xref: /linux/drivers/scsi/cxlflash/main.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 /**
39  * cxlflash_cmd_checkout() - checks out an AFU command
40  * @afu:	AFU to checkout from.
41  *
42  * Commands are checked out in a round-robin fashion. Note that since
43  * the command pool is larger than the hardware queue, the majority of
44  * times we will only loop once or twice before getting a command. The
45  * buffer and CDB within the command are initialized (zeroed) prior to
46  * returning.
47  *
48  * Return: The checked out command or NULL when command pool is empty.
49  */
50 struct afu_cmd *cxlflash_cmd_checkout(struct afu *afu)
51 {
52 	int k, dec = CXLFLASH_NUM_CMDS;
53 	struct afu_cmd *cmd;
54 
55 	while (dec--) {
56 		k = (afu->cmd_couts++ & (CXLFLASH_NUM_CMDS - 1));
57 
58 		cmd = &afu->cmd[k];
59 
60 		if (!atomic_dec_if_positive(&cmd->free)) {
61 			pr_debug("%s: returning found index=%d\n",
62 				 __func__, cmd->slot);
63 			memset(cmd->buf, 0, CMD_BUFSIZE);
64 			memset(cmd->rcb.cdb, 0, sizeof(cmd->rcb.cdb));
65 			return cmd;
66 		}
67 	}
68 
69 	return NULL;
70 }
71 
72 /**
73  * cxlflash_cmd_checkin() - checks in an AFU command
74  * @cmd:	AFU command to checkin.
75  *
76  * Safe to pass commands that have already been checked in. Several
77  * internal tracking fields are reset as part of the checkin. Note
78  * that these are intentionally reset prior to toggling the free bit
79  * to avoid clobbering values in the event that the command is checked
80  * out right away.
81  */
82 void cxlflash_cmd_checkin(struct afu_cmd *cmd)
83 {
84 	cmd->rcb.scp = NULL;
85 	cmd->rcb.timeout = 0;
86 	cmd->sa.ioasc = 0;
87 	cmd->cmd_tmf = false;
88 	cmd->sa.host_use[0] = 0; /* clears both completion and retry bytes */
89 
90 	if (unlikely(atomic_inc_return(&cmd->free) != 1)) {
91 		pr_err("%s: Freeing cmd (%d) that is not in use!\n",
92 		       __func__, cmd->slot);
93 		return;
94 	}
95 
96 	pr_debug("%s: released cmd %p index=%d\n", __func__, cmd, cmd->slot);
97 }
98 
99 /**
100  * process_cmd_err() - command error handler
101  * @cmd:	AFU command that experienced the error.
102  * @scp:	SCSI command associated with the AFU command in error.
103  *
104  * Translates error bits from AFU command to SCSI command results.
105  */
106 static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
107 {
108 	struct sisl_ioarcb *ioarcb;
109 	struct sisl_ioasa *ioasa;
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 		pr_debug("%s: cmd underrun cmd = %p scp = %p\n",
119 			 __func__, cmd, scp);
120 		scp->result = (DID_ERROR << 16);
121 	}
122 
123 	if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
124 		pr_debug("%s: cmd underrun cmd = %p scp = %p\n",
125 			 __func__, cmd, scp);
126 		scp->result = (DID_ERROR << 16);
127 	}
128 
129 	pr_debug("%s: cmd failed afu_rc=%d scsi_rc=%d fc_rc=%d "
130 		 "afu_extra=0x%X, scsi_entra=0x%X, fc_extra=0x%X\n",
131 		 __func__, ioasa->rc.afu_rc, ioasa->rc.scsi_rc,
132 		 ioasa->rc.fc_rc, ioasa->afu_extra, ioasa->scsi_extra,
133 		 ioasa->fc_extra);
134 
135 	if (ioasa->rc.scsi_rc) {
136 		/* We have a SCSI status */
137 		if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
138 			memcpy(scp->sense_buffer, ioasa->sense_data,
139 			       SISL_SENSE_DATA_LEN);
140 			scp->result = ioasa->rc.scsi_rc;
141 		} else
142 			scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
143 	}
144 
145 	/*
146 	 * We encountered an error. Set scp->result based on nature
147 	 * of error.
148 	 */
149 	if (ioasa->rc.fc_rc) {
150 		/* We have an FC status */
151 		switch (ioasa->rc.fc_rc) {
152 		case SISL_FC_RC_LINKDOWN:
153 			scp->result = (DID_REQUEUE << 16);
154 			break;
155 		case SISL_FC_RC_RESID:
156 			/* This indicates an FCP resid underrun */
157 			if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
158 				/* If the SISL_RC_FLAGS_OVERRUN flag was set,
159 				 * then we will handle this error else where.
160 				 * If not then we must handle it here.
161 				 * This is probably an AFU bug. We will
162 				 * attempt a retry to see if that resolves it.
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_MEDIUM_ERROR << 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 	u32 resid;
221 	ulong lock_flags;
222 	struct afu *afu = cmd->parent;
223 	struct cxlflash_cfg *cfg = afu->parent;
224 	bool cmd_is_tmf;
225 
226 	spin_lock_irqsave(&cmd->slock, lock_flags);
227 	cmd->sa.host_use_b[0] |= B_DONE;
228 	spin_unlock_irqrestore(&cmd->slock, lock_flags);
229 
230 	if (cmd->rcb.scp) {
231 		scp = cmd->rcb.scp;
232 		if (unlikely(cmd->sa.rc.afu_rc ||
233 			     cmd->sa.rc.scsi_rc ||
234 			     cmd->sa.rc.fc_rc))
235 			process_cmd_err(cmd, scp);
236 		else
237 			scp->result = (DID_OK << 16);
238 
239 		resid = cmd->sa.resid;
240 		cmd_is_tmf = cmd->cmd_tmf;
241 		cxlflash_cmd_checkin(cmd); /* Don't use cmd after here */
242 
243 		pr_debug("%s: calling scsi_set_resid, scp=%p "
244 			 "result=%X resid=%d\n", __func__,
245 			 scp, scp->result, resid);
246 
247 		scsi_set_resid(scp, resid);
248 		scsi_dma_unmap(scp);
249 		scp->scsi_done(scp);
250 
251 		if (cmd_is_tmf) {
252 			spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
253 			cfg->tmf_active = false;
254 			wake_up_all_locked(&cfg->tmf_waitq);
255 			spin_unlock_irqrestore(&cfg->tmf_waitq.lock,
256 					       lock_flags);
257 		}
258 	} else
259 		complete(&cmd->cevent);
260 }
261 
262 /**
263  * send_tmf() - sends a Task Management Function (TMF)
264  * @afu:	AFU to checkout from.
265  * @scp:	SCSI command from stack.
266  * @tmfcmd:	TMF command to send.
267  *
268  * Return:
269  *	0 on success
270  *	SCSI_MLQUEUE_HOST_BUSY when host is busy
271  */
272 static int send_tmf(struct afu *afu, struct scsi_cmnd *scp, u64 tmfcmd)
273 {
274 	struct afu_cmd *cmd;
275 
276 	u32 port_sel = scp->device->channel + 1;
277 	short lflag = 0;
278 	struct Scsi_Host *host = scp->device->host;
279 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
280 	ulong lock_flags;
281 	int rc = 0;
282 
283 	cmd = cxlflash_cmd_checkout(afu);
284 	if (unlikely(!cmd)) {
285 		pr_err("%s: could not get a free command\n", __func__);
286 		rc = SCSI_MLQUEUE_HOST_BUSY;
287 		goto out;
288 	}
289 
290 	/* If a Task Management Function is active, do not send one more.
291 	 */
292 	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
293 	if (cfg->tmf_active)
294 		wait_event_interruptible_locked_irq(cfg->tmf_waitq,
295 						    !cfg->tmf_active);
296 	cfg->tmf_active = true;
297 	cmd->cmd_tmf = true;
298 	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
299 
300 	cmd->rcb.ctx_id = afu->ctx_hndl;
301 	cmd->rcb.port_sel = port_sel;
302 	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
303 
304 	lflag = SISL_REQ_FLAGS_TMF_CMD;
305 
306 	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
307 			      SISL_REQ_FLAGS_SUP_UNDERRUN | lflag);
308 
309 	/* Stash the scp in the reserved field, for reuse during interrupt */
310 	cmd->rcb.scp = scp;
311 
312 	/* Copy the CDB from the cmd passed in */
313 	memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
314 
315 	/* Send the command */
316 	rc = cxlflash_send_cmd(afu, cmd);
317 	if (unlikely(rc)) {
318 		cxlflash_cmd_checkin(cmd);
319 		spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
320 		cfg->tmf_active = false;
321 		spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
322 		goto out;
323 	}
324 
325 	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
326 	wait_event_interruptible_locked_irq(cfg->tmf_waitq, !cfg->tmf_active);
327 	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
328 out:
329 	return rc;
330 }
331 
332 /**
333  * cxlflash_driver_info() - information handler for this host driver
334  * @host:	SCSI host associated with device.
335  *
336  * Return: A string describing the device.
337  */
338 static const char *cxlflash_driver_info(struct Scsi_Host *host)
339 {
340 	return CXLFLASH_ADAPTER_NAME;
341 }
342 
343 /**
344  * cxlflash_queuecommand() - sends a mid-layer request
345  * @host:	SCSI host associated with device.
346  * @scp:	SCSI command to send.
347  *
348  * Return:
349  *	0 on success
350  *	SCSI_MLQUEUE_HOST_BUSY when host is busy
351  */
352 static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
353 {
354 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
355 	struct afu *afu = cfg->afu;
356 	struct pci_dev *pdev = cfg->dev;
357 	struct afu_cmd *cmd;
358 	u32 port_sel = scp->device->channel + 1;
359 	int nseg, i, ncount;
360 	struct scatterlist *sg;
361 	ulong lock_flags;
362 	short lflag = 0;
363 	int rc = 0;
364 
365 	pr_debug("%s: (scp=%p) %d/%d/%d/%llu cdb=(%08X-%08X-%08X-%08X)\n",
366 		 __func__, scp, host->host_no, scp->device->channel,
367 		 scp->device->id, scp->device->lun,
368 		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
369 		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
370 		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
371 		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
372 
373 	/* If a Task Management Function is active, wait for it to complete
374 	 * before continuing with regular commands.
375 	 */
376 	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
377 	if (cfg->tmf_active) {
378 		spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
379 		rc = SCSI_MLQUEUE_HOST_BUSY;
380 		goto out;
381 	}
382 	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
383 
384 	switch (cfg->state) {
385 	case STATE_LIMBO:
386 		dev_dbg_ratelimited(&cfg->dev->dev, "%s: device in limbo!\n",
387 				    __func__);
388 		rc = SCSI_MLQUEUE_HOST_BUSY;
389 		goto out;
390 	case STATE_FAILTERM:
391 		dev_dbg_ratelimited(&cfg->dev->dev, "%s: device has failed!\n",
392 				    __func__);
393 		scp->result = (DID_NO_CONNECT << 16);
394 		scp->scsi_done(scp);
395 		rc = 0;
396 		goto out;
397 	default:
398 		break;
399 	}
400 
401 	cmd = cxlflash_cmd_checkout(afu);
402 	if (unlikely(!cmd)) {
403 		pr_err("%s: could not get a free command\n", __func__);
404 		rc = SCSI_MLQUEUE_HOST_BUSY;
405 		goto out;
406 	}
407 
408 	cmd->rcb.ctx_id = afu->ctx_hndl;
409 	cmd->rcb.port_sel = port_sel;
410 	cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
411 
412 	if (scp->sc_data_direction == DMA_TO_DEVICE)
413 		lflag = SISL_REQ_FLAGS_HOST_WRITE;
414 	else
415 		lflag = SISL_REQ_FLAGS_HOST_READ;
416 
417 	cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
418 			      SISL_REQ_FLAGS_SUP_UNDERRUN | lflag);
419 
420 	/* Stash the scp in the reserved field, for reuse during interrupt */
421 	cmd->rcb.scp = scp;
422 
423 	nseg = scsi_dma_map(scp);
424 	if (unlikely(nseg < 0)) {
425 		dev_err(&pdev->dev, "%s: Fail DMA map! nseg=%d\n",
426 			__func__, nseg);
427 		rc = SCSI_MLQUEUE_HOST_BUSY;
428 		goto out;
429 	}
430 
431 	ncount = scsi_sg_count(scp);
432 	scsi_for_each_sg(scp, sg, ncount, i) {
433 		cmd->rcb.data_len = sg_dma_len(sg);
434 		cmd->rcb.data_ea = sg_dma_address(sg);
435 	}
436 
437 	/* Copy the CDB from the scsi_cmnd passed in */
438 	memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
439 
440 	/* Send the command */
441 	rc = cxlflash_send_cmd(afu, cmd);
442 	if (unlikely(rc)) {
443 		cxlflash_cmd_checkin(cmd);
444 		scsi_dma_unmap(scp);
445 	}
446 
447 out:
448 	return rc;
449 }
450 
451 /**
452  * cxlflash_eh_device_reset_handler() - reset a single LUN
453  * @scp:	SCSI command to send.
454  *
455  * Return:
456  *	SUCCESS as defined in scsi/scsi.h
457  *	FAILED as defined in scsi/scsi.h
458  */
459 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
460 {
461 	int rc = SUCCESS;
462 	struct Scsi_Host *host = scp->device->host;
463 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
464 	struct afu *afu = cfg->afu;
465 	int rcr = 0;
466 
467 	pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
468 		 "cdb=(%08X-%08X-%08X-%08X)\n", __func__, scp,
469 		 host->host_no, scp->device->channel,
470 		 scp->device->id, scp->device->lun,
471 		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
472 		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
473 		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
474 		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
475 
476 	switch (cfg->state) {
477 	case STATE_NORMAL:
478 		rcr = send_tmf(afu, scp, TMF_LUN_RESET);
479 		if (unlikely(rcr))
480 			rc = FAILED;
481 		break;
482 	case STATE_LIMBO:
483 		wait_event(cfg->limbo_waitq, cfg->state != STATE_LIMBO);
484 		if (cfg->state == STATE_NORMAL)
485 			break;
486 		/* fall through */
487 	default:
488 		rc = FAILED;
489 		break;
490 	}
491 
492 	pr_debug("%s: returning rc=%d\n", __func__, rc);
493 	return rc;
494 }
495 
496 /**
497  * cxlflash_eh_host_reset_handler() - reset the host adapter
498  * @scp:	SCSI command from stack identifying host.
499  *
500  * Return:
501  *	SUCCESS as defined in scsi/scsi.h
502  *	FAILED as defined in scsi/scsi.h
503  */
504 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
505 {
506 	int rc = SUCCESS;
507 	int rcr = 0;
508 	struct Scsi_Host *host = scp->device->host;
509 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)host->hostdata;
510 
511 	pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
512 		 "cdb=(%08X-%08X-%08X-%08X)\n", __func__, scp,
513 		 host->host_no, scp->device->channel,
514 		 scp->device->id, scp->device->lun,
515 		 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
516 		 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
517 		 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
518 		 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
519 
520 	switch (cfg->state) {
521 	case STATE_NORMAL:
522 		cfg->state = STATE_LIMBO;
523 		scsi_block_requests(cfg->host);
524 		cxlflash_mark_contexts_error(cfg);
525 		rcr = cxlflash_afu_reset(cfg);
526 		if (rcr) {
527 			rc = FAILED;
528 			cfg->state = STATE_FAILTERM;
529 		} else
530 			cfg->state = STATE_NORMAL;
531 		wake_up_all(&cfg->limbo_waitq);
532 		scsi_unblock_requests(cfg->host);
533 		break;
534 	case STATE_LIMBO:
535 		wait_event(cfg->limbo_waitq, cfg->state != STATE_LIMBO);
536 		if (cfg->state == STATE_NORMAL)
537 			break;
538 		/* fall through */
539 	default:
540 		rc = FAILED;
541 		break;
542 	}
543 
544 	pr_debug("%s: returning rc=%d\n", __func__, rc);
545 	return rc;
546 }
547 
548 /**
549  * cxlflash_change_queue_depth() - change the queue depth for the device
550  * @sdev:	SCSI device destined for queue depth change.
551  * @qdepth:	Requested queue depth value to set.
552  *
553  * The requested queue depth is capped to the maximum supported value.
554  *
555  * Return: The actual queue depth set.
556  */
557 static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
558 {
559 
560 	if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
561 		qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
562 
563 	scsi_change_queue_depth(sdev, qdepth);
564 	return sdev->queue_depth;
565 }
566 
567 /**
568  * cxlflash_show_port_status() - queries and presents the current port status
569  * @dev:	Generic device associated with the host owning the port.
570  * @attr:	Device attribute representing the port.
571  * @buf:	Buffer of length PAGE_SIZE to report back port status in ASCII.
572  *
573  * Return: The size of the ASCII string returned in @buf.
574  */
575 static ssize_t cxlflash_show_port_status(struct device *dev,
576 					 struct device_attribute *attr,
577 					 char *buf)
578 {
579 	struct Scsi_Host *shost = class_to_shost(dev);
580 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
581 	struct afu *afu = cfg->afu;
582 
583 	char *disp_status;
584 	int rc;
585 	u32 port;
586 	u64 status;
587 	u64 *fc_regs;
588 
589 	rc = kstrtouint((attr->attr.name + 4), 10, &port);
590 	if (rc || (port >= NUM_FC_PORTS))
591 		return 0;
592 
593 	fc_regs = &afu->afu_map->global.fc_regs[port][0];
594 	status =
595 	    (readq_be(&fc_regs[FC_MTIP_STATUS / 8]) & FC_MTIP_STATUS_MASK);
596 
597 	if (status == FC_MTIP_STATUS_ONLINE)
598 		disp_status = "online";
599 	else if (status == FC_MTIP_STATUS_OFFLINE)
600 		disp_status = "offline";
601 	else
602 		disp_status = "unknown";
603 
604 	return snprintf(buf, PAGE_SIZE, "%s\n", disp_status);
605 }
606 
607 /**
608  * cxlflash_show_lun_mode() - presents the current LUN mode of the host
609  * @dev:	Generic device associated with the host.
610  * @attr:	Device attribute representing the lun mode.
611  * @buf:	Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
612  *
613  * Return: The size of the ASCII string returned in @buf.
614  */
615 static ssize_t cxlflash_show_lun_mode(struct device *dev,
616 				      struct device_attribute *attr, char *buf)
617 {
618 	struct Scsi_Host *shost = class_to_shost(dev);
619 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
620 	struct afu *afu = cfg->afu;
621 
622 	return snprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
623 }
624 
625 /**
626  * cxlflash_store_lun_mode() - sets the LUN mode of the host
627  * @dev:	Generic device associated with the host.
628  * @attr:	Device attribute representing the lun mode.
629  * @buf:	Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
630  * @count:	Length of data resizing in @buf.
631  *
632  * The CXL Flash AFU supports a dummy LUN mode where the external
633  * links and storage are not required. Space on the FPGA is used
634  * to create 1 or 2 small LUNs which are presented to the system
635  * as if they were a normal storage device. This feature is useful
636  * during development and also provides manufacturing with a way
637  * to test the AFU without an actual device.
638  *
639  * 0 = external LUN[s] (default)
640  * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
641  * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
642  * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
643  * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
644  *
645  * Return: The size of the ASCII string returned in @buf.
646  */
647 static ssize_t cxlflash_store_lun_mode(struct device *dev,
648 				       struct device_attribute *attr,
649 				       const char *buf, size_t count)
650 {
651 	struct Scsi_Host *shost = class_to_shost(dev);
652 	struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)shost->hostdata;
653 	struct afu *afu = cfg->afu;
654 	int rc;
655 	u32 lun_mode;
656 
657 	rc = kstrtouint(buf, 10, &lun_mode);
658 	if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
659 		afu->internal_lun = lun_mode;
660 		cxlflash_afu_reset(cfg);
661 		scsi_scan_host(cfg->host);
662 	}
663 
664 	return count;
665 }
666 
667 /**
668  * cxlflash_show_ioctl_version() - presents the current ioctl version of the host
669  * @dev:	Generic device associated with the host.
670  * @attr:	Device attribute representing the ioctl version.
671  * @buf:	Buffer of length PAGE_SIZE to report back the ioctl version.
672  *
673  * Return: The size of the ASCII string returned in @buf.
674  */
675 static ssize_t cxlflash_show_ioctl_version(struct device *dev,
676 					   struct device_attribute *attr,
677 					   char *buf)
678 {
679 	return scnprintf(buf, PAGE_SIZE, "%u\n", DK_CXLFLASH_VERSION_0);
680 }
681 
682 /**
683  * cxlflash_show_dev_mode() - presents the current mode of the device
684  * @dev:	Generic device associated with the device.
685  * @attr:	Device attribute representing the device mode.
686  * @buf:	Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
687  *
688  * Return: The size of the ASCII string returned in @buf.
689  */
690 static ssize_t cxlflash_show_dev_mode(struct device *dev,
691 				      struct device_attribute *attr, char *buf)
692 {
693 	struct scsi_device *sdev = to_scsi_device(dev);
694 
695 	return snprintf(buf, PAGE_SIZE, "%s\n",
696 			sdev->hostdata ? "superpipe" : "legacy");
697 }
698 
699 /**
700  * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
701  * @cxlflash:	Internal structure associated with the host.
702  */
703 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
704 {
705 	struct pci_dev *pdev = cfg->dev;
706 
707 	if (pci_channel_offline(pdev))
708 		wait_event_timeout(cfg->limbo_waitq,
709 				   !pci_channel_offline(pdev),
710 				   CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
711 }
712 
713 /*
714  * Host attributes
715  */
716 static DEVICE_ATTR(port0, S_IRUGO, cxlflash_show_port_status, NULL);
717 static DEVICE_ATTR(port1, S_IRUGO, cxlflash_show_port_status, NULL);
718 static DEVICE_ATTR(lun_mode, S_IRUGO | S_IWUSR, cxlflash_show_lun_mode,
719 		   cxlflash_store_lun_mode);
720 static DEVICE_ATTR(ioctl_version, S_IRUGO, cxlflash_show_ioctl_version, NULL);
721 
722 static struct device_attribute *cxlflash_host_attrs[] = {
723 	&dev_attr_port0,
724 	&dev_attr_port1,
725 	&dev_attr_lun_mode,
726 	&dev_attr_ioctl_version,
727 	NULL
728 };
729 
730 /*
731  * Device attributes
732  */
733 static DEVICE_ATTR(mode, S_IRUGO, cxlflash_show_dev_mode, NULL);
734 
735 static struct device_attribute *cxlflash_dev_attrs[] = {
736 	&dev_attr_mode,
737 	NULL
738 };
739 
740 /*
741  * Host template
742  */
743 static struct scsi_host_template driver_template = {
744 	.module = THIS_MODULE,
745 	.name = CXLFLASH_ADAPTER_NAME,
746 	.info = cxlflash_driver_info,
747 	.ioctl = cxlflash_ioctl,
748 	.proc_name = CXLFLASH_NAME,
749 	.queuecommand = cxlflash_queuecommand,
750 	.eh_device_reset_handler = cxlflash_eh_device_reset_handler,
751 	.eh_host_reset_handler = cxlflash_eh_host_reset_handler,
752 	.change_queue_depth = cxlflash_change_queue_depth,
753 	.cmd_per_lun = 16,
754 	.can_queue = CXLFLASH_MAX_CMDS,
755 	.this_id = -1,
756 	.sg_tablesize = SG_NONE,	/* No scatter gather support. */
757 	.max_sectors = CXLFLASH_MAX_SECTORS,
758 	.use_clustering = ENABLE_CLUSTERING,
759 	.shost_attrs = cxlflash_host_attrs,
760 	.sdev_attrs = cxlflash_dev_attrs,
761 };
762 
763 /*
764  * Device dependent values
765  */
766 static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS };
767 
768 /*
769  * PCI device binding table
770  */
771 static struct pci_device_id cxlflash_pci_table[] = {
772 	{PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
773 	 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
774 	{}
775 };
776 
777 MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
778 
779 /**
780  * free_mem() - free memory associated with the AFU
781  * @cxlflash:	Internal structure associated with the host.
782  */
783 static void free_mem(struct cxlflash_cfg *cfg)
784 {
785 	int i;
786 	char *buf = NULL;
787 	struct afu *afu = cfg->afu;
788 
789 	if (cfg->afu) {
790 		for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
791 			buf = afu->cmd[i].buf;
792 			if (!((u64)buf & (PAGE_SIZE - 1)))
793 				free_page((ulong)buf);
794 		}
795 
796 		free_pages((ulong)afu, get_order(sizeof(struct afu)));
797 		cfg->afu = NULL;
798 	}
799 }
800 
801 /**
802  * stop_afu() - stops the AFU command timers and unmaps the MMIO space
803  * @cxlflash:	Internal structure associated with the host.
804  *
805  * Safe to call with AFU in a partially allocated/initialized state.
806  */
807 static void stop_afu(struct cxlflash_cfg *cfg)
808 {
809 	int i;
810 	struct afu *afu = cfg->afu;
811 
812 	if (likely(afu)) {
813 		for (i = 0; i < CXLFLASH_NUM_CMDS; i++)
814 			complete(&afu->cmd[i].cevent);
815 
816 		if (likely(afu->afu_map)) {
817 			cxl_psa_unmap((void *)afu->afu_map);
818 			afu->afu_map = NULL;
819 		}
820 	}
821 }
822 
823 /**
824  * term_mc() - terminates the master context
825  * @cxlflash:	Internal structure associated with the host.
826  * @level:	Depth of allocation, where to begin waterfall tear down.
827  *
828  * Safe to call with AFU/MC in partially allocated/initialized state.
829  */
830 static void term_mc(struct cxlflash_cfg *cfg, enum undo_level level)
831 {
832 	int rc = 0;
833 	struct afu *afu = cfg->afu;
834 
835 	if (!afu || !cfg->mcctx) {
836 		pr_err("%s: returning from term_mc with NULL afu or MC\n",
837 		       __func__);
838 		return;
839 	}
840 
841 	switch (level) {
842 	case UNDO_START:
843 		rc = cxl_stop_context(cfg->mcctx);
844 		BUG_ON(rc);
845 	case UNMAP_THREE:
846 		cxl_unmap_afu_irq(cfg->mcctx, 3, afu);
847 	case UNMAP_TWO:
848 		cxl_unmap_afu_irq(cfg->mcctx, 2, afu);
849 	case UNMAP_ONE:
850 		cxl_unmap_afu_irq(cfg->mcctx, 1, afu);
851 	case FREE_IRQ:
852 		cxl_free_afu_irqs(cfg->mcctx);
853 	case RELEASE_CONTEXT:
854 		cfg->mcctx = NULL;
855 	}
856 }
857 
858 /**
859  * term_afu() - terminates the AFU
860  * @cxlflash:	Internal structure associated with the host.
861  *
862  * Safe to call with AFU/MC in partially allocated/initialized state.
863  */
864 static void term_afu(struct cxlflash_cfg *cfg)
865 {
866 	term_mc(cfg, UNDO_START);
867 
868 	if (cfg->afu)
869 		stop_afu(cfg);
870 
871 	pr_debug("%s: returning\n", __func__);
872 }
873 
874 /**
875  * cxlflash_remove() - PCI entry point to tear down host
876  * @pdev:	PCI device associated with the host.
877  *
878  * Safe to use as a cleanup in partially allocated/initialized state.
879  */
880 static void cxlflash_remove(struct pci_dev *pdev)
881 {
882 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
883 	ulong lock_flags;
884 
885 	/* If a Task Management Function is active, wait for it to complete
886 	 * before continuing with remove.
887 	 */
888 	spin_lock_irqsave(&cfg->tmf_waitq.lock, lock_flags);
889 	if (cfg->tmf_active)
890 		wait_event_interruptible_locked_irq(cfg->tmf_waitq,
891 						    !cfg->tmf_active);
892 	spin_unlock_irqrestore(&cfg->tmf_waitq.lock, lock_flags);
893 
894 	cfg->state = STATE_FAILTERM;
895 	cxlflash_stop_term_user_contexts(cfg);
896 
897 	switch (cfg->init_state) {
898 	case INIT_STATE_SCSI:
899 		cxlflash_term_local_luns(cfg);
900 		scsi_remove_host(cfg->host);
901 		scsi_host_put(cfg->host);
902 		/* Fall through */
903 	case INIT_STATE_AFU:
904 		term_afu(cfg);
905 	case INIT_STATE_PCI:
906 		pci_release_regions(cfg->dev);
907 		pci_disable_device(pdev);
908 	case INIT_STATE_NONE:
909 		flush_work(&cfg->work_q);
910 		free_mem(cfg);
911 		break;
912 	}
913 
914 	pr_debug("%s: returning\n", __func__);
915 }
916 
917 /**
918  * alloc_mem() - allocates the AFU and its command pool
919  * @cxlflash:	Internal structure associated with the host.
920  *
921  * A partially allocated state remains on failure.
922  *
923  * Return:
924  *	0 on success
925  *	-ENOMEM on failure to allocate memory
926  */
927 static int alloc_mem(struct cxlflash_cfg *cfg)
928 {
929 	int rc = 0;
930 	int i;
931 	char *buf = NULL;
932 
933 	/* This allocation is about 12K, i.e. only 1 64k page
934 	 * and upto 4 4k pages
935 	 */
936 	cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
937 					    get_order(sizeof(struct afu)));
938 	if (unlikely(!cfg->afu)) {
939 		pr_err("%s: cannot get %d free pages\n",
940 		       __func__, get_order(sizeof(struct afu)));
941 		rc = -ENOMEM;
942 		goto out;
943 	}
944 	cfg->afu->parent = cfg;
945 	cfg->afu->afu_map = NULL;
946 
947 	for (i = 0; i < CXLFLASH_NUM_CMDS; buf += CMD_BUFSIZE, i++) {
948 		if (!((u64)buf & (PAGE_SIZE - 1))) {
949 			buf = (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
950 			if (unlikely(!buf)) {
951 				pr_err("%s: Allocate command buffers fail!\n",
952 				       __func__);
953 				rc = -ENOMEM;
954 				free_mem(cfg);
955 				goto out;
956 			}
957 		}
958 
959 		cfg->afu->cmd[i].buf = buf;
960 		atomic_set(&cfg->afu->cmd[i].free, 1);
961 		cfg->afu->cmd[i].slot = i;
962 	}
963 
964 out:
965 	return rc;
966 }
967 
968 /**
969  * init_pci() - initializes the host as a PCI device
970  * @cxlflash:	Internal structure associated with the host.
971  *
972  * Return:
973  *	0 on success
974  *	-EIO on unable to communicate with device
975  *	A return code from the PCI sub-routines
976  */
977 static int init_pci(struct cxlflash_cfg *cfg)
978 {
979 	struct pci_dev *pdev = cfg->dev;
980 	int rc = 0;
981 
982 	cfg->cxlflash_regs_pci = pci_resource_start(pdev, 0);
983 	rc = pci_request_regions(pdev, CXLFLASH_NAME);
984 	if (rc < 0) {
985 		dev_err(&pdev->dev,
986 			"%s: Couldn't register memory range of registers\n",
987 			__func__);
988 		goto out;
989 	}
990 
991 	rc = pci_enable_device(pdev);
992 	if (rc || pci_channel_offline(pdev)) {
993 		if (pci_channel_offline(pdev)) {
994 			cxlflash_wait_for_pci_err_recovery(cfg);
995 			rc = pci_enable_device(pdev);
996 		}
997 
998 		if (rc) {
999 			dev_err(&pdev->dev, "%s: Cannot enable adapter\n",
1000 				__func__);
1001 			cxlflash_wait_for_pci_err_recovery(cfg);
1002 			goto out_release_regions;
1003 		}
1004 	}
1005 
1006 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1007 	if (rc < 0) {
1008 		dev_dbg(&pdev->dev, "%s: Failed to set 64 bit PCI DMA mask\n",
1009 			__func__);
1010 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1011 	}
1012 
1013 	if (rc < 0) {
1014 		dev_err(&pdev->dev, "%s: Failed to set PCI DMA mask\n",
1015 			__func__);
1016 		goto out_disable;
1017 	}
1018 
1019 	pci_set_master(pdev);
1020 
1021 	if (pci_channel_offline(pdev)) {
1022 		cxlflash_wait_for_pci_err_recovery(cfg);
1023 		if (pci_channel_offline(pdev)) {
1024 			rc = -EIO;
1025 			goto out_msi_disable;
1026 		}
1027 	}
1028 
1029 	rc = pci_save_state(pdev);
1030 
1031 	if (rc != PCIBIOS_SUCCESSFUL) {
1032 		dev_err(&pdev->dev, "%s: Failed to save PCI config space\n",
1033 			__func__);
1034 		rc = -EIO;
1035 		goto cleanup_nolog;
1036 	}
1037 
1038 out:
1039 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1040 	return rc;
1041 
1042 cleanup_nolog:
1043 out_msi_disable:
1044 	cxlflash_wait_for_pci_err_recovery(cfg);
1045 out_disable:
1046 	pci_disable_device(pdev);
1047 out_release_regions:
1048 	pci_release_regions(pdev);
1049 	goto out;
1050 
1051 }
1052 
1053 /**
1054  * init_scsi() - adds the host to the SCSI stack and kicks off host scan
1055  * @cxlflash:	Internal structure associated with the host.
1056  *
1057  * Return:
1058  *	0 on success
1059  *	A return code from adding the host
1060  */
1061 static int init_scsi(struct cxlflash_cfg *cfg)
1062 {
1063 	struct pci_dev *pdev = cfg->dev;
1064 	int rc = 0;
1065 
1066 	rc = scsi_add_host(cfg->host, &pdev->dev);
1067 	if (rc) {
1068 		dev_err(&pdev->dev, "%s: scsi_add_host failed (rc=%d)\n",
1069 			__func__, rc);
1070 		goto out;
1071 	}
1072 
1073 	scsi_scan_host(cfg->host);
1074 
1075 out:
1076 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1077 	return rc;
1078 }
1079 
1080 /**
1081  * set_port_online() - transitions the specified host FC port to online state
1082  * @fc_regs:	Top of MMIO region defined for specified port.
1083  *
1084  * The provided MMIO region must be mapped prior to call. Online state means
1085  * that the FC link layer has synced, completed the handshaking process, and
1086  * is ready for login to start.
1087  */
1088 static void set_port_online(u64 *fc_regs)
1089 {
1090 	u64 cmdcfg;
1091 
1092 	cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1093 	cmdcfg &= (~FC_MTIP_CMDCONFIG_OFFLINE);	/* clear OFF_LINE */
1094 	cmdcfg |= (FC_MTIP_CMDCONFIG_ONLINE);	/* set ON_LINE */
1095 	writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1096 }
1097 
1098 /**
1099  * set_port_offline() - transitions the specified host FC port to offline state
1100  * @fc_regs:	Top of MMIO region defined for specified port.
1101  *
1102  * The provided MMIO region must be mapped prior to call.
1103  */
1104 static void set_port_offline(u64 *fc_regs)
1105 {
1106 	u64 cmdcfg;
1107 
1108 	cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1109 	cmdcfg &= (~FC_MTIP_CMDCONFIG_ONLINE);	/* clear ON_LINE */
1110 	cmdcfg |= (FC_MTIP_CMDCONFIG_OFFLINE);	/* set OFF_LINE */
1111 	writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1112 }
1113 
1114 /**
1115  * wait_port_online() - waits for the specified host FC port come online
1116  * @fc_regs:	Top of MMIO region defined for specified port.
1117  * @delay_us:	Number of microseconds to delay between reading port status.
1118  * @nretry:	Number of cycles to retry reading port status.
1119  *
1120  * The provided MMIO region must be mapped prior to call. This will timeout
1121  * when the cable is not plugged in.
1122  *
1123  * Return:
1124  *	TRUE (1) when the specified port is online
1125  *	FALSE (0) when the specified port fails to come online after timeout
1126  *	-EINVAL when @delay_us is less than 1000
1127  */
1128 static int wait_port_online(u64 *fc_regs, u32 delay_us, u32 nretry)
1129 {
1130 	u64 status;
1131 
1132 	if (delay_us < 1000) {
1133 		pr_err("%s: invalid delay specified %d\n", __func__, delay_us);
1134 		return -EINVAL;
1135 	}
1136 
1137 	do {
1138 		msleep(delay_us / 1000);
1139 		status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1140 	} while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_ONLINE &&
1141 		 nretry--);
1142 
1143 	return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_ONLINE);
1144 }
1145 
1146 /**
1147  * wait_port_offline() - waits for the specified host FC port go offline
1148  * @fc_regs:	Top of MMIO region defined for specified port.
1149  * @delay_us:	Number of microseconds to delay between reading port status.
1150  * @nretry:	Number of cycles to retry reading port status.
1151  *
1152  * The provided MMIO region must be mapped prior to call.
1153  *
1154  * Return:
1155  *	TRUE (1) when the specified port is offline
1156  *	FALSE (0) when the specified port fails to go offline after timeout
1157  *	-EINVAL when @delay_us is less than 1000
1158  */
1159 static int wait_port_offline(u64 *fc_regs, u32 delay_us, u32 nretry)
1160 {
1161 	u64 status;
1162 
1163 	if (delay_us < 1000) {
1164 		pr_err("%s: invalid delay specified %d\n", __func__, delay_us);
1165 		return -EINVAL;
1166 	}
1167 
1168 	do {
1169 		msleep(delay_us / 1000);
1170 		status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1171 	} while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_OFFLINE &&
1172 		 nretry--);
1173 
1174 	return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_OFFLINE);
1175 }
1176 
1177 /**
1178  * afu_set_wwpn() - configures the WWPN for the specified host FC port
1179  * @afu:	AFU associated with the host that owns the specified FC port.
1180  * @port:	Port number being configured.
1181  * @fc_regs:	Top of MMIO region defined for specified port.
1182  * @wwpn:	The world-wide-port-number previously discovered for port.
1183  *
1184  * The provided MMIO region must be mapped prior to call. As part of the
1185  * sequence to configure the WWPN, the port is toggled offline and then back
1186  * online. This toggling action can cause this routine to delay up to a few
1187  * seconds. When configured to use the internal LUN feature of the AFU, a
1188  * failure to come online is overridden.
1189  *
1190  * Return:
1191  *	0 when the WWPN is successfully written and the port comes back online
1192  *	-1 when the port fails to go offline or come back up online
1193  */
1194 static int afu_set_wwpn(struct afu *afu, int port, u64 *fc_regs, u64 wwpn)
1195 {
1196 	int ret = 0;
1197 
1198 	set_port_offline(fc_regs);
1199 
1200 	if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1201 			       FC_PORT_STATUS_RETRY_CNT)) {
1202 		pr_debug("%s: wait on port %d to go offline timed out\n",
1203 			 __func__, port);
1204 		ret = -1; /* but continue on to leave the port back online */
1205 	}
1206 
1207 	if (ret == 0)
1208 		writeq_be(wwpn, &fc_regs[FC_PNAME / 8]);
1209 
1210 	set_port_online(fc_regs);
1211 
1212 	if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1213 			      FC_PORT_STATUS_RETRY_CNT)) {
1214 		pr_debug("%s: wait on port %d to go online timed out\n",
1215 			 __func__, port);
1216 		ret = -1;
1217 
1218 		/*
1219 		 * Override for internal lun!!!
1220 		 */
1221 		if (afu->internal_lun) {
1222 			pr_debug("%s: Overriding port %d online timeout!!!\n",
1223 				 __func__, port);
1224 			ret = 0;
1225 		}
1226 	}
1227 
1228 	pr_debug("%s: returning rc=%d\n", __func__, ret);
1229 
1230 	return ret;
1231 }
1232 
1233 /**
1234  * afu_link_reset() - resets the specified host FC port
1235  * @afu:	AFU associated with the host that owns the specified FC port.
1236  * @port:	Port number being configured.
1237  * @fc_regs:	Top of MMIO region defined for specified port.
1238  *
1239  * The provided MMIO region must be mapped prior to call. The sequence to
1240  * reset the port involves toggling it offline and then back online. This
1241  * action can cause this routine to delay up to a few seconds. An effort
1242  * is made to maintain link with the device by switching to host to use
1243  * the alternate port exclusively while the reset takes place.
1244  * failure to come online is overridden.
1245  */
1246 static void afu_link_reset(struct afu *afu, int port, u64 *fc_regs)
1247 {
1248 	u64 port_sel;
1249 
1250 	/* first switch the AFU to the other links, if any */
1251 	port_sel = readq_be(&afu->afu_map->global.regs.afu_port_sel);
1252 	port_sel &= ~(1ULL << port);
1253 	writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1254 	cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1255 
1256 	set_port_offline(fc_regs);
1257 	if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1258 			       FC_PORT_STATUS_RETRY_CNT))
1259 		pr_err("%s: wait on port %d to go offline timed out\n",
1260 		       __func__, port);
1261 
1262 	set_port_online(fc_regs);
1263 	if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1264 			      FC_PORT_STATUS_RETRY_CNT))
1265 		pr_err("%s: wait on port %d to go online timed out\n",
1266 		       __func__, port);
1267 
1268 	/* switch back to include this port */
1269 	port_sel |= (1ULL << port);
1270 	writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1271 	cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1272 
1273 	pr_debug("%s: returning port_sel=%lld\n", __func__, port_sel);
1274 }
1275 
1276 /*
1277  * Asynchronous interrupt information table
1278  */
1279 static const struct asyc_intr_info ainfo[] = {
1280 	{SISL_ASTATUS_FC0_OTHER, "other error", 0, CLR_FC_ERROR | LINK_RESET},
1281 	{SISL_ASTATUS_FC0_LOGO, "target initiated LOGO", 0, 0},
1282 	{SISL_ASTATUS_FC0_CRC_T, "CRC threshold exceeded", 0, LINK_RESET},
1283 	{SISL_ASTATUS_FC0_LOGI_R, "login timed out, retrying", 0, 0},
1284 	{SISL_ASTATUS_FC0_LOGI_F, "login failed", 0, CLR_FC_ERROR},
1285 	{SISL_ASTATUS_FC0_LOGI_S, "login succeeded", 0, 0},
1286 	{SISL_ASTATUS_FC0_LINK_DN, "link down", 0, 0},
1287 	{SISL_ASTATUS_FC0_LINK_UP, "link up", 0, 0},
1288 	{SISL_ASTATUS_FC1_OTHER, "other error", 1, CLR_FC_ERROR | LINK_RESET},
1289 	{SISL_ASTATUS_FC1_LOGO, "target initiated LOGO", 1, 0},
1290 	{SISL_ASTATUS_FC1_CRC_T, "CRC threshold exceeded", 1, LINK_RESET},
1291 	{SISL_ASTATUS_FC1_LOGI_R, "login timed out, retrying", 1, 0},
1292 	{SISL_ASTATUS_FC1_LOGI_F, "login failed", 1, CLR_FC_ERROR},
1293 	{SISL_ASTATUS_FC1_LOGI_S, "login succeeded", 1, 0},
1294 	{SISL_ASTATUS_FC1_LINK_DN, "link down", 1, 0},
1295 	{SISL_ASTATUS_FC1_LINK_UP, "link up", 1, 0},
1296 	{0x0, "", 0, 0}		/* terminator */
1297 };
1298 
1299 /**
1300  * find_ainfo() - locates and returns asynchronous interrupt information
1301  * @status:	Status code set by AFU on error.
1302  *
1303  * Return: The located information or NULL when the status code is invalid.
1304  */
1305 static const struct asyc_intr_info *find_ainfo(u64 status)
1306 {
1307 	const struct asyc_intr_info *info;
1308 
1309 	for (info = &ainfo[0]; info->status; info++)
1310 		if (info->status == status)
1311 			return info;
1312 
1313 	return NULL;
1314 }
1315 
1316 /**
1317  * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1318  * @afu:	AFU associated with the host.
1319  */
1320 static void afu_err_intr_init(struct afu *afu)
1321 {
1322 	int i;
1323 	u64 reg;
1324 
1325 	/* global async interrupts: AFU clears afu_ctrl on context exit
1326 	 * if async interrupts were sent to that context. This prevents
1327 	 * the AFU form sending further async interrupts when
1328 	 * there is
1329 	 * nobody to receive them.
1330 	 */
1331 
1332 	/* mask all */
1333 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_mask);
1334 	/* set LISN# to send and point to master context */
1335 	reg = ((u64) (((afu->ctx_hndl << 8) | SISL_MSI_ASYNC_ERROR)) << 40);
1336 
1337 	if (afu->internal_lun)
1338 		reg |= 1;	/* Bit 63 indicates local lun */
1339 	writeq_be(reg, &afu->afu_map->global.regs.afu_ctrl);
1340 	/* clear all */
1341 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1342 	/* unmask bits that are of interest */
1343 	/* note: afu can send an interrupt after this step */
1344 	writeq_be(SISL_ASTATUS_MASK, &afu->afu_map->global.regs.aintr_mask);
1345 	/* clear again in case a bit came on after previous clear but before */
1346 	/* unmask */
1347 	writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1348 
1349 	/* Clear/Set internal lun bits */
1350 	reg = readq_be(&afu->afu_map->global.fc_regs[0][FC_CONFIG2 / 8]);
1351 	reg &= SISL_FC_INTERNAL_MASK;
1352 	if (afu->internal_lun)
1353 		reg |= ((u64)(afu->internal_lun - 1) << SISL_FC_INTERNAL_SHIFT);
1354 	writeq_be(reg, &afu->afu_map->global.fc_regs[0][FC_CONFIG2 / 8]);
1355 
1356 	/* now clear FC errors */
1357 	for (i = 0; i < NUM_FC_PORTS; i++) {
1358 		writeq_be(0xFFFFFFFFU,
1359 			  &afu->afu_map->global.fc_regs[i][FC_ERROR / 8]);
1360 		writeq_be(0, &afu->afu_map->global.fc_regs[i][FC_ERRCAP / 8]);
1361 	}
1362 
1363 	/* sync interrupts for master's IOARRIN write */
1364 	/* note that unlike asyncs, there can be no pending sync interrupts */
1365 	/* at this time (this is a fresh context and master has not written */
1366 	/* IOARRIN yet), so there is nothing to clear. */
1367 
1368 	/* set LISN#, it is always sent to the context that wrote IOARRIN */
1369 	writeq_be(SISL_MSI_SYNC_ERROR, &afu->host_map->ctx_ctrl);
1370 	writeq_be(SISL_ISTATUS_MASK, &afu->host_map->intr_mask);
1371 }
1372 
1373 /**
1374  * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1375  * @irq:	Interrupt number.
1376  * @data:	Private data provided at interrupt registration, the AFU.
1377  *
1378  * Return: Always return IRQ_HANDLED.
1379  */
1380 static irqreturn_t cxlflash_sync_err_irq(int irq, void *data)
1381 {
1382 	struct afu *afu = (struct afu *)data;
1383 	u64 reg;
1384 	u64 reg_unmasked;
1385 
1386 	reg = readq_be(&afu->host_map->intr_status);
1387 	reg_unmasked = (reg & SISL_ISTATUS_UNMASK);
1388 
1389 	if (reg_unmasked == 0UL) {
1390 		pr_err("%s: %llX: spurious interrupt, intr_status %016llX\n",
1391 		       __func__, (u64)afu, reg);
1392 		goto cxlflash_sync_err_irq_exit;
1393 	}
1394 
1395 	pr_err("%s: %llX: unexpected interrupt, intr_status %016llX\n",
1396 	       __func__, (u64)afu, reg);
1397 
1398 	writeq_be(reg_unmasked, &afu->host_map->intr_clear);
1399 
1400 cxlflash_sync_err_irq_exit:
1401 	pr_debug("%s: returning rc=%d\n", __func__, IRQ_HANDLED);
1402 	return IRQ_HANDLED;
1403 }
1404 
1405 /**
1406  * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1407  * @irq:	Interrupt number.
1408  * @data:	Private data provided at interrupt registration, the AFU.
1409  *
1410  * Return: Always return IRQ_HANDLED.
1411  */
1412 static irqreturn_t cxlflash_rrq_irq(int irq, void *data)
1413 {
1414 	struct afu *afu = (struct afu *)data;
1415 	struct afu_cmd *cmd;
1416 	bool toggle = afu->toggle;
1417 	u64 entry,
1418 	    *hrrq_start = afu->hrrq_start,
1419 	    *hrrq_end = afu->hrrq_end,
1420 	    *hrrq_curr = afu->hrrq_curr;
1421 
1422 	/* Process however many RRQ entries that are ready */
1423 	while (true) {
1424 		entry = *hrrq_curr;
1425 
1426 		if ((entry & SISL_RESP_HANDLE_T_BIT) != toggle)
1427 			break;
1428 
1429 		cmd = (struct afu_cmd *)(entry & ~SISL_RESP_HANDLE_T_BIT);
1430 		cmd_complete(cmd);
1431 
1432 		/* Advance to next entry or wrap and flip the toggle bit */
1433 		if (hrrq_curr < hrrq_end)
1434 			hrrq_curr++;
1435 		else {
1436 			hrrq_curr = hrrq_start;
1437 			toggle ^= SISL_RESP_HANDLE_T_BIT;
1438 		}
1439 	}
1440 
1441 	afu->hrrq_curr = hrrq_curr;
1442 	afu->toggle = toggle;
1443 
1444 	return IRQ_HANDLED;
1445 }
1446 
1447 /**
1448  * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1449  * @irq:	Interrupt number.
1450  * @data:	Private data provided at interrupt registration, the AFU.
1451  *
1452  * Return: Always return IRQ_HANDLED.
1453  */
1454 static irqreturn_t cxlflash_async_err_irq(int irq, void *data)
1455 {
1456 	struct afu *afu = (struct afu *)data;
1457 	struct cxlflash_cfg *cfg;
1458 	u64 reg_unmasked;
1459 	const struct asyc_intr_info *info;
1460 	struct sisl_global_map *global = &afu->afu_map->global;
1461 	u64 reg;
1462 	u8 port;
1463 	int i;
1464 
1465 	cfg = afu->parent;
1466 
1467 	reg = readq_be(&global->regs.aintr_status);
1468 	reg_unmasked = (reg & SISL_ASTATUS_UNMASK);
1469 
1470 	if (reg_unmasked == 0) {
1471 		pr_err("%s: spurious interrupt, aintr_status 0x%016llX\n",
1472 		       __func__, reg);
1473 		goto out;
1474 	}
1475 
1476 	/* it is OK to clear AFU status before FC_ERROR */
1477 	writeq_be(reg_unmasked, &global->regs.aintr_clear);
1478 
1479 	/* check each bit that is on */
1480 	for (i = 0; reg_unmasked; i++, reg_unmasked = (reg_unmasked >> 1)) {
1481 		info = find_ainfo(1ULL << i);
1482 		if ((reg_unmasked & 0x1) || !info)
1483 			continue;
1484 
1485 		port = info->port;
1486 
1487 		pr_err("%s: FC Port %d -> %s, fc_status 0x%08llX\n",
1488 		       __func__, port, info->desc,
1489 		       readq_be(&global->fc_regs[port][FC_STATUS / 8]));
1490 
1491 		/*
1492 		 * do link reset first, some OTHER errors will set FC_ERROR
1493 		 * again if cleared before or w/o a reset
1494 		 */
1495 		if (info->action & LINK_RESET) {
1496 			pr_err("%s: FC Port %d: resetting link\n",
1497 			       __func__, port);
1498 			cfg->lr_state = LINK_RESET_REQUIRED;
1499 			cfg->lr_port = port;
1500 			schedule_work(&cfg->work_q);
1501 		}
1502 
1503 		if (info->action & CLR_FC_ERROR) {
1504 			reg = readq_be(&global->fc_regs[port][FC_ERROR / 8]);
1505 
1506 			/*
1507 			 * since all errors are unmasked, FC_ERROR and FC_ERRCAP
1508 			 * should be the same and tracing one is sufficient.
1509 			 */
1510 
1511 			pr_err("%s: fc %d: clearing fc_error 0x%08llX\n",
1512 			       __func__, port, reg);
1513 
1514 			writeq_be(reg, &global->fc_regs[port][FC_ERROR / 8]);
1515 			writeq_be(0, &global->fc_regs[port][FC_ERRCAP / 8]);
1516 		}
1517 	}
1518 
1519 out:
1520 	pr_debug("%s: returning rc=%d, afu=%p\n", __func__, IRQ_HANDLED, afu);
1521 	return IRQ_HANDLED;
1522 }
1523 
1524 /**
1525  * start_context() - starts the master context
1526  * @cxlflash:	Internal structure associated with the host.
1527  *
1528  * Return: A success or failure value from CXL services.
1529  */
1530 static int start_context(struct cxlflash_cfg *cfg)
1531 {
1532 	int rc = 0;
1533 
1534 	rc = cxl_start_context(cfg->mcctx,
1535 			       cfg->afu->work.work_element_descriptor,
1536 			       NULL);
1537 
1538 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1539 	return rc;
1540 }
1541 
1542 /**
1543  * read_vpd() - obtains the WWPNs from VPD
1544  * @cxlflash:	Internal structure associated with the host.
1545  * @wwpn:	Array of size NUM_FC_PORTS to pass back WWPNs
1546  *
1547  * Return:
1548  *	0 on success
1549  *	-ENODEV when VPD or WWPN keywords not found
1550  */
1551 static int read_vpd(struct cxlflash_cfg *cfg, u64 wwpn[])
1552 {
1553 	struct pci_dev *dev = cfg->parent_dev;
1554 	int rc = 0;
1555 	int ro_start, ro_size, i, j, k;
1556 	ssize_t vpd_size;
1557 	char vpd_data[CXLFLASH_VPD_LEN];
1558 	char tmp_buf[WWPN_BUF_LEN] = { 0 };
1559 	char *wwpn_vpd_tags[NUM_FC_PORTS] = { "V5", "V6" };
1560 
1561 	/* Get the VPD data from the device */
1562 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
1563 	if (unlikely(vpd_size <= 0)) {
1564 		pr_err("%s: Unable to read VPD (size = %ld)\n",
1565 		       __func__, vpd_size);
1566 		rc = -ENODEV;
1567 		goto out;
1568 	}
1569 
1570 	/* Get the read only section offset */
1571 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size,
1572 				    PCI_VPD_LRDT_RO_DATA);
1573 	if (unlikely(ro_start < 0)) {
1574 		pr_err("%s: VPD Read-only data not found\n", __func__);
1575 		rc = -ENODEV;
1576 		goto out;
1577 	}
1578 
1579 	/* Get the read only section size, cap when extends beyond read VPD */
1580 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
1581 	j = ro_size;
1582 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1583 	if (unlikely((i + j) > vpd_size)) {
1584 		pr_debug("%s: Might need to read more VPD (%d > %ld)\n",
1585 			 __func__, (i + j), vpd_size);
1586 		ro_size = vpd_size - i;
1587 	}
1588 
1589 	/*
1590 	 * Find the offset of the WWPN tag within the read only
1591 	 * VPD data and validate the found field (partials are
1592 	 * no good to us). Convert the ASCII data to an integer
1593 	 * value. Note that we must copy to a temporary buffer
1594 	 * because the conversion service requires that the ASCII
1595 	 * string be terminated.
1596 	 */
1597 	for (k = 0; k < NUM_FC_PORTS; k++) {
1598 		j = ro_size;
1599 		i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1600 
1601 		i = pci_vpd_find_info_keyword(vpd_data, i, j, wwpn_vpd_tags[k]);
1602 		if (unlikely(i < 0)) {
1603 			pr_err("%s: Port %d WWPN not found in VPD\n",
1604 			       __func__, k);
1605 			rc = -ENODEV;
1606 			goto out;
1607 		}
1608 
1609 		j = pci_vpd_info_field_size(&vpd_data[i]);
1610 		i += PCI_VPD_INFO_FLD_HDR_SIZE;
1611 		if (unlikely((i + j > vpd_size) || (j != WWPN_LEN))) {
1612 			pr_err("%s: Port %d WWPN incomplete or VPD corrupt\n",
1613 			       __func__, k);
1614 			rc = -ENODEV;
1615 			goto out;
1616 		}
1617 
1618 		memcpy(tmp_buf, &vpd_data[i], WWPN_LEN);
1619 		rc = kstrtoul(tmp_buf, WWPN_LEN, (ulong *)&wwpn[k]);
1620 		if (unlikely(rc)) {
1621 			pr_err("%s: Fail to convert port %d WWPN to integer\n",
1622 			       __func__, k);
1623 			rc = -ENODEV;
1624 			goto out;
1625 		}
1626 	}
1627 
1628 out:
1629 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1630 	return rc;
1631 }
1632 
1633 /**
1634  * cxlflash_context_reset() - timeout handler for AFU commands
1635  * @cmd:	AFU command that timed out.
1636  *
1637  * Sends a reset to the AFU.
1638  */
1639 void cxlflash_context_reset(struct afu_cmd *cmd)
1640 {
1641 	int nretry = 0;
1642 	u64 rrin = 0x1;
1643 	u64 room = 0;
1644 	struct afu *afu = cmd->parent;
1645 	ulong lock_flags;
1646 
1647 	pr_debug("%s: cmd=%p\n", __func__, cmd);
1648 
1649 	spin_lock_irqsave(&cmd->slock, lock_flags);
1650 
1651 	/* Already completed? */
1652 	if (cmd->sa.host_use_b[0] & B_DONE) {
1653 		spin_unlock_irqrestore(&cmd->slock, lock_flags);
1654 		return;
1655 	}
1656 
1657 	cmd->sa.host_use_b[0] |= (B_DONE | B_ERROR | B_TIMEOUT);
1658 	spin_unlock_irqrestore(&cmd->slock, lock_flags);
1659 
1660 	/*
1661 	 * We really want to send this reset at all costs, so spread
1662 	 * out wait time on successive retries for available room.
1663 	 */
1664 	do {
1665 		room = readq_be(&afu->host_map->cmd_room);
1666 		atomic64_set(&afu->room, room);
1667 		if (room)
1668 			goto write_rrin;
1669 		udelay(nretry);
1670 	} while (nretry++ < MC_ROOM_RETRY_CNT);
1671 
1672 	pr_err("%s: no cmd_room to send reset\n", __func__);
1673 	return;
1674 
1675 write_rrin:
1676 	nretry = 0;
1677 	writeq_be(rrin, &afu->host_map->ioarrin);
1678 	do {
1679 		rrin = readq_be(&afu->host_map->ioarrin);
1680 		if (rrin != 0x1)
1681 			break;
1682 		/* Double delay each time */
1683 		udelay(2 ^ nretry);
1684 	} while (nretry++ < MC_ROOM_RETRY_CNT);
1685 }
1686 
1687 /**
1688  * init_pcr() - initialize the provisioning and control registers
1689  * @cxlflash:	Internal structure associated with the host.
1690  *
1691  * Also sets up fast access to the mapped registers and initializes AFU
1692  * command fields that never change.
1693  */
1694 void init_pcr(struct cxlflash_cfg *cfg)
1695 {
1696 	struct afu *afu = cfg->afu;
1697 	struct sisl_ctrl_map *ctrl_map;
1698 	int i;
1699 
1700 	for (i = 0; i < MAX_CONTEXT; i++) {
1701 		ctrl_map = &afu->afu_map->ctrls[i].ctrl;
1702 		/* disrupt any clients that could be running */
1703 		/* e. g. clients that survived a master restart */
1704 		writeq_be(0, &ctrl_map->rht_start);
1705 		writeq_be(0, &ctrl_map->rht_cnt_id);
1706 		writeq_be(0, &ctrl_map->ctx_cap);
1707 	}
1708 
1709 	/* copy frequently used fields into afu */
1710 	afu->ctx_hndl = (u16) cxl_process_element(cfg->mcctx);
1711 	/* ctx_hndl is 16 bits in CAIA */
1712 	afu->host_map = &afu->afu_map->hosts[afu->ctx_hndl].host;
1713 	afu->ctrl_map = &afu->afu_map->ctrls[afu->ctx_hndl].ctrl;
1714 
1715 	/* Program the Endian Control for the master context */
1716 	writeq_be(SISL_ENDIAN_CTRL, &afu->host_map->endian_ctrl);
1717 
1718 	/* initialize cmd fields that never change */
1719 	for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
1720 		afu->cmd[i].rcb.ctx_id = afu->ctx_hndl;
1721 		afu->cmd[i].rcb.msi = SISL_MSI_RRQ_UPDATED;
1722 		afu->cmd[i].rcb.rrq = 0x0;
1723 	}
1724 }
1725 
1726 /**
1727  * init_global() - initialize AFU global registers
1728  * @cxlflash:	Internal structure associated with the host.
1729  */
1730 int init_global(struct cxlflash_cfg *cfg)
1731 {
1732 	struct afu *afu = cfg->afu;
1733 	u64 wwpn[NUM_FC_PORTS];	/* wwpn of AFU ports */
1734 	int i = 0, num_ports = 0;
1735 	int rc = 0;
1736 	u64 reg;
1737 
1738 	rc = read_vpd(cfg, &wwpn[0]);
1739 	if (rc) {
1740 		pr_err("%s: could not read vpd rc=%d\n", __func__, rc);
1741 		goto out;
1742 	}
1743 
1744 	pr_debug("%s: wwpn0=0x%llX wwpn1=0x%llX\n", __func__, wwpn[0], wwpn[1]);
1745 
1746 	/* set up RRQ in AFU for master issued cmds */
1747 	writeq_be((u64) afu->hrrq_start, &afu->host_map->rrq_start);
1748 	writeq_be((u64) afu->hrrq_end, &afu->host_map->rrq_end);
1749 
1750 	/* AFU configuration */
1751 	reg = readq_be(&afu->afu_map->global.regs.afu_config);
1752 	reg |= SISL_AFUCONF_AR_ALL|SISL_AFUCONF_ENDIAN;
1753 	/* enable all auto retry options and control endianness */
1754 	/* leave others at default: */
1755 	/* CTX_CAP write protected, mbox_r does not clear on read and */
1756 	/* checker on if dual afu */
1757 	writeq_be(reg, &afu->afu_map->global.regs.afu_config);
1758 
1759 	/* global port select: select either port */
1760 	if (afu->internal_lun) {
1761 		/* only use port 0 */
1762 		writeq_be(PORT0, &afu->afu_map->global.regs.afu_port_sel);
1763 		num_ports = NUM_FC_PORTS - 1;
1764 	} else {
1765 		writeq_be(BOTH_PORTS, &afu->afu_map->global.regs.afu_port_sel);
1766 		num_ports = NUM_FC_PORTS;
1767 	}
1768 
1769 	for (i = 0; i < num_ports; i++) {
1770 		/* unmask all errors (but they are still masked at AFU) */
1771 		writeq_be(0, &afu->afu_map->global.fc_regs[i][FC_ERRMSK / 8]);
1772 		/* clear CRC error cnt & set a threshold */
1773 		(void)readq_be(&afu->afu_map->global.
1774 			       fc_regs[i][FC_CNT_CRCERR / 8]);
1775 		writeq_be(MC_CRC_THRESH, &afu->afu_map->global.fc_regs[i]
1776 			  [FC_CRC_THRESH / 8]);
1777 
1778 		/* set WWPNs. If already programmed, wwpn[i] is 0 */
1779 		if (wwpn[i] != 0 &&
1780 		    afu_set_wwpn(afu, i,
1781 				 &afu->afu_map->global.fc_regs[i][0],
1782 				 wwpn[i])) {
1783 			pr_err("%s: failed to set WWPN on port %d\n",
1784 			       __func__, i);
1785 			rc = -EIO;
1786 			goto out;
1787 		}
1788 		/* Programming WWPN back to back causes additional
1789 		 * offline/online transitions and a PLOGI
1790 		 */
1791 		msleep(100);
1792 
1793 	}
1794 
1795 	/* set up master's own CTX_CAP to allow real mode, host translation */
1796 	/* tbls, afu cmds and read/write GSCSI cmds. */
1797 	/* First, unlock ctx_cap write by reading mbox */
1798 	(void)readq_be(&afu->ctrl_map->mbox_r);	/* unlock ctx_cap */
1799 	writeq_be((SISL_CTX_CAP_REAL_MODE | SISL_CTX_CAP_HOST_XLATE |
1800 		   SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD |
1801 		   SISL_CTX_CAP_AFU_CMD | SISL_CTX_CAP_GSCSI_CMD),
1802 		  &afu->ctrl_map->ctx_cap);
1803 	/* init heartbeat */
1804 	afu->hb = readq_be(&afu->afu_map->global.regs.afu_hb);
1805 
1806 out:
1807 	return rc;
1808 }
1809 
1810 /**
1811  * start_afu() - initializes and starts the AFU
1812  * @cxlflash:	Internal structure associated with the host.
1813  */
1814 static int start_afu(struct cxlflash_cfg *cfg)
1815 {
1816 	struct afu *afu = cfg->afu;
1817 	struct afu_cmd *cmd;
1818 
1819 	int i = 0;
1820 	int rc = 0;
1821 
1822 	for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
1823 		cmd = &afu->cmd[i];
1824 
1825 		init_completion(&cmd->cevent);
1826 		spin_lock_init(&cmd->slock);
1827 		cmd->parent = afu;
1828 	}
1829 
1830 	init_pcr(cfg);
1831 
1832 	/* initialize RRQ pointers */
1833 	afu->hrrq_start = &afu->rrq_entry[0];
1834 	afu->hrrq_end = &afu->rrq_entry[NUM_RRQ_ENTRY - 1];
1835 	afu->hrrq_curr = afu->hrrq_start;
1836 	afu->toggle = 1;
1837 
1838 	rc = init_global(cfg);
1839 
1840 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1841 	return rc;
1842 }
1843 
1844 /**
1845  * init_mc() - create and register as the master context
1846  * @cxlflash:	Internal structure associated with the host.
1847  *
1848  * Return:
1849  *	0 on success
1850  *	-ENOMEM when unable to obtain a context from CXL services
1851  *	A failure value from CXL services.
1852  */
1853 static int init_mc(struct cxlflash_cfg *cfg)
1854 {
1855 	struct cxl_context *ctx;
1856 	struct device *dev = &cfg->dev->dev;
1857 	struct afu *afu = cfg->afu;
1858 	int rc = 0;
1859 	enum undo_level level;
1860 
1861 	ctx = cxl_get_context(cfg->dev);
1862 	if (unlikely(!ctx))
1863 		return -ENOMEM;
1864 	cfg->mcctx = ctx;
1865 
1866 	/* Set it up as a master with the CXL */
1867 	cxl_set_master(ctx);
1868 
1869 	/* During initialization reset the AFU to start from a clean slate */
1870 	rc = cxl_afu_reset(cfg->mcctx);
1871 	if (unlikely(rc)) {
1872 		dev_err(dev, "%s: initial AFU reset failed rc=%d\n",
1873 			__func__, rc);
1874 		level = RELEASE_CONTEXT;
1875 		goto out;
1876 	}
1877 
1878 	rc = cxl_allocate_afu_irqs(ctx, 3);
1879 	if (unlikely(rc)) {
1880 		dev_err(dev, "%s: call to allocate_afu_irqs failed rc=%d!\n",
1881 			__func__, rc);
1882 		level = RELEASE_CONTEXT;
1883 		goto out;
1884 	}
1885 
1886 	rc = cxl_map_afu_irq(ctx, 1, cxlflash_sync_err_irq, afu,
1887 			     "SISL_MSI_SYNC_ERROR");
1888 	if (unlikely(rc <= 0)) {
1889 		dev_err(dev, "%s: IRQ 1 (SISL_MSI_SYNC_ERROR) map failed!\n",
1890 			__func__);
1891 		level = FREE_IRQ;
1892 		goto out;
1893 	}
1894 
1895 	rc = cxl_map_afu_irq(ctx, 2, cxlflash_rrq_irq, afu,
1896 			     "SISL_MSI_RRQ_UPDATED");
1897 	if (unlikely(rc <= 0)) {
1898 		dev_err(dev, "%s: IRQ 2 (SISL_MSI_RRQ_UPDATED) map failed!\n",
1899 			__func__);
1900 		level = UNMAP_ONE;
1901 		goto out;
1902 	}
1903 
1904 	rc = cxl_map_afu_irq(ctx, 3, cxlflash_async_err_irq, afu,
1905 			     "SISL_MSI_ASYNC_ERROR");
1906 	if (unlikely(rc <= 0)) {
1907 		dev_err(dev, "%s: IRQ 3 (SISL_MSI_ASYNC_ERROR) map failed!\n",
1908 			__func__);
1909 		level = UNMAP_TWO;
1910 		goto out;
1911 	}
1912 
1913 	rc = 0;
1914 
1915 	/* This performs the equivalent of the CXL_IOCTL_START_WORK.
1916 	 * The CXL_IOCTL_GET_PROCESS_ELEMENT is implicit in the process
1917 	 * element (pe) that is embedded in the context (ctx)
1918 	 */
1919 	rc = start_context(cfg);
1920 	if (unlikely(rc)) {
1921 		dev_err(dev, "%s: start context failed rc=%d\n", __func__, rc);
1922 		level = UNMAP_THREE;
1923 		goto out;
1924 	}
1925 ret:
1926 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1927 	return rc;
1928 out:
1929 	term_mc(cfg, level);
1930 	goto ret;
1931 }
1932 
1933 /**
1934  * init_afu() - setup as master context and start AFU
1935  * @cxlflash:	Internal structure associated with the host.
1936  *
1937  * This routine is a higher level of control for configuring the
1938  * AFU on probe and reset paths.
1939  *
1940  * Return:
1941  *	0 on success
1942  *	-ENOMEM when unable to map the AFU MMIO space
1943  *	A failure value from internal services.
1944  */
1945 static int init_afu(struct cxlflash_cfg *cfg)
1946 {
1947 	u64 reg;
1948 	int rc = 0;
1949 	struct afu *afu = cfg->afu;
1950 	struct device *dev = &cfg->dev->dev;
1951 
1952 	cxl_perst_reloads_same_image(cfg->cxl_afu, true);
1953 
1954 	rc = init_mc(cfg);
1955 	if (rc) {
1956 		dev_err(dev, "%s: call to init_mc failed, rc=%d!\n",
1957 			__func__, rc);
1958 		goto err1;
1959 	}
1960 
1961 	/* Map the entire MMIO space of the AFU.
1962 	 */
1963 	afu->afu_map = cxl_psa_map(cfg->mcctx);
1964 	if (!afu->afu_map) {
1965 		rc = -ENOMEM;
1966 		term_mc(cfg, UNDO_START);
1967 		dev_err(dev, "%s: call to cxl_psa_map failed!\n", __func__);
1968 		goto err1;
1969 	}
1970 
1971 	/* don't byte reverse on reading afu_version, else the string form */
1972 	/*     will be backwards */
1973 	reg = afu->afu_map->global.regs.afu_version;
1974 	memcpy(afu->version, &reg, 8);
1975 	afu->interface_version =
1976 	    readq_be(&afu->afu_map->global.regs.interface_version);
1977 	pr_debug("%s: afu version %s, interface version 0x%llX\n",
1978 		 __func__, afu->version, afu->interface_version);
1979 
1980 	rc = start_afu(cfg);
1981 	if (rc) {
1982 		dev_err(dev, "%s: call to start_afu failed, rc=%d!\n",
1983 			__func__, rc);
1984 		term_mc(cfg, UNDO_START);
1985 		cxl_psa_unmap((void *)afu->afu_map);
1986 		afu->afu_map = NULL;
1987 		goto err1;
1988 	}
1989 
1990 	afu_err_intr_init(cfg->afu);
1991 	atomic64_set(&afu->room, readq_be(&afu->host_map->cmd_room));
1992 
1993 	/* Restore the LUN mappings */
1994 	cxlflash_restore_luntable(cfg);
1995 err1:
1996 	pr_debug("%s: returning rc=%d\n", __func__, rc);
1997 	return rc;
1998 }
1999 
2000 /**
2001  * cxlflash_send_cmd() - sends an AFU command
2002  * @afu:	AFU associated with the host.
2003  * @cmd:	AFU command to send.
2004  *
2005  * Return:
2006  *	0 on success
2007  *	-1 on failure
2008  */
2009 int cxlflash_send_cmd(struct afu *afu, struct afu_cmd *cmd)
2010 {
2011 	struct cxlflash_cfg *cfg = afu->parent;
2012 	int nretry = 0;
2013 	int rc = 0;
2014 	u64 room;
2015 	long newval;
2016 
2017 	/*
2018 	 * This routine is used by critical users such an AFU sync and to
2019 	 * send a task management function (TMF). Thus we want to retry a
2020 	 * bit before returning an error. To avoid the performance penalty
2021 	 * of MMIO, we spread the update of 'room' over multiple commands.
2022 	 */
2023 retry:
2024 	newval = atomic64_dec_if_positive(&afu->room);
2025 	if (!newval) {
2026 		do {
2027 			room = readq_be(&afu->host_map->cmd_room);
2028 			atomic64_set(&afu->room, room);
2029 			if (room)
2030 				goto write_ioarrin;
2031 			udelay(nretry);
2032 		} while (nretry++ < MC_ROOM_RETRY_CNT);
2033 
2034 		pr_err("%s: no cmd_room to send 0x%X\n",
2035 		       __func__, cmd->rcb.cdb[0]);
2036 
2037 		goto no_room;
2038 	} else if (unlikely(newval < 0)) {
2039 		/* This should be rare. i.e. Only if two threads race and
2040 		 * decrement before the MMIO read is done. In this case
2041 		 * just benefit from the other thread having updated
2042 		 * afu->room.
2043 		 */
2044 		if (nretry++ < MC_ROOM_RETRY_CNT) {
2045 			udelay(nretry);
2046 			goto retry;
2047 		}
2048 
2049 		goto no_room;
2050 	}
2051 
2052 write_ioarrin:
2053 	writeq_be((u64)&cmd->rcb, &afu->host_map->ioarrin);
2054 out:
2055 	pr_debug("%s: cmd=%p len=%d ea=%p rc=%d\n", __func__, cmd,
2056 		 cmd->rcb.data_len, (void *)cmd->rcb.data_ea, rc);
2057 	return rc;
2058 
2059 no_room:
2060 	afu->read_room = true;
2061 	schedule_work(&cfg->work_q);
2062 	rc = SCSI_MLQUEUE_HOST_BUSY;
2063 	goto out;
2064 }
2065 
2066 /**
2067  * cxlflash_wait_resp() - polls for a response or timeout to a sent AFU command
2068  * @afu:	AFU associated with the host.
2069  * @cmd:	AFU command that was sent.
2070  */
2071 void cxlflash_wait_resp(struct afu *afu, struct afu_cmd *cmd)
2072 {
2073 	ulong timeout = jiffies + (cmd->rcb.timeout * 2 * HZ);
2074 
2075 	timeout = wait_for_completion_timeout(&cmd->cevent, timeout);
2076 	if (!timeout)
2077 		cxlflash_context_reset(cmd);
2078 
2079 	if (unlikely(cmd->sa.ioasc != 0))
2080 		pr_err("%s: CMD 0x%X failed, IOASC: flags 0x%X, afu_rc 0x%X, "
2081 		       "scsi_rc 0x%X, fc_rc 0x%X\n", __func__, cmd->rcb.cdb[0],
2082 		       cmd->sa.rc.flags, cmd->sa.rc.afu_rc, cmd->sa.rc.scsi_rc,
2083 		       cmd->sa.rc.fc_rc);
2084 }
2085 
2086 /**
2087  * cxlflash_afu_sync() - builds and sends an AFU sync command
2088  * @afu:	AFU associated with the host.
2089  * @ctx_hndl_u:	Identifies context requesting sync.
2090  * @res_hndl_u:	Identifies resource requesting sync.
2091  * @mode:	Type of sync to issue (lightweight, heavyweight, global).
2092  *
2093  * The AFU can only take 1 sync command at a time. This routine enforces this
2094  * limitation by using a mutex to provide exlusive access to the AFU during
2095  * the sync. This design point requires calling threads to not be on interrupt
2096  * context due to the possibility of sleeping during concurrent sync operations.
2097  *
2098  * AFU sync operations are only necessary and allowed when the device is
2099  * operating normally. When not operating normally, sync requests can occur as
2100  * part of cleaning up resources associated with an adapter prior to removal.
2101  * In this scenario, these requests are simply ignored (safe due to the AFU
2102  * going away).
2103  *
2104  * Return:
2105  *	0 on success
2106  *	-1 on failure
2107  */
2108 int cxlflash_afu_sync(struct afu *afu, ctx_hndl_t ctx_hndl_u,
2109 		      res_hndl_t res_hndl_u, u8 mode)
2110 {
2111 	struct cxlflash_cfg *cfg = afu->parent;
2112 	struct afu_cmd *cmd = NULL;
2113 	int rc = 0;
2114 	int retry_cnt = 0;
2115 	static DEFINE_MUTEX(sync_active);
2116 
2117 	if (cfg->state != STATE_NORMAL) {
2118 		pr_debug("%s: Sync not required! (%u)\n", __func__, cfg->state);
2119 		return 0;
2120 	}
2121 
2122 	mutex_lock(&sync_active);
2123 retry:
2124 	cmd = cxlflash_cmd_checkout(afu);
2125 	if (unlikely(!cmd)) {
2126 		retry_cnt++;
2127 		udelay(1000 * retry_cnt);
2128 		if (retry_cnt < MC_RETRY_CNT)
2129 			goto retry;
2130 		pr_err("%s: could not get a free command\n", __func__);
2131 		rc = -1;
2132 		goto out;
2133 	}
2134 
2135 	pr_debug("%s: afu=%p cmd=%p %d\n", __func__, afu, cmd, ctx_hndl_u);
2136 
2137 	memset(cmd->rcb.cdb, 0, sizeof(cmd->rcb.cdb));
2138 
2139 	cmd->rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
2140 	cmd->rcb.port_sel = 0x0;	/* NA */
2141 	cmd->rcb.lun_id = 0x0;	/* NA */
2142 	cmd->rcb.data_len = 0x0;
2143 	cmd->rcb.data_ea = 0x0;
2144 	cmd->rcb.timeout = MC_AFU_SYNC_TIMEOUT;
2145 
2146 	cmd->rcb.cdb[0] = 0xC0;	/* AFU Sync */
2147 	cmd->rcb.cdb[1] = mode;
2148 
2149 	/* The cdb is aligned, no unaligned accessors required */
2150 	*((u16 *)&cmd->rcb.cdb[2]) = swab16(ctx_hndl_u);
2151 	*((u32 *)&cmd->rcb.cdb[4]) = swab32(res_hndl_u);
2152 
2153 	rc = cxlflash_send_cmd(afu, cmd);
2154 	if (unlikely(rc))
2155 		goto out;
2156 
2157 	cxlflash_wait_resp(afu, cmd);
2158 
2159 	/* set on timeout */
2160 	if (unlikely((cmd->sa.ioasc != 0) ||
2161 		     (cmd->sa.host_use_b[0] & B_ERROR)))
2162 		rc = -1;
2163 out:
2164 	mutex_unlock(&sync_active);
2165 	if (cmd)
2166 		cxlflash_cmd_checkin(cmd);
2167 	pr_debug("%s: returning rc=%d\n", __func__, rc);
2168 	return rc;
2169 }
2170 
2171 /**
2172  * cxlflash_afu_reset() - resets the AFU
2173  * @cxlflash:	Internal structure associated with the host.
2174  *
2175  * Return:
2176  *	0 on success
2177  *	A failure value from internal services.
2178  */
2179 int cxlflash_afu_reset(struct cxlflash_cfg *cfg)
2180 {
2181 	int rc = 0;
2182 	/* Stop the context before the reset. Since the context is
2183 	 * no longer available restart it after the reset is complete
2184 	 */
2185 
2186 	term_afu(cfg);
2187 
2188 	rc = init_afu(cfg);
2189 
2190 	pr_debug("%s: returning rc=%d\n", __func__, rc);
2191 	return rc;
2192 }
2193 
2194 /**
2195  * cxlflash_worker_thread() - work thread handler for the AFU
2196  * @work:	Work structure contained within cxlflash associated with host.
2197  *
2198  * Handles the following events:
2199  * - Link reset which cannot be performed on interrupt context due to
2200  * blocking up to a few seconds
2201  * - Read AFU command room
2202  */
2203 static void cxlflash_worker_thread(struct work_struct *work)
2204 {
2205 	struct cxlflash_cfg *cfg = container_of(work, struct cxlflash_cfg,
2206 						work_q);
2207 	struct afu *afu = cfg->afu;
2208 	int port;
2209 	ulong lock_flags;
2210 
2211 	/* Avoid MMIO if the device has failed */
2212 
2213 	if (cfg->state != STATE_NORMAL)
2214 		return;
2215 
2216 	spin_lock_irqsave(cfg->host->host_lock, lock_flags);
2217 
2218 	if (cfg->lr_state == LINK_RESET_REQUIRED) {
2219 		port = cfg->lr_port;
2220 		if (port < 0)
2221 			pr_err("%s: invalid port index %d\n", __func__, port);
2222 		else {
2223 			spin_unlock_irqrestore(cfg->host->host_lock,
2224 					       lock_flags);
2225 
2226 			/* The reset can block... */
2227 			afu_link_reset(afu, port,
2228 				       &afu->afu_map->
2229 				       global.fc_regs[port][0]);
2230 			spin_lock_irqsave(cfg->host->host_lock, lock_flags);
2231 		}
2232 
2233 		cfg->lr_state = LINK_RESET_COMPLETE;
2234 	}
2235 
2236 	if (afu->read_room) {
2237 		atomic64_set(&afu->room, readq_be(&afu->host_map->cmd_room));
2238 		afu->read_room = false;
2239 	}
2240 
2241 	spin_unlock_irqrestore(cfg->host->host_lock, lock_flags);
2242 }
2243 
2244 /**
2245  * cxlflash_probe() - PCI entry point to add host
2246  * @pdev:	PCI device associated with the host.
2247  * @dev_id:	PCI device id associated with device.
2248  *
2249  * Return: 0 on success / non-zero on failure
2250  */
2251 static int cxlflash_probe(struct pci_dev *pdev,
2252 			  const struct pci_device_id *dev_id)
2253 {
2254 	struct Scsi_Host *host;
2255 	struct cxlflash_cfg *cfg = NULL;
2256 	struct device *phys_dev;
2257 	struct dev_dependent_vals *ddv;
2258 	int rc = 0;
2259 
2260 	dev_dbg(&pdev->dev, "%s: Found CXLFLASH with IRQ: %d\n",
2261 		__func__, pdev->irq);
2262 
2263 	ddv = (struct dev_dependent_vals *)dev_id->driver_data;
2264 	driver_template.max_sectors = ddv->max_sectors;
2265 
2266 	host = scsi_host_alloc(&driver_template, sizeof(struct cxlflash_cfg));
2267 	if (!host) {
2268 		dev_err(&pdev->dev, "%s: call to scsi_host_alloc failed!\n",
2269 			__func__);
2270 		rc = -ENOMEM;
2271 		goto out;
2272 	}
2273 
2274 	host->max_id = CXLFLASH_MAX_NUM_TARGETS_PER_BUS;
2275 	host->max_lun = CXLFLASH_MAX_NUM_LUNS_PER_TARGET;
2276 	host->max_channel = NUM_FC_PORTS - 1;
2277 	host->unique_id = host->host_no;
2278 	host->max_cmd_len = CXLFLASH_MAX_CDB_LEN;
2279 
2280 	cfg = (struct cxlflash_cfg *)host->hostdata;
2281 	cfg->host = host;
2282 	rc = alloc_mem(cfg);
2283 	if (rc) {
2284 		dev_err(&pdev->dev, "%s: call to scsi_host_alloc failed!\n",
2285 			__func__);
2286 		rc = -ENOMEM;
2287 		goto out;
2288 	}
2289 
2290 	cfg->init_state = INIT_STATE_NONE;
2291 	cfg->dev = pdev;
2292 
2293 	/*
2294 	 * The promoted LUNs move to the top of the LUN table. The rest stay
2295 	 * on the bottom half. The bottom half grows from the end
2296 	 * (index = 255), whereas the top half grows from the beginning
2297 	 * (index = 0).
2298 	 */
2299 	cfg->promote_lun_index  = 0;
2300 	cfg->last_lun_index[0] = CXLFLASH_NUM_VLUNS/2 - 1;
2301 	cfg->last_lun_index[1] = CXLFLASH_NUM_VLUNS/2 - 1;
2302 
2303 	cfg->dev_id = (struct pci_device_id *)dev_id;
2304 	cfg->mcctx = NULL;
2305 
2306 	init_waitqueue_head(&cfg->tmf_waitq);
2307 	init_waitqueue_head(&cfg->limbo_waitq);
2308 
2309 	INIT_WORK(&cfg->work_q, cxlflash_worker_thread);
2310 	cfg->lr_state = LINK_RESET_INVALID;
2311 	cfg->lr_port = -1;
2312 	mutex_init(&cfg->ctx_tbl_list_mutex);
2313 	mutex_init(&cfg->ctx_recovery_mutex);
2314 	INIT_LIST_HEAD(&cfg->ctx_err_recovery);
2315 	INIT_LIST_HEAD(&cfg->lluns);
2316 
2317 	pci_set_drvdata(pdev, cfg);
2318 
2319 	/* Use the special service provided to look up the physical
2320 	 * PCI device, since we are called on the probe of the virtual
2321 	 * PCI host bus (vphb)
2322 	 */
2323 	phys_dev = cxl_get_phys_dev(pdev);
2324 	if (!dev_is_pci(phys_dev)) {
2325 		pr_err("%s: not a pci dev\n", __func__);
2326 		rc = -ENODEV;
2327 		goto out_remove;
2328 	}
2329 	cfg->parent_dev = to_pci_dev(phys_dev);
2330 
2331 	cfg->cxl_afu = cxl_pci_to_afu(pdev);
2332 
2333 	rc = init_pci(cfg);
2334 	if (rc) {
2335 		dev_err(&pdev->dev, "%s: call to init_pci "
2336 			"failed rc=%d!\n", __func__, rc);
2337 		goto out_remove;
2338 	}
2339 	cfg->init_state = INIT_STATE_PCI;
2340 
2341 	rc = init_afu(cfg);
2342 	if (rc) {
2343 		dev_err(&pdev->dev, "%s: call to init_afu "
2344 			"failed rc=%d!\n", __func__, rc);
2345 		goto out_remove;
2346 	}
2347 	cfg->init_state = INIT_STATE_AFU;
2348 
2349 
2350 	rc = init_scsi(cfg);
2351 	if (rc) {
2352 		dev_err(&pdev->dev, "%s: call to init_scsi "
2353 			"failed rc=%d!\n", __func__, rc);
2354 		goto out_remove;
2355 	}
2356 	cfg->init_state = INIT_STATE_SCSI;
2357 
2358 out:
2359 	pr_debug("%s: returning rc=%d\n", __func__, rc);
2360 	return rc;
2361 
2362 out_remove:
2363 	cxlflash_remove(pdev);
2364 	goto out;
2365 }
2366 
2367 /**
2368  * cxlflash_pci_error_detected() - called when a PCI error is detected
2369  * @pdev:	PCI device struct.
2370  * @state:	PCI channel state.
2371  *
2372  * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
2373  */
2374 static pci_ers_result_t cxlflash_pci_error_detected(struct pci_dev *pdev,
2375 						    pci_channel_state_t state)
2376 {
2377 	int rc = 0;
2378 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
2379 	struct device *dev = &cfg->dev->dev;
2380 
2381 	dev_dbg(dev, "%s: pdev=%p state=%u\n", __func__, pdev, state);
2382 
2383 	switch (state) {
2384 	case pci_channel_io_frozen:
2385 		cfg->state = STATE_LIMBO;
2386 
2387 		/* Turn off legacy I/O */
2388 		scsi_block_requests(cfg->host);
2389 		rc = cxlflash_mark_contexts_error(cfg);
2390 		if (unlikely(rc))
2391 			dev_err(dev, "%s: Failed to mark user contexts!(%d)\n",
2392 				__func__, rc);
2393 		term_mc(cfg, UNDO_START);
2394 		stop_afu(cfg);
2395 
2396 		return PCI_ERS_RESULT_NEED_RESET;
2397 	case pci_channel_io_perm_failure:
2398 		cfg->state = STATE_FAILTERM;
2399 		wake_up_all(&cfg->limbo_waitq);
2400 		scsi_unblock_requests(cfg->host);
2401 		return PCI_ERS_RESULT_DISCONNECT;
2402 	default:
2403 		break;
2404 	}
2405 	return PCI_ERS_RESULT_NEED_RESET;
2406 }
2407 
2408 /**
2409  * cxlflash_pci_slot_reset() - called when PCI slot has been reset
2410  * @pdev:	PCI device struct.
2411  *
2412  * This routine is called by the pci error recovery code after the PCI
2413  * slot has been reset, just before we should resume normal operations.
2414  *
2415  * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
2416  */
2417 static pci_ers_result_t cxlflash_pci_slot_reset(struct pci_dev *pdev)
2418 {
2419 	int rc = 0;
2420 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
2421 	struct device *dev = &cfg->dev->dev;
2422 
2423 	dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
2424 
2425 	rc = init_afu(cfg);
2426 	if (unlikely(rc)) {
2427 		dev_err(dev, "%s: EEH recovery failed! (%d)\n", __func__, rc);
2428 		return PCI_ERS_RESULT_DISCONNECT;
2429 	}
2430 
2431 	return PCI_ERS_RESULT_RECOVERED;
2432 }
2433 
2434 /**
2435  * cxlflash_pci_resume() - called when normal operation can resume
2436  * @pdev:	PCI device struct
2437  */
2438 static void cxlflash_pci_resume(struct pci_dev *pdev)
2439 {
2440 	struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
2441 	struct device *dev = &cfg->dev->dev;
2442 
2443 	dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
2444 
2445 	cfg->state = STATE_NORMAL;
2446 	wake_up_all(&cfg->limbo_waitq);
2447 	scsi_unblock_requests(cfg->host);
2448 }
2449 
2450 static const struct pci_error_handlers cxlflash_err_handler = {
2451 	.error_detected = cxlflash_pci_error_detected,
2452 	.slot_reset = cxlflash_pci_slot_reset,
2453 	.resume = cxlflash_pci_resume,
2454 };
2455 
2456 /*
2457  * PCI device structure
2458  */
2459 static struct pci_driver cxlflash_driver = {
2460 	.name = CXLFLASH_NAME,
2461 	.id_table = cxlflash_pci_table,
2462 	.probe = cxlflash_probe,
2463 	.remove = cxlflash_remove,
2464 	.err_handler = &cxlflash_err_handler,
2465 };
2466 
2467 /**
2468  * init_cxlflash() - module entry point
2469  *
2470  * Return: 0 on success / non-zero on failure
2471  */
2472 static int __init init_cxlflash(void)
2473 {
2474 	pr_info("%s: IBM Power CXL Flash Adapter: %s\n",
2475 		__func__, CXLFLASH_DRIVER_DATE);
2476 
2477 	cxlflash_list_init();
2478 
2479 	return pci_register_driver(&cxlflash_driver);
2480 }
2481 
2482 /**
2483  * exit_cxlflash() - module exit point
2484  */
2485 static void __exit exit_cxlflash(void)
2486 {
2487 	cxlflash_term_global_luns();
2488 	cxlflash_free_errpage();
2489 
2490 	pci_unregister_driver(&cxlflash_driver);
2491 }
2492 
2493 module_init(init_cxlflash);
2494 module_exit(exit_cxlflash);
2495