xref: /linux/drivers/scsi/cxlflash/main.c (revision e2be04c7f9958dde770eeb8b30e829ca969b37bb)
1 /*
2  * CXL Flash Device Driver
3  *
4  * Written by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>, IBM Corporation
5  *             Matthew R. Ochs <mrochs@linux.vnet.ibm.com>, IBM Corporation
6  *
7  * Copyright (C) 2015 IBM Corporation
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * as published by the Free Software Foundation; either version
12  * 2 of the License, or (at your option) any later version.
13  */
14 
15 #include <linux/delay.h>
16 #include <linux/list.h>
17 #include <linux/module.h>
18 #include <linux/pci.h>
19 
20 #include <asm/unaligned.h>
21 
22 #include <misc/cxl.h>
23 
24 #include <scsi/scsi_cmnd.h>
25 #include <scsi/scsi_host.h>
26 #include <uapi/scsi/cxlflash_ioctl.h>
27 
28 #include "main.h"
29 #include "sislite.h"
30 #include "common.h"
31 
32 MODULE_DESCRIPTION(CXLFLASH_ADAPTER_NAME);
33 MODULE_AUTHOR("Manoj N. Kumar <manoj@linux.vnet.ibm.com>");
34 MODULE_AUTHOR("Matthew R. Ochs <mrochs@linux.vnet.ibm.com>");
35 MODULE_LICENSE("GPL");
36 
37 static struct class *cxlflash_class;
38 static u32 cxlflash_major;
39 static DECLARE_BITMAP(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
40 
41 /**
42  * process_cmd_err() - command error handler
43  * @cmd:	AFU command that experienced the error.
44  * @scp:	SCSI command associated with the AFU command in error.
45  *
46  * Translates error bits from AFU command to SCSI command results.
47  */
48 static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
49 {
50 	struct afu *afu = cmd->parent;
51 	struct cxlflash_cfg *cfg = afu->parent;
52 	struct device *dev = &cfg->dev->dev;
53 	struct sisl_ioarcb *ioarcb;
54 	struct sisl_ioasa *ioasa;
55 	u32 resid;
56 
57 	if (unlikely(!cmd))
58 		return;
59 
60 	ioarcb = &(cmd->rcb);
61 	ioasa = &(cmd->sa);
62 
63 	if (ioasa->rc.flags & SISL_RC_FLAGS_UNDERRUN) {
64 		resid = ioasa->resid;
65 		scsi_set_resid(scp, resid);
66 		dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
67 			__func__, cmd, scp, resid);
68 	}
69 
70 	if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
71 		dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p\n",
72 			__func__, cmd, scp);
73 		scp->result = (DID_ERROR << 16);
74 	}
75 
76 	dev_dbg(dev, "%s: cmd failed afu_rc=%02x scsi_rc=%02x fc_rc=%02x "
77 		"afu_extra=%02x scsi_extra=%02x fc_extra=%02x\n", __func__,
78 		ioasa->rc.afu_rc, ioasa->rc.scsi_rc, ioasa->rc.fc_rc,
79 		ioasa->afu_extra, ioasa->scsi_extra, ioasa->fc_extra);
80 
81 	if (ioasa->rc.scsi_rc) {
82 		/* We have a SCSI status */
83 		if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
84 			memcpy(scp->sense_buffer, ioasa->sense_data,
85 			       SISL_SENSE_DATA_LEN);
86 			scp->result = ioasa->rc.scsi_rc;
87 		} else
88 			scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
89 	}
90 
91 	/*
92 	 * We encountered an error. Set scp->result based on nature
93 	 * of error.
94 	 */
95 	if (ioasa->rc.fc_rc) {
96 		/* We have an FC status */
97 		switch (ioasa->rc.fc_rc) {
98 		case SISL_FC_RC_LINKDOWN:
99 			scp->result = (DID_REQUEUE << 16);
100 			break;
101 		case SISL_FC_RC_RESID:
102 			/* This indicates an FCP resid underrun */
103 			if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
104 				/* If the SISL_RC_FLAGS_OVERRUN flag was set,
105 				 * then we will handle this error else where.
106 				 * If not then we must handle it here.
107 				 * This is probably an AFU bug.
108 				 */
109 				scp->result = (DID_ERROR << 16);
110 			}
111 			break;
112 		case SISL_FC_RC_RESIDERR:
113 			/* Resid mismatch between adapter and device */
114 		case SISL_FC_RC_TGTABORT:
115 		case SISL_FC_RC_ABORTOK:
116 		case SISL_FC_RC_ABORTFAIL:
117 		case SISL_FC_RC_NOLOGI:
118 		case SISL_FC_RC_ABORTPEND:
119 		case SISL_FC_RC_WRABORTPEND:
120 		case SISL_FC_RC_NOEXP:
121 		case SISL_FC_RC_INUSE:
122 			scp->result = (DID_ERROR << 16);
123 			break;
124 		}
125 	}
126 
127 	if (ioasa->rc.afu_rc) {
128 		/* We have an AFU error */
129 		switch (ioasa->rc.afu_rc) {
130 		case SISL_AFU_RC_NO_CHANNELS:
131 			scp->result = (DID_NO_CONNECT << 16);
132 			break;
133 		case SISL_AFU_RC_DATA_DMA_ERR:
134 			switch (ioasa->afu_extra) {
135 			case SISL_AFU_DMA_ERR_PAGE_IN:
136 				/* Retry */
137 				scp->result = (DID_IMM_RETRY << 16);
138 				break;
139 			case SISL_AFU_DMA_ERR_INVALID_EA:
140 			default:
141 				scp->result = (DID_ERROR << 16);
142 			}
143 			break;
144 		case SISL_AFU_RC_OUT_OF_DATA_BUFS:
145 			/* Retry */
146 			scp->result = (DID_ALLOC_FAILURE << 16);
147 			break;
148 		default:
149 			scp->result = (DID_ERROR << 16);
150 		}
151 	}
152 }
153 
154 /**
155  * cmd_complete() - command completion handler
156  * @cmd:	AFU command that has completed.
157  *
158  * For SCSI commands this routine prepares and submits commands that have
159  * either completed or timed out to the SCSI stack. For internal commands
160  * (TMF or AFU), this routine simply notifies the originator that the
161  * command has completed.
162  */
163 static void cmd_complete(struct afu_cmd *cmd)
164 {
165 	struct scsi_cmnd *scp;
166 	ulong lock_flags;
167 	struct afu *afu = cmd->parent;
168 	struct cxlflash_cfg *cfg = afu->parent;
169 	struct device *dev = &cfg->dev->dev;
170 	struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
171 
172 	spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
173 	list_del(&cmd->list);
174 	spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
175 
176 	if (cmd->scp) {
177 		scp = cmd->scp;
178 		if (unlikely(cmd->sa.ioasc))
179 			process_cmd_err(cmd, scp);
180 		else
181 			scp->result = (DID_OK << 16);
182 
183 		dev_dbg_ratelimited(dev, "%s:scp=%p result=%08x ioasc=%08x\n",
184 				    __func__, scp, scp->result, cmd->sa.ioasc);
185 		scp->scsi_done(scp);
186 	} else if (cmd->cmd_tmf) {
187 		spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
188 		cfg->tmf_active = false;
189 		wake_up_all_locked(&cfg->tmf_waitq);
190 		spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
191 	} else
192 		complete(&cmd->cevent);
193 }
194 
195 /**
196  * flush_pending_cmds() - flush all pending commands on this hardware queue
197  * @hwq:	Hardware queue to flush.
198  *
199  * The hardware send queue lock associated with this hardware queue must be
200  * held when calling this routine.
201  */
202 static void flush_pending_cmds(struct hwq *hwq)
203 {
204 	struct cxlflash_cfg *cfg = hwq->afu->parent;
205 	struct afu_cmd *cmd, *tmp;
206 	struct scsi_cmnd *scp;
207 	ulong lock_flags;
208 
209 	list_for_each_entry_safe(cmd, tmp, &hwq->pending_cmds, list) {
210 		/* Bypass command when on a doneq, cmd_complete() will handle */
211 		if (!list_empty(&cmd->queue))
212 			continue;
213 
214 		list_del(&cmd->list);
215 
216 		if (cmd->scp) {
217 			scp = cmd->scp;
218 			scp->result = (DID_IMM_RETRY << 16);
219 			scp->scsi_done(scp);
220 		} else {
221 			cmd->cmd_aborted = true;
222 
223 			if (cmd->cmd_tmf) {
224 				spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
225 				cfg->tmf_active = false;
226 				wake_up_all_locked(&cfg->tmf_waitq);
227 				spin_unlock_irqrestore(&cfg->tmf_slock,
228 						       lock_flags);
229 			} else
230 				complete(&cmd->cevent);
231 		}
232 	}
233 }
234 
235 /**
236  * context_reset() - reset context via specified register
237  * @hwq:	Hardware queue owning the context to be reset.
238  * @reset_reg:	MMIO register to perform reset.
239  *
240  * When the reset is successful, the SISLite specification guarantees that
241  * the AFU has aborted all currently pending I/O. Accordingly, these commands
242  * must be flushed.
243  *
244  * Return: 0 on success, -errno on failure
245  */
246 static int context_reset(struct hwq *hwq, __be64 __iomem *reset_reg)
247 {
248 	struct cxlflash_cfg *cfg = hwq->afu->parent;
249 	struct device *dev = &cfg->dev->dev;
250 	int rc = -ETIMEDOUT;
251 	int nretry = 0;
252 	u64 val = 0x1;
253 	ulong lock_flags;
254 
255 	dev_dbg(dev, "%s: hwq=%p\n", __func__, hwq);
256 
257 	spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
258 
259 	writeq_be(val, reset_reg);
260 	do {
261 		val = readq_be(reset_reg);
262 		if ((val & 0x1) == 0x0) {
263 			rc = 0;
264 			break;
265 		}
266 
267 		/* Double delay each time */
268 		udelay(1 << nretry);
269 	} while (nretry++ < MC_ROOM_RETRY_CNT);
270 
271 	if (!rc)
272 		flush_pending_cmds(hwq);
273 
274 	spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
275 
276 	dev_dbg(dev, "%s: returning rc=%d, val=%016llx nretry=%d\n",
277 		__func__, rc, val, nretry);
278 	return rc;
279 }
280 
281 /**
282  * context_reset_ioarrin() - reset context via IOARRIN register
283  * @hwq:	Hardware queue owning the context to be reset.
284  *
285  * Return: 0 on success, -errno on failure
286  */
287 static int context_reset_ioarrin(struct hwq *hwq)
288 {
289 	return context_reset(hwq, &hwq->host_map->ioarrin);
290 }
291 
292 /**
293  * context_reset_sq() - reset context via SQ_CONTEXT_RESET register
294  * @hwq:	Hardware queue owning the context to be reset.
295  *
296  * Return: 0 on success, -errno on failure
297  */
298 static int context_reset_sq(struct hwq *hwq)
299 {
300 	return context_reset(hwq, &hwq->host_map->sq_ctx_reset);
301 }
302 
303 /**
304  * send_cmd_ioarrin() - sends an AFU command via IOARRIN register
305  * @afu:	AFU associated with the host.
306  * @cmd:	AFU command to send.
307  *
308  * Return:
309  *	0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
310  */
311 static int send_cmd_ioarrin(struct afu *afu, struct afu_cmd *cmd)
312 {
313 	struct cxlflash_cfg *cfg = afu->parent;
314 	struct device *dev = &cfg->dev->dev;
315 	struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
316 	int rc = 0;
317 	s64 room;
318 	ulong lock_flags;
319 
320 	/*
321 	 * To avoid the performance penalty of MMIO, spread the update of
322 	 * 'room' over multiple commands.
323 	 */
324 	spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
325 	if (--hwq->room < 0) {
326 		room = readq_be(&hwq->host_map->cmd_room);
327 		if (room <= 0) {
328 			dev_dbg_ratelimited(dev, "%s: no cmd_room to send "
329 					    "0x%02X, room=0x%016llX\n",
330 					    __func__, cmd->rcb.cdb[0], room);
331 			hwq->room = 0;
332 			rc = SCSI_MLQUEUE_HOST_BUSY;
333 			goto out;
334 		}
335 		hwq->room = room - 1;
336 	}
337 
338 	list_add(&cmd->list, &hwq->pending_cmds);
339 	writeq_be((u64)&cmd->rcb, &hwq->host_map->ioarrin);
340 out:
341 	spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
342 	dev_dbg(dev, "%s: cmd=%p len=%u ea=%016llx rc=%d\n", __func__,
343 		cmd, cmd->rcb.data_len, cmd->rcb.data_ea, rc);
344 	return rc;
345 }
346 
347 /**
348  * send_cmd_sq() - sends an AFU command via SQ ring
349  * @afu:	AFU associated with the host.
350  * @cmd:	AFU command to send.
351  *
352  * Return:
353  *	0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
354  */
355 static int send_cmd_sq(struct afu *afu, struct afu_cmd *cmd)
356 {
357 	struct cxlflash_cfg *cfg = afu->parent;
358 	struct device *dev = &cfg->dev->dev;
359 	struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
360 	int rc = 0;
361 	int newval;
362 	ulong lock_flags;
363 
364 	newval = atomic_dec_if_positive(&hwq->hsq_credits);
365 	if (newval <= 0) {
366 		rc = SCSI_MLQUEUE_HOST_BUSY;
367 		goto out;
368 	}
369 
370 	cmd->rcb.ioasa = &cmd->sa;
371 
372 	spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
373 
374 	*hwq->hsq_curr = cmd->rcb;
375 	if (hwq->hsq_curr < hwq->hsq_end)
376 		hwq->hsq_curr++;
377 	else
378 		hwq->hsq_curr = hwq->hsq_start;
379 
380 	list_add(&cmd->list, &hwq->pending_cmds);
381 	writeq_be((u64)hwq->hsq_curr, &hwq->host_map->sq_tail);
382 
383 	spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
384 out:
385 	dev_dbg(dev, "%s: cmd=%p len=%u ea=%016llx ioasa=%p rc=%d curr=%p "
386 	       "head=%016llx tail=%016llx\n", __func__, cmd, cmd->rcb.data_len,
387 	       cmd->rcb.data_ea, cmd->rcb.ioasa, rc, hwq->hsq_curr,
388 	       readq_be(&hwq->host_map->sq_head),
389 	       readq_be(&hwq->host_map->sq_tail));
390 	return rc;
391 }
392 
393 /**
394  * wait_resp() - polls for a response or timeout to a sent AFU command
395  * @afu:	AFU associated with the host.
396  * @cmd:	AFU command that was sent.
397  *
398  * Return: 0 on success, -errno on failure
399  */
400 static int wait_resp(struct afu *afu, struct afu_cmd *cmd)
401 {
402 	struct cxlflash_cfg *cfg = afu->parent;
403 	struct device *dev = &cfg->dev->dev;
404 	int rc = 0;
405 	ulong timeout = msecs_to_jiffies(cmd->rcb.timeout * 2 * 1000);
406 
407 	timeout = wait_for_completion_timeout(&cmd->cevent, timeout);
408 	if (!timeout)
409 		rc = -ETIMEDOUT;
410 
411 	if (cmd->cmd_aborted)
412 		rc = -EAGAIN;
413 
414 	if (unlikely(cmd->sa.ioasc != 0)) {
415 		dev_err(dev, "%s: cmd %02x failed, ioasc=%08x\n",
416 			__func__, cmd->rcb.cdb[0], cmd->sa.ioasc);
417 		rc = -EIO;
418 	}
419 
420 	return rc;
421 }
422 
423 /**
424  * cmd_to_target_hwq() - selects a target hardware queue for a SCSI command
425  * @host:	SCSI host associated with device.
426  * @scp:	SCSI command to send.
427  * @afu:	SCSI command to send.
428  *
429  * Hashes a command based upon the hardware queue mode.
430  *
431  * Return: Trusted index of target hardware queue
432  */
433 static u32 cmd_to_target_hwq(struct Scsi_Host *host, struct scsi_cmnd *scp,
434 			     struct afu *afu)
435 {
436 	u32 tag;
437 	u32 hwq = 0;
438 
439 	if (afu->num_hwqs == 1)
440 		return 0;
441 
442 	switch (afu->hwq_mode) {
443 	case HWQ_MODE_RR:
444 		hwq = afu->hwq_rr_count++ % afu->num_hwqs;
445 		break;
446 	case HWQ_MODE_TAG:
447 		tag = blk_mq_unique_tag(scp->request);
448 		hwq = blk_mq_unique_tag_to_hwq(tag);
449 		break;
450 	case HWQ_MODE_CPU:
451 		hwq = smp_processor_id() % afu->num_hwqs;
452 		break;
453 	default:
454 		WARN_ON_ONCE(1);
455 	}
456 
457 	return hwq;
458 }
459 
460 /**
461  * send_tmf() - sends a Task Management Function (TMF)
462  * @cfg:	Internal structure associated with the host.
463  * @sdev:	SCSI device destined for TMF.
464  * @tmfcmd:	TMF command to send.
465  *
466  * Return:
467  *	0 on success, SCSI_MLQUEUE_HOST_BUSY or -errno on failure
468  */
469 static int send_tmf(struct cxlflash_cfg *cfg, struct scsi_device *sdev,
470 		    u64 tmfcmd)
471 {
472 	struct afu *afu = cfg->afu;
473 	struct afu_cmd *cmd = NULL;
474 	struct device *dev = &cfg->dev->dev;
475 	struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
476 	char *buf = NULL;
477 	ulong lock_flags;
478 	int rc = 0;
479 	ulong to;
480 
481 	buf = kzalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
482 	if (unlikely(!buf)) {
483 		dev_err(dev, "%s: no memory for command\n", __func__);
484 		rc = -ENOMEM;
485 		goto out;
486 	}
487 
488 	cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
489 	INIT_LIST_HEAD(&cmd->queue);
490 
491 	/* When Task Management Function is active do not send another */
492 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
493 	if (cfg->tmf_active)
494 		wait_event_interruptible_lock_irq(cfg->tmf_waitq,
495 						  !cfg->tmf_active,
496 						  cfg->tmf_slock);
497 	cfg->tmf_active = true;
498 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
499 
500 	cmd->parent = afu;
501 	cmd->cmd_tmf = true;
502 	cmd->hwq_index = hwq->index;
503 
504 	cmd->rcb.ctx_id = hwq->ctx_hndl;
505 	cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
506 	cmd->rcb.port_sel = CHAN2PORTMASK(sdev->channel);
507 	cmd->rcb.lun_id = lun_to_lunid(sdev->lun);
508 	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
509 			      SISL_REQ_FLAGS_SUP_UNDERRUN |
510 			      SISL_REQ_FLAGS_TMF_CMD);
511 	memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
512 
513 	rc = afu->send_cmd(afu, cmd);
514 	if (unlikely(rc)) {
515 		spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
516 		cfg->tmf_active = false;
517 		spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
518 		goto out;
519 	}
520 
521 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
522 	to = msecs_to_jiffies(5000);
523 	to = wait_event_interruptible_lock_irq_timeout(cfg->tmf_waitq,
524 						       !cfg->tmf_active,
525 						       cfg->tmf_slock,
526 						       to);
527 	if (!to) {
528 		dev_err(dev, "%s: TMF timed out\n", __func__);
529 		rc = -ETIMEDOUT;
530 	} else if (cmd->cmd_aborted) {
531 		dev_err(dev, "%s: TMF aborted\n", __func__);
532 		rc = -EAGAIN;
533 	} else if (cmd->sa.ioasc) {
534 		dev_err(dev, "%s: TMF failed ioasc=%08x\n",
535 			__func__, cmd->sa.ioasc);
536 		rc = -EIO;
537 	}
538 	cfg->tmf_active = false;
539 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
540 out:
541 	kfree(buf);
542 	return rc;
543 }
544 
545 /**
546  * cxlflash_driver_info() - information handler for this host driver
547  * @host:	SCSI host associated with device.
548  *
549  * Return: A string describing the device.
550  */
551 static const char *cxlflash_driver_info(struct Scsi_Host *host)
552 {
553 	return CXLFLASH_ADAPTER_NAME;
554 }
555 
556 /**
557  * cxlflash_queuecommand() - sends a mid-layer request
558  * @host:	SCSI host associated with device.
559  * @scp:	SCSI command to send.
560  *
561  * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
562  */
563 static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
564 {
565 	struct cxlflash_cfg *cfg = shost_priv(host);
566 	struct afu *afu = cfg->afu;
567 	struct device *dev = &cfg->dev->dev;
568 	struct afu_cmd *cmd = sc_to_afuci(scp);
569 	struct scatterlist *sg = scsi_sglist(scp);
570 	int hwq_index = cmd_to_target_hwq(host, scp, afu);
571 	struct hwq *hwq = get_hwq(afu, hwq_index);
572 	u16 req_flags = SISL_REQ_FLAGS_SUP_UNDERRUN;
573 	ulong lock_flags;
574 	int rc = 0;
575 
576 	dev_dbg_ratelimited(dev, "%s: (scp=%p) %d/%d/%d/%llu "
577 			    "cdb=(%08x-%08x-%08x-%08x)\n",
578 			    __func__, scp, host->host_no, scp->device->channel,
579 			    scp->device->id, scp->device->lun,
580 			    get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
581 			    get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
582 			    get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
583 			    get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
584 
585 	/*
586 	 * If a Task Management Function is active, wait for it to complete
587 	 * before continuing with regular commands.
588 	 */
589 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
590 	if (cfg->tmf_active) {
591 		spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
592 		rc = SCSI_MLQUEUE_HOST_BUSY;
593 		goto out;
594 	}
595 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
596 
597 	switch (cfg->state) {
598 	case STATE_PROBING:
599 	case STATE_PROBED:
600 	case STATE_RESET:
601 		dev_dbg_ratelimited(dev, "%s: device is in reset\n", __func__);
602 		rc = SCSI_MLQUEUE_HOST_BUSY;
603 		goto out;
604 	case STATE_FAILTERM:
605 		dev_dbg_ratelimited(dev, "%s: device has failed\n", __func__);
606 		scp->result = (DID_NO_CONNECT << 16);
607 		scp->scsi_done(scp);
608 		rc = 0;
609 		goto out;
610 	default:
611 		break;
612 	}
613 
614 	if (likely(sg)) {
615 		cmd->rcb.data_len = sg->length;
616 		cmd->rcb.data_ea = (uintptr_t)sg_virt(sg);
617 	}
618 
619 	cmd->scp = scp;
620 	cmd->parent = afu;
621 	cmd->hwq_index = hwq_index;
622 
623 	cmd->rcb.ctx_id = hwq->ctx_hndl;
624 	cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
625 	cmd->rcb.port_sel = CHAN2PORTMASK(scp->device->channel);
626 	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
627 
628 	if (scp->sc_data_direction == DMA_TO_DEVICE)
629 		req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
630 
631 	cmd->rcb.req_flags = req_flags;
632 	memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
633 
634 	rc = afu->send_cmd(afu, cmd);
635 out:
636 	return rc;
637 }
638 
639 /**
640  * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
641  * @cfg:	Internal structure associated with the host.
642  */
643 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
644 {
645 	struct pci_dev *pdev = cfg->dev;
646 
647 	if (pci_channel_offline(pdev))
648 		wait_event_timeout(cfg->reset_waitq,
649 				   !pci_channel_offline(pdev),
650 				   CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
651 }
652 
653 /**
654  * free_mem() - free memory associated with the AFU
655  * @cfg:	Internal structure associated with the host.
656  */
657 static void free_mem(struct cxlflash_cfg *cfg)
658 {
659 	struct afu *afu = cfg->afu;
660 
661 	if (cfg->afu) {
662 		free_pages((ulong)afu, get_order(sizeof(struct afu)));
663 		cfg->afu = NULL;
664 	}
665 }
666 
667 /**
668  * cxlflash_reset_sync() - synchronizing point for asynchronous resets
669  * @cfg:	Internal structure associated with the host.
670  */
671 static void cxlflash_reset_sync(struct cxlflash_cfg *cfg)
672 {
673 	if (cfg->async_reset_cookie == 0)
674 		return;
675 
676 	/* Wait until all async calls prior to this cookie have completed */
677 	async_synchronize_cookie(cfg->async_reset_cookie + 1);
678 	cfg->async_reset_cookie = 0;
679 }
680 
681 /**
682  * stop_afu() - stops the AFU command timers and unmaps the MMIO space
683  * @cfg:	Internal structure associated with the host.
684  *
685  * Safe to call with AFU in a partially allocated/initialized state.
686  *
687  * Cancels scheduled worker threads, waits for any active internal AFU
688  * commands to timeout, disables IRQ polling and then unmaps the MMIO space.
689  */
690 static void stop_afu(struct cxlflash_cfg *cfg)
691 {
692 	struct afu *afu = cfg->afu;
693 	struct hwq *hwq;
694 	int i;
695 
696 	cancel_work_sync(&cfg->work_q);
697 	if (!current_is_async())
698 		cxlflash_reset_sync(cfg);
699 
700 	if (likely(afu)) {
701 		while (atomic_read(&afu->cmds_active))
702 			ssleep(1);
703 
704 		if (afu_is_irqpoll_enabled(afu)) {
705 			for (i = 0; i < afu->num_hwqs; i++) {
706 				hwq = get_hwq(afu, i);
707 
708 				irq_poll_disable(&hwq->irqpoll);
709 			}
710 		}
711 
712 		if (likely(afu->afu_map)) {
713 			cxl_psa_unmap((void __iomem *)afu->afu_map);
714 			afu->afu_map = NULL;
715 		}
716 	}
717 }
718 
719 /**
720  * term_intr() - disables all AFU interrupts
721  * @cfg:	Internal structure associated with the host.
722  * @level:	Depth of allocation, where to begin waterfall tear down.
723  * @index:	Index of the hardware queue.
724  *
725  * Safe to call with AFU/MC in partially allocated/initialized state.
726  */
727 static void term_intr(struct cxlflash_cfg *cfg, enum undo_level level,
728 		      u32 index)
729 {
730 	struct afu *afu = cfg->afu;
731 	struct device *dev = &cfg->dev->dev;
732 	struct hwq *hwq;
733 
734 	if (!afu) {
735 		dev_err(dev, "%s: returning with NULL afu\n", __func__);
736 		return;
737 	}
738 
739 	hwq = get_hwq(afu, index);
740 
741 	if (!hwq->ctx) {
742 		dev_err(dev, "%s: returning with NULL MC\n", __func__);
743 		return;
744 	}
745 
746 	switch (level) {
747 	case UNMAP_THREE:
748 		/* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
749 		if (index == PRIMARY_HWQ)
750 			cxl_unmap_afu_irq(hwq->ctx, 3, hwq);
751 	case UNMAP_TWO:
752 		cxl_unmap_afu_irq(hwq->ctx, 2, hwq);
753 	case UNMAP_ONE:
754 		cxl_unmap_afu_irq(hwq->ctx, 1, hwq);
755 	case FREE_IRQ:
756 		cxl_free_afu_irqs(hwq->ctx);
757 		/* fall through */
758 	case UNDO_NOOP:
759 		/* No action required */
760 		break;
761 	}
762 }
763 
764 /**
765  * term_mc() - terminates the master context
766  * @cfg:	Internal structure associated with the host.
767  * @index:	Index of the hardware queue.
768  *
769  * Safe to call with AFU/MC in partially allocated/initialized state.
770  */
771 static void term_mc(struct cxlflash_cfg *cfg, u32 index)
772 {
773 	struct afu *afu = cfg->afu;
774 	struct device *dev = &cfg->dev->dev;
775 	struct hwq *hwq;
776 	ulong lock_flags;
777 
778 	if (!afu) {
779 		dev_err(dev, "%s: returning with NULL afu\n", __func__);
780 		return;
781 	}
782 
783 	hwq = get_hwq(afu, index);
784 
785 	if (!hwq->ctx) {
786 		dev_err(dev, "%s: returning with NULL MC\n", __func__);
787 		return;
788 	}
789 
790 	WARN_ON(cxl_stop_context(hwq->ctx));
791 	if (index != PRIMARY_HWQ)
792 		WARN_ON(cxl_release_context(hwq->ctx));
793 	hwq->ctx = NULL;
794 
795 	spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
796 	flush_pending_cmds(hwq);
797 	spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
798 }
799 
800 /**
801  * term_afu() - terminates the AFU
802  * @cfg:	Internal structure associated with the host.
803  *
804  * Safe to call with AFU/MC in partially allocated/initialized state.
805  */
806 static void term_afu(struct cxlflash_cfg *cfg)
807 {
808 	struct device *dev = &cfg->dev->dev;
809 	int k;
810 
811 	/*
812 	 * Tear down is carefully orchestrated to ensure
813 	 * no interrupts can come in when the problem state
814 	 * area is unmapped.
815 	 *
816 	 * 1) Disable all AFU interrupts for each master
817 	 * 2) Unmap the problem state area
818 	 * 3) Stop each master context
819 	 */
820 	for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
821 		term_intr(cfg, UNMAP_THREE, k);
822 
823 	stop_afu(cfg);
824 
825 	for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
826 		term_mc(cfg, k);
827 
828 	dev_dbg(dev, "%s: returning\n", __func__);
829 }
830 
831 /**
832  * notify_shutdown() - notifies device of pending shutdown
833  * @cfg:	Internal structure associated with the host.
834  * @wait:	Whether to wait for shutdown processing to complete.
835  *
836  * This function will notify the AFU that the adapter is being shutdown
837  * and will wait for shutdown processing to complete if wait is true.
838  * This notification should flush pending I/Os to the device and halt
839  * further I/Os until the next AFU reset is issued and device restarted.
840  */
841 static void notify_shutdown(struct cxlflash_cfg *cfg, bool wait)
842 {
843 	struct afu *afu = cfg->afu;
844 	struct device *dev = &cfg->dev->dev;
845 	struct dev_dependent_vals *ddv;
846 	__be64 __iomem *fc_port_regs;
847 	u64 reg, status;
848 	int i, retry_cnt = 0;
849 
850 	ddv = (struct dev_dependent_vals *)cfg->dev_id->driver_data;
851 	if (!(ddv->flags & CXLFLASH_NOTIFY_SHUTDOWN))
852 		return;
853 
854 	if (!afu || !afu->afu_map) {
855 		dev_dbg(dev, "%s: Problem state area not mapped\n", __func__);
856 		return;
857 	}
858 
859 	/* Notify AFU */
860 	for (i = 0; i < cfg->num_fc_ports; i++) {
861 		fc_port_regs = get_fc_port_regs(cfg, i);
862 
863 		reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
864 		reg |= SISL_FC_SHUTDOWN_NORMAL;
865 		writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
866 	}
867 
868 	if (!wait)
869 		return;
870 
871 	/* Wait up to 1.5 seconds for shutdown processing to complete */
872 	for (i = 0; i < cfg->num_fc_ports; i++) {
873 		fc_port_regs = get_fc_port_regs(cfg, i);
874 		retry_cnt = 0;
875 
876 		while (true) {
877 			status = readq_be(&fc_port_regs[FC_STATUS / 8]);
878 			if (status & SISL_STATUS_SHUTDOWN_COMPLETE)
879 				break;
880 			if (++retry_cnt >= MC_RETRY_CNT) {
881 				dev_dbg(dev, "%s: port %d shutdown processing "
882 					"not yet completed\n", __func__, i);
883 				break;
884 			}
885 			msleep(100 * retry_cnt);
886 		}
887 	}
888 }
889 
890 /**
891  * cxlflash_get_minor() - gets the first available minor number
892  *
893  * Return: Unique minor number that can be used to create the character device.
894  */
895 static int cxlflash_get_minor(void)
896 {
897 	int minor;
898 	long bit;
899 
900 	bit = find_first_zero_bit(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
901 	if (bit >= CXLFLASH_MAX_ADAPTERS)
902 		return -1;
903 
904 	minor = bit & MINORMASK;
905 	set_bit(minor, cxlflash_minor);
906 	return minor;
907 }
908 
909 /**
910  * cxlflash_put_minor() - releases the minor number
911  * @minor:	Minor number that is no longer needed.
912  */
913 static void cxlflash_put_minor(int minor)
914 {
915 	clear_bit(minor, cxlflash_minor);
916 }
917 
918 /**
919  * cxlflash_release_chrdev() - release the character device for the host
920  * @cfg:	Internal structure associated with the host.
921  */
922 static void cxlflash_release_chrdev(struct cxlflash_cfg *cfg)
923 {
924 	device_unregister(cfg->chardev);
925 	cfg->chardev = NULL;
926 	cdev_del(&cfg->cdev);
927 	cxlflash_put_minor(MINOR(cfg->cdev.dev));
928 }
929 
930 /**
931  * cxlflash_remove() - PCI entry point to tear down host
932  * @pdev:	PCI device associated with the host.
933  *
934  * Safe to use as a cleanup in partially allocated/initialized state. Note that
935  * the reset_waitq is flushed as part of the stop/termination of user contexts.
936  */
937 static void cxlflash_remove(struct pci_dev *pdev)
938 {
939 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
940 	struct device *dev = &pdev->dev;
941 	ulong lock_flags;
942 
943 	if (!pci_is_enabled(pdev)) {
944 		dev_dbg(dev, "%s: Device is disabled\n", __func__);
945 		return;
946 	}
947 
948 	/* If a Task Management Function is active, wait for it to complete
949 	 * before continuing with remove.
950 	 */
951 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
952 	if (cfg->tmf_active)
953 		wait_event_interruptible_lock_irq(cfg->tmf_waitq,
954 						  !cfg->tmf_active,
955 						  cfg->tmf_slock);
956 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
957 
958 	/* Notify AFU and wait for shutdown processing to complete */
959 	notify_shutdown(cfg, true);
960 
961 	cfg->state = STATE_FAILTERM;
962 	cxlflash_stop_term_user_contexts(cfg);
963 
964 	switch (cfg->init_state) {
965 	case INIT_STATE_CDEV:
966 		cxlflash_release_chrdev(cfg);
967 	case INIT_STATE_SCSI:
968 		cxlflash_term_local_luns(cfg);
969 		scsi_remove_host(cfg->host);
970 	case INIT_STATE_AFU:
971 		term_afu(cfg);
972 	case INIT_STATE_PCI:
973 		pci_disable_device(pdev);
974 	case INIT_STATE_NONE:
975 		free_mem(cfg);
976 		scsi_host_put(cfg->host);
977 		break;
978 	}
979 
980 	dev_dbg(dev, "%s: returning\n", __func__);
981 }
982 
983 /**
984  * alloc_mem() - allocates the AFU and its command pool
985  * @cfg:	Internal structure associated with the host.
986  *
987  * A partially allocated state remains on failure.
988  *
989  * Return:
990  *	0 on success
991  *	-ENOMEM on failure to allocate memory
992  */
993 static int alloc_mem(struct cxlflash_cfg *cfg)
994 {
995 	int rc = 0;
996 	struct device *dev = &cfg->dev->dev;
997 
998 	/* AFU is ~28k, i.e. only one 64k page or up to seven 4k pages */
999 	cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1000 					    get_order(sizeof(struct afu)));
1001 	if (unlikely(!cfg->afu)) {
1002 		dev_err(dev, "%s: cannot get %d free pages\n",
1003 			__func__, get_order(sizeof(struct afu)));
1004 		rc = -ENOMEM;
1005 		goto out;
1006 	}
1007 	cfg->afu->parent = cfg;
1008 	cfg->afu->desired_hwqs = CXLFLASH_DEF_HWQS;
1009 	cfg->afu->afu_map = NULL;
1010 out:
1011 	return rc;
1012 }
1013 
1014 /**
1015  * init_pci() - initializes the host as a PCI device
1016  * @cfg:	Internal structure associated with the host.
1017  *
1018  * Return: 0 on success, -errno on failure
1019  */
1020 static int init_pci(struct cxlflash_cfg *cfg)
1021 {
1022 	struct pci_dev *pdev = cfg->dev;
1023 	struct device *dev = &cfg->dev->dev;
1024 	int rc = 0;
1025 
1026 	rc = pci_enable_device(pdev);
1027 	if (rc || pci_channel_offline(pdev)) {
1028 		if (pci_channel_offline(pdev)) {
1029 			cxlflash_wait_for_pci_err_recovery(cfg);
1030 			rc = pci_enable_device(pdev);
1031 		}
1032 
1033 		if (rc) {
1034 			dev_err(dev, "%s: Cannot enable adapter\n", __func__);
1035 			cxlflash_wait_for_pci_err_recovery(cfg);
1036 			goto out;
1037 		}
1038 	}
1039 
1040 out:
1041 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1042 	return rc;
1043 }
1044 
1045 /**
1046  * init_scsi() - adds the host to the SCSI stack and kicks off host scan
1047  * @cfg:	Internal structure associated with the host.
1048  *
1049  * Return: 0 on success, -errno on failure
1050  */
1051 static int init_scsi(struct cxlflash_cfg *cfg)
1052 {
1053 	struct pci_dev *pdev = cfg->dev;
1054 	struct device *dev = &cfg->dev->dev;
1055 	int rc = 0;
1056 
1057 	rc = scsi_add_host(cfg->host, &pdev->dev);
1058 	if (rc) {
1059 		dev_err(dev, "%s: scsi_add_host failed rc=%d\n", __func__, rc);
1060 		goto out;
1061 	}
1062 
1063 	scsi_scan_host(cfg->host);
1064 
1065 out:
1066 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1067 	return rc;
1068 }
1069 
1070 /**
1071  * set_port_online() - transitions the specified host FC port to online state
1072  * @fc_regs:	Top of MMIO region defined for specified port.
1073  *
1074  * The provided MMIO region must be mapped prior to call. Online state means
1075  * that the FC link layer has synced, completed the handshaking process, and
1076  * is ready for login to start.
1077  */
1078 static void set_port_online(__be64 __iomem *fc_regs)
1079 {
1080 	u64 cmdcfg;
1081 
1082 	cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1083 	cmdcfg &= (~FC_MTIP_CMDCONFIG_OFFLINE);	/* clear OFF_LINE */
1084 	cmdcfg |= (FC_MTIP_CMDCONFIG_ONLINE);	/* set ON_LINE */
1085 	writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1086 }
1087 
1088 /**
1089  * set_port_offline() - transitions the specified host FC port to offline state
1090  * @fc_regs:	Top of MMIO region defined for specified port.
1091  *
1092  * The provided MMIO region must be mapped prior to call.
1093  */
1094 static void set_port_offline(__be64 __iomem *fc_regs)
1095 {
1096 	u64 cmdcfg;
1097 
1098 	cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1099 	cmdcfg &= (~FC_MTIP_CMDCONFIG_ONLINE);	/* clear ON_LINE */
1100 	cmdcfg |= (FC_MTIP_CMDCONFIG_OFFLINE);	/* set OFF_LINE */
1101 	writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1102 }
1103 
1104 /**
1105  * wait_port_online() - waits for the specified host FC port come online
1106  * @fc_regs:	Top of MMIO region defined for specified port.
1107  * @delay_us:	Number of microseconds to delay between reading port status.
1108  * @nretry:	Number of cycles to retry reading port status.
1109  *
1110  * The provided MMIO region must be mapped prior to call. This will timeout
1111  * when the cable is not plugged in.
1112  *
1113  * Return:
1114  *	TRUE (1) when the specified port is online
1115  *	FALSE (0) when the specified port fails to come online after timeout
1116  */
1117 static bool wait_port_online(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1118 {
1119 	u64 status;
1120 
1121 	WARN_ON(delay_us < 1000);
1122 
1123 	do {
1124 		msleep(delay_us / 1000);
1125 		status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1126 		if (status == U64_MAX)
1127 			nretry /= 2;
1128 	} while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_ONLINE &&
1129 		 nretry--);
1130 
1131 	return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_ONLINE);
1132 }
1133 
1134 /**
1135  * wait_port_offline() - waits for the specified host FC port go offline
1136  * @fc_regs:	Top of MMIO region defined for specified port.
1137  * @delay_us:	Number of microseconds to delay between reading port status.
1138  * @nretry:	Number of cycles to retry reading port status.
1139  *
1140  * The provided MMIO region must be mapped prior to call.
1141  *
1142  * Return:
1143  *	TRUE (1) when the specified port is offline
1144  *	FALSE (0) when the specified port fails to go offline after timeout
1145  */
1146 static bool wait_port_offline(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1147 {
1148 	u64 status;
1149 
1150 	WARN_ON(delay_us < 1000);
1151 
1152 	do {
1153 		msleep(delay_us / 1000);
1154 		status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1155 		if (status == U64_MAX)
1156 			nretry /= 2;
1157 	} while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_OFFLINE &&
1158 		 nretry--);
1159 
1160 	return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_OFFLINE);
1161 }
1162 
1163 /**
1164  * afu_set_wwpn() - configures the WWPN for the specified host FC port
1165  * @afu:	AFU associated with the host that owns the specified FC port.
1166  * @port:	Port number being configured.
1167  * @fc_regs:	Top of MMIO region defined for specified port.
1168  * @wwpn:	The world-wide-port-number previously discovered for port.
1169  *
1170  * The provided MMIO region must be mapped prior to call. As part of the
1171  * sequence to configure the WWPN, the port is toggled offline and then back
1172  * online. This toggling action can cause this routine to delay up to a few
1173  * seconds. When configured to use the internal LUN feature of the AFU, a
1174  * failure to come online is overridden.
1175  */
1176 static void afu_set_wwpn(struct afu *afu, int port, __be64 __iomem *fc_regs,
1177 			 u64 wwpn)
1178 {
1179 	struct cxlflash_cfg *cfg = afu->parent;
1180 	struct device *dev = &cfg->dev->dev;
1181 
1182 	set_port_offline(fc_regs);
1183 	if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1184 			       FC_PORT_STATUS_RETRY_CNT)) {
1185 		dev_dbg(dev, "%s: wait on port %d to go offline timed out\n",
1186 			__func__, port);
1187 	}
1188 
1189 	writeq_be(wwpn, &fc_regs[FC_PNAME / 8]);
1190 
1191 	set_port_online(fc_regs);
1192 	if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1193 			      FC_PORT_STATUS_RETRY_CNT)) {
1194 		dev_dbg(dev, "%s: wait on port %d to go online timed out\n",
1195 			__func__, port);
1196 	}
1197 }
1198 
1199 /**
1200  * afu_link_reset() - resets the specified host FC port
1201  * @afu:	AFU associated with the host that owns the specified FC port.
1202  * @port:	Port number being configured.
1203  * @fc_regs:	Top of MMIO region defined for specified port.
1204  *
1205  * The provided MMIO region must be mapped prior to call. The sequence to
1206  * reset the port involves toggling it offline and then back online. This
1207  * action can cause this routine to delay up to a few seconds. An effort
1208  * is made to maintain link with the device by switching to host to use
1209  * the alternate port exclusively while the reset takes place.
1210  * failure to come online is overridden.
1211  */
1212 static void afu_link_reset(struct afu *afu, int port, __be64 __iomem *fc_regs)
1213 {
1214 	struct cxlflash_cfg *cfg = afu->parent;
1215 	struct device *dev = &cfg->dev->dev;
1216 	u64 port_sel;
1217 
1218 	/* first switch the AFU to the other links, if any */
1219 	port_sel = readq_be(&afu->afu_map->global.regs.afu_port_sel);
1220 	port_sel &= ~(1ULL << port);
1221 	writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1222 	cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1223 
1224 	set_port_offline(fc_regs);
1225 	if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1226 			       FC_PORT_STATUS_RETRY_CNT))
1227 		dev_err(dev, "%s: wait on port %d to go offline timed out\n",
1228 			__func__, port);
1229 
1230 	set_port_online(fc_regs);
1231 	if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1232 			      FC_PORT_STATUS_RETRY_CNT))
1233 		dev_err(dev, "%s: wait on port %d to go online timed out\n",
1234 			__func__, port);
1235 
1236 	/* switch back to include this port */
1237 	port_sel |= (1ULL << port);
1238 	writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1239 	cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1240 
1241 	dev_dbg(dev, "%s: returning port_sel=%016llx\n", __func__, port_sel);
1242 }
1243 
1244 /**
1245  * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1246  * @afu:	AFU associated with the host.
1247  */
1248 static void afu_err_intr_init(struct afu *afu)
1249 {
1250 	struct cxlflash_cfg *cfg = afu->parent;
1251 	__be64 __iomem *fc_port_regs;
1252 	int i;
1253 	struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
1254 	u64 reg;
1255 
1256 	/* global async interrupts: AFU clears afu_ctrl on context exit
1257 	 * if async interrupts were sent to that context. This prevents
1258 	 * the AFU form sending further async interrupts when
1259 	 * there is
1260 	 * nobody to receive them.
1261 	 */
1262 
1263 	/* mask all */
1264 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_mask);
1265 	/* set LISN# to send and point to primary master context */
1266 	reg = ((u64) (((hwq->ctx_hndl << 8) | SISL_MSI_ASYNC_ERROR)) << 40);
1267 
1268 	if (afu->internal_lun)
1269 		reg |= 1;	/* Bit 63 indicates local lun */
1270 	writeq_be(reg, &afu->afu_map->global.regs.afu_ctrl);
1271 	/* clear all */
1272 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1273 	/* unmask bits that are of interest */
1274 	/* note: afu can send an interrupt after this step */
1275 	writeq_be(SISL_ASTATUS_MASK, &afu->afu_map->global.regs.aintr_mask);
1276 	/* clear again in case a bit came on after previous clear but before */
1277 	/* unmask */
1278 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1279 
1280 	/* Clear/Set internal lun bits */
1281 	fc_port_regs = get_fc_port_regs(cfg, 0);
1282 	reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
1283 	reg &= SISL_FC_INTERNAL_MASK;
1284 	if (afu->internal_lun)
1285 		reg |= ((u64)(afu->internal_lun - 1) << SISL_FC_INTERNAL_SHIFT);
1286 	writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
1287 
1288 	/* now clear FC errors */
1289 	for (i = 0; i < cfg->num_fc_ports; i++) {
1290 		fc_port_regs = get_fc_port_regs(cfg, i);
1291 
1292 		writeq_be(0xFFFFFFFFU, &fc_port_regs[FC_ERROR / 8]);
1293 		writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1294 	}
1295 
1296 	/* sync interrupts for master's IOARRIN write */
1297 	/* note that unlike asyncs, there can be no pending sync interrupts */
1298 	/* at this time (this is a fresh context and master has not written */
1299 	/* IOARRIN yet), so there is nothing to clear. */
1300 
1301 	/* set LISN#, it is always sent to the context that wrote IOARRIN */
1302 	for (i = 0; i < afu->num_hwqs; i++) {
1303 		hwq = get_hwq(afu, i);
1304 
1305 		writeq_be(SISL_MSI_SYNC_ERROR, &hwq->host_map->ctx_ctrl);
1306 		writeq_be(SISL_ISTATUS_MASK, &hwq->host_map->intr_mask);
1307 	}
1308 }
1309 
1310 /**
1311  * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1312  * @irq:	Interrupt number.
1313  * @data:	Private data provided at interrupt registration, the AFU.
1314  *
1315  * Return: Always return IRQ_HANDLED.
1316  */
1317 static irqreturn_t cxlflash_sync_err_irq(int irq, void *data)
1318 {
1319 	struct hwq *hwq = (struct hwq *)data;
1320 	struct cxlflash_cfg *cfg = hwq->afu->parent;
1321 	struct device *dev = &cfg->dev->dev;
1322 	u64 reg;
1323 	u64 reg_unmasked;
1324 
1325 	reg = readq_be(&hwq->host_map->intr_status);
1326 	reg_unmasked = (reg & SISL_ISTATUS_UNMASK);
1327 
1328 	if (reg_unmasked == 0UL) {
1329 		dev_err(dev, "%s: spurious interrupt, intr_status=%016llx\n",
1330 			__func__, reg);
1331 		goto cxlflash_sync_err_irq_exit;
1332 	}
1333 
1334 	dev_err(dev, "%s: unexpected interrupt, intr_status=%016llx\n",
1335 		__func__, reg);
1336 
1337 	writeq_be(reg_unmasked, &hwq->host_map->intr_clear);
1338 
1339 cxlflash_sync_err_irq_exit:
1340 	return IRQ_HANDLED;
1341 }
1342 
1343 /**
1344  * process_hrrq() - process the read-response queue
1345  * @afu:	AFU associated with the host.
1346  * @doneq:	Queue of commands harvested from the RRQ.
1347  * @budget:	Threshold of RRQ entries to process.
1348  *
1349  * This routine must be called holding the disabled RRQ spin lock.
1350  *
1351  * Return: The number of entries processed.
1352  */
1353 static int process_hrrq(struct hwq *hwq, struct list_head *doneq, int budget)
1354 {
1355 	struct afu *afu = hwq->afu;
1356 	struct afu_cmd *cmd;
1357 	struct sisl_ioasa *ioasa;
1358 	struct sisl_ioarcb *ioarcb;
1359 	bool toggle = hwq->toggle;
1360 	int num_hrrq = 0;
1361 	u64 entry,
1362 	    *hrrq_start = hwq->hrrq_start,
1363 	    *hrrq_end = hwq->hrrq_end,
1364 	    *hrrq_curr = hwq->hrrq_curr;
1365 
1366 	/* Process ready RRQ entries up to the specified budget (if any) */
1367 	while (true) {
1368 		entry = *hrrq_curr;
1369 
1370 		if ((entry & SISL_RESP_HANDLE_T_BIT) != toggle)
1371 			break;
1372 
1373 		entry &= ~SISL_RESP_HANDLE_T_BIT;
1374 
1375 		if (afu_is_sq_cmd_mode(afu)) {
1376 			ioasa = (struct sisl_ioasa *)entry;
1377 			cmd = container_of(ioasa, struct afu_cmd, sa);
1378 		} else {
1379 			ioarcb = (struct sisl_ioarcb *)entry;
1380 			cmd = container_of(ioarcb, struct afu_cmd, rcb);
1381 		}
1382 
1383 		list_add_tail(&cmd->queue, doneq);
1384 
1385 		/* Advance to next entry or wrap and flip the toggle bit */
1386 		if (hrrq_curr < hrrq_end)
1387 			hrrq_curr++;
1388 		else {
1389 			hrrq_curr = hrrq_start;
1390 			toggle ^= SISL_RESP_HANDLE_T_BIT;
1391 		}
1392 
1393 		atomic_inc(&hwq->hsq_credits);
1394 		num_hrrq++;
1395 
1396 		if (budget > 0 && num_hrrq >= budget)
1397 			break;
1398 	}
1399 
1400 	hwq->hrrq_curr = hrrq_curr;
1401 	hwq->toggle = toggle;
1402 
1403 	return num_hrrq;
1404 }
1405 
1406 /**
1407  * process_cmd_doneq() - process a queue of harvested RRQ commands
1408  * @doneq:	Queue of completed commands.
1409  *
1410  * Note that upon return the queue can no longer be trusted.
1411  */
1412 static void process_cmd_doneq(struct list_head *doneq)
1413 {
1414 	struct afu_cmd *cmd, *tmp;
1415 
1416 	WARN_ON(list_empty(doneq));
1417 
1418 	list_for_each_entry_safe(cmd, tmp, doneq, queue)
1419 		cmd_complete(cmd);
1420 }
1421 
1422 /**
1423  * cxlflash_irqpoll() - process a queue of harvested RRQ commands
1424  * @irqpoll:	IRQ poll structure associated with queue to poll.
1425  * @budget:	Threshold of RRQ entries to process per poll.
1426  *
1427  * Return: The number of entries processed.
1428  */
1429 static int cxlflash_irqpoll(struct irq_poll *irqpoll, int budget)
1430 {
1431 	struct hwq *hwq = container_of(irqpoll, struct hwq, irqpoll);
1432 	unsigned long hrrq_flags;
1433 	LIST_HEAD(doneq);
1434 	int num_entries = 0;
1435 
1436 	spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1437 
1438 	num_entries = process_hrrq(hwq, &doneq, budget);
1439 	if (num_entries < budget)
1440 		irq_poll_complete(irqpoll);
1441 
1442 	spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1443 
1444 	process_cmd_doneq(&doneq);
1445 	return num_entries;
1446 }
1447 
1448 /**
1449  * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1450  * @irq:	Interrupt number.
1451  * @data:	Private data provided at interrupt registration, the AFU.
1452  *
1453  * Return: IRQ_HANDLED or IRQ_NONE when no ready entries found.
1454  */
1455 static irqreturn_t cxlflash_rrq_irq(int irq, void *data)
1456 {
1457 	struct hwq *hwq = (struct hwq *)data;
1458 	struct afu *afu = hwq->afu;
1459 	unsigned long hrrq_flags;
1460 	LIST_HEAD(doneq);
1461 	int num_entries = 0;
1462 
1463 	spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1464 
1465 	if (afu_is_irqpoll_enabled(afu)) {
1466 		irq_poll_sched(&hwq->irqpoll);
1467 		spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1468 		return IRQ_HANDLED;
1469 	}
1470 
1471 	num_entries = process_hrrq(hwq, &doneq, -1);
1472 	spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1473 
1474 	if (num_entries == 0)
1475 		return IRQ_NONE;
1476 
1477 	process_cmd_doneq(&doneq);
1478 	return IRQ_HANDLED;
1479 }
1480 
1481 /*
1482  * Asynchronous interrupt information table
1483  *
1484  * NOTE:
1485  *	- Order matters here as this array is indexed by bit position.
1486  *
1487  *	- The checkpatch script considers the BUILD_SISL_ASTATUS_FC_PORT macro
1488  *	  as complex and complains due to a lack of parentheses/braces.
1489  */
1490 #define ASTATUS_FC(_a, _b, _c, _d)					 \
1491 	{ SISL_ASTATUS_FC##_a##_##_b, _c, _a, (_d) }
1492 
1493 #define BUILD_SISL_ASTATUS_FC_PORT(_a)					 \
1494 	ASTATUS_FC(_a, LINK_UP, "link up", 0),				 \
1495 	ASTATUS_FC(_a, LINK_DN, "link down", 0),			 \
1496 	ASTATUS_FC(_a, LOGI_S, "login succeeded", SCAN_HOST),		 \
1497 	ASTATUS_FC(_a, LOGI_F, "login failed", CLR_FC_ERROR),		 \
1498 	ASTATUS_FC(_a, LOGI_R, "login timed out, retrying", LINK_RESET), \
1499 	ASTATUS_FC(_a, CRC_T, "CRC threshold exceeded", LINK_RESET),	 \
1500 	ASTATUS_FC(_a, LOGO, "target initiated LOGO", 0),		 \
1501 	ASTATUS_FC(_a, OTHER, "other error", CLR_FC_ERROR | LINK_RESET)
1502 
1503 static const struct asyc_intr_info ainfo[] = {
1504 	BUILD_SISL_ASTATUS_FC_PORT(1),
1505 	BUILD_SISL_ASTATUS_FC_PORT(0),
1506 	BUILD_SISL_ASTATUS_FC_PORT(3),
1507 	BUILD_SISL_ASTATUS_FC_PORT(2)
1508 };
1509 
1510 /**
1511  * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1512  * @irq:	Interrupt number.
1513  * @data:	Private data provided at interrupt registration, the AFU.
1514  *
1515  * Return: Always return IRQ_HANDLED.
1516  */
1517 static irqreturn_t cxlflash_async_err_irq(int irq, void *data)
1518 {
1519 	struct hwq *hwq = (struct hwq *)data;
1520 	struct afu *afu = hwq->afu;
1521 	struct cxlflash_cfg *cfg = afu->parent;
1522 	struct device *dev = &cfg->dev->dev;
1523 	const struct asyc_intr_info *info;
1524 	struct sisl_global_map __iomem *global = &afu->afu_map->global;
1525 	__be64 __iomem *fc_port_regs;
1526 	u64 reg_unmasked;
1527 	u64 reg;
1528 	u64 bit;
1529 	u8 port;
1530 
1531 	reg = readq_be(&global->regs.aintr_status);
1532 	reg_unmasked = (reg & SISL_ASTATUS_UNMASK);
1533 
1534 	if (unlikely(reg_unmasked == 0)) {
1535 		dev_err(dev, "%s: spurious interrupt, aintr_status=%016llx\n",
1536 			__func__, reg);
1537 		goto out;
1538 	}
1539 
1540 	/* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1541 	writeq_be(reg_unmasked, &global->regs.aintr_clear);
1542 
1543 	/* Check each bit that is on */
1544 	for_each_set_bit(bit, (ulong *)&reg_unmasked, BITS_PER_LONG) {
1545 		if (unlikely(bit >= ARRAY_SIZE(ainfo))) {
1546 			WARN_ON_ONCE(1);
1547 			continue;
1548 		}
1549 
1550 		info = &ainfo[bit];
1551 		if (unlikely(info->status != 1ULL << bit)) {
1552 			WARN_ON_ONCE(1);
1553 			continue;
1554 		}
1555 
1556 		port = info->port;
1557 		fc_port_regs = get_fc_port_regs(cfg, port);
1558 
1559 		dev_err(dev, "%s: FC Port %d -> %s, fc_status=%016llx\n",
1560 			__func__, port, info->desc,
1561 		       readq_be(&fc_port_regs[FC_STATUS / 8]));
1562 
1563 		/*
1564 		 * Do link reset first, some OTHER errors will set FC_ERROR
1565 		 * again if cleared before or w/o a reset
1566 		 */
1567 		if (info->action & LINK_RESET) {
1568 			dev_err(dev, "%s: FC Port %d: resetting link\n",
1569 				__func__, port);
1570 			cfg->lr_state = LINK_RESET_REQUIRED;
1571 			cfg->lr_port = port;
1572 			schedule_work(&cfg->work_q);
1573 		}
1574 
1575 		if (info->action & CLR_FC_ERROR) {
1576 			reg = readq_be(&fc_port_regs[FC_ERROR / 8]);
1577 
1578 			/*
1579 			 * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1580 			 * should be the same and tracing one is sufficient.
1581 			 */
1582 
1583 			dev_err(dev, "%s: fc %d: clearing fc_error=%016llx\n",
1584 				__func__, port, reg);
1585 
1586 			writeq_be(reg, &fc_port_regs[FC_ERROR / 8]);
1587 			writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1588 		}
1589 
1590 		if (info->action & SCAN_HOST) {
1591 			atomic_inc(&cfg->scan_host_needed);
1592 			schedule_work(&cfg->work_q);
1593 		}
1594 	}
1595 
1596 out:
1597 	return IRQ_HANDLED;
1598 }
1599 
1600 /**
1601  * start_context() - starts the master context
1602  * @cfg:	Internal structure associated with the host.
1603  * @index:	Index of the hardware queue.
1604  *
1605  * Return: A success or failure value from CXL services.
1606  */
1607 static int start_context(struct cxlflash_cfg *cfg, u32 index)
1608 {
1609 	struct device *dev = &cfg->dev->dev;
1610 	struct hwq *hwq = get_hwq(cfg->afu, index);
1611 	int rc = 0;
1612 
1613 	rc = cxl_start_context(hwq->ctx,
1614 			       hwq->work.work_element_descriptor,
1615 			       NULL);
1616 
1617 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1618 	return rc;
1619 }
1620 
1621 /**
1622  * read_vpd() - obtains the WWPNs from VPD
1623  * @cfg:	Internal structure associated with the host.
1624  * @wwpn:	Array of size MAX_FC_PORTS to pass back WWPNs
1625  *
1626  * Return: 0 on success, -errno on failure
1627  */
1628 static int read_vpd(struct cxlflash_cfg *cfg, u64 wwpn[])
1629 {
1630 	struct device *dev = &cfg->dev->dev;
1631 	struct pci_dev *pdev = cfg->dev;
1632 	int rc = 0;
1633 	int ro_start, ro_size, i, j, k;
1634 	ssize_t vpd_size;
1635 	char vpd_data[CXLFLASH_VPD_LEN];
1636 	char tmp_buf[WWPN_BUF_LEN] = { 0 };
1637 	char *wwpn_vpd_tags[MAX_FC_PORTS] = { "V5", "V6", "V7", "V8" };
1638 
1639 	/* Get the VPD data from the device */
1640 	vpd_size = cxl_read_adapter_vpd(pdev, vpd_data, sizeof(vpd_data));
1641 	if (unlikely(vpd_size <= 0)) {
1642 		dev_err(dev, "%s: Unable to read VPD (size = %ld)\n",
1643 			__func__, vpd_size);
1644 		rc = -ENODEV;
1645 		goto out;
1646 	}
1647 
1648 	/* Get the read only section offset */
1649 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size,
1650 				    PCI_VPD_LRDT_RO_DATA);
1651 	if (unlikely(ro_start < 0)) {
1652 		dev_err(dev, "%s: VPD Read-only data not found\n", __func__);
1653 		rc = -ENODEV;
1654 		goto out;
1655 	}
1656 
1657 	/* Get the read only section size, cap when extends beyond read VPD */
1658 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
1659 	j = ro_size;
1660 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1661 	if (unlikely((i + j) > vpd_size)) {
1662 		dev_dbg(dev, "%s: Might need to read more VPD (%d > %ld)\n",
1663 			__func__, (i + j), vpd_size);
1664 		ro_size = vpd_size - i;
1665 	}
1666 
1667 	/*
1668 	 * Find the offset of the WWPN tag within the read only
1669 	 * VPD data and validate the found field (partials are
1670 	 * no good to us). Convert the ASCII data to an integer
1671 	 * value. Note that we must copy to a temporary buffer
1672 	 * because the conversion service requires that the ASCII
1673 	 * string be terminated.
1674 	 */
1675 	for (k = 0; k < cfg->num_fc_ports; k++) {
1676 		j = ro_size;
1677 		i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1678 
1679 		i = pci_vpd_find_info_keyword(vpd_data, i, j, wwpn_vpd_tags[k]);
1680 		if (unlikely(i < 0)) {
1681 			dev_err(dev, "%s: Port %d WWPN not found in VPD\n",
1682 				__func__, k);
1683 			rc = -ENODEV;
1684 			goto out;
1685 		}
1686 
1687 		j = pci_vpd_info_field_size(&vpd_data[i]);
1688 		i += PCI_VPD_INFO_FLD_HDR_SIZE;
1689 		if (unlikely((i + j > vpd_size) || (j != WWPN_LEN))) {
1690 			dev_err(dev, "%s: Port %d WWPN incomplete or bad VPD\n",
1691 				__func__, k);
1692 			rc = -ENODEV;
1693 			goto out;
1694 		}
1695 
1696 		memcpy(tmp_buf, &vpd_data[i], WWPN_LEN);
1697 		rc = kstrtoul(tmp_buf, WWPN_LEN, (ulong *)&wwpn[k]);
1698 		if (unlikely(rc)) {
1699 			dev_err(dev, "%s: WWPN conversion failed for port %d\n",
1700 				__func__, k);
1701 			rc = -ENODEV;
1702 			goto out;
1703 		}
1704 
1705 		dev_dbg(dev, "%s: wwpn%d=%016llx\n", __func__, k, wwpn[k]);
1706 	}
1707 
1708 out:
1709 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1710 	return rc;
1711 }
1712 
1713 /**
1714  * init_pcr() - initialize the provisioning and control registers
1715  * @cfg:	Internal structure associated with the host.
1716  *
1717  * Also sets up fast access to the mapped registers and initializes AFU
1718  * command fields that never change.
1719  */
1720 static void init_pcr(struct cxlflash_cfg *cfg)
1721 {
1722 	struct afu *afu = cfg->afu;
1723 	struct sisl_ctrl_map __iomem *ctrl_map;
1724 	struct hwq *hwq;
1725 	int i;
1726 
1727 	for (i = 0; i < MAX_CONTEXT; i++) {
1728 		ctrl_map = &afu->afu_map->ctrls[i].ctrl;
1729 		/* Disrupt any clients that could be running */
1730 		/* e.g. clients that survived a master restart */
1731 		writeq_be(0, &ctrl_map->rht_start);
1732 		writeq_be(0, &ctrl_map->rht_cnt_id);
1733 		writeq_be(0, &ctrl_map->ctx_cap);
1734 	}
1735 
1736 	/* Copy frequently used fields into hwq */
1737 	for (i = 0; i < afu->num_hwqs; i++) {
1738 		hwq = get_hwq(afu, i);
1739 
1740 		hwq->ctx_hndl = (u16) cxl_process_element(hwq->ctx);
1741 		hwq->host_map = &afu->afu_map->hosts[hwq->ctx_hndl].host;
1742 		hwq->ctrl_map = &afu->afu_map->ctrls[hwq->ctx_hndl].ctrl;
1743 
1744 		/* Program the Endian Control for the master context */
1745 		writeq_be(SISL_ENDIAN_CTRL, &hwq->host_map->endian_ctrl);
1746 	}
1747 }
1748 
1749 /**
1750  * init_global() - initialize AFU global registers
1751  * @cfg:	Internal structure associated with the host.
1752  */
1753 static int init_global(struct cxlflash_cfg *cfg)
1754 {
1755 	struct afu *afu = cfg->afu;
1756 	struct device *dev = &cfg->dev->dev;
1757 	struct hwq *hwq;
1758 	struct sisl_host_map __iomem *hmap;
1759 	__be64 __iomem *fc_port_regs;
1760 	u64 wwpn[MAX_FC_PORTS];	/* wwpn of AFU ports */
1761 	int i = 0, num_ports = 0;
1762 	int rc = 0;
1763 	u64 reg;
1764 
1765 	rc = read_vpd(cfg, &wwpn[0]);
1766 	if (rc) {
1767 		dev_err(dev, "%s: could not read vpd rc=%d\n", __func__, rc);
1768 		goto out;
1769 	}
1770 
1771 	/* Set up RRQ and SQ in HWQ for master issued cmds */
1772 	for (i = 0; i < afu->num_hwqs; i++) {
1773 		hwq = get_hwq(afu, i);
1774 		hmap = hwq->host_map;
1775 
1776 		writeq_be((u64) hwq->hrrq_start, &hmap->rrq_start);
1777 		writeq_be((u64) hwq->hrrq_end, &hmap->rrq_end);
1778 
1779 		if (afu_is_sq_cmd_mode(afu)) {
1780 			writeq_be((u64)hwq->hsq_start, &hmap->sq_start);
1781 			writeq_be((u64)hwq->hsq_end, &hmap->sq_end);
1782 		}
1783 	}
1784 
1785 	/* AFU configuration */
1786 	reg = readq_be(&afu->afu_map->global.regs.afu_config);
1787 	reg |= SISL_AFUCONF_AR_ALL|SISL_AFUCONF_ENDIAN;
1788 	/* enable all auto retry options and control endianness */
1789 	/* leave others at default: */
1790 	/* CTX_CAP write protected, mbox_r does not clear on read and */
1791 	/* checker on if dual afu */
1792 	writeq_be(reg, &afu->afu_map->global.regs.afu_config);
1793 
1794 	/* Global port select: select either port */
1795 	if (afu->internal_lun) {
1796 		/* Only use port 0 */
1797 		writeq_be(PORT0, &afu->afu_map->global.regs.afu_port_sel);
1798 		num_ports = 0;
1799 	} else {
1800 		writeq_be(PORT_MASK(cfg->num_fc_ports),
1801 			  &afu->afu_map->global.regs.afu_port_sel);
1802 		num_ports = cfg->num_fc_ports;
1803 	}
1804 
1805 	for (i = 0; i < num_ports; i++) {
1806 		fc_port_regs = get_fc_port_regs(cfg, i);
1807 
1808 		/* Unmask all errors (but they are still masked at AFU) */
1809 		writeq_be(0, &fc_port_regs[FC_ERRMSK / 8]);
1810 		/* Clear CRC error cnt & set a threshold */
1811 		(void)readq_be(&fc_port_regs[FC_CNT_CRCERR / 8]);
1812 		writeq_be(MC_CRC_THRESH, &fc_port_regs[FC_CRC_THRESH / 8]);
1813 
1814 		/* Set WWPNs. If already programmed, wwpn[i] is 0 */
1815 		if (wwpn[i] != 0)
1816 			afu_set_wwpn(afu, i, &fc_port_regs[0], wwpn[i]);
1817 		/* Programming WWPN back to back causes additional
1818 		 * offline/online transitions and a PLOGI
1819 		 */
1820 		msleep(100);
1821 	}
1822 
1823 	/* Set up master's own CTX_CAP to allow real mode, host translation */
1824 	/* tables, afu cmds and read/write GSCSI cmds. */
1825 	/* First, unlock ctx_cap write by reading mbox */
1826 	for (i = 0; i < afu->num_hwqs; i++) {
1827 		hwq = get_hwq(afu, i);
1828 
1829 		(void)readq_be(&hwq->ctrl_map->mbox_r);	/* unlock ctx_cap */
1830 		writeq_be((SISL_CTX_CAP_REAL_MODE | SISL_CTX_CAP_HOST_XLATE |
1831 			SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD |
1832 			SISL_CTX_CAP_AFU_CMD | SISL_CTX_CAP_GSCSI_CMD),
1833 			&hwq->ctrl_map->ctx_cap);
1834 	}
1835 
1836 	/*
1837 	 * Determine write-same unmap support for host by evaluating the unmap
1838 	 * sector support bit of the context control register associated with
1839 	 * the primary hardware queue. Note that while this status is reflected
1840 	 * in a context register, the outcome can be assumed to be host-wide.
1841 	 */
1842 	hwq = get_hwq(afu, PRIMARY_HWQ);
1843 	reg = readq_be(&hwq->host_map->ctx_ctrl);
1844 	if (reg & SISL_CTX_CTRL_UNMAP_SECTOR)
1845 		cfg->ws_unmap = true;
1846 
1847 	/* Initialize heartbeat */
1848 	afu->hb = readq_be(&afu->afu_map->global.regs.afu_hb);
1849 out:
1850 	return rc;
1851 }
1852 
1853 /**
1854  * start_afu() - initializes and starts the AFU
1855  * @cfg:	Internal structure associated with the host.
1856  */
1857 static int start_afu(struct cxlflash_cfg *cfg)
1858 {
1859 	struct afu *afu = cfg->afu;
1860 	struct device *dev = &cfg->dev->dev;
1861 	struct hwq *hwq;
1862 	int rc = 0;
1863 	int i;
1864 
1865 	init_pcr(cfg);
1866 
1867 	/* Initialize each HWQ */
1868 	for (i = 0; i < afu->num_hwqs; i++) {
1869 		hwq = get_hwq(afu, i);
1870 
1871 		/* After an AFU reset, RRQ entries are stale, clear them */
1872 		memset(&hwq->rrq_entry, 0, sizeof(hwq->rrq_entry));
1873 
1874 		/* Initialize RRQ pointers */
1875 		hwq->hrrq_start = &hwq->rrq_entry[0];
1876 		hwq->hrrq_end = &hwq->rrq_entry[NUM_RRQ_ENTRY - 1];
1877 		hwq->hrrq_curr = hwq->hrrq_start;
1878 		hwq->toggle = 1;
1879 
1880 		/* Initialize spin locks */
1881 		spin_lock_init(&hwq->hrrq_slock);
1882 		spin_lock_init(&hwq->hsq_slock);
1883 
1884 		/* Initialize SQ */
1885 		if (afu_is_sq_cmd_mode(afu)) {
1886 			memset(&hwq->sq, 0, sizeof(hwq->sq));
1887 			hwq->hsq_start = &hwq->sq[0];
1888 			hwq->hsq_end = &hwq->sq[NUM_SQ_ENTRY - 1];
1889 			hwq->hsq_curr = hwq->hsq_start;
1890 
1891 			atomic_set(&hwq->hsq_credits, NUM_SQ_ENTRY - 1);
1892 		}
1893 
1894 		/* Initialize IRQ poll */
1895 		if (afu_is_irqpoll_enabled(afu))
1896 			irq_poll_init(&hwq->irqpoll, afu->irqpoll_weight,
1897 				      cxlflash_irqpoll);
1898 
1899 	}
1900 
1901 	rc = init_global(cfg);
1902 
1903 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1904 	return rc;
1905 }
1906 
1907 /**
1908  * init_intr() - setup interrupt handlers for the master context
1909  * @cfg:	Internal structure associated with the host.
1910  * @hwq:	Hardware queue to initialize.
1911  *
1912  * Return: 0 on success, -errno on failure
1913  */
1914 static enum undo_level init_intr(struct cxlflash_cfg *cfg,
1915 				 struct hwq *hwq)
1916 {
1917 	struct device *dev = &cfg->dev->dev;
1918 	struct cxl_context *ctx = hwq->ctx;
1919 	int rc = 0;
1920 	enum undo_level level = UNDO_NOOP;
1921 	bool is_primary_hwq = (hwq->index == PRIMARY_HWQ);
1922 	int num_irqs = is_primary_hwq ? 3 : 2;
1923 
1924 	rc = cxl_allocate_afu_irqs(ctx, num_irqs);
1925 	if (unlikely(rc)) {
1926 		dev_err(dev, "%s: allocate_afu_irqs failed rc=%d\n",
1927 			__func__, rc);
1928 		level = UNDO_NOOP;
1929 		goto out;
1930 	}
1931 
1932 	rc = cxl_map_afu_irq(ctx, 1, cxlflash_sync_err_irq, hwq,
1933 			     "SISL_MSI_SYNC_ERROR");
1934 	if (unlikely(rc <= 0)) {
1935 		dev_err(dev, "%s: SISL_MSI_SYNC_ERROR map failed\n", __func__);
1936 		level = FREE_IRQ;
1937 		goto out;
1938 	}
1939 
1940 	rc = cxl_map_afu_irq(ctx, 2, cxlflash_rrq_irq, hwq,
1941 			     "SISL_MSI_RRQ_UPDATED");
1942 	if (unlikely(rc <= 0)) {
1943 		dev_err(dev, "%s: SISL_MSI_RRQ_UPDATED map failed\n", __func__);
1944 		level = UNMAP_ONE;
1945 		goto out;
1946 	}
1947 
1948 	/* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
1949 	if (!is_primary_hwq)
1950 		goto out;
1951 
1952 	rc = cxl_map_afu_irq(ctx, 3, cxlflash_async_err_irq, hwq,
1953 			     "SISL_MSI_ASYNC_ERROR");
1954 	if (unlikely(rc <= 0)) {
1955 		dev_err(dev, "%s: SISL_MSI_ASYNC_ERROR map failed\n", __func__);
1956 		level = UNMAP_TWO;
1957 		goto out;
1958 	}
1959 out:
1960 	return level;
1961 }
1962 
1963 /**
1964  * init_mc() - create and register as the master context
1965  * @cfg:	Internal structure associated with the host.
1966  * index:	HWQ Index of the master context.
1967  *
1968  * Return: 0 on success, -errno on failure
1969  */
1970 static int init_mc(struct cxlflash_cfg *cfg, u32 index)
1971 {
1972 	struct cxl_context *ctx;
1973 	struct device *dev = &cfg->dev->dev;
1974 	struct hwq *hwq = get_hwq(cfg->afu, index);
1975 	int rc = 0;
1976 	enum undo_level level;
1977 
1978 	hwq->afu = cfg->afu;
1979 	hwq->index = index;
1980 	INIT_LIST_HEAD(&hwq->pending_cmds);
1981 
1982 	if (index == PRIMARY_HWQ)
1983 		ctx = cxl_get_context(cfg->dev);
1984 	else
1985 		ctx = cxl_dev_context_init(cfg->dev);
1986 	if (unlikely(!ctx)) {
1987 		rc = -ENOMEM;
1988 		goto err1;
1989 	}
1990 
1991 	WARN_ON(hwq->ctx);
1992 	hwq->ctx = ctx;
1993 
1994 	/* Set it up as a master with the CXL */
1995 	cxl_set_master(ctx);
1996 
1997 	/* Reset AFU when initializing primary context */
1998 	if (index == PRIMARY_HWQ) {
1999 		rc = cxl_afu_reset(ctx);
2000 		if (unlikely(rc)) {
2001 			dev_err(dev, "%s: AFU reset failed rc=%d\n",
2002 				      __func__, rc);
2003 			goto err1;
2004 		}
2005 	}
2006 
2007 	level = init_intr(cfg, hwq);
2008 	if (unlikely(level)) {
2009 		dev_err(dev, "%s: interrupt init failed rc=%d\n", __func__, rc);
2010 		goto err2;
2011 	}
2012 
2013 	/* This performs the equivalent of the CXL_IOCTL_START_WORK.
2014 	 * The CXL_IOCTL_GET_PROCESS_ELEMENT is implicit in the process
2015 	 * element (pe) that is embedded in the context (ctx)
2016 	 */
2017 	rc = start_context(cfg, index);
2018 	if (unlikely(rc)) {
2019 		dev_err(dev, "%s: start context failed rc=%d\n", __func__, rc);
2020 		level = UNMAP_THREE;
2021 		goto err2;
2022 	}
2023 
2024 out:
2025 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2026 	return rc;
2027 err2:
2028 	term_intr(cfg, level, index);
2029 	if (index != PRIMARY_HWQ)
2030 		cxl_release_context(ctx);
2031 err1:
2032 	hwq->ctx = NULL;
2033 	goto out;
2034 }
2035 
2036 /**
2037  * get_num_afu_ports() - determines and configures the number of AFU ports
2038  * @cfg:	Internal structure associated with the host.
2039  *
2040  * This routine determines the number of AFU ports by converting the global
2041  * port selection mask. The converted value is only valid following an AFU
2042  * reset (explicit or power-on). This routine must be invoked shortly after
2043  * mapping as other routines are dependent on the number of ports during the
2044  * initialization sequence.
2045  *
2046  * To support legacy AFUs that might not have reflected an initial global
2047  * port mask (value read is 0), default to the number of ports originally
2048  * supported by the cxlflash driver (2) before hardware with other port
2049  * offerings was introduced.
2050  */
2051 static void get_num_afu_ports(struct cxlflash_cfg *cfg)
2052 {
2053 	struct afu *afu = cfg->afu;
2054 	struct device *dev = &cfg->dev->dev;
2055 	u64 port_mask;
2056 	int num_fc_ports = LEGACY_FC_PORTS;
2057 
2058 	port_mask = readq_be(&afu->afu_map->global.regs.afu_port_sel);
2059 	if (port_mask != 0ULL)
2060 		num_fc_ports = min(ilog2(port_mask) + 1, MAX_FC_PORTS);
2061 
2062 	dev_dbg(dev, "%s: port_mask=%016llx num_fc_ports=%d\n",
2063 		__func__, port_mask, num_fc_ports);
2064 
2065 	cfg->num_fc_ports = num_fc_ports;
2066 	cfg->host->max_channel = PORTNUM2CHAN(num_fc_ports);
2067 }
2068 
2069 /**
2070  * init_afu() - setup as master context and start AFU
2071  * @cfg:	Internal structure associated with the host.
2072  *
2073  * This routine is a higher level of control for configuring the
2074  * AFU on probe and reset paths.
2075  *
2076  * Return: 0 on success, -errno on failure
2077  */
2078 static int init_afu(struct cxlflash_cfg *cfg)
2079 {
2080 	u64 reg;
2081 	int rc = 0;
2082 	struct afu *afu = cfg->afu;
2083 	struct device *dev = &cfg->dev->dev;
2084 	struct hwq *hwq;
2085 	int i;
2086 
2087 	cxl_perst_reloads_same_image(cfg->cxl_afu, true);
2088 
2089 	afu->num_hwqs = afu->desired_hwqs;
2090 	for (i = 0; i < afu->num_hwqs; i++) {
2091 		rc = init_mc(cfg, i);
2092 		if (rc) {
2093 			dev_err(dev, "%s: init_mc failed rc=%d index=%d\n",
2094 				__func__, rc, i);
2095 			goto err1;
2096 		}
2097 	}
2098 
2099 	/* Map the entire MMIO space of the AFU using the first context */
2100 	hwq = get_hwq(afu, PRIMARY_HWQ);
2101 	afu->afu_map = cxl_psa_map(hwq->ctx);
2102 	if (!afu->afu_map) {
2103 		dev_err(dev, "%s: cxl_psa_map failed\n", __func__);
2104 		rc = -ENOMEM;
2105 		goto err1;
2106 	}
2107 
2108 	/* No byte reverse on reading afu_version or string will be backwards */
2109 	reg = readq(&afu->afu_map->global.regs.afu_version);
2110 	memcpy(afu->version, &reg, sizeof(reg));
2111 	afu->interface_version =
2112 	    readq_be(&afu->afu_map->global.regs.interface_version);
2113 	if ((afu->interface_version + 1) == 0) {
2114 		dev_err(dev, "Back level AFU, please upgrade. AFU version %s "
2115 			"interface version %016llx\n", afu->version,
2116 		       afu->interface_version);
2117 		rc = -EINVAL;
2118 		goto err1;
2119 	}
2120 
2121 	if (afu_is_sq_cmd_mode(afu)) {
2122 		afu->send_cmd = send_cmd_sq;
2123 		afu->context_reset = context_reset_sq;
2124 	} else {
2125 		afu->send_cmd = send_cmd_ioarrin;
2126 		afu->context_reset = context_reset_ioarrin;
2127 	}
2128 
2129 	dev_dbg(dev, "%s: afu_ver=%s interface_ver=%016llx\n", __func__,
2130 		afu->version, afu->interface_version);
2131 
2132 	get_num_afu_ports(cfg);
2133 
2134 	rc = start_afu(cfg);
2135 	if (rc) {
2136 		dev_err(dev, "%s: start_afu failed, rc=%d\n", __func__, rc);
2137 		goto err1;
2138 	}
2139 
2140 	afu_err_intr_init(cfg->afu);
2141 	for (i = 0; i < afu->num_hwqs; i++) {
2142 		hwq = get_hwq(afu, i);
2143 
2144 		hwq->room = readq_be(&hwq->host_map->cmd_room);
2145 	}
2146 
2147 	/* Restore the LUN mappings */
2148 	cxlflash_restore_luntable(cfg);
2149 out:
2150 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2151 	return rc;
2152 
2153 err1:
2154 	for (i = afu->num_hwqs - 1; i >= 0; i--) {
2155 		term_intr(cfg, UNMAP_THREE, i);
2156 		term_mc(cfg, i);
2157 	}
2158 	goto out;
2159 }
2160 
2161 /**
2162  * afu_reset() - resets the AFU
2163  * @cfg:	Internal structure associated with the host.
2164  *
2165  * Return: 0 on success, -errno on failure
2166  */
2167 static int afu_reset(struct cxlflash_cfg *cfg)
2168 {
2169 	struct device *dev = &cfg->dev->dev;
2170 	int rc = 0;
2171 
2172 	/* Stop the context before the reset. Since the context is
2173 	 * no longer available restart it after the reset is complete
2174 	 */
2175 	term_afu(cfg);
2176 
2177 	rc = init_afu(cfg);
2178 
2179 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2180 	return rc;
2181 }
2182 
2183 /**
2184  * drain_ioctls() - wait until all currently executing ioctls have completed
2185  * @cfg:	Internal structure associated with the host.
2186  *
2187  * Obtain write access to read/write semaphore that wraps ioctl
2188  * handling to 'drain' ioctls currently executing.
2189  */
2190 static void drain_ioctls(struct cxlflash_cfg *cfg)
2191 {
2192 	down_write(&cfg->ioctl_rwsem);
2193 	up_write(&cfg->ioctl_rwsem);
2194 }
2195 
2196 /**
2197  * cxlflash_async_reset_host() - asynchronous host reset handler
2198  * @data:	Private data provided while scheduling reset.
2199  * @cookie:	Cookie that can be used for checkpointing.
2200  */
2201 static void cxlflash_async_reset_host(void *data, async_cookie_t cookie)
2202 {
2203 	struct cxlflash_cfg *cfg = data;
2204 	struct device *dev = &cfg->dev->dev;
2205 	int rc = 0;
2206 
2207 	if (cfg->state != STATE_RESET) {
2208 		dev_dbg(dev, "%s: Not performing a reset, state=%d\n",
2209 			__func__, cfg->state);
2210 		goto out;
2211 	}
2212 
2213 	drain_ioctls(cfg);
2214 	cxlflash_mark_contexts_error(cfg);
2215 	rc = afu_reset(cfg);
2216 	if (rc)
2217 		cfg->state = STATE_FAILTERM;
2218 	else
2219 		cfg->state = STATE_NORMAL;
2220 	wake_up_all(&cfg->reset_waitq);
2221 
2222 out:
2223 	scsi_unblock_requests(cfg->host);
2224 }
2225 
2226 /**
2227  * cxlflash_schedule_async_reset() - schedule an asynchronous host reset
2228  * @cfg:	Internal structure associated with the host.
2229  */
2230 static void cxlflash_schedule_async_reset(struct cxlflash_cfg *cfg)
2231 {
2232 	struct device *dev = &cfg->dev->dev;
2233 
2234 	if (cfg->state != STATE_NORMAL) {
2235 		dev_dbg(dev, "%s: Not performing reset state=%d\n",
2236 			__func__, cfg->state);
2237 		return;
2238 	}
2239 
2240 	cfg->state = STATE_RESET;
2241 	scsi_block_requests(cfg->host);
2242 	cfg->async_reset_cookie = async_schedule(cxlflash_async_reset_host,
2243 						 cfg);
2244 }
2245 
2246 /**
2247  * send_afu_cmd() - builds and sends an internal AFU command
2248  * @afu:	AFU associated with the host.
2249  * @rcb:	Pre-populated IOARCB describing command to send.
2250  *
2251  * The AFU can only take one internal AFU command at a time. This limitation is
2252  * enforced by using a mutex to provide exclusive access to the AFU during the
2253  * operation. This design point requires calling threads to not be on interrupt
2254  * context due to the possibility of sleeping during concurrent AFU operations.
2255  *
2256  * The command status is optionally passed back to the caller when the caller
2257  * populates the IOASA field of the IOARCB with a pointer to an IOASA structure.
2258  *
2259  * Return:
2260  *	0 on success, -errno on failure
2261  */
2262 static int send_afu_cmd(struct afu *afu, struct sisl_ioarcb *rcb)
2263 {
2264 	struct cxlflash_cfg *cfg = afu->parent;
2265 	struct device *dev = &cfg->dev->dev;
2266 	struct afu_cmd *cmd = NULL;
2267 	struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
2268 	char *buf = NULL;
2269 	int rc = 0;
2270 	int nretry = 0;
2271 	static DEFINE_MUTEX(sync_active);
2272 
2273 	if (cfg->state != STATE_NORMAL) {
2274 		dev_dbg(dev, "%s: Sync not required state=%u\n",
2275 			__func__, cfg->state);
2276 		return 0;
2277 	}
2278 
2279 	mutex_lock(&sync_active);
2280 	atomic_inc(&afu->cmds_active);
2281 	buf = kmalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
2282 	if (unlikely(!buf)) {
2283 		dev_err(dev, "%s: no memory for command\n", __func__);
2284 		rc = -ENOMEM;
2285 		goto out;
2286 	}
2287 
2288 	cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
2289 
2290 retry:
2291 	memset(cmd, 0, sizeof(*cmd));
2292 	memcpy(&cmd->rcb, rcb, sizeof(*rcb));
2293 	INIT_LIST_HEAD(&cmd->queue);
2294 	init_completion(&cmd->cevent);
2295 	cmd->parent = afu;
2296 	cmd->hwq_index = hwq->index;
2297 	cmd->rcb.ctx_id = hwq->ctx_hndl;
2298 
2299 	dev_dbg(dev, "%s: afu=%p cmd=%p type=%02x nretry=%d\n",
2300 		__func__, afu, cmd, cmd->rcb.cdb[0], nretry);
2301 
2302 	rc = afu->send_cmd(afu, cmd);
2303 	if (unlikely(rc)) {
2304 		rc = -ENOBUFS;
2305 		goto out;
2306 	}
2307 
2308 	rc = wait_resp(afu, cmd);
2309 	switch (rc) {
2310 	case -ETIMEDOUT:
2311 		rc = afu->context_reset(hwq);
2312 		if (rc) {
2313 			cxlflash_schedule_async_reset(cfg);
2314 			break;
2315 		}
2316 		/* fall through to retry */
2317 	case -EAGAIN:
2318 		if (++nretry < 2)
2319 			goto retry;
2320 		/* fall through to exit */
2321 	default:
2322 		break;
2323 	}
2324 
2325 	if (rcb->ioasa)
2326 		*rcb->ioasa = cmd->sa;
2327 out:
2328 	atomic_dec(&afu->cmds_active);
2329 	mutex_unlock(&sync_active);
2330 	kfree(buf);
2331 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2332 	return rc;
2333 }
2334 
2335 /**
2336  * cxlflash_afu_sync() - builds and sends an AFU sync command
2337  * @afu:	AFU associated with the host.
2338  * @ctx:	Identifies context requesting sync.
2339  * @res:	Identifies resource requesting sync.
2340  * @mode:	Type of sync to issue (lightweight, heavyweight, global).
2341  *
2342  * AFU sync operations are only necessary and allowed when the device is
2343  * operating normally. When not operating normally, sync requests can occur as
2344  * part of cleaning up resources associated with an adapter prior to removal.
2345  * In this scenario, these requests are simply ignored (safe due to the AFU
2346  * going away).
2347  *
2348  * Return:
2349  *	0 on success, -errno on failure
2350  */
2351 int cxlflash_afu_sync(struct afu *afu, ctx_hndl_t ctx, res_hndl_t res, u8 mode)
2352 {
2353 	struct cxlflash_cfg *cfg = afu->parent;
2354 	struct device *dev = &cfg->dev->dev;
2355 	struct sisl_ioarcb rcb = { 0 };
2356 
2357 	dev_dbg(dev, "%s: afu=%p ctx=%u res=%u mode=%u\n",
2358 		__func__, afu, ctx, res, mode);
2359 
2360 	rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
2361 	rcb.msi = SISL_MSI_RRQ_UPDATED;
2362 	rcb.timeout = MC_AFU_SYNC_TIMEOUT;
2363 
2364 	rcb.cdb[0] = SISL_AFU_CMD_SYNC;
2365 	rcb.cdb[1] = mode;
2366 	put_unaligned_be16(ctx, &rcb.cdb[2]);
2367 	put_unaligned_be32(res, &rcb.cdb[4]);
2368 
2369 	return send_afu_cmd(afu, &rcb);
2370 }
2371 
2372 /**
2373  * cxlflash_eh_abort_handler() - abort a SCSI command
2374  * @scp:	SCSI command to abort.
2375  *
2376  * CXL Flash devices do not support a single command abort. Reset the context
2377  * as per SISLite specification. Flush any pending commands in the hardware
2378  * queue before the reset.
2379  *
2380  * Return: SUCCESS/FAILED as defined in scsi/scsi.h
2381  */
2382 static int cxlflash_eh_abort_handler(struct scsi_cmnd *scp)
2383 {
2384 	int rc = FAILED;
2385 	struct Scsi_Host *host = scp->device->host;
2386 	struct cxlflash_cfg *cfg = shost_priv(host);
2387 	struct afu_cmd *cmd = sc_to_afuc(scp);
2388 	struct device *dev = &cfg->dev->dev;
2389 	struct afu *afu = cfg->afu;
2390 	struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
2391 
2392 	dev_dbg(dev, "%s: (scp=%p) %d/%d/%d/%llu "
2393 		"cdb=(%08x-%08x-%08x-%08x)\n", __func__, scp, host->host_no,
2394 		scp->device->channel, scp->device->id, scp->device->lun,
2395 		get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
2396 		get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
2397 		get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
2398 		get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
2399 
2400 	/* When the state is not normal, another reset/reload is in progress.
2401 	 * Return failed and the mid-layer will invoke host reset handler.
2402 	 */
2403 	if (cfg->state != STATE_NORMAL) {
2404 		dev_dbg(dev, "%s: Invalid state for abort, state=%d\n",
2405 			__func__, cfg->state);
2406 		goto out;
2407 	}
2408 
2409 	rc = afu->context_reset(hwq);
2410 	if (unlikely(rc))
2411 		goto out;
2412 
2413 	rc = SUCCESS;
2414 
2415 out:
2416 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2417 	return rc;
2418 }
2419 
2420 /**
2421  * cxlflash_eh_device_reset_handler() - reset a single LUN
2422  * @scp:	SCSI command to send.
2423  *
2424  * Return:
2425  *	SUCCESS as defined in scsi/scsi.h
2426  *	FAILED as defined in scsi/scsi.h
2427  */
2428 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
2429 {
2430 	int rc = SUCCESS;
2431 	struct scsi_device *sdev = scp->device;
2432 	struct Scsi_Host *host = sdev->host;
2433 	struct cxlflash_cfg *cfg = shost_priv(host);
2434 	struct device *dev = &cfg->dev->dev;
2435 	int rcr = 0;
2436 
2437 	dev_dbg(dev, "%s: %d/%d/%d/%llu\n", __func__,
2438 		host->host_no, sdev->channel, sdev->id, sdev->lun);
2439 retry:
2440 	switch (cfg->state) {
2441 	case STATE_NORMAL:
2442 		rcr = send_tmf(cfg, sdev, TMF_LUN_RESET);
2443 		if (unlikely(rcr))
2444 			rc = FAILED;
2445 		break;
2446 	case STATE_RESET:
2447 		wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2448 		goto retry;
2449 	default:
2450 		rc = FAILED;
2451 		break;
2452 	}
2453 
2454 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2455 	return rc;
2456 }
2457 
2458 /**
2459  * cxlflash_eh_host_reset_handler() - reset the host adapter
2460  * @scp:	SCSI command from stack identifying host.
2461  *
2462  * Following a reset, the state is evaluated again in case an EEH occurred
2463  * during the reset. In such a scenario, the host reset will either yield
2464  * until the EEH recovery is complete or return success or failure based
2465  * upon the current device state.
2466  *
2467  * Return:
2468  *	SUCCESS as defined in scsi/scsi.h
2469  *	FAILED as defined in scsi/scsi.h
2470  */
2471 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
2472 {
2473 	int rc = SUCCESS;
2474 	int rcr = 0;
2475 	struct Scsi_Host *host = scp->device->host;
2476 	struct cxlflash_cfg *cfg = shost_priv(host);
2477 	struct device *dev = &cfg->dev->dev;
2478 
2479 	dev_dbg(dev, "%s: %d\n", __func__, host->host_no);
2480 
2481 	switch (cfg->state) {
2482 	case STATE_NORMAL:
2483 		cfg->state = STATE_RESET;
2484 		drain_ioctls(cfg);
2485 		cxlflash_mark_contexts_error(cfg);
2486 		rcr = afu_reset(cfg);
2487 		if (rcr) {
2488 			rc = FAILED;
2489 			cfg->state = STATE_FAILTERM;
2490 		} else
2491 			cfg->state = STATE_NORMAL;
2492 		wake_up_all(&cfg->reset_waitq);
2493 		ssleep(1);
2494 		/* fall through */
2495 	case STATE_RESET:
2496 		wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2497 		if (cfg->state == STATE_NORMAL)
2498 			break;
2499 		/* fall through */
2500 	default:
2501 		rc = FAILED;
2502 		break;
2503 	}
2504 
2505 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2506 	return rc;
2507 }
2508 
2509 /**
2510  * cxlflash_change_queue_depth() - change the queue depth for the device
2511  * @sdev:	SCSI device destined for queue depth change.
2512  * @qdepth:	Requested queue depth value to set.
2513  *
2514  * The requested queue depth is capped to the maximum supported value.
2515  *
2516  * Return: The actual queue depth set.
2517  */
2518 static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
2519 {
2520 
2521 	if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
2522 		qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
2523 
2524 	scsi_change_queue_depth(sdev, qdepth);
2525 	return sdev->queue_depth;
2526 }
2527 
2528 /**
2529  * cxlflash_show_port_status() - queries and presents the current port status
2530  * @port:	Desired port for status reporting.
2531  * @cfg:	Internal structure associated with the host.
2532  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2533  *
2534  * Return: The size of the ASCII string returned in @buf or -EINVAL.
2535  */
2536 static ssize_t cxlflash_show_port_status(u32 port,
2537 					 struct cxlflash_cfg *cfg,
2538 					 char *buf)
2539 {
2540 	struct device *dev = &cfg->dev->dev;
2541 	char *disp_status;
2542 	u64 status;
2543 	__be64 __iomem *fc_port_regs;
2544 
2545 	WARN_ON(port >= MAX_FC_PORTS);
2546 
2547 	if (port >= cfg->num_fc_ports) {
2548 		dev_info(dev, "%s: Port %d not supported on this card.\n",
2549 			__func__, port);
2550 		return -EINVAL;
2551 	}
2552 
2553 	fc_port_regs = get_fc_port_regs(cfg, port);
2554 	status = readq_be(&fc_port_regs[FC_MTIP_STATUS / 8]);
2555 	status &= FC_MTIP_STATUS_MASK;
2556 
2557 	if (status == FC_MTIP_STATUS_ONLINE)
2558 		disp_status = "online";
2559 	else if (status == FC_MTIP_STATUS_OFFLINE)
2560 		disp_status = "offline";
2561 	else
2562 		disp_status = "unknown";
2563 
2564 	return scnprintf(buf, PAGE_SIZE, "%s\n", disp_status);
2565 }
2566 
2567 /**
2568  * port0_show() - queries and presents the current status of port 0
2569  * @dev:	Generic device associated with the host owning the port.
2570  * @attr:	Device attribute representing the port.
2571  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2572  *
2573  * Return: The size of the ASCII string returned in @buf.
2574  */
2575 static ssize_t port0_show(struct device *dev,
2576 			  struct device_attribute *attr,
2577 			  char *buf)
2578 {
2579 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2580 
2581 	return cxlflash_show_port_status(0, cfg, buf);
2582 }
2583 
2584 /**
2585  * port1_show() - queries and presents the current status of port 1
2586  * @dev:	Generic device associated with the host owning the port.
2587  * @attr:	Device attribute representing the port.
2588  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2589  *
2590  * Return: The size of the ASCII string returned in @buf.
2591  */
2592 static ssize_t port1_show(struct device *dev,
2593 			  struct device_attribute *attr,
2594 			  char *buf)
2595 {
2596 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2597 
2598 	return cxlflash_show_port_status(1, cfg, buf);
2599 }
2600 
2601 /**
2602  * port2_show() - queries and presents the current status of port 2
2603  * @dev:	Generic device associated with the host owning the port.
2604  * @attr:	Device attribute representing the port.
2605  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2606  *
2607  * Return: The size of the ASCII string returned in @buf.
2608  */
2609 static ssize_t port2_show(struct device *dev,
2610 			  struct device_attribute *attr,
2611 			  char *buf)
2612 {
2613 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2614 
2615 	return cxlflash_show_port_status(2, cfg, buf);
2616 }
2617 
2618 /**
2619  * port3_show() - queries and presents the current status of port 3
2620  * @dev:	Generic device associated with the host owning the port.
2621  * @attr:	Device attribute representing the port.
2622  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2623  *
2624  * Return: The size of the ASCII string returned in @buf.
2625  */
2626 static ssize_t port3_show(struct device *dev,
2627 			  struct device_attribute *attr,
2628 			  char *buf)
2629 {
2630 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2631 
2632 	return cxlflash_show_port_status(3, cfg, buf);
2633 }
2634 
2635 /**
2636  * lun_mode_show() - presents the current LUN mode of the host
2637  * @dev:	Generic device associated with the host.
2638  * @attr:	Device attribute representing the LUN mode.
2639  * @buf:	Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2640  *
2641  * Return: The size of the ASCII string returned in @buf.
2642  */
2643 static ssize_t lun_mode_show(struct device *dev,
2644 			     struct device_attribute *attr, char *buf)
2645 {
2646 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2647 	struct afu *afu = cfg->afu;
2648 
2649 	return scnprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
2650 }
2651 
2652 /**
2653  * lun_mode_store() - sets the LUN mode of the host
2654  * @dev:	Generic device associated with the host.
2655  * @attr:	Device attribute representing the LUN mode.
2656  * @buf:	Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2657  * @count:	Length of data resizing in @buf.
2658  *
2659  * The CXL Flash AFU supports a dummy LUN mode where the external
2660  * links and storage are not required. Space on the FPGA is used
2661  * to create 1 or 2 small LUNs which are presented to the system
2662  * as if they were a normal storage device. This feature is useful
2663  * during development and also provides manufacturing with a way
2664  * to test the AFU without an actual device.
2665  *
2666  * 0 = external LUN[s] (default)
2667  * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2668  * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2669  * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2670  * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2671  *
2672  * Return: The size of the ASCII string returned in @buf.
2673  */
2674 static ssize_t lun_mode_store(struct device *dev,
2675 			      struct device_attribute *attr,
2676 			      const char *buf, size_t count)
2677 {
2678 	struct Scsi_Host *shost = class_to_shost(dev);
2679 	struct cxlflash_cfg *cfg = shost_priv(shost);
2680 	struct afu *afu = cfg->afu;
2681 	int rc;
2682 	u32 lun_mode;
2683 
2684 	rc = kstrtouint(buf, 10, &lun_mode);
2685 	if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
2686 		afu->internal_lun = lun_mode;
2687 
2688 		/*
2689 		 * When configured for internal LUN, there is only one channel,
2690 		 * channel number 0, else there will be one less than the number
2691 		 * of fc ports for this card.
2692 		 */
2693 		if (afu->internal_lun)
2694 			shost->max_channel = 0;
2695 		else
2696 			shost->max_channel = PORTNUM2CHAN(cfg->num_fc_ports);
2697 
2698 		afu_reset(cfg);
2699 		scsi_scan_host(cfg->host);
2700 	}
2701 
2702 	return count;
2703 }
2704 
2705 /**
2706  * ioctl_version_show() - presents the current ioctl version of the host
2707  * @dev:	Generic device associated with the host.
2708  * @attr:	Device attribute representing the ioctl version.
2709  * @buf:	Buffer of length PAGE_SIZE to report back the ioctl version.
2710  *
2711  * Return: The size of the ASCII string returned in @buf.
2712  */
2713 static ssize_t ioctl_version_show(struct device *dev,
2714 				  struct device_attribute *attr, char *buf)
2715 {
2716 	ssize_t bytes = 0;
2717 
2718 	bytes = scnprintf(buf, PAGE_SIZE,
2719 			  "disk: %u\n", DK_CXLFLASH_VERSION_0);
2720 	bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2721 			   "host: %u\n", HT_CXLFLASH_VERSION_0);
2722 
2723 	return bytes;
2724 }
2725 
2726 /**
2727  * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2728  * @port:	Desired port for status reporting.
2729  * @cfg:	Internal structure associated with the host.
2730  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2731  *
2732  * Return: The size of the ASCII string returned in @buf or -EINVAL.
2733  */
2734 static ssize_t cxlflash_show_port_lun_table(u32 port,
2735 					    struct cxlflash_cfg *cfg,
2736 					    char *buf)
2737 {
2738 	struct device *dev = &cfg->dev->dev;
2739 	__be64 __iomem *fc_port_luns;
2740 	int i;
2741 	ssize_t bytes = 0;
2742 
2743 	WARN_ON(port >= MAX_FC_PORTS);
2744 
2745 	if (port >= cfg->num_fc_ports) {
2746 		dev_info(dev, "%s: Port %d not supported on this card.\n",
2747 			__func__, port);
2748 		return -EINVAL;
2749 	}
2750 
2751 	fc_port_luns = get_fc_port_luns(cfg, port);
2752 
2753 	for (i = 0; i < CXLFLASH_NUM_VLUNS; i++)
2754 		bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2755 				   "%03d: %016llx\n",
2756 				   i, readq_be(&fc_port_luns[i]));
2757 	return bytes;
2758 }
2759 
2760 /**
2761  * port0_lun_table_show() - presents the current LUN table of port 0
2762  * @dev:	Generic device associated with the host owning the port.
2763  * @attr:	Device attribute representing the port.
2764  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2765  *
2766  * Return: The size of the ASCII string returned in @buf.
2767  */
2768 static ssize_t port0_lun_table_show(struct device *dev,
2769 				    struct device_attribute *attr,
2770 				    char *buf)
2771 {
2772 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2773 
2774 	return cxlflash_show_port_lun_table(0, cfg, buf);
2775 }
2776 
2777 /**
2778  * port1_lun_table_show() - presents the current LUN table of port 1
2779  * @dev:	Generic device associated with the host owning the port.
2780  * @attr:	Device attribute representing the port.
2781  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2782  *
2783  * Return: The size of the ASCII string returned in @buf.
2784  */
2785 static ssize_t port1_lun_table_show(struct device *dev,
2786 				    struct device_attribute *attr,
2787 				    char *buf)
2788 {
2789 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2790 
2791 	return cxlflash_show_port_lun_table(1, cfg, buf);
2792 }
2793 
2794 /**
2795  * port2_lun_table_show() - presents the current LUN table of port 2
2796  * @dev:	Generic device associated with the host owning the port.
2797  * @attr:	Device attribute representing the port.
2798  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2799  *
2800  * Return: The size of the ASCII string returned in @buf.
2801  */
2802 static ssize_t port2_lun_table_show(struct device *dev,
2803 				    struct device_attribute *attr,
2804 				    char *buf)
2805 {
2806 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2807 
2808 	return cxlflash_show_port_lun_table(2, cfg, buf);
2809 }
2810 
2811 /**
2812  * port3_lun_table_show() - presents the current LUN table of port 3
2813  * @dev:	Generic device associated with the host owning the port.
2814  * @attr:	Device attribute representing the port.
2815  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2816  *
2817  * Return: The size of the ASCII string returned in @buf.
2818  */
2819 static ssize_t port3_lun_table_show(struct device *dev,
2820 				    struct device_attribute *attr,
2821 				    char *buf)
2822 {
2823 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2824 
2825 	return cxlflash_show_port_lun_table(3, cfg, buf);
2826 }
2827 
2828 /**
2829  * irqpoll_weight_show() - presents the current IRQ poll weight for the host
2830  * @dev:	Generic device associated with the host.
2831  * @attr:	Device attribute representing the IRQ poll weight.
2832  * @buf:	Buffer of length PAGE_SIZE to report back the current IRQ poll
2833  *		weight in ASCII.
2834  *
2835  * An IRQ poll weight of 0 indicates polling is disabled.
2836  *
2837  * Return: The size of the ASCII string returned in @buf.
2838  */
2839 static ssize_t irqpoll_weight_show(struct device *dev,
2840 				   struct device_attribute *attr, char *buf)
2841 {
2842 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2843 	struct afu *afu = cfg->afu;
2844 
2845 	return scnprintf(buf, PAGE_SIZE, "%u\n", afu->irqpoll_weight);
2846 }
2847 
2848 /**
2849  * irqpoll_weight_store() - sets the current IRQ poll weight for the host
2850  * @dev:	Generic device associated with the host.
2851  * @attr:	Device attribute representing the IRQ poll weight.
2852  * @buf:	Buffer of length PAGE_SIZE containing the desired IRQ poll
2853  *		weight in ASCII.
2854  * @count:	Length of data resizing in @buf.
2855  *
2856  * An IRQ poll weight of 0 indicates polling is disabled.
2857  *
2858  * Return: The size of the ASCII string returned in @buf.
2859  */
2860 static ssize_t irqpoll_weight_store(struct device *dev,
2861 				    struct device_attribute *attr,
2862 				    const char *buf, size_t count)
2863 {
2864 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2865 	struct device *cfgdev = &cfg->dev->dev;
2866 	struct afu *afu = cfg->afu;
2867 	struct hwq *hwq;
2868 	u32 weight;
2869 	int rc, i;
2870 
2871 	rc = kstrtouint(buf, 10, &weight);
2872 	if (rc)
2873 		return -EINVAL;
2874 
2875 	if (weight > 256) {
2876 		dev_info(cfgdev,
2877 			 "Invalid IRQ poll weight. It must be 256 or less.\n");
2878 		return -EINVAL;
2879 	}
2880 
2881 	if (weight == afu->irqpoll_weight) {
2882 		dev_info(cfgdev,
2883 			 "Current IRQ poll weight has the same weight.\n");
2884 		return -EINVAL;
2885 	}
2886 
2887 	if (afu_is_irqpoll_enabled(afu)) {
2888 		for (i = 0; i < afu->num_hwqs; i++) {
2889 			hwq = get_hwq(afu, i);
2890 
2891 			irq_poll_disable(&hwq->irqpoll);
2892 		}
2893 	}
2894 
2895 	afu->irqpoll_weight = weight;
2896 
2897 	if (weight > 0) {
2898 		for (i = 0; i < afu->num_hwqs; i++) {
2899 			hwq = get_hwq(afu, i);
2900 
2901 			irq_poll_init(&hwq->irqpoll, weight, cxlflash_irqpoll);
2902 		}
2903 	}
2904 
2905 	return count;
2906 }
2907 
2908 /**
2909  * num_hwqs_show() - presents the number of hardware queues for the host
2910  * @dev:	Generic device associated with the host.
2911  * @attr:	Device attribute representing the number of hardware queues.
2912  * @buf:	Buffer of length PAGE_SIZE to report back the number of hardware
2913  *		queues in ASCII.
2914  *
2915  * Return: The size of the ASCII string returned in @buf.
2916  */
2917 static ssize_t num_hwqs_show(struct device *dev,
2918 			     struct device_attribute *attr, char *buf)
2919 {
2920 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2921 	struct afu *afu = cfg->afu;
2922 
2923 	return scnprintf(buf, PAGE_SIZE, "%u\n", afu->num_hwqs);
2924 }
2925 
2926 /**
2927  * num_hwqs_store() - sets the number of hardware queues for the host
2928  * @dev:	Generic device associated with the host.
2929  * @attr:	Device attribute representing the number of hardware queues.
2930  * @buf:	Buffer of length PAGE_SIZE containing the number of hardware
2931  *		queues in ASCII.
2932  * @count:	Length of data resizing in @buf.
2933  *
2934  * n > 0: num_hwqs = n
2935  * n = 0: num_hwqs = num_online_cpus()
2936  * n < 0: num_online_cpus() / abs(n)
2937  *
2938  * Return: The size of the ASCII string returned in @buf.
2939  */
2940 static ssize_t num_hwqs_store(struct device *dev,
2941 			      struct device_attribute *attr,
2942 			      const char *buf, size_t count)
2943 {
2944 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2945 	struct afu *afu = cfg->afu;
2946 	int rc;
2947 	int nhwqs, num_hwqs;
2948 
2949 	rc = kstrtoint(buf, 10, &nhwqs);
2950 	if (rc)
2951 		return -EINVAL;
2952 
2953 	if (nhwqs >= 1)
2954 		num_hwqs = nhwqs;
2955 	else if (nhwqs == 0)
2956 		num_hwqs = num_online_cpus();
2957 	else
2958 		num_hwqs = num_online_cpus() / abs(nhwqs);
2959 
2960 	afu->desired_hwqs = min(num_hwqs, CXLFLASH_MAX_HWQS);
2961 	WARN_ON_ONCE(afu->desired_hwqs == 0);
2962 
2963 retry:
2964 	switch (cfg->state) {
2965 	case STATE_NORMAL:
2966 		cfg->state = STATE_RESET;
2967 		drain_ioctls(cfg);
2968 		cxlflash_mark_contexts_error(cfg);
2969 		rc = afu_reset(cfg);
2970 		if (rc)
2971 			cfg->state = STATE_FAILTERM;
2972 		else
2973 			cfg->state = STATE_NORMAL;
2974 		wake_up_all(&cfg->reset_waitq);
2975 		break;
2976 	case STATE_RESET:
2977 		wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2978 		if (cfg->state == STATE_NORMAL)
2979 			goto retry;
2980 	default:
2981 		/* Ideally should not happen */
2982 		dev_err(dev, "%s: Device is not ready, state=%d\n",
2983 			__func__, cfg->state);
2984 		break;
2985 	}
2986 
2987 	return count;
2988 }
2989 
2990 static const char *hwq_mode_name[MAX_HWQ_MODE] = { "rr", "tag", "cpu" };
2991 
2992 /**
2993  * hwq_mode_show() - presents the HWQ steering mode for the host
2994  * @dev:	Generic device associated with the host.
2995  * @attr:	Device attribute representing the HWQ steering mode.
2996  * @buf:	Buffer of length PAGE_SIZE to report back the HWQ steering mode
2997  *		as a character string.
2998  *
2999  * Return: The size of the ASCII string returned in @buf.
3000  */
3001 static ssize_t hwq_mode_show(struct device *dev,
3002 			     struct device_attribute *attr, char *buf)
3003 {
3004 	struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
3005 	struct afu *afu = cfg->afu;
3006 
3007 	return scnprintf(buf, PAGE_SIZE, "%s\n", hwq_mode_name[afu->hwq_mode]);
3008 }
3009 
3010 /**
3011  * hwq_mode_store() - sets the HWQ steering mode for the host
3012  * @dev:	Generic device associated with the host.
3013  * @attr:	Device attribute representing the HWQ steering mode.
3014  * @buf:	Buffer of length PAGE_SIZE containing the HWQ steering mode
3015  *		as a character string.
3016  * @count:	Length of data resizing in @buf.
3017  *
3018  * rr = Round-Robin
3019  * tag = Block MQ Tagging
3020  * cpu = CPU Affinity
3021  *
3022  * Return: The size of the ASCII string returned in @buf.
3023  */
3024 static ssize_t hwq_mode_store(struct device *dev,
3025 			      struct device_attribute *attr,
3026 			      const char *buf, size_t count)
3027 {
3028 	struct Scsi_Host *shost = class_to_shost(dev);
3029 	struct cxlflash_cfg *cfg = shost_priv(shost);
3030 	struct device *cfgdev = &cfg->dev->dev;
3031 	struct afu *afu = cfg->afu;
3032 	int i;
3033 	u32 mode = MAX_HWQ_MODE;
3034 
3035 	for (i = 0; i < MAX_HWQ_MODE; i++) {
3036 		if (!strncmp(hwq_mode_name[i], buf, strlen(hwq_mode_name[i]))) {
3037 			mode = i;
3038 			break;
3039 		}
3040 	}
3041 
3042 	if (mode >= MAX_HWQ_MODE) {
3043 		dev_info(cfgdev, "Invalid HWQ steering mode.\n");
3044 		return -EINVAL;
3045 	}
3046 
3047 	if ((mode == HWQ_MODE_TAG) && !shost_use_blk_mq(shost)) {
3048 		dev_info(cfgdev, "SCSI-MQ is not enabled, use a different "
3049 			 "HWQ steering mode.\n");
3050 		return -EINVAL;
3051 	}
3052 
3053 	afu->hwq_mode = mode;
3054 
3055 	return count;
3056 }
3057 
3058 /**
3059  * mode_show() - presents the current mode of the device
3060  * @dev:	Generic device associated with the device.
3061  * @attr:	Device attribute representing the device mode.
3062  * @buf:	Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
3063  *
3064  * Return: The size of the ASCII string returned in @buf.
3065  */
3066 static ssize_t mode_show(struct device *dev,
3067 			 struct device_attribute *attr, char *buf)
3068 {
3069 	struct scsi_device *sdev = to_scsi_device(dev);
3070 
3071 	return scnprintf(buf, PAGE_SIZE, "%s\n",
3072 			 sdev->hostdata ? "superpipe" : "legacy");
3073 }
3074 
3075 /*
3076  * Host attributes
3077  */
3078 static DEVICE_ATTR_RO(port0);
3079 static DEVICE_ATTR_RO(port1);
3080 static DEVICE_ATTR_RO(port2);
3081 static DEVICE_ATTR_RO(port3);
3082 static DEVICE_ATTR_RW(lun_mode);
3083 static DEVICE_ATTR_RO(ioctl_version);
3084 static DEVICE_ATTR_RO(port0_lun_table);
3085 static DEVICE_ATTR_RO(port1_lun_table);
3086 static DEVICE_ATTR_RO(port2_lun_table);
3087 static DEVICE_ATTR_RO(port3_lun_table);
3088 static DEVICE_ATTR_RW(irqpoll_weight);
3089 static DEVICE_ATTR_RW(num_hwqs);
3090 static DEVICE_ATTR_RW(hwq_mode);
3091 
3092 static struct device_attribute *cxlflash_host_attrs[] = {
3093 	&dev_attr_port0,
3094 	&dev_attr_port1,
3095 	&dev_attr_port2,
3096 	&dev_attr_port3,
3097 	&dev_attr_lun_mode,
3098 	&dev_attr_ioctl_version,
3099 	&dev_attr_port0_lun_table,
3100 	&dev_attr_port1_lun_table,
3101 	&dev_attr_port2_lun_table,
3102 	&dev_attr_port3_lun_table,
3103 	&dev_attr_irqpoll_weight,
3104 	&dev_attr_num_hwqs,
3105 	&dev_attr_hwq_mode,
3106 	NULL
3107 };
3108 
3109 /*
3110  * Device attributes
3111  */
3112 static DEVICE_ATTR_RO(mode);
3113 
3114 static struct device_attribute *cxlflash_dev_attrs[] = {
3115 	&dev_attr_mode,
3116 	NULL
3117 };
3118 
3119 /*
3120  * Host template
3121  */
3122 static struct scsi_host_template driver_template = {
3123 	.module = THIS_MODULE,
3124 	.name = CXLFLASH_ADAPTER_NAME,
3125 	.info = cxlflash_driver_info,
3126 	.ioctl = cxlflash_ioctl,
3127 	.proc_name = CXLFLASH_NAME,
3128 	.queuecommand = cxlflash_queuecommand,
3129 	.eh_abort_handler = cxlflash_eh_abort_handler,
3130 	.eh_device_reset_handler = cxlflash_eh_device_reset_handler,
3131 	.eh_host_reset_handler = cxlflash_eh_host_reset_handler,
3132 	.change_queue_depth = cxlflash_change_queue_depth,
3133 	.cmd_per_lun = CXLFLASH_MAX_CMDS_PER_LUN,
3134 	.can_queue = CXLFLASH_MAX_CMDS,
3135 	.cmd_size = sizeof(struct afu_cmd) + __alignof__(struct afu_cmd) - 1,
3136 	.this_id = -1,
3137 	.sg_tablesize = 1,	/* No scatter gather support */
3138 	.max_sectors = CXLFLASH_MAX_SECTORS,
3139 	.use_clustering = ENABLE_CLUSTERING,
3140 	.shost_attrs = cxlflash_host_attrs,
3141 	.sdev_attrs = cxlflash_dev_attrs,
3142 };
3143 
3144 /*
3145  * Device dependent values
3146  */
3147 static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS,
3148 					0ULL };
3149 static struct dev_dependent_vals dev_flash_gt_vals = { CXLFLASH_MAX_SECTORS,
3150 					CXLFLASH_NOTIFY_SHUTDOWN };
3151 static struct dev_dependent_vals dev_briard_vals = { CXLFLASH_MAX_SECTORS,
3152 					CXLFLASH_NOTIFY_SHUTDOWN };
3153 
3154 /*
3155  * PCI device binding table
3156  */
3157 static struct pci_device_id cxlflash_pci_table[] = {
3158 	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
3159 	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
3160 	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_FLASH_GT,
3161 	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_flash_gt_vals},
3162 	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_BRIARD,
3163 	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_briard_vals},
3164 	{}
3165 };
3166 
3167 MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
3168 
3169 /**
3170  * cxlflash_worker_thread() - work thread handler for the AFU
3171  * @work:	Work structure contained within cxlflash associated with host.
3172  *
3173  * Handles the following events:
3174  * - Link reset which cannot be performed on interrupt context due to
3175  * blocking up to a few seconds
3176  * - Rescan the host
3177  */
3178 static void cxlflash_worker_thread(struct work_struct *work)
3179 {
3180 	struct cxlflash_cfg *cfg = container_of(work, struct cxlflash_cfg,
3181 						work_q);
3182 	struct afu *afu = cfg->afu;
3183 	struct device *dev = &cfg->dev->dev;
3184 	__be64 __iomem *fc_port_regs;
3185 	int port;
3186 	ulong lock_flags;
3187 
3188 	/* Avoid MMIO if the device has failed */
3189 
3190 	if (cfg->state != STATE_NORMAL)
3191 		return;
3192 
3193 	spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3194 
3195 	if (cfg->lr_state == LINK_RESET_REQUIRED) {
3196 		port = cfg->lr_port;
3197 		if (port < 0)
3198 			dev_err(dev, "%s: invalid port index %d\n",
3199 				__func__, port);
3200 		else {
3201 			spin_unlock_irqrestore(cfg->host->host_lock,
3202 					       lock_flags);
3203 
3204 			/* The reset can block... */
3205 			fc_port_regs = get_fc_port_regs(cfg, port);
3206 			afu_link_reset(afu, port, fc_port_regs);
3207 			spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3208 		}
3209 
3210 		cfg->lr_state = LINK_RESET_COMPLETE;
3211 	}
3212 
3213 	spin_unlock_irqrestore(cfg->host->host_lock, lock_flags);
3214 
3215 	if (atomic_dec_if_positive(&cfg->scan_host_needed) >= 0)
3216 		scsi_scan_host(cfg->host);
3217 }
3218 
3219 /**
3220  * cxlflash_chr_open() - character device open handler
3221  * @inode:	Device inode associated with this character device.
3222  * @file:	File pointer for this device.
3223  *
3224  * Only users with admin privileges are allowed to open the character device.
3225  *
3226  * Return: 0 on success, -errno on failure
3227  */
3228 static int cxlflash_chr_open(struct inode *inode, struct file *file)
3229 {
3230 	struct cxlflash_cfg *cfg;
3231 
3232 	if (!capable(CAP_SYS_ADMIN))
3233 		return -EACCES;
3234 
3235 	cfg = container_of(inode->i_cdev, struct cxlflash_cfg, cdev);
3236 	file->private_data = cfg;
3237 
3238 	return 0;
3239 }
3240 
3241 /**
3242  * decode_hioctl() - translates encoded host ioctl to easily identifiable string
3243  * @cmd:        The host ioctl command to decode.
3244  *
3245  * Return: A string identifying the decoded host ioctl.
3246  */
3247 static char *decode_hioctl(int cmd)
3248 {
3249 	switch (cmd) {
3250 	case HT_CXLFLASH_LUN_PROVISION:
3251 		return __stringify_1(HT_CXLFLASH_LUN_PROVISION);
3252 	}
3253 
3254 	return "UNKNOWN";
3255 }
3256 
3257 /**
3258  * cxlflash_lun_provision() - host LUN provisioning handler
3259  * @cfg:	Internal structure associated with the host.
3260  * @arg:	Kernel copy of userspace ioctl data structure.
3261  *
3262  * Return: 0 on success, -errno on failure
3263  */
3264 static int cxlflash_lun_provision(struct cxlflash_cfg *cfg,
3265 				  struct ht_cxlflash_lun_provision *lunprov)
3266 {
3267 	struct afu *afu = cfg->afu;
3268 	struct device *dev = &cfg->dev->dev;
3269 	struct sisl_ioarcb rcb;
3270 	struct sisl_ioasa asa;
3271 	__be64 __iomem *fc_port_regs;
3272 	u16 port = lunprov->port;
3273 	u16 scmd = lunprov->hdr.subcmd;
3274 	u16 type;
3275 	u64 reg;
3276 	u64 size;
3277 	u64 lun_id;
3278 	int rc = 0;
3279 
3280 	if (!afu_is_lun_provision(afu)) {
3281 		rc = -ENOTSUPP;
3282 		goto out;
3283 	}
3284 
3285 	if (port >= cfg->num_fc_ports) {
3286 		rc = -EINVAL;
3287 		goto out;
3288 	}
3289 
3290 	switch (scmd) {
3291 	case HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN:
3292 		type = SISL_AFU_LUN_PROVISION_CREATE;
3293 		size = lunprov->size;
3294 		lun_id = 0;
3295 		break;
3296 	case HT_CXLFLASH_LUN_PROVISION_SUBCMD_DELETE_LUN:
3297 		type = SISL_AFU_LUN_PROVISION_DELETE;
3298 		size = 0;
3299 		lun_id = lunprov->lun_id;
3300 		break;
3301 	case HT_CXLFLASH_LUN_PROVISION_SUBCMD_QUERY_PORT:
3302 		fc_port_regs = get_fc_port_regs(cfg, port);
3303 
3304 		reg = readq_be(&fc_port_regs[FC_MAX_NUM_LUNS / 8]);
3305 		lunprov->max_num_luns = reg;
3306 		reg = readq_be(&fc_port_regs[FC_CUR_NUM_LUNS / 8]);
3307 		lunprov->cur_num_luns = reg;
3308 		reg = readq_be(&fc_port_regs[FC_MAX_CAP_PORT / 8]);
3309 		lunprov->max_cap_port = reg;
3310 		reg = readq_be(&fc_port_regs[FC_CUR_CAP_PORT / 8]);
3311 		lunprov->cur_cap_port = reg;
3312 
3313 		goto out;
3314 	default:
3315 		rc = -EINVAL;
3316 		goto out;
3317 	}
3318 
3319 	memset(&rcb, 0, sizeof(rcb));
3320 	memset(&asa, 0, sizeof(asa));
3321 	rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
3322 	rcb.lun_id = lun_id;
3323 	rcb.msi = SISL_MSI_RRQ_UPDATED;
3324 	rcb.timeout = MC_LUN_PROV_TIMEOUT;
3325 	rcb.ioasa = &asa;
3326 
3327 	rcb.cdb[0] = SISL_AFU_CMD_LUN_PROVISION;
3328 	rcb.cdb[1] = type;
3329 	rcb.cdb[2] = port;
3330 	put_unaligned_be64(size, &rcb.cdb[8]);
3331 
3332 	rc = send_afu_cmd(afu, &rcb);
3333 	if (rc) {
3334 		dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3335 			__func__, rc, asa.ioasc, asa.afu_extra);
3336 		goto out;
3337 	}
3338 
3339 	if (scmd == HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN) {
3340 		lunprov->lun_id = (u64)asa.lunid_hi << 32 | asa.lunid_lo;
3341 		memcpy(lunprov->wwid, asa.wwid, sizeof(lunprov->wwid));
3342 	}
3343 out:
3344 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3345 	return rc;
3346 }
3347 
3348 /**
3349  * cxlflash_afu_debug() - host AFU debug handler
3350  * @cfg:	Internal structure associated with the host.
3351  * @arg:	Kernel copy of userspace ioctl data structure.
3352  *
3353  * For debug requests requiring a data buffer, always provide an aligned
3354  * (cache line) buffer to the AFU to appease any alignment requirements.
3355  *
3356  * Return: 0 on success, -errno on failure
3357  */
3358 static int cxlflash_afu_debug(struct cxlflash_cfg *cfg,
3359 			      struct ht_cxlflash_afu_debug *afu_dbg)
3360 {
3361 	struct afu *afu = cfg->afu;
3362 	struct device *dev = &cfg->dev->dev;
3363 	struct sisl_ioarcb rcb;
3364 	struct sisl_ioasa asa;
3365 	char *buf = NULL;
3366 	char *kbuf = NULL;
3367 	void __user *ubuf = (__force void __user *)afu_dbg->data_ea;
3368 	u16 req_flags = SISL_REQ_FLAGS_AFU_CMD;
3369 	u32 ulen = afu_dbg->data_len;
3370 	bool is_write = afu_dbg->hdr.flags & HT_CXLFLASH_HOST_WRITE;
3371 	int rc = 0;
3372 
3373 	if (!afu_is_afu_debug(afu)) {
3374 		rc = -ENOTSUPP;
3375 		goto out;
3376 	}
3377 
3378 	if (ulen) {
3379 		req_flags |= SISL_REQ_FLAGS_SUP_UNDERRUN;
3380 
3381 		if (ulen > HT_CXLFLASH_AFU_DEBUG_MAX_DATA_LEN) {
3382 			rc = -EINVAL;
3383 			goto out;
3384 		}
3385 
3386 		if (unlikely(!access_ok(is_write ? VERIFY_READ : VERIFY_WRITE,
3387 					ubuf, ulen))) {
3388 			rc = -EFAULT;
3389 			goto out;
3390 		}
3391 
3392 		buf = kmalloc(ulen + cache_line_size() - 1, GFP_KERNEL);
3393 		if (unlikely(!buf)) {
3394 			rc = -ENOMEM;
3395 			goto out;
3396 		}
3397 
3398 		kbuf = PTR_ALIGN(buf, cache_line_size());
3399 
3400 		if (is_write) {
3401 			req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
3402 
3403 			if (copy_from_user(kbuf, ubuf, ulen)) {
3404 				rc = -EFAULT;
3405 				goto out;
3406 			}
3407 		}
3408 	}
3409 
3410 	memset(&rcb, 0, sizeof(rcb));
3411 	memset(&asa, 0, sizeof(asa));
3412 
3413 	rcb.req_flags = req_flags;
3414 	rcb.msi = SISL_MSI_RRQ_UPDATED;
3415 	rcb.timeout = MC_AFU_DEBUG_TIMEOUT;
3416 	rcb.ioasa = &asa;
3417 
3418 	if (ulen) {
3419 		rcb.data_len = ulen;
3420 		rcb.data_ea = (uintptr_t)kbuf;
3421 	}
3422 
3423 	rcb.cdb[0] = SISL_AFU_CMD_DEBUG;
3424 	memcpy(&rcb.cdb[4], afu_dbg->afu_subcmd,
3425 	       HT_CXLFLASH_AFU_DEBUG_SUBCMD_LEN);
3426 
3427 	rc = send_afu_cmd(afu, &rcb);
3428 	if (rc) {
3429 		dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3430 			__func__, rc, asa.ioasc, asa.afu_extra);
3431 		goto out;
3432 	}
3433 
3434 	if (ulen && !is_write) {
3435 		if (copy_to_user(ubuf, kbuf, ulen))
3436 			rc = -EFAULT;
3437 	}
3438 out:
3439 	kfree(buf);
3440 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3441 	return rc;
3442 }
3443 
3444 /**
3445  * cxlflash_chr_ioctl() - character device IOCTL handler
3446  * @file:	File pointer for this device.
3447  * @cmd:	IOCTL command.
3448  * @arg:	Userspace ioctl data structure.
3449  *
3450  * A read/write semaphore is used to implement a 'drain' of currently
3451  * running ioctls. The read semaphore is taken at the beginning of each
3452  * ioctl thread and released upon concluding execution. Additionally the
3453  * semaphore should be released and then reacquired in any ioctl execution
3454  * path which will wait for an event to occur that is outside the scope of
3455  * the ioctl (i.e. an adapter reset). To drain the ioctls currently running,
3456  * a thread simply needs to acquire the write semaphore.
3457  *
3458  * Return: 0 on success, -errno on failure
3459  */
3460 static long cxlflash_chr_ioctl(struct file *file, unsigned int cmd,
3461 			       unsigned long arg)
3462 {
3463 	typedef int (*hioctl) (struct cxlflash_cfg *, void *);
3464 
3465 	struct cxlflash_cfg *cfg = file->private_data;
3466 	struct device *dev = &cfg->dev->dev;
3467 	char buf[sizeof(union cxlflash_ht_ioctls)];
3468 	void __user *uarg = (void __user *)arg;
3469 	struct ht_cxlflash_hdr *hdr;
3470 	size_t size = 0;
3471 	bool known_ioctl = false;
3472 	int idx = 0;
3473 	int rc = 0;
3474 	hioctl do_ioctl = NULL;
3475 
3476 	static const struct {
3477 		size_t size;
3478 		hioctl ioctl;
3479 	} ioctl_tbl[] = {	/* NOTE: order matters here */
3480 	{ sizeof(struct ht_cxlflash_lun_provision),
3481 		(hioctl)cxlflash_lun_provision },
3482 	{ sizeof(struct ht_cxlflash_afu_debug),
3483 		(hioctl)cxlflash_afu_debug },
3484 	};
3485 
3486 	/* Hold read semaphore so we can drain if needed */
3487 	down_read(&cfg->ioctl_rwsem);
3488 
3489 	dev_dbg(dev, "%s: cmd=%u idx=%d tbl_size=%lu\n",
3490 		__func__, cmd, idx, sizeof(ioctl_tbl));
3491 
3492 	switch (cmd) {
3493 	case HT_CXLFLASH_LUN_PROVISION:
3494 	case HT_CXLFLASH_AFU_DEBUG:
3495 		known_ioctl = true;
3496 		idx = _IOC_NR(HT_CXLFLASH_LUN_PROVISION) - _IOC_NR(cmd);
3497 		size = ioctl_tbl[idx].size;
3498 		do_ioctl = ioctl_tbl[idx].ioctl;
3499 
3500 		if (likely(do_ioctl))
3501 			break;
3502 
3503 		/* fall through */
3504 	default:
3505 		rc = -EINVAL;
3506 		goto out;
3507 	}
3508 
3509 	if (unlikely(copy_from_user(&buf, uarg, size))) {
3510 		dev_err(dev, "%s: copy_from_user() fail "
3511 			"size=%lu cmd=%d (%s) uarg=%p\n",
3512 			__func__, size, cmd, decode_hioctl(cmd), uarg);
3513 		rc = -EFAULT;
3514 		goto out;
3515 	}
3516 
3517 	hdr = (struct ht_cxlflash_hdr *)&buf;
3518 	if (hdr->version != HT_CXLFLASH_VERSION_0) {
3519 		dev_dbg(dev, "%s: Version %u not supported for %s\n",
3520 			__func__, hdr->version, decode_hioctl(cmd));
3521 		rc = -EINVAL;
3522 		goto out;
3523 	}
3524 
3525 	if (hdr->rsvd[0] || hdr->rsvd[1] || hdr->return_flags) {
3526 		dev_dbg(dev, "%s: Reserved/rflags populated\n", __func__);
3527 		rc = -EINVAL;
3528 		goto out;
3529 	}
3530 
3531 	rc = do_ioctl(cfg, (void *)&buf);
3532 	if (likely(!rc))
3533 		if (unlikely(copy_to_user(uarg, &buf, size))) {
3534 			dev_err(dev, "%s: copy_to_user() fail "
3535 				"size=%lu cmd=%d (%s) uarg=%p\n",
3536 				__func__, size, cmd, decode_hioctl(cmd), uarg);
3537 			rc = -EFAULT;
3538 		}
3539 
3540 	/* fall through to exit */
3541 
3542 out:
3543 	up_read(&cfg->ioctl_rwsem);
3544 	if (unlikely(rc && known_ioctl))
3545 		dev_err(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3546 			__func__, decode_hioctl(cmd), cmd, rc);
3547 	else
3548 		dev_dbg(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3549 			__func__, decode_hioctl(cmd), cmd, rc);
3550 	return rc;
3551 }
3552 
3553 /*
3554  * Character device file operations
3555  */
3556 static const struct file_operations cxlflash_chr_fops = {
3557 	.owner          = THIS_MODULE,
3558 	.open           = cxlflash_chr_open,
3559 	.unlocked_ioctl	= cxlflash_chr_ioctl,
3560 	.compat_ioctl	= cxlflash_chr_ioctl,
3561 };
3562 
3563 /**
3564  * init_chrdev() - initialize the character device for the host
3565  * @cfg:	Internal structure associated with the host.
3566  *
3567  * Return: 0 on success, -errno on failure
3568  */
3569 static int init_chrdev(struct cxlflash_cfg *cfg)
3570 {
3571 	struct device *dev = &cfg->dev->dev;
3572 	struct device *char_dev;
3573 	dev_t devno;
3574 	int minor;
3575 	int rc = 0;
3576 
3577 	minor = cxlflash_get_minor();
3578 	if (unlikely(minor < 0)) {
3579 		dev_err(dev, "%s: Exhausted allowed adapters\n", __func__);
3580 		rc = -ENOSPC;
3581 		goto out;
3582 	}
3583 
3584 	devno = MKDEV(cxlflash_major, minor);
3585 	cdev_init(&cfg->cdev, &cxlflash_chr_fops);
3586 
3587 	rc = cdev_add(&cfg->cdev, devno, 1);
3588 	if (rc) {
3589 		dev_err(dev, "%s: cdev_add failed rc=%d\n", __func__, rc);
3590 		goto err1;
3591 	}
3592 
3593 	char_dev = device_create(cxlflash_class, NULL, devno,
3594 				 NULL, "cxlflash%d", minor);
3595 	if (IS_ERR(char_dev)) {
3596 		rc = PTR_ERR(char_dev);
3597 		dev_err(dev, "%s: device_create failed rc=%d\n",
3598 			__func__, rc);
3599 		goto err2;
3600 	}
3601 
3602 	cfg->chardev = char_dev;
3603 out:
3604 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3605 	return rc;
3606 err2:
3607 	cdev_del(&cfg->cdev);
3608 err1:
3609 	cxlflash_put_minor(minor);
3610 	goto out;
3611 }
3612 
3613 /**
3614  * cxlflash_probe() - PCI entry point to add host
3615  * @pdev:	PCI device associated with the host.
3616  * @dev_id:	PCI device id associated with device.
3617  *
3618  * The device will initially start out in a 'probing' state and
3619  * transition to the 'normal' state at the end of a successful
3620  * probe. Should an EEH event occur during probe, the notification
3621  * thread (error_detected()) will wait until the probe handler
3622  * is nearly complete. At that time, the device will be moved to
3623  * a 'probed' state and the EEH thread woken up to drive the slot
3624  * reset and recovery (device moves to 'normal' state). Meanwhile,
3625  * the probe will be allowed to exit successfully.
3626  *
3627  * Return: 0 on success, -errno on failure
3628  */
3629 static int cxlflash_probe(struct pci_dev *pdev,
3630 			  const struct pci_device_id *dev_id)
3631 {
3632 	struct Scsi_Host *host;
3633 	struct cxlflash_cfg *cfg = NULL;
3634 	struct device *dev = &pdev->dev;
3635 	struct dev_dependent_vals *ddv;
3636 	int rc = 0;
3637 	int k;
3638 
3639 	dev_dbg(&pdev->dev, "%s: Found CXLFLASH with IRQ: %d\n",
3640 		__func__, pdev->irq);
3641 
3642 	ddv = (struct dev_dependent_vals *)dev_id->driver_data;
3643 	driver_template.max_sectors = ddv->max_sectors;
3644 
3645 	host = scsi_host_alloc(&driver_template, sizeof(struct cxlflash_cfg));
3646 	if (!host) {
3647 		dev_err(dev, "%s: scsi_host_alloc failed\n", __func__);
3648 		rc = -ENOMEM;
3649 		goto out;
3650 	}
3651 
3652 	host->max_id = CXLFLASH_MAX_NUM_TARGETS_PER_BUS;
3653 	host->max_lun = CXLFLASH_MAX_NUM_LUNS_PER_TARGET;
3654 	host->unique_id = host->host_no;
3655 	host->max_cmd_len = CXLFLASH_MAX_CDB_LEN;
3656 
3657 	cfg = shost_priv(host);
3658 	cfg->host = host;
3659 	rc = alloc_mem(cfg);
3660 	if (rc) {
3661 		dev_err(dev, "%s: alloc_mem failed\n", __func__);
3662 		rc = -ENOMEM;
3663 		scsi_host_put(cfg->host);
3664 		goto out;
3665 	}
3666 
3667 	cfg->init_state = INIT_STATE_NONE;
3668 	cfg->dev = pdev;
3669 	cfg->cxl_fops = cxlflash_cxl_fops;
3670 
3671 	/*
3672 	 * Promoted LUNs move to the top of the LUN table. The rest stay on
3673 	 * the bottom half. The bottom half grows from the end (index = 255),
3674 	 * whereas the top half grows from the beginning (index = 0).
3675 	 *
3676 	 * Initialize the last LUN index for all possible ports.
3677 	 */
3678 	cfg->promote_lun_index = 0;
3679 
3680 	for (k = 0; k < MAX_FC_PORTS; k++)
3681 		cfg->last_lun_index[k] = CXLFLASH_NUM_VLUNS/2 - 1;
3682 
3683 	cfg->dev_id = (struct pci_device_id *)dev_id;
3684 
3685 	init_waitqueue_head(&cfg->tmf_waitq);
3686 	init_waitqueue_head(&cfg->reset_waitq);
3687 
3688 	INIT_WORK(&cfg->work_q, cxlflash_worker_thread);
3689 	cfg->lr_state = LINK_RESET_INVALID;
3690 	cfg->lr_port = -1;
3691 	spin_lock_init(&cfg->tmf_slock);
3692 	mutex_init(&cfg->ctx_tbl_list_mutex);
3693 	mutex_init(&cfg->ctx_recovery_mutex);
3694 	init_rwsem(&cfg->ioctl_rwsem);
3695 	INIT_LIST_HEAD(&cfg->ctx_err_recovery);
3696 	INIT_LIST_HEAD(&cfg->lluns);
3697 
3698 	pci_set_drvdata(pdev, cfg);
3699 
3700 	cfg->cxl_afu = cxl_pci_to_afu(pdev);
3701 
3702 	rc = init_pci(cfg);
3703 	if (rc) {
3704 		dev_err(dev, "%s: init_pci failed rc=%d\n", __func__, rc);
3705 		goto out_remove;
3706 	}
3707 	cfg->init_state = INIT_STATE_PCI;
3708 
3709 	rc = init_afu(cfg);
3710 	if (rc && !wq_has_sleeper(&cfg->reset_waitq)) {
3711 		dev_err(dev, "%s: init_afu failed rc=%d\n", __func__, rc);
3712 		goto out_remove;
3713 	}
3714 	cfg->init_state = INIT_STATE_AFU;
3715 
3716 	rc = init_scsi(cfg);
3717 	if (rc) {
3718 		dev_err(dev, "%s: init_scsi failed rc=%d\n", __func__, rc);
3719 		goto out_remove;
3720 	}
3721 	cfg->init_state = INIT_STATE_SCSI;
3722 
3723 	rc = init_chrdev(cfg);
3724 	if (rc) {
3725 		dev_err(dev, "%s: init_chrdev failed rc=%d\n", __func__, rc);
3726 		goto out_remove;
3727 	}
3728 	cfg->init_state = INIT_STATE_CDEV;
3729 
3730 	if (wq_has_sleeper(&cfg->reset_waitq)) {
3731 		cfg->state = STATE_PROBED;
3732 		wake_up_all(&cfg->reset_waitq);
3733 	} else
3734 		cfg->state = STATE_NORMAL;
3735 out:
3736 	dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3737 	return rc;
3738 
3739 out_remove:
3740 	cxlflash_remove(pdev);
3741 	goto out;
3742 }
3743 
3744 /**
3745  * cxlflash_pci_error_detected() - called when a PCI error is detected
3746  * @pdev:	PCI device struct.
3747  * @state:	PCI channel state.
3748  *
3749  * When an EEH occurs during an active reset, wait until the reset is
3750  * complete and then take action based upon the device state.
3751  *
3752  * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
3753  */
3754 static pci_ers_result_t cxlflash_pci_error_detected(struct pci_dev *pdev,
3755 						    pci_channel_state_t state)
3756 {
3757 	int rc = 0;
3758 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3759 	struct device *dev = &cfg->dev->dev;
3760 
3761 	dev_dbg(dev, "%s: pdev=%p state=%u\n", __func__, pdev, state);
3762 
3763 	switch (state) {
3764 	case pci_channel_io_frozen:
3765 		wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
3766 					     cfg->state != STATE_PROBING);
3767 		if (cfg->state == STATE_FAILTERM)
3768 			return PCI_ERS_RESULT_DISCONNECT;
3769 
3770 		cfg->state = STATE_RESET;
3771 		scsi_block_requests(cfg->host);
3772 		drain_ioctls(cfg);
3773 		rc = cxlflash_mark_contexts_error(cfg);
3774 		if (unlikely(rc))
3775 			dev_err(dev, "%s: Failed to mark user contexts rc=%d\n",
3776 				__func__, rc);
3777 		term_afu(cfg);
3778 		return PCI_ERS_RESULT_NEED_RESET;
3779 	case pci_channel_io_perm_failure:
3780 		cfg->state = STATE_FAILTERM;
3781 		wake_up_all(&cfg->reset_waitq);
3782 		scsi_unblock_requests(cfg->host);
3783 		return PCI_ERS_RESULT_DISCONNECT;
3784 	default:
3785 		break;
3786 	}
3787 	return PCI_ERS_RESULT_NEED_RESET;
3788 }
3789 
3790 /**
3791  * cxlflash_pci_slot_reset() - called when PCI slot has been reset
3792  * @pdev:	PCI device struct.
3793  *
3794  * This routine is called by the pci error recovery code after the PCI
3795  * slot has been reset, just before we should resume normal operations.
3796  *
3797  * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
3798  */
3799 static pci_ers_result_t cxlflash_pci_slot_reset(struct pci_dev *pdev)
3800 {
3801 	int rc = 0;
3802 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3803 	struct device *dev = &cfg->dev->dev;
3804 
3805 	dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3806 
3807 	rc = init_afu(cfg);
3808 	if (unlikely(rc)) {
3809 		dev_err(dev, "%s: EEH recovery failed rc=%d\n", __func__, rc);
3810 		return PCI_ERS_RESULT_DISCONNECT;
3811 	}
3812 
3813 	return PCI_ERS_RESULT_RECOVERED;
3814 }
3815 
3816 /**
3817  * cxlflash_pci_resume() - called when normal operation can resume
3818  * @pdev:	PCI device struct
3819  */
3820 static void cxlflash_pci_resume(struct pci_dev *pdev)
3821 {
3822 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3823 	struct device *dev = &cfg->dev->dev;
3824 
3825 	dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3826 
3827 	cfg->state = STATE_NORMAL;
3828 	wake_up_all(&cfg->reset_waitq);
3829 	scsi_unblock_requests(cfg->host);
3830 }
3831 
3832 /**
3833  * cxlflash_devnode() - provides devtmpfs for devices in the cxlflash class
3834  * @dev:	Character device.
3835  * @mode:	Mode that can be used to verify access.
3836  *
3837  * Return: Allocated string describing the devtmpfs structure.
3838  */
3839 static char *cxlflash_devnode(struct device *dev, umode_t *mode)
3840 {
3841 	return kasprintf(GFP_KERNEL, "cxlflash/%s", dev_name(dev));
3842 }
3843 
3844 /**
3845  * cxlflash_class_init() - create character device class
3846  *
3847  * Return: 0 on success, -errno on failure
3848  */
3849 static int cxlflash_class_init(void)
3850 {
3851 	dev_t devno;
3852 	int rc = 0;
3853 
3854 	rc = alloc_chrdev_region(&devno, 0, CXLFLASH_MAX_ADAPTERS, "cxlflash");
3855 	if (unlikely(rc)) {
3856 		pr_err("%s: alloc_chrdev_region failed rc=%d\n", __func__, rc);
3857 		goto out;
3858 	}
3859 
3860 	cxlflash_major = MAJOR(devno);
3861 
3862 	cxlflash_class = class_create(THIS_MODULE, "cxlflash");
3863 	if (IS_ERR(cxlflash_class)) {
3864 		rc = PTR_ERR(cxlflash_class);
3865 		pr_err("%s: class_create failed rc=%d\n", __func__, rc);
3866 		goto err;
3867 	}
3868 
3869 	cxlflash_class->devnode = cxlflash_devnode;
3870 out:
3871 	pr_debug("%s: returning rc=%d\n", __func__, rc);
3872 	return rc;
3873 err:
3874 	unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3875 	goto out;
3876 }
3877 
3878 /**
3879  * cxlflash_class_exit() - destroy character device class
3880  */
3881 static void cxlflash_class_exit(void)
3882 {
3883 	dev_t devno = MKDEV(cxlflash_major, 0);
3884 
3885 	class_destroy(cxlflash_class);
3886 	unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3887 }
3888 
3889 static const struct pci_error_handlers cxlflash_err_handler = {
3890 	.error_detected = cxlflash_pci_error_detected,
3891 	.slot_reset = cxlflash_pci_slot_reset,
3892 	.resume = cxlflash_pci_resume,
3893 };
3894 
3895 /*
3896  * PCI device structure
3897  */
3898 static struct pci_driver cxlflash_driver = {
3899 	.name = CXLFLASH_NAME,
3900 	.id_table = cxlflash_pci_table,
3901 	.probe = cxlflash_probe,
3902 	.remove = cxlflash_remove,
3903 	.shutdown = cxlflash_remove,
3904 	.err_handler = &cxlflash_err_handler,
3905 };
3906 
3907 /**
3908  * init_cxlflash() - module entry point
3909  *
3910  * Return: 0 on success, -errno on failure
3911  */
3912 static int __init init_cxlflash(void)
3913 {
3914 	int rc;
3915 
3916 	check_sizes();
3917 	cxlflash_list_init();
3918 	rc = cxlflash_class_init();
3919 	if (unlikely(rc))
3920 		goto out;
3921 
3922 	rc = pci_register_driver(&cxlflash_driver);
3923 	if (unlikely(rc))
3924 		goto err;
3925 out:
3926 	pr_debug("%s: returning rc=%d\n", __func__, rc);
3927 	return rc;
3928 err:
3929 	cxlflash_class_exit();
3930 	goto out;
3931 }
3932 
3933 /**
3934  * exit_cxlflash() - module exit point
3935  */
3936 static void __exit exit_cxlflash(void)
3937 {
3938 	cxlflash_term_global_luns();
3939 	cxlflash_free_errpage();
3940 
3941 	pci_unregister_driver(&cxlflash_driver);
3942 	cxlflash_class_exit();
3943 }
3944 
3945 module_init(init_cxlflash);
3946 module_exit(exit_cxlflash);
3947