq_ptr->nr_input_qs;
122 again:
123 	ccq = do_eqbs(q->irq_ptr->sch_token, state, nr, &tmp_start, &tmp_count,
124 		      auto_ack);
125 
126 	switch (ccq) {
127 	case 0:
128 	case 32:
129 		/* all done, or next buffer state different */
130 		return count - tmp_count;
131 	case 96:
132 		/* not all buffers processed */
133 		qperf_inc(q, eqbs_partial);
134 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "EQBS part:%02x",
135 			tmp_count);
136 		return count - tmp_count;
137 	case 97:
138 		/* no buffer processed */
139 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "EQBS again:%2d", ccq);
140 		goto again;
141 	default:
142 		DBF_ERROR("%4x ccq:%3d", SCH_NO(q), ccq);
143 		DBF_ERROR("%4x EQBS ERROR", SCH_NO(q));
144 		DBF_ERROR("%3d%3d%2d", count, tmp_count, nr);
145 		q->handler(q->irq_ptr->cdev, QDIO_ERROR_GET_BUF_STATE, q->nr,
146 			   q->first_to_kick, count, q->irq_ptr->int_parm);
147 		return 0;
148 	}
149 }
150 
151 /**
152  * qdio_do_sqbs - set buffer states for QEBSM
153  * @q: queue to manipulate
154  * @state: new state of the buffers
155  * @start: first buffer number to change
156  * @count: how many buffers to change
157  *
158  * Returns the number of successfully changed buffers.
159  * Does retrying until the specified count of buffer states is set or an
160  * error occurs.
161  */
162 static int qdio_do_sqbs(struct qdio_q *q, unsigned char state, int start,
163 			int count)
164 {
165 	unsigned int ccq = 0;
166 	int tmp_count = count, tmp_start = start;
167 	int nr = q->nr;
168 
169 	if (!count)
170 		return 0;
171 	qperf_inc(q, sqbs);
172 
173 	if (!q->is_input_q)
174 		nr += q->irq_ptr->nr_input_qs;
175 again:
176 	ccq = do_sqbs(q->irq_ptr->sch_token, state, nr, &tmp_start, &tmp_count);
177 
178 	switch (ccq) {
179 	case 0:
180 	case 32:
181 		/* all done, or active buffer adapter-owned */
182 		WARN_ON_ONCE(tmp_count);
183 		return count - tmp_count;
184 	case 96:
185 		/* not all buffers processed */
186 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "SQBS again:%2d", ccq);
187 		qperf_inc(q, sqbs_partial);
188 		goto again;
189 	default:
190 		DBF_ERROR("%4x ccq:%3d", SCH_NO(q), ccq);
191 		DBF_ERROR("%4x SQBS ERROR", SCH_NO(q));
192 		DBF_ERROR("%3d%3d%2d", count, tmp_count, nr);
193 		q->handler(q->irq_ptr->cdev, QDIO_ERROR_SET_BUF_STATE, q->nr,
194 			   q->first_to_kick, count, q->irq_ptr->int_parm);
195 		return 0;
196 	}
197 }
198 
199 /*
200  * Returns number of examined buffers and their common state in *state.
201  * Requested number of buffers-to-examine must be > 0.
202  */
203 static inline int get_buf_states(struct qdio_q *q, unsigned int bufnr,
204 				 unsigned char *state, unsigned int count,
205 				 int auto_ack, int merge_pending)
206 {
207 	unsigned char __state = 0;
208 	int i;
209 
210 	if (is_qebsm(q))
211 		return qdio_do_eqbs(q, state, bufnr, count, auto_ack);
212 
213 	/* get initial state: */
214 	__state = q->slsb.val[bufnr];
215 	if (merge_pending && __state == SLSB_P_OUTPUT_PENDING)
216 		__state = SLSB_P_OUTPUT_EMPTY;
217 
218 	for (i = 1; i < count; i++) {
219 		bufnr = next_buf(bufnr);
220 
221 		/* merge PENDING into EMPTY: */
222 		if (merge_pending &&
223 		    q->slsb.val[bufnr] == SLSB_P_OUTPUT_PENDING &&
224 		    __state == SLSB_P_OUTPUT_EMPTY)
225 			continue;
226 
227 		/* stop if next state differs from initial state: */
228 		if (q->slsb.val[bufnr] != __state)
229 			break;
230 	}
231 	*state = __state;
232 	return i;
233 }
234 
235 static inline int get_buf_state(struct qdio_q *q, unsigned int bufnr,
236 				unsigned char *state, int auto_ack)
237 {
238 	return get_buf_states(q, bufnr, state, 1, auto_ack, 0);
239 }
240 
241 /* wrap-around safe setting of slsb states, returns number of changed buffers */
242 static inline int set_buf_states(struct qdio_q *q, int bufnr,
243 				 unsigned char state, int count)
244 {
245 	int i;
246 
247 	if (is_qebsm(q))
248 		return qdio_do_sqbs(q, state, bufnr, count);
249 
250 	for (i = 0; i < count; i++) {
251 		xchg(&q->slsb.val[bufnr], state);
252 		bufnr = next_buf(bufnr);
253 	}
254 	return count;
255 }
256 
257 static inline int set_buf_state(struct qdio_q *q, int bufnr,
258 				unsigned char state)
259 {
260 	return set_buf_states(q, bufnr, state, 1);
261 }
262 
263 /* set slsb states to initial state */
264 static void qdio_init_buf_states(struct qdio_irq *irq_ptr)
265 {
266 	struct qdio_q *q;
267 	int i;
268 
269 	for_each_input_queue(irq_ptr, q, i)
270 		set_buf_states(q, 0, SLSB_P_INPUT_NOT_INIT,
271 			       QDIO_MAX_BUFFERS_PER_Q);
272 	for_each_output_queue(irq_ptr, q, i)
273 		set_buf_states(q, 0, SLSB_P_OUTPUT_NOT_INIT,
274 			       QDIO_MAX_BUFFERS_PER_Q);
275 }
276 
277 static inline int qdio_siga_sync(struct qdio_q *q, unsigned int output,
278 			  unsigned int input)
279 {
280 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
281 	unsigned int fc = QDIO_SIGA_SYNC;
282 	int cc;
283 
284 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-s:%1d", q->nr);
285 	qperf_inc(q, siga_sync);
286 
287 	if (is_qebsm(q)) {
288 		schid = q->irq_ptr->sch_token;
289 		fc |= QDIO_SIGA_QEBSM_FLAG;
290 	}
291 
292 	cc = do_siga_sync(schid, output, input, fc);
293 	if (unlikely(cc))
294 		DBF_ERROR("%4x SIGA-S:%2d", SCH_NO(q), cc);
295 	return (cc) ? -EIO : 0;
296 }
297 
298 static inline int qdio_siga_sync_q(struct qdio_q *q)
299 {
300 	if (q->is_input_q)
301 		return qdio_siga_sync(q, 0, q->mask);
302 	else
303 		return qdio_siga_sync(q, q->mask, 0);
304 }
305 
306 static int qdio_siga_output(struct qdio_q *q, unsigned int *busy_bit,
307 	unsigned long aob)
308 {
309 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
310 	unsigned int fc = QDIO_SIGA_WRITE;
311 	u64 start_time = 0;
312 	int retries = 0, cc;
313 	unsigned long laob = 0;
314 
315 	WARN_ON_ONCE(aob && ((queue_type(q) != QDIO_IQDIO_QFMT) ||
316 			     !q->u.out.use_cq));
317 	if (q->u.out.use_cq && aob != 0) {
318 		fc = QDIO_SIGA_WRITEQ;
319 		laob = aob;
320 	}
321 
322 	if (is_qebsm(q)) {
323 		schid = q->irq_ptr->sch_token;
324 		fc |= QDIO_SIGA_QEBSM_FLAG;
325 	}
326 again:
327 	cc = do_siga_output(schid, q->mask, busy_bit, fc, laob);
328 
329 	/* hipersocket busy condition */
330 	if (unlikely(*busy_bit)) {
331 		retries++;
332 
333 		if (!start_time) {
334 			start_time = get_tod_clock_fast();
335 			goto again;
336 		}
337 		if (get_tod_clock_fast() - start_time < QDIO_BUSY_BIT_PATIENCE)
338 			goto again;
339 	}
340 	if (retries) {
341 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr,
342 			      "%4x cc2 BB1:%1d", SCH_NO(q), q->nr);
343 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "count:%u", retries);
344 	}
345 	return cc;
346 }
347 
348 static inline int qdio_siga_input(struct qdio_q *q)
349 {
350 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
351 	unsigned int fc = QDIO_SIGA_READ;
352 	int cc;
353 
354 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-r:%1d", q->nr);
355 	qperf_inc(q, siga_read);
356 
357 	if (is_qebsm(q)) {
358 		schid = q->irq_ptr->sch_token;
359 		fc |= QDIO_SIGA_QEBSM_FLAG;
360 	}
361 
362 	cc = do_siga_input(schid, q->mask, fc);
363 	if (unlikely(cc))
364 		DBF_ERROR("%4x SIGA-R:%2d", SCH_NO(q), cc);
365 	return (cc) ? -EIO : 0;
366 }
367 
368 #define qdio_siga_sync_out(q) qdio_siga_sync(q, ~0U, 0)
369 #define qdio_siga_sync_all(q) qdio_siga_sync(q, ~0U, ~0U)
370 
371 static inline void qdio_sync_queues(struct qdio_q *q)
372 {
373 	/* PCI capable outbound queues will also be scanned so sync them too */
374 	if (pci_out_supported(q))
375 		qdio_siga_sync_all(q);
376 	else
377 		qdio_siga_sync_q(q);
378 }
379 
380 int debug_get_buf_state(struct qdio_q *q, unsigned int bufnr,
381 			unsigned char *state)
382 {
383 	if (need_siga_sync(q))
384 		qdio_siga_sync_q(q);
385 	return get_buf_states(q, bufnr, state, 1, 0, 0);
386 }
387 
388 static inline void qdio_stop_polling(struct qdio_q *q)
389 {
390 	if (!q->u.in.polling)
391 		return;
392 
393 	q->u.in.polling = 0;
394 	qperf_inc(q, stop_polling);
395 
396 	/* show the card that we are not polling anymore */
397 	if (is_qebsm(q)) {
398 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
399 			       q->u.in.ack_count);
400 		q->u.in.ack_count = 0;
401 	} else
402 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
403 }
404 
405 static inline void account_sbals(struct qdio_q *q, unsigned int count)
406 {
407 	int pos;
408 
409 	q->q_stats.nr_sbal_total += count;
410 	if (count == QDIO_MAX_BUFFERS_MASK) {
411 		q->q_stats.nr_sbals[7]++;
412 		return;
413 	}
414 	pos = ilog2(count);
415 	q->q_stats.nr_sbals[pos]++;
416 }
417 
418 static void process_buffer_error(struct qdio_q *q, int count)
419 {
420 	unsigned char state = (q->is_input_q) ? SLSB_P_INPUT_NOT_INIT :
421 					SLSB_P_OUTPUT_NOT_INIT;
422 
423 	q->qdio_error = QDIO_ERROR_SLSB_STATE;
424 
425 	/* special handling for no target buffer empty */
426 	if (queue_type(q) == QDIO_IQDIO_QFMT && !q->is_input_q &&
427 	    q->sbal[q->first_to_check]->element[15].sflags == 0x10) {
428 		qperf_inc(q, target_full);
429 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "OUTFULL FTC:%02x",
430 			      q->first_to_check);
431 		goto set;
432 	}
433 
434 	DBF_ERROR("%4x BUF ERROR", SCH_NO(q));
435 	DBF_ERROR((q->is_input_q) ? "IN:%2d" : "OUT:%2d", q->nr);
436 	DBF_ERROR("FTC:%3d C:%3d", q->first_to_check, count);
437 	DBF_ERROR("F14:%2x F15:%2x",
438 		  q->sbal[q->first_to_check]->element[14].sflags,
439 		  q->sbal[q->first_to_check]->element[15].sflags);
440 
441 set:
442 	/*
443 	 * Interrupts may be avoided as long as the error is present
444 	 * so change the buffer state immediately to avoid starvation.
445 	 */
446 	set_buf_states(q, q->first_to_check, state, count);
447 }
448 
449 static inline void inbound_primed(struct qdio_q *q, int count)
450 {
451 	int new;
452 
453 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in prim:%1d %02x", q->nr, count);
454 
455 	/* for QEBSM the ACK was already set by EQBS */
456 	if (is_qebsm(q)) {
457 		if (!q->u.in.polling) {
458 			q->u.in.polling = 1;
459 			q->u.in.ack_count = count;
460 			q->u.in.ack_start = q->first_to_check;
461 			return;
462 		}
463 
464 		/* delete the previous ACK's */
465 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
466 			       q->u.in.ack_count);
467 		q->u.in.ack_count = count;
468 		q->u.in.ack_start = q->first_to_check;
469 		return;
470 	}
471 
472 	/*
473 	 * ACK the newest buffer. The ACK will be removed in qdio_stop_polling
474 	 * or by the next inbound run.
475 	 */
476 	new = add_buf(q->first_to_check, count - 1);
477 	if (q->u.in.polling) {
478 		/* reset the previous ACK but first set the new one */
479 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
480 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
481 	} else {
482 		q->u.in.polling = 1;
483 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
484 	}
485 
486 	q->u.in.ack_start = new;
487 	count--;
488 	if (!count)
489 		return;
490 	/* need to change ALL buffers to get more interrupts */
491 	set_buf_states(q, q->first_to_check, SLSB_P_INPUT_NOT_INIT, count);
492 }
493 
494 static int get_inbound_buffer_frontier(struct qdio_q *q)
495 {
496 	unsigned char state = 0;
497 	int count;
498 
499 	q->timestamp = get_tod_clock_fast();
500 
501 	/*
502 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
503 	 * would return 0.
504 	 */
505 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
506 	if (!count)
507 		goto out;
508 
509 	/*
510 	 * No siga sync here, as a PCI or we after a thin interrupt
511 	 * already sync'ed the queues.
512 	 */
513 	count = get_buf_states(q, q->first_to_check, &state, count, 1, 0);
514 	if (!count)
515 		goto out;
516 
517 	switch (state) {
518 	case SLSB_P_INPUT_PRIMED:
519 		inbound_primed(q, count);
520 		q->first_to_check = add_buf(q->first_to_check, count);
521 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
522 			qperf_inc(q, inbound_queue_full);
523 		if (q->irq_ptr->perf_stat_enabled)
524 			account_sbals(q, count);
525 		break;
526 	case SLSB_P_INPUT_ERROR:
527 		process_buffer_error(q, count);
528 		q->first_to_check = add_buf(q->first_to_check, count);
529 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
530 			qperf_inc(q, inbound_queue_full);
531 		if (q->irq_ptr->perf_stat_enabled)
532 			account_sbals_error(q, count);
533 		break;
534 	case SLSB_CU_INPUT_EMPTY:
535 	case SLSB_P_INPUT_NOT_INIT:
536 	case SLSB_P_INPUT_ACK:
537 		if (q->irq_ptr->perf_stat_enabled)
538 			q->q_stats.nr_sbal_nop++;
539 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in nop:%1d %#02x",
540 			q->nr, q->first_to_check);
541 		break;
542 	default:
543 		WARN_ON_ONCE(1);
544 	}
545 out:
546 	return q->first_to_check;
547 }
548 
549 static int qdio_inbound_q_moved(struct qdio_q *q)
550 {
551 	int bufnr;
552 
553 	bufnr = get_inbound_buffer_frontier(q);
554 
555 	if (bufnr != q->last_move) {
556 		q->last_move = bufnr;
557 		if (!is_thinint_irq(q->irq_ptr) && MACHINE_IS_LPAR)
558 			q->u.in.timestamp = get_tod_clock();
559 		return 1;
560 	} else
561 		return 0;
562 }
563 
564 static inline int qdio_inbound_q_done(struct qdio_q *q)
565 {
566 	unsigned char state = 0;
567 
568 	if (!atomic_read(&q->nr_buf_used))
569 		return 1;
570 
571 	if (need_siga_sync(q))
572 		qdio_siga_sync_q(q);
573 	get_buf_state(q, q->first_to_check, &state, 0);
574 
575 	if (state == SLSB_P_INPUT_PRIMED || state == SLSB_P_INPUT_ERROR)
576 		/* more work coming */
577 		return 0;
578 
579 	if (is_thinint_irq(q->irq_ptr))
580 		return 1;
581 
582 	/* don't poll under z/VM */
583 	if (MACHINE_IS_VM)
584 		return 1;
585 
586 	/*
587 	 * At this point we know, that inbound first_to_check
588 	 * has (probably) not moved (see qdio_inbound_processing).
589 	 */
590 	if (get_tod_clock_fast() > q->u.in.timestamp + QDIO_INPUT_THRESHOLD) {
591 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in done:%02x",
592 			      q->first_to_check);
593 		return 1;
594 	} else
595 		return 0;
596 }
597 
598 static inline int contains_aobs(struct qdio_q *q)
599 {
600 	return !q->is_input_q && q->u.out.use_cq;
601 }
602 
603 static inline void qdio_handle_aobs(struct qdio_q *q, int start, int count)
604 {
605 	unsigned char state = 0;
606 	int j, b = start;
607 
608 	if (!contains_aobs(q))
609 		return;
610 
611 	for (j = 0; j < count; ++j) {
612 		get_buf_state(q, b, &state, 0);
613 		if (state == SLSB_P_OUTPUT_PENDING) {
614 			struct qaob *aob = q->u.out.aobs[b];
615 			if (aob == NULL)
616 				continue;
617 
618 			q->u.out.sbal_state[b].flags |=
619 				QDIO_OUTBUF_STATE_FLAG_PENDING;
620 			q->u.out.aobs[b] = NULL;
621 		} else if (state == SLSB_P_OUTPUT_EMPTY) {
622 			q->u.out.sbal_state[b].aob = NULL;
623 		}
624 		b = next_buf(b);
625 	}
626 }
627 
628 static inline unsigned long qdio_aob_for_buffer(struct qdio_output_q *q,
629 					int bufnr)
630 {
631 	unsigned long phys_aob = 0;
632 
633 	if (!q->use_cq)
634 		return 0;
635 
636 	if (!q->aobs[bufnr]) {
637 		struct qaob *aob = qdio_allocate_aob();
638 		q->aobs[bufnr] = aob;
639 	}
640 	if (q->aobs[bufnr]) {
641 		q->sbal_state[bufnr].aob = q->aobs[bufnr];
642 		q->aobs[bufnr]->user1 = (u64) q->sbal_state[bufnr].user;
643 		phys_aob = virt_to_phys(q->aobs[bufnr]);
644 		WARN_ON_ONCE(phys_aob & 0xFF);
645 	}
646 
647 	q->sbal_state[bufnr].flags = 0;
648 	return phys_aob;
649 }
650 
651 static void qdio_kick_handler(struct qdio_q *q)
652 {
653 	int start = q->first_to_kick;
654 	int end = q->first_to_check;
655 	int count;
656 
657 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
658 		return;
659 
660 	count = sub_buf(end, start);
661 
662 	if (q->is_input_q) {
663 		qperf_inc(q, inbound_handler);
664 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "kih s:%02x c:%02x", start, count);
665 	} else {
666 		qperf_inc(q, outbound_handler);
667 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "koh: s:%02x c:%02x",
668 			      start, count);
669 	}
670 
671 	qdio_handle_aobs(q, start, count);
672 
673 	q->handler(q->irq_ptr->cdev, q->qdio_error, q->nr, start, count,
674 		   q->irq_ptr->int_parm);
675 
676 	/* for the next time */
677 	q->first_to_kick = end;
678 	q->qdio_error = 0;
679 }
680 
681 static inline int qdio_tasklet_schedule(struct qdio_q *q)
682 {
683 	if (likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE)) {
684 		tasklet_schedule(&q->tasklet);
685 		return 0;
686 	}
687 	return -EPERM;
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 (!qdio_tasklet_schedule(q))
703 			return;
704 	}
705 
706 	qdio_stop_polling(q);
707 	/*
708 	 * We need to check again to not lose initiative after
709 	 * resetting the ACK state.
710 	 */
711 	if (!qdio_inbound_q_done(q)) {
712 		qperf_inc(q, tasklet_inbound_resched2);
713 		qdio_tasklet_schedule(q);
714 	}
715 }
716 
717 void qdio_inbound_processing(unsigned long data)
718 {
719 	struct qdio_q *q = (struct qdio_q *)data;
720 	__qdio_inbound_processing(q);
721 }
722 
723 static int get_outbound_buffer_frontier(struct qdio_q *q)
724 {
725 	unsigned char state = 0;
726 	int count;
727 
728 	q->timestamp = get_tod_clock_fast();
729 
730 	if (need_siga_sync(q))
731 		if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
732 		    !pci_out_supported(q)) ||
733 		    (queue_type(q) == QDIO_IQDIO_QFMT &&
734 		    multicast_outbound(q)))
735 			qdio_siga_sync_q(q);
736 
737 	/*
738 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
739 	 * would return 0.
740 	 */
741 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
742 	if (!count)
743 		goto out;
744 
745 	count = get_buf_states(q, q->first_to_check, &state, count, 0,
746 			       q->u.out.use_cq);
747 	if (!count)
748 		goto out;
749 
750 	switch (state) {
751 	case SLSB_P_OUTPUT_EMPTY:
752 		/* the adapter got it */
753 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr,
754 			"out empty:%1d %02x", q->nr, count);
755 
756 		atomic_sub(count, &q->nr_buf_used);
757 		q->first_to_check = add_buf(q->first_to_check, count);
758 		if (q->irq_ptr->perf_stat_enabled)
759 			account_sbals(q, count);
760 
761 		break;
762 	case SLSB_P_OUTPUT_ERROR:
763 		process_buffer_error(q, count);
764 		q->first_to_check = add_buf(q->first_to_check, count);
765 		atomic_sub(count, &q->nr_buf_used);
766 		if (q->irq_ptr->perf_stat_enabled)
767 			account_sbals_error(q, count);
768 		break;
769 	case SLSB_CU_OUTPUT_PRIMED:
770 		/* the adapter has not fetched the output yet */
771 		if (q->irq_ptr->perf_stat_enabled)
772 			q->q_stats.nr_sbal_nop++;
773 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out primed:%1d",
774 			      q->nr);
775 		break;
776 	case SLSB_P_OUTPUT_NOT_INIT:
777 	case SLSB_P_OUTPUT_HALTED:
778 		break;
779 	default:
780 		WARN_ON_ONCE(1);
781 	}
782 
783 out:
784 	return q->first_to_check;
785 }
786 
787 /* all buffers processed? */
788 static inline int qdio_outbound_q_done(struct qdio_q *q)
789 {
790 	return atomic_read(&q->nr_buf_used) == 0;
791 }
792 
793 static inline int qdio_outbound_q_moved(struct qdio_q *q)
794 {
795 	int bufnr;
796 
797 	bufnr = get_outbound_buffer_frontier(q);
798 
799 	if (bufnr != q->last_move) {
800 		q->last_move = bufnr;
801 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out moved:%1d", q->nr);
802 		return 1;
803 	} else
804 		return 0;
805 }
806 
807 static int qdio_kick_outbound_q(struct qdio_q *q, unsigned long aob)
808 {
809 	int retries = 0, cc;
810 	unsigned int busy_bit;
811 
812 	if (!need_siga_out(q))
813 		return 0;
814 
815 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w:%1d", q->nr);
816 retry:
817 	qperf_inc(q, siga_write);
818 
819 	cc = qdio_siga_output(q, &busy_bit, aob);
820 	switch (cc) {
821 	case 0:
822 		break;
823 	case 2:
824 		if (busy_bit) {
825 			while (++retries < QDIO_BUSY_BIT_RETRIES) {
826 				mdelay(QDIO_BUSY_BIT_RETRY_DELAY);
827 				goto retry;
828 			}
829 			DBF_ERROR("%4x cc2 BBC:%1d", SCH_NO(q), q->nr);
830 			cc = -EBUSY;
831 		} else {
832 			DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w cc2:%1d", q->nr);
833 			cc = -ENOBUFS;
834 		}
835 		break;
836 	case 1:
837 	case 3:
838 		DBF_ERROR("%4x SIGA-W:%1d", SCH_NO(q), cc);
839 		cc = -EIO;
840 		break;
841 	}
842 	if (retries) {
843 		DBF_ERROR("%4x cc2 BB2:%1d", SCH_NO(q), q->nr);
844 		DBF_ERROR("count:%u", retries);
845 	}
846 	return cc;
847 }
848 
849 static void __qdio_outbound_processing(struct qdio_q *q)
850 {
851 	qperf_inc(q, tasklet_outbound);
852 	WARN_ON_ONCE(atomic_read(&q->nr_buf_used) < 0);
853 
854 	if (qdio_outbound_q_moved(q))
855 		qdio_kick_handler(q);
856 
857 	if (queue_type(q) == QDIO_ZFCP_QFMT)
858 		if (!pci_out_supported(q) && !qdio_outbound_q_done(q))
859 			goto sched;
860 
861 	if (q->u.out.pci_out_enabled)
862 		return;
863 
864 	/*
865 	 * Now we know that queue type is either qeth without pci enabled
866 	 * or HiperSockets. Make sure buffer switch from PRIMED to EMPTY
867 	 * is noticed and outbound_handler is called after some time.
868 	 */
869 	if (qdio_outbound_q_done(q))
870 		del_timer_sync(&q->u.out.timer);
871 	else
872 		if (!timer_pending(&q->u.out.timer) &&
873 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
874 			mod_timer(&q->u.out.timer, jiffies + 10 * HZ);
875 	return;
876 
877 sched:
878 	qdio_tasklet_schedule(q);
879 }
880 
881 /* outbound tasklet */
882 void qdio_outbound_processing(unsigned long data)
883 {
884 	struct qdio_q *q = (struct qdio_q *)data;
885 	__qdio_outbound_processing(q);
886 }
887 
888 void qdio_outbound_timer(struct timer_list *t)
889 {
890 	struct qdio_q *q = from_timer(q, t, u.out.timer);
891 
892 	qdio_tasklet_schedule(q);
893 }
894 
895 static inline void qdio_check_outbound_after_thinint(struct qdio_q *q)
896 {
897 	struct qdio_q *out;
898 	int i;
899 
900 	if (!pci_out_supported(q))
901 		return;
902 
903 	for_each_output_queue(q->irq_ptr, out, i)
904 		if (!qdio_outbound_q_done(out))
905 			qdio_tasklet_schedule(out);
906 }
907 
908 static void __tiqdio_inbound_processing(struct qdio_q *q)
909 {
910 	qperf_inc(q, tasklet_inbound);
911 	if (need_siga_sync(q) && need_siga_sync_after_ai(q))
912 		qdio_sync_queues(q);
913 
914 	/*
915 	 * The interrupt could be caused by a PCI request. Check the
916 	 * PCI capable outbound queues.
917 	 */
918 	qdio_check_outbound_after_thinint(q);
919 
920 	if (!qdio_inbound_q_moved(q))
921 		return;
922 
923 	qdio_kick_handler(q);
924 
925 	if (!qdio_inbound_q_done(q)) {
926 		qperf_inc(q, tasklet_inbound_resched);
927 		if (!qdio_tasklet_schedule(q))
928 			return;
929 	}
930 
931 	qdio_stop_polling(q);
932 	/*
933 	 * We need to check again to not lose initiative after
934 	 * resetting the ACK state.
935 	 */
936 	if (!qdio_inbound_q_done(q)) {
937 		qperf_inc(q, tasklet_inbound_resched2);
938 		qdio_tasklet_schedule(q);
939 	}
940 }
941 
942 void tiqdio_inbound_processing(unsigned long data)
943 {
944 	struct qdio_q *q = (struct qdio_q *)data;
945 	__tiqdio_inbound_processing(q);
946 }
947 
948 static inline void qdio_set_state(struct qdio_irq *irq_ptr,
949 				  enum qdio_irq_states state)
950 {
951 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "newstate: %1d", state);
952 
953 	irq_ptr->state = state;
954 	mb();
955 }
956 
957 static void qdio_irq_check_sense(struct qdio_irq *irq_ptr, struct irb *irb)
958 {
959 	if (irb->esw.esw0.erw.cons) {
960 		DBF_ERROR("%4x sense:", irq_ptr->schid.sch_no);
961 		DBF_ERROR_HEX(irb, 64);
962 		DBF_ERROR_HEX(irb->ecw, 64);
963 	}
964 }
965 
966 /* PCI interrupt handler */
967 static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
968 {
969 	int i;
970 	struct qdio_q *q;
971 
972 	if (unlikely(irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
973 		return;
974 
975 	for_each_input_queue(irq_ptr, q, i) {
976 		if (q->u.in.queue_start_poll) {
977 			/* skip if polling is enabled or already in work */
978 			if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
979 				     &q->u.in.queue_irq_state)) {
980 				qperf_inc(q, int_discarded);
981 				continue;
982 			}
983 			q->u.in.queue_start_poll(q->irq_ptr->cdev, q->nr,
984 						 q->irq_ptr->int_parm);
985 		} else {
986 			tasklet_schedule(&q->tasklet);
987 		}
988 	}
989 
990 	if (!(irq_ptr->qib.ac & QIB_AC_OUTBOUND_PCI_SUPPORTED))
991 		return;
992 
993 	for_each_output_queue(irq_ptr, q, i) {
994 		if (qdio_outbound_q_done(q))
995 			continue;
996 		if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
997 			qdio_siga_sync_q(q);
998 		qdio_tasklet_schedule(q);
999 	}
1000 }
1001 
1002 static void qdio_handle_activate_check(struct ccw_device *cdev,
1003 				unsigned long intparm, int cstat, int dstat)
1004 {
1005 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1006 	struct qdio_q *q;
1007 	int count;
1008 
1009 	DBF_ERROR("%4x ACT CHECK", irq_ptr->schid.sch_no);
1010 	DBF_ERROR("intp :%lx", intparm);
1011 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1012 
1013 	if (irq_ptr->nr_input_qs) {
1014 		q = irq_ptr->input_qs[0];
1015 	} else if (irq_ptr->nr_output_qs) {
1016 		q = irq_ptr->output_qs[0];
1017 	} else {
1018 		dump_stack();
1019 		goto no_handler;
1020 	}
1021 
1022 	count = sub_buf(q->first_to_check, q->first_to_kick);
1023 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_ACTIVATE,
1024 		   q->nr, q->first_to_kick, count, irq_ptr->int_parm);
1025 no_handler:
1026 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1027 	/*
1028 	 * In case of z/VM LGR (Live Guest Migration) QDIO recovery will happen.
1029 	 * Therefore we call the LGR detection function here.
1030 	 */
1031 	lgr_info_log();
1032 }
1033 
1034 static void qdio_establish_handle_irq(struct ccw_device *cdev, int cstat,
1035 				      int dstat)
1036 {
1037 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1038 
1039 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "qest irq");
1040 
1041 	if (cstat)
1042 		goto error;
1043 	if (dstat & ~(DEV_STAT_DEV_END | DEV_STAT_CHN_END))
1044 		goto error;
1045 	if (!(dstat & DEV_STAT_DEV_END))
1046 		goto error;
1047 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ESTABLISHED);
1048 	return;
1049 
1050 error:
1051 	DBF_ERROR("%4x EQ:error", irq_ptr->schid.sch_no);
1052 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1053 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1054 }
1055 
1056 /* qdio interrupt handler */
1057 void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
1058 		      struct irb *irb)
1059 {
1060 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1061 	struct subchannel_id schid;
1062 	int cstat, dstat;
1063 
1064 	if (!intparm || !irq_ptr) {
1065 		ccw_device_get_schid(cdev, &schid);
1066 		DBF_ERROR("qint:%4x", schid.sch_no);
1067 		return;
1068 	}
1069 
1070 	if (irq_ptr->perf_stat_enabled)
1071 		irq_ptr->perf_stat.qdio_int++;
1072 
1073 	if (IS_ERR(irb)) {
1074 		DBF_ERROR("%4x IO error", irq_ptr->schid.sch_no);
1075 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1076 		wake_up(&cdev->private->wait_q);
1077 		return;
1078 	}
1079 	qdio_irq_check_sense(irq_ptr, irb);
1080 	cstat = irb->scsw.cmd.cstat;
1081 	dstat = irb->scsw.cmd.dstat;
1082 
1083 	switch (irq_ptr->state) {
1084 	case QDIO_IRQ_STATE_INACTIVE:
1085 		qdio_establish_handle_irq(cdev, cstat, dstat);
1086 		break;
1087 	case QDIO_IRQ_STATE_CLEANUP:
1088 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1089 		break;
1090 	case QDIO_IRQ_STATE_ESTABLISHED:
1091 	case QDIO_IRQ_STATE_ACTIVE:
1092 		if (cstat & SCHN_STAT_PCI) {
1093 			qdio_int_handler_pci(irq_ptr);
1094 			return;
1095 		}
1096 		if (cstat || dstat)
1097 			qdio_handle_activate_check(cdev, intparm, cstat,
1098 						   dstat);
1099 		break;
1100 	case QDIO_IRQ_STATE_STOPPED:
1101 		break;
1102 	default:
1103 		WARN_ON_ONCE(1);
1104 	}
1105 	wake_up(&cdev->private->wait_q);
1106 }
1107 
1108 /**
1109  * qdio_get_ssqd_desc - get qdio subchannel description
1110  * @cdev: ccw device to get description for
1111  * @data: where to store the ssqd
1112  *
1113  * Returns 0 or an error code. The results of the chsc are stored in the
1114  * specified structure.
1115  */
1116 int qdio_get_ssqd_desc(struct ccw_device *cdev,
1117 		       struct qdio_ssqd_desc *data)
1118 {
1119 	struct subchannel_id schid;
1120 
1121 	if (!cdev || !cdev->private)
1122 		return -EINVAL;
1123 
1124 	ccw_device_get_schid(cdev, &schid);
1125 	DBF_EVENT("get ssqd:%4x", schid.sch_no);
1126 	return qdio_setup_get_ssqd(NULL, &schid, data);
1127 }
1128 EXPORT_SYMBOL_GPL(qdio_get_ssqd_desc);
1129 
1130 static void qdio_shutdown_queues(struct ccw_device *cdev)
1131 {
1132 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1133 	struct qdio_q *q;
1134 	int i;
1135 
1136 	for_each_input_queue(irq_ptr, q, i)
1137 		tasklet_kill(&q->tasklet);
1138 
1139 	for_each_output_queue(irq_ptr, q, i) {
1140 		del_timer_sync(&q->u.out.timer);
1141 		tasklet_kill(&q->tasklet);
1142 	}
1143 }
1144 
1145 /**
1146  * qdio_shutdown - shut down a qdio subchannel
1147  * @cdev: associated ccw device
1148  * @how: use halt or clear to shutdown
1149  */
1150 int qdio_shutdown(struct ccw_device *cdev, int how)
1151 {
1152 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1153 	struct subchannel_id schid;
1154 	int rc;
1155 
1156 	if (!irq_ptr)
1157 		return -ENODEV;
1158 
1159 	WARN_ON_ONCE(irqs_disabled());
1160 	ccw_device_get_schid(cdev, &schid);
1161 	DBF_EVENT("qshutdown:%4x", schid.sch_no);
1162 
1163 	mutex_lock(&irq_ptr->setup_mutex);
1164 	/*
1165 	 * Subchannel was already shot down. We cannot prevent being called
1166 	 * twice since cio may trigger a shutdown asynchronously.
1167 	 */
1168 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1169 		mutex_unlock(&irq_ptr->setup_mutex);
1170 		return 0;
1171 	}
1172 
1173 	/*
1174 	 * Indicate that the device is going down. Scheduling the queue
1175 	 * tasklets is forbidden from here on.
1176 	 */
1177 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1178 
1179 	tiqdio_remove_input_queues(irq_ptr);
1180 	qdio_shutdown_queues(cdev);
1181 	qdio_shutdown_debug_entries(irq_ptr);
1182 
1183 	/* cleanup subchannel */
1184 	spin_lock_irq(get_ccwdev_lock(cdev));
1185 
1186 	if (how & QDIO_FLAG_CLEANUP_USING_CLEAR)
1187 		rc = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
1188 	else
1189 		/* default behaviour is halt */
1190 		rc = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
1191 	if (rc) {
1192 		DBF_ERROR("%4x SHUTD ERR", irq_ptr->schid.sch_no);
1193 		DBF_ERROR("rc:%4d", rc);
1194 		goto no_cleanup;
1195 	}
1196 
1197 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
1198 	spin_unlock_irq(get_ccwdev_lock(cdev));
1199 	wait_event_interruptible_timeout(cdev->private->wait_q,
1200 		irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
1201 		irq_ptr->state == QDIO_IRQ_STATE_ERR,
1202 		10 * HZ);
1203 	spin_lock_irq(get_ccwdev_lock(cdev));
1204 
1205 no_cleanup:
1206 	qdio_shutdown_thinint(irq_ptr);
1207 
1208 	/* restore interrupt handler */
1209 	if ((void *)cdev->handler == (void *)qdio_int_handler) {
1210 		cdev->handler = irq_ptr->orig_handler;
1211 		cdev->private->intparm = 0;
1212 	}
1213 	spin_unlock_irq(get_ccwdev_lock(cdev));
1214 
1215 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1216 	mutex_unlock(&irq_ptr->setup_mutex);
1217 	if (rc)
1218 		return rc;
1219 	return 0;
1220 }
1221 EXPORT_SYMBOL_GPL(qdio_shutdown);
1222 
1223 /**
1224  * qdio_free - free data structures for a qdio subchannel
1225  * @cdev: associated ccw device
1226  */
1227 int qdio_free(struct ccw_device *cdev)
1228 {
1229 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1230 	struct subchannel_id schid;
1231 
1232 	if (!irq_ptr)
1233 		return -ENODEV;
1234 
1235 	ccw_device_get_schid(cdev, &schid);
1236 	DBF_EVENT("qfree:%4x", schid.sch_no);
1237 	DBF_DEV_EVENT(DBF_ERR, irq_ptr, "dbf abandoned");
1238 	mutex_lock(&irq_ptr->setup_mutex);
1239 
1240 	irq_ptr->debug_area = NULL;
1241 	cdev->private->qdio_data = NULL;
1242 	mutex_unlock(&irq_ptr->setup_mutex);
1243 
1244 	qdio_release_memory(irq_ptr);
1245 	return 0;
1246 }
1247 EXPORT_SYMBOL_GPL(qdio_free);
1248 
1249 /**
1250  * qdio_allocate - allocate qdio queues and associated data
1251  * @init_data: initialization data
1252  */
1253 int qdio_allocate(struct qdio_initialize *init_data)
1254 {
1255 	struct subchannel_id schid;
1256 	struct qdio_irq *irq_ptr;
1257 
1258 	ccw_device_get_schid(init_data->cdev, &schid);
1259 	DBF_EVENT("qallocate:%4x", 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 	if (qdio_allocate_dbf(init_data, irq_ptr))
1280 		goto out_rel;
1281 
1282 	/*
1283 	 * Allocate a page for the chsc calls in qdio_establish.
1284 	 * Must be pre-allocated since a zfcp recovery will call
1285 	 * qdio_establish. In case of low memory and swap on a zfcp disk
1286 	 * we may not be able to allocate memory otherwise.
1287 	 */
1288 	irq_ptr->chsc_page = get_zeroed_page(GFP_KERNEL);
1289 	if (!irq_ptr->chsc_page)
1290 		goto out_rel;
1291 
1292 	/* qdr is used in ccw1.cda which is u32 */
1293 	irq_ptr->qdr = (struct qdr *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1294 	if (!irq_ptr->qdr)
1295 		goto out_rel;
1296 
1297 	if (qdio_allocate_qs(irq_ptr, init_data->no_input_qs,
1298 			     init_data->no_output_qs))
1299 		goto out_rel;
1300 
1301 	init_data->cdev->private->qdio_data = irq_ptr;
1302 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1303 	return 0;
1304 out_rel:
1305 	qdio_release_memory(irq_ptr);
1306 out_err:
1307 	return -ENOMEM;
1308 }
1309 EXPORT_SYMBOL_GPL(qdio_allocate);
1310 
1311 static void qdio_detect_hsicq(struct qdio_irq *irq_ptr)
1312 {
1313 	struct qdio_q *q = irq_ptr->input_qs[0];
1314 	int i, use_cq = 0;
1315 
1316 	if (irq_ptr->nr_input_qs > 1 && queue_type(q) == QDIO_IQDIO_QFMT)
1317 		use_cq = 1;
1318 
1319 	for_each_output_queue(irq_ptr, q, i) {
1320 		if (use_cq) {
1321 			if (qdio_enable_async_operation(&q->u.out) < 0) {
1322 				use_cq = 0;
1323 				continue;
1324 			}
1325 		} else
1326 			qdio_disable_async_operation(&q->u.out);
1327 	}
1328 	DBF_EVENT("use_cq:%d", use_cq);
1329 }
1330 
1331 /**
1332  * qdio_establish - establish queues on a qdio subchannel
1333  * @init_data: initialization data
1334  */
1335 int qdio_establish(struct qdio_initialize *init_data)
1336 {
1337 	struct ccw_device *cdev = init_data->cdev;
1338 	struct subchannel_id schid;
1339 	struct qdio_irq *irq_ptr;
1340 	int rc;
1341 
1342 	ccw_device_get_schid(cdev, &schid);
1343 	DBF_EVENT("qestablish:%4x", schid.sch_no);
1344 
1345 	irq_ptr = cdev->private->qdio_data;
1346 	if (!irq_ptr)
1347 		return -ENODEV;
1348 
1349 	mutex_lock(&irq_ptr->setup_mutex);
1350 	qdio_setup_irq(init_data);
1351 
1352 	rc = qdio_establish_thinint(irq_ptr);
1353 	if (rc) {
1354 		mutex_unlock(&irq_ptr->setup_mutex);
1355 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1356 		return rc;
1357 	}
1358 
1359 	/* establish q */
1360 	irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
1361 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1362 	irq_ptr->ccw.count = irq_ptr->equeue.count;
1363 	irq_ptr->ccw.cda = (u32)((addr_t)irq_ptr->qdr);
1364 
1365 	spin_lock_irq(get_ccwdev_lock(cdev));
1366 	ccw_device_set_options_mask(cdev, 0);
1367 
1368 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ESTABLISH, 0, 0);
1369 	spin_unlock_irq(get_ccwdev_lock(cdev));
1370 	if (rc) {
1371 		DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
1372 		DBF_ERROR("rc:%4x", rc);
1373 		mutex_unlock(&irq_ptr->setup_mutex);
1374 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1375 		return rc;
1376 	}
1377 
1378 	wait_event_interruptible_timeout(cdev->private->wait_q,
1379 		irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
1380 		irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
1381 
1382 	if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
1383 		mutex_unlock(&irq_ptr->setup_mutex);
1384 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1385 		return -EIO;
1386 	}
1387 
1388 	qdio_setup_ssqd_info(irq_ptr);
1389 
1390 	qdio_detect_hsicq(irq_ptr);
1391 
1392 	/* qebsm is now setup if available, initialize buffer states */
1393 	qdio_init_buf_states(irq_ptr);
1394 
1395 	mutex_unlock(&irq_ptr->setup_mutex);
1396 	qdio_print_subchannel_info(irq_ptr, cdev);
1397 	qdio_setup_debug_entries(irq_ptr, cdev);
1398 	return 0;
1399 }
1400 EXPORT_SYMBOL_GPL(qdio_establish);
1401 
1402 /**
1403  * qdio_activate - activate queues on a qdio subchannel
1404  * @cdev: associated cdev
1405  */
1406 int qdio_activate(struct ccw_device *cdev)
1407 {
1408 	struct subchannel_id schid;
1409 	struct qdio_irq *irq_ptr;
1410 	int rc;
1411 
1412 	ccw_device_get_schid(cdev, &schid);
1413 	DBF_EVENT("qactivate:%4x", schid.sch_no);
1414 
1415 	irq_ptr = cdev->private->qdio_data;
1416 	if (!irq_ptr)
1417 		return -ENODEV;
1418 
1419 	mutex_lock(&irq_ptr->setup_mutex);
1420 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1421 		rc = -EBUSY;
1422 		goto out;
1423 	}
1424 
1425 	irq_ptr->ccw.cmd_code = irq_ptr->aqueue.cmd;
1426 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1427 	irq_ptr->ccw.count = irq_ptr->aqueue.count;
1428 	irq_ptr->ccw.cda = 0;
1429 
1430 	spin_lock_irq(get_ccwdev_lock(cdev));
1431 	ccw_device_set_options(cdev, CCWDEV_REPORT_ALL);
1432 
1433 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ACTIVATE,
1434 			      0, DOIO_DENY_PREFETCH);
1435 	spin_unlock_irq(get_ccwdev_lock(cdev));
1436 	if (rc) {
1437 		DBF_ERROR("%4x act IO ERR", irq_ptr->schid.sch_no);
1438 		DBF_ERROR("rc:%4x", rc);
1439 		goto out;
1440 	}
1441 
1442 	if (is_thinint_irq(irq_ptr))
1443 		tiqdio_add_input_queues(irq_ptr);
1444 
1445 	/* wait for subchannel to become active */
1446 	msleep(5);
1447 
1448 	switch (irq_ptr->state) {
1449 	case QDIO_IRQ_STATE_STOPPED:
1450 	case QDIO_IRQ_STATE_ERR:
1451 		rc = -EIO;
1452 		break;
1453 	default:
1454 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ACTIVE);
1455 		rc = 0;
1456 	}
1457 out:
1458 	mutex_unlock(&irq_ptr->setup_mutex);
1459 	return rc;
1460 }
1461 EXPORT_SYMBOL_GPL(qdio_activate);
1462 
1463 static inline int buf_in_between(int bufnr, int start, int count)
1464 {
1465 	int end = add_buf(start, count);
1466 
1467 	if (end > start) {
1468 		if (bufnr >= start && bufnr < end)
1469 			return 1;
1470 		else
1471 			return 0;
1472 	}
1473 
1474 	/* wrap-around case */
1475 	if ((bufnr >= start && bufnr <= QDIO_MAX_BUFFERS_PER_Q) ||
1476 	    (bufnr < end))
1477 		return 1;
1478 	else
1479 		return 0;
1480 }
1481 
1482 /**
1483  * handle_inbound - reset processed input buffers
1484  * @q: queue containing the buffers
1485  * @callflags: flags
1486  * @bufnr: first buffer to process
1487  * @count: how many buffers are emptied
1488  */
1489 static int handle_inbound(struct qdio_q *q, unsigned int callflags,
1490 			  int bufnr, int count)
1491 {
1492 	int diff;
1493 
1494 	qperf_inc(q, inbound_call);
1495 
1496 	if (!q->u.in.polling)
1497 		goto set;
1498 
1499 	/* protect against stop polling setting an ACK for an emptied slsb */
1500 	if (count == QDIO_MAX_BUFFERS_PER_Q) {
1501 		/* overwriting everything, just delete polling status */
1502 		q->u.in.polling = 0;
1503 		q->u.in.ack_count = 0;
1504 		goto set;
1505 	} else if (buf_in_between(q->u.in.ack_start, bufnr, count)) {
1506 		if (is_qebsm(q)) {
1507 			/* partial overwrite, just update ack_start */
1508 			diff = add_buf(bufnr, count);
1509 			diff = sub_buf(diff, q->u.in.ack_start);
1510 			q->u.in.ack_count -= diff;
1511 			if (q->u.in.ack_count <= 0) {
1512 				q->u.in.polling = 0;
1513 				q->u.in.ack_count = 0;
1514 				goto set;
1515 			}
1516 			q->u.in.ack_start = add_buf(q->u.in.ack_start, diff);
1517 		}
1518 		else
1519 			/* the only ACK will be deleted, so stop polling */
1520 			q->u.in.polling = 0;
1521 	}
1522 
1523 set:
1524 	count = set_buf_states(q, bufnr, SLSB_CU_INPUT_EMPTY, count);
1525 	atomic_add(count, &q->nr_buf_used);
1526 
1527 	if (need_siga_in(q))
1528 		return qdio_siga_input(q);
1529 
1530 	return 0;
1531 }
1532 
1533 /**
1534  * handle_outbound - process filled outbound buffers
1535  * @q: queue containing the buffers
1536  * @callflags: flags
1537  * @bufnr: first buffer to process
1538  * @count: how many buffers are filled
1539  */
1540 static int handle_outbound(struct qdio_q *q, unsigned int callflags,
1541 			   int bufnr, int count)
1542 {
1543 	unsigned char state = 0;
1544 	int used, rc = 0;
1545 
1546 	qperf_inc(q, outbound_call);
1547 
1548 	count = set_buf_states(q, bufnr, SLSB_CU_OUTPUT_PRIMED, count);
1549 	used = atomic_add_return(count, &q->nr_buf_used);
1550 
1551 	if (used == QDIO_MAX_BUFFERS_PER_Q)
1552 		qperf_inc(q, outbound_queue_full);
1553 
1554 	if (callflags & QDIO_FLAG_PCI_OUT) {
1555 		q->u.out.pci_out_enabled = 1;
1556 		qperf_inc(q, pci_request_int);
1557 	} else
1558 		q->u.out.pci_out_enabled = 0;
1559 
1560 	if (queue_type(q) == QDIO_IQDIO_QFMT) {
1561 		unsigned long phys_aob = 0;
1562 
1563 		/* One SIGA-W per buffer required for unicast HSI */
1564 		WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
1565 
1566 		phys_aob = qdio_aob_for_buffer(&q->u.out, bufnr);
1567 
1568 		rc = qdio_kick_outbound_q(q, phys_aob);
1569 	} else if (need_siga_sync(q)) {
1570 		rc = qdio_siga_sync_q(q);
1571 	} else {
1572 		/* try to fast requeue buffers */
1573 		get_buf_state(q, prev_buf(bufnr), &state, 0);
1574 		if (state != SLSB_CU_OUTPUT_PRIMED)
1575 			rc = qdio_kick_outbound_q(q, 0);
1576 		else
1577 			qperf_inc(q, fast_requeue);
1578 	}
1579 
1580 	/* in case of SIGA errors we must process the error immediately */
1581 	if (used >= q->u.out.scan_threshold || rc)
1582 		qdio_tasklet_schedule(q);
1583 	else
1584 		/* free the SBALs in case of no further traffic */
1585 		if (!timer_pending(&q->u.out.timer) &&
1586 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
1587 			mod_timer(&q->u.out.timer, jiffies + HZ);
1588 	return rc;
1589 }
1590 
1591 /**
1592  * do_QDIO - process input or output buffers
1593  * @cdev: associated ccw_device for the qdio subchannel
1594  * @callflags: input or output and special flags from the program
1595  * @q_nr: queue number
1596  * @bufnr: buffer number
1597  * @count: how many buffers to process
1598  */
1599 int do_QDIO(struct ccw_device *cdev, unsigned int callflags,
1600 	    int q_nr, unsigned int bufnr, unsigned int count)
1601 {
1602 	struct qdio_irq *irq_ptr;
1603 
1604 	if (bufnr >= QDIO_MAX_BUFFERS_PER_Q || count > QDIO_MAX_BUFFERS_PER_Q)
1605 		return -EINVAL;
1606 
1607 	irq_ptr = cdev->private->qdio_data;
1608 	if (!irq_ptr)
1609 		return -ENODEV;
1610 
1611 	DBF_DEV_EVENT(DBF_INFO, irq_ptr,
1612 		      "do%02x b:%02x c:%02x", callflags, bufnr, count);
1613 
1614 	if (irq_ptr->state != QDIO_IRQ_STATE_ACTIVE)
1615 		return -EIO;
1616 	if (!count)
1617 		return 0;
1618 	if (callflags & QDIO_FLAG_SYNC_INPUT)
1619 		return handle_inbound(irq_ptr->input_qs[q_nr],
1620 				      callflags, bufnr, count);
1621 	else if (callflags & QDIO_FLAG_SYNC_OUTPUT)
1622 		return handle_outbound(irq_ptr->output_qs[q_nr],
1623 				       callflags, bufnr, count);
1624 	return -EINVAL;
1625 }
1626 EXPORT_SYMBOL_GPL(do_QDIO);
1627 
1628 /**
1629  * qdio_start_irq - process input buffers
1630  * @cdev: associated ccw_device for the qdio subchannel
1631  * @nr: input queue number
1632  *
1633  * Return codes
1634  *   0 - success
1635  *   1 - irqs not started since new data is available
1636  */
1637 int qdio_start_irq(struct ccw_device *cdev, int nr)
1638 {
1639 	struct qdio_q *q;
1640 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1641 
1642 	if (!irq_ptr)
1643 		return -ENODEV;
1644 	q = irq_ptr->input_qs[nr];
1645 
1646 	clear_nonshared_ind(irq_ptr);
1647 	qdio_stop_polling(q);
1648 	clear_bit(QDIO_QUEUE_IRQS_DISABLED, &q->u.in.queue_irq_state);
1649 
1650 	/*
1651 	 * We need to check again to not lose initiative after
1652 	 * resetting the ACK state.
1653 	 */
1654 	if (test_nonshared_ind(irq_ptr))
1655 		goto rescan;
1656 	if (!qdio_inbound_q_done(q))
1657 		goto rescan;
1658 	return 0;
1659 
1660 rescan:
1661 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1662 			     &q->u.in.queue_irq_state))
1663 		return 0;
1664 	else
1665 		return 1;
1666 
1667 }
1668 EXPORT_SYMBOL(qdio_start_irq);
1669 
1670 /**
1671  * qdio_get_next_buffers - process input buffers
1672  * @cdev: associated ccw_device for the qdio subchannel
1673  * @nr: input queue number
1674  * @bufnr: first filled buffer number
1675  * @error: buffers are in error state
1676  *
1677  * Return codes
1678  *   < 0 - error
1679  *   = 0 - no new buffers found
1680  *   > 0 - number of processed buffers
1681  */
1682 int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
1683 			  int *error)
1684 {
1685 	struct qdio_q *q;
1686 	int start, end;
1687 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1688 
1689 	if (!irq_ptr)
1690 		return -ENODEV;
1691 	q = irq_ptr->input_qs[nr];
1692 
1693 	/*
1694 	 * Cannot rely on automatic sync after interrupt since queues may
1695 	 * also be examined without interrupt.
1696 	 */
1697 	if (need_siga_sync(q))
1698 		qdio_sync_queues(q);
1699 
1700 	/* check the PCI capable outbound queues. */
1701 	qdio_check_outbound_after_thinint(q);
1702 
1703 	if (!qdio_inbound_q_moved(q))
1704 		return 0;
1705 
1706 	/* Note: upper-layer MUST stop processing immediately here ... */
1707 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
1708 		return -EIO;
1709 
1710 	start = q->first_to_kick;
1711 	end = q->first_to_check;
1712 	*bufnr = start;
1713 	*error = q->qdio_error;
1714 
1715 	/* for the next time */
1716 	q->first_to_kick = end;
1717 	q->qdio_error = 0;
1718 	return sub_buf(end, start);
1719 }
1720 EXPORT_SYMBOL(qdio_get_next_buffers);
1721 
1722 /**
1723  * qdio_stop_irq - disable interrupt processing for the device
1724  * @cdev: associated ccw_device for the qdio subchannel
1725  * @nr: input queue number
1726  *
1727  * Return codes
1728  *   0 - interrupts were already disabled
1729  *   1 - interrupts successfully disabled
1730  */
1731 int qdio_stop_irq(struct ccw_device *cdev, int nr)
1732 {
1733 	struct qdio_q *q;
1734 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1735 
1736 	if (!irq_ptr)
1737 		return -ENODEV;
1738 	q = irq_ptr->input_qs[nr];
1739 
1740 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1741 			     &q->u.in.queue_irq_state))
1742 		return 0;
1743 	else
1744 		return 1;
1745 }
1746 EXPORT_SYMBOL(qdio_stop_irq);
1747 
1748 /**
1749  * qdio_pnso_brinfo() - perform network subchannel op #0 - bridge info.
1750  * @schid:		Subchannel ID.
1751  * @cnc:		Boolean Change-Notification Control
1752  * @response:		Response code will be stored at this address
1753  * @cb: 		Callback function will be executed for each element
1754  *			of the address list
1755  * @priv:		Pointer to pass to the callback function.
1756  *
1757  * Performs "Store-network-bridging-information list" operation and calls
1758  * the callback function for every entry in the list. If "change-
1759  * notification-control" is set, further changes in the address list
1760  * will be reported via the IPA command.
1761  */
1762 int qdio_pnso_brinfo(struct subchannel_id schid,
1763 		int cnc, u16 *response,
1764 		void (*cb)(void *priv, enum qdio_brinfo_entry_type type,
1765 				void *entry),
1766 		void *priv)
1767 {
1768 	struct chsc_pnso_area *rr;
1769 	int rc;
1770 	u32 prev_instance = 0;
1771 	int isfirstblock = 1;
1772 	int i, size, elems;
1773 
1774 	rr = (struct chsc_pnso_area *)get_zeroed_page(GFP_KERNEL);
1775 	if (rr == NULL)
1776 		return -ENOMEM;
1777 	do {
1778 		/* on the first iteration, naihdr.resume_token will be zero */
1779 		rc = chsc_pnso_brinfo(schid, rr, rr->naihdr.resume_token, cnc);
1780 		if (rc != 0 && rc != -EBUSY)
1781 			goto out;
1782 		if (rr->response.code != 1) {
1783 			rc = -EIO;
1784 			continue;
1785 		} else
1786 			rc = 0;
1787 
1788 		if (cb == NULL)
1789 			continue;
1790 
1791 		size = rr->naihdr.naids;
1792 		elems = (rr->response.length -
1793 				sizeof(struct chsc_header) -
1794 				sizeof(struct chsc_brinfo_naihdr)) /
1795 				size;
1796 
1797 		if (!isfirstblock && (rr->naihdr.instance != prev_instance)) {
1798 			/* Inform the caller that they need to scrap */
1799 			/* the data that was already reported via cb */
1800 				rc = -EAGAIN;
1801 				break;
1802 		}
1803 		isfirstblock = 0;
1804 		prev_instance = rr->naihdr.instance;
1805 		for (i = 0; i < elems; i++)
1806 			switch (size) {
1807 			case sizeof(struct qdio_brinfo_entry_l3_ipv6):
1808 				(*cb)(priv, l3_ipv6_addr,
1809 						&rr->entries.l3_ipv6[i]);
1810 				break;
1811 			case sizeof(struct qdio_brinfo_entry_l3_ipv4):
1812 				(*cb)(priv, l3_ipv4_addr,
1813 						&rr->entries.l3_ipv4[i]);
1814 				break;
1815 			case sizeof(struct qdio_brinfo_entry_l2):
1816 				(*cb)(priv, l2_addr_lnid,
1817 						&rr->entries.l2[i]);
1818 				break;
1819 			default:
1820 				WARN_ON_ONCE(1);
1821 				rc = -EIO;
1822 				goto out;
1823 			}
1824 	} while (rr->response.code == 0x0107 ||  /* channel busy */
1825 		  (rr->response.code == 1 && /* list stored */
1826 		   /* resume token is non-zero => list incomplete */
1827 		   (rr->naihdr.resume_token.t1 || rr->naihdr.resume_token.t2)));
1828 	(*response) = rr->response.code;
1829 
1830 out:
1831 	free_page((unsigned long)rr);
1832 	return rc;
1833 }
1834 EXPORT_SYMBOL_GPL(qdio_pnso_brinfo);
1835 
1836 static int __init init_QDIO(void)
1837 {
1838 	int rc;
1839 
1840 	rc = qdio_debug_init();
1841 	if (rc)
1842 		return rc;
1843 	rc = qdio_setup_init();
1844 	if (rc)
1845 		goto out_debug;
1846 	rc = tiqdio_allocate_memory();
1847 	if (rc)
1848 		goto out_cache;
1849 	rc = tiqdio_register_thinints();
1850 	if (rc)
1851 		goto out_ti;
1852 	return 0;
1853 
1854 out_ti:
1855 	tiqdio_free_memory();
1856 out_cache:
1857 	qdio_setup_exit();
1858 out_debug:
1859 	qdio_debug_exit();
1860 	return rc;
1861 }
1862 
1863 static void __exit exit_QDIO(void)
1864 {
1865 	tiqdio_unregister_thinints();
1866 	tiqdio_free_memory();
1867 	qdio_setup_exit();
1868 	qdio_debug_exit();
1869 }
1870 
1871 module_init(init_QDIO);
1872 module_exit(exit_QDIO);
1873