xref: /openbmc/linux/drivers/s390/cio/qdio_main.c (revision 9726bfcd)
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 = 1;
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 
216 	/* Bail out early if there is no work on the queue: */
217 	if (__state & SLSB_OWNER_CU)
218 		goto out;
219 
220 	if (merge_pending && __state == SLSB_P_OUTPUT_PENDING)
221 		__state = SLSB_P_OUTPUT_EMPTY;
222 
223 	for (; i < count; i++) {
224 		bufnr = next_buf(bufnr);
225 
226 		/* merge PENDING into EMPTY: */
227 		if (merge_pending &&
228 		    q->slsb.val[bufnr] == SLSB_P_OUTPUT_PENDING &&
229 		    __state == SLSB_P_OUTPUT_EMPTY)
230 			continue;
231 
232 		/* stop if next state differs from initial state: */
233 		if (q->slsb.val[bufnr] != __state)
234 			break;
235 	}
236 
237 out:
238 	*state = __state;
239 	return i;
240 }
241 
242 static inline int get_buf_state(struct qdio_q *q, unsigned int bufnr,
243 				unsigned char *state, int auto_ack)
244 {
245 	return get_buf_states(q, bufnr, state, 1, auto_ack, 0);
246 }
247 
248 /* wrap-around safe setting of slsb states, returns number of changed buffers */
249 static inline int set_buf_states(struct qdio_q *q, int bufnr,
250 				 unsigned char state, int count)
251 {
252 	int i;
253 
254 	if (is_qebsm(q))
255 		return qdio_do_sqbs(q, state, bufnr, count);
256 
257 	for (i = 0; i < count; i++) {
258 		xchg(&q->slsb.val[bufnr], state);
259 		bufnr = next_buf(bufnr);
260 	}
261 	return count;
262 }
263 
264 static inline int set_buf_state(struct qdio_q *q, int bufnr,
265 				unsigned char state)
266 {
267 	return set_buf_states(q, bufnr, state, 1);
268 }
269 
270 /* set slsb states to initial state */
271 static void qdio_init_buf_states(struct qdio_irq *irq_ptr)
272 {
273 	struct qdio_q *q;
274 	int i;
275 
276 	for_each_input_queue(irq_ptr, q, i)
277 		set_buf_states(q, 0, SLSB_P_INPUT_NOT_INIT,
278 			       QDIO_MAX_BUFFERS_PER_Q);
279 	for_each_output_queue(irq_ptr, q, i)
280 		set_buf_states(q, 0, SLSB_P_OUTPUT_NOT_INIT,
281 			       QDIO_MAX_BUFFERS_PER_Q);
282 }
283 
284 static inline int qdio_siga_sync(struct qdio_q *q, unsigned int output,
285 			  unsigned int input)
286 {
287 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
288 	unsigned int fc = QDIO_SIGA_SYNC;
289 	int cc;
290 
291 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-s:%1d", q->nr);
292 	qperf_inc(q, siga_sync);
293 
294 	if (is_qebsm(q)) {
295 		schid = q->irq_ptr->sch_token;
296 		fc |= QDIO_SIGA_QEBSM_FLAG;
297 	}
298 
299 	cc = do_siga_sync(schid, output, input, fc);
300 	if (unlikely(cc))
301 		DBF_ERROR("%4x SIGA-S:%2d", SCH_NO(q), cc);
302 	return (cc) ? -EIO : 0;
303 }
304 
305 static inline int qdio_siga_sync_q(struct qdio_q *q)
306 {
307 	if (q->is_input_q)
308 		return qdio_siga_sync(q, 0, q->mask);
309 	else
310 		return qdio_siga_sync(q, q->mask, 0);
311 }
312 
313 static int qdio_siga_output(struct qdio_q *q, unsigned int *busy_bit,
314 	unsigned long aob)
315 {
316 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
317 	unsigned int fc = QDIO_SIGA_WRITE;
318 	u64 start_time = 0;
319 	int retries = 0, cc;
320 	unsigned long laob = 0;
321 
322 	if (aob) {
323 		fc = QDIO_SIGA_WRITEQ;
324 		laob = aob;
325 	}
326 
327 	if (is_qebsm(q)) {
328 		schid = q->irq_ptr->sch_token;
329 		fc |= QDIO_SIGA_QEBSM_FLAG;
330 	}
331 again:
332 	cc = do_siga_output(schid, q->mask, busy_bit, fc, laob);
333 
334 	/* hipersocket busy condition */
335 	if (unlikely(*busy_bit)) {
336 		retries++;
337 
338 		if (!start_time) {
339 			start_time = get_tod_clock_fast();
340 			goto again;
341 		}
342 		if (get_tod_clock_fast() - start_time < QDIO_BUSY_BIT_PATIENCE)
343 			goto again;
344 	}
345 	if (retries) {
346 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr,
347 			      "%4x cc2 BB1:%1d", SCH_NO(q), q->nr);
348 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "count:%u", retries);
349 	}
350 	return cc;
351 }
352 
353 static inline int qdio_siga_input(struct qdio_q *q)
354 {
355 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
356 	unsigned int fc = QDIO_SIGA_READ;
357 	int cc;
358 
359 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-r:%1d", q->nr);
360 	qperf_inc(q, siga_read);
361 
362 	if (is_qebsm(q)) {
363 		schid = q->irq_ptr->sch_token;
364 		fc |= QDIO_SIGA_QEBSM_FLAG;
365 	}
366 
367 	cc = do_siga_input(schid, q->mask, fc);
368 	if (unlikely(cc))
369 		DBF_ERROR("%4x SIGA-R:%2d", SCH_NO(q), cc);
370 	return (cc) ? -EIO : 0;
371 }
372 
373 #define qdio_siga_sync_out(q) qdio_siga_sync(q, ~0U, 0)
374 #define qdio_siga_sync_all(q) qdio_siga_sync(q, ~0U, ~0U)
375 
376 static inline void qdio_sync_queues(struct qdio_q *q)
377 {
378 	/* PCI capable outbound queues will also be scanned so sync them too */
379 	if (pci_out_supported(q->irq_ptr))
380 		qdio_siga_sync_all(q);
381 	else
382 		qdio_siga_sync_q(q);
383 }
384 
385 int debug_get_buf_state(struct qdio_q *q, unsigned int bufnr,
386 			unsigned char *state)
387 {
388 	if (need_siga_sync(q))
389 		qdio_siga_sync_q(q);
390 	return get_buf_state(q, bufnr, state, 0);
391 }
392 
393 static inline void qdio_stop_polling(struct qdio_q *q)
394 {
395 	if (!q->u.in.polling)
396 		return;
397 
398 	q->u.in.polling = 0;
399 	qperf_inc(q, stop_polling);
400 
401 	/* show the card that we are not polling anymore */
402 	if (is_qebsm(q)) {
403 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
404 			       q->u.in.ack_count);
405 		q->u.in.ack_count = 0;
406 	} else
407 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
408 }
409 
410 static inline void account_sbals(struct qdio_q *q, unsigned int count)
411 {
412 	int pos;
413 
414 	q->q_stats.nr_sbal_total += count;
415 	if (count == QDIO_MAX_BUFFERS_MASK) {
416 		q->q_stats.nr_sbals[7]++;
417 		return;
418 	}
419 	pos = ilog2(count);
420 	q->q_stats.nr_sbals[pos]++;
421 }
422 
423 static void process_buffer_error(struct qdio_q *q, unsigned int start,
424 				 int count)
425 {
426 	unsigned char state = (q->is_input_q) ? SLSB_P_INPUT_NOT_INIT :
427 					SLSB_P_OUTPUT_NOT_INIT;
428 
429 	q->qdio_error = QDIO_ERROR_SLSB_STATE;
430 
431 	/* special handling for no target buffer empty */
432 	if (queue_type(q) == QDIO_IQDIO_QFMT && !q->is_input_q &&
433 	    q->sbal[start]->element[15].sflags == 0x10) {
434 		qperf_inc(q, target_full);
435 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "OUTFULL FTC:%02x", start);
436 		goto set;
437 	}
438 
439 	DBF_ERROR("%4x BUF ERROR", SCH_NO(q));
440 	DBF_ERROR((q->is_input_q) ? "IN:%2d" : "OUT:%2d", q->nr);
441 	DBF_ERROR("FTC:%3d C:%3d", start, count);
442 	DBF_ERROR("F14:%2x F15:%2x",
443 		  q->sbal[start]->element[14].sflags,
444 		  q->sbal[start]->element[15].sflags);
445 
446 set:
447 	/*
448 	 * Interrupts may be avoided as long as the error is present
449 	 * so change the buffer state immediately to avoid starvation.
450 	 */
451 	set_buf_states(q, start, state, count);
452 }
453 
454 static inline void inbound_primed(struct qdio_q *q, unsigned int start,
455 				  int count)
456 {
457 	int new;
458 
459 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in prim:%1d %02x", q->nr, count);
460 
461 	/* for QEBSM the ACK was already set by EQBS */
462 	if (is_qebsm(q)) {
463 		if (!q->u.in.polling) {
464 			q->u.in.polling = 1;
465 			q->u.in.ack_count = count;
466 			q->u.in.ack_start = start;
467 			return;
468 		}
469 
470 		/* delete the previous ACK's */
471 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
472 			       q->u.in.ack_count);
473 		q->u.in.ack_count = count;
474 		q->u.in.ack_start = start;
475 		return;
476 	}
477 
478 	/*
479 	 * ACK the newest buffer. The ACK will be removed in qdio_stop_polling
480 	 * or by the next inbound run.
481 	 */
482 	new = add_buf(start, count - 1);
483 	if (q->u.in.polling) {
484 		/* reset the previous ACK but first set the new one */
485 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
486 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
487 	} else {
488 		q->u.in.polling = 1;
489 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
490 	}
491 
492 	q->u.in.ack_start = new;
493 	count--;
494 	if (!count)
495 		return;
496 	/* need to change ALL buffers to get more interrupts */
497 	set_buf_states(q, start, SLSB_P_INPUT_NOT_INIT, count);
498 }
499 
500 static int get_inbound_buffer_frontier(struct qdio_q *q, unsigned int start)
501 {
502 	unsigned char state = 0;
503 	int count;
504 
505 	q->timestamp = get_tod_clock_fast();
506 
507 	/*
508 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
509 	 * would return 0.
510 	 */
511 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
512 	if (!count)
513 		return 0;
514 
515 	/*
516 	 * No siga sync here, as a PCI or we after a thin interrupt
517 	 * already sync'ed the queues.
518 	 */
519 	count = get_buf_states(q, start, &state, count, 1, 0);
520 	if (!count)
521 		return 0;
522 
523 	switch (state) {
524 	case SLSB_P_INPUT_PRIMED:
525 		inbound_primed(q, start, count);
526 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
527 			qperf_inc(q, inbound_queue_full);
528 		if (q->irq_ptr->perf_stat_enabled)
529 			account_sbals(q, count);
530 		return count;
531 	case SLSB_P_INPUT_ERROR:
532 		process_buffer_error(q, start, count);
533 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
534 			qperf_inc(q, inbound_queue_full);
535 		if (q->irq_ptr->perf_stat_enabled)
536 			account_sbals_error(q, count);
537 		return count;
538 	case SLSB_CU_INPUT_EMPTY:
539 	case SLSB_P_INPUT_NOT_INIT:
540 	case SLSB_P_INPUT_ACK:
541 		if (q->irq_ptr->perf_stat_enabled)
542 			q->q_stats.nr_sbal_nop++;
543 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in nop:%1d %#02x",
544 			      q->nr, start);
545 		return 0;
546 	default:
547 		WARN_ON_ONCE(1);
548 		return 0;
549 	}
550 }
551 
552 static int qdio_inbound_q_moved(struct qdio_q *q, unsigned int start)
553 {
554 	int count;
555 
556 	count = get_inbound_buffer_frontier(q, start);
557 
558 	if (count && !is_thinint_irq(q->irq_ptr) && MACHINE_IS_LPAR)
559 		q->u.in.timestamp = get_tod_clock();
560 
561 	return count;
562 }
563 
564 static inline int qdio_inbound_q_done(struct qdio_q *q, unsigned int start)
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, start, &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", start);
592 		return 1;
593 	} else
594 		return 0;
595 }
596 
597 static inline void qdio_handle_aobs(struct qdio_q *q, int start, int count)
598 {
599 	unsigned char state = 0;
600 	int j, b = start;
601 
602 	for (j = 0; j < count; ++j) {
603 		get_buf_state(q, b, &state, 0);
604 		if (state == SLSB_P_OUTPUT_PENDING) {
605 			struct qaob *aob = q->u.out.aobs[b];
606 			if (aob == NULL)
607 				continue;
608 
609 			q->u.out.sbal_state[b].flags |=
610 				QDIO_OUTBUF_STATE_FLAG_PENDING;
611 			q->u.out.aobs[b] = NULL;
612 		}
613 		b = next_buf(b);
614 	}
615 }
616 
617 static inline unsigned long qdio_aob_for_buffer(struct qdio_output_q *q,
618 					int bufnr)
619 {
620 	unsigned long phys_aob = 0;
621 
622 	if (!q->aobs[bufnr]) {
623 		struct qaob *aob = qdio_allocate_aob();
624 		q->aobs[bufnr] = aob;
625 	}
626 	if (q->aobs[bufnr]) {
627 		q->aobs[bufnr]->user1 = (u64) q->sbal_state[bufnr].user;
628 		phys_aob = virt_to_phys(q->aobs[bufnr]);
629 		WARN_ON_ONCE(phys_aob & 0xFF);
630 	}
631 
632 	q->sbal_state[bufnr].flags = 0;
633 	return phys_aob;
634 }
635 
636 static void qdio_kick_handler(struct qdio_q *q, unsigned int count)
637 {
638 	int start = q->first_to_kick;
639 
640 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
641 		return;
642 
643 	if (q->is_input_q) {
644 		qperf_inc(q, inbound_handler);
645 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "kih s:%02x c:%02x", start, count);
646 	} else {
647 		qperf_inc(q, outbound_handler);
648 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "koh: s:%02x c:%02x",
649 			      start, count);
650 		if (q->u.out.use_cq)
651 			qdio_handle_aobs(q, start, count);
652 	}
653 
654 	q->handler(q->irq_ptr->cdev, q->qdio_error, q->nr, start, count,
655 		   q->irq_ptr->int_parm);
656 
657 	/* for the next time */
658 	q->first_to_kick = add_buf(start, count);
659 	q->qdio_error = 0;
660 }
661 
662 static inline int qdio_tasklet_schedule(struct qdio_q *q)
663 {
664 	if (likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE)) {
665 		tasklet_schedule(&q->tasklet);
666 		return 0;
667 	}
668 	return -EPERM;
669 }
670 
671 static void __qdio_inbound_processing(struct qdio_q *q)
672 {
673 	unsigned int start = q->first_to_check;
674 	int count;
675 
676 	qperf_inc(q, tasklet_inbound);
677 
678 	count = qdio_inbound_q_moved(q, start);
679 	if (count == 0)
680 		return;
681 
682 	start = add_buf(start, count);
683 	q->first_to_check = start;
684 	qdio_kick_handler(q, count);
685 
686 	if (!qdio_inbound_q_done(q, start)) {
687 		/* means poll time is not yet over */
688 		qperf_inc(q, tasklet_inbound_resched);
689 		if (!qdio_tasklet_schedule(q))
690 			return;
691 	}
692 
693 	qdio_stop_polling(q);
694 	/*
695 	 * We need to check again to not lose initiative after
696 	 * resetting the ACK state.
697 	 */
698 	if (!qdio_inbound_q_done(q, start)) {
699 		qperf_inc(q, tasklet_inbound_resched2);
700 		qdio_tasklet_schedule(q);
701 	}
702 }
703 
704 void qdio_inbound_processing(unsigned long data)
705 {
706 	struct qdio_q *q = (struct qdio_q *)data;
707 	__qdio_inbound_processing(q);
708 }
709 
710 static int get_outbound_buffer_frontier(struct qdio_q *q, unsigned int start)
711 {
712 	unsigned char state = 0;
713 	int count;
714 
715 	q->timestamp = get_tod_clock_fast();
716 
717 	if (need_siga_sync(q))
718 		if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
719 		    !pci_out_supported(q->irq_ptr)) ||
720 		    (queue_type(q) == QDIO_IQDIO_QFMT &&
721 		    multicast_outbound(q)))
722 			qdio_siga_sync_q(q);
723 
724 	count = atomic_read(&q->nr_buf_used);
725 	if (!count)
726 		return 0;
727 
728 	count = get_buf_states(q, start, &state, count, 0, q->u.out.use_cq);
729 	if (!count)
730 		return 0;
731 
732 	switch (state) {
733 	case SLSB_P_OUTPUT_EMPTY:
734 	case SLSB_P_OUTPUT_PENDING:
735 		/* the adapter got it */
736 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr,
737 			"out empty:%1d %02x", q->nr, count);
738 
739 		atomic_sub(count, &q->nr_buf_used);
740 		if (q->irq_ptr->perf_stat_enabled)
741 			account_sbals(q, count);
742 		return count;
743 	case SLSB_P_OUTPUT_ERROR:
744 		process_buffer_error(q, start, count);
745 		atomic_sub(count, &q->nr_buf_used);
746 		if (q->irq_ptr->perf_stat_enabled)
747 			account_sbals_error(q, count);
748 		return count;
749 	case SLSB_CU_OUTPUT_PRIMED:
750 		/* the adapter has not fetched the output yet */
751 		if (q->irq_ptr->perf_stat_enabled)
752 			q->q_stats.nr_sbal_nop++;
753 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out primed:%1d",
754 			      q->nr);
755 		return 0;
756 	case SLSB_P_OUTPUT_NOT_INIT:
757 	case SLSB_P_OUTPUT_HALTED:
758 		return 0;
759 	default:
760 		WARN_ON_ONCE(1);
761 		return 0;
762 	}
763 }
764 
765 /* all buffers processed? */
766 static inline int qdio_outbound_q_done(struct qdio_q *q)
767 {
768 	return atomic_read(&q->nr_buf_used) == 0;
769 }
770 
771 static inline int qdio_outbound_q_moved(struct qdio_q *q, unsigned int start)
772 {
773 	int count;
774 
775 	count = get_outbound_buffer_frontier(q, start);
776 
777 	if (count)
778 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out moved:%1d", q->nr);
779 
780 	return count;
781 }
782 
783 static int qdio_kick_outbound_q(struct qdio_q *q, unsigned long aob)
784 {
785 	int retries = 0, cc;
786 	unsigned int busy_bit;
787 
788 	if (!need_siga_out(q))
789 		return 0;
790 
791 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w:%1d", q->nr);
792 retry:
793 	qperf_inc(q, siga_write);
794 
795 	cc = qdio_siga_output(q, &busy_bit, aob);
796 	switch (cc) {
797 	case 0:
798 		break;
799 	case 2:
800 		if (busy_bit) {
801 			while (++retries < QDIO_BUSY_BIT_RETRIES) {
802 				mdelay(QDIO_BUSY_BIT_RETRY_DELAY);
803 				goto retry;
804 			}
805 			DBF_ERROR("%4x cc2 BBC:%1d", SCH_NO(q), q->nr);
806 			cc = -EBUSY;
807 		} else {
808 			DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w cc2:%1d", q->nr);
809 			cc = -ENOBUFS;
810 		}
811 		break;
812 	case 1:
813 	case 3:
814 		DBF_ERROR("%4x SIGA-W:%1d", SCH_NO(q), cc);
815 		cc = -EIO;
816 		break;
817 	}
818 	if (retries) {
819 		DBF_ERROR("%4x cc2 BB2:%1d", SCH_NO(q), q->nr);
820 		DBF_ERROR("count:%u", retries);
821 	}
822 	return cc;
823 }
824 
825 static void __qdio_outbound_processing(struct qdio_q *q)
826 {
827 	unsigned int start = q->first_to_check;
828 	int count;
829 
830 	qperf_inc(q, tasklet_outbound);
831 	WARN_ON_ONCE(atomic_read(&q->nr_buf_used) < 0);
832 
833 	count = qdio_outbound_q_moved(q, start);
834 	if (count) {
835 		q->first_to_check = add_buf(start, count);
836 		qdio_kick_handler(q, count);
837 	}
838 
839 	if (queue_type(q) == QDIO_ZFCP_QFMT && !pci_out_supported(q->irq_ptr) &&
840 	    !qdio_outbound_q_done(q))
841 		goto sched;
842 
843 	if (q->u.out.pci_out_enabled)
844 		return;
845 
846 	/*
847 	 * Now we know that queue type is either qeth without pci enabled
848 	 * or HiperSockets. Make sure buffer switch from PRIMED to EMPTY
849 	 * is noticed and outbound_handler is called after some time.
850 	 */
851 	if (qdio_outbound_q_done(q))
852 		del_timer_sync(&q->u.out.timer);
853 	else
854 		if (!timer_pending(&q->u.out.timer) &&
855 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
856 			mod_timer(&q->u.out.timer, jiffies + 10 * HZ);
857 	return;
858 
859 sched:
860 	qdio_tasklet_schedule(q);
861 }
862 
863 /* outbound tasklet */
864 void qdio_outbound_processing(unsigned long data)
865 {
866 	struct qdio_q *q = (struct qdio_q *)data;
867 	__qdio_outbound_processing(q);
868 }
869 
870 void qdio_outbound_timer(struct timer_list *t)
871 {
872 	struct qdio_q *q = from_timer(q, t, u.out.timer);
873 
874 	qdio_tasklet_schedule(q);
875 }
876 
877 static inline void qdio_check_outbound_pci_queues(struct qdio_irq *irq)
878 {
879 	struct qdio_q *out;
880 	int i;
881 
882 	if (!pci_out_supported(irq))
883 		return;
884 
885 	for_each_output_queue(irq, out, i)
886 		if (!qdio_outbound_q_done(out))
887 			qdio_tasklet_schedule(out);
888 }
889 
890 static void __tiqdio_inbound_processing(struct qdio_q *q)
891 {
892 	unsigned int start = q->first_to_check;
893 	int count;
894 
895 	qperf_inc(q, tasklet_inbound);
896 	if (need_siga_sync(q) && need_siga_sync_after_ai(q))
897 		qdio_sync_queues(q);
898 
899 	/* The interrupt could be caused by a PCI request: */
900 	qdio_check_outbound_pci_queues(q->irq_ptr);
901 
902 	count = qdio_inbound_q_moved(q, start);
903 	if (count == 0)
904 		return;
905 
906 	start = add_buf(start, count);
907 	q->first_to_check = start;
908 	qdio_kick_handler(q, count);
909 
910 	if (!qdio_inbound_q_done(q, start)) {
911 		qperf_inc(q, tasklet_inbound_resched);
912 		if (!qdio_tasklet_schedule(q))
913 			return;
914 	}
915 
916 	qdio_stop_polling(q);
917 	/*
918 	 * We need to check again to not lose initiative after
919 	 * resetting the ACK state.
920 	 */
921 	if (!qdio_inbound_q_done(q, start)) {
922 		qperf_inc(q, tasklet_inbound_resched2);
923 		qdio_tasklet_schedule(q);
924 	}
925 }
926 
927 void tiqdio_inbound_processing(unsigned long data)
928 {
929 	struct qdio_q *q = (struct qdio_q *)data;
930 	__tiqdio_inbound_processing(q);
931 }
932 
933 static inline void qdio_set_state(struct qdio_irq *irq_ptr,
934 				  enum qdio_irq_states state)
935 {
936 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "newstate: %1d", state);
937 
938 	irq_ptr->state = state;
939 	mb();
940 }
941 
942 static void qdio_irq_check_sense(struct qdio_irq *irq_ptr, struct irb *irb)
943 {
944 	if (irb->esw.esw0.erw.cons) {
945 		DBF_ERROR("%4x sense:", irq_ptr->schid.sch_no);
946 		DBF_ERROR_HEX(irb, 64);
947 		DBF_ERROR_HEX(irb->ecw, 64);
948 	}
949 }
950 
951 /* PCI interrupt handler */
952 static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
953 {
954 	int i;
955 	struct qdio_q *q;
956 
957 	if (unlikely(irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
958 		return;
959 
960 	for_each_input_queue(irq_ptr, q, i) {
961 		if (q->u.in.queue_start_poll) {
962 			/* skip if polling is enabled or already in work */
963 			if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
964 				     &q->u.in.queue_irq_state)) {
965 				qperf_inc(q, int_discarded);
966 				continue;
967 			}
968 			q->u.in.queue_start_poll(q->irq_ptr->cdev, q->nr,
969 						 q->irq_ptr->int_parm);
970 		} else {
971 			tasklet_schedule(&q->tasklet);
972 		}
973 	}
974 
975 	if (!pci_out_supported(irq_ptr))
976 		return;
977 
978 	for_each_output_queue(irq_ptr, q, i) {
979 		if (qdio_outbound_q_done(q))
980 			continue;
981 		if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
982 			qdio_siga_sync_q(q);
983 		qdio_tasklet_schedule(q);
984 	}
985 }
986 
987 static void qdio_handle_activate_check(struct ccw_device *cdev,
988 				unsigned long intparm, int cstat, int dstat)
989 {
990 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
991 	struct qdio_q *q;
992 	int count;
993 
994 	DBF_ERROR("%4x ACT CHECK", irq_ptr->schid.sch_no);
995 	DBF_ERROR("intp :%lx", intparm);
996 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
997 
998 	if (irq_ptr->nr_input_qs) {
999 		q = irq_ptr->input_qs[0];
1000 	} else if (irq_ptr->nr_output_qs) {
1001 		q = irq_ptr->output_qs[0];
1002 	} else {
1003 		dump_stack();
1004 		goto no_handler;
1005 	}
1006 
1007 	count = sub_buf(q->first_to_check, q->first_to_kick);
1008 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_ACTIVATE,
1009 		   q->nr, q->first_to_kick, count, irq_ptr->int_parm);
1010 no_handler:
1011 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1012 	/*
1013 	 * In case of z/VM LGR (Live Guest Migration) QDIO recovery will happen.
1014 	 * Therefore we call the LGR detection function here.
1015 	 */
1016 	lgr_info_log();
1017 }
1018 
1019 static void qdio_establish_handle_irq(struct ccw_device *cdev, int cstat,
1020 				      int dstat)
1021 {
1022 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1023 
1024 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "qest irq");
1025 
1026 	if (cstat)
1027 		goto error;
1028 	if (dstat & ~(DEV_STAT_DEV_END | DEV_STAT_CHN_END))
1029 		goto error;
1030 	if (!(dstat & DEV_STAT_DEV_END))
1031 		goto error;
1032 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ESTABLISHED);
1033 	return;
1034 
1035 error:
1036 	DBF_ERROR("%4x EQ:error", irq_ptr->schid.sch_no);
1037 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1038 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1039 }
1040 
1041 /* qdio interrupt handler */
1042 void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
1043 		      struct irb *irb)
1044 {
1045 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1046 	struct subchannel_id schid;
1047 	int cstat, dstat;
1048 
1049 	if (!intparm || !irq_ptr) {
1050 		ccw_device_get_schid(cdev, &schid);
1051 		DBF_ERROR("qint:%4x", schid.sch_no);
1052 		return;
1053 	}
1054 
1055 	if (irq_ptr->perf_stat_enabled)
1056 		irq_ptr->perf_stat.qdio_int++;
1057 
1058 	if (IS_ERR(irb)) {
1059 		DBF_ERROR("%4x IO error", irq_ptr->schid.sch_no);
1060 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1061 		wake_up(&cdev->private->wait_q);
1062 		return;
1063 	}
1064 	qdio_irq_check_sense(irq_ptr, irb);
1065 	cstat = irb->scsw.cmd.cstat;
1066 	dstat = irb->scsw.cmd.dstat;
1067 
1068 	switch (irq_ptr->state) {
1069 	case QDIO_IRQ_STATE_INACTIVE:
1070 		qdio_establish_handle_irq(cdev, cstat, dstat);
1071 		break;
1072 	case QDIO_IRQ_STATE_CLEANUP:
1073 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1074 		break;
1075 	case QDIO_IRQ_STATE_ESTABLISHED:
1076 	case QDIO_IRQ_STATE_ACTIVE:
1077 		if (cstat & SCHN_STAT_PCI) {
1078 			qdio_int_handler_pci(irq_ptr);
1079 			return;
1080 		}
1081 		if (cstat || dstat)
1082 			qdio_handle_activate_check(cdev, intparm, cstat,
1083 						   dstat);
1084 		break;
1085 	case QDIO_IRQ_STATE_STOPPED:
1086 		break;
1087 	default:
1088 		WARN_ON_ONCE(1);
1089 	}
1090 	wake_up(&cdev->private->wait_q);
1091 }
1092 
1093 /**
1094  * qdio_get_ssqd_desc - get qdio subchannel description
1095  * @cdev: ccw device to get description for
1096  * @data: where to store the ssqd
1097  *
1098  * Returns 0 or an error code. The results of the chsc are stored in the
1099  * specified structure.
1100  */
1101 int qdio_get_ssqd_desc(struct ccw_device *cdev,
1102 		       struct qdio_ssqd_desc *data)
1103 {
1104 	struct subchannel_id schid;
1105 
1106 	if (!cdev || !cdev->private)
1107 		return -EINVAL;
1108 
1109 	ccw_device_get_schid(cdev, &schid);
1110 	DBF_EVENT("get ssqd:%4x", schid.sch_no);
1111 	return qdio_setup_get_ssqd(NULL, &schid, data);
1112 }
1113 EXPORT_SYMBOL_GPL(qdio_get_ssqd_desc);
1114 
1115 static void qdio_shutdown_queues(struct ccw_device *cdev)
1116 {
1117 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1118 	struct qdio_q *q;
1119 	int i;
1120 
1121 	for_each_input_queue(irq_ptr, q, i)
1122 		tasklet_kill(&q->tasklet);
1123 
1124 	for_each_output_queue(irq_ptr, q, i) {
1125 		del_timer_sync(&q->u.out.timer);
1126 		tasklet_kill(&q->tasklet);
1127 	}
1128 }
1129 
1130 /**
1131  * qdio_shutdown - shut down a qdio subchannel
1132  * @cdev: associated ccw device
1133  * @how: use halt or clear to shutdown
1134  */
1135 int qdio_shutdown(struct ccw_device *cdev, int how)
1136 {
1137 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1138 	struct subchannel_id schid;
1139 	int rc;
1140 
1141 	if (!irq_ptr)
1142 		return -ENODEV;
1143 
1144 	WARN_ON_ONCE(irqs_disabled());
1145 	ccw_device_get_schid(cdev, &schid);
1146 	DBF_EVENT("qshutdown:%4x", schid.sch_no);
1147 
1148 	mutex_lock(&irq_ptr->setup_mutex);
1149 	/*
1150 	 * Subchannel was already shot down. We cannot prevent being called
1151 	 * twice since cio may trigger a shutdown asynchronously.
1152 	 */
1153 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1154 		mutex_unlock(&irq_ptr->setup_mutex);
1155 		return 0;
1156 	}
1157 
1158 	/*
1159 	 * Indicate that the device is going down. Scheduling the queue
1160 	 * tasklets is forbidden from here on.
1161 	 */
1162 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1163 
1164 	tiqdio_remove_input_queues(irq_ptr);
1165 	qdio_shutdown_queues(cdev);
1166 	qdio_shutdown_debug_entries(irq_ptr);
1167 
1168 	/* cleanup subchannel */
1169 	spin_lock_irq(get_ccwdev_lock(cdev));
1170 
1171 	if (how & QDIO_FLAG_CLEANUP_USING_CLEAR)
1172 		rc = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
1173 	else
1174 		/* default behaviour is halt */
1175 		rc = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
1176 	if (rc) {
1177 		DBF_ERROR("%4x SHUTD ERR", irq_ptr->schid.sch_no);
1178 		DBF_ERROR("rc:%4d", rc);
1179 		goto no_cleanup;
1180 	}
1181 
1182 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
1183 	spin_unlock_irq(get_ccwdev_lock(cdev));
1184 	wait_event_interruptible_timeout(cdev->private->wait_q,
1185 		irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
1186 		irq_ptr->state == QDIO_IRQ_STATE_ERR,
1187 		10 * HZ);
1188 	spin_lock_irq(get_ccwdev_lock(cdev));
1189 
1190 no_cleanup:
1191 	qdio_shutdown_thinint(irq_ptr);
1192 
1193 	/* restore interrupt handler */
1194 	if ((void *)cdev->handler == (void *)qdio_int_handler) {
1195 		cdev->handler = irq_ptr->orig_handler;
1196 		cdev->private->intparm = 0;
1197 	}
1198 	spin_unlock_irq(get_ccwdev_lock(cdev));
1199 
1200 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1201 	mutex_unlock(&irq_ptr->setup_mutex);
1202 	if (rc)
1203 		return rc;
1204 	return 0;
1205 }
1206 EXPORT_SYMBOL_GPL(qdio_shutdown);
1207 
1208 /**
1209  * qdio_free - free data structures for a qdio subchannel
1210  * @cdev: associated ccw device
1211  */
1212 int qdio_free(struct ccw_device *cdev)
1213 {
1214 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1215 	struct subchannel_id schid;
1216 
1217 	if (!irq_ptr)
1218 		return -ENODEV;
1219 
1220 	ccw_device_get_schid(cdev, &schid);
1221 	DBF_EVENT("qfree:%4x", schid.sch_no);
1222 	DBF_DEV_EVENT(DBF_ERR, irq_ptr, "dbf abandoned");
1223 	mutex_lock(&irq_ptr->setup_mutex);
1224 
1225 	irq_ptr->debug_area = NULL;
1226 	cdev->private->qdio_data = NULL;
1227 	mutex_unlock(&irq_ptr->setup_mutex);
1228 
1229 	qdio_release_memory(irq_ptr);
1230 	return 0;
1231 }
1232 EXPORT_SYMBOL_GPL(qdio_free);
1233 
1234 /**
1235  * qdio_allocate - allocate qdio queues and associated data
1236  * @init_data: initialization data
1237  */
1238 int qdio_allocate(struct qdio_initialize *init_data)
1239 {
1240 	struct subchannel_id schid;
1241 	struct qdio_irq *irq_ptr;
1242 
1243 	ccw_device_get_schid(init_data->cdev, &schid);
1244 	DBF_EVENT("qallocate:%4x", schid.sch_no);
1245 
1246 	if ((init_data->no_input_qs && !init_data->input_handler) ||
1247 	    (init_data->no_output_qs && !init_data->output_handler))
1248 		return -EINVAL;
1249 
1250 	if ((init_data->no_input_qs > QDIO_MAX_QUEUES_PER_IRQ) ||
1251 	    (init_data->no_output_qs > QDIO_MAX_QUEUES_PER_IRQ))
1252 		return -EINVAL;
1253 
1254 	if ((!init_data->input_sbal_addr_array) ||
1255 	    (!init_data->output_sbal_addr_array))
1256 		return -EINVAL;
1257 
1258 	/* irq_ptr must be in GFP_DMA since it contains ccw1.cda */
1259 	irq_ptr = (void *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1260 	if (!irq_ptr)
1261 		goto out_err;
1262 
1263 	mutex_init(&irq_ptr->setup_mutex);
1264 	if (qdio_allocate_dbf(init_data, irq_ptr))
1265 		goto out_rel;
1266 
1267 	/*
1268 	 * Allocate a page for the chsc calls in qdio_establish.
1269 	 * Must be pre-allocated since a zfcp recovery will call
1270 	 * qdio_establish. In case of low memory and swap on a zfcp disk
1271 	 * we may not be able to allocate memory otherwise.
1272 	 */
1273 	irq_ptr->chsc_page = get_zeroed_page(GFP_KERNEL);
1274 	if (!irq_ptr->chsc_page)
1275 		goto out_rel;
1276 
1277 	/* qdr is used in ccw1.cda which is u32 */
1278 	irq_ptr->qdr = (struct qdr *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1279 	if (!irq_ptr->qdr)
1280 		goto out_rel;
1281 
1282 	if (qdio_allocate_qs(irq_ptr, init_data->no_input_qs,
1283 			     init_data->no_output_qs))
1284 		goto out_rel;
1285 
1286 	init_data->cdev->private->qdio_data = irq_ptr;
1287 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1288 	return 0;
1289 out_rel:
1290 	qdio_release_memory(irq_ptr);
1291 out_err:
1292 	return -ENOMEM;
1293 }
1294 EXPORT_SYMBOL_GPL(qdio_allocate);
1295 
1296 static void qdio_detect_hsicq(struct qdio_irq *irq_ptr)
1297 {
1298 	struct qdio_q *q = irq_ptr->input_qs[0];
1299 	int i, use_cq = 0;
1300 
1301 	if (irq_ptr->nr_input_qs > 1 && queue_type(q) == QDIO_IQDIO_QFMT)
1302 		use_cq = 1;
1303 
1304 	for_each_output_queue(irq_ptr, q, i) {
1305 		if (use_cq) {
1306 			if (multicast_outbound(q))
1307 				continue;
1308 			if (qdio_enable_async_operation(&q->u.out) < 0) {
1309 				use_cq = 0;
1310 				continue;
1311 			}
1312 		} else
1313 			qdio_disable_async_operation(&q->u.out);
1314 	}
1315 	DBF_EVENT("use_cq:%d", use_cq);
1316 }
1317 
1318 /**
1319  * qdio_establish - establish queues on a qdio subchannel
1320  * @init_data: initialization data
1321  */
1322 int qdio_establish(struct qdio_initialize *init_data)
1323 {
1324 	struct ccw_device *cdev = init_data->cdev;
1325 	struct subchannel_id schid;
1326 	struct qdio_irq *irq_ptr;
1327 	int rc;
1328 
1329 	ccw_device_get_schid(cdev, &schid);
1330 	DBF_EVENT("qestablish:%4x", schid.sch_no);
1331 
1332 	irq_ptr = cdev->private->qdio_data;
1333 	if (!irq_ptr)
1334 		return -ENODEV;
1335 
1336 	mutex_lock(&irq_ptr->setup_mutex);
1337 	qdio_setup_irq(init_data);
1338 
1339 	rc = qdio_establish_thinint(irq_ptr);
1340 	if (rc) {
1341 		mutex_unlock(&irq_ptr->setup_mutex);
1342 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1343 		return rc;
1344 	}
1345 
1346 	/* establish q */
1347 	irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
1348 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1349 	irq_ptr->ccw.count = irq_ptr->equeue.count;
1350 	irq_ptr->ccw.cda = (u32)((addr_t)irq_ptr->qdr);
1351 
1352 	spin_lock_irq(get_ccwdev_lock(cdev));
1353 	ccw_device_set_options_mask(cdev, 0);
1354 
1355 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ESTABLISH, 0, 0);
1356 	spin_unlock_irq(get_ccwdev_lock(cdev));
1357 	if (rc) {
1358 		DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
1359 		DBF_ERROR("rc:%4x", rc);
1360 		mutex_unlock(&irq_ptr->setup_mutex);
1361 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1362 		return rc;
1363 	}
1364 
1365 	wait_event_interruptible_timeout(cdev->private->wait_q,
1366 		irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
1367 		irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
1368 
1369 	if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
1370 		mutex_unlock(&irq_ptr->setup_mutex);
1371 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1372 		return -EIO;
1373 	}
1374 
1375 	qdio_setup_ssqd_info(irq_ptr);
1376 
1377 	qdio_detect_hsicq(irq_ptr);
1378 
1379 	/* qebsm is now setup if available, initialize buffer states */
1380 	qdio_init_buf_states(irq_ptr);
1381 
1382 	mutex_unlock(&irq_ptr->setup_mutex);
1383 	qdio_print_subchannel_info(irq_ptr, cdev);
1384 	qdio_setup_debug_entries(irq_ptr, cdev);
1385 	return 0;
1386 }
1387 EXPORT_SYMBOL_GPL(qdio_establish);
1388 
1389 /**
1390  * qdio_activate - activate queues on a qdio subchannel
1391  * @cdev: associated cdev
1392  */
1393 int qdio_activate(struct ccw_device *cdev)
1394 {
1395 	struct subchannel_id schid;
1396 	struct qdio_irq *irq_ptr;
1397 	int rc;
1398 
1399 	ccw_device_get_schid(cdev, &schid);
1400 	DBF_EVENT("qactivate:%4x", schid.sch_no);
1401 
1402 	irq_ptr = cdev->private->qdio_data;
1403 	if (!irq_ptr)
1404 		return -ENODEV;
1405 
1406 	mutex_lock(&irq_ptr->setup_mutex);
1407 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1408 		rc = -EBUSY;
1409 		goto out;
1410 	}
1411 
1412 	irq_ptr->ccw.cmd_code = irq_ptr->aqueue.cmd;
1413 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1414 	irq_ptr->ccw.count = irq_ptr->aqueue.count;
1415 	irq_ptr->ccw.cda = 0;
1416 
1417 	spin_lock_irq(get_ccwdev_lock(cdev));
1418 	ccw_device_set_options(cdev, CCWDEV_REPORT_ALL);
1419 
1420 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ACTIVATE,
1421 			      0, DOIO_DENY_PREFETCH);
1422 	spin_unlock_irq(get_ccwdev_lock(cdev));
1423 	if (rc) {
1424 		DBF_ERROR("%4x act IO ERR", irq_ptr->schid.sch_no);
1425 		DBF_ERROR("rc:%4x", rc);
1426 		goto out;
1427 	}
1428 
1429 	if (is_thinint_irq(irq_ptr))
1430 		tiqdio_add_input_queues(irq_ptr);
1431 
1432 	/* wait for subchannel to become active */
1433 	msleep(5);
1434 
1435 	switch (irq_ptr->state) {
1436 	case QDIO_IRQ_STATE_STOPPED:
1437 	case QDIO_IRQ_STATE_ERR:
1438 		rc = -EIO;
1439 		break;
1440 	default:
1441 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ACTIVE);
1442 		rc = 0;
1443 	}
1444 out:
1445 	mutex_unlock(&irq_ptr->setup_mutex);
1446 	return rc;
1447 }
1448 EXPORT_SYMBOL_GPL(qdio_activate);
1449 
1450 static inline int buf_in_between(int bufnr, int start, int count)
1451 {
1452 	int end = add_buf(start, count);
1453 
1454 	if (end > start) {
1455 		if (bufnr >= start && bufnr < end)
1456 			return 1;
1457 		else
1458 			return 0;
1459 	}
1460 
1461 	/* wrap-around case */
1462 	if ((bufnr >= start && bufnr <= QDIO_MAX_BUFFERS_PER_Q) ||
1463 	    (bufnr < end))
1464 		return 1;
1465 	else
1466 		return 0;
1467 }
1468 
1469 /**
1470  * handle_inbound - reset processed input buffers
1471  * @q: queue containing the buffers
1472  * @callflags: flags
1473  * @bufnr: first buffer to process
1474  * @count: how many buffers are emptied
1475  */
1476 static int handle_inbound(struct qdio_q *q, unsigned int callflags,
1477 			  int bufnr, int count)
1478 {
1479 	int diff;
1480 
1481 	qperf_inc(q, inbound_call);
1482 
1483 	if (!q->u.in.polling)
1484 		goto set;
1485 
1486 	/* protect against stop polling setting an ACK for an emptied slsb */
1487 	if (count == QDIO_MAX_BUFFERS_PER_Q) {
1488 		/* overwriting everything, just delete polling status */
1489 		q->u.in.polling = 0;
1490 		q->u.in.ack_count = 0;
1491 		goto set;
1492 	} else if (buf_in_between(q->u.in.ack_start, bufnr, count)) {
1493 		if (is_qebsm(q)) {
1494 			/* partial overwrite, just update ack_start */
1495 			diff = add_buf(bufnr, count);
1496 			diff = sub_buf(diff, q->u.in.ack_start);
1497 			q->u.in.ack_count -= diff;
1498 			if (q->u.in.ack_count <= 0) {
1499 				q->u.in.polling = 0;
1500 				q->u.in.ack_count = 0;
1501 				goto set;
1502 			}
1503 			q->u.in.ack_start = add_buf(q->u.in.ack_start, diff);
1504 		}
1505 		else
1506 			/* the only ACK will be deleted, so stop polling */
1507 			q->u.in.polling = 0;
1508 	}
1509 
1510 set:
1511 	count = set_buf_states(q, bufnr, SLSB_CU_INPUT_EMPTY, count);
1512 	atomic_add(count, &q->nr_buf_used);
1513 
1514 	if (need_siga_in(q))
1515 		return qdio_siga_input(q);
1516 
1517 	return 0;
1518 }
1519 
1520 /**
1521  * handle_outbound - process filled outbound buffers
1522  * @q: queue containing the buffers
1523  * @callflags: flags
1524  * @bufnr: first buffer to process
1525  * @count: how many buffers are filled
1526  */
1527 static int handle_outbound(struct qdio_q *q, unsigned int callflags,
1528 			   int bufnr, int count)
1529 {
1530 	unsigned char state = 0;
1531 	int used, rc = 0;
1532 
1533 	qperf_inc(q, outbound_call);
1534 
1535 	count = set_buf_states(q, bufnr, SLSB_CU_OUTPUT_PRIMED, count);
1536 	used = atomic_add_return(count, &q->nr_buf_used);
1537 
1538 	if (used == QDIO_MAX_BUFFERS_PER_Q)
1539 		qperf_inc(q, outbound_queue_full);
1540 
1541 	if (callflags & QDIO_FLAG_PCI_OUT) {
1542 		q->u.out.pci_out_enabled = 1;
1543 		qperf_inc(q, pci_request_int);
1544 	} else
1545 		q->u.out.pci_out_enabled = 0;
1546 
1547 	if (queue_type(q) == QDIO_IQDIO_QFMT) {
1548 		unsigned long phys_aob = 0;
1549 
1550 		/* One SIGA-W per buffer required for unicast HSI */
1551 		WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
1552 
1553 		if (q->u.out.use_cq)
1554 			phys_aob = qdio_aob_for_buffer(&q->u.out, bufnr);
1555 
1556 		rc = qdio_kick_outbound_q(q, phys_aob);
1557 	} else if (need_siga_sync(q)) {
1558 		rc = qdio_siga_sync_q(q);
1559 	} else if (count < QDIO_MAX_BUFFERS_PER_Q &&
1560 		   get_buf_state(q, prev_buf(bufnr), &state, 0) > 0 &&
1561 		   state == SLSB_CU_OUTPUT_PRIMED) {
1562 		/* The previous buffer is not processed yet, tack on. */
1563 		qperf_inc(q, fast_requeue);
1564 	} else {
1565 		rc = qdio_kick_outbound_q(q, 0);
1566 	}
1567 
1568 	/* in case of SIGA errors we must process the error immediately */
1569 	if (used >= q->u.out.scan_threshold || rc)
1570 		qdio_tasklet_schedule(q);
1571 	else
1572 		/* free the SBALs in case of no further traffic */
1573 		if (!timer_pending(&q->u.out.timer) &&
1574 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
1575 			mod_timer(&q->u.out.timer, jiffies + HZ);
1576 	return rc;
1577 }
1578 
1579 /**
1580  * do_QDIO - process input or output buffers
1581  * @cdev: associated ccw_device for the qdio subchannel
1582  * @callflags: input or output and special flags from the program
1583  * @q_nr: queue number
1584  * @bufnr: buffer number
1585  * @count: how many buffers to process
1586  */
1587 int do_QDIO(struct ccw_device *cdev, unsigned int callflags,
1588 	    int q_nr, unsigned int bufnr, unsigned int count)
1589 {
1590 	struct qdio_irq *irq_ptr;
1591 
1592 	if (bufnr >= QDIO_MAX_BUFFERS_PER_Q || count > QDIO_MAX_BUFFERS_PER_Q)
1593 		return -EINVAL;
1594 
1595 	irq_ptr = cdev->private->qdio_data;
1596 	if (!irq_ptr)
1597 		return -ENODEV;
1598 
1599 	DBF_DEV_EVENT(DBF_INFO, irq_ptr,
1600 		      "do%02x b:%02x c:%02x", callflags, bufnr, count);
1601 
1602 	if (irq_ptr->state != QDIO_IRQ_STATE_ACTIVE)
1603 		return -EIO;
1604 	if (!count)
1605 		return 0;
1606 	if (callflags & QDIO_FLAG_SYNC_INPUT)
1607 		return handle_inbound(irq_ptr->input_qs[q_nr],
1608 				      callflags, bufnr, count);
1609 	else if (callflags & QDIO_FLAG_SYNC_OUTPUT)
1610 		return handle_outbound(irq_ptr->output_qs[q_nr],
1611 				       callflags, bufnr, count);
1612 	return -EINVAL;
1613 }
1614 EXPORT_SYMBOL_GPL(do_QDIO);
1615 
1616 /**
1617  * qdio_start_irq - process input buffers
1618  * @cdev: associated ccw_device for the qdio subchannel
1619  * @nr: input queue number
1620  *
1621  * Return codes
1622  *   0 - success
1623  *   1 - irqs not started since new data is available
1624  */
1625 int qdio_start_irq(struct ccw_device *cdev, int nr)
1626 {
1627 	struct qdio_q *q;
1628 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1629 
1630 	if (!irq_ptr)
1631 		return -ENODEV;
1632 	q = irq_ptr->input_qs[nr];
1633 
1634 	clear_nonshared_ind(irq_ptr);
1635 	qdio_stop_polling(q);
1636 	clear_bit(QDIO_QUEUE_IRQS_DISABLED, &q->u.in.queue_irq_state);
1637 
1638 	/*
1639 	 * We need to check again to not lose initiative after
1640 	 * resetting the ACK state.
1641 	 */
1642 	if (test_nonshared_ind(irq_ptr))
1643 		goto rescan;
1644 	if (!qdio_inbound_q_done(q, q->first_to_check))
1645 		goto rescan;
1646 	return 0;
1647 
1648 rescan:
1649 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1650 			     &q->u.in.queue_irq_state))
1651 		return 0;
1652 	else
1653 		return 1;
1654 
1655 }
1656 EXPORT_SYMBOL(qdio_start_irq);
1657 
1658 /**
1659  * qdio_get_next_buffers - process input buffers
1660  * @cdev: associated ccw_device for the qdio subchannel
1661  * @nr: input queue number
1662  * @bufnr: first filled buffer number
1663  * @error: buffers are in error state
1664  *
1665  * Return codes
1666  *   < 0 - error
1667  *   = 0 - no new buffers found
1668  *   > 0 - number of processed buffers
1669  */
1670 int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
1671 			  int *error)
1672 {
1673 	struct qdio_q *q;
1674 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1675 	unsigned int start;
1676 	int count;
1677 
1678 	if (!irq_ptr)
1679 		return -ENODEV;
1680 	q = irq_ptr->input_qs[nr];
1681 	start = q->first_to_check;
1682 
1683 	/*
1684 	 * Cannot rely on automatic sync after interrupt since queues may
1685 	 * also be examined without interrupt.
1686 	 */
1687 	if (need_siga_sync(q))
1688 		qdio_sync_queues(q);
1689 
1690 	qdio_check_outbound_pci_queues(irq_ptr);
1691 
1692 	count = qdio_inbound_q_moved(q, start);
1693 	if (count == 0)
1694 		return 0;
1695 
1696 	start = add_buf(start, count);
1697 	q->first_to_check = start;
1698 
1699 	/* Note: upper-layer MUST stop processing immediately here ... */
1700 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
1701 		return -EIO;
1702 
1703 	*bufnr = q->first_to_kick;
1704 	*error = q->qdio_error;
1705 
1706 	/* for the next time */
1707 	q->first_to_kick = add_buf(q->first_to_kick, count);
1708 	q->qdio_error = 0;
1709 
1710 	return count;
1711 }
1712 EXPORT_SYMBOL(qdio_get_next_buffers);
1713 
1714 /**
1715  * qdio_stop_irq - disable interrupt processing for the device
1716  * @cdev: associated ccw_device for the qdio subchannel
1717  * @nr: input queue number
1718  *
1719  * Return codes
1720  *   0 - interrupts were already disabled
1721  *   1 - interrupts successfully disabled
1722  */
1723 int qdio_stop_irq(struct ccw_device *cdev, int nr)
1724 {
1725 	struct qdio_q *q;
1726 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1727 
1728 	if (!irq_ptr)
1729 		return -ENODEV;
1730 	q = irq_ptr->input_qs[nr];
1731 
1732 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1733 			     &q->u.in.queue_irq_state))
1734 		return 0;
1735 	else
1736 		return 1;
1737 }
1738 EXPORT_SYMBOL(qdio_stop_irq);
1739 
1740 /**
1741  * qdio_pnso_brinfo() - perform network subchannel op #0 - bridge info.
1742  * @schid:		Subchannel ID.
1743  * @cnc:		Boolean Change-Notification Control
1744  * @response:		Response code will be stored at this address
1745  * @cb: 		Callback function will be executed for each element
1746  *			of the address list
1747  * @priv:		Pointer to pass to the callback function.
1748  *
1749  * Performs "Store-network-bridging-information list" operation and calls
1750  * the callback function for every entry in the list. If "change-
1751  * notification-control" is set, further changes in the address list
1752  * will be reported via the IPA command.
1753  */
1754 int qdio_pnso_brinfo(struct subchannel_id schid,
1755 		int cnc, u16 *response,
1756 		void (*cb)(void *priv, enum qdio_brinfo_entry_type type,
1757 				void *entry),
1758 		void *priv)
1759 {
1760 	struct chsc_pnso_area *rr;
1761 	int rc;
1762 	u32 prev_instance = 0;
1763 	int isfirstblock = 1;
1764 	int i, size, elems;
1765 
1766 	rr = (struct chsc_pnso_area *)get_zeroed_page(GFP_KERNEL);
1767 	if (rr == NULL)
1768 		return -ENOMEM;
1769 	do {
1770 		/* on the first iteration, naihdr.resume_token will be zero */
1771 		rc = chsc_pnso_brinfo(schid, rr, rr->naihdr.resume_token, cnc);
1772 		if (rc != 0 && rc != -EBUSY)
1773 			goto out;
1774 		if (rr->response.code != 1) {
1775 			rc = -EIO;
1776 			continue;
1777 		} else
1778 			rc = 0;
1779 
1780 		if (cb == NULL)
1781 			continue;
1782 
1783 		size = rr->naihdr.naids;
1784 		elems = (rr->response.length -
1785 				sizeof(struct chsc_header) -
1786 				sizeof(struct chsc_brinfo_naihdr)) /
1787 				size;
1788 
1789 		if (!isfirstblock && (rr->naihdr.instance != prev_instance)) {
1790 			/* Inform the caller that they need to scrap */
1791 			/* the data that was already reported via cb */
1792 				rc = -EAGAIN;
1793 				break;
1794 		}
1795 		isfirstblock = 0;
1796 		prev_instance = rr->naihdr.instance;
1797 		for (i = 0; i < elems; i++)
1798 			switch (size) {
1799 			case sizeof(struct qdio_brinfo_entry_l3_ipv6):
1800 				(*cb)(priv, l3_ipv6_addr,
1801 						&rr->entries.l3_ipv6[i]);
1802 				break;
1803 			case sizeof(struct qdio_brinfo_entry_l3_ipv4):
1804 				(*cb)(priv, l3_ipv4_addr,
1805 						&rr->entries.l3_ipv4[i]);
1806 				break;
1807 			case sizeof(struct qdio_brinfo_entry_l2):
1808 				(*cb)(priv, l2_addr_lnid,
1809 						&rr->entries.l2[i]);
1810 				break;
1811 			default:
1812 				WARN_ON_ONCE(1);
1813 				rc = -EIO;
1814 				goto out;
1815 			}
1816 	} while (rr->response.code == 0x0107 ||  /* channel busy */
1817 		  (rr->response.code == 1 && /* list stored */
1818 		   /* resume token is non-zero => list incomplete */
1819 		   (rr->naihdr.resume_token.t1 || rr->naihdr.resume_token.t2)));
1820 	(*response) = rr->response.code;
1821 
1822 out:
1823 	free_page((unsigned long)rr);
1824 	return rc;
1825 }
1826 EXPORT_SYMBOL_GPL(qdio_pnso_brinfo);
1827 
1828 static int __init init_QDIO(void)
1829 {
1830 	int rc;
1831 
1832 	rc = qdio_debug_init();
1833 	if (rc)
1834 		return rc;
1835 	rc = qdio_setup_init();
1836 	if (rc)
1837 		goto out_debug;
1838 	rc = tiqdio_allocate_memory();
1839 	if (rc)
1840 		goto out_cache;
1841 	rc = tiqdio_register_thinints();
1842 	if (rc)
1843 		goto out_ti;
1844 	return 0;
1845 
1846 out_ti:
1847 	tiqdio_free_memory();
1848 out_cache:
1849 	qdio_setup_exit();
1850 out_debug:
1851 	qdio_debug_exit();
1852 	return rc;
1853 }
1854 
1855 static void __exit exit_QDIO(void)
1856 {
1857 	tiqdio_unregister_thinints();
1858 	tiqdio_free_memory();
1859 	qdio_setup_exit();
1860 	qdio_debug_exit();
1861 }
1862 
1863 module_init(init_QDIO);
1864 module_exit(exit_QDIO);
1865