2 * CXL Flash Device Driver
4 * Written by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>, IBM Corporation
5 * Matthew R. Ochs <mrochs@linux.vnet.ibm.com>, IBM Corporation
7 * Copyright (C) 2015 IBM Corporation
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.
15 #include <linux/delay.h>
16 #include <linux/list.h>
17 #include <linux/module.h>
18 #include <linux/pci.h>
20 #include <asm/unaligned.h>
22 #include <scsi/scsi_cmnd.h>
23 #include <scsi/scsi_host.h>
24 #include <uapi/scsi/cxlflash_ioctl.h>
30 MODULE_DESCRIPTION(CXLFLASH_ADAPTER_NAME);
31 MODULE_AUTHOR("Manoj N. Kumar <manoj@linux.vnet.ibm.com>");
32 MODULE_AUTHOR("Matthew R. Ochs <mrochs@linux.vnet.ibm.com>");
33 MODULE_LICENSE("GPL");
35 static struct class *cxlflash_class;
36 static u32 cxlflash_major;
37 static DECLARE_BITMAP(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
40 * process_cmd_err() - command error handler
41 * @cmd: AFU command that experienced the error.
42 * @scp: SCSI command associated with the AFU command in error.
44 * Translates error bits from AFU command to SCSI command results.
46 static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
48 struct afu *afu = cmd->parent;
49 struct cxlflash_cfg *cfg = afu->parent;
50 struct device *dev = &cfg->dev->dev;
51 struct sisl_ioarcb *ioarcb;
52 struct sisl_ioasa *ioasa;
61 if (ioasa->rc.flags & SISL_RC_FLAGS_UNDERRUN) {
63 scsi_set_resid(scp, resid);
64 dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
65 __func__, cmd, scp, resid);
68 if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
69 dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p\n",
71 scp->result = (DID_ERROR << 16);
74 dev_dbg(dev, "%s: cmd failed afu_rc=%02x scsi_rc=%02x fc_rc=%02x "
75 "afu_extra=%02x scsi_extra=%02x fc_extra=%02x\n", __func__,
76 ioasa->rc.afu_rc, ioasa->rc.scsi_rc, ioasa->rc.fc_rc,
77 ioasa->afu_extra, ioasa->scsi_extra, ioasa->fc_extra);
79 if (ioasa->rc.scsi_rc) {
80 /* We have a SCSI status */
81 if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
82 memcpy(scp->sense_buffer, ioasa->sense_data,
84 scp->result = ioasa->rc.scsi_rc;
86 scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
90 * We encountered an error. Set scp->result based on nature
93 if (ioasa->rc.fc_rc) {
94 /* We have an FC status */
95 switch (ioasa->rc.fc_rc) {
96 case SISL_FC_RC_LINKDOWN:
97 scp->result = (DID_REQUEUE << 16);
99 case SISL_FC_RC_RESID:
100 /* This indicates an FCP resid underrun */
101 if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
102 /* If the SISL_RC_FLAGS_OVERRUN flag was set,
103 * then we will handle this error else where.
104 * If not then we must handle it here.
105 * This is probably an AFU bug.
107 scp->result = (DID_ERROR << 16);
110 case SISL_FC_RC_RESIDERR:
111 /* Resid mismatch between adapter and device */
112 case SISL_FC_RC_TGTABORT:
113 case SISL_FC_RC_ABORTOK:
114 case SISL_FC_RC_ABORTFAIL:
115 case SISL_FC_RC_NOLOGI:
116 case SISL_FC_RC_ABORTPEND:
117 case SISL_FC_RC_WRABORTPEND:
118 case SISL_FC_RC_NOEXP:
119 case SISL_FC_RC_INUSE:
120 scp->result = (DID_ERROR << 16);
125 if (ioasa->rc.afu_rc) {
126 /* We have an AFU error */
127 switch (ioasa->rc.afu_rc) {
128 case SISL_AFU_RC_NO_CHANNELS:
129 scp->result = (DID_NO_CONNECT << 16);
131 case SISL_AFU_RC_DATA_DMA_ERR:
132 switch (ioasa->afu_extra) {
133 case SISL_AFU_DMA_ERR_PAGE_IN:
135 scp->result = (DID_IMM_RETRY << 16);
137 case SISL_AFU_DMA_ERR_INVALID_EA:
139 scp->result = (DID_ERROR << 16);
142 case SISL_AFU_RC_OUT_OF_DATA_BUFS:
144 scp->result = (DID_ALLOC_FAILURE << 16);
147 scp->result = (DID_ERROR << 16);
153 * cmd_complete() - command completion handler
154 * @cmd: AFU command that has completed.
156 * For SCSI commands this routine prepares and submits commands that have
157 * either completed or timed out to the SCSI stack. For internal commands
158 * (TMF or AFU), this routine simply notifies the originator that the
159 * command has completed.
161 static void cmd_complete(struct afu_cmd *cmd)
163 struct scsi_cmnd *scp;
165 struct afu *afu = cmd->parent;
166 struct cxlflash_cfg *cfg = afu->parent;
167 struct device *dev = &cfg->dev->dev;
168 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
170 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
171 list_del(&cmd->list);
172 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
176 if (unlikely(cmd->sa.ioasc))
177 process_cmd_err(cmd, scp);
179 scp->result = (DID_OK << 16);
181 dev_dbg_ratelimited(dev, "%s:scp=%p result=%08x ioasc=%08x\n",
182 __func__, scp, scp->result, cmd->sa.ioasc);
184 } else if (cmd->cmd_tmf) {
185 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
186 cfg->tmf_active = false;
187 wake_up_all_locked(&cfg->tmf_waitq);
188 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
190 complete(&cmd->cevent);
194 * flush_pending_cmds() - flush all pending commands on this hardware queue
195 * @hwq: Hardware queue to flush.
197 * The hardware send queue lock associated with this hardware queue must be
198 * held when calling this routine.
200 static void flush_pending_cmds(struct hwq *hwq)
202 struct cxlflash_cfg *cfg = hwq->afu->parent;
203 struct afu_cmd *cmd, *tmp;
204 struct scsi_cmnd *scp;
207 list_for_each_entry_safe(cmd, tmp, &hwq->pending_cmds, list) {
208 /* Bypass command when on a doneq, cmd_complete() will handle */
209 if (!list_empty(&cmd->queue))
212 list_del(&cmd->list);
216 scp->result = (DID_IMM_RETRY << 16);
219 cmd->cmd_aborted = true;
222 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
223 cfg->tmf_active = false;
224 wake_up_all_locked(&cfg->tmf_waitq);
225 spin_unlock_irqrestore(&cfg->tmf_slock,
228 complete(&cmd->cevent);
234 * context_reset() - reset context via specified register
235 * @hwq: Hardware queue owning the context to be reset.
236 * @reset_reg: MMIO register to perform reset.
238 * When the reset is successful, the SISLite specification guarantees that
239 * the AFU has aborted all currently pending I/O. Accordingly, these commands
242 * Return: 0 on success, -errno on failure
244 static int context_reset(struct hwq *hwq, __be64 __iomem *reset_reg)
246 struct cxlflash_cfg *cfg = hwq->afu->parent;
247 struct device *dev = &cfg->dev->dev;
253 dev_dbg(dev, "%s: hwq=%p\n", __func__, hwq);
255 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
257 writeq_be(val, reset_reg);
259 val = readq_be(reset_reg);
260 if ((val & 0x1) == 0x0) {
265 /* Double delay each time */
267 } while (nretry++ < MC_ROOM_RETRY_CNT);
270 flush_pending_cmds(hwq);
272 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
274 dev_dbg(dev, "%s: returning rc=%d, val=%016llx nretry=%d\n",
275 __func__, rc, val, nretry);
280 * context_reset_ioarrin() - reset context via IOARRIN register
281 * @hwq: Hardware queue owning the context to be reset.
283 * Return: 0 on success, -errno on failure
285 static int context_reset_ioarrin(struct hwq *hwq)
287 return context_reset(hwq, &hwq->host_map->ioarrin);
291 * context_reset_sq() - reset context via SQ_CONTEXT_RESET register
292 * @hwq: Hardware queue owning the context to be reset.
294 * Return: 0 on success, -errno on failure
296 static int context_reset_sq(struct hwq *hwq)
298 return context_reset(hwq, &hwq->host_map->sq_ctx_reset);
302 * send_cmd_ioarrin() - sends an AFU command via IOARRIN register
303 * @afu: AFU associated with the host.
304 * @cmd: AFU command to send.
307 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
309 static int send_cmd_ioarrin(struct afu *afu, struct afu_cmd *cmd)
311 struct cxlflash_cfg *cfg = afu->parent;
312 struct device *dev = &cfg->dev->dev;
313 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
319 * To avoid the performance penalty of MMIO, spread the update of
320 * 'room' over multiple commands.
322 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
323 if (--hwq->room < 0) {
324 room = readq_be(&hwq->host_map->cmd_room);
326 dev_dbg_ratelimited(dev, "%s: no cmd_room to send "
327 "0x%02X, room=0x%016llX\n",
328 __func__, cmd->rcb.cdb[0], room);
330 rc = SCSI_MLQUEUE_HOST_BUSY;
333 hwq->room = room - 1;
336 list_add(&cmd->list, &hwq->pending_cmds);
337 writeq_be((u64)&cmd->rcb, &hwq->host_map->ioarrin);
339 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
340 dev_dbg_ratelimited(dev, "%s: cmd=%p len=%u ea=%016llx rc=%d\n",
341 __func__, cmd, cmd->rcb.data_len, cmd->rcb.data_ea, rc);
346 * send_cmd_sq() - sends an AFU command via SQ ring
347 * @afu: AFU associated with the host.
348 * @cmd: AFU command to send.
351 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
353 static int send_cmd_sq(struct afu *afu, struct afu_cmd *cmd)
355 struct cxlflash_cfg *cfg = afu->parent;
356 struct device *dev = &cfg->dev->dev;
357 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
362 newval = atomic_dec_if_positive(&hwq->hsq_credits);
364 rc = SCSI_MLQUEUE_HOST_BUSY;
368 cmd->rcb.ioasa = &cmd->sa;
370 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
372 *hwq->hsq_curr = cmd->rcb;
373 if (hwq->hsq_curr < hwq->hsq_end)
376 hwq->hsq_curr = hwq->hsq_start;
378 list_add(&cmd->list, &hwq->pending_cmds);
379 writeq_be((u64)hwq->hsq_curr, &hwq->host_map->sq_tail);
381 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
383 dev_dbg(dev, "%s: cmd=%p len=%u ea=%016llx ioasa=%p rc=%d curr=%p "
384 "head=%016llx tail=%016llx\n", __func__, cmd, cmd->rcb.data_len,
385 cmd->rcb.data_ea, cmd->rcb.ioasa, rc, hwq->hsq_curr,
386 readq_be(&hwq->host_map->sq_head),
387 readq_be(&hwq->host_map->sq_tail));
392 * wait_resp() - polls for a response or timeout to a sent AFU command
393 * @afu: AFU associated with the host.
394 * @cmd: AFU command that was sent.
396 * Return: 0 on success, -errno on failure
398 static int wait_resp(struct afu *afu, struct afu_cmd *cmd)
400 struct cxlflash_cfg *cfg = afu->parent;
401 struct device *dev = &cfg->dev->dev;
403 ulong timeout = msecs_to_jiffies(cmd->rcb.timeout * 2 * 1000);
405 timeout = wait_for_completion_timeout(&cmd->cevent, timeout);
409 if (cmd->cmd_aborted)
412 if (unlikely(cmd->sa.ioasc != 0)) {
413 dev_err(dev, "%s: cmd %02x failed, ioasc=%08x\n",
414 __func__, cmd->rcb.cdb[0], cmd->sa.ioasc);
422 * cmd_to_target_hwq() - selects a target hardware queue for a SCSI command
423 * @host: SCSI host associated with device.
424 * @scp: SCSI command to send.
425 * @afu: SCSI command to send.
427 * Hashes a command based upon the hardware queue mode.
429 * Return: Trusted index of target hardware queue
431 static u32 cmd_to_target_hwq(struct Scsi_Host *host, struct scsi_cmnd *scp,
437 if (afu->num_hwqs == 1)
440 switch (afu->hwq_mode) {
442 hwq = afu->hwq_rr_count++ % afu->num_hwqs;
445 tag = blk_mq_unique_tag(scp->request);
446 hwq = blk_mq_unique_tag_to_hwq(tag);
449 hwq = smp_processor_id() % afu->num_hwqs;
459 * send_tmf() - sends a Task Management Function (TMF)
460 * @cfg: Internal structure associated with the host.
461 * @sdev: SCSI device destined for TMF.
462 * @tmfcmd: TMF command to send.
465 * 0 on success, SCSI_MLQUEUE_HOST_BUSY or -errno on failure
467 static int send_tmf(struct cxlflash_cfg *cfg, struct scsi_device *sdev,
470 struct afu *afu = cfg->afu;
471 struct afu_cmd *cmd = NULL;
472 struct device *dev = &cfg->dev->dev;
473 struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
474 bool needs_deletion = false;
480 buf = kzalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
481 if (unlikely(!buf)) {
482 dev_err(dev, "%s: no memory for command\n", __func__);
487 cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
488 INIT_LIST_HEAD(&cmd->queue);
490 /* When Task Management Function is active do not send another */
491 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
493 wait_event_interruptible_lock_irq(cfg->tmf_waitq,
496 cfg->tmf_active = true;
497 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
501 cmd->hwq_index = hwq->index;
503 cmd->rcb.ctx_id = hwq->ctx_hndl;
504 cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
505 cmd->rcb.port_sel = CHAN2PORTMASK(sdev->channel);
506 cmd->rcb.lun_id = lun_to_lunid(sdev->lun);
507 cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
508 SISL_REQ_FLAGS_SUP_UNDERRUN |
509 SISL_REQ_FLAGS_TMF_CMD);
510 memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
512 rc = afu->send_cmd(afu, cmd);
514 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
515 cfg->tmf_active = false;
516 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
520 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
521 to = msecs_to_jiffies(5000);
522 to = wait_event_interruptible_lock_irq_timeout(cfg->tmf_waitq,
527 dev_err(dev, "%s: TMF timed out\n", __func__);
529 needs_deletion = true;
530 } else if (cmd->cmd_aborted) {
531 dev_err(dev, "%s: TMF aborted\n", __func__);
533 } else if (cmd->sa.ioasc) {
534 dev_err(dev, "%s: TMF failed ioasc=%08x\n",
535 __func__, cmd->sa.ioasc);
538 cfg->tmf_active = false;
539 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
541 if (needs_deletion) {
542 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
543 list_del(&cmd->list);
544 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
552 * cxlflash_driver_info() - information handler for this host driver
553 * @host: SCSI host associated with device.
555 * Return: A string describing the device.
557 static const char *cxlflash_driver_info(struct Scsi_Host *host)
559 return CXLFLASH_ADAPTER_NAME;
563 * cxlflash_queuecommand() - sends a mid-layer request
564 * @host: SCSI host associated with device.
565 * @scp: SCSI command to send.
567 * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
569 static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
571 struct cxlflash_cfg *cfg = shost_priv(host);
572 struct afu *afu = cfg->afu;
573 struct device *dev = &cfg->dev->dev;
574 struct afu_cmd *cmd = sc_to_afuci(scp);
575 struct scatterlist *sg = scsi_sglist(scp);
576 int hwq_index = cmd_to_target_hwq(host, scp, afu);
577 struct hwq *hwq = get_hwq(afu, hwq_index);
578 u16 req_flags = SISL_REQ_FLAGS_SUP_UNDERRUN;
582 dev_dbg_ratelimited(dev, "%s: (scp=%p) %d/%d/%d/%llu "
583 "cdb=(%08x-%08x-%08x-%08x)\n",
584 __func__, scp, host->host_no, scp->device->channel,
585 scp->device->id, scp->device->lun,
586 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
587 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
588 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
589 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
592 * If a Task Management Function is active, wait for it to complete
593 * before continuing with regular commands.
595 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
596 if (cfg->tmf_active) {
597 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
598 rc = SCSI_MLQUEUE_HOST_BUSY;
601 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
603 switch (cfg->state) {
607 dev_dbg_ratelimited(dev, "%s: device is in reset\n", __func__);
608 rc = SCSI_MLQUEUE_HOST_BUSY;
611 dev_dbg_ratelimited(dev, "%s: device has failed\n", __func__);
612 scp->result = (DID_NO_CONNECT << 16);
617 atomic_inc(&afu->cmds_active);
622 cmd->rcb.data_len = sg->length;
623 cmd->rcb.data_ea = (uintptr_t)sg_virt(sg);
628 cmd->hwq_index = hwq_index;
631 cmd->rcb.ctx_id = hwq->ctx_hndl;
632 cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
633 cmd->rcb.port_sel = CHAN2PORTMASK(scp->device->channel);
634 cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
636 if (scp->sc_data_direction == DMA_TO_DEVICE)
637 req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
639 cmd->rcb.req_flags = req_flags;
640 memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
642 rc = afu->send_cmd(afu, cmd);
643 atomic_dec(&afu->cmds_active);
649 * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
650 * @cfg: Internal structure associated with the host.
652 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
654 struct pci_dev *pdev = cfg->dev;
656 if (pci_channel_offline(pdev))
657 wait_event_timeout(cfg->reset_waitq,
658 !pci_channel_offline(pdev),
659 CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
663 * free_mem() - free memory associated with the AFU
664 * @cfg: Internal structure associated with the host.
666 static void free_mem(struct cxlflash_cfg *cfg)
668 struct afu *afu = cfg->afu;
671 free_pages((ulong)afu, get_order(sizeof(struct afu)));
677 * cxlflash_reset_sync() - synchronizing point for asynchronous resets
678 * @cfg: Internal structure associated with the host.
680 static void cxlflash_reset_sync(struct cxlflash_cfg *cfg)
682 if (cfg->async_reset_cookie == 0)
685 /* Wait until all async calls prior to this cookie have completed */
686 async_synchronize_cookie(cfg->async_reset_cookie + 1);
687 cfg->async_reset_cookie = 0;
691 * stop_afu() - stops the AFU command timers and unmaps the MMIO space
692 * @cfg: Internal structure associated with the host.
694 * Safe to call with AFU in a partially allocated/initialized state.
696 * Cancels scheduled worker threads, waits for any active internal AFU
697 * commands to timeout, disables IRQ polling and then unmaps the MMIO space.
699 static void stop_afu(struct cxlflash_cfg *cfg)
701 struct afu *afu = cfg->afu;
705 cancel_work_sync(&cfg->work_q);
706 if (!current_is_async())
707 cxlflash_reset_sync(cfg);
710 while (atomic_read(&afu->cmds_active))
713 if (afu_is_irqpoll_enabled(afu)) {
714 for (i = 0; i < afu->num_hwqs; i++) {
715 hwq = get_hwq(afu, i);
717 irq_poll_disable(&hwq->irqpoll);
721 if (likely(afu->afu_map)) {
722 cfg->ops->psa_unmap(afu->afu_map);
729 * term_intr() - disables all AFU interrupts
730 * @cfg: Internal structure associated with the host.
731 * @level: Depth of allocation, where to begin waterfall tear down.
732 * @index: Index of the hardware queue.
734 * Safe to call with AFU/MC in partially allocated/initialized state.
736 static void term_intr(struct cxlflash_cfg *cfg, enum undo_level level,
739 struct afu *afu = cfg->afu;
740 struct device *dev = &cfg->dev->dev;
744 dev_err(dev, "%s: returning with NULL afu\n", __func__);
748 hwq = get_hwq(afu, index);
750 if (!hwq->ctx_cookie) {
751 dev_err(dev, "%s: returning with NULL MC\n", __func__);
757 /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
758 if (index == PRIMARY_HWQ)
759 cfg->ops->unmap_afu_irq(hwq->ctx_cookie, 3, hwq);
761 cfg->ops->unmap_afu_irq(hwq->ctx_cookie, 2, hwq);
763 cfg->ops->unmap_afu_irq(hwq->ctx_cookie, 1, hwq);
765 cfg->ops->free_afu_irqs(hwq->ctx_cookie);
768 /* No action required */
774 * term_mc() - terminates the master context
775 * @cfg: Internal structure associated with the host.
776 * @index: Index of the hardware queue.
778 * Safe to call with AFU/MC in partially allocated/initialized state.
780 static void term_mc(struct cxlflash_cfg *cfg, u32 index)
782 struct afu *afu = cfg->afu;
783 struct device *dev = &cfg->dev->dev;
788 dev_err(dev, "%s: returning with NULL afu\n", __func__);
792 hwq = get_hwq(afu, index);
794 if (!hwq->ctx_cookie) {
795 dev_err(dev, "%s: returning with NULL MC\n", __func__);
799 WARN_ON(cfg->ops->stop_context(hwq->ctx_cookie));
800 if (index != PRIMARY_HWQ)
801 WARN_ON(cfg->ops->release_context(hwq->ctx_cookie));
802 hwq->ctx_cookie = NULL;
804 spin_lock_irqsave(&hwq->hrrq_slock, lock_flags);
805 hwq->hrrq_online = false;
806 spin_unlock_irqrestore(&hwq->hrrq_slock, lock_flags);
808 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
809 flush_pending_cmds(hwq);
810 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
814 * term_afu() - terminates the AFU
815 * @cfg: Internal structure associated with the host.
817 * Safe to call with AFU/MC in partially allocated/initialized state.
819 static void term_afu(struct cxlflash_cfg *cfg)
821 struct device *dev = &cfg->dev->dev;
825 * Tear down is carefully orchestrated to ensure
826 * no interrupts can come in when the problem state
829 * 1) Disable all AFU interrupts for each master
830 * 2) Unmap the problem state area
831 * 3) Stop each master context
833 for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
834 term_intr(cfg, UNMAP_THREE, k);
838 for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
841 dev_dbg(dev, "%s: returning\n", __func__);
845 * notify_shutdown() - notifies device of pending shutdown
846 * @cfg: Internal structure associated with the host.
847 * @wait: Whether to wait for shutdown processing to complete.
849 * This function will notify the AFU that the adapter is being shutdown
850 * and will wait for shutdown processing to complete if wait is true.
851 * This notification should flush pending I/Os to the device and halt
852 * further I/Os until the next AFU reset is issued and device restarted.
854 static void notify_shutdown(struct cxlflash_cfg *cfg, bool wait)
856 struct afu *afu = cfg->afu;
857 struct device *dev = &cfg->dev->dev;
858 struct dev_dependent_vals *ddv;
859 __be64 __iomem *fc_port_regs;
861 int i, retry_cnt = 0;
863 ddv = (struct dev_dependent_vals *)cfg->dev_id->driver_data;
864 if (!(ddv->flags & CXLFLASH_NOTIFY_SHUTDOWN))
867 if (!afu || !afu->afu_map) {
868 dev_dbg(dev, "%s: Problem state area not mapped\n", __func__);
873 for (i = 0; i < cfg->num_fc_ports; i++) {
874 fc_port_regs = get_fc_port_regs(cfg, i);
876 reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
877 reg |= SISL_FC_SHUTDOWN_NORMAL;
878 writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
884 /* Wait up to 1.5 seconds for shutdown processing to complete */
885 for (i = 0; i < cfg->num_fc_ports; i++) {
886 fc_port_regs = get_fc_port_regs(cfg, i);
890 status = readq_be(&fc_port_regs[FC_STATUS / 8]);
891 if (status & SISL_STATUS_SHUTDOWN_COMPLETE)
893 if (++retry_cnt >= MC_RETRY_CNT) {
894 dev_dbg(dev, "%s: port %d shutdown processing "
895 "not yet completed\n", __func__, i);
898 msleep(100 * retry_cnt);
904 * cxlflash_get_minor() - gets the first available minor number
906 * Return: Unique minor number that can be used to create the character device.
908 static int cxlflash_get_minor(void)
913 bit = find_first_zero_bit(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
914 if (bit >= CXLFLASH_MAX_ADAPTERS)
917 minor = bit & MINORMASK;
918 set_bit(minor, cxlflash_minor);
923 * cxlflash_put_minor() - releases the minor number
924 * @minor: Minor number that is no longer needed.
926 static void cxlflash_put_minor(int minor)
928 clear_bit(minor, cxlflash_minor);
932 * cxlflash_release_chrdev() - release the character device for the host
933 * @cfg: Internal structure associated with the host.
935 static void cxlflash_release_chrdev(struct cxlflash_cfg *cfg)
937 device_unregister(cfg->chardev);
939 cdev_del(&cfg->cdev);
940 cxlflash_put_minor(MINOR(cfg->cdev.dev));
944 * cxlflash_remove() - PCI entry point to tear down host
945 * @pdev: PCI device associated with the host.
947 * Safe to use as a cleanup in partially allocated/initialized state. Note that
948 * the reset_waitq is flushed as part of the stop/termination of user contexts.
950 static void cxlflash_remove(struct pci_dev *pdev)
952 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
953 struct device *dev = &pdev->dev;
956 if (!pci_is_enabled(pdev)) {
957 dev_dbg(dev, "%s: Device is disabled\n", __func__);
961 /* Yield to running recovery threads before continuing with remove */
962 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
963 cfg->state != STATE_PROBING);
964 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
966 wait_event_interruptible_lock_irq(cfg->tmf_waitq,
969 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
971 /* Notify AFU and wait for shutdown processing to complete */
972 notify_shutdown(cfg, true);
974 cfg->state = STATE_FAILTERM;
975 cxlflash_stop_term_user_contexts(cfg);
977 switch (cfg->init_state) {
978 case INIT_STATE_CDEV:
979 cxlflash_release_chrdev(cfg);
980 case INIT_STATE_SCSI:
981 cxlflash_term_local_luns(cfg);
982 scsi_remove_host(cfg->host);
986 cfg->ops->destroy_afu(cfg->afu_cookie);
987 pci_disable_device(pdev);
988 case INIT_STATE_NONE:
990 scsi_host_put(cfg->host);
994 dev_dbg(dev, "%s: returning\n", __func__);
998 * alloc_mem() - allocates the AFU and its command pool
999 * @cfg: Internal structure associated with the host.
1001 * A partially allocated state remains on failure.
1005 * -ENOMEM on failure to allocate memory
1007 static int alloc_mem(struct cxlflash_cfg *cfg)
1010 struct device *dev = &cfg->dev->dev;
1012 /* AFU is ~28k, i.e. only one 64k page or up to seven 4k pages */
1013 cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1014 get_order(sizeof(struct afu)));
1015 if (unlikely(!cfg->afu)) {
1016 dev_err(dev, "%s: cannot get %d free pages\n",
1017 __func__, get_order(sizeof(struct afu)));
1021 cfg->afu->parent = cfg;
1022 cfg->afu->desired_hwqs = CXLFLASH_DEF_HWQS;
1023 cfg->afu->afu_map = NULL;
1029 * init_pci() - initializes the host as a PCI device
1030 * @cfg: Internal structure associated with the host.
1032 * Return: 0 on success, -errno on failure
1034 static int init_pci(struct cxlflash_cfg *cfg)
1036 struct pci_dev *pdev = cfg->dev;
1037 struct device *dev = &cfg->dev->dev;
1040 rc = pci_enable_device(pdev);
1041 if (rc || pci_channel_offline(pdev)) {
1042 if (pci_channel_offline(pdev)) {
1043 cxlflash_wait_for_pci_err_recovery(cfg);
1044 rc = pci_enable_device(pdev);
1048 dev_err(dev, "%s: Cannot enable adapter\n", __func__);
1049 cxlflash_wait_for_pci_err_recovery(cfg);
1055 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1060 * init_scsi() - adds the host to the SCSI stack and kicks off host scan
1061 * @cfg: Internal structure associated with the host.
1063 * Return: 0 on success, -errno on failure
1065 static int init_scsi(struct cxlflash_cfg *cfg)
1067 struct pci_dev *pdev = cfg->dev;
1068 struct device *dev = &cfg->dev->dev;
1071 rc = scsi_add_host(cfg->host, &pdev->dev);
1073 dev_err(dev, "%s: scsi_add_host failed rc=%d\n", __func__, rc);
1077 scsi_scan_host(cfg->host);
1080 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1085 * set_port_online() - transitions the specified host FC port to online state
1086 * @fc_regs: Top of MMIO region defined for specified port.
1088 * The provided MMIO region must be mapped prior to call. Online state means
1089 * that the FC link layer has synced, completed the handshaking process, and
1090 * is ready for login to start.
1092 static void set_port_online(__be64 __iomem *fc_regs)
1096 cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1097 cmdcfg &= (~FC_MTIP_CMDCONFIG_OFFLINE); /* clear OFF_LINE */
1098 cmdcfg |= (FC_MTIP_CMDCONFIG_ONLINE); /* set ON_LINE */
1099 writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1103 * set_port_offline() - transitions the specified host FC port to offline state
1104 * @fc_regs: Top of MMIO region defined for specified port.
1106 * The provided MMIO region must be mapped prior to call.
1108 static void set_port_offline(__be64 __iomem *fc_regs)
1112 cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1113 cmdcfg &= (~FC_MTIP_CMDCONFIG_ONLINE); /* clear ON_LINE */
1114 cmdcfg |= (FC_MTIP_CMDCONFIG_OFFLINE); /* set OFF_LINE */
1115 writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1119 * wait_port_online() - waits for the specified host FC port come online
1120 * @fc_regs: Top of MMIO region defined for specified port.
1121 * @delay_us: Number of microseconds to delay between reading port status.
1122 * @nretry: Number of cycles to retry reading port status.
1124 * The provided MMIO region must be mapped prior to call. This will timeout
1125 * when the cable is not plugged in.
1128 * TRUE (1) when the specified port is online
1129 * FALSE (0) when the specified port fails to come online after timeout
1131 static bool wait_port_online(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1135 WARN_ON(delay_us < 1000);
1138 msleep(delay_us / 1000);
1139 status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1140 if (status == U64_MAX)
1142 } while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_ONLINE &&
1145 return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_ONLINE);
1149 * wait_port_offline() - waits for the specified host FC port go offline
1150 * @fc_regs: Top of MMIO region defined for specified port.
1151 * @delay_us: Number of microseconds to delay between reading port status.
1152 * @nretry: Number of cycles to retry reading port status.
1154 * The provided MMIO region must be mapped prior to call.
1157 * TRUE (1) when the specified port is offline
1158 * FALSE (0) when the specified port fails to go offline after timeout
1160 static bool wait_port_offline(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1164 WARN_ON(delay_us < 1000);
1167 msleep(delay_us / 1000);
1168 status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1169 if (status == U64_MAX)
1171 } while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_OFFLINE &&
1174 return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_OFFLINE);
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.
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.
1190 static void afu_set_wwpn(struct afu *afu, int port, __be64 __iomem *fc_regs,
1193 struct cxlflash_cfg *cfg = afu->parent;
1194 struct device *dev = &cfg->dev->dev;
1196 set_port_offline(fc_regs);
1197 if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1198 FC_PORT_STATUS_RETRY_CNT)) {
1199 dev_dbg(dev, "%s: wait on port %d to go offline timed out\n",
1203 writeq_be(wwpn, &fc_regs[FC_PNAME / 8]);
1205 set_port_online(fc_regs);
1206 if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1207 FC_PORT_STATUS_RETRY_CNT)) {
1208 dev_dbg(dev, "%s: wait on port %d to go online timed out\n",
1214 * afu_link_reset() - resets the specified host FC port
1215 * @afu: AFU associated with the host that owns the specified FC port.
1216 * @port: Port number being configured.
1217 * @fc_regs: Top of MMIO region defined for specified port.
1219 * The provided MMIO region must be mapped prior to call. The sequence to
1220 * reset the port involves toggling it offline and then back online. This
1221 * action can cause this routine to delay up to a few seconds. An effort
1222 * is made to maintain link with the device by switching to host to use
1223 * the alternate port exclusively while the reset takes place.
1224 * failure to come online is overridden.
1226 static void afu_link_reset(struct afu *afu, int port, __be64 __iomem *fc_regs)
1228 struct cxlflash_cfg *cfg = afu->parent;
1229 struct device *dev = &cfg->dev->dev;
1232 /* first switch the AFU to the other links, if any */
1233 port_sel = readq_be(&afu->afu_map->global.regs.afu_port_sel);
1234 port_sel &= ~(1ULL << port);
1235 writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1236 cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1238 set_port_offline(fc_regs);
1239 if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1240 FC_PORT_STATUS_RETRY_CNT))
1241 dev_err(dev, "%s: wait on port %d to go offline timed out\n",
1244 set_port_online(fc_regs);
1245 if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1246 FC_PORT_STATUS_RETRY_CNT))
1247 dev_err(dev, "%s: wait on port %d to go online timed out\n",
1250 /* switch back to include this port */
1251 port_sel |= (1ULL << port);
1252 writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1253 cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1255 dev_dbg(dev, "%s: returning port_sel=%016llx\n", __func__, port_sel);
1259 * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1260 * @afu: AFU associated with the host.
1262 static void afu_err_intr_init(struct afu *afu)
1264 struct cxlflash_cfg *cfg = afu->parent;
1265 __be64 __iomem *fc_port_regs;
1267 struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
1270 /* global async interrupts: AFU clears afu_ctrl on context exit
1271 * if async interrupts were sent to that context. This prevents
1272 * the AFU form sending further async interrupts when
1274 * nobody to receive them.
1278 writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_mask);
1279 /* set LISN# to send and point to primary master context */
1280 reg = ((u64) (((hwq->ctx_hndl << 8) | SISL_MSI_ASYNC_ERROR)) << 40);
1282 if (afu->internal_lun)
1283 reg |= 1; /* Bit 63 indicates local lun */
1284 writeq_be(reg, &afu->afu_map->global.regs.afu_ctrl);
1286 writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1287 /* unmask bits that are of interest */
1288 /* note: afu can send an interrupt after this step */
1289 writeq_be(SISL_ASTATUS_MASK, &afu->afu_map->global.regs.aintr_mask);
1290 /* clear again in case a bit came on after previous clear but before */
1292 writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1294 /* Clear/Set internal lun bits */
1295 fc_port_regs = get_fc_port_regs(cfg, 0);
1296 reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
1297 reg &= SISL_FC_INTERNAL_MASK;
1298 if (afu->internal_lun)
1299 reg |= ((u64)(afu->internal_lun - 1) << SISL_FC_INTERNAL_SHIFT);
1300 writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
1302 /* now clear FC errors */
1303 for (i = 0; i < cfg->num_fc_ports; i++) {
1304 fc_port_regs = get_fc_port_regs(cfg, i);
1306 writeq_be(0xFFFFFFFFU, &fc_port_regs[FC_ERROR / 8]);
1307 writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1310 /* sync interrupts for master's IOARRIN write */
1311 /* note that unlike asyncs, there can be no pending sync interrupts */
1312 /* at this time (this is a fresh context and master has not written */
1313 /* IOARRIN yet), so there is nothing to clear. */
1315 /* set LISN#, it is always sent to the context that wrote IOARRIN */
1316 for (i = 0; i < afu->num_hwqs; i++) {
1317 hwq = get_hwq(afu, i);
1319 reg = readq_be(&hwq->host_map->ctx_ctrl);
1320 WARN_ON((reg & SISL_CTX_CTRL_LISN_MASK) != 0);
1321 reg |= SISL_MSI_SYNC_ERROR;
1322 writeq_be(reg, &hwq->host_map->ctx_ctrl);
1323 writeq_be(SISL_ISTATUS_MASK, &hwq->host_map->intr_mask);
1328 * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1329 * @irq: Interrupt number.
1330 * @data: Private data provided at interrupt registration, the AFU.
1332 * Return: Always return IRQ_HANDLED.
1334 static irqreturn_t cxlflash_sync_err_irq(int irq, void *data)
1336 struct hwq *hwq = (struct hwq *)data;
1337 struct cxlflash_cfg *cfg = hwq->afu->parent;
1338 struct device *dev = &cfg->dev->dev;
1342 reg = readq_be(&hwq->host_map->intr_status);
1343 reg_unmasked = (reg & SISL_ISTATUS_UNMASK);
1345 if (reg_unmasked == 0UL) {
1346 dev_err(dev, "%s: spurious interrupt, intr_status=%016llx\n",
1348 goto cxlflash_sync_err_irq_exit;
1351 dev_err(dev, "%s: unexpected interrupt, intr_status=%016llx\n",
1354 writeq_be(reg_unmasked, &hwq->host_map->intr_clear);
1356 cxlflash_sync_err_irq_exit:
1361 * process_hrrq() - process the read-response queue
1362 * @afu: AFU associated with the host.
1363 * @doneq: Queue of commands harvested from the RRQ.
1364 * @budget: Threshold of RRQ entries to process.
1366 * This routine must be called holding the disabled RRQ spin lock.
1368 * Return: The number of entries processed.
1370 static int process_hrrq(struct hwq *hwq, struct list_head *doneq, int budget)
1372 struct afu *afu = hwq->afu;
1373 struct afu_cmd *cmd;
1374 struct sisl_ioasa *ioasa;
1375 struct sisl_ioarcb *ioarcb;
1376 bool toggle = hwq->toggle;
1379 *hrrq_start = hwq->hrrq_start,
1380 *hrrq_end = hwq->hrrq_end,
1381 *hrrq_curr = hwq->hrrq_curr;
1383 /* Process ready RRQ entries up to the specified budget (if any) */
1387 if ((entry & SISL_RESP_HANDLE_T_BIT) != toggle)
1390 entry &= ~SISL_RESP_HANDLE_T_BIT;
1392 if (afu_is_sq_cmd_mode(afu)) {
1393 ioasa = (struct sisl_ioasa *)entry;
1394 cmd = container_of(ioasa, struct afu_cmd, sa);
1396 ioarcb = (struct sisl_ioarcb *)entry;
1397 cmd = container_of(ioarcb, struct afu_cmd, rcb);
1400 list_add_tail(&cmd->queue, doneq);
1402 /* Advance to next entry or wrap and flip the toggle bit */
1403 if (hrrq_curr < hrrq_end)
1406 hrrq_curr = hrrq_start;
1407 toggle ^= SISL_RESP_HANDLE_T_BIT;
1410 atomic_inc(&hwq->hsq_credits);
1413 if (budget > 0 && num_hrrq >= budget)
1417 hwq->hrrq_curr = hrrq_curr;
1418 hwq->toggle = toggle;
1424 * process_cmd_doneq() - process a queue of harvested RRQ commands
1425 * @doneq: Queue of completed commands.
1427 * Note that upon return the queue can no longer be trusted.
1429 static void process_cmd_doneq(struct list_head *doneq)
1431 struct afu_cmd *cmd, *tmp;
1433 WARN_ON(list_empty(doneq));
1435 list_for_each_entry_safe(cmd, tmp, doneq, queue)
1440 * cxlflash_irqpoll() - process a queue of harvested RRQ commands
1441 * @irqpoll: IRQ poll structure associated with queue to poll.
1442 * @budget: Threshold of RRQ entries to process per poll.
1444 * Return: The number of entries processed.
1446 static int cxlflash_irqpoll(struct irq_poll *irqpoll, int budget)
1448 struct hwq *hwq = container_of(irqpoll, struct hwq, irqpoll);
1449 unsigned long hrrq_flags;
1451 int num_entries = 0;
1453 spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1455 num_entries = process_hrrq(hwq, &doneq, budget);
1456 if (num_entries < budget)
1457 irq_poll_complete(irqpoll);
1459 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1461 process_cmd_doneq(&doneq);
1466 * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1467 * @irq: Interrupt number.
1468 * @data: Private data provided at interrupt registration, the AFU.
1470 * Return: IRQ_HANDLED or IRQ_NONE when no ready entries found.
1472 static irqreturn_t cxlflash_rrq_irq(int irq, void *data)
1474 struct hwq *hwq = (struct hwq *)data;
1475 struct afu *afu = hwq->afu;
1476 unsigned long hrrq_flags;
1478 int num_entries = 0;
1480 spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1482 /* Silently drop spurious interrupts when queue is not online */
1483 if (!hwq->hrrq_online) {
1484 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1488 if (afu_is_irqpoll_enabled(afu)) {
1489 irq_poll_sched(&hwq->irqpoll);
1490 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1494 num_entries = process_hrrq(hwq, &doneq, -1);
1495 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1497 if (num_entries == 0)
1500 process_cmd_doneq(&doneq);
1505 * Asynchronous interrupt information table
1508 * - Order matters here as this array is indexed by bit position.
1510 * - The checkpatch script considers the BUILD_SISL_ASTATUS_FC_PORT macro
1511 * as complex and complains due to a lack of parentheses/braces.
1513 #define ASTATUS_FC(_a, _b, _c, _d) \
1514 { SISL_ASTATUS_FC##_a##_##_b, _c, _a, (_d) }
1516 #define BUILD_SISL_ASTATUS_FC_PORT(_a) \
1517 ASTATUS_FC(_a, LINK_UP, "link up", 0), \
1518 ASTATUS_FC(_a, LINK_DN, "link down", 0), \
1519 ASTATUS_FC(_a, LOGI_S, "login succeeded", SCAN_HOST), \
1520 ASTATUS_FC(_a, LOGI_F, "login failed", CLR_FC_ERROR), \
1521 ASTATUS_FC(_a, LOGI_R, "login timed out, retrying", LINK_RESET), \
1522 ASTATUS_FC(_a, CRC_T, "CRC threshold exceeded", LINK_RESET), \
1523 ASTATUS_FC(_a, LOGO, "target initiated LOGO", 0), \
1524 ASTATUS_FC(_a, OTHER, "other error", CLR_FC_ERROR | LINK_RESET)
1526 static const struct asyc_intr_info ainfo[] = {
1527 BUILD_SISL_ASTATUS_FC_PORT(1),
1528 BUILD_SISL_ASTATUS_FC_PORT(0),
1529 BUILD_SISL_ASTATUS_FC_PORT(3),
1530 BUILD_SISL_ASTATUS_FC_PORT(2)
1534 * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1535 * @irq: Interrupt number.
1536 * @data: Private data provided at interrupt registration, the AFU.
1538 * Return: Always return IRQ_HANDLED.
1540 static irqreturn_t cxlflash_async_err_irq(int irq, void *data)
1542 struct hwq *hwq = (struct hwq *)data;
1543 struct afu *afu = hwq->afu;
1544 struct cxlflash_cfg *cfg = afu->parent;
1545 struct device *dev = &cfg->dev->dev;
1546 const struct asyc_intr_info *info;
1547 struct sisl_global_map __iomem *global = &afu->afu_map->global;
1548 __be64 __iomem *fc_port_regs;
1554 reg = readq_be(&global->regs.aintr_status);
1555 reg_unmasked = (reg & SISL_ASTATUS_UNMASK);
1557 if (unlikely(reg_unmasked == 0)) {
1558 dev_err(dev, "%s: spurious interrupt, aintr_status=%016llx\n",
1563 /* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1564 writeq_be(reg_unmasked, &global->regs.aintr_clear);
1566 /* Check each bit that is on */
1567 for_each_set_bit(bit, (ulong *)®_unmasked, BITS_PER_LONG) {
1568 if (unlikely(bit >= ARRAY_SIZE(ainfo))) {
1574 if (unlikely(info->status != 1ULL << bit)) {
1580 fc_port_regs = get_fc_port_regs(cfg, port);
1582 dev_err(dev, "%s: FC Port %d -> %s, fc_status=%016llx\n",
1583 __func__, port, info->desc,
1584 readq_be(&fc_port_regs[FC_STATUS / 8]));
1587 * Do link reset first, some OTHER errors will set FC_ERROR
1588 * again if cleared before or w/o a reset
1590 if (info->action & LINK_RESET) {
1591 dev_err(dev, "%s: FC Port %d: resetting link\n",
1593 cfg->lr_state = LINK_RESET_REQUIRED;
1594 cfg->lr_port = port;
1595 schedule_work(&cfg->work_q);
1598 if (info->action & CLR_FC_ERROR) {
1599 reg = readq_be(&fc_port_regs[FC_ERROR / 8]);
1602 * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1603 * should be the same and tracing one is sufficient.
1606 dev_err(dev, "%s: fc %d: clearing fc_error=%016llx\n",
1607 __func__, port, reg);
1609 writeq_be(reg, &fc_port_regs[FC_ERROR / 8]);
1610 writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1613 if (info->action & SCAN_HOST) {
1614 atomic_inc(&cfg->scan_host_needed);
1615 schedule_work(&cfg->work_q);
1624 * read_vpd() - obtains the WWPNs from VPD
1625 * @cfg: Internal structure associated with the host.
1626 * @wwpn: Array of size MAX_FC_PORTS to pass back WWPNs
1628 * Return: 0 on success, -errno on failure
1630 static int read_vpd(struct cxlflash_cfg *cfg, u64 wwpn[])
1632 struct device *dev = &cfg->dev->dev;
1633 struct pci_dev *pdev = cfg->dev;
1635 int ro_start, ro_size, i, j, k;
1637 char vpd_data[CXLFLASH_VPD_LEN];
1638 char tmp_buf[WWPN_BUF_LEN] = { 0 };
1639 const struct dev_dependent_vals *ddv = (struct dev_dependent_vals *)
1640 cfg->dev_id->driver_data;
1641 const bool wwpn_vpd_required = ddv->flags & CXLFLASH_WWPN_VPD_REQUIRED;
1642 const char *wwpn_vpd_tags[MAX_FC_PORTS] = { "V5", "V6", "V7", "V8" };
1644 /* Get the VPD data from the device */
1645 vpd_size = cfg->ops->read_adapter_vpd(pdev, vpd_data, sizeof(vpd_data));
1646 if (unlikely(vpd_size <= 0)) {
1647 dev_err(dev, "%s: Unable to read VPD (size = %ld)\n",
1648 __func__, vpd_size);
1653 /* Get the read only section offset */
1654 ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size,
1655 PCI_VPD_LRDT_RO_DATA);
1656 if (unlikely(ro_start < 0)) {
1657 dev_err(dev, "%s: VPD Read-only data not found\n", __func__);
1662 /* Get the read only section size, cap when extends beyond read VPD */
1663 ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
1665 i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1666 if (unlikely((i + j) > vpd_size)) {
1667 dev_dbg(dev, "%s: Might need to read more VPD (%d > %ld)\n",
1668 __func__, (i + j), vpd_size);
1669 ro_size = vpd_size - i;
1673 * Find the offset of the WWPN tag within the read only
1674 * VPD data and validate the found field (partials are
1675 * no good to us). Convert the ASCII data to an integer
1676 * value. Note that we must copy to a temporary buffer
1677 * because the conversion service requires that the ASCII
1678 * string be terminated.
1680 * Allow for WWPN not being found for all devices, setting
1681 * the returned WWPN to zero when not found. Notify with a
1682 * log error for cards that should have had WWPN keywords
1683 * in the VPD - cards requiring WWPN will not have their
1684 * ports programmed and operate in an undefined state.
1686 for (k = 0; k < cfg->num_fc_ports; k++) {
1688 i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1690 i = pci_vpd_find_info_keyword(vpd_data, i, j, wwpn_vpd_tags[k]);
1692 if (wwpn_vpd_required)
1693 dev_err(dev, "%s: Port %d WWPN not found\n",
1699 j = pci_vpd_info_field_size(&vpd_data[i]);
1700 i += PCI_VPD_INFO_FLD_HDR_SIZE;
1701 if (unlikely((i + j > vpd_size) || (j != WWPN_LEN))) {
1702 dev_err(dev, "%s: Port %d WWPN incomplete or bad VPD\n",
1708 memcpy(tmp_buf, &vpd_data[i], WWPN_LEN);
1709 rc = kstrtoul(tmp_buf, WWPN_LEN, (ulong *)&wwpn[k]);
1711 dev_err(dev, "%s: WWPN conversion failed for port %d\n",
1717 dev_dbg(dev, "%s: wwpn%d=%016llx\n", __func__, k, wwpn[k]);
1721 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1726 * init_pcr() - initialize the provisioning and control registers
1727 * @cfg: Internal structure associated with the host.
1729 * Also sets up fast access to the mapped registers and initializes AFU
1730 * command fields that never change.
1732 static void init_pcr(struct cxlflash_cfg *cfg)
1734 struct afu *afu = cfg->afu;
1735 struct sisl_ctrl_map __iomem *ctrl_map;
1740 for (i = 0; i < MAX_CONTEXT; i++) {
1741 ctrl_map = &afu->afu_map->ctrls[i].ctrl;
1742 /* Disrupt any clients that could be running */
1743 /* e.g. clients that survived a master restart */
1744 writeq_be(0, &ctrl_map->rht_start);
1745 writeq_be(0, &ctrl_map->rht_cnt_id);
1746 writeq_be(0, &ctrl_map->ctx_cap);
1749 /* Copy frequently used fields into hwq */
1750 for (i = 0; i < afu->num_hwqs; i++) {
1751 hwq = get_hwq(afu, i);
1752 cookie = hwq->ctx_cookie;
1754 hwq->ctx_hndl = (u16) cfg->ops->process_element(cookie);
1755 hwq->host_map = &afu->afu_map->hosts[hwq->ctx_hndl].host;
1756 hwq->ctrl_map = &afu->afu_map->ctrls[hwq->ctx_hndl].ctrl;
1758 /* Program the Endian Control for the master context */
1759 writeq_be(SISL_ENDIAN_CTRL, &hwq->host_map->endian_ctrl);
1764 * init_global() - initialize AFU global registers
1765 * @cfg: Internal structure associated with the host.
1767 static int init_global(struct cxlflash_cfg *cfg)
1769 struct afu *afu = cfg->afu;
1770 struct device *dev = &cfg->dev->dev;
1772 struct sisl_host_map __iomem *hmap;
1773 __be64 __iomem *fc_port_regs;
1774 u64 wwpn[MAX_FC_PORTS]; /* wwpn of AFU ports */
1775 int i = 0, num_ports = 0;
1781 rc = read_vpd(cfg, &wwpn[0]);
1783 dev_err(dev, "%s: could not read vpd rc=%d\n", __func__, rc);
1787 /* Set up RRQ and SQ in HWQ for master issued cmds */
1788 for (i = 0; i < afu->num_hwqs; i++) {
1789 hwq = get_hwq(afu, i);
1790 hmap = hwq->host_map;
1792 writeq_be((u64) hwq->hrrq_start, &hmap->rrq_start);
1793 writeq_be((u64) hwq->hrrq_end, &hmap->rrq_end);
1794 hwq->hrrq_online = true;
1796 if (afu_is_sq_cmd_mode(afu)) {
1797 writeq_be((u64)hwq->hsq_start, &hmap->sq_start);
1798 writeq_be((u64)hwq->hsq_end, &hmap->sq_end);
1802 /* AFU configuration */
1803 reg = readq_be(&afu->afu_map->global.regs.afu_config);
1804 reg |= SISL_AFUCONF_AR_ALL|SISL_AFUCONF_ENDIAN;
1805 /* enable all auto retry options and control endianness */
1806 /* leave others at default: */
1807 /* CTX_CAP write protected, mbox_r does not clear on read and */
1808 /* checker on if dual afu */
1809 writeq_be(reg, &afu->afu_map->global.regs.afu_config);
1811 /* Global port select: select either port */
1812 if (afu->internal_lun) {
1813 /* Only use port 0 */
1814 writeq_be(PORT0, &afu->afu_map->global.regs.afu_port_sel);
1817 writeq_be(PORT_MASK(cfg->num_fc_ports),
1818 &afu->afu_map->global.regs.afu_port_sel);
1819 num_ports = cfg->num_fc_ports;
1822 for (i = 0; i < num_ports; i++) {
1823 fc_port_regs = get_fc_port_regs(cfg, i);
1825 /* Unmask all errors (but they are still masked at AFU) */
1826 writeq_be(0, &fc_port_regs[FC_ERRMSK / 8]);
1827 /* Clear CRC error cnt & set a threshold */
1828 (void)readq_be(&fc_port_regs[FC_CNT_CRCERR / 8]);
1829 writeq_be(MC_CRC_THRESH, &fc_port_regs[FC_CRC_THRESH / 8]);
1831 /* Set WWPNs. If already programmed, wwpn[i] is 0 */
1833 afu_set_wwpn(afu, i, &fc_port_regs[0], wwpn[i]);
1834 /* Programming WWPN back to back causes additional
1835 * offline/online transitions and a PLOGI
1840 if (afu_is_ocxl_lisn(afu)) {
1841 /* Set up the LISN effective address for each master */
1842 for (i = 0; i < afu->num_hwqs; i++) {
1843 hwq = get_hwq(afu, i);
1844 ctx = hwq->ctx_cookie;
1846 for (j = 0; j < hwq->num_irqs; j++) {
1847 reg = cfg->ops->get_irq_objhndl(ctx, j);
1848 writeq_be(reg, &hwq->ctrl_map->lisn_ea[j]);
1851 reg = hwq->ctx_hndl;
1852 writeq_be(SISL_LISN_PASID(reg, reg),
1853 &hwq->ctrl_map->lisn_pasid[0]);
1854 writeq_be(SISL_LISN_PASID(0UL, reg),
1855 &hwq->ctrl_map->lisn_pasid[1]);
1859 /* Set up master's own CTX_CAP to allow real mode, host translation */
1860 /* tables, afu cmds and read/write GSCSI cmds. */
1861 /* First, unlock ctx_cap write by reading mbox */
1862 for (i = 0; i < afu->num_hwqs; i++) {
1863 hwq = get_hwq(afu, i);
1865 (void)readq_be(&hwq->ctrl_map->mbox_r); /* unlock ctx_cap */
1866 writeq_be((SISL_CTX_CAP_REAL_MODE | SISL_CTX_CAP_HOST_XLATE |
1867 SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD |
1868 SISL_CTX_CAP_AFU_CMD | SISL_CTX_CAP_GSCSI_CMD),
1869 &hwq->ctrl_map->ctx_cap);
1873 * Determine write-same unmap support for host by evaluating the unmap
1874 * sector support bit of the context control register associated with
1875 * the primary hardware queue. Note that while this status is reflected
1876 * in a context register, the outcome can be assumed to be host-wide.
1878 hwq = get_hwq(afu, PRIMARY_HWQ);
1879 reg = readq_be(&hwq->host_map->ctx_ctrl);
1880 if (reg & SISL_CTX_CTRL_UNMAP_SECTOR)
1881 cfg->ws_unmap = true;
1883 /* Initialize heartbeat */
1884 afu->hb = readq_be(&afu->afu_map->global.regs.afu_hb);
1890 * start_afu() - initializes and starts the AFU
1891 * @cfg: Internal structure associated with the host.
1893 static int start_afu(struct cxlflash_cfg *cfg)
1895 struct afu *afu = cfg->afu;
1896 struct device *dev = &cfg->dev->dev;
1903 /* Initialize each HWQ */
1904 for (i = 0; i < afu->num_hwqs; i++) {
1905 hwq = get_hwq(afu, i);
1907 /* After an AFU reset, RRQ entries are stale, clear them */
1908 memset(&hwq->rrq_entry, 0, sizeof(hwq->rrq_entry));
1910 /* Initialize RRQ pointers */
1911 hwq->hrrq_start = &hwq->rrq_entry[0];
1912 hwq->hrrq_end = &hwq->rrq_entry[NUM_RRQ_ENTRY - 1];
1913 hwq->hrrq_curr = hwq->hrrq_start;
1916 /* Initialize spin locks */
1917 spin_lock_init(&hwq->hrrq_slock);
1918 spin_lock_init(&hwq->hsq_slock);
1921 if (afu_is_sq_cmd_mode(afu)) {
1922 memset(&hwq->sq, 0, sizeof(hwq->sq));
1923 hwq->hsq_start = &hwq->sq[0];
1924 hwq->hsq_end = &hwq->sq[NUM_SQ_ENTRY - 1];
1925 hwq->hsq_curr = hwq->hsq_start;
1927 atomic_set(&hwq->hsq_credits, NUM_SQ_ENTRY - 1);
1930 /* Initialize IRQ poll */
1931 if (afu_is_irqpoll_enabled(afu))
1932 irq_poll_init(&hwq->irqpoll, afu->irqpoll_weight,
1937 rc = init_global(cfg);
1939 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1944 * init_intr() - setup interrupt handlers for the master context
1945 * @cfg: Internal structure associated with the host.
1946 * @hwq: Hardware queue to initialize.
1948 * Return: 0 on success, -errno on failure
1950 static enum undo_level init_intr(struct cxlflash_cfg *cfg,
1953 struct device *dev = &cfg->dev->dev;
1954 void *ctx = hwq->ctx_cookie;
1956 enum undo_level level = UNDO_NOOP;
1957 bool is_primary_hwq = (hwq->index == PRIMARY_HWQ);
1958 int num_irqs = hwq->num_irqs;
1960 rc = cfg->ops->allocate_afu_irqs(ctx, num_irqs);
1962 dev_err(dev, "%s: allocate_afu_irqs failed rc=%d\n",
1968 rc = cfg->ops->map_afu_irq(ctx, 1, cxlflash_sync_err_irq, hwq,
1969 "SISL_MSI_SYNC_ERROR");
1970 if (unlikely(rc <= 0)) {
1971 dev_err(dev, "%s: SISL_MSI_SYNC_ERROR map failed\n", __func__);
1976 rc = cfg->ops->map_afu_irq(ctx, 2, cxlflash_rrq_irq, hwq,
1977 "SISL_MSI_RRQ_UPDATED");
1978 if (unlikely(rc <= 0)) {
1979 dev_err(dev, "%s: SISL_MSI_RRQ_UPDATED map failed\n", __func__);
1984 /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
1985 if (!is_primary_hwq)
1988 rc = cfg->ops->map_afu_irq(ctx, 3, cxlflash_async_err_irq, hwq,
1989 "SISL_MSI_ASYNC_ERROR");
1990 if (unlikely(rc <= 0)) {
1991 dev_err(dev, "%s: SISL_MSI_ASYNC_ERROR map failed\n", __func__);
2000 * init_mc() - create and register as the master context
2001 * @cfg: Internal structure associated with the host.
2002 * index: HWQ Index of the master context.
2004 * Return: 0 on success, -errno on failure
2006 static int init_mc(struct cxlflash_cfg *cfg, u32 index)
2009 struct device *dev = &cfg->dev->dev;
2010 struct hwq *hwq = get_hwq(cfg->afu, index);
2013 enum undo_level level;
2015 hwq->afu = cfg->afu;
2017 INIT_LIST_HEAD(&hwq->pending_cmds);
2019 if (index == PRIMARY_HWQ) {
2020 ctx = cfg->ops->get_context(cfg->dev, cfg->afu_cookie);
2023 ctx = cfg->ops->dev_context_init(cfg->dev, cfg->afu_cookie);
2026 if (IS_ERR_OR_NULL(ctx)) {
2031 WARN_ON(hwq->ctx_cookie);
2032 hwq->ctx_cookie = ctx;
2033 hwq->num_irqs = num_irqs;
2035 /* Set it up as a master with the CXL */
2036 cfg->ops->set_master(ctx);
2038 /* Reset AFU when initializing primary context */
2039 if (index == PRIMARY_HWQ) {
2040 rc = cfg->ops->afu_reset(ctx);
2042 dev_err(dev, "%s: AFU reset failed rc=%d\n",
2048 level = init_intr(cfg, hwq);
2049 if (unlikely(level)) {
2050 dev_err(dev, "%s: interrupt init failed rc=%d\n", __func__, rc);
2054 /* Finally, activate the context by starting it */
2055 rc = cfg->ops->start_context(hwq->ctx_cookie);
2057 dev_err(dev, "%s: start context failed rc=%d\n", __func__, rc);
2058 level = UNMAP_THREE;
2063 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2066 term_intr(cfg, level, index);
2067 if (index != PRIMARY_HWQ)
2068 cfg->ops->release_context(ctx);
2070 hwq->ctx_cookie = NULL;
2075 * get_num_afu_ports() - determines and configures the number of AFU ports
2076 * @cfg: Internal structure associated with the host.
2078 * This routine determines the number of AFU ports by converting the global
2079 * port selection mask. The converted value is only valid following an AFU
2080 * reset (explicit or power-on). This routine must be invoked shortly after
2081 * mapping as other routines are dependent on the number of ports during the
2082 * initialization sequence.
2084 * To support legacy AFUs that might not have reflected an initial global
2085 * port mask (value read is 0), default to the number of ports originally
2086 * supported by the cxlflash driver (2) before hardware with other port
2087 * offerings was introduced.
2089 static void get_num_afu_ports(struct cxlflash_cfg *cfg)
2091 struct afu *afu = cfg->afu;
2092 struct device *dev = &cfg->dev->dev;
2094 int num_fc_ports = LEGACY_FC_PORTS;
2096 port_mask = readq_be(&afu->afu_map->global.regs.afu_port_sel);
2097 if (port_mask != 0ULL)
2098 num_fc_ports = min(ilog2(port_mask) + 1, MAX_FC_PORTS);
2100 dev_dbg(dev, "%s: port_mask=%016llx num_fc_ports=%d\n",
2101 __func__, port_mask, num_fc_ports);
2103 cfg->num_fc_ports = num_fc_ports;
2104 cfg->host->max_channel = PORTNUM2CHAN(num_fc_ports);
2108 * init_afu() - setup as master context and start AFU
2109 * @cfg: Internal structure associated with the host.
2111 * This routine is a higher level of control for configuring the
2112 * AFU on probe and reset paths.
2114 * Return: 0 on success, -errno on failure
2116 static int init_afu(struct cxlflash_cfg *cfg)
2120 struct afu *afu = cfg->afu;
2121 struct device *dev = &cfg->dev->dev;
2125 cfg->ops->perst_reloads_same_image(cfg->afu_cookie, true);
2127 mutex_init(&afu->sync_active);
2128 afu->num_hwqs = afu->desired_hwqs;
2129 for (i = 0; i < afu->num_hwqs; i++) {
2130 rc = init_mc(cfg, i);
2132 dev_err(dev, "%s: init_mc failed rc=%d index=%d\n",
2138 /* Map the entire MMIO space of the AFU using the first context */
2139 hwq = get_hwq(afu, PRIMARY_HWQ);
2140 afu->afu_map = cfg->ops->psa_map(hwq->ctx_cookie);
2141 if (!afu->afu_map) {
2142 dev_err(dev, "%s: psa_map failed\n", __func__);
2147 /* No byte reverse on reading afu_version or string will be backwards */
2148 reg = readq(&afu->afu_map->global.regs.afu_version);
2149 memcpy(afu->version, ®, sizeof(reg));
2150 afu->interface_version =
2151 readq_be(&afu->afu_map->global.regs.interface_version);
2152 if ((afu->interface_version + 1) == 0) {
2153 dev_err(dev, "Back level AFU, please upgrade. AFU version %s "
2154 "interface version %016llx\n", afu->version,
2155 afu->interface_version);
2160 if (afu_is_sq_cmd_mode(afu)) {
2161 afu->send_cmd = send_cmd_sq;
2162 afu->context_reset = context_reset_sq;
2164 afu->send_cmd = send_cmd_ioarrin;
2165 afu->context_reset = context_reset_ioarrin;
2168 dev_dbg(dev, "%s: afu_ver=%s interface_ver=%016llx\n", __func__,
2169 afu->version, afu->interface_version);
2171 get_num_afu_ports(cfg);
2173 rc = start_afu(cfg);
2175 dev_err(dev, "%s: start_afu failed, rc=%d\n", __func__, rc);
2179 afu_err_intr_init(cfg->afu);
2180 for (i = 0; i < afu->num_hwqs; i++) {
2181 hwq = get_hwq(afu, i);
2183 hwq->room = readq_be(&hwq->host_map->cmd_room);
2186 /* Restore the LUN mappings */
2187 cxlflash_restore_luntable(cfg);
2189 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2193 for (i = afu->num_hwqs - 1; i >= 0; i--) {
2194 term_intr(cfg, UNMAP_THREE, i);
2201 * afu_reset() - resets the AFU
2202 * @cfg: Internal structure associated with the host.
2204 * Return: 0 on success, -errno on failure
2206 static int afu_reset(struct cxlflash_cfg *cfg)
2208 struct device *dev = &cfg->dev->dev;
2211 /* Stop the context before the reset. Since the context is
2212 * no longer available restart it after the reset is complete
2218 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2223 * drain_ioctls() - wait until all currently executing ioctls have completed
2224 * @cfg: Internal structure associated with the host.
2226 * Obtain write access to read/write semaphore that wraps ioctl
2227 * handling to 'drain' ioctls currently executing.
2229 static void drain_ioctls(struct cxlflash_cfg *cfg)
2231 down_write(&cfg->ioctl_rwsem);
2232 up_write(&cfg->ioctl_rwsem);
2236 * cxlflash_async_reset_host() - asynchronous host reset handler
2237 * @data: Private data provided while scheduling reset.
2238 * @cookie: Cookie that can be used for checkpointing.
2240 static void cxlflash_async_reset_host(void *data, async_cookie_t cookie)
2242 struct cxlflash_cfg *cfg = data;
2243 struct device *dev = &cfg->dev->dev;
2246 if (cfg->state != STATE_RESET) {
2247 dev_dbg(dev, "%s: Not performing a reset, state=%d\n",
2248 __func__, cfg->state);
2253 cxlflash_mark_contexts_error(cfg);
2254 rc = afu_reset(cfg);
2256 cfg->state = STATE_FAILTERM;
2258 cfg->state = STATE_NORMAL;
2259 wake_up_all(&cfg->reset_waitq);
2262 scsi_unblock_requests(cfg->host);
2266 * cxlflash_schedule_async_reset() - schedule an asynchronous host reset
2267 * @cfg: Internal structure associated with the host.
2269 static void cxlflash_schedule_async_reset(struct cxlflash_cfg *cfg)
2271 struct device *dev = &cfg->dev->dev;
2273 if (cfg->state != STATE_NORMAL) {
2274 dev_dbg(dev, "%s: Not performing reset state=%d\n",
2275 __func__, cfg->state);
2279 cfg->state = STATE_RESET;
2280 scsi_block_requests(cfg->host);
2281 cfg->async_reset_cookie = async_schedule(cxlflash_async_reset_host,
2286 * send_afu_cmd() - builds and sends an internal AFU command
2287 * @afu: AFU associated with the host.
2288 * @rcb: Pre-populated IOARCB describing command to send.
2290 * The AFU can only take one internal AFU command at a time. This limitation is
2291 * enforced by using a mutex to provide exclusive access to the AFU during the
2292 * operation. This design point requires calling threads to not be on interrupt
2293 * context due to the possibility of sleeping during concurrent AFU operations.
2295 * The command status is optionally passed back to the caller when the caller
2296 * populates the IOASA field of the IOARCB with a pointer to an IOASA structure.
2299 * 0 on success, -errno on failure
2301 static int send_afu_cmd(struct afu *afu, struct sisl_ioarcb *rcb)
2303 struct cxlflash_cfg *cfg = afu->parent;
2304 struct device *dev = &cfg->dev->dev;
2305 struct afu_cmd *cmd = NULL;
2306 struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
2312 if (cfg->state != STATE_NORMAL) {
2313 dev_dbg(dev, "%s: Sync not required state=%u\n",
2314 __func__, cfg->state);
2318 mutex_lock(&afu->sync_active);
2319 atomic_inc(&afu->cmds_active);
2320 buf = kmalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
2321 if (unlikely(!buf)) {
2322 dev_err(dev, "%s: no memory for command\n", __func__);
2327 cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
2330 memset(cmd, 0, sizeof(*cmd));
2331 memcpy(&cmd->rcb, rcb, sizeof(*rcb));
2332 INIT_LIST_HEAD(&cmd->queue);
2333 init_completion(&cmd->cevent);
2335 cmd->hwq_index = hwq->index;
2336 cmd->rcb.ctx_id = hwq->ctx_hndl;
2338 dev_dbg(dev, "%s: afu=%p cmd=%p type=%02x nretry=%d\n",
2339 __func__, afu, cmd, cmd->rcb.cdb[0], nretry);
2341 rc = afu->send_cmd(afu, cmd);
2347 rc = wait_resp(afu, cmd);
2350 rc = afu->context_reset(hwq);
2352 /* Delete the command from pending_cmds list */
2353 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
2354 list_del(&cmd->list);
2355 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
2357 cxlflash_schedule_async_reset(cfg);
2360 /* fall through to retry */
2364 /* fall through to exit */
2370 *rcb->ioasa = cmd->sa;
2372 atomic_dec(&afu->cmds_active);
2373 mutex_unlock(&afu->sync_active);
2375 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2380 * cxlflash_afu_sync() - builds and sends an AFU sync command
2381 * @afu: AFU associated with the host.
2382 * @ctx: Identifies context requesting sync.
2383 * @res: Identifies resource requesting sync.
2384 * @mode: Type of sync to issue (lightweight, heavyweight, global).
2386 * AFU sync operations are only necessary and allowed when the device is
2387 * operating normally. When not operating normally, sync requests can occur as
2388 * part of cleaning up resources associated with an adapter prior to removal.
2389 * In this scenario, these requests are simply ignored (safe due to the AFU
2393 * 0 on success, -errno on failure
2395 int cxlflash_afu_sync(struct afu *afu, ctx_hndl_t ctx, res_hndl_t res, u8 mode)
2397 struct cxlflash_cfg *cfg = afu->parent;
2398 struct device *dev = &cfg->dev->dev;
2399 struct sisl_ioarcb rcb = { 0 };
2401 dev_dbg(dev, "%s: afu=%p ctx=%u res=%u mode=%u\n",
2402 __func__, afu, ctx, res, mode);
2404 rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
2405 rcb.msi = SISL_MSI_RRQ_UPDATED;
2406 rcb.timeout = MC_AFU_SYNC_TIMEOUT;
2408 rcb.cdb[0] = SISL_AFU_CMD_SYNC;
2410 put_unaligned_be16(ctx, &rcb.cdb[2]);
2411 put_unaligned_be32(res, &rcb.cdb[4]);
2413 return send_afu_cmd(afu, &rcb);
2417 * cxlflash_eh_abort_handler() - abort a SCSI command
2418 * @scp: SCSI command to abort.
2420 * CXL Flash devices do not support a single command abort. Reset the context
2421 * as per SISLite specification. Flush any pending commands in the hardware
2422 * queue before the reset.
2424 * Return: SUCCESS/FAILED as defined in scsi/scsi.h
2426 static int cxlflash_eh_abort_handler(struct scsi_cmnd *scp)
2429 struct Scsi_Host *host = scp->device->host;
2430 struct cxlflash_cfg *cfg = shost_priv(host);
2431 struct afu_cmd *cmd = sc_to_afuc(scp);
2432 struct device *dev = &cfg->dev->dev;
2433 struct afu *afu = cfg->afu;
2434 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
2436 dev_dbg(dev, "%s: (scp=%p) %d/%d/%d/%llu "
2437 "cdb=(%08x-%08x-%08x-%08x)\n", __func__, scp, host->host_no,
2438 scp->device->channel, scp->device->id, scp->device->lun,
2439 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
2440 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
2441 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
2442 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
2444 /* When the state is not normal, another reset/reload is in progress.
2445 * Return failed and the mid-layer will invoke host reset handler.
2447 if (cfg->state != STATE_NORMAL) {
2448 dev_dbg(dev, "%s: Invalid state for abort, state=%d\n",
2449 __func__, cfg->state);
2453 rc = afu->context_reset(hwq);
2460 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2465 * cxlflash_eh_device_reset_handler() - reset a single LUN
2466 * @scp: SCSI command to send.
2469 * SUCCESS as defined in scsi/scsi.h
2470 * FAILED as defined in scsi/scsi.h
2472 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
2475 struct scsi_device *sdev = scp->device;
2476 struct Scsi_Host *host = sdev->host;
2477 struct cxlflash_cfg *cfg = shost_priv(host);
2478 struct device *dev = &cfg->dev->dev;
2481 dev_dbg(dev, "%s: %d/%d/%d/%llu\n", __func__,
2482 host->host_no, sdev->channel, sdev->id, sdev->lun);
2484 switch (cfg->state) {
2486 rcr = send_tmf(cfg, sdev, TMF_LUN_RESET);
2491 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2498 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2503 * cxlflash_eh_host_reset_handler() - reset the host adapter
2504 * @scp: SCSI command from stack identifying host.
2506 * Following a reset, the state is evaluated again in case an EEH occurred
2507 * during the reset. In such a scenario, the host reset will either yield
2508 * until the EEH recovery is complete or return success or failure based
2509 * upon the current device state.
2512 * SUCCESS as defined in scsi/scsi.h
2513 * FAILED as defined in scsi/scsi.h
2515 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
2519 struct Scsi_Host *host = scp->device->host;
2520 struct cxlflash_cfg *cfg = shost_priv(host);
2521 struct device *dev = &cfg->dev->dev;
2523 dev_dbg(dev, "%s: %d\n", __func__, host->host_no);
2525 switch (cfg->state) {
2527 cfg->state = STATE_RESET;
2529 cxlflash_mark_contexts_error(cfg);
2530 rcr = afu_reset(cfg);
2533 cfg->state = STATE_FAILTERM;
2535 cfg->state = STATE_NORMAL;
2536 wake_up_all(&cfg->reset_waitq);
2540 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2541 if (cfg->state == STATE_NORMAL)
2549 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2554 * cxlflash_change_queue_depth() - change the queue depth for the device
2555 * @sdev: SCSI device destined for queue depth change.
2556 * @qdepth: Requested queue depth value to set.
2558 * The requested queue depth is capped to the maximum supported value.
2560 * Return: The actual queue depth set.
2562 static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
2565 if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
2566 qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
2568 scsi_change_queue_depth(sdev, qdepth);
2569 return sdev->queue_depth;
2573 * cxlflash_show_port_status() - queries and presents the current port status
2574 * @port: Desired port for status reporting.
2575 * @cfg: Internal structure associated with the host.
2576 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2578 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2580 static ssize_t cxlflash_show_port_status(u32 port,
2581 struct cxlflash_cfg *cfg,
2584 struct device *dev = &cfg->dev->dev;
2587 __be64 __iomem *fc_port_regs;
2589 WARN_ON(port >= MAX_FC_PORTS);
2591 if (port >= cfg->num_fc_ports) {
2592 dev_info(dev, "%s: Port %d not supported on this card.\n",
2597 fc_port_regs = get_fc_port_regs(cfg, port);
2598 status = readq_be(&fc_port_regs[FC_MTIP_STATUS / 8]);
2599 status &= FC_MTIP_STATUS_MASK;
2601 if (status == FC_MTIP_STATUS_ONLINE)
2602 disp_status = "online";
2603 else if (status == FC_MTIP_STATUS_OFFLINE)
2604 disp_status = "offline";
2606 disp_status = "unknown";
2608 return scnprintf(buf, PAGE_SIZE, "%s\n", disp_status);
2612 * port0_show() - queries and presents the current status of port 0
2613 * @dev: Generic device associated with the host owning the port.
2614 * @attr: Device attribute representing the port.
2615 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2617 * Return: The size of the ASCII string returned in @buf.
2619 static ssize_t port0_show(struct device *dev,
2620 struct device_attribute *attr,
2623 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2625 return cxlflash_show_port_status(0, cfg, buf);
2629 * port1_show() - queries and presents the current status of port 1
2630 * @dev: Generic device associated with the host owning the port.
2631 * @attr: Device attribute representing the port.
2632 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2634 * Return: The size of the ASCII string returned in @buf.
2636 static ssize_t port1_show(struct device *dev,
2637 struct device_attribute *attr,
2640 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2642 return cxlflash_show_port_status(1, cfg, buf);
2646 * port2_show() - queries and presents the current status of port 2
2647 * @dev: Generic device associated with the host owning the port.
2648 * @attr: Device attribute representing the port.
2649 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2651 * Return: The size of the ASCII string returned in @buf.
2653 static ssize_t port2_show(struct device *dev,
2654 struct device_attribute *attr,
2657 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2659 return cxlflash_show_port_status(2, cfg, buf);
2663 * port3_show() - queries and presents the current status of port 3
2664 * @dev: Generic device associated with the host owning the port.
2665 * @attr: Device attribute representing the port.
2666 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2668 * Return: The size of the ASCII string returned in @buf.
2670 static ssize_t port3_show(struct device *dev,
2671 struct device_attribute *attr,
2674 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2676 return cxlflash_show_port_status(3, cfg, buf);
2680 * lun_mode_show() - presents the current LUN mode of the host
2681 * @dev: Generic device associated with the host.
2682 * @attr: Device attribute representing the LUN mode.
2683 * @buf: Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2685 * Return: The size of the ASCII string returned in @buf.
2687 static ssize_t lun_mode_show(struct device *dev,
2688 struct device_attribute *attr, char *buf)
2690 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2691 struct afu *afu = cfg->afu;
2693 return scnprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
2697 * lun_mode_store() - sets the LUN mode of the host
2698 * @dev: Generic device associated with the host.
2699 * @attr: Device attribute representing the LUN mode.
2700 * @buf: Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2701 * @count: Length of data resizing in @buf.
2703 * The CXL Flash AFU supports a dummy LUN mode where the external
2704 * links and storage are not required. Space on the FPGA is used
2705 * to create 1 or 2 small LUNs which are presented to the system
2706 * as if they were a normal storage device. This feature is useful
2707 * during development and also provides manufacturing with a way
2708 * to test the AFU without an actual device.
2710 * 0 = external LUN[s] (default)
2711 * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2712 * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2713 * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2714 * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2716 * Return: The size of the ASCII string returned in @buf.
2718 static ssize_t lun_mode_store(struct device *dev,
2719 struct device_attribute *attr,
2720 const char *buf, size_t count)
2722 struct Scsi_Host *shost = class_to_shost(dev);
2723 struct cxlflash_cfg *cfg = shost_priv(shost);
2724 struct afu *afu = cfg->afu;
2728 rc = kstrtouint(buf, 10, &lun_mode);
2729 if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
2730 afu->internal_lun = lun_mode;
2733 * When configured for internal LUN, there is only one channel,
2734 * channel number 0, else there will be one less than the number
2735 * of fc ports for this card.
2737 if (afu->internal_lun)
2738 shost->max_channel = 0;
2740 shost->max_channel = PORTNUM2CHAN(cfg->num_fc_ports);
2743 scsi_scan_host(cfg->host);
2750 * ioctl_version_show() - presents the current ioctl version of the host
2751 * @dev: Generic device associated with the host.
2752 * @attr: Device attribute representing the ioctl version.
2753 * @buf: Buffer of length PAGE_SIZE to report back the ioctl version.
2755 * Return: The size of the ASCII string returned in @buf.
2757 static ssize_t ioctl_version_show(struct device *dev,
2758 struct device_attribute *attr, char *buf)
2762 bytes = scnprintf(buf, PAGE_SIZE,
2763 "disk: %u\n", DK_CXLFLASH_VERSION_0);
2764 bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2765 "host: %u\n", HT_CXLFLASH_VERSION_0);
2771 * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2772 * @port: Desired port for status reporting.
2773 * @cfg: Internal structure associated with the host.
2774 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2776 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2778 static ssize_t cxlflash_show_port_lun_table(u32 port,
2779 struct cxlflash_cfg *cfg,
2782 struct device *dev = &cfg->dev->dev;
2783 __be64 __iomem *fc_port_luns;
2787 WARN_ON(port >= MAX_FC_PORTS);
2789 if (port >= cfg->num_fc_ports) {
2790 dev_info(dev, "%s: Port %d not supported on this card.\n",
2795 fc_port_luns = get_fc_port_luns(cfg, port);
2797 for (i = 0; i < CXLFLASH_NUM_VLUNS; i++)
2798 bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2800 i, readq_be(&fc_port_luns[i]));
2805 * port0_lun_table_show() - presents the current LUN table of port 0
2806 * @dev: Generic device associated with the host owning the port.
2807 * @attr: Device attribute representing the port.
2808 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2810 * Return: The size of the ASCII string returned in @buf.
2812 static ssize_t port0_lun_table_show(struct device *dev,
2813 struct device_attribute *attr,
2816 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2818 return cxlflash_show_port_lun_table(0, cfg, buf);
2822 * port1_lun_table_show() - presents the current LUN table of port 1
2823 * @dev: Generic device associated with the host owning the port.
2824 * @attr: Device attribute representing the port.
2825 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2827 * Return: The size of the ASCII string returned in @buf.
2829 static ssize_t port1_lun_table_show(struct device *dev,
2830 struct device_attribute *attr,
2833 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2835 return cxlflash_show_port_lun_table(1, cfg, buf);
2839 * port2_lun_table_show() - presents the current LUN table of port 2
2840 * @dev: Generic device associated with the host owning the port.
2841 * @attr: Device attribute representing the port.
2842 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2844 * Return: The size of the ASCII string returned in @buf.
2846 static ssize_t port2_lun_table_show(struct device *dev,
2847 struct device_attribute *attr,
2850 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2852 return cxlflash_show_port_lun_table(2, cfg, buf);
2856 * port3_lun_table_show() - presents the current LUN table of port 3
2857 * @dev: Generic device associated with the host owning the port.
2858 * @attr: Device attribute representing the port.
2859 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2861 * Return: The size of the ASCII string returned in @buf.
2863 static ssize_t port3_lun_table_show(struct device *dev,
2864 struct device_attribute *attr,
2867 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2869 return cxlflash_show_port_lun_table(3, cfg, buf);
2873 * irqpoll_weight_show() - presents the current IRQ poll weight for the host
2874 * @dev: Generic device associated with the host.
2875 * @attr: Device attribute representing the IRQ poll weight.
2876 * @buf: Buffer of length PAGE_SIZE to report back the current IRQ poll
2879 * An IRQ poll weight of 0 indicates polling is disabled.
2881 * Return: The size of the ASCII string returned in @buf.
2883 static ssize_t irqpoll_weight_show(struct device *dev,
2884 struct device_attribute *attr, char *buf)
2886 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2887 struct afu *afu = cfg->afu;
2889 return scnprintf(buf, PAGE_SIZE, "%u\n", afu->irqpoll_weight);
2893 * irqpoll_weight_store() - sets the current IRQ poll weight for the host
2894 * @dev: Generic device associated with the host.
2895 * @attr: Device attribute representing the IRQ poll weight.
2896 * @buf: Buffer of length PAGE_SIZE containing the desired IRQ poll
2898 * @count: Length of data resizing in @buf.
2900 * An IRQ poll weight of 0 indicates polling is disabled.
2902 * Return: The size of the ASCII string returned in @buf.
2904 static ssize_t irqpoll_weight_store(struct device *dev,
2905 struct device_attribute *attr,
2906 const char *buf, size_t count)
2908 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2909 struct device *cfgdev = &cfg->dev->dev;
2910 struct afu *afu = cfg->afu;
2915 rc = kstrtouint(buf, 10, &weight);
2921 "Invalid IRQ poll weight. It must be 256 or less.\n");
2925 if (weight == afu->irqpoll_weight) {
2927 "Current IRQ poll weight has the same weight.\n");
2931 if (afu_is_irqpoll_enabled(afu)) {
2932 for (i = 0; i < afu->num_hwqs; i++) {
2933 hwq = get_hwq(afu, i);
2935 irq_poll_disable(&hwq->irqpoll);
2939 afu->irqpoll_weight = weight;
2942 for (i = 0; i < afu->num_hwqs; i++) {
2943 hwq = get_hwq(afu, i);
2945 irq_poll_init(&hwq->irqpoll, weight, cxlflash_irqpoll);
2953 * num_hwqs_show() - presents the number of hardware queues for the host
2954 * @dev: Generic device associated with the host.
2955 * @attr: Device attribute representing the number of hardware queues.
2956 * @buf: Buffer of length PAGE_SIZE to report back the number of hardware
2959 * Return: The size of the ASCII string returned in @buf.
2961 static ssize_t num_hwqs_show(struct device *dev,
2962 struct device_attribute *attr, char *buf)
2964 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2965 struct afu *afu = cfg->afu;
2967 return scnprintf(buf, PAGE_SIZE, "%u\n", afu->num_hwqs);
2971 * num_hwqs_store() - sets the number of hardware queues for the host
2972 * @dev: Generic device associated with the host.
2973 * @attr: Device attribute representing the number of hardware queues.
2974 * @buf: Buffer of length PAGE_SIZE containing the number of hardware
2976 * @count: Length of data resizing in @buf.
2978 * n > 0: num_hwqs = n
2979 * n = 0: num_hwqs = num_online_cpus()
2980 * n < 0: num_online_cpus() / abs(n)
2982 * Return: The size of the ASCII string returned in @buf.
2984 static ssize_t num_hwqs_store(struct device *dev,
2985 struct device_attribute *attr,
2986 const char *buf, size_t count)
2988 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2989 struct afu *afu = cfg->afu;
2991 int nhwqs, num_hwqs;
2993 rc = kstrtoint(buf, 10, &nhwqs);
2999 else if (nhwqs == 0)
3000 num_hwqs = num_online_cpus();
3002 num_hwqs = num_online_cpus() / abs(nhwqs);
3004 afu->desired_hwqs = min(num_hwqs, CXLFLASH_MAX_HWQS);
3005 WARN_ON_ONCE(afu->desired_hwqs == 0);
3008 switch (cfg->state) {
3010 cfg->state = STATE_RESET;
3012 cxlflash_mark_contexts_error(cfg);
3013 rc = afu_reset(cfg);
3015 cfg->state = STATE_FAILTERM;
3017 cfg->state = STATE_NORMAL;
3018 wake_up_all(&cfg->reset_waitq);
3021 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
3022 if (cfg->state == STATE_NORMAL)
3025 /* Ideally should not happen */
3026 dev_err(dev, "%s: Device is not ready, state=%d\n",
3027 __func__, cfg->state);
3034 static const char *hwq_mode_name[MAX_HWQ_MODE] = { "rr", "tag", "cpu" };
3037 * hwq_mode_show() - presents the HWQ steering mode for the host
3038 * @dev: Generic device associated with the host.
3039 * @attr: Device attribute representing the HWQ steering mode.
3040 * @buf: Buffer of length PAGE_SIZE to report back the HWQ steering mode
3041 * as a character string.
3043 * Return: The size of the ASCII string returned in @buf.
3045 static ssize_t hwq_mode_show(struct device *dev,
3046 struct device_attribute *attr, char *buf)
3048 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
3049 struct afu *afu = cfg->afu;
3051 return scnprintf(buf, PAGE_SIZE, "%s\n", hwq_mode_name[afu->hwq_mode]);
3055 * hwq_mode_store() - sets the HWQ steering mode for the host
3056 * @dev: Generic device associated with the host.
3057 * @attr: Device attribute representing the HWQ steering mode.
3058 * @buf: Buffer of length PAGE_SIZE containing the HWQ steering mode
3059 * as a character string.
3060 * @count: Length of data resizing in @buf.
3063 * tag = Block MQ Tagging
3064 * cpu = CPU Affinity
3066 * Return: The size of the ASCII string returned in @buf.
3068 static ssize_t hwq_mode_store(struct device *dev,
3069 struct device_attribute *attr,
3070 const char *buf, size_t count)
3072 struct Scsi_Host *shost = class_to_shost(dev);
3073 struct cxlflash_cfg *cfg = shost_priv(shost);
3074 struct device *cfgdev = &cfg->dev->dev;
3075 struct afu *afu = cfg->afu;
3077 u32 mode = MAX_HWQ_MODE;
3079 for (i = 0; i < MAX_HWQ_MODE; i++) {
3080 if (!strncmp(hwq_mode_name[i], buf, strlen(hwq_mode_name[i]))) {
3086 if (mode >= MAX_HWQ_MODE) {
3087 dev_info(cfgdev, "Invalid HWQ steering mode.\n");
3091 if ((mode == HWQ_MODE_TAG) && !shost_use_blk_mq(shost)) {
3092 dev_info(cfgdev, "SCSI-MQ is not enabled, use a different "
3093 "HWQ steering mode.\n");
3097 afu->hwq_mode = mode;
3103 * mode_show() - presents the current mode of the device
3104 * @dev: Generic device associated with the device.
3105 * @attr: Device attribute representing the device mode.
3106 * @buf: Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
3108 * Return: The size of the ASCII string returned in @buf.
3110 static ssize_t mode_show(struct device *dev,
3111 struct device_attribute *attr, char *buf)
3113 struct scsi_device *sdev = to_scsi_device(dev);
3115 return scnprintf(buf, PAGE_SIZE, "%s\n",
3116 sdev->hostdata ? "superpipe" : "legacy");
3122 static DEVICE_ATTR_RO(port0);
3123 static DEVICE_ATTR_RO(port1);
3124 static DEVICE_ATTR_RO(port2);
3125 static DEVICE_ATTR_RO(port3);
3126 static DEVICE_ATTR_RW(lun_mode);
3127 static DEVICE_ATTR_RO(ioctl_version);
3128 static DEVICE_ATTR_RO(port0_lun_table);
3129 static DEVICE_ATTR_RO(port1_lun_table);
3130 static DEVICE_ATTR_RO(port2_lun_table);
3131 static DEVICE_ATTR_RO(port3_lun_table);
3132 static DEVICE_ATTR_RW(irqpoll_weight);
3133 static DEVICE_ATTR_RW(num_hwqs);
3134 static DEVICE_ATTR_RW(hwq_mode);
3136 static struct device_attribute *cxlflash_host_attrs[] = {
3142 &dev_attr_ioctl_version,
3143 &dev_attr_port0_lun_table,
3144 &dev_attr_port1_lun_table,
3145 &dev_attr_port2_lun_table,
3146 &dev_attr_port3_lun_table,
3147 &dev_attr_irqpoll_weight,
3156 static DEVICE_ATTR_RO(mode);
3158 static struct device_attribute *cxlflash_dev_attrs[] = {
3166 static struct scsi_host_template driver_template = {
3167 .module = THIS_MODULE,
3168 .name = CXLFLASH_ADAPTER_NAME,
3169 .info = cxlflash_driver_info,
3170 .ioctl = cxlflash_ioctl,
3171 .proc_name = CXLFLASH_NAME,
3172 .queuecommand = cxlflash_queuecommand,
3173 .eh_abort_handler = cxlflash_eh_abort_handler,
3174 .eh_device_reset_handler = cxlflash_eh_device_reset_handler,
3175 .eh_host_reset_handler = cxlflash_eh_host_reset_handler,
3176 .change_queue_depth = cxlflash_change_queue_depth,
3177 .cmd_per_lun = CXLFLASH_MAX_CMDS_PER_LUN,
3178 .can_queue = CXLFLASH_MAX_CMDS,
3179 .cmd_size = sizeof(struct afu_cmd) + __alignof__(struct afu_cmd) - 1,
3181 .sg_tablesize = 1, /* No scatter gather support */
3182 .max_sectors = CXLFLASH_MAX_SECTORS,
3183 .use_clustering = ENABLE_CLUSTERING,
3184 .shost_attrs = cxlflash_host_attrs,
3185 .sdev_attrs = cxlflash_dev_attrs,
3189 * Device dependent values
3191 static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS,
3192 CXLFLASH_WWPN_VPD_REQUIRED };
3193 static struct dev_dependent_vals dev_flash_gt_vals = { CXLFLASH_MAX_SECTORS,
3194 CXLFLASH_NOTIFY_SHUTDOWN };
3195 static struct dev_dependent_vals dev_briard_vals = { CXLFLASH_MAX_SECTORS,
3196 (CXLFLASH_NOTIFY_SHUTDOWN |
3197 CXLFLASH_OCXL_DEV) };
3200 * PCI device binding table
3202 static struct pci_device_id cxlflash_pci_table[] = {
3203 {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
3204 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
3205 {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_FLASH_GT,
3206 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_flash_gt_vals},
3207 {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_BRIARD,
3208 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_briard_vals},
3212 MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
3215 * cxlflash_worker_thread() - work thread handler for the AFU
3216 * @work: Work structure contained within cxlflash associated with host.
3218 * Handles the following events:
3219 * - Link reset which cannot be performed on interrupt context due to
3220 * blocking up to a few seconds
3223 static void cxlflash_worker_thread(struct work_struct *work)
3225 struct cxlflash_cfg *cfg = container_of(work, struct cxlflash_cfg,
3227 struct afu *afu = cfg->afu;
3228 struct device *dev = &cfg->dev->dev;
3229 __be64 __iomem *fc_port_regs;
3233 /* Avoid MMIO if the device has failed */
3235 if (cfg->state != STATE_NORMAL)
3238 spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3240 if (cfg->lr_state == LINK_RESET_REQUIRED) {
3241 port = cfg->lr_port;
3243 dev_err(dev, "%s: invalid port index %d\n",
3246 spin_unlock_irqrestore(cfg->host->host_lock,
3249 /* The reset can block... */
3250 fc_port_regs = get_fc_port_regs(cfg, port);
3251 afu_link_reset(afu, port, fc_port_regs);
3252 spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3255 cfg->lr_state = LINK_RESET_COMPLETE;
3258 spin_unlock_irqrestore(cfg->host->host_lock, lock_flags);
3260 if (atomic_dec_if_positive(&cfg->scan_host_needed) >= 0)
3261 scsi_scan_host(cfg->host);
3265 * cxlflash_chr_open() - character device open handler
3266 * @inode: Device inode associated with this character device.
3267 * @file: File pointer for this device.
3269 * Only users with admin privileges are allowed to open the character device.
3271 * Return: 0 on success, -errno on failure
3273 static int cxlflash_chr_open(struct inode *inode, struct file *file)
3275 struct cxlflash_cfg *cfg;
3277 if (!capable(CAP_SYS_ADMIN))
3280 cfg = container_of(inode->i_cdev, struct cxlflash_cfg, cdev);
3281 file->private_data = cfg;
3287 * decode_hioctl() - translates encoded host ioctl to easily identifiable string
3288 * @cmd: The host ioctl command to decode.
3290 * Return: A string identifying the decoded host ioctl.
3292 static char *decode_hioctl(int cmd)
3295 case HT_CXLFLASH_LUN_PROVISION:
3296 return __stringify_1(HT_CXLFLASH_LUN_PROVISION);
3303 * cxlflash_lun_provision() - host LUN provisioning handler
3304 * @cfg: Internal structure associated with the host.
3305 * @arg: Kernel copy of userspace ioctl data structure.
3307 * Return: 0 on success, -errno on failure
3309 static int cxlflash_lun_provision(struct cxlflash_cfg *cfg,
3310 struct ht_cxlflash_lun_provision *lunprov)
3312 struct afu *afu = cfg->afu;
3313 struct device *dev = &cfg->dev->dev;
3314 struct sisl_ioarcb rcb;
3315 struct sisl_ioasa asa;
3316 __be64 __iomem *fc_port_regs;
3317 u16 port = lunprov->port;
3318 u16 scmd = lunprov->hdr.subcmd;
3325 if (!afu_is_lun_provision(afu)) {
3330 if (port >= cfg->num_fc_ports) {
3336 case HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN:
3337 type = SISL_AFU_LUN_PROVISION_CREATE;
3338 size = lunprov->size;
3341 case HT_CXLFLASH_LUN_PROVISION_SUBCMD_DELETE_LUN:
3342 type = SISL_AFU_LUN_PROVISION_DELETE;
3344 lun_id = lunprov->lun_id;
3346 case HT_CXLFLASH_LUN_PROVISION_SUBCMD_QUERY_PORT:
3347 fc_port_regs = get_fc_port_regs(cfg, port);
3349 reg = readq_be(&fc_port_regs[FC_MAX_NUM_LUNS / 8]);
3350 lunprov->max_num_luns = reg;
3351 reg = readq_be(&fc_port_regs[FC_CUR_NUM_LUNS / 8]);
3352 lunprov->cur_num_luns = reg;
3353 reg = readq_be(&fc_port_regs[FC_MAX_CAP_PORT / 8]);
3354 lunprov->max_cap_port = reg;
3355 reg = readq_be(&fc_port_regs[FC_CUR_CAP_PORT / 8]);
3356 lunprov->cur_cap_port = reg;
3364 memset(&rcb, 0, sizeof(rcb));
3365 memset(&asa, 0, sizeof(asa));
3366 rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
3367 rcb.lun_id = lun_id;
3368 rcb.msi = SISL_MSI_RRQ_UPDATED;
3369 rcb.timeout = MC_LUN_PROV_TIMEOUT;
3372 rcb.cdb[0] = SISL_AFU_CMD_LUN_PROVISION;
3375 put_unaligned_be64(size, &rcb.cdb[8]);
3377 rc = send_afu_cmd(afu, &rcb);
3379 dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3380 __func__, rc, asa.ioasc, asa.afu_extra);
3384 if (scmd == HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN) {
3385 lunprov->lun_id = (u64)asa.lunid_hi << 32 | asa.lunid_lo;
3386 memcpy(lunprov->wwid, asa.wwid, sizeof(lunprov->wwid));
3389 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3394 * cxlflash_afu_debug() - host AFU debug handler
3395 * @cfg: Internal structure associated with the host.
3396 * @arg: Kernel copy of userspace ioctl data structure.
3398 * For debug requests requiring a data buffer, always provide an aligned
3399 * (cache line) buffer to the AFU to appease any alignment requirements.
3401 * Return: 0 on success, -errno on failure
3403 static int cxlflash_afu_debug(struct cxlflash_cfg *cfg,
3404 struct ht_cxlflash_afu_debug *afu_dbg)
3406 struct afu *afu = cfg->afu;
3407 struct device *dev = &cfg->dev->dev;
3408 struct sisl_ioarcb rcb;
3409 struct sisl_ioasa asa;
3412 void __user *ubuf = (__force void __user *)afu_dbg->data_ea;
3413 u16 req_flags = SISL_REQ_FLAGS_AFU_CMD;
3414 u32 ulen = afu_dbg->data_len;
3415 bool is_write = afu_dbg->hdr.flags & HT_CXLFLASH_HOST_WRITE;
3418 if (!afu_is_afu_debug(afu)) {
3424 req_flags |= SISL_REQ_FLAGS_SUP_UNDERRUN;
3426 if (ulen > HT_CXLFLASH_AFU_DEBUG_MAX_DATA_LEN) {
3431 buf = kmalloc(ulen + cache_line_size() - 1, GFP_KERNEL);
3432 if (unlikely(!buf)) {
3437 kbuf = PTR_ALIGN(buf, cache_line_size());
3440 req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
3442 if (copy_from_user(kbuf, ubuf, ulen)) {
3449 memset(&rcb, 0, sizeof(rcb));
3450 memset(&asa, 0, sizeof(asa));
3452 rcb.req_flags = req_flags;
3453 rcb.msi = SISL_MSI_RRQ_UPDATED;
3454 rcb.timeout = MC_AFU_DEBUG_TIMEOUT;
3458 rcb.data_len = ulen;
3459 rcb.data_ea = (uintptr_t)kbuf;
3462 rcb.cdb[0] = SISL_AFU_CMD_DEBUG;
3463 memcpy(&rcb.cdb[4], afu_dbg->afu_subcmd,
3464 HT_CXLFLASH_AFU_DEBUG_SUBCMD_LEN);
3466 rc = send_afu_cmd(afu, &rcb);
3468 dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3469 __func__, rc, asa.ioasc, asa.afu_extra);
3473 if (ulen && !is_write) {
3474 if (copy_to_user(ubuf, kbuf, ulen))
3479 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3484 * cxlflash_chr_ioctl() - character device IOCTL handler
3485 * @file: File pointer for this device.
3486 * @cmd: IOCTL command.
3487 * @arg: Userspace ioctl data structure.
3489 * A read/write semaphore is used to implement a 'drain' of currently
3490 * running ioctls. The read semaphore is taken at the beginning of each
3491 * ioctl thread and released upon concluding execution. Additionally the
3492 * semaphore should be released and then reacquired in any ioctl execution
3493 * path which will wait for an event to occur that is outside the scope of
3494 * the ioctl (i.e. an adapter reset). To drain the ioctls currently running,
3495 * a thread simply needs to acquire the write semaphore.
3497 * Return: 0 on success, -errno on failure
3499 static long cxlflash_chr_ioctl(struct file *file, unsigned int cmd,
3502 typedef int (*hioctl) (struct cxlflash_cfg *, void *);
3504 struct cxlflash_cfg *cfg = file->private_data;
3505 struct device *dev = &cfg->dev->dev;
3506 char buf[sizeof(union cxlflash_ht_ioctls)];
3507 void __user *uarg = (void __user *)arg;
3508 struct ht_cxlflash_hdr *hdr;
3510 bool known_ioctl = false;
3513 hioctl do_ioctl = NULL;
3515 static const struct {
3518 } ioctl_tbl[] = { /* NOTE: order matters here */
3519 { sizeof(struct ht_cxlflash_lun_provision),
3520 (hioctl)cxlflash_lun_provision },
3521 { sizeof(struct ht_cxlflash_afu_debug),
3522 (hioctl)cxlflash_afu_debug },
3525 /* Hold read semaphore so we can drain if needed */
3526 down_read(&cfg->ioctl_rwsem);
3528 dev_dbg(dev, "%s: cmd=%u idx=%d tbl_size=%lu\n",
3529 __func__, cmd, idx, sizeof(ioctl_tbl));
3532 case HT_CXLFLASH_LUN_PROVISION:
3533 case HT_CXLFLASH_AFU_DEBUG:
3535 idx = _IOC_NR(HT_CXLFLASH_LUN_PROVISION) - _IOC_NR(cmd);
3536 size = ioctl_tbl[idx].size;
3537 do_ioctl = ioctl_tbl[idx].ioctl;
3539 if (likely(do_ioctl))
3548 if (unlikely(copy_from_user(&buf, uarg, size))) {
3549 dev_err(dev, "%s: copy_from_user() fail "
3550 "size=%lu cmd=%d (%s) uarg=%p\n",
3551 __func__, size, cmd, decode_hioctl(cmd), uarg);
3556 hdr = (struct ht_cxlflash_hdr *)&buf;
3557 if (hdr->version != HT_CXLFLASH_VERSION_0) {
3558 dev_dbg(dev, "%s: Version %u not supported for %s\n",
3559 __func__, hdr->version, decode_hioctl(cmd));
3564 if (hdr->rsvd[0] || hdr->rsvd[1] || hdr->return_flags) {
3565 dev_dbg(dev, "%s: Reserved/rflags populated\n", __func__);
3570 rc = do_ioctl(cfg, (void *)&buf);
3572 if (unlikely(copy_to_user(uarg, &buf, size))) {
3573 dev_err(dev, "%s: copy_to_user() fail "
3574 "size=%lu cmd=%d (%s) uarg=%p\n",
3575 __func__, size, cmd, decode_hioctl(cmd), uarg);
3579 /* fall through to exit */
3582 up_read(&cfg->ioctl_rwsem);
3583 if (unlikely(rc && known_ioctl))
3584 dev_err(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3585 __func__, decode_hioctl(cmd), cmd, rc);
3587 dev_dbg(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3588 __func__, decode_hioctl(cmd), cmd, rc);
3593 * Character device file operations
3595 static const struct file_operations cxlflash_chr_fops = {
3596 .owner = THIS_MODULE,
3597 .open = cxlflash_chr_open,
3598 .unlocked_ioctl = cxlflash_chr_ioctl,
3599 .compat_ioctl = cxlflash_chr_ioctl,
3603 * init_chrdev() - initialize the character device for the host
3604 * @cfg: Internal structure associated with the host.
3606 * Return: 0 on success, -errno on failure
3608 static int init_chrdev(struct cxlflash_cfg *cfg)
3610 struct device *dev = &cfg->dev->dev;
3611 struct device *char_dev;
3616 minor = cxlflash_get_minor();
3617 if (unlikely(minor < 0)) {
3618 dev_err(dev, "%s: Exhausted allowed adapters\n", __func__);
3623 devno = MKDEV(cxlflash_major, minor);
3624 cdev_init(&cfg->cdev, &cxlflash_chr_fops);
3626 rc = cdev_add(&cfg->cdev, devno, 1);
3628 dev_err(dev, "%s: cdev_add failed rc=%d\n", __func__, rc);
3632 char_dev = device_create(cxlflash_class, NULL, devno,
3633 NULL, "cxlflash%d", minor);
3634 if (IS_ERR(char_dev)) {
3635 rc = PTR_ERR(char_dev);
3636 dev_err(dev, "%s: device_create failed rc=%d\n",
3641 cfg->chardev = char_dev;
3643 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3646 cdev_del(&cfg->cdev);
3648 cxlflash_put_minor(minor);
3653 * cxlflash_probe() - PCI entry point to add host
3654 * @pdev: PCI device associated with the host.
3655 * @dev_id: PCI device id associated with device.
3657 * The device will initially start out in a 'probing' state and
3658 * transition to the 'normal' state at the end of a successful
3659 * probe. Should an EEH event occur during probe, the notification
3660 * thread (error_detected()) will wait until the probe handler
3661 * is nearly complete. At that time, the device will be moved to
3662 * a 'probed' state and the EEH thread woken up to drive the slot
3663 * reset and recovery (device moves to 'normal' state). Meanwhile,
3664 * the probe will be allowed to exit successfully.
3666 * Return: 0 on success, -errno on failure
3668 static int cxlflash_probe(struct pci_dev *pdev,
3669 const struct pci_device_id *dev_id)
3671 struct Scsi_Host *host;
3672 struct cxlflash_cfg *cfg = NULL;
3673 struct device *dev = &pdev->dev;
3674 struct dev_dependent_vals *ddv;
3678 dev_dbg(&pdev->dev, "%s: Found CXLFLASH with IRQ: %d\n",
3679 __func__, pdev->irq);
3681 ddv = (struct dev_dependent_vals *)dev_id->driver_data;
3682 driver_template.max_sectors = ddv->max_sectors;
3684 host = scsi_host_alloc(&driver_template, sizeof(struct cxlflash_cfg));
3686 dev_err(dev, "%s: scsi_host_alloc failed\n", __func__);
3691 host->max_id = CXLFLASH_MAX_NUM_TARGETS_PER_BUS;
3692 host->max_lun = CXLFLASH_MAX_NUM_LUNS_PER_TARGET;
3693 host->unique_id = host->host_no;
3694 host->max_cmd_len = CXLFLASH_MAX_CDB_LEN;
3696 cfg = shost_priv(host);
3697 cfg->state = STATE_PROBING;
3699 rc = alloc_mem(cfg);
3701 dev_err(dev, "%s: alloc_mem failed\n", __func__);
3703 scsi_host_put(cfg->host);
3707 cfg->init_state = INIT_STATE_NONE;
3709 cfg->cxl_fops = cxlflash_cxl_fops;
3710 cfg->ops = cxlflash_assign_ops(ddv);
3711 WARN_ON_ONCE(!cfg->ops);
3714 * Promoted LUNs move to the top of the LUN table. The rest stay on
3715 * the bottom half. The bottom half grows from the end (index = 255),
3716 * whereas the top half grows from the beginning (index = 0).
3718 * Initialize the last LUN index for all possible ports.
3720 cfg->promote_lun_index = 0;
3722 for (k = 0; k < MAX_FC_PORTS; k++)
3723 cfg->last_lun_index[k] = CXLFLASH_NUM_VLUNS/2 - 1;
3725 cfg->dev_id = (struct pci_device_id *)dev_id;
3727 init_waitqueue_head(&cfg->tmf_waitq);
3728 init_waitqueue_head(&cfg->reset_waitq);
3730 INIT_WORK(&cfg->work_q, cxlflash_worker_thread);
3731 cfg->lr_state = LINK_RESET_INVALID;
3733 spin_lock_init(&cfg->tmf_slock);
3734 mutex_init(&cfg->ctx_tbl_list_mutex);
3735 mutex_init(&cfg->ctx_recovery_mutex);
3736 init_rwsem(&cfg->ioctl_rwsem);
3737 INIT_LIST_HEAD(&cfg->ctx_err_recovery);
3738 INIT_LIST_HEAD(&cfg->lluns);
3740 pci_set_drvdata(pdev, cfg);
3744 dev_err(dev, "%s: init_pci failed rc=%d\n", __func__, rc);
3747 cfg->init_state = INIT_STATE_PCI;
3749 cfg->afu_cookie = cfg->ops->create_afu(pdev);
3750 if (unlikely(!cfg->afu_cookie)) {
3751 dev_err(dev, "%s: create_afu failed\n", __func__);
3757 if (rc && !wq_has_sleeper(&cfg->reset_waitq)) {
3758 dev_err(dev, "%s: init_afu failed rc=%d\n", __func__, rc);
3761 cfg->init_state = INIT_STATE_AFU;
3763 rc = init_scsi(cfg);
3765 dev_err(dev, "%s: init_scsi failed rc=%d\n", __func__, rc);
3768 cfg->init_state = INIT_STATE_SCSI;
3770 rc = init_chrdev(cfg);
3772 dev_err(dev, "%s: init_chrdev failed rc=%d\n", __func__, rc);
3775 cfg->init_state = INIT_STATE_CDEV;
3777 if (wq_has_sleeper(&cfg->reset_waitq)) {
3778 cfg->state = STATE_PROBED;
3779 wake_up_all(&cfg->reset_waitq);
3781 cfg->state = STATE_NORMAL;
3783 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3787 cfg->state = STATE_PROBED;
3788 cxlflash_remove(pdev);
3793 * cxlflash_pci_error_detected() - called when a PCI error is detected
3794 * @pdev: PCI device struct.
3795 * @state: PCI channel state.
3797 * When an EEH occurs during an active reset, wait until the reset is
3798 * complete and then take action based upon the device state.
3800 * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
3802 static pci_ers_result_t cxlflash_pci_error_detected(struct pci_dev *pdev,
3803 pci_channel_state_t state)
3806 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3807 struct device *dev = &cfg->dev->dev;
3809 dev_dbg(dev, "%s: pdev=%p state=%u\n", __func__, pdev, state);
3812 case pci_channel_io_frozen:
3813 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
3814 cfg->state != STATE_PROBING);
3815 if (cfg->state == STATE_FAILTERM)
3816 return PCI_ERS_RESULT_DISCONNECT;
3818 cfg->state = STATE_RESET;
3819 scsi_block_requests(cfg->host);
3821 rc = cxlflash_mark_contexts_error(cfg);
3823 dev_err(dev, "%s: Failed to mark user contexts rc=%d\n",
3826 return PCI_ERS_RESULT_NEED_RESET;
3827 case pci_channel_io_perm_failure:
3828 cfg->state = STATE_FAILTERM;
3829 wake_up_all(&cfg->reset_waitq);
3830 scsi_unblock_requests(cfg->host);
3831 return PCI_ERS_RESULT_DISCONNECT;
3835 return PCI_ERS_RESULT_NEED_RESET;
3839 * cxlflash_pci_slot_reset() - called when PCI slot has been reset
3840 * @pdev: PCI device struct.
3842 * This routine is called by the pci error recovery code after the PCI
3843 * slot has been reset, just before we should resume normal operations.
3845 * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
3847 static pci_ers_result_t cxlflash_pci_slot_reset(struct pci_dev *pdev)
3850 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3851 struct device *dev = &cfg->dev->dev;
3853 dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3857 dev_err(dev, "%s: EEH recovery failed rc=%d\n", __func__, rc);
3858 return PCI_ERS_RESULT_DISCONNECT;
3861 return PCI_ERS_RESULT_RECOVERED;
3865 * cxlflash_pci_resume() - called when normal operation can resume
3866 * @pdev: PCI device struct
3868 static void cxlflash_pci_resume(struct pci_dev *pdev)
3870 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3871 struct device *dev = &cfg->dev->dev;
3873 dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3875 cfg->state = STATE_NORMAL;
3876 wake_up_all(&cfg->reset_waitq);
3877 scsi_unblock_requests(cfg->host);
3881 * cxlflash_devnode() - provides devtmpfs for devices in the cxlflash class
3882 * @dev: Character device.
3883 * @mode: Mode that can be used to verify access.
3885 * Return: Allocated string describing the devtmpfs structure.
3887 static char *cxlflash_devnode(struct device *dev, umode_t *mode)
3889 return kasprintf(GFP_KERNEL, "cxlflash/%s", dev_name(dev));
3893 * cxlflash_class_init() - create character device class
3895 * Return: 0 on success, -errno on failure
3897 static int cxlflash_class_init(void)
3902 rc = alloc_chrdev_region(&devno, 0, CXLFLASH_MAX_ADAPTERS, "cxlflash");
3904 pr_err("%s: alloc_chrdev_region failed rc=%d\n", __func__, rc);
3908 cxlflash_major = MAJOR(devno);
3910 cxlflash_class = class_create(THIS_MODULE, "cxlflash");
3911 if (IS_ERR(cxlflash_class)) {
3912 rc = PTR_ERR(cxlflash_class);
3913 pr_err("%s: class_create failed rc=%d\n", __func__, rc);
3917 cxlflash_class->devnode = cxlflash_devnode;
3919 pr_debug("%s: returning rc=%d\n", __func__, rc);
3922 unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3927 * cxlflash_class_exit() - destroy character device class
3929 static void cxlflash_class_exit(void)
3931 dev_t devno = MKDEV(cxlflash_major, 0);
3933 class_destroy(cxlflash_class);
3934 unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3937 static const struct pci_error_handlers cxlflash_err_handler = {
3938 .error_detected = cxlflash_pci_error_detected,
3939 .slot_reset = cxlflash_pci_slot_reset,
3940 .resume = cxlflash_pci_resume,
3944 * PCI device structure
3946 static struct pci_driver cxlflash_driver = {
3947 .name = CXLFLASH_NAME,
3948 .id_table = cxlflash_pci_table,
3949 .probe = cxlflash_probe,
3950 .remove = cxlflash_remove,
3951 .shutdown = cxlflash_remove,
3952 .err_handler = &cxlflash_err_handler,
3956 * init_cxlflash() - module entry point
3958 * Return: 0 on success, -errno on failure
3960 static int __init init_cxlflash(void)
3965 cxlflash_list_init();
3966 rc = cxlflash_class_init();
3970 rc = pci_register_driver(&cxlflash_driver);
3974 pr_debug("%s: returning rc=%d\n", __func__, rc);
3977 cxlflash_class_exit();
3982 * exit_cxlflash() - module exit point
3984 static void __exit exit_cxlflash(void)
3986 cxlflash_term_global_luns();
3987 cxlflash_free_errpage();
3989 pci_unregister_driver(&cxlflash_driver);
3990 cxlflash_class_exit();
3993 module_init(init_cxlflash);
3994 module_exit(exit_cxlflash);