xref: /openbmc/linux/drivers/s390/cio/qdio_main.c (revision 9e3bd0f6)
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 	}
651 
652 	q->handler(q->irq_ptr->cdev, q->qdio_error, q->nr, start, count,
653 		   q->irq_ptr->int_parm);
654 
655 	/* for the next time */
656 	q->first_to_kick = add_buf(start, count);
657 	q->qdio_error = 0;
658 }
659 
660 static inline int qdio_tasklet_schedule(struct qdio_q *q)
661 {
662 	if (likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE)) {
663 		tasklet_schedule(&q->tasklet);
664 		return 0;
665 	}
666 	return -EPERM;
667 }
668 
669 static void __qdio_inbound_processing(struct qdio_q *q)
670 {
671 	unsigned int start = q->first_to_check;
672 	int count;
673 
674 	qperf_inc(q, tasklet_inbound);
675 
676 	count = qdio_inbound_q_moved(q, start);
677 	if (count == 0)
678 		return;
679 
680 	start = add_buf(start, count);
681 	q->first_to_check = start;
682 	qdio_kick_handler(q, count);
683 
684 	if (!qdio_inbound_q_done(q, start)) {
685 		/* means poll time is not yet over */
686 		qperf_inc(q, tasklet_inbound_resched);
687 		if (!qdio_tasklet_schedule(q))
688 			return;
689 	}
690 
691 	qdio_stop_polling(q);
692 	/*
693 	 * We need to check again to not lose initiative after
694 	 * resetting the ACK state.
695 	 */
696 	if (!qdio_inbound_q_done(q, start)) {
697 		qperf_inc(q, tasklet_inbound_resched2);
698 		qdio_tasklet_schedule(q);
699 	}
700 }
701 
702 void qdio_inbound_processing(unsigned long data)
703 {
704 	struct qdio_q *q = (struct qdio_q *)data;
705 	__qdio_inbound_processing(q);
706 }
707 
708 static int get_outbound_buffer_frontier(struct qdio_q *q, unsigned int start)
709 {
710 	unsigned char state = 0;
711 	int count;
712 
713 	q->timestamp = get_tod_clock_fast();
714 
715 	if (need_siga_sync(q))
716 		if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
717 		    !pci_out_supported(q->irq_ptr)) ||
718 		    (queue_type(q) == QDIO_IQDIO_QFMT &&
719 		    multicast_outbound(q)))
720 			qdio_siga_sync_q(q);
721 
722 	count = atomic_read(&q->nr_buf_used);
723 	if (!count)
724 		return 0;
725 
726 	count = get_buf_states(q, start, &state, count, 0, q->u.out.use_cq);
727 	if (!count)
728 		return 0;
729 
730 	switch (state) {
731 	case SLSB_P_OUTPUT_EMPTY:
732 	case SLSB_P_OUTPUT_PENDING:
733 		/* the adapter got it */
734 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr,
735 			"out empty:%1d %02x", q->nr, count);
736 
737 		atomic_sub(count, &q->nr_buf_used);
738 		if (q->irq_ptr->perf_stat_enabled)
739 			account_sbals(q, count);
740 		return count;
741 	case SLSB_P_OUTPUT_ERROR:
742 		process_buffer_error(q, start, count);
743 		atomic_sub(count, &q->nr_buf_used);
744 		if (q->irq_ptr->perf_stat_enabled)
745 			account_sbals_error(q, count);
746 		return count;
747 	case SLSB_CU_OUTPUT_PRIMED:
748 		/* the adapter has not fetched the output yet */
749 		if (q->irq_ptr->perf_stat_enabled)
750 			q->q_stats.nr_sbal_nop++;
751 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out primed:%1d",
752 			      q->nr);
753 		return 0;
754 	case SLSB_P_OUTPUT_NOT_INIT:
755 	case SLSB_P_OUTPUT_HALTED:
756 		return 0;
757 	default:
758 		WARN_ON_ONCE(1);
759 		return 0;
760 	}
761 }
762 
763 /* all buffers processed? */
764 static inline int qdio_outbound_q_done(struct qdio_q *q)
765 {
766 	return atomic_read(&q->nr_buf_used) == 0;
767 }
768 
769 static inline int qdio_outbound_q_moved(struct qdio_q *q, unsigned int start)
770 {
771 	int count;
772 
773 	count = get_outbound_buffer_frontier(q, start);
774 
775 	if (count) {
776 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out moved:%1d", q->nr);
777 		if (q->u.out.use_cq)
778 			qdio_handle_aobs(q, start, count);
779 	}
780 
781 	return count;
782 }
783 
784 static int qdio_kick_outbound_q(struct qdio_q *q, unsigned long aob)
785 {
786 	int retries = 0, cc;
787 	unsigned int busy_bit;
788 
789 	if (!need_siga_out(q))
790 		return 0;
791 
792 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w:%1d", q->nr);
793 retry:
794 	qperf_inc(q, siga_write);
795 
796 	cc = qdio_siga_output(q, &busy_bit, aob);
797 	switch (cc) {
798 	case 0:
799 		break;
800 	case 2:
801 		if (busy_bit) {
802 			while (++retries < QDIO_BUSY_BIT_RETRIES) {
803 				mdelay(QDIO_BUSY_BIT_RETRY_DELAY);
804 				goto retry;
805 			}
806 			DBF_ERROR("%4x cc2 BBC:%1d", SCH_NO(q), q->nr);
807 			cc = -EBUSY;
808 		} else {
809 			DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w cc2:%1d", q->nr);
810 			cc = -ENOBUFS;
811 		}
812 		break;
813 	case 1:
814 	case 3:
815 		DBF_ERROR("%4x SIGA-W:%1d", SCH_NO(q), cc);
816 		cc = -EIO;
817 		break;
818 	}
819 	if (retries) {
820 		DBF_ERROR("%4x cc2 BB2:%1d", SCH_NO(q), q->nr);
821 		DBF_ERROR("count:%u", retries);
822 	}
823 	return cc;
824 }
825 
826 static void __qdio_outbound_processing(struct qdio_q *q)
827 {
828 	unsigned int start = q->first_to_check;
829 	int count;
830 
831 	qperf_inc(q, tasklet_outbound);
832 	WARN_ON_ONCE(atomic_read(&q->nr_buf_used) < 0);
833 
834 	count = qdio_outbound_q_moved(q, start);
835 	if (count) {
836 		q->first_to_check = add_buf(start, count);
837 		qdio_kick_handler(q, count);
838 	}
839 
840 	if (queue_type(q) == QDIO_ZFCP_QFMT && !pci_out_supported(q->irq_ptr) &&
841 	    !qdio_outbound_q_done(q))
842 		goto sched;
843 
844 	if (q->u.out.pci_out_enabled)
845 		return;
846 
847 	/*
848 	 * Now we know that queue type is either qeth without pci enabled
849 	 * or HiperSockets. Make sure buffer switch from PRIMED to EMPTY
850 	 * is noticed and outbound_handler is called after some time.
851 	 */
852 	if (qdio_outbound_q_done(q))
853 		del_timer_sync(&q->u.out.timer);
854 	else
855 		if (!timer_pending(&q->u.out.timer) &&
856 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
857 			mod_timer(&q->u.out.timer, jiffies + 10 * HZ);
858 	return;
859 
860 sched:
861 	qdio_tasklet_schedule(q);
862 }
863 
864 /* outbound tasklet */
865 void qdio_outbound_processing(unsigned long data)
866 {
867 	struct qdio_q *q = (struct qdio_q *)data;
868 	__qdio_outbound_processing(q);
869 }
870 
871 void qdio_outbound_timer(struct timer_list *t)
872 {
873 	struct qdio_q *q = from_timer(q, t, u.out.timer);
874 
875 	qdio_tasklet_schedule(q);
876 }
877 
878 static inline void qdio_check_outbound_pci_queues(struct qdio_irq *irq)
879 {
880 	struct qdio_q *out;
881 	int i;
882 
883 	if (!pci_out_supported(irq) || !irq->scan_threshold)
884 		return;
885 
886 	for_each_output_queue(irq, out, i)
887 		if (!qdio_outbound_q_done(out))
888 			qdio_tasklet_schedule(out);
889 }
890 
891 static void __tiqdio_inbound_processing(struct qdio_q *q)
892 {
893 	unsigned int start = q->first_to_check;
894 	int count;
895 
896 	qperf_inc(q, tasklet_inbound);
897 	if (need_siga_sync(q) && need_siga_sync_after_ai(q))
898 		qdio_sync_queues(q);
899 
900 	/* The interrupt could be caused by a PCI request: */
901 	qdio_check_outbound_pci_queues(q->irq_ptr);
902 
903 	count = qdio_inbound_q_moved(q, start);
904 	if (count == 0)
905 		return;
906 
907 	start = add_buf(start, count);
908 	q->first_to_check = start;
909 	qdio_kick_handler(q, count);
910 
911 	if (!qdio_inbound_q_done(q, start)) {
912 		qperf_inc(q, tasklet_inbound_resched);
913 		if (!qdio_tasklet_schedule(q))
914 			return;
915 	}
916 
917 	qdio_stop_polling(q);
918 	/*
919 	 * We need to check again to not lose initiative after
920 	 * resetting the ACK state.
921 	 */
922 	if (!qdio_inbound_q_done(q, start)) {
923 		qperf_inc(q, tasklet_inbound_resched2);
924 		qdio_tasklet_schedule(q);
925 	}
926 }
927 
928 void tiqdio_inbound_processing(unsigned long data)
929 {
930 	struct qdio_q *q = (struct qdio_q *)data;
931 	__tiqdio_inbound_processing(q);
932 }
933 
934 static inline void qdio_set_state(struct qdio_irq *irq_ptr,
935 				  enum qdio_irq_states state)
936 {
937 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "newstate: %1d", state);
938 
939 	irq_ptr->state = state;
940 	mb();
941 }
942 
943 static void qdio_irq_check_sense(struct qdio_irq *irq_ptr, struct irb *irb)
944 {
945 	if (irb->esw.esw0.erw.cons) {
946 		DBF_ERROR("%4x sense:", irq_ptr->schid.sch_no);
947 		DBF_ERROR_HEX(irb, 64);
948 		DBF_ERROR_HEX(irb->ecw, 64);
949 	}
950 }
951 
952 /* PCI interrupt handler */
953 static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
954 {
955 	int i;
956 	struct qdio_q *q;
957 
958 	if (unlikely(irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
959 		return;
960 
961 	for_each_input_queue(irq_ptr, q, i) {
962 		if (q->u.in.queue_start_poll) {
963 			/* skip if polling is enabled or already in work */
964 			if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
965 				     &q->u.in.queue_irq_state)) {
966 				qperf_inc(q, int_discarded);
967 				continue;
968 			}
969 			q->u.in.queue_start_poll(q->irq_ptr->cdev, q->nr,
970 						 q->irq_ptr->int_parm);
971 		} else {
972 			tasklet_schedule(&q->tasklet);
973 		}
974 	}
975 
976 	if (!pci_out_supported(irq_ptr) || !irq_ptr->scan_threshold)
977 		return;
978 
979 	for_each_output_queue(irq_ptr, q, i) {
980 		if (qdio_outbound_q_done(q))
981 			continue;
982 		if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
983 			qdio_siga_sync_q(q);
984 		qdio_tasklet_schedule(q);
985 	}
986 }
987 
988 static void qdio_handle_activate_check(struct ccw_device *cdev,
989 				unsigned long intparm, int cstat, int dstat)
990 {
991 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
992 	struct qdio_q *q;
993 	int count;
994 
995 	DBF_ERROR("%4x ACT CHECK", irq_ptr->schid.sch_no);
996 	DBF_ERROR("intp :%lx", intparm);
997 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
998 
999 	if (irq_ptr->nr_input_qs) {
1000 		q = irq_ptr->input_qs[0];
1001 	} else if (irq_ptr->nr_output_qs) {
1002 		q = irq_ptr->output_qs[0];
1003 	} else {
1004 		dump_stack();
1005 		goto no_handler;
1006 	}
1007 
1008 	count = sub_buf(q->first_to_check, q->first_to_kick);
1009 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_ACTIVATE,
1010 		   q->nr, q->first_to_kick, count, irq_ptr->int_parm);
1011 no_handler:
1012 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1013 	/*
1014 	 * In case of z/VM LGR (Live Guest Migration) QDIO recovery will happen.
1015 	 * Therefore we call the LGR detection function here.
1016 	 */
1017 	lgr_info_log();
1018 }
1019 
1020 static void qdio_establish_handle_irq(struct ccw_device *cdev, int cstat,
1021 				      int dstat)
1022 {
1023 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1024 
1025 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "qest irq");
1026 
1027 	if (cstat)
1028 		goto error;
1029 	if (dstat & ~(DEV_STAT_DEV_END | DEV_STAT_CHN_END))
1030 		goto error;
1031 	if (!(dstat & DEV_STAT_DEV_END))
1032 		goto error;
1033 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ESTABLISHED);
1034 	return;
1035 
1036 error:
1037 	DBF_ERROR("%4x EQ:error", irq_ptr->schid.sch_no);
1038 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1039 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1040 }
1041 
1042 /* qdio interrupt handler */
1043 void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
1044 		      struct irb *irb)
1045 {
1046 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1047 	struct subchannel_id schid;
1048 	int cstat, dstat;
1049 
1050 	if (!intparm || !irq_ptr) {
1051 		ccw_device_get_schid(cdev, &schid);
1052 		DBF_ERROR("qint:%4x", schid.sch_no);
1053 		return;
1054 	}
1055 
1056 	if (irq_ptr->perf_stat_enabled)
1057 		irq_ptr->perf_stat.qdio_int++;
1058 
1059 	if (IS_ERR(irb)) {
1060 		DBF_ERROR("%4x IO error", irq_ptr->schid.sch_no);
1061 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1062 		wake_up(&cdev->private->wait_q);
1063 		return;
1064 	}
1065 	qdio_irq_check_sense(irq_ptr, irb);
1066 	cstat = irb->scsw.cmd.cstat;
1067 	dstat = irb->scsw.cmd.dstat;
1068 
1069 	switch (irq_ptr->state) {
1070 	case QDIO_IRQ_STATE_INACTIVE:
1071 		qdio_establish_handle_irq(cdev, cstat, dstat);
1072 		break;
1073 	case QDIO_IRQ_STATE_CLEANUP:
1074 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1075 		break;
1076 	case QDIO_IRQ_STATE_ESTABLISHED:
1077 	case QDIO_IRQ_STATE_ACTIVE:
1078 		if (cstat & SCHN_STAT_PCI) {
1079 			qdio_int_handler_pci(irq_ptr);
1080 			return;
1081 		}
1082 		if (cstat || dstat)
1083 			qdio_handle_activate_check(cdev, intparm, cstat,
1084 						   dstat);
1085 		break;
1086 	case QDIO_IRQ_STATE_STOPPED:
1087 		break;
1088 	default:
1089 		WARN_ON_ONCE(1);
1090 	}
1091 	wake_up(&cdev->private->wait_q);
1092 }
1093 
1094 /**
1095  * qdio_get_ssqd_desc - get qdio subchannel description
1096  * @cdev: ccw device to get description for
1097  * @data: where to store the ssqd
1098  *
1099  * Returns 0 or an error code. The results of the chsc are stored in the
1100  * specified structure.
1101  */
1102 int qdio_get_ssqd_desc(struct ccw_device *cdev,
1103 		       struct qdio_ssqd_desc *data)
1104 {
1105 	struct subchannel_id schid;
1106 
1107 	if (!cdev || !cdev->private)
1108 		return -EINVAL;
1109 
1110 	ccw_device_get_schid(cdev, &schid);
1111 	DBF_EVENT("get ssqd:%4x", schid.sch_no);
1112 	return qdio_setup_get_ssqd(NULL, &schid, data);
1113 }
1114 EXPORT_SYMBOL_GPL(qdio_get_ssqd_desc);
1115 
1116 static void qdio_shutdown_queues(struct ccw_device *cdev)
1117 {
1118 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1119 	struct qdio_q *q;
1120 	int i;
1121 
1122 	for_each_input_queue(irq_ptr, q, i)
1123 		tasklet_kill(&q->tasklet);
1124 
1125 	for_each_output_queue(irq_ptr, q, i) {
1126 		del_timer_sync(&q->u.out.timer);
1127 		tasklet_kill(&q->tasklet);
1128 	}
1129 }
1130 
1131 /**
1132  * qdio_shutdown - shut down a qdio subchannel
1133  * @cdev: associated ccw device
1134  * @how: use halt or clear to shutdown
1135  */
1136 int qdio_shutdown(struct ccw_device *cdev, int how)
1137 {
1138 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1139 	struct subchannel_id schid;
1140 	int rc;
1141 
1142 	if (!irq_ptr)
1143 		return -ENODEV;
1144 
1145 	WARN_ON_ONCE(irqs_disabled());
1146 	ccw_device_get_schid(cdev, &schid);
1147 	DBF_EVENT("qshutdown:%4x", schid.sch_no);
1148 
1149 	mutex_lock(&irq_ptr->setup_mutex);
1150 	/*
1151 	 * Subchannel was already shot down. We cannot prevent being called
1152 	 * twice since cio may trigger a shutdown asynchronously.
1153 	 */
1154 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1155 		mutex_unlock(&irq_ptr->setup_mutex);
1156 		return 0;
1157 	}
1158 
1159 	/*
1160 	 * Indicate that the device is going down. Scheduling the queue
1161 	 * tasklets is forbidden from here on.
1162 	 */
1163 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1164 
1165 	tiqdio_remove_input_queues(irq_ptr);
1166 	qdio_shutdown_queues(cdev);
1167 	qdio_shutdown_debug_entries(irq_ptr);
1168 
1169 	/* cleanup subchannel */
1170 	spin_lock_irq(get_ccwdev_lock(cdev));
1171 
1172 	if (how & QDIO_FLAG_CLEANUP_USING_CLEAR)
1173 		rc = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
1174 	else
1175 		/* default behaviour is halt */
1176 		rc = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
1177 	if (rc) {
1178 		DBF_ERROR("%4x SHUTD ERR", irq_ptr->schid.sch_no);
1179 		DBF_ERROR("rc:%4d", rc);
1180 		goto no_cleanup;
1181 	}
1182 
1183 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
1184 	spin_unlock_irq(get_ccwdev_lock(cdev));
1185 	wait_event_interruptible_timeout(cdev->private->wait_q,
1186 		irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
1187 		irq_ptr->state == QDIO_IRQ_STATE_ERR,
1188 		10 * HZ);
1189 	spin_lock_irq(get_ccwdev_lock(cdev));
1190 
1191 no_cleanup:
1192 	qdio_shutdown_thinint(irq_ptr);
1193 
1194 	/* restore interrupt handler */
1195 	if ((void *)cdev->handler == (void *)qdio_int_handler) {
1196 		cdev->handler = irq_ptr->orig_handler;
1197 		cdev->private->intparm = 0;
1198 	}
1199 	spin_unlock_irq(get_ccwdev_lock(cdev));
1200 
1201 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1202 	mutex_unlock(&irq_ptr->setup_mutex);
1203 	if (rc)
1204 		return rc;
1205 	return 0;
1206 }
1207 EXPORT_SYMBOL_GPL(qdio_shutdown);
1208 
1209 /**
1210  * qdio_free - free data structures for a qdio subchannel
1211  * @cdev: associated ccw device
1212  */
1213 int qdio_free(struct ccw_device *cdev)
1214 {
1215 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1216 	struct subchannel_id schid;
1217 
1218 	if (!irq_ptr)
1219 		return -ENODEV;
1220 
1221 	ccw_device_get_schid(cdev, &schid);
1222 	DBF_EVENT("qfree:%4x", schid.sch_no);
1223 	DBF_DEV_EVENT(DBF_ERR, irq_ptr, "dbf abandoned");
1224 	mutex_lock(&irq_ptr->setup_mutex);
1225 
1226 	irq_ptr->debug_area = NULL;
1227 	cdev->private->qdio_data = NULL;
1228 	mutex_unlock(&irq_ptr->setup_mutex);
1229 
1230 	qdio_release_memory(irq_ptr);
1231 	return 0;
1232 }
1233 EXPORT_SYMBOL_GPL(qdio_free);
1234 
1235 /**
1236  * qdio_allocate - allocate qdio queues and associated data
1237  * @init_data: initialization data
1238  */
1239 int qdio_allocate(struct qdio_initialize *init_data)
1240 {
1241 	struct subchannel_id schid;
1242 	struct qdio_irq *irq_ptr;
1243 
1244 	ccw_device_get_schid(init_data->cdev, &schid);
1245 	DBF_EVENT("qallocate:%4x", schid.sch_no);
1246 
1247 	if ((init_data->no_input_qs && !init_data->input_handler) ||
1248 	    (init_data->no_output_qs && !init_data->output_handler))
1249 		return -EINVAL;
1250 
1251 	if ((init_data->no_input_qs > QDIO_MAX_QUEUES_PER_IRQ) ||
1252 	    (init_data->no_output_qs > QDIO_MAX_QUEUES_PER_IRQ))
1253 		return -EINVAL;
1254 
1255 	if ((!init_data->input_sbal_addr_array) ||
1256 	    (!init_data->output_sbal_addr_array))
1257 		return -EINVAL;
1258 
1259 	/* irq_ptr must be in GFP_DMA since it contains ccw1.cda */
1260 	irq_ptr = (void *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1261 	if (!irq_ptr)
1262 		goto out_err;
1263 
1264 	mutex_init(&irq_ptr->setup_mutex);
1265 	if (qdio_allocate_dbf(init_data, irq_ptr))
1266 		goto out_rel;
1267 
1268 	/*
1269 	 * Allocate a page for the chsc calls in qdio_establish.
1270 	 * Must be pre-allocated since a zfcp recovery will call
1271 	 * qdio_establish. In case of low memory and swap on a zfcp disk
1272 	 * we may not be able to allocate memory otherwise.
1273 	 */
1274 	irq_ptr->chsc_page = get_zeroed_page(GFP_KERNEL);
1275 	if (!irq_ptr->chsc_page)
1276 		goto out_rel;
1277 
1278 	/* qdr is used in ccw1.cda which is u32 */
1279 	irq_ptr->qdr = (struct qdr *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1280 	if (!irq_ptr->qdr)
1281 		goto out_rel;
1282 
1283 	if (qdio_allocate_qs(irq_ptr, init_data->no_input_qs,
1284 			     init_data->no_output_qs))
1285 		goto out_rel;
1286 
1287 	init_data->cdev->private->qdio_data = irq_ptr;
1288 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1289 	return 0;
1290 out_rel:
1291 	qdio_release_memory(irq_ptr);
1292 out_err:
1293 	return -ENOMEM;
1294 }
1295 EXPORT_SYMBOL_GPL(qdio_allocate);
1296 
1297 static void qdio_detect_hsicq(struct qdio_irq *irq_ptr)
1298 {
1299 	struct qdio_q *q = irq_ptr->input_qs[0];
1300 	int i, use_cq = 0;
1301 
1302 	if (irq_ptr->nr_input_qs > 1 && queue_type(q) == QDIO_IQDIO_QFMT)
1303 		use_cq = 1;
1304 
1305 	for_each_output_queue(irq_ptr, q, i) {
1306 		if (use_cq) {
1307 			if (multicast_outbound(q))
1308 				continue;
1309 			if (qdio_enable_async_operation(&q->u.out) < 0) {
1310 				use_cq = 0;
1311 				continue;
1312 			}
1313 		} else
1314 			qdio_disable_async_operation(&q->u.out);
1315 	}
1316 	DBF_EVENT("use_cq:%d", use_cq);
1317 }
1318 
1319 /**
1320  * qdio_establish - establish queues on a qdio subchannel
1321  * @init_data: initialization data
1322  */
1323 int qdio_establish(struct qdio_initialize *init_data)
1324 {
1325 	struct ccw_device *cdev = init_data->cdev;
1326 	struct subchannel_id schid;
1327 	struct qdio_irq *irq_ptr;
1328 	int rc;
1329 
1330 	ccw_device_get_schid(cdev, &schid);
1331 	DBF_EVENT("qestablish:%4x", schid.sch_no);
1332 
1333 	irq_ptr = cdev->private->qdio_data;
1334 	if (!irq_ptr)
1335 		return -ENODEV;
1336 
1337 	mutex_lock(&irq_ptr->setup_mutex);
1338 	qdio_setup_irq(init_data);
1339 
1340 	rc = qdio_establish_thinint(irq_ptr);
1341 	if (rc) {
1342 		mutex_unlock(&irq_ptr->setup_mutex);
1343 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1344 		return rc;
1345 	}
1346 
1347 	/* establish q */
1348 	irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
1349 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1350 	irq_ptr->ccw.count = irq_ptr->equeue.count;
1351 	irq_ptr->ccw.cda = (u32)((addr_t)irq_ptr->qdr);
1352 
1353 	spin_lock_irq(get_ccwdev_lock(cdev));
1354 	ccw_device_set_options_mask(cdev, 0);
1355 
1356 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ESTABLISH, 0, 0);
1357 	spin_unlock_irq(get_ccwdev_lock(cdev));
1358 	if (rc) {
1359 		DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
1360 		DBF_ERROR("rc:%4x", rc);
1361 		mutex_unlock(&irq_ptr->setup_mutex);
1362 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1363 		return rc;
1364 	}
1365 
1366 	wait_event_interruptible_timeout(cdev->private->wait_q,
1367 		irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
1368 		irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
1369 
1370 	if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
1371 		mutex_unlock(&irq_ptr->setup_mutex);
1372 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1373 		return -EIO;
1374 	}
1375 
1376 	qdio_setup_ssqd_info(irq_ptr);
1377 
1378 	qdio_detect_hsicq(irq_ptr);
1379 
1380 	/* qebsm is now setup if available, initialize buffer states */
1381 	qdio_init_buf_states(irq_ptr);
1382 
1383 	mutex_unlock(&irq_ptr->setup_mutex);
1384 	qdio_print_subchannel_info(irq_ptr, cdev);
1385 	qdio_setup_debug_entries(irq_ptr, cdev);
1386 	return 0;
1387 }
1388 EXPORT_SYMBOL_GPL(qdio_establish);
1389 
1390 /**
1391  * qdio_activate - activate queues on a qdio subchannel
1392  * @cdev: associated cdev
1393  */
1394 int qdio_activate(struct ccw_device *cdev)
1395 {
1396 	struct subchannel_id schid;
1397 	struct qdio_irq *irq_ptr;
1398 	int rc;
1399 
1400 	ccw_device_get_schid(cdev, &schid);
1401 	DBF_EVENT("qactivate:%4x", schid.sch_no);
1402 
1403 	irq_ptr = cdev->private->qdio_data;
1404 	if (!irq_ptr)
1405 		return -ENODEV;
1406 
1407 	mutex_lock(&irq_ptr->setup_mutex);
1408 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1409 		rc = -EBUSY;
1410 		goto out;
1411 	}
1412 
1413 	irq_ptr->ccw.cmd_code = irq_ptr->aqueue.cmd;
1414 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1415 	irq_ptr->ccw.count = irq_ptr->aqueue.count;
1416 	irq_ptr->ccw.cda = 0;
1417 
1418 	spin_lock_irq(get_ccwdev_lock(cdev));
1419 	ccw_device_set_options(cdev, CCWDEV_REPORT_ALL);
1420 
1421 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ACTIVATE,
1422 			      0, DOIO_DENY_PREFETCH);
1423 	spin_unlock_irq(get_ccwdev_lock(cdev));
1424 	if (rc) {
1425 		DBF_ERROR("%4x act IO ERR", irq_ptr->schid.sch_no);
1426 		DBF_ERROR("rc:%4x", rc);
1427 		goto out;
1428 	}
1429 
1430 	if (is_thinint_irq(irq_ptr))
1431 		tiqdio_add_input_queues(irq_ptr);
1432 
1433 	/* wait for subchannel to become active */
1434 	msleep(5);
1435 
1436 	switch (irq_ptr->state) {
1437 	case QDIO_IRQ_STATE_STOPPED:
1438 	case QDIO_IRQ_STATE_ERR:
1439 		rc = -EIO;
1440 		break;
1441 	default:
1442 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ACTIVE);
1443 		rc = 0;
1444 	}
1445 out:
1446 	mutex_unlock(&irq_ptr->setup_mutex);
1447 	return rc;
1448 }
1449 EXPORT_SYMBOL_GPL(qdio_activate);
1450 
1451 static inline int buf_in_between(int bufnr, int start, int count)
1452 {
1453 	int end = add_buf(start, count);
1454 
1455 	if (end > start) {
1456 		if (bufnr >= start && bufnr < end)
1457 			return 1;
1458 		else
1459 			return 0;
1460 	}
1461 
1462 	/* wrap-around case */
1463 	if ((bufnr >= start && bufnr <= QDIO_MAX_BUFFERS_PER_Q) ||
1464 	    (bufnr < end))
1465 		return 1;
1466 	else
1467 		return 0;
1468 }
1469 
1470 /**
1471  * handle_inbound - reset processed input buffers
1472  * @q: queue containing the buffers
1473  * @callflags: flags
1474  * @bufnr: first buffer to process
1475  * @count: how many buffers are emptied
1476  */
1477 static int handle_inbound(struct qdio_q *q, unsigned int callflags,
1478 			  int bufnr, int count)
1479 {
1480 	int diff;
1481 
1482 	qperf_inc(q, inbound_call);
1483 
1484 	if (!q->u.in.polling)
1485 		goto set;
1486 
1487 	/* protect against stop polling setting an ACK for an emptied slsb */
1488 	if (count == QDIO_MAX_BUFFERS_PER_Q) {
1489 		/* overwriting everything, just delete polling status */
1490 		q->u.in.polling = 0;
1491 		q->u.in.ack_count = 0;
1492 		goto set;
1493 	} else if (buf_in_between(q->u.in.ack_start, bufnr, count)) {
1494 		if (is_qebsm(q)) {
1495 			/* partial overwrite, just update ack_start */
1496 			diff = add_buf(bufnr, count);
1497 			diff = sub_buf(diff, q->u.in.ack_start);
1498 			q->u.in.ack_count -= diff;
1499 			if (q->u.in.ack_count <= 0) {
1500 				q->u.in.polling = 0;
1501 				q->u.in.ack_count = 0;
1502 				goto set;
1503 			}
1504 			q->u.in.ack_start = add_buf(q->u.in.ack_start, diff);
1505 		}
1506 		else
1507 			/* the only ACK will be deleted, so stop polling */
1508 			q->u.in.polling = 0;
1509 	}
1510 
1511 set:
1512 	count = set_buf_states(q, bufnr, SLSB_CU_INPUT_EMPTY, count);
1513 	atomic_add(count, &q->nr_buf_used);
1514 
1515 	if (need_siga_in(q))
1516 		return qdio_siga_input(q);
1517 
1518 	return 0;
1519 }
1520 
1521 /**
1522  * handle_outbound - process filled outbound buffers
1523  * @q: queue containing the buffers
1524  * @callflags: flags
1525  * @bufnr: first buffer to process
1526  * @count: how many buffers are filled
1527  */
1528 static int handle_outbound(struct qdio_q *q, unsigned int callflags,
1529 			   int bufnr, int count)
1530 {
1531 	const unsigned int scan_threshold = q->irq_ptr->scan_threshold;
1532 	unsigned char state = 0;
1533 	int used, rc = 0;
1534 
1535 	qperf_inc(q, outbound_call);
1536 
1537 	count = set_buf_states(q, bufnr, SLSB_CU_OUTPUT_PRIMED, count);
1538 	used = atomic_add_return(count, &q->nr_buf_used);
1539 
1540 	if (used == QDIO_MAX_BUFFERS_PER_Q)
1541 		qperf_inc(q, outbound_queue_full);
1542 
1543 	if (callflags & QDIO_FLAG_PCI_OUT) {
1544 		q->u.out.pci_out_enabled = 1;
1545 		qperf_inc(q, pci_request_int);
1546 	} else
1547 		q->u.out.pci_out_enabled = 0;
1548 
1549 	if (queue_type(q) == QDIO_IQDIO_QFMT) {
1550 		unsigned long phys_aob = 0;
1551 
1552 		/* One SIGA-W per buffer required for unicast HSI */
1553 		WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
1554 
1555 		if (q->u.out.use_cq)
1556 			phys_aob = qdio_aob_for_buffer(&q->u.out, bufnr);
1557 
1558 		rc = qdio_kick_outbound_q(q, phys_aob);
1559 	} else if (need_siga_sync(q)) {
1560 		rc = qdio_siga_sync_q(q);
1561 	} else if (count < QDIO_MAX_BUFFERS_PER_Q &&
1562 		   get_buf_state(q, prev_buf(bufnr), &state, 0) > 0 &&
1563 		   state == SLSB_CU_OUTPUT_PRIMED) {
1564 		/* The previous buffer is not processed yet, tack on. */
1565 		qperf_inc(q, fast_requeue);
1566 	} else {
1567 		rc = qdio_kick_outbound_q(q, 0);
1568 	}
1569 
1570 	/* Let drivers implement their own completion scanning: */
1571 	if (!scan_threshold)
1572 		return rc;
1573 
1574 	/* in case of SIGA errors we must process the error immediately */
1575 	if (used >= scan_threshold || rc)
1576 		qdio_tasklet_schedule(q);
1577 	else
1578 		/* free the SBALs in case of no further traffic */
1579 		if (!timer_pending(&q->u.out.timer) &&
1580 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
1581 			mod_timer(&q->u.out.timer, jiffies + HZ);
1582 	return rc;
1583 }
1584 
1585 /**
1586  * do_QDIO - process input or output buffers
1587  * @cdev: associated ccw_device for the qdio subchannel
1588  * @callflags: input or output and special flags from the program
1589  * @q_nr: queue number
1590  * @bufnr: buffer number
1591  * @count: how many buffers to process
1592  */
1593 int do_QDIO(struct ccw_device *cdev, unsigned int callflags,
1594 	    int q_nr, unsigned int bufnr, unsigned int count)
1595 {
1596 	struct qdio_irq *irq_ptr;
1597 
1598 	if (bufnr >= QDIO_MAX_BUFFERS_PER_Q || count > QDIO_MAX_BUFFERS_PER_Q)
1599 		return -EINVAL;
1600 
1601 	irq_ptr = cdev->private->qdio_data;
1602 	if (!irq_ptr)
1603 		return -ENODEV;
1604 
1605 	DBF_DEV_EVENT(DBF_INFO, irq_ptr,
1606 		      "do%02x b:%02x c:%02x", callflags, bufnr, count);
1607 
1608 	if (irq_ptr->state != QDIO_IRQ_STATE_ACTIVE)
1609 		return -EIO;
1610 	if (!count)
1611 		return 0;
1612 	if (callflags & QDIO_FLAG_SYNC_INPUT)
1613 		return handle_inbound(irq_ptr->input_qs[q_nr],
1614 				      callflags, bufnr, count);
1615 	else if (callflags & QDIO_FLAG_SYNC_OUTPUT)
1616 		return handle_outbound(irq_ptr->output_qs[q_nr],
1617 				       callflags, bufnr, count);
1618 	return -EINVAL;
1619 }
1620 EXPORT_SYMBOL_GPL(do_QDIO);
1621 
1622 /**
1623  * qdio_start_irq - process input buffers
1624  * @cdev: associated ccw_device for the qdio subchannel
1625  * @nr: input queue number
1626  *
1627  * Return codes
1628  *   0 - success
1629  *   1 - irqs not started since new data is available
1630  */
1631 int qdio_start_irq(struct ccw_device *cdev, int nr)
1632 {
1633 	struct qdio_q *q;
1634 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1635 
1636 	if (!irq_ptr)
1637 		return -ENODEV;
1638 	q = irq_ptr->input_qs[nr];
1639 
1640 	clear_nonshared_ind(irq_ptr);
1641 	qdio_stop_polling(q);
1642 	clear_bit(QDIO_QUEUE_IRQS_DISABLED, &q->u.in.queue_irq_state);
1643 
1644 	/*
1645 	 * We need to check again to not lose initiative after
1646 	 * resetting the ACK state.
1647 	 */
1648 	if (test_nonshared_ind(irq_ptr))
1649 		goto rescan;
1650 	if (!qdio_inbound_q_done(q, q->first_to_check))
1651 		goto rescan;
1652 	return 0;
1653 
1654 rescan:
1655 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1656 			     &q->u.in.queue_irq_state))
1657 		return 0;
1658 	else
1659 		return 1;
1660 
1661 }
1662 EXPORT_SYMBOL(qdio_start_irq);
1663 
1664 static int __qdio_inspect_queue(struct qdio_q *q, unsigned int *bufnr,
1665 				unsigned int *error)
1666 {
1667 	unsigned int start = q->first_to_check;
1668 	int count;
1669 
1670 	count = q->is_input_q ? qdio_inbound_q_moved(q, start) :
1671 				qdio_outbound_q_moved(q, start);
1672 	if (count == 0)
1673 		return 0;
1674 
1675 	*bufnr = start;
1676 	*error = q->qdio_error;
1677 
1678 	/* for the next time */
1679 	q->first_to_check = add_buf(start, count);
1680 	q->qdio_error = 0;
1681 
1682 	return count;
1683 }
1684 
1685 int qdio_inspect_queue(struct ccw_device *cdev, unsigned int nr, bool is_input,
1686 		       unsigned int *bufnr, unsigned int *error)
1687 {
1688 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1689 	struct qdio_q *q;
1690 
1691 	if (!irq_ptr)
1692 		return -ENODEV;
1693 	q = is_input ? irq_ptr->input_qs[nr] : irq_ptr->output_qs[nr];
1694 
1695 	if (need_siga_sync(q))
1696 		qdio_siga_sync_q(q);
1697 
1698 	return __qdio_inspect_queue(q, bufnr, error);
1699 }
1700 EXPORT_SYMBOL_GPL(qdio_inspect_queue);
1701 
1702 /**
1703  * qdio_get_next_buffers - process input buffers
1704  * @cdev: associated ccw_device for the qdio subchannel
1705  * @nr: input queue number
1706  * @bufnr: first filled buffer number
1707  * @error: buffers are in error state
1708  *
1709  * Return codes
1710  *   < 0 - error
1711  *   = 0 - no new buffers found
1712  *   > 0 - number of processed buffers
1713  */
1714 int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
1715 			  int *error)
1716 {
1717 	struct qdio_q *q;
1718 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1719 
1720 	if (!irq_ptr)
1721 		return -ENODEV;
1722 	q = irq_ptr->input_qs[nr];
1723 
1724 	/*
1725 	 * Cannot rely on automatic sync after interrupt since queues may
1726 	 * also be examined without interrupt.
1727 	 */
1728 	if (need_siga_sync(q))
1729 		qdio_sync_queues(q);
1730 
1731 	qdio_check_outbound_pci_queues(irq_ptr);
1732 
1733 	/* Note: upper-layer MUST stop processing immediately here ... */
1734 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
1735 		return -EIO;
1736 
1737 	return __qdio_inspect_queue(q, bufnr, error);
1738 }
1739 EXPORT_SYMBOL(qdio_get_next_buffers);
1740 
1741 /**
1742  * qdio_stop_irq - disable interrupt processing for the device
1743  * @cdev: associated ccw_device for the qdio subchannel
1744  * @nr: input queue number
1745  *
1746  * Return codes
1747  *   0 - interrupts were already disabled
1748  *   1 - interrupts successfully disabled
1749  */
1750 int qdio_stop_irq(struct ccw_device *cdev, int nr)
1751 {
1752 	struct qdio_q *q;
1753 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1754 
1755 	if (!irq_ptr)
1756 		return -ENODEV;
1757 	q = irq_ptr->input_qs[nr];
1758 
1759 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1760 			     &q->u.in.queue_irq_state))
1761 		return 0;
1762 	else
1763 		return 1;
1764 }
1765 EXPORT_SYMBOL(qdio_stop_irq);
1766 
1767 /**
1768  * qdio_pnso_brinfo() - perform network subchannel op #0 - bridge info.
1769  * @schid:		Subchannel ID.
1770  * @cnc:		Boolean Change-Notification Control
1771  * @response:		Response code will be stored at this address
1772  * @cb: 		Callback function will be executed for each element
1773  *			of the address list
1774  * @priv:		Pointer to pass to the callback function.
1775  *
1776  * Performs "Store-network-bridging-information list" operation and calls
1777  * the callback function for every entry in the list. If "change-
1778  * notification-control" is set, further changes in the address list
1779  * will be reported via the IPA command.
1780  */
1781 int qdio_pnso_brinfo(struct subchannel_id schid,
1782 		int cnc, u16 *response,
1783 		void (*cb)(void *priv, enum qdio_brinfo_entry_type type,
1784 				void *entry),
1785 		void *priv)
1786 {
1787 	struct chsc_pnso_area *rr;
1788 	int rc;
1789 	u32 prev_instance = 0;
1790 	int isfirstblock = 1;
1791 	int i, size, elems;
1792 
1793 	rr = (struct chsc_pnso_area *)get_zeroed_page(GFP_KERNEL);
1794 	if (rr == NULL)
1795 		return -ENOMEM;
1796 	do {
1797 		/* on the first iteration, naihdr.resume_token will be zero */
1798 		rc = chsc_pnso_brinfo(schid, rr, rr->naihdr.resume_token, cnc);
1799 		if (rc != 0 && rc != -EBUSY)
1800 			goto out;
1801 		if (rr->response.code != 1) {
1802 			rc = -EIO;
1803 			continue;
1804 		} else
1805 			rc = 0;
1806 
1807 		if (cb == NULL)
1808 			continue;
1809 
1810 		size = rr->naihdr.naids;
1811 		elems = (rr->response.length -
1812 				sizeof(struct chsc_header) -
1813 				sizeof(struct chsc_brinfo_naihdr)) /
1814 				size;
1815 
1816 		if (!isfirstblock && (rr->naihdr.instance != prev_instance)) {
1817 			/* Inform the caller that they need to scrap */
1818 			/* the data that was already reported via cb */
1819 				rc = -EAGAIN;
1820 				break;
1821 		}
1822 		isfirstblock = 0;
1823 		prev_instance = rr->naihdr.instance;
1824 		for (i = 0; i < elems; i++)
1825 			switch (size) {
1826 			case sizeof(struct qdio_brinfo_entry_l3_ipv6):
1827 				(*cb)(priv, l3_ipv6_addr,
1828 						&rr->entries.l3_ipv6[i]);
1829 				break;
1830 			case sizeof(struct qdio_brinfo_entry_l3_ipv4):
1831 				(*cb)(priv, l3_ipv4_addr,
1832 						&rr->entries.l3_ipv4[i]);
1833 				break;
1834 			case sizeof(struct qdio_brinfo_entry_l2):
1835 				(*cb)(priv, l2_addr_lnid,
1836 						&rr->entries.l2[i]);
1837 				break;
1838 			default:
1839 				WARN_ON_ONCE(1);
1840 				rc = -EIO;
1841 				goto out;
1842 			}
1843 	} while (rr->response.code == 0x0107 ||  /* channel busy */
1844 		  (rr->response.code == 1 && /* list stored */
1845 		   /* resume token is non-zero => list incomplete */
1846 		   (rr->naihdr.resume_token.t1 || rr->naihdr.resume_token.t2)));
1847 	(*response) = rr->response.code;
1848 
1849 out:
1850 	free_page((unsigned long)rr);
1851 	return rc;
1852 }
1853 EXPORT_SYMBOL_GPL(qdio_pnso_brinfo);
1854 
1855 static int __init init_QDIO(void)
1856 {
1857 	int rc;
1858 
1859 	rc = qdio_debug_init();
1860 	if (rc)
1861 		return rc;
1862 	rc = qdio_setup_init();
1863 	if (rc)
1864 		goto out_debug;
1865 	rc = tiqdio_allocate_memory();
1866 	if (rc)
1867 		goto out_cache;
1868 	rc = tiqdio_register_thinints();
1869 	if (rc)
1870 		goto out_ti;
1871 	return 0;
1872 
1873 out_ti:
1874 	tiqdio_free_memory();
1875 out_cache:
1876 	qdio_setup_exit();
1877 out_debug:
1878 	qdio_debug_exit();
1879 	return rc;
1880 }
1881 
1882 static void __exit exit_QDIO(void)
1883 {
1884 	tiqdio_unregister_thinints();
1885 	tiqdio_free_memory();
1886 	qdio_setup_exit();
1887 	qdio_debug_exit();
1888 }
1889 
1890 module_init(init_QDIO);
1891 module_exit(exit_QDIO);
1892