xref: /openbmc/linux/drivers/s390/cio/qdio_main.c (revision 98ddec80)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Linux for s390 qdio support, buffer handling, qdio API and module support.
4  *
5  * Copyright IBM Corp. 2000, 2008
6  * Author(s): Utz Bacher <utz.bacher@de.ibm.com>
7  *	      Jan Glauber <jang@linux.vnet.ibm.com>
8  * 2.6 cio integration by Cornelia Huck <cornelia.huck@de.ibm.com>
9  */
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/timer.h>
14 #include <linux/delay.h>
15 #include <linux/gfp.h>
16 #include <linux/io.h>
17 #include <linux/atomic.h>
18 #include <asm/debug.h>
19 #include <asm/qdio.h>
20 #include <asm/ipl.h>
21 
22 #include "cio.h"
23 #include "css.h"
24 #include "device.h"
25 #include "qdio.h"
26 #include "qdio_debug.h"
27 
28 MODULE_AUTHOR("Utz Bacher <utz.bacher@de.ibm.com>,"\
29 	"Jan Glauber <jang@linux.vnet.ibm.com>");
30 MODULE_DESCRIPTION("QDIO base support");
31 MODULE_LICENSE("GPL");
32 
33 static inline int do_siga_sync(unsigned long schid,
34 			       unsigned int out_mask, unsigned int in_mask,
35 			       unsigned int fc)
36 {
37 	register unsigned long __fc asm ("0") = fc;
38 	register unsigned long __schid asm ("1") = schid;
39 	register unsigned long out asm ("2") = out_mask;
40 	register unsigned long in asm ("3") = in_mask;
41 	int cc;
42 
43 	asm volatile(
44 		"	siga	0\n"
45 		"	ipm	%0\n"
46 		"	srl	%0,28\n"
47 		: "=d" (cc)
48 		: "d" (__fc), "d" (__schid), "d" (out), "d" (in) : "cc");
49 	return cc;
50 }
51 
52 static inline int do_siga_input(unsigned long schid, unsigned int mask,
53 				unsigned int fc)
54 {
55 	register unsigned long __fc asm ("0") = fc;
56 	register unsigned long __schid asm ("1") = schid;
57 	register unsigned long __mask asm ("2") = mask;
58 	int cc;
59 
60 	asm volatile(
61 		"	siga	0\n"
62 		"	ipm	%0\n"
63 		"	srl	%0,28\n"
64 		: "=d" (cc)
65 		: "d" (__fc), "d" (__schid), "d" (__mask) : "cc");
66 	return cc;
67 }
68 
69 /**
70  * do_siga_output - perform SIGA-w/wt function
71  * @schid: subchannel id or in case of QEBSM the subchannel token
72  * @mask: which output queues to process
73  * @bb: busy bit indicator, set only if SIGA-w/wt could not access a buffer
74  * @fc: function code to perform
75  * @aob: asynchronous operation block
76  *
77  * Returns condition code.
78  * Note: For IQDC unicast queues only the highest priority queue is processed.
79  */
80 static inline int do_siga_output(unsigned long schid, unsigned long mask,
81 				 unsigned int *bb, unsigned int fc,
82 				 unsigned long aob)
83 {
84 	register unsigned long __fc asm("0") = fc;
85 	register unsigned long __schid asm("1") = schid;
86 	register unsigned long __mask asm("2") = mask;
87 	register unsigned long __aob asm("3") = aob;
88 	int cc;
89 
90 	asm volatile(
91 		"	siga	0\n"
92 		"	ipm	%0\n"
93 		"	srl	%0,28\n"
94 		: "=d" (cc), "+d" (__fc), "+d" (__aob)
95 		: "d" (__schid), "d" (__mask)
96 		: "cc");
97 	*bb = __fc >> 31;
98 	return cc;
99 }
100 
101 /**
102  * qdio_do_eqbs - extract buffer states for QEBSM
103  * @q: queue to manipulate
104  * @state: state of the extracted buffers
105  * @start: buffer number to start at
106  * @count: count of buffers to examine
107  * @auto_ack: automatically acknowledge buffers
108  *
109  * Returns the number of successfully extracted equal buffer states.
110  * Stops processing if a state is different from the last buffers state.
111  */
112 static int qdio_do_eqbs(struct qdio_q *q, unsigned char *state,
113 			int start, int count, int auto_ack)
114 {
115 	int tmp_count = count, tmp_start = start, nr = q->nr;
116 	unsigned int ccq = 0;
117 
118 	qperf_inc(q, eqbs);
119 
120 	if (!q->is_input_q)
121 		nr += q->irq_ptr->nr_input_qs;
122 again:
123 	ccq = do_eqbs(q->irq_ptr->sch_token, state, nr, &tmp_start, &tmp_count,
124 		      auto_ack);
125 
126 	switch (ccq) {
127 	case 0:
128 	case 32:
129 		/* all done, or next buffer state different */
130 		return count - tmp_count;
131 	case 96:
132 		/* not all buffers processed */
133 		qperf_inc(q, eqbs_partial);
134 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "EQBS part:%02x",
135 			tmp_count);
136 		return count - tmp_count;
137 	case 97:
138 		/* no buffer processed */
139 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "EQBS again:%2d", ccq);
140 		goto again;
141 	default:
142 		DBF_ERROR("%4x ccq:%3d", SCH_NO(q), ccq);
143 		DBF_ERROR("%4x EQBS ERROR", SCH_NO(q));
144 		DBF_ERROR("%3d%3d%2d", count, tmp_count, nr);
145 		q->handler(q->irq_ptr->cdev, QDIO_ERROR_GET_BUF_STATE, q->nr,
146 			   q->first_to_kick, count, q->irq_ptr->int_parm);
147 		return 0;
148 	}
149 }
150 
151 /**
152  * qdio_do_sqbs - set buffer states for QEBSM
153  * @q: queue to manipulate
154  * @state: new state of the buffers
155  * @start: first buffer number to change
156  * @count: how many buffers to change
157  *
158  * Returns the number of successfully changed buffers.
159  * Does retrying until the specified count of buffer states is set or an
160  * error occurs.
161  */
162 static int qdio_do_sqbs(struct qdio_q *q, unsigned char state, int start,
163 			int count)
164 {
165 	unsigned int ccq = 0;
166 	int tmp_count = count, tmp_start = start;
167 	int nr = q->nr;
168 
169 	if (!count)
170 		return 0;
171 	qperf_inc(q, sqbs);
172 
173 	if (!q->is_input_q)
174 		nr += q->irq_ptr->nr_input_qs;
175 again:
176 	ccq = do_sqbs(q->irq_ptr->sch_token, state, nr, &tmp_start, &tmp_count);
177 
178 	switch (ccq) {
179 	case 0:
180 	case 32:
181 		/* all done, or active buffer adapter-owned */
182 		WARN_ON_ONCE(tmp_count);
183 		return count - tmp_count;
184 	case 96:
185 		/* not all buffers processed */
186 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "SQBS again:%2d", ccq);
187 		qperf_inc(q, sqbs_partial);
188 		goto again;
189 	default:
190 		DBF_ERROR("%4x ccq:%3d", SCH_NO(q), ccq);
191 		DBF_ERROR("%4x SQBS ERROR", SCH_NO(q));
192 		DBF_ERROR("%3d%3d%2d", count, tmp_count, nr);
193 		q->handler(q->irq_ptr->cdev, QDIO_ERROR_SET_BUF_STATE, q->nr,
194 			   q->first_to_kick, count, q->irq_ptr->int_parm);
195 		return 0;
196 	}
197 }
198 
199 /*
200  * Returns number of examined buffers and their common state in *state.
201  * Requested number of buffers-to-examine must be > 0.
202  */
203 static inline int get_buf_states(struct qdio_q *q, unsigned int bufnr,
204 				 unsigned char *state, unsigned int count,
205 				 int auto_ack, int merge_pending)
206 {
207 	unsigned char __state = 0;
208 	int i;
209 
210 	if (is_qebsm(q))
211 		return qdio_do_eqbs(q, state, bufnr, count, auto_ack);
212 
213 	/* get initial state: */
214 	__state = q->slsb.val[bufnr];
215 	if (merge_pending && __state == SLSB_P_OUTPUT_PENDING)
216 		__state = SLSB_P_OUTPUT_EMPTY;
217 
218 	for (i = 1; i < count; i++) {
219 		bufnr = next_buf(bufnr);
220 
221 		/* merge PENDING into EMPTY: */
222 		if (merge_pending &&
223 		    q->slsb.val[bufnr] == SLSB_P_OUTPUT_PENDING &&
224 		    __state == SLSB_P_OUTPUT_EMPTY)
225 			continue;
226 
227 		/* stop if next state differs from initial state: */
228 		if (q->slsb.val[bufnr] != __state)
229 			break;
230 	}
231 	*state = __state;
232 	return i;
233 }
234 
235 static inline int get_buf_state(struct qdio_q *q, unsigned int bufnr,
236 				unsigned char *state, int auto_ack)
237 {
238 	return get_buf_states(q, bufnr, state, 1, auto_ack, 0);
239 }
240 
241 /* wrap-around safe setting of slsb states, returns number of changed buffers */
242 static inline int set_buf_states(struct qdio_q *q, int bufnr,
243 				 unsigned char state, int count)
244 {
245 	int i;
246 
247 	if (is_qebsm(q))
248 		return qdio_do_sqbs(q, state, bufnr, count);
249 
250 	for (i = 0; i < count; i++) {
251 		xchg(&q->slsb.val[bufnr], state);
252 		bufnr = next_buf(bufnr);
253 	}
254 	return count;
255 }
256 
257 static inline int set_buf_state(struct qdio_q *q, int bufnr,
258 				unsigned char state)
259 {
260 	return set_buf_states(q, bufnr, state, 1);
261 }
262 
263 /* set slsb states to initial state */
264 static void qdio_init_buf_states(struct qdio_irq *irq_ptr)
265 {
266 	struct qdio_q *q;
267 	int i;
268 
269 	for_each_input_queue(irq_ptr, q, i)
270 		set_buf_states(q, 0, SLSB_P_INPUT_NOT_INIT,
271 			       QDIO_MAX_BUFFERS_PER_Q);
272 	for_each_output_queue(irq_ptr, q, i)
273 		set_buf_states(q, 0, SLSB_P_OUTPUT_NOT_INIT,
274 			       QDIO_MAX_BUFFERS_PER_Q);
275 }
276 
277 static inline int qdio_siga_sync(struct qdio_q *q, unsigned int output,
278 			  unsigned int input)
279 {
280 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
281 	unsigned int fc = QDIO_SIGA_SYNC;
282 	int cc;
283 
284 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-s:%1d", q->nr);
285 	qperf_inc(q, siga_sync);
286 
287 	if (is_qebsm(q)) {
288 		schid = q->irq_ptr->sch_token;
289 		fc |= QDIO_SIGA_QEBSM_FLAG;
290 	}
291 
292 	cc = do_siga_sync(schid, output, input, fc);
293 	if (unlikely(cc))
294 		DBF_ERROR("%4x SIGA-S:%2d", SCH_NO(q), cc);
295 	return (cc) ? -EIO : 0;
296 }
297 
298 static inline int qdio_siga_sync_q(struct qdio_q *q)
299 {
300 	if (q->is_input_q)
301 		return qdio_siga_sync(q, 0, q->mask);
302 	else
303 		return qdio_siga_sync(q, q->mask, 0);
304 }
305 
306 static int qdio_siga_output(struct qdio_q *q, unsigned int *busy_bit,
307 	unsigned long aob)
308 {
309 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
310 	unsigned int fc = QDIO_SIGA_WRITE;
311 	u64 start_time = 0;
312 	int retries = 0, cc;
313 	unsigned long laob = 0;
314 
315 	WARN_ON_ONCE(aob && ((queue_type(q) != QDIO_IQDIO_QFMT) ||
316 			     !q->u.out.use_cq));
317 	if (q->u.out.use_cq && aob != 0) {
318 		fc = QDIO_SIGA_WRITEQ;
319 		laob = aob;
320 	}
321 
322 	if (is_qebsm(q)) {
323 		schid = q->irq_ptr->sch_token;
324 		fc |= QDIO_SIGA_QEBSM_FLAG;
325 	}
326 again:
327 	cc = do_siga_output(schid, q->mask, busy_bit, fc, laob);
328 
329 	/* hipersocket busy condition */
330 	if (unlikely(*busy_bit)) {
331 		retries++;
332 
333 		if (!start_time) {
334 			start_time = get_tod_clock_fast();
335 			goto again;
336 		}
337 		if (get_tod_clock_fast() - start_time < QDIO_BUSY_BIT_PATIENCE)
338 			goto again;
339 	}
340 	if (retries) {
341 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr,
342 			      "%4x cc2 BB1:%1d", SCH_NO(q), q->nr);
343 		DBF_DEV_EVENT(DBF_WARN, q->irq_ptr, "count:%u", retries);
344 	}
345 	return cc;
346 }
347 
348 static inline int qdio_siga_input(struct qdio_q *q)
349 {
350 	unsigned long schid = *((u32 *) &q->irq_ptr->schid);
351 	unsigned int fc = QDIO_SIGA_READ;
352 	int cc;
353 
354 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-r:%1d", q->nr);
355 	qperf_inc(q, siga_read);
356 
357 	if (is_qebsm(q)) {
358 		schid = q->irq_ptr->sch_token;
359 		fc |= QDIO_SIGA_QEBSM_FLAG;
360 	}
361 
362 	cc = do_siga_input(schid, q->mask, fc);
363 	if (unlikely(cc))
364 		DBF_ERROR("%4x SIGA-R:%2d", SCH_NO(q), cc);
365 	return (cc) ? -EIO : 0;
366 }
367 
368 #define qdio_siga_sync_out(q) qdio_siga_sync(q, ~0U, 0)
369 #define qdio_siga_sync_all(q) qdio_siga_sync(q, ~0U, ~0U)
370 
371 static inline void qdio_sync_queues(struct qdio_q *q)
372 {
373 	/* PCI capable outbound queues will also be scanned so sync them too */
374 	if (pci_out_supported(q))
375 		qdio_siga_sync_all(q);
376 	else
377 		qdio_siga_sync_q(q);
378 }
379 
380 int debug_get_buf_state(struct qdio_q *q, unsigned int bufnr,
381 			unsigned char *state)
382 {
383 	if (need_siga_sync(q))
384 		qdio_siga_sync_q(q);
385 	return get_buf_states(q, bufnr, state, 1, 0, 0);
386 }
387 
388 static inline void qdio_stop_polling(struct qdio_q *q)
389 {
390 	if (!q->u.in.polling)
391 		return;
392 
393 	q->u.in.polling = 0;
394 	qperf_inc(q, stop_polling);
395 
396 	/* show the card that we are not polling anymore */
397 	if (is_qebsm(q)) {
398 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
399 			       q->u.in.ack_count);
400 		q->u.in.ack_count = 0;
401 	} else
402 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
403 }
404 
405 static inline void account_sbals(struct qdio_q *q, unsigned int count)
406 {
407 	int pos;
408 
409 	q->q_stats.nr_sbal_total += count;
410 	if (count == QDIO_MAX_BUFFERS_MASK) {
411 		q->q_stats.nr_sbals[7]++;
412 		return;
413 	}
414 	pos = ilog2(count);
415 	q->q_stats.nr_sbals[pos]++;
416 }
417 
418 static void process_buffer_error(struct qdio_q *q, int count)
419 {
420 	unsigned char state = (q->is_input_q) ? SLSB_P_INPUT_NOT_INIT :
421 					SLSB_P_OUTPUT_NOT_INIT;
422 
423 	q->qdio_error = QDIO_ERROR_SLSB_STATE;
424 
425 	/* special handling for no target buffer empty */
426 	if (queue_type(q) == QDIO_IQDIO_QFMT && !q->is_input_q &&
427 	    q->sbal[q->first_to_check]->element[15].sflags == 0x10) {
428 		qperf_inc(q, target_full);
429 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "OUTFULL FTC:%02x",
430 			      q->first_to_check);
431 		goto set;
432 	}
433 
434 	DBF_ERROR("%4x BUF ERROR", SCH_NO(q));
435 	DBF_ERROR((q->is_input_q) ? "IN:%2d" : "OUT:%2d", q->nr);
436 	DBF_ERROR("FTC:%3d C:%3d", q->first_to_check, count);
437 	DBF_ERROR("F14:%2x F15:%2x",
438 		  q->sbal[q->first_to_check]->element[14].sflags,
439 		  q->sbal[q->first_to_check]->element[15].sflags);
440 
441 set:
442 	/*
443 	 * Interrupts may be avoided as long as the error is present
444 	 * so change the buffer state immediately to avoid starvation.
445 	 */
446 	set_buf_states(q, q->first_to_check, state, count);
447 }
448 
449 static inline void inbound_primed(struct qdio_q *q, int count)
450 {
451 	int new;
452 
453 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in prim:%1d %02x", q->nr, count);
454 
455 	/* for QEBSM the ACK was already set by EQBS */
456 	if (is_qebsm(q)) {
457 		if (!q->u.in.polling) {
458 			q->u.in.polling = 1;
459 			q->u.in.ack_count = count;
460 			q->u.in.ack_start = q->first_to_check;
461 			return;
462 		}
463 
464 		/* delete the previous ACK's */
465 		set_buf_states(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT,
466 			       q->u.in.ack_count);
467 		q->u.in.ack_count = count;
468 		q->u.in.ack_start = q->first_to_check;
469 		return;
470 	}
471 
472 	/*
473 	 * ACK the newest buffer. The ACK will be removed in qdio_stop_polling
474 	 * or by the next inbound run.
475 	 */
476 	new = add_buf(q->first_to_check, count - 1);
477 	if (q->u.in.polling) {
478 		/* reset the previous ACK but first set the new one */
479 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
480 		set_buf_state(q, q->u.in.ack_start, SLSB_P_INPUT_NOT_INIT);
481 	} else {
482 		q->u.in.polling = 1;
483 		set_buf_state(q, new, SLSB_P_INPUT_ACK);
484 	}
485 
486 	q->u.in.ack_start = new;
487 	count--;
488 	if (!count)
489 		return;
490 	/* need to change ALL buffers to get more interrupts */
491 	set_buf_states(q, q->first_to_check, SLSB_P_INPUT_NOT_INIT, count);
492 }
493 
494 static int get_inbound_buffer_frontier(struct qdio_q *q)
495 {
496 	unsigned char state = 0;
497 	int count;
498 
499 	q->timestamp = get_tod_clock_fast();
500 
501 	/*
502 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
503 	 * would return 0.
504 	 */
505 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
506 	if (!count)
507 		goto out;
508 
509 	/*
510 	 * No siga sync here, as a PCI or we after a thin interrupt
511 	 * already sync'ed the queues.
512 	 */
513 	count = get_buf_states(q, q->first_to_check, &state, count, 1, 0);
514 	if (!count)
515 		goto out;
516 
517 	switch (state) {
518 	case SLSB_P_INPUT_PRIMED:
519 		inbound_primed(q, count);
520 		q->first_to_check = add_buf(q->first_to_check, count);
521 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
522 			qperf_inc(q, inbound_queue_full);
523 		if (q->irq_ptr->perf_stat_enabled)
524 			account_sbals(q, count);
525 		break;
526 	case SLSB_P_INPUT_ERROR:
527 		process_buffer_error(q, count);
528 		q->first_to_check = add_buf(q->first_to_check, count);
529 		if (atomic_sub_return(count, &q->nr_buf_used) == 0)
530 			qperf_inc(q, inbound_queue_full);
531 		if (q->irq_ptr->perf_stat_enabled)
532 			account_sbals_error(q, count);
533 		break;
534 	case SLSB_CU_INPUT_EMPTY:
535 	case SLSB_P_INPUT_NOT_INIT:
536 	case SLSB_P_INPUT_ACK:
537 		if (q->irq_ptr->perf_stat_enabled)
538 			q->q_stats.nr_sbal_nop++;
539 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in nop:%1d %#02x",
540 			q->nr, q->first_to_check);
541 		break;
542 	default:
543 		WARN_ON_ONCE(1);
544 	}
545 out:
546 	return q->first_to_check;
547 }
548 
549 static int qdio_inbound_q_moved(struct qdio_q *q)
550 {
551 	int bufnr;
552 
553 	bufnr = get_inbound_buffer_frontier(q);
554 
555 	if (bufnr != q->last_move) {
556 		q->last_move = bufnr;
557 		if (!is_thinint_irq(q->irq_ptr) && MACHINE_IS_LPAR)
558 			q->u.in.timestamp = get_tod_clock();
559 		return 1;
560 	} else
561 		return 0;
562 }
563 
564 static inline int qdio_inbound_q_done(struct qdio_q *q)
565 {
566 	unsigned char state = 0;
567 
568 	if (!atomic_read(&q->nr_buf_used))
569 		return 1;
570 
571 	if (need_siga_sync(q))
572 		qdio_siga_sync_q(q);
573 	get_buf_state(q, q->first_to_check, &state, 0);
574 
575 	if (state == SLSB_P_INPUT_PRIMED || state == SLSB_P_INPUT_ERROR)
576 		/* more work coming */
577 		return 0;
578 
579 	if (is_thinint_irq(q->irq_ptr))
580 		return 1;
581 
582 	/* don't poll under z/VM */
583 	if (MACHINE_IS_VM)
584 		return 1;
585 
586 	/*
587 	 * At this point we know, that inbound first_to_check
588 	 * has (probably) not moved (see qdio_inbound_processing).
589 	 */
590 	if (get_tod_clock_fast() > q->u.in.timestamp + QDIO_INPUT_THRESHOLD) {
591 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "in done:%02x",
592 			      q->first_to_check);
593 		return 1;
594 	} else
595 		return 0;
596 }
597 
598 static inline int contains_aobs(struct qdio_q *q)
599 {
600 	return !q->is_input_q && q->u.out.use_cq;
601 }
602 
603 static inline void qdio_handle_aobs(struct qdio_q *q, int start, int count)
604 {
605 	unsigned char state = 0;
606 	int j, b = start;
607 
608 	if (!contains_aobs(q))
609 		return;
610 
611 	for (j = 0; j < count; ++j) {
612 		get_buf_state(q, b, &state, 0);
613 		if (state == SLSB_P_OUTPUT_PENDING) {
614 			struct qaob *aob = q->u.out.aobs[b];
615 			if (aob == NULL)
616 				continue;
617 
618 			q->u.out.sbal_state[b].flags |=
619 				QDIO_OUTBUF_STATE_FLAG_PENDING;
620 			q->u.out.aobs[b] = NULL;
621 		} else if (state == SLSB_P_OUTPUT_EMPTY) {
622 			q->u.out.sbal_state[b].aob = NULL;
623 		}
624 		b = next_buf(b);
625 	}
626 }
627 
628 static inline unsigned long qdio_aob_for_buffer(struct qdio_output_q *q,
629 					int bufnr)
630 {
631 	unsigned long phys_aob = 0;
632 
633 	if (!q->use_cq)
634 		goto out;
635 
636 	if (!q->aobs[bufnr]) {
637 		struct qaob *aob = qdio_allocate_aob();
638 		q->aobs[bufnr] = aob;
639 	}
640 	if (q->aobs[bufnr]) {
641 		q->sbal_state[bufnr].flags = QDIO_OUTBUF_STATE_FLAG_NONE;
642 		q->sbal_state[bufnr].aob = q->aobs[bufnr];
643 		q->aobs[bufnr]->user1 = (u64) q->sbal_state[bufnr].user;
644 		phys_aob = virt_to_phys(q->aobs[bufnr]);
645 		WARN_ON_ONCE(phys_aob & 0xFF);
646 	}
647 
648 out:
649 	return phys_aob;
650 }
651 
652 static void qdio_kick_handler(struct qdio_q *q)
653 {
654 	int start = q->first_to_kick;
655 	int end = q->first_to_check;
656 	int count;
657 
658 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
659 		return;
660 
661 	count = sub_buf(end, start);
662 
663 	if (q->is_input_q) {
664 		qperf_inc(q, inbound_handler);
665 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "kih s:%02x c:%02x", start, count);
666 	} else {
667 		qperf_inc(q, outbound_handler);
668 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "koh: s:%02x c:%02x",
669 			      start, count);
670 	}
671 
672 	qdio_handle_aobs(q, start, count);
673 
674 	q->handler(q->irq_ptr->cdev, q->qdio_error, q->nr, start, count,
675 		   q->irq_ptr->int_parm);
676 
677 	/* for the next time */
678 	q->first_to_kick = end;
679 	q->qdio_error = 0;
680 }
681 
682 static inline int qdio_tasklet_schedule(struct qdio_q *q)
683 {
684 	if (likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE)) {
685 		tasklet_schedule(&q->tasklet);
686 		return 0;
687 	}
688 	return -EPERM;
689 }
690 
691 static void __qdio_inbound_processing(struct qdio_q *q)
692 {
693 	qperf_inc(q, tasklet_inbound);
694 
695 	if (!qdio_inbound_q_moved(q))
696 		return;
697 
698 	qdio_kick_handler(q);
699 
700 	if (!qdio_inbound_q_done(q)) {
701 		/* means poll time is not yet over */
702 		qperf_inc(q, tasklet_inbound_resched);
703 		if (!qdio_tasklet_schedule(q))
704 			return;
705 	}
706 
707 	qdio_stop_polling(q);
708 	/*
709 	 * We need to check again to not lose initiative after
710 	 * resetting the ACK state.
711 	 */
712 	if (!qdio_inbound_q_done(q)) {
713 		qperf_inc(q, tasklet_inbound_resched2);
714 		qdio_tasklet_schedule(q);
715 	}
716 }
717 
718 void qdio_inbound_processing(unsigned long data)
719 {
720 	struct qdio_q *q = (struct qdio_q *)data;
721 	__qdio_inbound_processing(q);
722 }
723 
724 static int get_outbound_buffer_frontier(struct qdio_q *q)
725 {
726 	unsigned char state = 0;
727 	int count;
728 
729 	q->timestamp = get_tod_clock_fast();
730 
731 	if (need_siga_sync(q))
732 		if (((queue_type(q) != QDIO_IQDIO_QFMT) &&
733 		    !pci_out_supported(q)) ||
734 		    (queue_type(q) == QDIO_IQDIO_QFMT &&
735 		    multicast_outbound(q)))
736 			qdio_siga_sync_q(q);
737 
738 	/*
739 	 * Don't check 128 buffers, as otherwise qdio_inbound_q_moved
740 	 * would return 0.
741 	 */
742 	count = min(atomic_read(&q->nr_buf_used), QDIO_MAX_BUFFERS_MASK);
743 	if (!count)
744 		goto out;
745 
746 	count = get_buf_states(q, q->first_to_check, &state, count, 0,
747 			       q->u.out.use_cq);
748 	if (!count)
749 		goto out;
750 
751 	switch (state) {
752 	case SLSB_P_OUTPUT_EMPTY:
753 		/* the adapter got it */
754 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr,
755 			"out empty:%1d %02x", q->nr, count);
756 
757 		atomic_sub(count, &q->nr_buf_used);
758 		q->first_to_check = add_buf(q->first_to_check, count);
759 		if (q->irq_ptr->perf_stat_enabled)
760 			account_sbals(q, count);
761 
762 		break;
763 	case SLSB_P_OUTPUT_ERROR:
764 		process_buffer_error(q, count);
765 		q->first_to_check = add_buf(q->first_to_check, count);
766 		atomic_sub(count, &q->nr_buf_used);
767 		if (q->irq_ptr->perf_stat_enabled)
768 			account_sbals_error(q, count);
769 		break;
770 	case SLSB_CU_OUTPUT_PRIMED:
771 		/* the adapter has not fetched the output yet */
772 		if (q->irq_ptr->perf_stat_enabled)
773 			q->q_stats.nr_sbal_nop++;
774 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out primed:%1d",
775 			      q->nr);
776 		break;
777 	case SLSB_P_OUTPUT_NOT_INIT:
778 	case SLSB_P_OUTPUT_HALTED:
779 		break;
780 	default:
781 		WARN_ON_ONCE(1);
782 	}
783 
784 out:
785 	return q->first_to_check;
786 }
787 
788 /* all buffers processed? */
789 static inline int qdio_outbound_q_done(struct qdio_q *q)
790 {
791 	return atomic_read(&q->nr_buf_used) == 0;
792 }
793 
794 static inline int qdio_outbound_q_moved(struct qdio_q *q)
795 {
796 	int bufnr;
797 
798 	bufnr = get_outbound_buffer_frontier(q);
799 
800 	if (bufnr != q->last_move) {
801 		q->last_move = bufnr;
802 		DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "out moved:%1d", q->nr);
803 		return 1;
804 	} else
805 		return 0;
806 }
807 
808 static int qdio_kick_outbound_q(struct qdio_q *q, unsigned long aob)
809 {
810 	int retries = 0, cc;
811 	unsigned int busy_bit;
812 
813 	if (!need_siga_out(q))
814 		return 0;
815 
816 	DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w:%1d", q->nr);
817 retry:
818 	qperf_inc(q, siga_write);
819 
820 	cc = qdio_siga_output(q, &busy_bit, aob);
821 	switch (cc) {
822 	case 0:
823 		break;
824 	case 2:
825 		if (busy_bit) {
826 			while (++retries < QDIO_BUSY_BIT_RETRIES) {
827 				mdelay(QDIO_BUSY_BIT_RETRY_DELAY);
828 				goto retry;
829 			}
830 			DBF_ERROR("%4x cc2 BBC:%1d", SCH_NO(q), q->nr);
831 			cc = -EBUSY;
832 		} else {
833 			DBF_DEV_EVENT(DBF_INFO, q->irq_ptr, "siga-w cc2:%1d", q->nr);
834 			cc = -ENOBUFS;
835 		}
836 		break;
837 	case 1:
838 	case 3:
839 		DBF_ERROR("%4x SIGA-W:%1d", SCH_NO(q), cc);
840 		cc = -EIO;
841 		break;
842 	}
843 	if (retries) {
844 		DBF_ERROR("%4x cc2 BB2:%1d", SCH_NO(q), q->nr);
845 		DBF_ERROR("count:%u", retries);
846 	}
847 	return cc;
848 }
849 
850 static void __qdio_outbound_processing(struct qdio_q *q)
851 {
852 	qperf_inc(q, tasklet_outbound);
853 	WARN_ON_ONCE(atomic_read(&q->nr_buf_used) < 0);
854 
855 	if (qdio_outbound_q_moved(q))
856 		qdio_kick_handler(q);
857 
858 	if (queue_type(q) == QDIO_ZFCP_QFMT)
859 		if (!pci_out_supported(q) && !qdio_outbound_q_done(q))
860 			goto sched;
861 
862 	if (q->u.out.pci_out_enabled)
863 		return;
864 
865 	/*
866 	 * Now we know that queue type is either qeth without pci enabled
867 	 * or HiperSockets. Make sure buffer switch from PRIMED to EMPTY
868 	 * is noticed and outbound_handler is called after some time.
869 	 */
870 	if (qdio_outbound_q_done(q))
871 		del_timer_sync(&q->u.out.timer);
872 	else
873 		if (!timer_pending(&q->u.out.timer) &&
874 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
875 			mod_timer(&q->u.out.timer, jiffies + 10 * HZ);
876 	return;
877 
878 sched:
879 	qdio_tasklet_schedule(q);
880 }
881 
882 /* outbound tasklet */
883 void qdio_outbound_processing(unsigned long data)
884 {
885 	struct qdio_q *q = (struct qdio_q *)data;
886 	__qdio_outbound_processing(q);
887 }
888 
889 void qdio_outbound_timer(struct timer_list *t)
890 {
891 	struct qdio_q *q = from_timer(q, t, u.out.timer);
892 
893 	qdio_tasklet_schedule(q);
894 }
895 
896 static inline void qdio_check_outbound_after_thinint(struct qdio_q *q)
897 {
898 	struct qdio_q *out;
899 	int i;
900 
901 	if (!pci_out_supported(q))
902 		return;
903 
904 	for_each_output_queue(q->irq_ptr, out, i)
905 		if (!qdio_outbound_q_done(out))
906 			qdio_tasklet_schedule(out);
907 }
908 
909 static void __tiqdio_inbound_processing(struct qdio_q *q)
910 {
911 	qperf_inc(q, tasklet_inbound);
912 	if (need_siga_sync(q) && need_siga_sync_after_ai(q))
913 		qdio_sync_queues(q);
914 
915 	/*
916 	 * The interrupt could be caused by a PCI request. Check the
917 	 * PCI capable outbound queues.
918 	 */
919 	qdio_check_outbound_after_thinint(q);
920 
921 	if (!qdio_inbound_q_moved(q))
922 		return;
923 
924 	qdio_kick_handler(q);
925 
926 	if (!qdio_inbound_q_done(q)) {
927 		qperf_inc(q, tasklet_inbound_resched);
928 		if (!qdio_tasklet_schedule(q))
929 			return;
930 	}
931 
932 	qdio_stop_polling(q);
933 	/*
934 	 * We need to check again to not lose initiative after
935 	 * resetting the ACK state.
936 	 */
937 	if (!qdio_inbound_q_done(q)) {
938 		qperf_inc(q, tasklet_inbound_resched2);
939 		qdio_tasklet_schedule(q);
940 	}
941 }
942 
943 void tiqdio_inbound_processing(unsigned long data)
944 {
945 	struct qdio_q *q = (struct qdio_q *)data;
946 	__tiqdio_inbound_processing(q);
947 }
948 
949 static inline void qdio_set_state(struct qdio_irq *irq_ptr,
950 				  enum qdio_irq_states state)
951 {
952 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "newstate: %1d", state);
953 
954 	irq_ptr->state = state;
955 	mb();
956 }
957 
958 static void qdio_irq_check_sense(struct qdio_irq *irq_ptr, struct irb *irb)
959 {
960 	if (irb->esw.esw0.erw.cons) {
961 		DBF_ERROR("%4x sense:", irq_ptr->schid.sch_no);
962 		DBF_ERROR_HEX(irb, 64);
963 		DBF_ERROR_HEX(irb->ecw, 64);
964 	}
965 }
966 
967 /* PCI interrupt handler */
968 static void qdio_int_handler_pci(struct qdio_irq *irq_ptr)
969 {
970 	int i;
971 	struct qdio_q *q;
972 
973 	if (unlikely(irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
974 		return;
975 
976 	for_each_input_queue(irq_ptr, q, i) {
977 		if (q->u.in.queue_start_poll) {
978 			/* skip if polling is enabled or already in work */
979 			if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
980 				     &q->u.in.queue_irq_state)) {
981 				qperf_inc(q, int_discarded);
982 				continue;
983 			}
984 			q->u.in.queue_start_poll(q->irq_ptr->cdev, q->nr,
985 						 q->irq_ptr->int_parm);
986 		} else {
987 			tasklet_schedule(&q->tasklet);
988 		}
989 	}
990 
991 	if (!(irq_ptr->qib.ac & QIB_AC_OUTBOUND_PCI_SUPPORTED))
992 		return;
993 
994 	for_each_output_queue(irq_ptr, q, i) {
995 		if (qdio_outbound_q_done(q))
996 			continue;
997 		if (need_siga_sync(q) && need_siga_sync_out_after_pci(q))
998 			qdio_siga_sync_q(q);
999 		qdio_tasklet_schedule(q);
1000 	}
1001 }
1002 
1003 static void qdio_handle_activate_check(struct ccw_device *cdev,
1004 				unsigned long intparm, int cstat, int dstat)
1005 {
1006 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1007 	struct qdio_q *q;
1008 	int count;
1009 
1010 	DBF_ERROR("%4x ACT CHECK", irq_ptr->schid.sch_no);
1011 	DBF_ERROR("intp :%lx", intparm);
1012 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1013 
1014 	if (irq_ptr->nr_input_qs) {
1015 		q = irq_ptr->input_qs[0];
1016 	} else if (irq_ptr->nr_output_qs) {
1017 		q = irq_ptr->output_qs[0];
1018 	} else {
1019 		dump_stack();
1020 		goto no_handler;
1021 	}
1022 
1023 	count = sub_buf(q->first_to_check, q->first_to_kick);
1024 	q->handler(q->irq_ptr->cdev, QDIO_ERROR_ACTIVATE,
1025 		   q->nr, q->first_to_kick, count, irq_ptr->int_parm);
1026 no_handler:
1027 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1028 	/*
1029 	 * In case of z/VM LGR (Live Guest Migration) QDIO recovery will happen.
1030 	 * Therefore we call the LGR detection function here.
1031 	 */
1032 	lgr_info_log();
1033 }
1034 
1035 static void qdio_establish_handle_irq(struct ccw_device *cdev, int cstat,
1036 				      int dstat)
1037 {
1038 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1039 
1040 	DBF_DEV_EVENT(DBF_INFO, irq_ptr, "qest irq");
1041 
1042 	if (cstat)
1043 		goto error;
1044 	if (dstat & ~(DEV_STAT_DEV_END | DEV_STAT_CHN_END))
1045 		goto error;
1046 	if (!(dstat & DEV_STAT_DEV_END))
1047 		goto error;
1048 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ESTABLISHED);
1049 	return;
1050 
1051 error:
1052 	DBF_ERROR("%4x EQ:error", irq_ptr->schid.sch_no);
1053 	DBF_ERROR("ds: %2x cs:%2x", dstat, cstat);
1054 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1055 }
1056 
1057 /* qdio interrupt handler */
1058 void qdio_int_handler(struct ccw_device *cdev, unsigned long intparm,
1059 		      struct irb *irb)
1060 {
1061 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1062 	struct subchannel_id schid;
1063 	int cstat, dstat;
1064 
1065 	if (!intparm || !irq_ptr) {
1066 		ccw_device_get_schid(cdev, &schid);
1067 		DBF_ERROR("qint:%4x", schid.sch_no);
1068 		return;
1069 	}
1070 
1071 	if (irq_ptr->perf_stat_enabled)
1072 		irq_ptr->perf_stat.qdio_int++;
1073 
1074 	if (IS_ERR(irb)) {
1075 		DBF_ERROR("%4x IO error", irq_ptr->schid.sch_no);
1076 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ERR);
1077 		wake_up(&cdev->private->wait_q);
1078 		return;
1079 	}
1080 	qdio_irq_check_sense(irq_ptr, irb);
1081 	cstat = irb->scsw.cmd.cstat;
1082 	dstat = irb->scsw.cmd.dstat;
1083 
1084 	switch (irq_ptr->state) {
1085 	case QDIO_IRQ_STATE_INACTIVE:
1086 		qdio_establish_handle_irq(cdev, cstat, dstat);
1087 		break;
1088 	case QDIO_IRQ_STATE_CLEANUP:
1089 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1090 		break;
1091 	case QDIO_IRQ_STATE_ESTABLISHED:
1092 	case QDIO_IRQ_STATE_ACTIVE:
1093 		if (cstat & SCHN_STAT_PCI) {
1094 			qdio_int_handler_pci(irq_ptr);
1095 			return;
1096 		}
1097 		if (cstat || dstat)
1098 			qdio_handle_activate_check(cdev, intparm, cstat,
1099 						   dstat);
1100 		break;
1101 	case QDIO_IRQ_STATE_STOPPED:
1102 		break;
1103 	default:
1104 		WARN_ON_ONCE(1);
1105 	}
1106 	wake_up(&cdev->private->wait_q);
1107 }
1108 
1109 /**
1110  * qdio_get_ssqd_desc - get qdio subchannel description
1111  * @cdev: ccw device to get description for
1112  * @data: where to store the ssqd
1113  *
1114  * Returns 0 or an error code. The results of the chsc are stored in the
1115  * specified structure.
1116  */
1117 int qdio_get_ssqd_desc(struct ccw_device *cdev,
1118 		       struct qdio_ssqd_desc *data)
1119 {
1120 	struct subchannel_id schid;
1121 
1122 	if (!cdev || !cdev->private)
1123 		return -EINVAL;
1124 
1125 	ccw_device_get_schid(cdev, &schid);
1126 	DBF_EVENT("get ssqd:%4x", schid.sch_no);
1127 	return qdio_setup_get_ssqd(NULL, &schid, data);
1128 }
1129 EXPORT_SYMBOL_GPL(qdio_get_ssqd_desc);
1130 
1131 static void qdio_shutdown_queues(struct ccw_device *cdev)
1132 {
1133 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1134 	struct qdio_q *q;
1135 	int i;
1136 
1137 	for_each_input_queue(irq_ptr, q, i)
1138 		tasklet_kill(&q->tasklet);
1139 
1140 	for_each_output_queue(irq_ptr, q, i) {
1141 		del_timer_sync(&q->u.out.timer);
1142 		tasklet_kill(&q->tasklet);
1143 	}
1144 }
1145 
1146 /**
1147  * qdio_shutdown - shut down a qdio subchannel
1148  * @cdev: associated ccw device
1149  * @how: use halt or clear to shutdown
1150  */
1151 int qdio_shutdown(struct ccw_device *cdev, int how)
1152 {
1153 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1154 	struct subchannel_id schid;
1155 	int rc;
1156 
1157 	if (!irq_ptr)
1158 		return -ENODEV;
1159 
1160 	WARN_ON_ONCE(irqs_disabled());
1161 	ccw_device_get_schid(cdev, &schid);
1162 	DBF_EVENT("qshutdown:%4x", schid.sch_no);
1163 
1164 	mutex_lock(&irq_ptr->setup_mutex);
1165 	/*
1166 	 * Subchannel was already shot down. We cannot prevent being called
1167 	 * twice since cio may trigger a shutdown asynchronously.
1168 	 */
1169 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1170 		mutex_unlock(&irq_ptr->setup_mutex);
1171 		return 0;
1172 	}
1173 
1174 	/*
1175 	 * Indicate that the device is going down. Scheduling the queue
1176 	 * tasklets is forbidden from here on.
1177 	 */
1178 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_STOPPED);
1179 
1180 	tiqdio_remove_input_queues(irq_ptr);
1181 	qdio_shutdown_queues(cdev);
1182 	qdio_shutdown_debug_entries(irq_ptr);
1183 
1184 	/* cleanup subchannel */
1185 	spin_lock_irq(get_ccwdev_lock(cdev));
1186 
1187 	if (how & QDIO_FLAG_CLEANUP_USING_CLEAR)
1188 		rc = ccw_device_clear(cdev, QDIO_DOING_CLEANUP);
1189 	else
1190 		/* default behaviour is halt */
1191 		rc = ccw_device_halt(cdev, QDIO_DOING_CLEANUP);
1192 	if (rc) {
1193 		DBF_ERROR("%4x SHUTD ERR", irq_ptr->schid.sch_no);
1194 		DBF_ERROR("rc:%4d", rc);
1195 		goto no_cleanup;
1196 	}
1197 
1198 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_CLEANUP);
1199 	spin_unlock_irq(get_ccwdev_lock(cdev));
1200 	wait_event_interruptible_timeout(cdev->private->wait_q,
1201 		irq_ptr->state == QDIO_IRQ_STATE_INACTIVE ||
1202 		irq_ptr->state == QDIO_IRQ_STATE_ERR,
1203 		10 * HZ);
1204 	spin_lock_irq(get_ccwdev_lock(cdev));
1205 
1206 no_cleanup:
1207 	qdio_shutdown_thinint(irq_ptr);
1208 
1209 	/* restore interrupt handler */
1210 	if ((void *)cdev->handler == (void *)qdio_int_handler) {
1211 		cdev->handler = irq_ptr->orig_handler;
1212 		cdev->private->intparm = 0;
1213 	}
1214 	spin_unlock_irq(get_ccwdev_lock(cdev));
1215 
1216 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1217 	mutex_unlock(&irq_ptr->setup_mutex);
1218 	if (rc)
1219 		return rc;
1220 	return 0;
1221 }
1222 EXPORT_SYMBOL_GPL(qdio_shutdown);
1223 
1224 /**
1225  * qdio_free - free data structures for a qdio subchannel
1226  * @cdev: associated ccw device
1227  */
1228 int qdio_free(struct ccw_device *cdev)
1229 {
1230 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1231 	struct subchannel_id schid;
1232 
1233 	if (!irq_ptr)
1234 		return -ENODEV;
1235 
1236 	ccw_device_get_schid(cdev, &schid);
1237 	DBF_EVENT("qfree:%4x", schid.sch_no);
1238 	DBF_DEV_EVENT(DBF_ERR, irq_ptr, "dbf abandoned");
1239 	mutex_lock(&irq_ptr->setup_mutex);
1240 
1241 	irq_ptr->debug_area = NULL;
1242 	cdev->private->qdio_data = NULL;
1243 	mutex_unlock(&irq_ptr->setup_mutex);
1244 
1245 	qdio_release_memory(irq_ptr);
1246 	return 0;
1247 }
1248 EXPORT_SYMBOL_GPL(qdio_free);
1249 
1250 /**
1251  * qdio_allocate - allocate qdio queues and associated data
1252  * @init_data: initialization data
1253  */
1254 int qdio_allocate(struct qdio_initialize *init_data)
1255 {
1256 	struct subchannel_id schid;
1257 	struct qdio_irq *irq_ptr;
1258 
1259 	ccw_device_get_schid(init_data->cdev, &schid);
1260 	DBF_EVENT("qallocate:%4x", schid.sch_no);
1261 
1262 	if ((init_data->no_input_qs && !init_data->input_handler) ||
1263 	    (init_data->no_output_qs && !init_data->output_handler))
1264 		return -EINVAL;
1265 
1266 	if ((init_data->no_input_qs > QDIO_MAX_QUEUES_PER_IRQ) ||
1267 	    (init_data->no_output_qs > QDIO_MAX_QUEUES_PER_IRQ))
1268 		return -EINVAL;
1269 
1270 	if ((!init_data->input_sbal_addr_array) ||
1271 	    (!init_data->output_sbal_addr_array))
1272 		return -EINVAL;
1273 
1274 	/* irq_ptr must be in GFP_DMA since it contains ccw1.cda */
1275 	irq_ptr = (void *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1276 	if (!irq_ptr)
1277 		goto out_err;
1278 
1279 	mutex_init(&irq_ptr->setup_mutex);
1280 	if (qdio_allocate_dbf(init_data, irq_ptr))
1281 		goto out_rel;
1282 
1283 	/*
1284 	 * Allocate a page for the chsc calls in qdio_establish.
1285 	 * Must be pre-allocated since a zfcp recovery will call
1286 	 * qdio_establish. In case of low memory and swap on a zfcp disk
1287 	 * we may not be able to allocate memory otherwise.
1288 	 */
1289 	irq_ptr->chsc_page = get_zeroed_page(GFP_KERNEL);
1290 	if (!irq_ptr->chsc_page)
1291 		goto out_rel;
1292 
1293 	/* qdr is used in ccw1.cda which is u32 */
1294 	irq_ptr->qdr = (struct qdr *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1295 	if (!irq_ptr->qdr)
1296 		goto out_rel;
1297 
1298 	if (qdio_allocate_qs(irq_ptr, init_data->no_input_qs,
1299 			     init_data->no_output_qs))
1300 		goto out_rel;
1301 
1302 	init_data->cdev->private->qdio_data = irq_ptr;
1303 	qdio_set_state(irq_ptr, QDIO_IRQ_STATE_INACTIVE);
1304 	return 0;
1305 out_rel:
1306 	qdio_release_memory(irq_ptr);
1307 out_err:
1308 	return -ENOMEM;
1309 }
1310 EXPORT_SYMBOL_GPL(qdio_allocate);
1311 
1312 static void qdio_detect_hsicq(struct qdio_irq *irq_ptr)
1313 {
1314 	struct qdio_q *q = irq_ptr->input_qs[0];
1315 	int i, use_cq = 0;
1316 
1317 	if (irq_ptr->nr_input_qs > 1 && queue_type(q) == QDIO_IQDIO_QFMT)
1318 		use_cq = 1;
1319 
1320 	for_each_output_queue(irq_ptr, q, i) {
1321 		if (use_cq) {
1322 			if (qdio_enable_async_operation(&q->u.out) < 0) {
1323 				use_cq = 0;
1324 				continue;
1325 			}
1326 		} else
1327 			qdio_disable_async_operation(&q->u.out);
1328 	}
1329 	DBF_EVENT("use_cq:%d", use_cq);
1330 }
1331 
1332 /**
1333  * qdio_establish - establish queues on a qdio subchannel
1334  * @init_data: initialization data
1335  */
1336 int qdio_establish(struct qdio_initialize *init_data)
1337 {
1338 	struct ccw_device *cdev = init_data->cdev;
1339 	struct subchannel_id schid;
1340 	struct qdio_irq *irq_ptr;
1341 	int rc;
1342 
1343 	ccw_device_get_schid(cdev, &schid);
1344 	DBF_EVENT("qestablish:%4x", schid.sch_no);
1345 
1346 	irq_ptr = cdev->private->qdio_data;
1347 	if (!irq_ptr)
1348 		return -ENODEV;
1349 
1350 	mutex_lock(&irq_ptr->setup_mutex);
1351 	qdio_setup_irq(init_data);
1352 
1353 	rc = qdio_establish_thinint(irq_ptr);
1354 	if (rc) {
1355 		mutex_unlock(&irq_ptr->setup_mutex);
1356 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1357 		return rc;
1358 	}
1359 
1360 	/* establish q */
1361 	irq_ptr->ccw.cmd_code = irq_ptr->equeue.cmd;
1362 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1363 	irq_ptr->ccw.count = irq_ptr->equeue.count;
1364 	irq_ptr->ccw.cda = (u32)((addr_t)irq_ptr->qdr);
1365 
1366 	spin_lock_irq(get_ccwdev_lock(cdev));
1367 	ccw_device_set_options_mask(cdev, 0);
1368 
1369 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ESTABLISH, 0, 0);
1370 	spin_unlock_irq(get_ccwdev_lock(cdev));
1371 	if (rc) {
1372 		DBF_ERROR("%4x est IO ERR", irq_ptr->schid.sch_no);
1373 		DBF_ERROR("rc:%4x", rc);
1374 		mutex_unlock(&irq_ptr->setup_mutex);
1375 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1376 		return rc;
1377 	}
1378 
1379 	wait_event_interruptible_timeout(cdev->private->wait_q,
1380 		irq_ptr->state == QDIO_IRQ_STATE_ESTABLISHED ||
1381 		irq_ptr->state == QDIO_IRQ_STATE_ERR, HZ);
1382 
1383 	if (irq_ptr->state != QDIO_IRQ_STATE_ESTABLISHED) {
1384 		mutex_unlock(&irq_ptr->setup_mutex);
1385 		qdio_shutdown(cdev, QDIO_FLAG_CLEANUP_USING_CLEAR);
1386 		return -EIO;
1387 	}
1388 
1389 	qdio_setup_ssqd_info(irq_ptr);
1390 
1391 	qdio_detect_hsicq(irq_ptr);
1392 
1393 	/* qebsm is now setup if available, initialize buffer states */
1394 	qdio_init_buf_states(irq_ptr);
1395 
1396 	mutex_unlock(&irq_ptr->setup_mutex);
1397 	qdio_print_subchannel_info(irq_ptr, cdev);
1398 	qdio_setup_debug_entries(irq_ptr, cdev);
1399 	return 0;
1400 }
1401 EXPORT_SYMBOL_GPL(qdio_establish);
1402 
1403 /**
1404  * qdio_activate - activate queues on a qdio subchannel
1405  * @cdev: associated cdev
1406  */
1407 int qdio_activate(struct ccw_device *cdev)
1408 {
1409 	struct subchannel_id schid;
1410 	struct qdio_irq *irq_ptr;
1411 	int rc;
1412 
1413 	ccw_device_get_schid(cdev, &schid);
1414 	DBF_EVENT("qactivate:%4x", schid.sch_no);
1415 
1416 	irq_ptr = cdev->private->qdio_data;
1417 	if (!irq_ptr)
1418 		return -ENODEV;
1419 
1420 	mutex_lock(&irq_ptr->setup_mutex);
1421 	if (irq_ptr->state == QDIO_IRQ_STATE_INACTIVE) {
1422 		rc = -EBUSY;
1423 		goto out;
1424 	}
1425 
1426 	irq_ptr->ccw.cmd_code = irq_ptr->aqueue.cmd;
1427 	irq_ptr->ccw.flags = CCW_FLAG_SLI;
1428 	irq_ptr->ccw.count = irq_ptr->aqueue.count;
1429 	irq_ptr->ccw.cda = 0;
1430 
1431 	spin_lock_irq(get_ccwdev_lock(cdev));
1432 	ccw_device_set_options(cdev, CCWDEV_REPORT_ALL);
1433 
1434 	rc = ccw_device_start(cdev, &irq_ptr->ccw, QDIO_DOING_ACTIVATE,
1435 			      0, DOIO_DENY_PREFETCH);
1436 	spin_unlock_irq(get_ccwdev_lock(cdev));
1437 	if (rc) {
1438 		DBF_ERROR("%4x act IO ERR", irq_ptr->schid.sch_no);
1439 		DBF_ERROR("rc:%4x", rc);
1440 		goto out;
1441 	}
1442 
1443 	if (is_thinint_irq(irq_ptr))
1444 		tiqdio_add_input_queues(irq_ptr);
1445 
1446 	/* wait for subchannel to become active */
1447 	msleep(5);
1448 
1449 	switch (irq_ptr->state) {
1450 	case QDIO_IRQ_STATE_STOPPED:
1451 	case QDIO_IRQ_STATE_ERR:
1452 		rc = -EIO;
1453 		break;
1454 	default:
1455 		qdio_set_state(irq_ptr, QDIO_IRQ_STATE_ACTIVE);
1456 		rc = 0;
1457 	}
1458 out:
1459 	mutex_unlock(&irq_ptr->setup_mutex);
1460 	return rc;
1461 }
1462 EXPORT_SYMBOL_GPL(qdio_activate);
1463 
1464 static inline int buf_in_between(int bufnr, int start, int count)
1465 {
1466 	int end = add_buf(start, count);
1467 
1468 	if (end > start) {
1469 		if (bufnr >= start && bufnr < end)
1470 			return 1;
1471 		else
1472 			return 0;
1473 	}
1474 
1475 	/* wrap-around case */
1476 	if ((bufnr >= start && bufnr <= QDIO_MAX_BUFFERS_PER_Q) ||
1477 	    (bufnr < end))
1478 		return 1;
1479 	else
1480 		return 0;
1481 }
1482 
1483 /**
1484  * handle_inbound - reset processed input buffers
1485  * @q: queue containing the buffers
1486  * @callflags: flags
1487  * @bufnr: first buffer to process
1488  * @count: how many buffers are emptied
1489  */
1490 static int handle_inbound(struct qdio_q *q, unsigned int callflags,
1491 			  int bufnr, int count)
1492 {
1493 	int diff;
1494 
1495 	qperf_inc(q, inbound_call);
1496 
1497 	if (!q->u.in.polling)
1498 		goto set;
1499 
1500 	/* protect against stop polling setting an ACK for an emptied slsb */
1501 	if (count == QDIO_MAX_BUFFERS_PER_Q) {
1502 		/* overwriting everything, just delete polling status */
1503 		q->u.in.polling = 0;
1504 		q->u.in.ack_count = 0;
1505 		goto set;
1506 	} else if (buf_in_between(q->u.in.ack_start, bufnr, count)) {
1507 		if (is_qebsm(q)) {
1508 			/* partial overwrite, just update ack_start */
1509 			diff = add_buf(bufnr, count);
1510 			diff = sub_buf(diff, q->u.in.ack_start);
1511 			q->u.in.ack_count -= diff;
1512 			if (q->u.in.ack_count <= 0) {
1513 				q->u.in.polling = 0;
1514 				q->u.in.ack_count = 0;
1515 				goto set;
1516 			}
1517 			q->u.in.ack_start = add_buf(q->u.in.ack_start, diff);
1518 		}
1519 		else
1520 			/* the only ACK will be deleted, so stop polling */
1521 			q->u.in.polling = 0;
1522 	}
1523 
1524 set:
1525 	count = set_buf_states(q, bufnr, SLSB_CU_INPUT_EMPTY, count);
1526 	atomic_add(count, &q->nr_buf_used);
1527 
1528 	if (need_siga_in(q))
1529 		return qdio_siga_input(q);
1530 
1531 	return 0;
1532 }
1533 
1534 /**
1535  * handle_outbound - process filled outbound buffers
1536  * @q: queue containing the buffers
1537  * @callflags: flags
1538  * @bufnr: first buffer to process
1539  * @count: how many buffers are filled
1540  */
1541 static int handle_outbound(struct qdio_q *q, unsigned int callflags,
1542 			   int bufnr, int count)
1543 {
1544 	unsigned char state = 0;
1545 	int used, rc = 0;
1546 
1547 	qperf_inc(q, outbound_call);
1548 
1549 	count = set_buf_states(q, bufnr, SLSB_CU_OUTPUT_PRIMED, count);
1550 	used = atomic_add_return(count, &q->nr_buf_used);
1551 
1552 	if (used == QDIO_MAX_BUFFERS_PER_Q)
1553 		qperf_inc(q, outbound_queue_full);
1554 
1555 	if (callflags & QDIO_FLAG_PCI_OUT) {
1556 		q->u.out.pci_out_enabled = 1;
1557 		qperf_inc(q, pci_request_int);
1558 	} else
1559 		q->u.out.pci_out_enabled = 0;
1560 
1561 	if (queue_type(q) == QDIO_IQDIO_QFMT) {
1562 		unsigned long phys_aob = 0;
1563 
1564 		/* One SIGA-W per buffer required for unicast HSI */
1565 		WARN_ON_ONCE(count > 1 && !multicast_outbound(q));
1566 
1567 		phys_aob = qdio_aob_for_buffer(&q->u.out, bufnr);
1568 
1569 		rc = qdio_kick_outbound_q(q, phys_aob);
1570 	} else if (need_siga_sync(q)) {
1571 		rc = qdio_siga_sync_q(q);
1572 	} else {
1573 		/* try to fast requeue buffers */
1574 		get_buf_state(q, prev_buf(bufnr), &state, 0);
1575 		if (state != SLSB_CU_OUTPUT_PRIMED)
1576 			rc = qdio_kick_outbound_q(q, 0);
1577 		else
1578 			qperf_inc(q, fast_requeue);
1579 	}
1580 
1581 	/* in case of SIGA errors we must process the error immediately */
1582 	if (used >= q->u.out.scan_threshold || rc)
1583 		qdio_tasklet_schedule(q);
1584 	else
1585 		/* free the SBALs in case of no further traffic */
1586 		if (!timer_pending(&q->u.out.timer) &&
1587 		    likely(q->irq_ptr->state == QDIO_IRQ_STATE_ACTIVE))
1588 			mod_timer(&q->u.out.timer, jiffies + HZ);
1589 	return rc;
1590 }
1591 
1592 /**
1593  * do_QDIO - process input or output buffers
1594  * @cdev: associated ccw_device for the qdio subchannel
1595  * @callflags: input or output and special flags from the program
1596  * @q_nr: queue number
1597  * @bufnr: buffer number
1598  * @count: how many buffers to process
1599  */
1600 int do_QDIO(struct ccw_device *cdev, unsigned int callflags,
1601 	    int q_nr, unsigned int bufnr, unsigned int count)
1602 {
1603 	struct qdio_irq *irq_ptr;
1604 
1605 	if (bufnr >= QDIO_MAX_BUFFERS_PER_Q || count > QDIO_MAX_BUFFERS_PER_Q)
1606 		return -EINVAL;
1607 
1608 	irq_ptr = cdev->private->qdio_data;
1609 	if (!irq_ptr)
1610 		return -ENODEV;
1611 
1612 	DBF_DEV_EVENT(DBF_INFO, irq_ptr,
1613 		      "do%02x b:%02x c:%02x", callflags, bufnr, count);
1614 
1615 	if (irq_ptr->state != QDIO_IRQ_STATE_ACTIVE)
1616 		return -EIO;
1617 	if (!count)
1618 		return 0;
1619 	if (callflags & QDIO_FLAG_SYNC_INPUT)
1620 		return handle_inbound(irq_ptr->input_qs[q_nr],
1621 				      callflags, bufnr, count);
1622 	else if (callflags & QDIO_FLAG_SYNC_OUTPUT)
1623 		return handle_outbound(irq_ptr->output_qs[q_nr],
1624 				       callflags, bufnr, count);
1625 	return -EINVAL;
1626 }
1627 EXPORT_SYMBOL_GPL(do_QDIO);
1628 
1629 /**
1630  * qdio_start_irq - process input buffers
1631  * @cdev: associated ccw_device for the qdio subchannel
1632  * @nr: input queue number
1633  *
1634  * Return codes
1635  *   0 - success
1636  *   1 - irqs not started since new data is available
1637  */
1638 int qdio_start_irq(struct ccw_device *cdev, int nr)
1639 {
1640 	struct qdio_q *q;
1641 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1642 
1643 	if (!irq_ptr)
1644 		return -ENODEV;
1645 	q = irq_ptr->input_qs[nr];
1646 
1647 	clear_nonshared_ind(irq_ptr);
1648 	qdio_stop_polling(q);
1649 	clear_bit(QDIO_QUEUE_IRQS_DISABLED, &q->u.in.queue_irq_state);
1650 
1651 	/*
1652 	 * We need to check again to not lose initiative after
1653 	 * resetting the ACK state.
1654 	 */
1655 	if (test_nonshared_ind(irq_ptr))
1656 		goto rescan;
1657 	if (!qdio_inbound_q_done(q))
1658 		goto rescan;
1659 	return 0;
1660 
1661 rescan:
1662 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1663 			     &q->u.in.queue_irq_state))
1664 		return 0;
1665 	else
1666 		return 1;
1667 
1668 }
1669 EXPORT_SYMBOL(qdio_start_irq);
1670 
1671 /**
1672  * qdio_get_next_buffers - process input buffers
1673  * @cdev: associated ccw_device for the qdio subchannel
1674  * @nr: input queue number
1675  * @bufnr: first filled buffer number
1676  * @error: buffers are in error state
1677  *
1678  * Return codes
1679  *   < 0 - error
1680  *   = 0 - no new buffers found
1681  *   > 0 - number of processed buffers
1682  */
1683 int qdio_get_next_buffers(struct ccw_device *cdev, int nr, int *bufnr,
1684 			  int *error)
1685 {
1686 	struct qdio_q *q;
1687 	int start, end;
1688 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1689 
1690 	if (!irq_ptr)
1691 		return -ENODEV;
1692 	q = irq_ptr->input_qs[nr];
1693 
1694 	/*
1695 	 * Cannot rely on automatic sync after interrupt since queues may
1696 	 * also be examined without interrupt.
1697 	 */
1698 	if (need_siga_sync(q))
1699 		qdio_sync_queues(q);
1700 
1701 	/* check the PCI capable outbound queues. */
1702 	qdio_check_outbound_after_thinint(q);
1703 
1704 	if (!qdio_inbound_q_moved(q))
1705 		return 0;
1706 
1707 	/* Note: upper-layer MUST stop processing immediately here ... */
1708 	if (unlikely(q->irq_ptr->state != QDIO_IRQ_STATE_ACTIVE))
1709 		return -EIO;
1710 
1711 	start = q->first_to_kick;
1712 	end = q->first_to_check;
1713 	*bufnr = start;
1714 	*error = q->qdio_error;
1715 
1716 	/* for the next time */
1717 	q->first_to_kick = end;
1718 	q->qdio_error = 0;
1719 	return sub_buf(end, start);
1720 }
1721 EXPORT_SYMBOL(qdio_get_next_buffers);
1722 
1723 /**
1724  * qdio_stop_irq - disable interrupt processing for the device
1725  * @cdev: associated ccw_device for the qdio subchannel
1726  * @nr: input queue number
1727  *
1728  * Return codes
1729  *   0 - interrupts were already disabled
1730  *   1 - interrupts successfully disabled
1731  */
1732 int qdio_stop_irq(struct ccw_device *cdev, int nr)
1733 {
1734 	struct qdio_q *q;
1735 	struct qdio_irq *irq_ptr = cdev->private->qdio_data;
1736 
1737 	if (!irq_ptr)
1738 		return -ENODEV;
1739 	q = irq_ptr->input_qs[nr];
1740 
1741 	if (test_and_set_bit(QDIO_QUEUE_IRQS_DISABLED,
1742 			     &q->u.in.queue_irq_state))
1743 		return 0;
1744 	else
1745 		return 1;
1746 }
1747 EXPORT_SYMBOL(qdio_stop_irq);
1748 
1749 /**
1750  * qdio_pnso_brinfo() - perform network subchannel op #0 - bridge info.
1751  * @schid:		Subchannel ID.
1752  * @cnc:		Boolean Change-Notification Control
1753  * @response:		Response code will be stored at this address
1754  * @cb: 		Callback function will be executed for each element
1755  *			of the address list
1756  * @priv:		Pointer to pass to the callback function.
1757  *
1758  * Performs "Store-network-bridging-information list" operation and calls
1759  * the callback function for every entry in the list. If "change-
1760  * notification-control" is set, further changes in the address list
1761  * will be reported via the IPA command.
1762  */
1763 int qdio_pnso_brinfo(struct subchannel_id schid,
1764 		int cnc, u16 *response,
1765 		void (*cb)(void *priv, enum qdio_brinfo_entry_type type,
1766 				void *entry),
1767 		void *priv)
1768 {
1769 	struct chsc_pnso_area *rr;
1770 	int rc;
1771 	u32 prev_instance = 0;
1772 	int isfirstblock = 1;
1773 	int i, size, elems;
1774 
1775 	rr = (struct chsc_pnso_area *)get_zeroed_page(GFP_KERNEL);
1776 	if (rr == NULL)
1777 		return -ENOMEM;
1778 	do {
1779 		/* on the first iteration, naihdr.resume_token will be zero */
1780 		rc = chsc_pnso_brinfo(schid, rr, rr->naihdr.resume_token, cnc);
1781 		if (rc != 0 && rc != -EBUSY)
1782 			goto out;
1783 		if (rr->response.code != 1) {
1784 			rc = -EIO;
1785 			continue;
1786 		} else
1787 			rc = 0;
1788 
1789 		if (cb == NULL)
1790 			continue;
1791 
1792 		size = rr->naihdr.naids;
1793 		elems = (rr->response.length -
1794 				sizeof(struct chsc_header) -
1795 				sizeof(struct chsc_brinfo_naihdr)) /
1796 				size;
1797 
1798 		if (!isfirstblock && (rr->naihdr.instance != prev_instance)) {
1799 			/* Inform the caller that they need to scrap */
1800 			/* the data that was already reported via cb */
1801 				rc = -EAGAIN;
1802 				break;
1803 		}
1804 		isfirstblock = 0;
1805 		prev_instance = rr->naihdr.instance;
1806 		for (i = 0; i < elems; i++)
1807 			switch (size) {
1808 			case sizeof(struct qdio_brinfo_entry_l3_ipv6):
1809 				(*cb)(priv, l3_ipv6_addr,
1810 						&rr->entries.l3_ipv6[i]);
1811 				break;
1812 			case sizeof(struct qdio_brinfo_entry_l3_ipv4):
1813 				(*cb)(priv, l3_ipv4_addr,
1814 						&rr->entries.l3_ipv4[i]);
1815 				break;
1816 			case sizeof(struct qdio_brinfo_entry_l2):
1817 				(*cb)(priv, l2_addr_lnid,
1818 						&rr->entries.l2[i]);
1819 				break;
1820 			default:
1821 				WARN_ON_ONCE(1);
1822 				rc = -EIO;
1823 				goto out;
1824 			}
1825 	} while (rr->response.code == 0x0107 ||  /* channel busy */
1826 		  (rr->response.code == 1 && /* list stored */
1827 		   /* resume token is non-zero => list incomplete */
1828 		   (rr->naihdr.resume_token.t1 || rr->naihdr.resume_token.t2)));
1829 	(*response) = rr->response.code;
1830 
1831 out:
1832 	free_page((unsigned long)rr);
1833 	return rc;
1834 }
1835 EXPORT_SYMBOL_GPL(qdio_pnso_brinfo);
1836 
1837 static int __init init_QDIO(void)
1838 {
1839 	int rc;
1840 
1841 	rc = qdio_debug_init();
1842 	if (rc)
1843 		return rc;
1844 	rc = qdio_setup_init();
1845 	if (rc)
1846 		goto out_debug;
1847 	rc = tiqdio_allocate_memory();
1848 	if (rc)
1849 		goto out_cache;
1850 	rc = tiqdio_register_thinints();
1851 	if (rc)
1852 		goto out_ti;
1853 	return 0;
1854 
1855 out_ti:
1856 	tiqdio_free_memory();
1857 out_cache:
1858 	qdio_setup_exit();
1859 out_debug:
1860 	qdio_debug_exit();
1861 	return rc;
1862 }
1863 
1864 static void __exit exit_QDIO(void)
1865 {
1866 	tiqdio_unregister_thinints();
1867 	tiqdio_free_memory();
1868 	qdio_setup_exit();
1869 	qdio_debug_exit();
1870 }
1871 
1872 module_init(init_QDIO);
1873 module_exit(exit_QDIO);
1874