xref: /openbmc/linux/drivers/scsi/cxlflash/main.c (revision 4f205687)
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 /**
38  * cmd_checkout() - checks out an AFU command
39  * @afu:	AFU to checkout from.
40  *
41  * Commands are checked out in a round-robin fashion. Note that since
42  * the command pool is larger than the hardware queue, the majority of
43  * times we will only loop once or twice before getting a command. The
44  * buffer and CDB within the command are initialized (zeroed) prior to
45  * returning.
46  *
47  * Return: The checked out command or NULL when command pool is empty.
48  */
49 static struct afu_cmd *cmd_checkout(struct afu *afu)
50 {
51 	int k, dec = CXLFLASH_NUM_CMDS;
52 	struct afu_cmd *cmd;
53 
54 	while (dec--) {
55 		k = (afu->cmd_couts++ & (CXLFLASH_NUM_CMDS - 1));
56 
57 		cmd = &afu->cmd[k];
58 
59 		if (!atomic_dec_if_positive(&cmd->free)) {
60 			pr_devel("%s: returning found index=%d cmd=%p\n",
61 				 __func__, cmd->slot, cmd);
62 			memset(cmd->buf, 0, CMD_BUFSIZE);
63 			memset(cmd->rcb.cdb, 0, sizeof(cmd->rcb.cdb));
64 			return cmd;
65 		}
66 	}
67 
68 	return NULL;
69 }
70 
71 /**
72  * cmd_checkin() - checks in an AFU command
73  * @cmd:	AFU command to checkin.
74  *
75  * Safe to pass commands that have already been checked in. Several
76  * internal tracking fields are reset as part of the checkin. Note
77  * that these are intentionally reset prior to toggling the free bit
78  * to avoid clobbering values in the event that the command is checked
79  * out right away.
80  */
81 static void cmd_checkin(struct afu_cmd *cmd)
82 {
83 	cmd->rcb.scp = NULL;
84 	cmd->rcb.timeout = 0;
85 	cmd->sa.ioasc = 0;
86 	cmd->cmd_tmf = false;
87 	cmd->sa.host_use[0] = 0; /* clears both completion and retry bytes */
88 
89 	if (unlikely(atomic_inc_return(&cmd->free) != 1)) {
90 		pr_err("%s: Freeing cmd (%d) that is not in use!\n",
91 		       __func__, cmd->slot);
92 		return;
93 	}
94 
95 	pr_devel("%s: released cmd %p index=%d\n", __func__, cmd, cmd->slot);
96 }
97 
98 /**
99  * process_cmd_err() - command error handler
100  * @cmd:	AFU command that experienced the error.
101  * @scp:	SCSI command associated with the AFU command in error.
102  *
103  * Translates error bits from AFU command to SCSI command results.
104  */
105 static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
106 {
107 	struct sisl_ioarcb *ioarcb;
108 	struct sisl_ioasa *ioasa;
109 	u32 resid;
110 
111 	if (unlikely(!cmd))
112 		return;
113 
114 	ioarcb = &(cmd->rcb);
115 	ioasa = &(cmd->sa);
116 
117 	if (ioasa->rc.flags & SISL_RC_FLAGS_UNDERRUN) {
118 		resid = ioasa->resid;
119 		scsi_set_resid(scp, resid);
120 		pr_debug("%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
121 			 __func__, cmd, scp, resid);
122 	}
123 
124 	if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
125 		pr_debug("%s: cmd underrun cmd = %p scp = %p\n",
126 			 __func__, cmd, scp);
127 		scp->result = (DID_ERROR << 16);
128 	}
129 
130 	pr_debug("%s: cmd failed afu_rc=%d scsi_rc=%d fc_rc=%d "
131 		 "afu_extra=0x%X, scsi_extra=0x%X, fc_extra=0x%X\n",
132 		 __func__, ioasa->rc.afu_rc, ioasa->rc.scsi_rc,
133 		 ioasa->rc.fc_rc, ioasa->afu_extra, ioasa->scsi_extra,
134 		 ioasa->fc_extra);
135 
136 	if (ioasa->rc.scsi_rc) {
137 		/* We have a SCSI status */
138 		if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
139 			memcpy(scp->sense_buffer, ioasa->sense_data,
140 			       SISL_SENSE_DATA_LEN);
141 			scp->result = ioasa->rc.scsi_rc;
142 		} else
143 			scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
144 	}
145 
146 	/*
147 	 * We encountered an error. Set scp->result based on nature
148 	 * of error.
149 	 */
150 	if (ioasa->rc.fc_rc) {
151 		/* We have an FC status */
152 		switch (ioasa->rc.fc_rc) {
153 		case SISL_FC_RC_LINKDOWN:
154 			scp->result = (DID_REQUEUE << 16);
155 			break;
156 		case SISL_FC_RC_RESID:
157 			/* This indicates an FCP resid underrun */
158 			if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
159 				/* If the SISL_RC_FLAGS_OVERRUN flag was set,
160 				 * then we will handle this error else where.
161 				 * If not then we must handle it here.
162 				 * This is probably an AFU bug.
163 				 */
164 				scp->result = (DID_ERROR << 16);
165 			}
166 			break;
167 		case SISL_FC_RC_RESIDERR:
168 			/* Resid mismatch between adapter and device */
169 		case SISL_FC_RC_TGTABORT:
170 		case SISL_FC_RC_ABORTOK:
171 		case SISL_FC_RC_ABORTFAIL:
172 		case SISL_FC_RC_NOLOGI:
173 		case SISL_FC_RC_ABORTPEND:
174 		case SISL_FC_RC_WRABORTPEND:
175 		case SISL_FC_RC_NOEXP:
176 		case SISL_FC_RC_INUSE:
177 			scp->result = (DID_ERROR << 16);
178 			break;
179 		}
180 	}
181 
182 	if (ioasa->rc.afu_rc) {
183 		/* We have an AFU error */
184 		switch (ioasa->rc.afu_rc) {
185 		case SISL_AFU_RC_NO_CHANNELS:
186 			scp->result = (DID_NO_CONNECT << 16);
187 			break;
188 		case SISL_AFU_RC_DATA_DMA_ERR:
189 			switch (ioasa->afu_extra) {
190 			case SISL_AFU_DMA_ERR_PAGE_IN:
191 				/* Retry */
192 				scp->result = (DID_IMM_RETRY << 16);
193 				break;
194 			case SISL_AFU_DMA_ERR_INVALID_EA:
195 			default:
196 				scp->result = (DID_ERROR << 16);
197 			}
198 			break;
199 		case SISL_AFU_RC_OUT_OF_DATA_BUFS:
200 			/* Retry */
201 			scp->result = (DID_ALLOC_FAILURE << 16);
202 			break;
203 		default:
204 			scp->result = (DID_ERROR << 16);
205 		}
206 	}
207 }
208 
209 /**
210  * cmd_complete() - command completion handler
211  * @cmd:	AFU command that has completed.
212  *
213  * Prepares and submits command that has either completed or timed out to
214  * the SCSI stack. Checks AFU command back into command pool for non-internal
215  * (rcb.scp populated) commands.
216  */
217 static void cmd_complete(struct afu_cmd *cmd)
218 {
219 	struct scsi_cmnd *scp;
220 	ulong lock_flags;
221 	struct afu *afu = cmd->parent;
222 	struct cxlflash_cfg *cfg = afu->parent;
223 	bool cmd_is_tmf;
224 
225 	spin_lock_irqsave(&cmd->slock, lock_flags);
226 	cmd->sa.host_use_b[0] |= B_DONE;
227 	spin_unlock_irqrestore(&cmd->slock, lock_flags);
228 
229 	if (cmd->rcb.scp) {
230 		scp = cmd->rcb.scp;
231 		if (unlikely(cmd->sa.ioasc))
232 			process_cmd_err(cmd, scp);
233 		else
234 			scp->result = (DID_OK << 16);
235 
236 		cmd_is_tmf = cmd->cmd_tmf;
237 		cmd_checkin(cmd); /* Don't use cmd after here */
238 
239 		pr_debug_ratelimited("%s: calling scsi_done scp=%p result=%X "
240 				     "ioasc=%d\n", __func__, scp, scp->result,
241 				     cmd->sa.ioasc);
242 
243 		scsi_dma_unmap(scp);
244 		scp->scsi_done(scp);
245 
246 		if (cmd_is_tmf) {
247 			spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
248 			cfg->tmf_active = false;
249 			wake_up_all_locked(&cfg->tmf_waitq);
250 			spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
251 		}
252 	} else
253 		complete(&cmd->cevent);
254 }
255 
256 /**
257  * context_reset() - timeout handler for AFU commands
258  * @cmd:	AFU command that timed out.
259  *
260  * Sends a reset to the AFU.
261  */
262 static void context_reset(struct afu_cmd *cmd)
263 {
264 	int nretry = 0;
265 	u64 rrin = 0x1;
266 	u64 room = 0;
267 	struct afu *afu = cmd->parent;
268 	ulong lock_flags;
269 
270 	pr_debug("%s: cmd=%p\n", __func__, cmd);
271 
272 	spin_lock_irqsave(&cmd->slock, lock_flags);
273 
274 	/* Already completed? */
275 	if (cmd->sa.host_use_b[0] & B_DONE) {
276 		spin_unlock_irqrestore(&cmd->slock, lock_flags);
277 		return;
278 	}
279 
280 	cmd->sa.host_use_b[0] |= (B_DONE | B_ERROR | B_TIMEOUT);
281 	spin_unlock_irqrestore(&cmd->slock, lock_flags);
282 
283 	/*
284 	 * We really want to send this reset at all costs, so spread
285 	 * out wait time on successive retries for available room.
286 	 */
287 	do {
288 		room = readq_be(&afu->host_map->cmd_room);
289 		atomic64_set(&afu->room, room);
290 		if (room)
291 			goto write_rrin;
292 		udelay(1 << nretry);
293 	} while (nretry++ < MC_ROOM_RETRY_CNT);
294 
295 	pr_err("%s: no cmd_room to send reset\n", __func__);
296 	return;
297 
298 write_rrin:
299 	nretry = 0;
300 	writeq_be(rrin, &afu->host_map->ioarrin);
301 	do {
302 		rrin = readq_be(&afu->host_map->ioarrin);
303 		if (rrin != 0x1)
304 			break;
305 		/* Double delay each time */
306 		udelay(1 << nretry);
307 	} while (nretry++ < MC_ROOM_RETRY_CNT);
308 }
309 
310 /**
311  * send_cmd() - sends an AFU command
312  * @afu:	AFU associated with the host.
313  * @cmd:	AFU command to send.
314  *
315  * Return:
316  *	0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
317  */
318 static int send_cmd(struct afu *afu, struct afu_cmd *cmd)
319 {
320 	struct cxlflash_cfg *cfg = afu->parent;
321 	struct device *dev = &cfg->dev->dev;
322 	int nretry = 0;
323 	int rc = 0;
324 	u64 room;
325 	long newval;
326 
327 	/*
328 	 * This routine is used by critical users such an AFU sync and to
329 	 * send a task management function (TMF). Thus we want to retry a
330 	 * bit before returning an error. To avoid the performance penalty
331 	 * of MMIO, we spread the update of 'room' over multiple commands.
332 	 */
333 retry:
334 	newval = atomic64_dec_if_positive(&afu->room);
335 	if (!newval) {
336 		do {
337 			room = readq_be(&afu->host_map->cmd_room);
338 			atomic64_set(&afu->room, room);
339 			if (room)
340 				goto write_ioarrin;
341 			udelay(1 << nretry);
342 		} while (nretry++ < MC_ROOM_RETRY_CNT);
343 
344 		dev_err(dev, "%s: no cmd_room to send 0x%X\n",
345 		       __func__, cmd->rcb.cdb[0]);
346 
347 		goto no_room;
348 	} else if (unlikely(newval < 0)) {
349 		/* This should be rare. i.e. Only if two threads race and
350 		 * decrement before the MMIO read is done. In this case
351 		 * just benefit from the other thread having updated
352 		 * afu->room.
353 		 */
354 		if (nretry++ < MC_ROOM_RETRY_CNT) {
355 			udelay(1 << nretry);
356 			goto retry;
357 		}
358 
359 		goto no_room;
360 	}
361 
362 write_ioarrin:
363 	writeq_be((u64)&cmd->rcb, &afu->host_map->ioarrin);
364 out:
365 	pr_devel("%s: cmd=%p len=%d ea=%p rc=%d\n", __func__, cmd,
366 		 cmd->rcb.data_len, (void *)cmd->rcb.data_ea, rc);
367 	return rc;
368 
369 no_room:
370 	afu->read_room = true;
371 	kref_get(&cfg->afu->mapcount);
372 	schedule_work(&cfg->work_q);
373 	rc = SCSI_MLQUEUE_HOST_BUSY;
374 	goto out;
375 }
376 
377 /**
378  * wait_resp() - polls for a response or timeout to a sent AFU command
379  * @afu:	AFU associated with the host.
380  * @cmd:	AFU command that was sent.
381  */
382 static void wait_resp(struct afu *afu, struct afu_cmd *cmd)
383 {
384 	ulong timeout = msecs_to_jiffies(cmd->rcb.timeout * 2 * 1000);
385 
386 	timeout = wait_for_completion_timeout(&cmd->cevent, timeout);
387 	if (!timeout)
388 		context_reset(cmd);
389 
390 	if (unlikely(cmd->sa.ioasc != 0))
391 		pr_err("%s: CMD 0x%X failed, IOASC: flags 0x%X, afu_rc 0x%X, "
392 		       "scsi_rc 0x%X, fc_rc 0x%X\n", __func__, cmd->rcb.cdb[0],
393 		       cmd->sa.rc.flags, cmd->sa.rc.afu_rc, cmd->sa.rc.scsi_rc,
394 		       cmd->sa.rc.fc_rc);
395 }
396 
397 /**
398  * send_tmf() - sends a Task Management Function (TMF)
399  * @afu:	AFU to checkout from.
400  * @scp:	SCSI command from stack.
401  * @tmfcmd:	TMF command to send.
402  *
403  * Return:
404  *	0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
405  */
406 static int send_tmf(struct afu *afu, struct scsi_cmnd *scp, u64 tmfcmd)
407 {
408 	struct afu_cmd *cmd;
409 
410 	u32 port_sel = scp->device->channel + 1;
411 	short lflag = 0;
412 	struct Scsi_Host *host = scp->device->host;
413 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
414 	struct device *dev = &cfg->dev->dev;
415 	ulong lock_flags;
416 	int rc = 0;
417 	ulong to;
418 
419 	cmd = cmd_checkout(afu);
420 	if (unlikely(!cmd)) {
421 		dev_err(dev, "%s: could not get a free command\n", __func__);
422 		rc = SCSI_MLQUEUE_HOST_BUSY;
423 		goto out;
424 	}
425 
426 	/* When Task Management Function is active do not send another */
427 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
428 	if (cfg->tmf_active)
429 		wait_event_interruptible_lock_irq(cfg->tmf_waitq,
430 						  !cfg->tmf_active,
431 						  cfg->tmf_slock);
432 	cfg->tmf_active = true;
433 	cmd->cmd_tmf = true;
434 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
435 
436 	cmd->rcb.ctx_id = afu->ctx_hndl;
437 	cmd->rcb.port_sel = port_sel;
438 	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
439 
440 	lflag = SISL_REQ_FLAGS_TMF_CMD;
441 
442 	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
443 			      SISL_REQ_FLAGS_SUP_UNDERRUN | lflag);
444 
445 	/* Stash the scp in the reserved field, for reuse during interrupt */
446 	cmd->rcb.scp = scp;
447 
448 	/* Copy the CDB from the cmd passed in */
449 	memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
450 
451 	/* Send the command */
452 	rc = send_cmd(afu, cmd);
453 	if (unlikely(rc)) {
454 		cmd_checkin(cmd);
455 		spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
456 		cfg->tmf_active = false;
457 		spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
458 		goto out;
459 	}
460 
461 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
462 	to = msecs_to_jiffies(5000);
463 	to = wait_event_interruptible_lock_irq_timeout(cfg->tmf_waitq,
464 						       !cfg->tmf_active,
465 						       cfg->tmf_slock,
466 						       to);
467 	if (!to) {
468 		cfg->tmf_active = false;
469 		dev_err(dev, "%s: TMF timed out!\n", __func__);
470 		rc = -1;
471 	}
472 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
473 out:
474 	return rc;
475 }
476 
477 static void afu_unmap(struct kref *ref)
478 {
479 	struct afu *afu = container_of(ref, struct afu, mapcount);
480 
481 	if (likely(afu->afu_map)) {
482 		cxl_psa_unmap((void __iomem *)afu->afu_map);
483 		afu->afu_map = NULL;
484 	}
485 }
486 
487 /**
488  * cxlflash_driver_info() - information handler for this host driver
489  * @host:	SCSI host associated with device.
490  *
491  * Return: A string describing the device.
492  */
493 static const char *cxlflash_driver_info(struct Scsi_Host *host)
494 {
495 	return CXLFLASH_ADAPTER_NAME;
496 }
497 
498 /**
499  * cxlflash_queuecommand() - sends a mid-layer request
500  * @host:	SCSI host associated with device.
501  * @scp:	SCSI command to send.
502  *
503  * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
504  */
505 static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
506 {
507 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
508 	struct afu *afu = cfg->afu;
509 	struct device *dev = &cfg->dev->dev;
510 	struct afu_cmd *cmd;
511 	u32 port_sel = scp->device->channel + 1;
512 	int nseg, i, ncount;
513 	struct scatterlist *sg;
514 	ulong lock_flags;
515 	short lflag = 0;
516 	int rc = 0;
517 	int kref_got = 0;
518 
519 	dev_dbg_ratelimited(dev, "%s: (scp=%p) %d/%d/%d/%llu "
520 			    "cdb=(%08X-%08X-%08X-%08X)\n",
521 			    __func__, scp, host->host_no, scp->device->channel,
522 			    scp->device->id, scp->device->lun,
523 			    get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
524 			    get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
525 			    get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
526 			    get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
527 
528 	/*
529 	 * If a Task Management Function is active, wait for it to complete
530 	 * before continuing with regular commands.
531 	 */
532 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
533 	if (cfg->tmf_active) {
534 		spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
535 		rc = SCSI_MLQUEUE_HOST_BUSY;
536 		goto out;
537 	}
538 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
539 
540 	switch (cfg->state) {
541 	case STATE_RESET:
542 		dev_dbg_ratelimited(dev, "%s: device is in reset!\n", __func__);
543 		rc = SCSI_MLQUEUE_HOST_BUSY;
544 		goto out;
545 	case STATE_FAILTERM:
546 		dev_dbg_ratelimited(dev, "%s: device has failed!\n", __func__);
547 		scp->result = (DID_NO_CONNECT << 16);
548 		scp->scsi_done(scp);
549 		rc = 0;
550 		goto out;
551 	default:
552 		break;
553 	}
554 
555 	cmd = cmd_checkout(afu);
556 	if (unlikely(!cmd)) {
557 		dev_err(dev, "%s: could not get a free command\n", __func__);
558 		rc = SCSI_MLQUEUE_HOST_BUSY;
559 		goto out;
560 	}
561 
562 	kref_get(&cfg->afu->mapcount);
563 	kref_got = 1;
564 
565 	cmd->rcb.ctx_id = afu->ctx_hndl;
566 	cmd->rcb.port_sel = port_sel;
567 	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
568 
569 	if (scp->sc_data_direction == DMA_TO_DEVICE)
570 		lflag = SISL_REQ_FLAGS_HOST_WRITE;
571 	else
572 		lflag = SISL_REQ_FLAGS_HOST_READ;
573 
574 	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
575 			      SISL_REQ_FLAGS_SUP_UNDERRUN | lflag);
576 
577 	/* Stash the scp in the reserved field, for reuse during interrupt */
578 	cmd->rcb.scp = scp;
579 
580 	nseg = scsi_dma_map(scp);
581 	if (unlikely(nseg < 0)) {
582 		dev_err(dev, "%s: Fail DMA map! nseg=%d\n",
583 			__func__, nseg);
584 		rc = SCSI_MLQUEUE_HOST_BUSY;
585 		goto out;
586 	}
587 
588 	ncount = scsi_sg_count(scp);
589 	scsi_for_each_sg(scp, sg, ncount, i) {
590 		cmd->rcb.data_len = sg_dma_len(sg);
591 		cmd->rcb.data_ea = sg_dma_address(sg);
592 	}
593 
594 	/* Copy the CDB from the scsi_cmnd passed in */
595 	memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
596 
597 	/* Send the command */
598 	rc = send_cmd(afu, cmd);
599 	if (unlikely(rc)) {
600 		cmd_checkin(cmd);
601 		scsi_dma_unmap(scp);
602 	}
603 
604 out:
605 	if (kref_got)
606 		kref_put(&afu->mapcount, afu_unmap);
607 	pr_devel("%s: returning rc=%d\n", __func__, rc);
608 	return rc;
609 }
610 
611 /**
612  * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
613  * @cfg:	Internal structure associated with the host.
614  */
615 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
616 {
617 	struct pci_dev *pdev = cfg->dev;
618 
619 	if (pci_channel_offline(pdev))
620 		wait_event_timeout(cfg->reset_waitq,
621 				   !pci_channel_offline(pdev),
622 				   CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
623 }
624 
625 /**
626  * free_mem() - free memory associated with the AFU
627  * @cfg:	Internal structure associated with the host.
628  */
629 static void free_mem(struct cxlflash_cfg *cfg)
630 {
631 	int i;
632 	char *buf = NULL;
633 	struct afu *afu = cfg->afu;
634 
635 	if (cfg->afu) {
636 		for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
637 			buf = afu->cmd[i].buf;
638 			if (!((u64)buf & (PAGE_SIZE - 1)))
639 				free_page((ulong)buf);
640 		}
641 
642 		free_pages((ulong)afu, get_order(sizeof(struct afu)));
643 		cfg->afu = NULL;
644 	}
645 }
646 
647 /**
648  * stop_afu() - stops the AFU command timers and unmaps the MMIO space
649  * @cfg:	Internal structure associated with the host.
650  *
651  * Safe to call with AFU in a partially allocated/initialized state.
652  *
653  * Cleans up all state associated with the command queue, and unmaps
654  * the MMIO space.
655  *
656  *  - complete() will take care of commands we initiated (they'll be checked
657  *  in as part of the cleanup that occurs after the completion)
658  *
659  *  - cmd_checkin() will take care of entries that we did not initiate and that
660  *  have not (and will not) complete because they are sitting on a [now stale]
661  *  hardware queue
662  */
663 static void stop_afu(struct cxlflash_cfg *cfg)
664 {
665 	int i;
666 	struct afu *afu = cfg->afu;
667 	struct afu_cmd *cmd;
668 
669 	if (likely(afu)) {
670 		for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
671 			cmd = &afu->cmd[i];
672 			complete(&cmd->cevent);
673 			if (!atomic_read(&cmd->free))
674 				cmd_checkin(cmd);
675 		}
676 
677 		if (likely(afu->afu_map)) {
678 			cxl_psa_unmap((void __iomem *)afu->afu_map);
679 			afu->afu_map = NULL;
680 		}
681 		kref_put(&afu->mapcount, afu_unmap);
682 	}
683 }
684 
685 /**
686  * term_intr() - disables all AFU interrupts
687  * @cfg:	Internal structure associated with the host.
688  * @level:	Depth of allocation, where to begin waterfall tear down.
689  *
690  * Safe to call with AFU/MC in partially allocated/initialized state.
691  */
692 static void term_intr(struct cxlflash_cfg *cfg, enum undo_level level)
693 {
694 	struct afu *afu = cfg->afu;
695 	struct device *dev = &cfg->dev->dev;
696 
697 	if (!afu || !cfg->mcctx) {
698 		dev_err(dev, "%s: returning with NULL afu or MC\n", __func__);
699 		return;
700 	}
701 
702 	switch (level) {
703 	case UNMAP_THREE:
704 		cxl_unmap_afu_irq(cfg->mcctx, 3, afu);
705 	case UNMAP_TWO:
706 		cxl_unmap_afu_irq(cfg->mcctx, 2, afu);
707 	case UNMAP_ONE:
708 		cxl_unmap_afu_irq(cfg->mcctx, 1, afu);
709 	case FREE_IRQ:
710 		cxl_free_afu_irqs(cfg->mcctx);
711 		/* fall through */
712 	case UNDO_NOOP:
713 		/* No action required */
714 		break;
715 	}
716 }
717 
718 /**
719  * term_mc() - terminates the master context
720  * @cfg:	Internal structure associated with the host.
721  * @level:	Depth of allocation, where to begin waterfall tear down.
722  *
723  * Safe to call with AFU/MC in partially allocated/initialized state.
724  */
725 static void term_mc(struct cxlflash_cfg *cfg)
726 {
727 	int rc = 0;
728 	struct afu *afu = cfg->afu;
729 	struct device *dev = &cfg->dev->dev;
730 
731 	if (!afu || !cfg->mcctx) {
732 		dev_err(dev, "%s: returning with NULL afu or MC\n", __func__);
733 		return;
734 	}
735 
736 	rc = cxl_stop_context(cfg->mcctx);
737 	WARN_ON(rc);
738 	cfg->mcctx = NULL;
739 }
740 
741 /**
742  * term_afu() - terminates the AFU
743  * @cfg:	Internal structure associated with the host.
744  *
745  * Safe to call with AFU/MC in partially allocated/initialized state.
746  */
747 static void term_afu(struct cxlflash_cfg *cfg)
748 {
749 	/*
750 	 * Tear down is carefully orchestrated to ensure
751 	 * no interrupts can come in when the problem state
752 	 * area is unmapped.
753 	 *
754 	 * 1) Disable all AFU interrupts
755 	 * 2) Unmap the problem state area
756 	 * 3) Stop the master context
757 	 */
758 	term_intr(cfg, UNMAP_THREE);
759 	if (cfg->afu)
760 		stop_afu(cfg);
761 
762 	term_mc(cfg);
763 
764 	pr_debug("%s: returning\n", __func__);
765 }
766 
767 /**
768  * cxlflash_remove() - PCI entry point to tear down host
769  * @pdev:	PCI device associated with the host.
770  *
771  * Safe to use as a cleanup in partially allocated/initialized state.
772  */
773 static void cxlflash_remove(struct pci_dev *pdev)
774 {
775 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
776 	ulong lock_flags;
777 
778 	/* If a Task Management Function is active, wait for it to complete
779 	 * before continuing with remove.
780 	 */
781 	spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
782 	if (cfg->tmf_active)
783 		wait_event_interruptible_lock_irq(cfg->tmf_waitq,
784 						  !cfg->tmf_active,
785 						  cfg->tmf_slock);
786 	spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
787 
788 	cfg->state = STATE_FAILTERM;
789 	cxlflash_stop_term_user_contexts(cfg);
790 
791 	switch (cfg->init_state) {
792 	case INIT_STATE_SCSI:
793 		cxlflash_term_local_luns(cfg);
794 		scsi_remove_host(cfg->host);
795 		/* fall through */
796 	case INIT_STATE_AFU:
797 		cancel_work_sync(&cfg->work_q);
798 		term_afu(cfg);
799 	case INIT_STATE_PCI:
800 		pci_disable_device(pdev);
801 	case INIT_STATE_NONE:
802 		free_mem(cfg);
803 		scsi_host_put(cfg->host);
804 		break;
805 	}
806 
807 	pr_debug("%s: returning\n", __func__);
808 }
809 
810 /**
811  * alloc_mem() - allocates the AFU and its command pool
812  * @cfg:	Internal structure associated with the host.
813  *
814  * A partially allocated state remains on failure.
815  *
816  * Return:
817  *	0 on success
818  *	-ENOMEM on failure to allocate memory
819  */
820 static int alloc_mem(struct cxlflash_cfg *cfg)
821 {
822 	int rc = 0;
823 	int i;
824 	char *buf = NULL;
825 	struct device *dev = &cfg->dev->dev;
826 
827 	/* AFU is ~12k, i.e. only one 64k page or up to four 4k pages */
828 	cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
829 					    get_order(sizeof(struct afu)));
830 	if (unlikely(!cfg->afu)) {
831 		dev_err(dev, "%s: cannot get %d free pages\n",
832 			__func__, get_order(sizeof(struct afu)));
833 		rc = -ENOMEM;
834 		goto out;
835 	}
836 	cfg->afu->parent = cfg;
837 	cfg->afu->afu_map = NULL;
838 
839 	for (i = 0; i < CXLFLASH_NUM_CMDS; buf += CMD_BUFSIZE, i++) {
840 		if (!((u64)buf & (PAGE_SIZE - 1))) {
841 			buf = (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
842 			if (unlikely(!buf)) {
843 				dev_err(dev,
844 					"%s: Allocate command buffers fail!\n",
845 				       __func__);
846 				rc = -ENOMEM;
847 				free_mem(cfg);
848 				goto out;
849 			}
850 		}
851 
852 		cfg->afu->cmd[i].buf = buf;
853 		atomic_set(&cfg->afu->cmd[i].free, 1);
854 		cfg->afu->cmd[i].slot = i;
855 	}
856 
857 out:
858 	return rc;
859 }
860 
861 /**
862  * init_pci() - initializes the host as a PCI device
863  * @cfg:	Internal structure associated with the host.
864  *
865  * Return: 0 on success, -errno on failure
866  */
867 static int init_pci(struct cxlflash_cfg *cfg)
868 {
869 	struct pci_dev *pdev = cfg->dev;
870 	int rc = 0;
871 
872 	rc = pci_enable_device(pdev);
873 	if (rc || pci_channel_offline(pdev)) {
874 		if (pci_channel_offline(pdev)) {
875 			cxlflash_wait_for_pci_err_recovery(cfg);
876 			rc = pci_enable_device(pdev);
877 		}
878 
879 		if (rc) {
880 			dev_err(&pdev->dev, "%s: Cannot enable adapter\n",
881 				__func__);
882 			cxlflash_wait_for_pci_err_recovery(cfg);
883 			goto out;
884 		}
885 	}
886 
887 out:
888 	pr_debug("%s: returning rc=%d\n", __func__, rc);
889 	return rc;
890 }
891 
892 /**
893  * init_scsi() - adds the host to the SCSI stack and kicks off host scan
894  * @cfg:	Internal structure associated with the host.
895  *
896  * Return: 0 on success, -errno on failure
897  */
898 static int init_scsi(struct cxlflash_cfg *cfg)
899 {
900 	struct pci_dev *pdev = cfg->dev;
901 	int rc = 0;
902 
903 	rc = scsi_add_host(cfg->host, &pdev->dev);
904 	if (rc) {
905 		dev_err(&pdev->dev, "%s: scsi_add_host failed (rc=%d)\n",
906 			__func__, rc);
907 		goto out;
908 	}
909 
910 	scsi_scan_host(cfg->host);
911 
912 out:
913 	pr_debug("%s: returning rc=%d\n", __func__, rc);
914 	return rc;
915 }
916 
917 /**
918  * set_port_online() - transitions the specified host FC port to online state
919  * @fc_regs:	Top of MMIO region defined for specified port.
920  *
921  * The provided MMIO region must be mapped prior to call. Online state means
922  * that the FC link layer has synced, completed the handshaking process, and
923  * is ready for login to start.
924  */
925 static void set_port_online(__be64 __iomem *fc_regs)
926 {
927 	u64 cmdcfg;
928 
929 	cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
930 	cmdcfg &= (~FC_MTIP_CMDCONFIG_OFFLINE);	/* clear OFF_LINE */
931 	cmdcfg |= (FC_MTIP_CMDCONFIG_ONLINE);	/* set ON_LINE */
932 	writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
933 }
934 
935 /**
936  * set_port_offline() - transitions the specified host FC port to offline state
937  * @fc_regs:	Top of MMIO region defined for specified port.
938  *
939  * The provided MMIO region must be mapped prior to call.
940  */
941 static void set_port_offline(__be64 __iomem *fc_regs)
942 {
943 	u64 cmdcfg;
944 
945 	cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
946 	cmdcfg &= (~FC_MTIP_CMDCONFIG_ONLINE);	/* clear ON_LINE */
947 	cmdcfg |= (FC_MTIP_CMDCONFIG_OFFLINE);	/* set OFF_LINE */
948 	writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
949 }
950 
951 /**
952  * wait_port_online() - waits for the specified host FC port come online
953  * @fc_regs:	Top of MMIO region defined for specified port.
954  * @delay_us:	Number of microseconds to delay between reading port status.
955  * @nretry:	Number of cycles to retry reading port status.
956  *
957  * The provided MMIO region must be mapped prior to call. This will timeout
958  * when the cable is not plugged in.
959  *
960  * Return:
961  *	TRUE (1) when the specified port is online
962  *	FALSE (0) when the specified port fails to come online after timeout
963  *	-EINVAL when @delay_us is less than 1000
964  */
965 static int wait_port_online(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
966 {
967 	u64 status;
968 
969 	if (delay_us < 1000) {
970 		pr_err("%s: invalid delay specified %d\n", __func__, delay_us);
971 		return -EINVAL;
972 	}
973 
974 	do {
975 		msleep(delay_us / 1000);
976 		status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
977 	} while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_ONLINE &&
978 		 nretry--);
979 
980 	return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_ONLINE);
981 }
982 
983 /**
984  * wait_port_offline() - waits for the specified host FC port go offline
985  * @fc_regs:	Top of MMIO region defined for specified port.
986  * @delay_us:	Number of microseconds to delay between reading port status.
987  * @nretry:	Number of cycles to retry reading port status.
988  *
989  * The provided MMIO region must be mapped prior to call.
990  *
991  * Return:
992  *	TRUE (1) when the specified port is offline
993  *	FALSE (0) when the specified port fails to go offline after timeout
994  *	-EINVAL when @delay_us is less than 1000
995  */
996 static int wait_port_offline(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
997 {
998 	u64 status;
999 
1000 	if (delay_us < 1000) {
1001 		pr_err("%s: invalid delay specified %d\n", __func__, delay_us);
1002 		return -EINVAL;
1003 	}
1004 
1005 	do {
1006 		msleep(delay_us / 1000);
1007 		status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1008 	} while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_OFFLINE &&
1009 		 nretry--);
1010 
1011 	return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_OFFLINE);
1012 }
1013 
1014 /**
1015  * afu_set_wwpn() - configures the WWPN for the specified host FC port
1016  * @afu:	AFU associated with the host that owns the specified FC port.
1017  * @port:	Port number being configured.
1018  * @fc_regs:	Top of MMIO region defined for specified port.
1019  * @wwpn:	The world-wide-port-number previously discovered for port.
1020  *
1021  * The provided MMIO region must be mapped prior to call. As part of the
1022  * sequence to configure the WWPN, the port is toggled offline and then back
1023  * online. This toggling action can cause this routine to delay up to a few
1024  * seconds. When configured to use the internal LUN feature of the AFU, a
1025  * failure to come online is overridden.
1026  *
1027  * Return:
1028  *	0 when the WWPN is successfully written and the port comes back online
1029  *	-1 when the port fails to go offline or come back up online
1030  */
1031 static int afu_set_wwpn(struct afu *afu, int port, __be64 __iomem *fc_regs,
1032 			u64 wwpn)
1033 {
1034 	int rc = 0;
1035 
1036 	set_port_offline(fc_regs);
1037 
1038 	if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1039 			       FC_PORT_STATUS_RETRY_CNT)) {
1040 		pr_debug("%s: wait on port %d to go offline timed out\n",
1041 			 __func__, port);
1042 		rc = -1; /* but continue on to leave the port back online */
1043 	}
1044 
1045 	if (rc == 0)
1046 		writeq_be(wwpn, &fc_regs[FC_PNAME / 8]);
1047 
1048 	/* Always return success after programming WWPN */
1049 	rc = 0;
1050 
1051 	set_port_online(fc_regs);
1052 
1053 	if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1054 			      FC_PORT_STATUS_RETRY_CNT)) {
1055 		pr_err("%s: wait on port %d to go online timed out\n",
1056 		       __func__, port);
1057 	}
1058 
1059 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1060 
1061 	return rc;
1062 }
1063 
1064 /**
1065  * afu_link_reset() - resets the specified host FC port
1066  * @afu:	AFU associated with the host that owns the specified FC port.
1067  * @port:	Port number being configured.
1068  * @fc_regs:	Top of MMIO region defined for specified port.
1069  *
1070  * The provided MMIO region must be mapped prior to call. The sequence to
1071  * reset the port involves toggling it offline and then back online. This
1072  * action can cause this routine to delay up to a few seconds. An effort
1073  * is made to maintain link with the device by switching to host to use
1074  * the alternate port exclusively while the reset takes place.
1075  * failure to come online is overridden.
1076  */
1077 static void afu_link_reset(struct afu *afu, int port, __be64 __iomem *fc_regs)
1078 {
1079 	u64 port_sel;
1080 
1081 	/* first switch the AFU to the other links, if any */
1082 	port_sel = readq_be(&afu->afu_map->global.regs.afu_port_sel);
1083 	port_sel &= ~(1ULL << port);
1084 	writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1085 	cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1086 
1087 	set_port_offline(fc_regs);
1088 	if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1089 			       FC_PORT_STATUS_RETRY_CNT))
1090 		pr_err("%s: wait on port %d to go offline timed out\n",
1091 		       __func__, port);
1092 
1093 	set_port_online(fc_regs);
1094 	if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1095 			      FC_PORT_STATUS_RETRY_CNT))
1096 		pr_err("%s: wait on port %d to go online timed out\n",
1097 		       __func__, port);
1098 
1099 	/* switch back to include this port */
1100 	port_sel |= (1ULL << port);
1101 	writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1102 	cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1103 
1104 	pr_debug("%s: returning port_sel=%lld\n", __func__, port_sel);
1105 }
1106 
1107 /*
1108  * Asynchronous interrupt information table
1109  */
1110 static const struct asyc_intr_info ainfo[] = {
1111 	{SISL_ASTATUS_FC0_OTHER, "other error", 0, CLR_FC_ERROR | LINK_RESET},
1112 	{SISL_ASTATUS_FC0_LOGO, "target initiated LOGO", 0, 0},
1113 	{SISL_ASTATUS_FC0_CRC_T, "CRC threshold exceeded", 0, LINK_RESET},
1114 	{SISL_ASTATUS_FC0_LOGI_R, "login timed out, retrying", 0, LINK_RESET},
1115 	{SISL_ASTATUS_FC0_LOGI_F, "login failed", 0, CLR_FC_ERROR},
1116 	{SISL_ASTATUS_FC0_LOGI_S, "login succeeded", 0, SCAN_HOST},
1117 	{SISL_ASTATUS_FC0_LINK_DN, "link down", 0, 0},
1118 	{SISL_ASTATUS_FC0_LINK_UP, "link up", 0, SCAN_HOST},
1119 	{SISL_ASTATUS_FC1_OTHER, "other error", 1, CLR_FC_ERROR | LINK_RESET},
1120 	{SISL_ASTATUS_FC1_LOGO, "target initiated LOGO", 1, 0},
1121 	{SISL_ASTATUS_FC1_CRC_T, "CRC threshold exceeded", 1, LINK_RESET},
1122 	{SISL_ASTATUS_FC1_LOGI_R, "login timed out, retrying", 1, LINK_RESET},
1123 	{SISL_ASTATUS_FC1_LOGI_F, "login failed", 1, CLR_FC_ERROR},
1124 	{SISL_ASTATUS_FC1_LOGI_S, "login succeeded", 1, SCAN_HOST},
1125 	{SISL_ASTATUS_FC1_LINK_DN, "link down", 1, 0},
1126 	{SISL_ASTATUS_FC1_LINK_UP, "link up", 1, SCAN_HOST},
1127 	{0x0, "", 0, 0}		/* terminator */
1128 };
1129 
1130 /**
1131  * find_ainfo() - locates and returns asynchronous interrupt information
1132  * @status:	Status code set by AFU on error.
1133  *
1134  * Return: The located information or NULL when the status code is invalid.
1135  */
1136 static const struct asyc_intr_info *find_ainfo(u64 status)
1137 {
1138 	const struct asyc_intr_info *info;
1139 
1140 	for (info = &ainfo[0]; info->status; info++)
1141 		if (info->status == status)
1142 			return info;
1143 
1144 	return NULL;
1145 }
1146 
1147 /**
1148  * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1149  * @afu:	AFU associated with the host.
1150  */
1151 static void afu_err_intr_init(struct afu *afu)
1152 {
1153 	int i;
1154 	u64 reg;
1155 
1156 	/* global async interrupts: AFU clears afu_ctrl on context exit
1157 	 * if async interrupts were sent to that context. This prevents
1158 	 * the AFU form sending further async interrupts when
1159 	 * there is
1160 	 * nobody to receive them.
1161 	 */
1162 
1163 	/* mask all */
1164 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_mask);
1165 	/* set LISN# to send and point to master context */
1166 	reg = ((u64) (((afu->ctx_hndl << 8) | SISL_MSI_ASYNC_ERROR)) << 40);
1167 
1168 	if (afu->internal_lun)
1169 		reg |= 1;	/* Bit 63 indicates local lun */
1170 	writeq_be(reg, &afu->afu_map->global.regs.afu_ctrl);
1171 	/* clear all */
1172 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1173 	/* unmask bits that are of interest */
1174 	/* note: afu can send an interrupt after this step */
1175 	writeq_be(SISL_ASTATUS_MASK, &afu->afu_map->global.regs.aintr_mask);
1176 	/* clear again in case a bit came on after previous clear but before */
1177 	/* unmask */
1178 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1179 
1180 	/* Clear/Set internal lun bits */
1181 	reg = readq_be(&afu->afu_map->global.fc_regs[0][FC_CONFIG2 / 8]);
1182 	reg &= SISL_FC_INTERNAL_MASK;
1183 	if (afu->internal_lun)
1184 		reg |= ((u64)(afu->internal_lun - 1) << SISL_FC_INTERNAL_SHIFT);
1185 	writeq_be(reg, &afu->afu_map->global.fc_regs[0][FC_CONFIG2 / 8]);
1186 
1187 	/* now clear FC errors */
1188 	for (i = 0; i < NUM_FC_PORTS; i++) {
1189 		writeq_be(0xFFFFFFFFU,
1190 			  &afu->afu_map->global.fc_regs[i][FC_ERROR / 8]);
1191 		writeq_be(0, &afu->afu_map->global.fc_regs[i][FC_ERRCAP / 8]);
1192 	}
1193 
1194 	/* sync interrupts for master's IOARRIN write */
1195 	/* note that unlike asyncs, there can be no pending sync interrupts */
1196 	/* at this time (this is a fresh context and master has not written */
1197 	/* IOARRIN yet), so there is nothing to clear. */
1198 
1199 	/* set LISN#, it is always sent to the context that wrote IOARRIN */
1200 	writeq_be(SISL_MSI_SYNC_ERROR, &afu->host_map->ctx_ctrl);
1201 	writeq_be(SISL_ISTATUS_MASK, &afu->host_map->intr_mask);
1202 }
1203 
1204 /**
1205  * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1206  * @irq:	Interrupt number.
1207  * @data:	Private data provided at interrupt registration, the AFU.
1208  *
1209  * Return: Always return IRQ_HANDLED.
1210  */
1211 static irqreturn_t cxlflash_sync_err_irq(int irq, void *data)
1212 {
1213 	struct afu *afu = (struct afu *)data;
1214 	u64 reg;
1215 	u64 reg_unmasked;
1216 
1217 	reg = readq_be(&afu->host_map->intr_status);
1218 	reg_unmasked = (reg & SISL_ISTATUS_UNMASK);
1219 
1220 	if (reg_unmasked == 0UL) {
1221 		pr_err("%s: %llX: spurious interrupt, intr_status %016llX\n",
1222 		       __func__, (u64)afu, reg);
1223 		goto cxlflash_sync_err_irq_exit;
1224 	}
1225 
1226 	pr_err("%s: %llX: unexpected interrupt, intr_status %016llX\n",
1227 	       __func__, (u64)afu, reg);
1228 
1229 	writeq_be(reg_unmasked, &afu->host_map->intr_clear);
1230 
1231 cxlflash_sync_err_irq_exit:
1232 	pr_debug("%s: returning rc=%d\n", __func__, IRQ_HANDLED);
1233 	return IRQ_HANDLED;
1234 }
1235 
1236 /**
1237  * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1238  * @irq:	Interrupt number.
1239  * @data:	Private data provided at interrupt registration, the AFU.
1240  *
1241  * Return: Always return IRQ_HANDLED.
1242  */
1243 static irqreturn_t cxlflash_rrq_irq(int irq, void *data)
1244 {
1245 	struct afu *afu = (struct afu *)data;
1246 	struct afu_cmd *cmd;
1247 	bool toggle = afu->toggle;
1248 	u64 entry,
1249 	    *hrrq_start = afu->hrrq_start,
1250 	    *hrrq_end = afu->hrrq_end,
1251 	    *hrrq_curr = afu->hrrq_curr;
1252 
1253 	/* Process however many RRQ entries that are ready */
1254 	while (true) {
1255 		entry = *hrrq_curr;
1256 
1257 		if ((entry & SISL_RESP_HANDLE_T_BIT) != toggle)
1258 			break;
1259 
1260 		cmd = (struct afu_cmd *)(entry & ~SISL_RESP_HANDLE_T_BIT);
1261 		cmd_complete(cmd);
1262 
1263 		/* Advance to next entry or wrap and flip the toggle bit */
1264 		if (hrrq_curr < hrrq_end)
1265 			hrrq_curr++;
1266 		else {
1267 			hrrq_curr = hrrq_start;
1268 			toggle ^= SISL_RESP_HANDLE_T_BIT;
1269 		}
1270 	}
1271 
1272 	afu->hrrq_curr = hrrq_curr;
1273 	afu->toggle = toggle;
1274 
1275 	return IRQ_HANDLED;
1276 }
1277 
1278 /**
1279  * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1280  * @irq:	Interrupt number.
1281  * @data:	Private data provided at interrupt registration, the AFU.
1282  *
1283  * Return: Always return IRQ_HANDLED.
1284  */
1285 static irqreturn_t cxlflash_async_err_irq(int irq, void *data)
1286 {
1287 	struct afu *afu = (struct afu *)data;
1288 	struct cxlflash_cfg *cfg = afu->parent;
1289 	struct device *dev = &cfg->dev->dev;
1290 	u64 reg_unmasked;
1291 	const struct asyc_intr_info *info;
1292 	struct sisl_global_map __iomem *global = &afu->afu_map->global;
1293 	u64 reg;
1294 	u8 port;
1295 	int i;
1296 
1297 	reg = readq_be(&global->regs.aintr_status);
1298 	reg_unmasked = (reg & SISL_ASTATUS_UNMASK);
1299 
1300 	if (reg_unmasked == 0) {
1301 		dev_err(dev, "%s: spurious interrupt, aintr_status 0x%016llX\n",
1302 			__func__, reg);
1303 		goto out;
1304 	}
1305 
1306 	/* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1307 	writeq_be(reg_unmasked, &global->regs.aintr_clear);
1308 
1309 	/* Check each bit that is on */
1310 	for (i = 0; reg_unmasked; i++, reg_unmasked = (reg_unmasked >> 1)) {
1311 		info = find_ainfo(1ULL << i);
1312 		if (((reg_unmasked & 0x1) == 0) || !info)
1313 			continue;
1314 
1315 		port = info->port;
1316 
1317 		dev_err(dev, "%s: FC Port %d -> %s, fc_status 0x%08llX\n",
1318 			__func__, port, info->desc,
1319 		       readq_be(&global->fc_regs[port][FC_STATUS / 8]));
1320 
1321 		/*
1322 		 * Do link reset first, some OTHER errors will set FC_ERROR
1323 		 * again if cleared before or w/o a reset
1324 		 */
1325 		if (info->action & LINK_RESET) {
1326 			dev_err(dev, "%s: FC Port %d: resetting link\n",
1327 				__func__, port);
1328 			cfg->lr_state = LINK_RESET_REQUIRED;
1329 			cfg->lr_port = port;
1330 			kref_get(&cfg->afu->mapcount);
1331 			schedule_work(&cfg->work_q);
1332 		}
1333 
1334 		if (info->action & CLR_FC_ERROR) {
1335 			reg = readq_be(&global->fc_regs[port][FC_ERROR / 8]);
1336 
1337 			/*
1338 			 * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1339 			 * should be the same and tracing one is sufficient.
1340 			 */
1341 
1342 			dev_err(dev, "%s: fc %d: clearing fc_error 0x%08llX\n",
1343 				__func__, port, reg);
1344 
1345 			writeq_be(reg, &global->fc_regs[port][FC_ERROR / 8]);
1346 			writeq_be(0, &global->fc_regs[port][FC_ERRCAP / 8]);
1347 		}
1348 
1349 		if (info->action & SCAN_HOST) {
1350 			atomic_inc(&cfg->scan_host_needed);
1351 			kref_get(&cfg->afu->mapcount);
1352 			schedule_work(&cfg->work_q);
1353 		}
1354 	}
1355 
1356 out:
1357 	dev_dbg(dev, "%s: returning IRQ_HANDLED, afu=%p\n", __func__, afu);
1358 	return IRQ_HANDLED;
1359 }
1360 
1361 /**
1362  * start_context() - starts the master context
1363  * @cfg:	Internal structure associated with the host.
1364  *
1365  * Return: A success or failure value from CXL services.
1366  */
1367 static int start_context(struct cxlflash_cfg *cfg)
1368 {
1369 	int rc = 0;
1370 
1371 	rc = cxl_start_context(cfg->mcctx,
1372 			       cfg->afu->work.work_element_descriptor,
1373 			       NULL);
1374 
1375 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1376 	return rc;
1377 }
1378 
1379 /**
1380  * read_vpd() - obtains the WWPNs from VPD
1381  * @cfg:	Internal structure associated with the host.
1382  * @wwpn:	Array of size NUM_FC_PORTS to pass back WWPNs
1383  *
1384  * Return: 0 on success, -errno on failure
1385  */
1386 static int read_vpd(struct cxlflash_cfg *cfg, u64 wwpn[])
1387 {
1388 	struct pci_dev *dev = cfg->dev;
1389 	int rc = 0;
1390 	int ro_start, ro_size, i, j, k;
1391 	ssize_t vpd_size;
1392 	char vpd_data[CXLFLASH_VPD_LEN];
1393 	char tmp_buf[WWPN_BUF_LEN] = { 0 };
1394 	char *wwpn_vpd_tags[NUM_FC_PORTS] = { "V5", "V6" };
1395 
1396 	/* Get the VPD data from the device */
1397 	vpd_size = cxl_read_adapter_vpd(dev, vpd_data, sizeof(vpd_data));
1398 	if (unlikely(vpd_size <= 0)) {
1399 		dev_err(&dev->dev, "%s: Unable to read VPD (size = %ld)\n",
1400 		       __func__, vpd_size);
1401 		rc = -ENODEV;
1402 		goto out;
1403 	}
1404 
1405 	/* Get the read only section offset */
1406 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size,
1407 				    PCI_VPD_LRDT_RO_DATA);
1408 	if (unlikely(ro_start < 0)) {
1409 		dev_err(&dev->dev, "%s: VPD Read-only data not found\n",
1410 			__func__);
1411 		rc = -ENODEV;
1412 		goto out;
1413 	}
1414 
1415 	/* Get the read only section size, cap when extends beyond read VPD */
1416 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
1417 	j = ro_size;
1418 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1419 	if (unlikely((i + j) > vpd_size)) {
1420 		pr_debug("%s: Might need to read more VPD (%d > %ld)\n",
1421 			 __func__, (i + j), vpd_size);
1422 		ro_size = vpd_size - i;
1423 	}
1424 
1425 	/*
1426 	 * Find the offset of the WWPN tag within the read only
1427 	 * VPD data and validate the found field (partials are
1428 	 * no good to us). Convert the ASCII data to an integer
1429 	 * value. Note that we must copy to a temporary buffer
1430 	 * because the conversion service requires that the ASCII
1431 	 * string be terminated.
1432 	 */
1433 	for (k = 0; k < NUM_FC_PORTS; k++) {
1434 		j = ro_size;
1435 		i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1436 
1437 		i = pci_vpd_find_info_keyword(vpd_data, i, j, wwpn_vpd_tags[k]);
1438 		if (unlikely(i < 0)) {
1439 			dev_err(&dev->dev, "%s: Port %d WWPN not found "
1440 				"in VPD\n", __func__, k);
1441 			rc = -ENODEV;
1442 			goto out;
1443 		}
1444 
1445 		j = pci_vpd_info_field_size(&vpd_data[i]);
1446 		i += PCI_VPD_INFO_FLD_HDR_SIZE;
1447 		if (unlikely((i + j > vpd_size) || (j != WWPN_LEN))) {
1448 			dev_err(&dev->dev, "%s: Port %d WWPN incomplete or "
1449 				"VPD corrupt\n",
1450 			       __func__, k);
1451 			rc = -ENODEV;
1452 			goto out;
1453 		}
1454 
1455 		memcpy(tmp_buf, &vpd_data[i], WWPN_LEN);
1456 		rc = kstrtoul(tmp_buf, WWPN_LEN, (ulong *)&wwpn[k]);
1457 		if (unlikely(rc)) {
1458 			dev_err(&dev->dev, "%s: Fail to convert port %d WWPN "
1459 				"to integer\n", __func__, k);
1460 			rc = -ENODEV;
1461 			goto out;
1462 		}
1463 	}
1464 
1465 out:
1466 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1467 	return rc;
1468 }
1469 
1470 /**
1471  * init_pcr() - initialize the provisioning and control registers
1472  * @cfg:	Internal structure associated with the host.
1473  *
1474  * Also sets up fast access to the mapped registers and initializes AFU
1475  * command fields that never change.
1476  */
1477 static void init_pcr(struct cxlflash_cfg *cfg)
1478 {
1479 	struct afu *afu = cfg->afu;
1480 	struct sisl_ctrl_map __iomem *ctrl_map;
1481 	int i;
1482 
1483 	for (i = 0; i < MAX_CONTEXT; i++) {
1484 		ctrl_map = &afu->afu_map->ctrls[i].ctrl;
1485 		/* Disrupt any clients that could be running */
1486 		/* e.g. clients that survived a master restart */
1487 		writeq_be(0, &ctrl_map->rht_start);
1488 		writeq_be(0, &ctrl_map->rht_cnt_id);
1489 		writeq_be(0, &ctrl_map->ctx_cap);
1490 	}
1491 
1492 	/* Copy frequently used fields into afu */
1493 	afu->ctx_hndl = (u16) cxl_process_element(cfg->mcctx);
1494 	afu->host_map = &afu->afu_map->hosts[afu->ctx_hndl].host;
1495 	afu->ctrl_map = &afu->afu_map->ctrls[afu->ctx_hndl].ctrl;
1496 
1497 	/* Program the Endian Control for the master context */
1498 	writeq_be(SISL_ENDIAN_CTRL, &afu->host_map->endian_ctrl);
1499 
1500 	/* Initialize cmd fields that never change */
1501 	for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
1502 		afu->cmd[i].rcb.ctx_id = afu->ctx_hndl;
1503 		afu->cmd[i].rcb.msi = SISL_MSI_RRQ_UPDATED;
1504 		afu->cmd[i].rcb.rrq = 0x0;
1505 	}
1506 }
1507 
1508 /**
1509  * init_global() - initialize AFU global registers
1510  * @cfg:	Internal structure associated with the host.
1511  */
1512 static int init_global(struct cxlflash_cfg *cfg)
1513 {
1514 	struct afu *afu = cfg->afu;
1515 	struct device *dev = &cfg->dev->dev;
1516 	u64 wwpn[NUM_FC_PORTS];	/* wwpn of AFU ports */
1517 	int i = 0, num_ports = 0;
1518 	int rc = 0;
1519 	u64 reg;
1520 
1521 	rc = read_vpd(cfg, &wwpn[0]);
1522 	if (rc) {
1523 		dev_err(dev, "%s: could not read vpd rc=%d\n", __func__, rc);
1524 		goto out;
1525 	}
1526 
1527 	pr_debug("%s: wwpn0=0x%llX wwpn1=0x%llX\n", __func__, wwpn[0], wwpn[1]);
1528 
1529 	/* Set up RRQ in AFU for master issued cmds */
1530 	writeq_be((u64) afu->hrrq_start, &afu->host_map->rrq_start);
1531 	writeq_be((u64) afu->hrrq_end, &afu->host_map->rrq_end);
1532 
1533 	/* AFU configuration */
1534 	reg = readq_be(&afu->afu_map->global.regs.afu_config);
1535 	reg |= SISL_AFUCONF_AR_ALL|SISL_AFUCONF_ENDIAN;
1536 	/* enable all auto retry options and control endianness */
1537 	/* leave others at default: */
1538 	/* CTX_CAP write protected, mbox_r does not clear on read and */
1539 	/* checker on if dual afu */
1540 	writeq_be(reg, &afu->afu_map->global.regs.afu_config);
1541 
1542 	/* Global port select: select either port */
1543 	if (afu->internal_lun) {
1544 		/* Only use port 0 */
1545 		writeq_be(PORT0, &afu->afu_map->global.regs.afu_port_sel);
1546 		num_ports = NUM_FC_PORTS - 1;
1547 	} else {
1548 		writeq_be(BOTH_PORTS, &afu->afu_map->global.regs.afu_port_sel);
1549 		num_ports = NUM_FC_PORTS;
1550 	}
1551 
1552 	for (i = 0; i < num_ports; i++) {
1553 		/* Unmask all errors (but they are still masked at AFU) */
1554 		writeq_be(0, &afu->afu_map->global.fc_regs[i][FC_ERRMSK / 8]);
1555 		/* Clear CRC error cnt & set a threshold */
1556 		(void)readq_be(&afu->afu_map->global.
1557 			       fc_regs[i][FC_CNT_CRCERR / 8]);
1558 		writeq_be(MC_CRC_THRESH, &afu->afu_map->global.fc_regs[i]
1559 			  [FC_CRC_THRESH / 8]);
1560 
1561 		/* Set WWPNs. If already programmed, wwpn[i] is 0 */
1562 		if (wwpn[i] != 0 &&
1563 		    afu_set_wwpn(afu, i,
1564 				 &afu->afu_map->global.fc_regs[i][0],
1565 				 wwpn[i])) {
1566 			dev_err(dev, "%s: failed to set WWPN on port %d\n",
1567 			       __func__, i);
1568 			rc = -EIO;
1569 			goto out;
1570 		}
1571 		/* Programming WWPN back to back causes additional
1572 		 * offline/online transitions and a PLOGI
1573 		 */
1574 		msleep(100);
1575 	}
1576 
1577 	/* Set up master's own CTX_CAP to allow real mode, host translation */
1578 	/* tables, afu cmds and read/write GSCSI cmds. */
1579 	/* First, unlock ctx_cap write by reading mbox */
1580 	(void)readq_be(&afu->ctrl_map->mbox_r);	/* unlock ctx_cap */
1581 	writeq_be((SISL_CTX_CAP_REAL_MODE | SISL_CTX_CAP_HOST_XLATE |
1582 		   SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD |
1583 		   SISL_CTX_CAP_AFU_CMD | SISL_CTX_CAP_GSCSI_CMD),
1584 		  &afu->ctrl_map->ctx_cap);
1585 	/* Initialize heartbeat */
1586 	afu->hb = readq_be(&afu->afu_map->global.regs.afu_hb);
1587 
1588 out:
1589 	return rc;
1590 }
1591 
1592 /**
1593  * start_afu() - initializes and starts the AFU
1594  * @cfg:	Internal structure associated with the host.
1595  */
1596 static int start_afu(struct cxlflash_cfg *cfg)
1597 {
1598 	struct afu *afu = cfg->afu;
1599 	struct afu_cmd *cmd;
1600 
1601 	int i = 0;
1602 	int rc = 0;
1603 
1604 	for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
1605 		cmd = &afu->cmd[i];
1606 
1607 		init_completion(&cmd->cevent);
1608 		spin_lock_init(&cmd->slock);
1609 		cmd->parent = afu;
1610 	}
1611 
1612 	init_pcr(cfg);
1613 
1614 	/* After an AFU reset, RRQ entries are stale, clear them */
1615 	memset(&afu->rrq_entry, 0, sizeof(afu->rrq_entry));
1616 
1617 	/* Initialize RRQ pointers */
1618 	afu->hrrq_start = &afu->rrq_entry[0];
1619 	afu->hrrq_end = &afu->rrq_entry[NUM_RRQ_ENTRY - 1];
1620 	afu->hrrq_curr = afu->hrrq_start;
1621 	afu->toggle = 1;
1622 
1623 	rc = init_global(cfg);
1624 
1625 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1626 	return rc;
1627 }
1628 
1629 /**
1630  * init_intr() - setup interrupt handlers for the master context
1631  * @cfg:	Internal structure associated with the host.
1632  *
1633  * Return: 0 on success, -errno on failure
1634  */
1635 static enum undo_level init_intr(struct cxlflash_cfg *cfg,
1636 				 struct cxl_context *ctx)
1637 {
1638 	struct afu *afu = cfg->afu;
1639 	struct device *dev = &cfg->dev->dev;
1640 	int rc = 0;
1641 	enum undo_level level = UNDO_NOOP;
1642 
1643 	rc = cxl_allocate_afu_irqs(ctx, 3);
1644 	if (unlikely(rc)) {
1645 		dev_err(dev, "%s: call to allocate_afu_irqs failed rc=%d!\n",
1646 			__func__, rc);
1647 		level = UNDO_NOOP;
1648 		goto out;
1649 	}
1650 
1651 	rc = cxl_map_afu_irq(ctx, 1, cxlflash_sync_err_irq, afu,
1652 			     "SISL_MSI_SYNC_ERROR");
1653 	if (unlikely(rc <= 0)) {
1654 		dev_err(dev, "%s: IRQ 1 (SISL_MSI_SYNC_ERROR) map failed!\n",
1655 			__func__);
1656 		level = FREE_IRQ;
1657 		goto out;
1658 	}
1659 
1660 	rc = cxl_map_afu_irq(ctx, 2, cxlflash_rrq_irq, afu,
1661 			     "SISL_MSI_RRQ_UPDATED");
1662 	if (unlikely(rc <= 0)) {
1663 		dev_err(dev, "%s: IRQ 2 (SISL_MSI_RRQ_UPDATED) map failed!\n",
1664 			__func__);
1665 		level = UNMAP_ONE;
1666 		goto out;
1667 	}
1668 
1669 	rc = cxl_map_afu_irq(ctx, 3, cxlflash_async_err_irq, afu,
1670 			     "SISL_MSI_ASYNC_ERROR");
1671 	if (unlikely(rc <= 0)) {
1672 		dev_err(dev, "%s: IRQ 3 (SISL_MSI_ASYNC_ERROR) map failed!\n",
1673 			__func__);
1674 		level = UNMAP_TWO;
1675 		goto out;
1676 	}
1677 out:
1678 	return level;
1679 }
1680 
1681 /**
1682  * init_mc() - create and register as the master context
1683  * @cfg:	Internal structure associated with the host.
1684  *
1685  * Return: 0 on success, -errno on failure
1686  */
1687 static int init_mc(struct cxlflash_cfg *cfg)
1688 {
1689 	struct cxl_context *ctx;
1690 	struct device *dev = &cfg->dev->dev;
1691 	int rc = 0;
1692 	enum undo_level level;
1693 
1694 	ctx = cxl_get_context(cfg->dev);
1695 	if (unlikely(!ctx)) {
1696 		rc = -ENOMEM;
1697 		goto ret;
1698 	}
1699 	cfg->mcctx = ctx;
1700 
1701 	/* Set it up as a master with the CXL */
1702 	cxl_set_master(ctx);
1703 
1704 	/* During initialization reset the AFU to start from a clean slate */
1705 	rc = cxl_afu_reset(cfg->mcctx);
1706 	if (unlikely(rc)) {
1707 		dev_err(dev, "%s: initial AFU reset failed rc=%d\n",
1708 			__func__, rc);
1709 		goto ret;
1710 	}
1711 
1712 	level = init_intr(cfg, ctx);
1713 	if (unlikely(level)) {
1714 		dev_err(dev, "%s: setting up interrupts failed rc=%d\n",
1715 			__func__, rc);
1716 		goto out;
1717 	}
1718 
1719 	/* This performs the equivalent of the CXL_IOCTL_START_WORK.
1720 	 * The CXL_IOCTL_GET_PROCESS_ELEMENT is implicit in the process
1721 	 * element (pe) that is embedded in the context (ctx)
1722 	 */
1723 	rc = start_context(cfg);
1724 	if (unlikely(rc)) {
1725 		dev_err(dev, "%s: start context failed rc=%d\n", __func__, rc);
1726 		level = UNMAP_THREE;
1727 		goto out;
1728 	}
1729 ret:
1730 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1731 	return rc;
1732 out:
1733 	term_intr(cfg, level);
1734 	goto ret;
1735 }
1736 
1737 /**
1738  * init_afu() - setup as master context and start AFU
1739  * @cfg:	Internal structure associated with the host.
1740  *
1741  * This routine is a higher level of control for configuring the
1742  * AFU on probe and reset paths.
1743  *
1744  * Return: 0 on success, -errno on failure
1745  */
1746 static int init_afu(struct cxlflash_cfg *cfg)
1747 {
1748 	u64 reg;
1749 	int rc = 0;
1750 	struct afu *afu = cfg->afu;
1751 	struct device *dev = &cfg->dev->dev;
1752 
1753 	cxl_perst_reloads_same_image(cfg->cxl_afu, true);
1754 
1755 	rc = init_mc(cfg);
1756 	if (rc) {
1757 		dev_err(dev, "%s: call to init_mc failed, rc=%d!\n",
1758 			__func__, rc);
1759 		goto out;
1760 	}
1761 
1762 	/* Map the entire MMIO space of the AFU */
1763 	afu->afu_map = cxl_psa_map(cfg->mcctx);
1764 	if (!afu->afu_map) {
1765 		dev_err(dev, "%s: call to cxl_psa_map failed!\n", __func__);
1766 		rc = -ENOMEM;
1767 		goto err1;
1768 	}
1769 	kref_init(&afu->mapcount);
1770 
1771 	/* No byte reverse on reading afu_version or string will be backwards */
1772 	reg = readq(&afu->afu_map->global.regs.afu_version);
1773 	memcpy(afu->version, &reg, sizeof(reg));
1774 	afu->interface_version =
1775 	    readq_be(&afu->afu_map->global.regs.interface_version);
1776 	if ((afu->interface_version + 1) == 0) {
1777 		pr_err("Back level AFU, please upgrade. AFU version %s "
1778 		       "interface version 0x%llx\n", afu->version,
1779 		       afu->interface_version);
1780 		rc = -EINVAL;
1781 		goto err2;
1782 	}
1783 
1784 	pr_debug("%s: afu version %s, interface version 0x%llX\n", __func__,
1785 		 afu->version, afu->interface_version);
1786 
1787 	rc = start_afu(cfg);
1788 	if (rc) {
1789 		dev_err(dev, "%s: call to start_afu failed, rc=%d!\n",
1790 			__func__, rc);
1791 		goto err2;
1792 	}
1793 
1794 	afu_err_intr_init(cfg->afu);
1795 	atomic64_set(&afu->room, readq_be(&afu->host_map->cmd_room));
1796 
1797 	/* Restore the LUN mappings */
1798 	cxlflash_restore_luntable(cfg);
1799 out:
1800 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1801 	return rc;
1802 
1803 err2:
1804 	kref_put(&afu->mapcount, afu_unmap);
1805 err1:
1806 	term_intr(cfg, UNMAP_THREE);
1807 	term_mc(cfg);
1808 	goto out;
1809 }
1810 
1811 /**
1812  * cxlflash_afu_sync() - builds and sends an AFU sync command
1813  * @afu:	AFU associated with the host.
1814  * @ctx_hndl_u:	Identifies context requesting sync.
1815  * @res_hndl_u:	Identifies resource requesting sync.
1816  * @mode:	Type of sync to issue (lightweight, heavyweight, global).
1817  *
1818  * The AFU can only take 1 sync command at a time. This routine enforces this
1819  * limitation by using a mutex to provide exclusive access to the AFU during
1820  * the sync. This design point requires calling threads to not be on interrupt
1821  * context due to the possibility of sleeping during concurrent sync operations.
1822  *
1823  * AFU sync operations are only necessary and allowed when the device is
1824  * operating normally. When not operating normally, sync requests can occur as
1825  * part of cleaning up resources associated with an adapter prior to removal.
1826  * In this scenario, these requests are simply ignored (safe due to the AFU
1827  * going away).
1828  *
1829  * Return:
1830  *	0 on success
1831  *	-1 on failure
1832  */
1833 int cxlflash_afu_sync(struct afu *afu, ctx_hndl_t ctx_hndl_u,
1834 		      res_hndl_t res_hndl_u, u8 mode)
1835 {
1836 	struct cxlflash_cfg *cfg = afu->parent;
1837 	struct device *dev = &cfg->dev->dev;
1838 	struct afu_cmd *cmd = NULL;
1839 	int rc = 0;
1840 	int retry_cnt = 0;
1841 	static DEFINE_MUTEX(sync_active);
1842 
1843 	if (cfg->state != STATE_NORMAL) {
1844 		pr_debug("%s: Sync not required! (%u)\n", __func__, cfg->state);
1845 		return 0;
1846 	}
1847 
1848 	mutex_lock(&sync_active);
1849 retry:
1850 	cmd = cmd_checkout(afu);
1851 	if (unlikely(!cmd)) {
1852 		retry_cnt++;
1853 		udelay(1000 * retry_cnt);
1854 		if (retry_cnt < MC_RETRY_CNT)
1855 			goto retry;
1856 		dev_err(dev, "%s: could not get a free command\n", __func__);
1857 		rc = -1;
1858 		goto out;
1859 	}
1860 
1861 	pr_debug("%s: afu=%p cmd=%p %d\n", __func__, afu, cmd, ctx_hndl_u);
1862 
1863 	memset(cmd->rcb.cdb, 0, sizeof(cmd->rcb.cdb));
1864 
1865 	cmd->rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
1866 	cmd->rcb.port_sel = 0x0;	/* NA */
1867 	cmd->rcb.lun_id = 0x0;	/* NA */
1868 	cmd->rcb.data_len = 0x0;
1869 	cmd->rcb.data_ea = 0x0;
1870 	cmd->rcb.timeout = MC_AFU_SYNC_TIMEOUT;
1871 
1872 	cmd->rcb.cdb[0] = 0xC0;	/* AFU Sync */
1873 	cmd->rcb.cdb[1] = mode;
1874 
1875 	/* The cdb is aligned, no unaligned accessors required */
1876 	*((__be16 *)&cmd->rcb.cdb[2]) = cpu_to_be16(ctx_hndl_u);
1877 	*((__be32 *)&cmd->rcb.cdb[4]) = cpu_to_be32(res_hndl_u);
1878 
1879 	rc = send_cmd(afu, cmd);
1880 	if (unlikely(rc))
1881 		goto out;
1882 
1883 	wait_resp(afu, cmd);
1884 
1885 	/* Set on timeout */
1886 	if (unlikely((cmd->sa.ioasc != 0) ||
1887 		     (cmd->sa.host_use_b[0] & B_ERROR)))
1888 		rc = -1;
1889 out:
1890 	mutex_unlock(&sync_active);
1891 	if (cmd)
1892 		cmd_checkin(cmd);
1893 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1894 	return rc;
1895 }
1896 
1897 /**
1898  * afu_reset() - resets the AFU
1899  * @cfg:	Internal structure associated with the host.
1900  *
1901  * Return: 0 on success, -errno on failure
1902  */
1903 static int afu_reset(struct cxlflash_cfg *cfg)
1904 {
1905 	int rc = 0;
1906 	/* Stop the context before the reset. Since the context is
1907 	 * no longer available restart it after the reset is complete
1908 	 */
1909 
1910 	term_afu(cfg);
1911 
1912 	rc = init_afu(cfg);
1913 
1914 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1915 	return rc;
1916 }
1917 
1918 /**
1919  * cxlflash_eh_device_reset_handler() - reset a single LUN
1920  * @scp:	SCSI command to send.
1921  *
1922  * Return:
1923  *	SUCCESS as defined in scsi/scsi.h
1924  *	FAILED as defined in scsi/scsi.h
1925  */
1926 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
1927 {
1928 	int rc = SUCCESS;
1929 	struct Scsi_Host *host = scp->device->host;
1930 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
1931 	struct afu *afu = cfg->afu;
1932 	int rcr = 0;
1933 
1934 	pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
1935 		 "cdb=(%08X-%08X-%08X-%08X)\n", __func__, scp,
1936 		 host->host_no, scp->device->channel,
1937 		 scp->device->id, scp->device->lun,
1938 		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
1939 		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
1940 		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
1941 		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
1942 
1943 retry:
1944 	switch (cfg->state) {
1945 	case STATE_NORMAL:
1946 		rcr = send_tmf(afu, scp, TMF_LUN_RESET);
1947 		if (unlikely(rcr))
1948 			rc = FAILED;
1949 		break;
1950 	case STATE_RESET:
1951 		wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
1952 		goto retry;
1953 	default:
1954 		rc = FAILED;
1955 		break;
1956 	}
1957 
1958 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1959 	return rc;
1960 }
1961 
1962 /**
1963  * cxlflash_eh_host_reset_handler() - reset the host adapter
1964  * @scp:	SCSI command from stack identifying host.
1965  *
1966  * Return:
1967  *	SUCCESS as defined in scsi/scsi.h
1968  *	FAILED as defined in scsi/scsi.h
1969  */
1970 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
1971 {
1972 	int rc = SUCCESS;
1973 	int rcr = 0;
1974 	struct Scsi_Host *host = scp->device->host;
1975 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
1976 
1977 	pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
1978 		 "cdb=(%08X-%08X-%08X-%08X)\n", __func__, scp,
1979 		 host->host_no, scp->device->channel,
1980 		 scp->device->id, scp->device->lun,
1981 		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
1982 		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
1983 		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
1984 		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
1985 
1986 	switch (cfg->state) {
1987 	case STATE_NORMAL:
1988 		cfg->state = STATE_RESET;
1989 		cxlflash_mark_contexts_error(cfg);
1990 		rcr = afu_reset(cfg);
1991 		if (rcr) {
1992 			rc = FAILED;
1993 			cfg->state = STATE_FAILTERM;
1994 		} else
1995 			cfg->state = STATE_NORMAL;
1996 		wake_up_all(&cfg->reset_waitq);
1997 		break;
1998 	case STATE_RESET:
1999 		wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2000 		if (cfg->state == STATE_NORMAL)
2001 			break;
2002 		/* fall through */
2003 	default:
2004 		rc = FAILED;
2005 		break;
2006 	}
2007 
2008 	pr_debug("%s: returning rc=%d\n", __func__, rc);
2009 	return rc;
2010 }
2011 
2012 /**
2013  * cxlflash_change_queue_depth() - change the queue depth for the device
2014  * @sdev:	SCSI device destined for queue depth change.
2015  * @qdepth:	Requested queue depth value to set.
2016  *
2017  * The requested queue depth is capped to the maximum supported value.
2018  *
2019  * Return: The actual queue depth set.
2020  */
2021 static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
2022 {
2023 
2024 	if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
2025 		qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
2026 
2027 	scsi_change_queue_depth(sdev, qdepth);
2028 	return sdev->queue_depth;
2029 }
2030 
2031 /**
2032  * cxlflash_show_port_status() - queries and presents the current port status
2033  * @port:	Desired port for status reporting.
2034  * @afu:	AFU owning the specified port.
2035  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2036  *
2037  * Return: The size of the ASCII string returned in @buf.
2038  */
2039 static ssize_t cxlflash_show_port_status(u32 port, struct afu *afu, char *buf)
2040 {
2041 	char *disp_status;
2042 	u64 status;
2043 	__be64 __iomem *fc_regs;
2044 
2045 	if (port >= NUM_FC_PORTS)
2046 		return 0;
2047 
2048 	fc_regs = &afu->afu_map->global.fc_regs[port][0];
2049 	status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
2050 	status &= FC_MTIP_STATUS_MASK;
2051 
2052 	if (status == FC_MTIP_STATUS_ONLINE)
2053 		disp_status = "online";
2054 	else if (status == FC_MTIP_STATUS_OFFLINE)
2055 		disp_status = "offline";
2056 	else
2057 		disp_status = "unknown";
2058 
2059 	return scnprintf(buf, PAGE_SIZE, "%s\n", disp_status);
2060 }
2061 
2062 /**
2063  * port0_show() - queries and presents the current status of port 0
2064  * @dev:	Generic device associated with the host owning the port.
2065  * @attr:	Device attribute representing the port.
2066  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2067  *
2068  * Return: The size of the ASCII string returned in @buf.
2069  */
2070 static ssize_t port0_show(struct device *dev,
2071 			  struct device_attribute *attr,
2072 			  char *buf)
2073 {
2074 	struct Scsi_Host *shost = class_to_shost(dev);
2075 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
2076 	struct afu *afu = cfg->afu;
2077 
2078 	return cxlflash_show_port_status(0, afu, buf);
2079 }
2080 
2081 /**
2082  * port1_show() - queries and presents the current status of port 1
2083  * @dev:	Generic device associated with the host owning the port.
2084  * @attr:	Device attribute representing the port.
2085  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2086  *
2087  * Return: The size of the ASCII string returned in @buf.
2088  */
2089 static ssize_t port1_show(struct device *dev,
2090 			  struct device_attribute *attr,
2091 			  char *buf)
2092 {
2093 	struct Scsi_Host *shost = class_to_shost(dev);
2094 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
2095 	struct afu *afu = cfg->afu;
2096 
2097 	return cxlflash_show_port_status(1, afu, buf);
2098 }
2099 
2100 /**
2101  * lun_mode_show() - presents the current LUN mode of the host
2102  * @dev:	Generic device associated with the host.
2103  * @attr:	Device attribute representing the LUN mode.
2104  * @buf:	Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2105  *
2106  * Return: The size of the ASCII string returned in @buf.
2107  */
2108 static ssize_t lun_mode_show(struct device *dev,
2109 			     struct device_attribute *attr, char *buf)
2110 {
2111 	struct Scsi_Host *shost = class_to_shost(dev);
2112 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
2113 	struct afu *afu = cfg->afu;
2114 
2115 	return scnprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
2116 }
2117 
2118 /**
2119  * lun_mode_store() - sets the LUN mode of the host
2120  * @dev:	Generic device associated with the host.
2121  * @attr:	Device attribute representing the LUN mode.
2122  * @buf:	Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2123  * @count:	Length of data resizing in @buf.
2124  *
2125  * The CXL Flash AFU supports a dummy LUN mode where the external
2126  * links and storage are not required. Space on the FPGA is used
2127  * to create 1 or 2 small LUNs which are presented to the system
2128  * as if they were a normal storage device. This feature is useful
2129  * during development and also provides manufacturing with a way
2130  * to test the AFU without an actual device.
2131  *
2132  * 0 = external LUN[s] (default)
2133  * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2134  * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2135  * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2136  * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2137  *
2138  * Return: The size of the ASCII string returned in @buf.
2139  */
2140 static ssize_t lun_mode_store(struct device *dev,
2141 			      struct device_attribute *attr,
2142 			      const char *buf, size_t count)
2143 {
2144 	struct Scsi_Host *shost = class_to_shost(dev);
2145 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
2146 	struct afu *afu = cfg->afu;
2147 	int rc;
2148 	u32 lun_mode;
2149 
2150 	rc = kstrtouint(buf, 10, &lun_mode);
2151 	if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
2152 		afu->internal_lun = lun_mode;
2153 
2154 		/*
2155 		 * When configured for internal LUN, there is only one channel,
2156 		 * channel number 0, else there will be 2 (default).
2157 		 */
2158 		if (afu->internal_lun)
2159 			shost->max_channel = 0;
2160 		else
2161 			shost->max_channel = NUM_FC_PORTS - 1;
2162 
2163 		afu_reset(cfg);
2164 		scsi_scan_host(cfg->host);
2165 	}
2166 
2167 	return count;
2168 }
2169 
2170 /**
2171  * ioctl_version_show() - presents the current ioctl version of the host
2172  * @dev:	Generic device associated with the host.
2173  * @attr:	Device attribute representing the ioctl version.
2174  * @buf:	Buffer of length PAGE_SIZE to report back the ioctl version.
2175  *
2176  * Return: The size of the ASCII string returned in @buf.
2177  */
2178 static ssize_t ioctl_version_show(struct device *dev,
2179 				  struct device_attribute *attr, char *buf)
2180 {
2181 	return scnprintf(buf, PAGE_SIZE, "%u\n", DK_CXLFLASH_VERSION_0);
2182 }
2183 
2184 /**
2185  * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2186  * @port:	Desired port for status reporting.
2187  * @afu:	AFU owning the specified port.
2188  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2189  *
2190  * Return: The size of the ASCII string returned in @buf.
2191  */
2192 static ssize_t cxlflash_show_port_lun_table(u32 port,
2193 					    struct afu *afu,
2194 					    char *buf)
2195 {
2196 	int i;
2197 	ssize_t bytes = 0;
2198 	__be64 __iomem *fc_port;
2199 
2200 	if (port >= NUM_FC_PORTS)
2201 		return 0;
2202 
2203 	fc_port = &afu->afu_map->global.fc_port[port][0];
2204 
2205 	for (i = 0; i < CXLFLASH_NUM_VLUNS; i++)
2206 		bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2207 				   "%03d: %016llX\n", i, readq_be(&fc_port[i]));
2208 	return bytes;
2209 }
2210 
2211 /**
2212  * port0_lun_table_show() - presents the current LUN table of port 0
2213  * @dev:	Generic device associated with the host owning the port.
2214  * @attr:	Device attribute representing the port.
2215  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2216  *
2217  * Return: The size of the ASCII string returned in @buf.
2218  */
2219 static ssize_t port0_lun_table_show(struct device *dev,
2220 				    struct device_attribute *attr,
2221 				    char *buf)
2222 {
2223 	struct Scsi_Host *shost = class_to_shost(dev);
2224 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
2225 	struct afu *afu = cfg->afu;
2226 
2227 	return cxlflash_show_port_lun_table(0, afu, buf);
2228 }
2229 
2230 /**
2231  * port1_lun_table_show() - presents the current LUN table of port 1
2232  * @dev:	Generic device associated with the host owning the port.
2233  * @attr:	Device attribute representing the port.
2234  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
2235  *
2236  * Return: The size of the ASCII string returned in @buf.
2237  */
2238 static ssize_t port1_lun_table_show(struct device *dev,
2239 				    struct device_attribute *attr,
2240 				    char *buf)
2241 {
2242 	struct Scsi_Host *shost = class_to_shost(dev);
2243 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
2244 	struct afu *afu = cfg->afu;
2245 
2246 	return cxlflash_show_port_lun_table(1, afu, buf);
2247 }
2248 
2249 /**
2250  * mode_show() - presents the current mode of the device
2251  * @dev:	Generic device associated with the device.
2252  * @attr:	Device attribute representing the device mode.
2253  * @buf:	Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
2254  *
2255  * Return: The size of the ASCII string returned in @buf.
2256  */
2257 static ssize_t mode_show(struct device *dev,
2258 			 struct device_attribute *attr, char *buf)
2259 {
2260 	struct scsi_device *sdev = to_scsi_device(dev);
2261 
2262 	return scnprintf(buf, PAGE_SIZE, "%s\n",
2263 			 sdev->hostdata ? "superpipe" : "legacy");
2264 }
2265 
2266 /*
2267  * Host attributes
2268  */
2269 static DEVICE_ATTR_RO(port0);
2270 static DEVICE_ATTR_RO(port1);
2271 static DEVICE_ATTR_RW(lun_mode);
2272 static DEVICE_ATTR_RO(ioctl_version);
2273 static DEVICE_ATTR_RO(port0_lun_table);
2274 static DEVICE_ATTR_RO(port1_lun_table);
2275 
2276 static struct device_attribute *cxlflash_host_attrs[] = {
2277 	&dev_attr_port0,
2278 	&dev_attr_port1,
2279 	&dev_attr_lun_mode,
2280 	&dev_attr_ioctl_version,
2281 	&dev_attr_port0_lun_table,
2282 	&dev_attr_port1_lun_table,
2283 	NULL
2284 };
2285 
2286 /*
2287  * Device attributes
2288  */
2289 static DEVICE_ATTR_RO(mode);
2290 
2291 static struct device_attribute *cxlflash_dev_attrs[] = {
2292 	&dev_attr_mode,
2293 	NULL
2294 };
2295 
2296 /*
2297  * Host template
2298  */
2299 static struct scsi_host_template driver_template = {
2300 	.module = THIS_MODULE,
2301 	.name = CXLFLASH_ADAPTER_NAME,
2302 	.info = cxlflash_driver_info,
2303 	.ioctl = cxlflash_ioctl,
2304 	.proc_name = CXLFLASH_NAME,
2305 	.queuecommand = cxlflash_queuecommand,
2306 	.eh_device_reset_handler = cxlflash_eh_device_reset_handler,
2307 	.eh_host_reset_handler = cxlflash_eh_host_reset_handler,
2308 	.change_queue_depth = cxlflash_change_queue_depth,
2309 	.cmd_per_lun = CXLFLASH_MAX_CMDS_PER_LUN,
2310 	.can_queue = CXLFLASH_MAX_CMDS,
2311 	.this_id = -1,
2312 	.sg_tablesize = SG_NONE,	/* No scatter gather support */
2313 	.max_sectors = CXLFLASH_MAX_SECTORS,
2314 	.use_clustering = ENABLE_CLUSTERING,
2315 	.shost_attrs = cxlflash_host_attrs,
2316 	.sdev_attrs = cxlflash_dev_attrs,
2317 };
2318 
2319 /*
2320  * Device dependent values
2321  */
2322 static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS };
2323 static struct dev_dependent_vals dev_flash_gt_vals = { CXLFLASH_MAX_SECTORS };
2324 
2325 /*
2326  * PCI device binding table
2327  */
2328 static struct pci_device_id cxlflash_pci_table[] = {
2329 	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
2330 	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
2331 	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_FLASH_GT,
2332 	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_flash_gt_vals},
2333 	{}
2334 };
2335 
2336 MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
2337 
2338 /**
2339  * cxlflash_worker_thread() - work thread handler for the AFU
2340  * @work:	Work structure contained within cxlflash associated with host.
2341  *
2342  * Handles the following events:
2343  * - Link reset which cannot be performed on interrupt context due to
2344  * blocking up to a few seconds
2345  * - Read AFU command room
2346  * - Rescan the host
2347  */
2348 static void cxlflash_worker_thread(struct work_struct *work)
2349 {
2350 	struct cxlflash_cfg *cfg = container_of(work, struct cxlflash_cfg,
2351 						work_q);
2352 	struct afu *afu = cfg->afu;
2353 	struct device *dev = &cfg->dev->dev;
2354 	int port;
2355 	ulong lock_flags;
2356 
2357 	/* Avoid MMIO if the device has failed */
2358 
2359 	if (cfg->state != STATE_NORMAL)
2360 		return;
2361 
2362 	spin_lock_irqsave(cfg->host->host_lock, lock_flags);
2363 
2364 	if (cfg->lr_state == LINK_RESET_REQUIRED) {
2365 		port = cfg->lr_port;
2366 		if (port < 0)
2367 			dev_err(dev, "%s: invalid port index %d\n",
2368 				__func__, port);
2369 		else {
2370 			spin_unlock_irqrestore(cfg->host->host_lock,
2371 					       lock_flags);
2372 
2373 			/* The reset can block... */
2374 			afu_link_reset(afu, port,
2375 				       &afu->afu_map->global.fc_regs[port][0]);
2376 			spin_lock_irqsave(cfg->host->host_lock, lock_flags);
2377 		}
2378 
2379 		cfg->lr_state = LINK_RESET_COMPLETE;
2380 	}
2381 
2382 	if (afu->read_room) {
2383 		atomic64_set(&afu->room, readq_be(&afu->host_map->cmd_room));
2384 		afu->read_room = false;
2385 	}
2386 
2387 	spin_unlock_irqrestore(cfg->host->host_lock, lock_flags);
2388 
2389 	if (atomic_dec_if_positive(&cfg->scan_host_needed) >= 0)
2390 		scsi_scan_host(cfg->host);
2391 	kref_put(&afu->mapcount, afu_unmap);
2392 }
2393 
2394 /**
2395  * cxlflash_probe() - PCI entry point to add host
2396  * @pdev:	PCI device associated with the host.
2397  * @dev_id:	PCI device id associated with device.
2398  *
2399  * Return: 0 on success, -errno on failure
2400  */
2401 static int cxlflash_probe(struct pci_dev *pdev,
2402 			  const struct pci_device_id *dev_id)
2403 {
2404 	struct Scsi_Host *host;
2405 	struct cxlflash_cfg *cfg = NULL;
2406 	struct dev_dependent_vals *ddv;
2407 	int rc = 0;
2408 
2409 	dev_dbg(&pdev->dev, "%s: Found CXLFLASH with IRQ: %d\n",
2410 		__func__, pdev->irq);
2411 
2412 	ddv = (struct dev_dependent_vals *)dev_id->driver_data;
2413 	driver_template.max_sectors = ddv->max_sectors;
2414 
2415 	host = scsi_host_alloc(&driver_template, sizeof(struct cxlflash_cfg));
2416 	if (!host) {
2417 		dev_err(&pdev->dev, "%s: call to scsi_host_alloc failed!\n",
2418 			__func__);
2419 		rc = -ENOMEM;
2420 		goto out;
2421 	}
2422 
2423 	host->max_id = CXLFLASH_MAX_NUM_TARGETS_PER_BUS;
2424 	host->max_lun = CXLFLASH_MAX_NUM_LUNS_PER_TARGET;
2425 	host->max_channel = NUM_FC_PORTS - 1;
2426 	host->unique_id = host->host_no;
2427 	host->max_cmd_len = CXLFLASH_MAX_CDB_LEN;
2428 
2429 	cfg = (struct cxlflash_cfg *)host->hostdata;
2430 	cfg->host = host;
2431 	rc = alloc_mem(cfg);
2432 	if (rc) {
2433 		dev_err(&pdev->dev, "%s: call to alloc_mem failed!\n",
2434 			__func__);
2435 		rc = -ENOMEM;
2436 		scsi_host_put(cfg->host);
2437 		goto out;
2438 	}
2439 
2440 	cfg->init_state = INIT_STATE_NONE;
2441 	cfg->dev = pdev;
2442 	cfg->cxl_fops = cxlflash_cxl_fops;
2443 
2444 	/*
2445 	 * The promoted LUNs move to the top of the LUN table. The rest stay
2446 	 * on the bottom half. The bottom half grows from the end
2447 	 * (index = 255), whereas the top half grows from the beginning
2448 	 * (index = 0).
2449 	 */
2450 	cfg->promote_lun_index  = 0;
2451 	cfg->last_lun_index[0] = CXLFLASH_NUM_VLUNS/2 - 1;
2452 	cfg->last_lun_index[1] = CXLFLASH_NUM_VLUNS/2 - 1;
2453 
2454 	cfg->dev_id = (struct pci_device_id *)dev_id;
2455 
2456 	init_waitqueue_head(&cfg->tmf_waitq);
2457 	init_waitqueue_head(&cfg->reset_waitq);
2458 
2459 	INIT_WORK(&cfg->work_q, cxlflash_worker_thread);
2460 	cfg->lr_state = LINK_RESET_INVALID;
2461 	cfg->lr_port = -1;
2462 	spin_lock_init(&cfg->tmf_slock);
2463 	mutex_init(&cfg->ctx_tbl_list_mutex);
2464 	mutex_init(&cfg->ctx_recovery_mutex);
2465 	init_rwsem(&cfg->ioctl_rwsem);
2466 	INIT_LIST_HEAD(&cfg->ctx_err_recovery);
2467 	INIT_LIST_HEAD(&cfg->lluns);
2468 
2469 	pci_set_drvdata(pdev, cfg);
2470 
2471 	cfg->cxl_afu = cxl_pci_to_afu(pdev);
2472 
2473 	rc = init_pci(cfg);
2474 	if (rc) {
2475 		dev_err(&pdev->dev, "%s: call to init_pci "
2476 			"failed rc=%d!\n", __func__, rc);
2477 		goto out_remove;
2478 	}
2479 	cfg->init_state = INIT_STATE_PCI;
2480 
2481 	rc = init_afu(cfg);
2482 	if (rc) {
2483 		dev_err(&pdev->dev, "%s: call to init_afu "
2484 			"failed rc=%d!\n", __func__, rc);
2485 		goto out_remove;
2486 	}
2487 	cfg->init_state = INIT_STATE_AFU;
2488 
2489 	rc = init_scsi(cfg);
2490 	if (rc) {
2491 		dev_err(&pdev->dev, "%s: call to init_scsi "
2492 			"failed rc=%d!\n", __func__, rc);
2493 		goto out_remove;
2494 	}
2495 	cfg->init_state = INIT_STATE_SCSI;
2496 
2497 out:
2498 	pr_debug("%s: returning rc=%d\n", __func__, rc);
2499 	return rc;
2500 
2501 out_remove:
2502 	cxlflash_remove(pdev);
2503 	goto out;
2504 }
2505 
2506 /**
2507  * drain_ioctls() - wait until all currently executing ioctls have completed
2508  * @cfg:	Internal structure associated with the host.
2509  *
2510  * Obtain write access to read/write semaphore that wraps ioctl
2511  * handling to 'drain' ioctls currently executing.
2512  */
2513 static void drain_ioctls(struct cxlflash_cfg *cfg)
2514 {
2515 	down_write(&cfg->ioctl_rwsem);
2516 	up_write(&cfg->ioctl_rwsem);
2517 }
2518 
2519 /**
2520  * cxlflash_pci_error_detected() - called when a PCI error is detected
2521  * @pdev:	PCI device struct.
2522  * @state:	PCI channel state.
2523  *
2524  * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
2525  */
2526 static pci_ers_result_t cxlflash_pci_error_detected(struct pci_dev *pdev,
2527 						    pci_channel_state_t state)
2528 {
2529 	int rc = 0;
2530 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
2531 	struct device *dev = &cfg->dev->dev;
2532 
2533 	dev_dbg(dev, "%s: pdev=%p state=%u\n", __func__, pdev, state);
2534 
2535 	switch (state) {
2536 	case pci_channel_io_frozen:
2537 		cfg->state = STATE_RESET;
2538 		scsi_block_requests(cfg->host);
2539 		drain_ioctls(cfg);
2540 		rc = cxlflash_mark_contexts_error(cfg);
2541 		if (unlikely(rc))
2542 			dev_err(dev, "%s: Failed to mark user contexts!(%d)\n",
2543 				__func__, rc);
2544 		term_afu(cfg);
2545 		return PCI_ERS_RESULT_NEED_RESET;
2546 	case pci_channel_io_perm_failure:
2547 		cfg->state = STATE_FAILTERM;
2548 		wake_up_all(&cfg->reset_waitq);
2549 		scsi_unblock_requests(cfg->host);
2550 		return PCI_ERS_RESULT_DISCONNECT;
2551 	default:
2552 		break;
2553 	}
2554 	return PCI_ERS_RESULT_NEED_RESET;
2555 }
2556 
2557 /**
2558  * cxlflash_pci_slot_reset() - called when PCI slot has been reset
2559  * @pdev:	PCI device struct.
2560  *
2561  * This routine is called by the pci error recovery code after the PCI
2562  * slot has been reset, just before we should resume normal operations.
2563  *
2564  * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
2565  */
2566 static pci_ers_result_t cxlflash_pci_slot_reset(struct pci_dev *pdev)
2567 {
2568 	int rc = 0;
2569 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
2570 	struct device *dev = &cfg->dev->dev;
2571 
2572 	dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
2573 
2574 	rc = init_afu(cfg);
2575 	if (unlikely(rc)) {
2576 		dev_err(dev, "%s: EEH recovery failed! (%d)\n", __func__, rc);
2577 		return PCI_ERS_RESULT_DISCONNECT;
2578 	}
2579 
2580 	return PCI_ERS_RESULT_RECOVERED;
2581 }
2582 
2583 /**
2584  * cxlflash_pci_resume() - called when normal operation can resume
2585  * @pdev:	PCI device struct
2586  */
2587 static void cxlflash_pci_resume(struct pci_dev *pdev)
2588 {
2589 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
2590 	struct device *dev = &cfg->dev->dev;
2591 
2592 	dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
2593 
2594 	cfg->state = STATE_NORMAL;
2595 	wake_up_all(&cfg->reset_waitq);
2596 	scsi_unblock_requests(cfg->host);
2597 }
2598 
2599 static const struct pci_error_handlers cxlflash_err_handler = {
2600 	.error_detected = cxlflash_pci_error_detected,
2601 	.slot_reset = cxlflash_pci_slot_reset,
2602 	.resume = cxlflash_pci_resume,
2603 };
2604 
2605 /*
2606  * PCI device structure
2607  */
2608 static struct pci_driver cxlflash_driver = {
2609 	.name = CXLFLASH_NAME,
2610 	.id_table = cxlflash_pci_table,
2611 	.probe = cxlflash_probe,
2612 	.remove = cxlflash_remove,
2613 	.err_handler = &cxlflash_err_handler,
2614 };
2615 
2616 /**
2617  * init_cxlflash() - module entry point
2618  *
2619  * Return: 0 on success, -errno on failure
2620  */
2621 static int __init init_cxlflash(void)
2622 {
2623 	pr_info("%s: %s\n", __func__, CXLFLASH_ADAPTER_NAME);
2624 
2625 	cxlflash_list_init();
2626 
2627 	return pci_register_driver(&cxlflash_driver);
2628 }
2629 
2630 /**
2631  * exit_cxlflash() - module exit point
2632  */
2633 static void __exit exit_cxlflash(void)
2634 {
2635 	cxlflash_term_global_luns();
2636 	cxlflash_free_errpage();
2637 
2638 	pci_unregister_driver(&cxlflash_driver);
2639 }
2640 
2641 module_init(init_cxlflash);
2642 module_exit(exit_cxlflash);
2643