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>
24 #include <scsi/scsi_cmnd.h>
25 #include <scsi/scsi_host.h>
26 #include <uapi/scsi/cxlflash_ioctl.h>
32 MODULE_DESCRIPTION(CXLFLASH_ADAPTER_NAME);
33 MODULE_AUTHOR("Manoj N. Kumar <manoj@linux.vnet.ibm.com>");
34 MODULE_AUTHOR("Matthew R. Ochs <mrochs@linux.vnet.ibm.com>");
35 MODULE_LICENSE("GPL");
37 static struct class *cxlflash_class;
38 static u32 cxlflash_major;
39 static DECLARE_BITMAP(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
42 * process_cmd_err() - command error handler
43 * @cmd: AFU command that experienced the error.
44 * @scp: SCSI command associated with the AFU command in error.
46 * Translates error bits from AFU command to SCSI command results.
48 static void process_cmd_err(struct afu_cmd *cmd, struct scsi_cmnd *scp)
50 struct afu *afu = cmd->parent;
51 struct cxlflash_cfg *cfg = afu->parent;
52 struct device *dev = &cfg->dev->dev;
53 struct sisl_ioarcb *ioarcb;
54 struct sisl_ioasa *ioasa;
63 if (ioasa->rc.flags & SISL_RC_FLAGS_UNDERRUN) {
65 scsi_set_resid(scp, resid);
66 dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
67 __func__, cmd, scp, resid);
70 if (ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN) {
71 dev_dbg(dev, "%s: cmd underrun cmd = %p scp = %p\n",
73 scp->result = (DID_ERROR << 16);
76 dev_dbg(dev, "%s: cmd failed afu_rc=%02x scsi_rc=%02x fc_rc=%02x "
77 "afu_extra=%02x scsi_extra=%02x fc_extra=%02x\n", __func__,
78 ioasa->rc.afu_rc, ioasa->rc.scsi_rc, ioasa->rc.fc_rc,
79 ioasa->afu_extra, ioasa->scsi_extra, ioasa->fc_extra);
81 if (ioasa->rc.scsi_rc) {
82 /* We have a SCSI status */
83 if (ioasa->rc.flags & SISL_RC_FLAGS_SENSE_VALID) {
84 memcpy(scp->sense_buffer, ioasa->sense_data,
86 scp->result = ioasa->rc.scsi_rc;
88 scp->result = ioasa->rc.scsi_rc | (DID_ERROR << 16);
92 * We encountered an error. Set scp->result based on nature
95 if (ioasa->rc.fc_rc) {
96 /* We have an FC status */
97 switch (ioasa->rc.fc_rc) {
98 case SISL_FC_RC_LINKDOWN:
99 scp->result = (DID_REQUEUE << 16);
101 case SISL_FC_RC_RESID:
102 /* This indicates an FCP resid underrun */
103 if (!(ioasa->rc.flags & SISL_RC_FLAGS_OVERRUN)) {
104 /* If the SISL_RC_FLAGS_OVERRUN flag was set,
105 * then we will handle this error else where.
106 * If not then we must handle it here.
107 * This is probably an AFU bug.
109 scp->result = (DID_ERROR << 16);
112 case SISL_FC_RC_RESIDERR:
113 /* Resid mismatch between adapter and device */
114 case SISL_FC_RC_TGTABORT:
115 case SISL_FC_RC_ABORTOK:
116 case SISL_FC_RC_ABORTFAIL:
117 case SISL_FC_RC_NOLOGI:
118 case SISL_FC_RC_ABORTPEND:
119 case SISL_FC_RC_WRABORTPEND:
120 case SISL_FC_RC_NOEXP:
121 case SISL_FC_RC_INUSE:
122 scp->result = (DID_ERROR << 16);
127 if (ioasa->rc.afu_rc) {
128 /* We have an AFU error */
129 switch (ioasa->rc.afu_rc) {
130 case SISL_AFU_RC_NO_CHANNELS:
131 scp->result = (DID_NO_CONNECT << 16);
133 case SISL_AFU_RC_DATA_DMA_ERR:
134 switch (ioasa->afu_extra) {
135 case SISL_AFU_DMA_ERR_PAGE_IN:
137 scp->result = (DID_IMM_RETRY << 16);
139 case SISL_AFU_DMA_ERR_INVALID_EA:
141 scp->result = (DID_ERROR << 16);
144 case SISL_AFU_RC_OUT_OF_DATA_BUFS:
146 scp->result = (DID_ALLOC_FAILURE << 16);
149 scp->result = (DID_ERROR << 16);
155 * cmd_complete() - command completion handler
156 * @cmd: AFU command that has completed.
158 * For SCSI commands this routine prepares and submits commands that have
159 * either completed or timed out to the SCSI stack. For internal commands
160 * (TMF or AFU), this routine simply notifies the originator that the
161 * command has completed.
163 static void cmd_complete(struct afu_cmd *cmd)
165 struct scsi_cmnd *scp;
167 struct afu *afu = cmd->parent;
168 struct cxlflash_cfg *cfg = afu->parent;
169 struct device *dev = &cfg->dev->dev;
170 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
172 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
173 list_del(&cmd->list);
174 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
178 if (unlikely(cmd->sa.ioasc))
179 process_cmd_err(cmd, scp);
181 scp->result = (DID_OK << 16);
183 dev_dbg_ratelimited(dev, "%s:scp=%p result=%08x ioasc=%08x\n",
184 __func__, scp, scp->result, cmd->sa.ioasc);
186 } else if (cmd->cmd_tmf) {
187 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
188 cfg->tmf_active = false;
189 wake_up_all_locked(&cfg->tmf_waitq);
190 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
192 complete(&cmd->cevent);
196 * flush_pending_cmds() - flush all pending commands on this hardware queue
197 * @hwq: Hardware queue to flush.
199 * The hardware send queue lock associated with this hardware queue must be
200 * held when calling this routine.
202 static void flush_pending_cmds(struct hwq *hwq)
204 struct cxlflash_cfg *cfg = hwq->afu->parent;
205 struct afu_cmd *cmd, *tmp;
206 struct scsi_cmnd *scp;
209 list_for_each_entry_safe(cmd, tmp, &hwq->pending_cmds, list) {
210 /* Bypass command when on a doneq, cmd_complete() will handle */
211 if (!list_empty(&cmd->queue))
214 list_del(&cmd->list);
218 scp->result = (DID_IMM_RETRY << 16);
221 cmd->cmd_aborted = true;
224 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
225 cfg->tmf_active = false;
226 wake_up_all_locked(&cfg->tmf_waitq);
227 spin_unlock_irqrestore(&cfg->tmf_slock,
230 complete(&cmd->cevent);
236 * context_reset() - reset context via specified register
237 * @hwq: Hardware queue owning the context to be reset.
238 * @reset_reg: MMIO register to perform reset.
240 * When the reset is successful, the SISLite specification guarantees that
241 * the AFU has aborted all currently pending I/O. Accordingly, these commands
244 * Return: 0 on success, -errno on failure
246 static int context_reset(struct hwq *hwq, __be64 __iomem *reset_reg)
248 struct cxlflash_cfg *cfg = hwq->afu->parent;
249 struct device *dev = &cfg->dev->dev;
255 dev_dbg(dev, "%s: hwq=%p\n", __func__, hwq);
257 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
259 writeq_be(val, reset_reg);
261 val = readq_be(reset_reg);
262 if ((val & 0x1) == 0x0) {
267 /* Double delay each time */
269 } while (nretry++ < MC_ROOM_RETRY_CNT);
272 flush_pending_cmds(hwq);
274 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
276 dev_dbg(dev, "%s: returning rc=%d, val=%016llx nretry=%d\n",
277 __func__, rc, val, nretry);
282 * context_reset_ioarrin() - reset context via IOARRIN register
283 * @hwq: Hardware queue owning the context to be reset.
285 * Return: 0 on success, -errno on failure
287 static int context_reset_ioarrin(struct hwq *hwq)
289 return context_reset(hwq, &hwq->host_map->ioarrin);
293 * context_reset_sq() - reset context via SQ_CONTEXT_RESET register
294 * @hwq: Hardware queue owning the context to be reset.
296 * Return: 0 on success, -errno on failure
298 static int context_reset_sq(struct hwq *hwq)
300 return context_reset(hwq, &hwq->host_map->sq_ctx_reset);
304 * send_cmd_ioarrin() - sends an AFU command via IOARRIN register
305 * @afu: AFU associated with the host.
306 * @cmd: AFU command to send.
309 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
311 static int send_cmd_ioarrin(struct afu *afu, struct afu_cmd *cmd)
313 struct cxlflash_cfg *cfg = afu->parent;
314 struct device *dev = &cfg->dev->dev;
315 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
321 * To avoid the performance penalty of MMIO, spread the update of
322 * 'room' over multiple commands.
324 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
325 if (--hwq->room < 0) {
326 room = readq_be(&hwq->host_map->cmd_room);
328 dev_dbg_ratelimited(dev, "%s: no cmd_room to send "
329 "0x%02X, room=0x%016llX\n",
330 __func__, cmd->rcb.cdb[0], room);
332 rc = SCSI_MLQUEUE_HOST_BUSY;
335 hwq->room = room - 1;
338 list_add(&cmd->list, &hwq->pending_cmds);
339 writeq_be((u64)&cmd->rcb, &hwq->host_map->ioarrin);
341 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
342 dev_dbg(dev, "%s: cmd=%p len=%u ea=%016llx rc=%d\n", __func__,
343 cmd, cmd->rcb.data_len, cmd->rcb.data_ea, rc);
348 * send_cmd_sq() - sends an AFU command via SQ ring
349 * @afu: AFU associated with the host.
350 * @cmd: AFU command to send.
353 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
355 static int send_cmd_sq(struct afu *afu, struct afu_cmd *cmd)
357 struct cxlflash_cfg *cfg = afu->parent;
358 struct device *dev = &cfg->dev->dev;
359 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
364 newval = atomic_dec_if_positive(&hwq->hsq_credits);
366 rc = SCSI_MLQUEUE_HOST_BUSY;
370 cmd->rcb.ioasa = &cmd->sa;
372 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
374 *hwq->hsq_curr = cmd->rcb;
375 if (hwq->hsq_curr < hwq->hsq_end)
378 hwq->hsq_curr = hwq->hsq_start;
380 list_add(&cmd->list, &hwq->pending_cmds);
381 writeq_be((u64)hwq->hsq_curr, &hwq->host_map->sq_tail);
383 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
385 dev_dbg(dev, "%s: cmd=%p len=%u ea=%016llx ioasa=%p rc=%d curr=%p "
386 "head=%016llx tail=%016llx\n", __func__, cmd, cmd->rcb.data_len,
387 cmd->rcb.data_ea, cmd->rcb.ioasa, rc, hwq->hsq_curr,
388 readq_be(&hwq->host_map->sq_head),
389 readq_be(&hwq->host_map->sq_tail));
394 * wait_resp() - polls for a response or timeout to a sent AFU command
395 * @afu: AFU associated with the host.
396 * @cmd: AFU command that was sent.
398 * Return: 0 on success, -errno on failure
400 static int wait_resp(struct afu *afu, struct afu_cmd *cmd)
402 struct cxlflash_cfg *cfg = afu->parent;
403 struct device *dev = &cfg->dev->dev;
405 ulong timeout = msecs_to_jiffies(cmd->rcb.timeout * 2 * 1000);
407 timeout = wait_for_completion_timeout(&cmd->cevent, timeout);
411 if (cmd->cmd_aborted)
414 if (unlikely(cmd->sa.ioasc != 0)) {
415 dev_err(dev, "%s: cmd %02x failed, ioasc=%08x\n",
416 __func__, cmd->rcb.cdb[0], cmd->sa.ioasc);
424 * cmd_to_target_hwq() - selects a target hardware queue for a SCSI command
425 * @host: SCSI host associated with device.
426 * @scp: SCSI command to send.
427 * @afu: SCSI command to send.
429 * Hashes a command based upon the hardware queue mode.
431 * Return: Trusted index of target hardware queue
433 static u32 cmd_to_target_hwq(struct Scsi_Host *host, struct scsi_cmnd *scp,
439 if (afu->num_hwqs == 1)
442 switch (afu->hwq_mode) {
444 hwq = afu->hwq_rr_count++ % afu->num_hwqs;
447 tag = blk_mq_unique_tag(scp->request);
448 hwq = blk_mq_unique_tag_to_hwq(tag);
451 hwq = smp_processor_id() % afu->num_hwqs;
461 * send_tmf() - sends a Task Management Function (TMF)
462 * @cfg: Internal structure associated with the host.
463 * @sdev: SCSI device destined for TMF.
464 * @tmfcmd: TMF command to send.
467 * 0 on success, SCSI_MLQUEUE_HOST_BUSY or -errno on failure
469 static int send_tmf(struct cxlflash_cfg *cfg, struct scsi_device *sdev,
472 struct afu *afu = cfg->afu;
473 struct afu_cmd *cmd = NULL;
474 struct device *dev = &cfg->dev->dev;
475 struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
481 buf = kzalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
482 if (unlikely(!buf)) {
483 dev_err(dev, "%s: no memory for command\n", __func__);
488 cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
489 INIT_LIST_HEAD(&cmd->queue);
491 /* When Task Management Function is active do not send another */
492 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
494 wait_event_interruptible_lock_irq(cfg->tmf_waitq,
497 cfg->tmf_active = true;
498 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
502 cmd->hwq_index = hwq->index;
504 cmd->rcb.ctx_id = hwq->ctx_hndl;
505 cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
506 cmd->rcb.port_sel = CHAN2PORTMASK(sdev->channel);
507 cmd->rcb.lun_id = lun_to_lunid(sdev->lun);
508 cmd->rcb.req_flags = (SISL_REQ_FLAGS_PORT_LUN_ID |
509 SISL_REQ_FLAGS_SUP_UNDERRUN |
510 SISL_REQ_FLAGS_TMF_CMD);
511 memcpy(cmd->rcb.cdb, &tmfcmd, sizeof(tmfcmd));
513 rc = afu->send_cmd(afu, cmd);
515 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
516 cfg->tmf_active = false;
517 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
521 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
522 to = msecs_to_jiffies(5000);
523 to = wait_event_interruptible_lock_irq_timeout(cfg->tmf_waitq,
528 dev_err(dev, "%s: TMF timed out\n", __func__);
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);
546 * cxlflash_driver_info() - information handler for this host driver
547 * @host: SCSI host associated with device.
549 * Return: A string describing the device.
551 static const char *cxlflash_driver_info(struct Scsi_Host *host)
553 return CXLFLASH_ADAPTER_NAME;
557 * cxlflash_queuecommand() - sends a mid-layer request
558 * @host: SCSI host associated with device.
559 * @scp: SCSI command to send.
561 * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
563 static int cxlflash_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scp)
565 struct cxlflash_cfg *cfg = shost_priv(host);
566 struct afu *afu = cfg->afu;
567 struct device *dev = &cfg->dev->dev;
568 struct afu_cmd *cmd = sc_to_afuci(scp);
569 struct scatterlist *sg = scsi_sglist(scp);
570 int hwq_index = cmd_to_target_hwq(host, scp, afu);
571 struct hwq *hwq = get_hwq(afu, hwq_index);
572 u16 req_flags = SISL_REQ_FLAGS_SUP_UNDERRUN;
576 dev_dbg_ratelimited(dev, "%s: (scp=%p) %d/%d/%d/%llu "
577 "cdb=(%08x-%08x-%08x-%08x)\n",
578 __func__, scp, host->host_no, scp->device->channel,
579 scp->device->id, scp->device->lun,
580 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
581 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
582 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
583 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
586 * If a Task Management Function is active, wait for it to complete
587 * before continuing with regular commands.
589 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
590 if (cfg->tmf_active) {
591 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
592 rc = SCSI_MLQUEUE_HOST_BUSY;
595 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
597 switch (cfg->state) {
601 dev_dbg_ratelimited(dev, "%s: device is in reset\n", __func__);
602 rc = SCSI_MLQUEUE_HOST_BUSY;
605 dev_dbg_ratelimited(dev, "%s: device has failed\n", __func__);
606 scp->result = (DID_NO_CONNECT << 16);
615 cmd->rcb.data_len = sg->length;
616 cmd->rcb.data_ea = (uintptr_t)sg_virt(sg);
621 cmd->hwq_index = hwq_index;
624 cmd->rcb.ctx_id = hwq->ctx_hndl;
625 cmd->rcb.msi = SISL_MSI_RRQ_UPDATED;
626 cmd->rcb.port_sel = CHAN2PORTMASK(scp->device->channel);
627 cmd->rcb.lun_id = lun_to_lunid(scp->device->lun);
629 if (scp->sc_data_direction == DMA_TO_DEVICE)
630 req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
632 cmd->rcb.req_flags = req_flags;
633 memcpy(cmd->rcb.cdb, scp->cmnd, sizeof(cmd->rcb.cdb));
635 rc = afu->send_cmd(afu, cmd);
641 * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
642 * @cfg: Internal structure associated with the host.
644 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg *cfg)
646 struct pci_dev *pdev = cfg->dev;
648 if (pci_channel_offline(pdev))
649 wait_event_timeout(cfg->reset_waitq,
650 !pci_channel_offline(pdev),
651 CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT);
655 * free_mem() - free memory associated with the AFU
656 * @cfg: Internal structure associated with the host.
658 static void free_mem(struct cxlflash_cfg *cfg)
660 struct afu *afu = cfg->afu;
663 free_pages((ulong)afu, get_order(sizeof(struct afu)));
669 * cxlflash_reset_sync() - synchronizing point for asynchronous resets
670 * @cfg: Internal structure associated with the host.
672 static void cxlflash_reset_sync(struct cxlflash_cfg *cfg)
674 if (cfg->async_reset_cookie == 0)
677 /* Wait until all async calls prior to this cookie have completed */
678 async_synchronize_cookie(cfg->async_reset_cookie + 1);
679 cfg->async_reset_cookie = 0;
683 * stop_afu() - stops the AFU command timers and unmaps the MMIO space
684 * @cfg: Internal structure associated with the host.
686 * Safe to call with AFU in a partially allocated/initialized state.
688 * Cancels scheduled worker threads, waits for any active internal AFU
689 * commands to timeout, disables IRQ polling and then unmaps the MMIO space.
691 static void stop_afu(struct cxlflash_cfg *cfg)
693 struct afu *afu = cfg->afu;
697 cancel_work_sync(&cfg->work_q);
698 if (!current_is_async())
699 cxlflash_reset_sync(cfg);
702 while (atomic_read(&afu->cmds_active))
705 if (afu_is_irqpoll_enabled(afu)) {
706 for (i = 0; i < afu->num_hwqs; i++) {
707 hwq = get_hwq(afu, i);
709 irq_poll_disable(&hwq->irqpoll);
713 if (likely(afu->afu_map)) {
714 cxl_psa_unmap((void __iomem *)afu->afu_map);
721 * term_intr() - disables all AFU interrupts
722 * @cfg: Internal structure associated with the host.
723 * @level: Depth of allocation, where to begin waterfall tear down.
724 * @index: Index of the hardware queue.
726 * Safe to call with AFU/MC in partially allocated/initialized state.
728 static void term_intr(struct cxlflash_cfg *cfg, enum undo_level level,
731 struct afu *afu = cfg->afu;
732 struct device *dev = &cfg->dev->dev;
736 dev_err(dev, "%s: returning with NULL afu\n", __func__);
740 hwq = get_hwq(afu, index);
743 dev_err(dev, "%s: returning with NULL MC\n", __func__);
749 /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
750 if (index == PRIMARY_HWQ)
751 cxl_unmap_afu_irq(hwq->ctx, 3, hwq);
753 cxl_unmap_afu_irq(hwq->ctx, 2, hwq);
755 cxl_unmap_afu_irq(hwq->ctx, 1, hwq);
757 cxl_free_afu_irqs(hwq->ctx);
760 /* No action required */
766 * term_mc() - terminates the master context
767 * @cfg: Internal structure associated with the host.
768 * @index: Index of the hardware queue.
770 * Safe to call with AFU/MC in partially allocated/initialized state.
772 static void term_mc(struct cxlflash_cfg *cfg, u32 index)
774 struct afu *afu = cfg->afu;
775 struct device *dev = &cfg->dev->dev;
780 dev_err(dev, "%s: returning with NULL afu\n", __func__);
784 hwq = get_hwq(afu, index);
787 dev_err(dev, "%s: returning with NULL MC\n", __func__);
791 WARN_ON(cxl_stop_context(hwq->ctx));
792 if (index != PRIMARY_HWQ)
793 WARN_ON(cxl_release_context(hwq->ctx));
796 spin_lock_irqsave(&hwq->hsq_slock, lock_flags);
797 flush_pending_cmds(hwq);
798 spin_unlock_irqrestore(&hwq->hsq_slock, lock_flags);
802 * term_afu() - terminates the AFU
803 * @cfg: Internal structure associated with the host.
805 * Safe to call with AFU/MC in partially allocated/initialized state.
807 static void term_afu(struct cxlflash_cfg *cfg)
809 struct device *dev = &cfg->dev->dev;
813 * Tear down is carefully orchestrated to ensure
814 * no interrupts can come in when the problem state
817 * 1) Disable all AFU interrupts for each master
818 * 2) Unmap the problem state area
819 * 3) Stop each master context
821 for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
822 term_intr(cfg, UNMAP_THREE, k);
826 for (k = cfg->afu->num_hwqs - 1; k >= 0; k--)
829 dev_dbg(dev, "%s: returning\n", __func__);
833 * notify_shutdown() - notifies device of pending shutdown
834 * @cfg: Internal structure associated with the host.
835 * @wait: Whether to wait for shutdown processing to complete.
837 * This function will notify the AFU that the adapter is being shutdown
838 * and will wait for shutdown processing to complete if wait is true.
839 * This notification should flush pending I/Os to the device and halt
840 * further I/Os until the next AFU reset is issued and device restarted.
842 static void notify_shutdown(struct cxlflash_cfg *cfg, bool wait)
844 struct afu *afu = cfg->afu;
845 struct device *dev = &cfg->dev->dev;
846 struct dev_dependent_vals *ddv;
847 __be64 __iomem *fc_port_regs;
849 int i, retry_cnt = 0;
851 ddv = (struct dev_dependent_vals *)cfg->dev_id->driver_data;
852 if (!(ddv->flags & CXLFLASH_NOTIFY_SHUTDOWN))
855 if (!afu || !afu->afu_map) {
856 dev_dbg(dev, "%s: Problem state area not mapped\n", __func__);
861 for (i = 0; i < cfg->num_fc_ports; i++) {
862 fc_port_regs = get_fc_port_regs(cfg, i);
864 reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
865 reg |= SISL_FC_SHUTDOWN_NORMAL;
866 writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
872 /* Wait up to 1.5 seconds for shutdown processing to complete */
873 for (i = 0; i < cfg->num_fc_ports; i++) {
874 fc_port_regs = get_fc_port_regs(cfg, i);
878 status = readq_be(&fc_port_regs[FC_STATUS / 8]);
879 if (status & SISL_STATUS_SHUTDOWN_COMPLETE)
881 if (++retry_cnt >= MC_RETRY_CNT) {
882 dev_dbg(dev, "%s: port %d shutdown processing "
883 "not yet completed\n", __func__, i);
886 msleep(100 * retry_cnt);
892 * cxlflash_get_minor() - gets the first available minor number
894 * Return: Unique minor number that can be used to create the character device.
896 static int cxlflash_get_minor(void)
901 bit = find_first_zero_bit(cxlflash_minor, CXLFLASH_MAX_ADAPTERS);
902 if (bit >= CXLFLASH_MAX_ADAPTERS)
905 minor = bit & MINORMASK;
906 set_bit(minor, cxlflash_minor);
911 * cxlflash_put_minor() - releases the minor number
912 * @minor: Minor number that is no longer needed.
914 static void cxlflash_put_minor(int minor)
916 clear_bit(minor, cxlflash_minor);
920 * cxlflash_release_chrdev() - release the character device for the host
921 * @cfg: Internal structure associated with the host.
923 static void cxlflash_release_chrdev(struct cxlflash_cfg *cfg)
925 device_unregister(cfg->chardev);
927 cdev_del(&cfg->cdev);
928 cxlflash_put_minor(MINOR(cfg->cdev.dev));
932 * cxlflash_remove() - PCI entry point to tear down host
933 * @pdev: PCI device associated with the host.
935 * Safe to use as a cleanup in partially allocated/initialized state. Note that
936 * the reset_waitq is flushed as part of the stop/termination of user contexts.
938 static void cxlflash_remove(struct pci_dev *pdev)
940 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
941 struct device *dev = &pdev->dev;
944 if (!pci_is_enabled(pdev)) {
945 dev_dbg(dev, "%s: Device is disabled\n", __func__);
949 /* Yield to running recovery threads before continuing with remove */
950 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
951 cfg->state != STATE_PROBING);
952 spin_lock_irqsave(&cfg->tmf_slock, lock_flags);
954 wait_event_interruptible_lock_irq(cfg->tmf_waitq,
957 spin_unlock_irqrestore(&cfg->tmf_slock, lock_flags);
959 /* Notify AFU and wait for shutdown processing to complete */
960 notify_shutdown(cfg, true);
962 cfg->state = STATE_FAILTERM;
963 cxlflash_stop_term_user_contexts(cfg);
965 switch (cfg->init_state) {
966 case INIT_STATE_CDEV:
967 cxlflash_release_chrdev(cfg);
968 case INIT_STATE_SCSI:
969 cxlflash_term_local_luns(cfg);
970 scsi_remove_host(cfg->host);
974 pci_disable_device(pdev);
975 case INIT_STATE_NONE:
977 scsi_host_put(cfg->host);
981 dev_dbg(dev, "%s: returning\n", __func__);
985 * alloc_mem() - allocates the AFU and its command pool
986 * @cfg: Internal structure associated with the host.
988 * A partially allocated state remains on failure.
992 * -ENOMEM on failure to allocate memory
994 static int alloc_mem(struct cxlflash_cfg *cfg)
997 struct device *dev = &cfg->dev->dev;
999 /* AFU is ~28k, i.e. only one 64k page or up to seven 4k pages */
1000 cfg->afu = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1001 get_order(sizeof(struct afu)));
1002 if (unlikely(!cfg->afu)) {
1003 dev_err(dev, "%s: cannot get %d free pages\n",
1004 __func__, get_order(sizeof(struct afu)));
1008 cfg->afu->parent = cfg;
1009 cfg->afu->desired_hwqs = CXLFLASH_DEF_HWQS;
1010 cfg->afu->afu_map = NULL;
1016 * init_pci() - initializes the host as a PCI device
1017 * @cfg: Internal structure associated with the host.
1019 * Return: 0 on success, -errno on failure
1021 static int init_pci(struct cxlflash_cfg *cfg)
1023 struct pci_dev *pdev = cfg->dev;
1024 struct device *dev = &cfg->dev->dev;
1027 rc = pci_enable_device(pdev);
1028 if (rc || pci_channel_offline(pdev)) {
1029 if (pci_channel_offline(pdev)) {
1030 cxlflash_wait_for_pci_err_recovery(cfg);
1031 rc = pci_enable_device(pdev);
1035 dev_err(dev, "%s: Cannot enable adapter\n", __func__);
1036 cxlflash_wait_for_pci_err_recovery(cfg);
1042 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1047 * init_scsi() - adds the host to the SCSI stack and kicks off host scan
1048 * @cfg: Internal structure associated with the host.
1050 * Return: 0 on success, -errno on failure
1052 static int init_scsi(struct cxlflash_cfg *cfg)
1054 struct pci_dev *pdev = cfg->dev;
1055 struct device *dev = &cfg->dev->dev;
1058 rc = scsi_add_host(cfg->host, &pdev->dev);
1060 dev_err(dev, "%s: scsi_add_host failed rc=%d\n", __func__, rc);
1064 scsi_scan_host(cfg->host);
1067 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1072 * set_port_online() - transitions the specified host FC port to online state
1073 * @fc_regs: Top of MMIO region defined for specified port.
1075 * The provided MMIO region must be mapped prior to call. Online state means
1076 * that the FC link layer has synced, completed the handshaking process, and
1077 * is ready for login to start.
1079 static void set_port_online(__be64 __iomem *fc_regs)
1083 cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1084 cmdcfg &= (~FC_MTIP_CMDCONFIG_OFFLINE); /* clear OFF_LINE */
1085 cmdcfg |= (FC_MTIP_CMDCONFIG_ONLINE); /* set ON_LINE */
1086 writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1090 * set_port_offline() - transitions the specified host FC port to offline state
1091 * @fc_regs: Top of MMIO region defined for specified port.
1093 * The provided MMIO region must be mapped prior to call.
1095 static void set_port_offline(__be64 __iomem *fc_regs)
1099 cmdcfg = readq_be(&fc_regs[FC_MTIP_CMDCONFIG / 8]);
1100 cmdcfg &= (~FC_MTIP_CMDCONFIG_ONLINE); /* clear ON_LINE */
1101 cmdcfg |= (FC_MTIP_CMDCONFIG_OFFLINE); /* set OFF_LINE */
1102 writeq_be(cmdcfg, &fc_regs[FC_MTIP_CMDCONFIG / 8]);
1106 * wait_port_online() - waits for the specified host FC port come online
1107 * @fc_regs: Top of MMIO region defined for specified port.
1108 * @delay_us: Number of microseconds to delay between reading port status.
1109 * @nretry: Number of cycles to retry reading port status.
1111 * The provided MMIO region must be mapped prior to call. This will timeout
1112 * when the cable is not plugged in.
1115 * TRUE (1) when the specified port is online
1116 * FALSE (0) when the specified port fails to come online after timeout
1118 static bool wait_port_online(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1122 WARN_ON(delay_us < 1000);
1125 msleep(delay_us / 1000);
1126 status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1127 if (status == U64_MAX)
1129 } while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_ONLINE &&
1132 return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_ONLINE);
1136 * wait_port_offline() - waits for the specified host FC port go offline
1137 * @fc_regs: Top of MMIO region defined for specified port.
1138 * @delay_us: Number of microseconds to delay between reading port status.
1139 * @nretry: Number of cycles to retry reading port status.
1141 * The provided MMIO region must be mapped prior to call.
1144 * TRUE (1) when the specified port is offline
1145 * FALSE (0) when the specified port fails to go offline after timeout
1147 static bool wait_port_offline(__be64 __iomem *fc_regs, u32 delay_us, u32 nretry)
1151 WARN_ON(delay_us < 1000);
1154 msleep(delay_us / 1000);
1155 status = readq_be(&fc_regs[FC_MTIP_STATUS / 8]);
1156 if (status == U64_MAX)
1158 } while ((status & FC_MTIP_STATUS_MASK) != FC_MTIP_STATUS_OFFLINE &&
1161 return ((status & FC_MTIP_STATUS_MASK) == FC_MTIP_STATUS_OFFLINE);
1165 * afu_set_wwpn() - configures the WWPN for the specified host FC port
1166 * @afu: AFU associated with the host that owns the specified FC port.
1167 * @port: Port number being configured.
1168 * @fc_regs: Top of MMIO region defined for specified port.
1169 * @wwpn: The world-wide-port-number previously discovered for port.
1171 * The provided MMIO region must be mapped prior to call. As part of the
1172 * sequence to configure the WWPN, the port is toggled offline and then back
1173 * online. This toggling action can cause this routine to delay up to a few
1174 * seconds. When configured to use the internal LUN feature of the AFU, a
1175 * failure to come online is overridden.
1177 static void afu_set_wwpn(struct afu *afu, int port, __be64 __iomem *fc_regs,
1180 struct cxlflash_cfg *cfg = afu->parent;
1181 struct device *dev = &cfg->dev->dev;
1183 set_port_offline(fc_regs);
1184 if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1185 FC_PORT_STATUS_RETRY_CNT)) {
1186 dev_dbg(dev, "%s: wait on port %d to go offline timed out\n",
1190 writeq_be(wwpn, &fc_regs[FC_PNAME / 8]);
1192 set_port_online(fc_regs);
1193 if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1194 FC_PORT_STATUS_RETRY_CNT)) {
1195 dev_dbg(dev, "%s: wait on port %d to go online timed out\n",
1201 * afu_link_reset() - resets the specified host FC port
1202 * @afu: AFU associated with the host that owns the specified FC port.
1203 * @port: Port number being configured.
1204 * @fc_regs: Top of MMIO region defined for specified port.
1206 * The provided MMIO region must be mapped prior to call. The sequence to
1207 * reset the port involves toggling it offline and then back online. This
1208 * action can cause this routine to delay up to a few seconds. An effort
1209 * is made to maintain link with the device by switching to host to use
1210 * the alternate port exclusively while the reset takes place.
1211 * failure to come online is overridden.
1213 static void afu_link_reset(struct afu *afu, int port, __be64 __iomem *fc_regs)
1215 struct cxlflash_cfg *cfg = afu->parent;
1216 struct device *dev = &cfg->dev->dev;
1219 /* first switch the AFU to the other links, if any */
1220 port_sel = readq_be(&afu->afu_map->global.regs.afu_port_sel);
1221 port_sel &= ~(1ULL << port);
1222 writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1223 cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1225 set_port_offline(fc_regs);
1226 if (!wait_port_offline(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1227 FC_PORT_STATUS_RETRY_CNT))
1228 dev_err(dev, "%s: wait on port %d to go offline timed out\n",
1231 set_port_online(fc_regs);
1232 if (!wait_port_online(fc_regs, FC_PORT_STATUS_RETRY_INTERVAL_US,
1233 FC_PORT_STATUS_RETRY_CNT))
1234 dev_err(dev, "%s: wait on port %d to go online timed out\n",
1237 /* switch back to include this port */
1238 port_sel |= (1ULL << port);
1239 writeq_be(port_sel, &afu->afu_map->global.regs.afu_port_sel);
1240 cxlflash_afu_sync(afu, 0, 0, AFU_GSYNC);
1242 dev_dbg(dev, "%s: returning port_sel=%016llx\n", __func__, port_sel);
1246 * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1247 * @afu: AFU associated with the host.
1249 static void afu_err_intr_init(struct afu *afu)
1251 struct cxlflash_cfg *cfg = afu->parent;
1252 __be64 __iomem *fc_port_regs;
1254 struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
1257 /* global async interrupts: AFU clears afu_ctrl on context exit
1258 * if async interrupts were sent to that context. This prevents
1259 * the AFU form sending further async interrupts when
1261 * nobody to receive them.
1265 writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_mask);
1266 /* set LISN# to send and point to primary master context */
1267 reg = ((u64) (((hwq->ctx_hndl << 8) | SISL_MSI_ASYNC_ERROR)) << 40);
1269 if (afu->internal_lun)
1270 reg |= 1; /* Bit 63 indicates local lun */
1271 writeq_be(reg, &afu->afu_map->global.regs.afu_ctrl);
1273 writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1274 /* unmask bits that are of interest */
1275 /* note: afu can send an interrupt after this step */
1276 writeq_be(SISL_ASTATUS_MASK, &afu->afu_map->global.regs.aintr_mask);
1277 /* clear again in case a bit came on after previous clear but before */
1279 writeq_be(-1ULL, &afu->afu_map->global.regs.aintr_clear);
1281 /* Clear/Set internal lun bits */
1282 fc_port_regs = get_fc_port_regs(cfg, 0);
1283 reg = readq_be(&fc_port_regs[FC_CONFIG2 / 8]);
1284 reg &= SISL_FC_INTERNAL_MASK;
1285 if (afu->internal_lun)
1286 reg |= ((u64)(afu->internal_lun - 1) << SISL_FC_INTERNAL_SHIFT);
1287 writeq_be(reg, &fc_port_regs[FC_CONFIG2 / 8]);
1289 /* now clear FC errors */
1290 for (i = 0; i < cfg->num_fc_ports; i++) {
1291 fc_port_regs = get_fc_port_regs(cfg, i);
1293 writeq_be(0xFFFFFFFFU, &fc_port_regs[FC_ERROR / 8]);
1294 writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1297 /* sync interrupts for master's IOARRIN write */
1298 /* note that unlike asyncs, there can be no pending sync interrupts */
1299 /* at this time (this is a fresh context and master has not written */
1300 /* IOARRIN yet), so there is nothing to clear. */
1302 /* set LISN#, it is always sent to the context that wrote IOARRIN */
1303 for (i = 0; i < afu->num_hwqs; i++) {
1304 hwq = get_hwq(afu, i);
1306 reg = readq_be(&hwq->host_map->ctx_ctrl);
1307 WARN_ON((reg & SISL_CTX_CTRL_LISN_MASK) != 0);
1308 reg |= SISL_MSI_SYNC_ERROR;
1309 writeq_be(reg, &hwq->host_map->ctx_ctrl);
1310 writeq_be(SISL_ISTATUS_MASK, &hwq->host_map->intr_mask);
1315 * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1316 * @irq: Interrupt number.
1317 * @data: Private data provided at interrupt registration, the AFU.
1319 * Return: Always return IRQ_HANDLED.
1321 static irqreturn_t cxlflash_sync_err_irq(int irq, void *data)
1323 struct hwq *hwq = (struct hwq *)data;
1324 struct cxlflash_cfg *cfg = hwq->afu->parent;
1325 struct device *dev = &cfg->dev->dev;
1329 reg = readq_be(&hwq->host_map->intr_status);
1330 reg_unmasked = (reg & SISL_ISTATUS_UNMASK);
1332 if (reg_unmasked == 0UL) {
1333 dev_err(dev, "%s: spurious interrupt, intr_status=%016llx\n",
1335 goto cxlflash_sync_err_irq_exit;
1338 dev_err(dev, "%s: unexpected interrupt, intr_status=%016llx\n",
1341 writeq_be(reg_unmasked, &hwq->host_map->intr_clear);
1343 cxlflash_sync_err_irq_exit:
1348 * process_hrrq() - process the read-response queue
1349 * @afu: AFU associated with the host.
1350 * @doneq: Queue of commands harvested from the RRQ.
1351 * @budget: Threshold of RRQ entries to process.
1353 * This routine must be called holding the disabled RRQ spin lock.
1355 * Return: The number of entries processed.
1357 static int process_hrrq(struct hwq *hwq, struct list_head *doneq, int budget)
1359 struct afu *afu = hwq->afu;
1360 struct afu_cmd *cmd;
1361 struct sisl_ioasa *ioasa;
1362 struct sisl_ioarcb *ioarcb;
1363 bool toggle = hwq->toggle;
1366 *hrrq_start = hwq->hrrq_start,
1367 *hrrq_end = hwq->hrrq_end,
1368 *hrrq_curr = hwq->hrrq_curr;
1370 /* Process ready RRQ entries up to the specified budget (if any) */
1374 if ((entry & SISL_RESP_HANDLE_T_BIT) != toggle)
1377 entry &= ~SISL_RESP_HANDLE_T_BIT;
1379 if (afu_is_sq_cmd_mode(afu)) {
1380 ioasa = (struct sisl_ioasa *)entry;
1381 cmd = container_of(ioasa, struct afu_cmd, sa);
1383 ioarcb = (struct sisl_ioarcb *)entry;
1384 cmd = container_of(ioarcb, struct afu_cmd, rcb);
1387 list_add_tail(&cmd->queue, doneq);
1389 /* Advance to next entry or wrap and flip the toggle bit */
1390 if (hrrq_curr < hrrq_end)
1393 hrrq_curr = hrrq_start;
1394 toggle ^= SISL_RESP_HANDLE_T_BIT;
1397 atomic_inc(&hwq->hsq_credits);
1400 if (budget > 0 && num_hrrq >= budget)
1404 hwq->hrrq_curr = hrrq_curr;
1405 hwq->toggle = toggle;
1411 * process_cmd_doneq() - process a queue of harvested RRQ commands
1412 * @doneq: Queue of completed commands.
1414 * Note that upon return the queue can no longer be trusted.
1416 static void process_cmd_doneq(struct list_head *doneq)
1418 struct afu_cmd *cmd, *tmp;
1420 WARN_ON(list_empty(doneq));
1422 list_for_each_entry_safe(cmd, tmp, doneq, queue)
1427 * cxlflash_irqpoll() - process a queue of harvested RRQ commands
1428 * @irqpoll: IRQ poll structure associated with queue to poll.
1429 * @budget: Threshold of RRQ entries to process per poll.
1431 * Return: The number of entries processed.
1433 static int cxlflash_irqpoll(struct irq_poll *irqpoll, int budget)
1435 struct hwq *hwq = container_of(irqpoll, struct hwq, irqpoll);
1436 unsigned long hrrq_flags;
1438 int num_entries = 0;
1440 spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1442 num_entries = process_hrrq(hwq, &doneq, budget);
1443 if (num_entries < budget)
1444 irq_poll_complete(irqpoll);
1446 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1448 process_cmd_doneq(&doneq);
1453 * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1454 * @irq: Interrupt number.
1455 * @data: Private data provided at interrupt registration, the AFU.
1457 * Return: IRQ_HANDLED or IRQ_NONE when no ready entries found.
1459 static irqreturn_t cxlflash_rrq_irq(int irq, void *data)
1461 struct hwq *hwq = (struct hwq *)data;
1462 struct afu *afu = hwq->afu;
1463 unsigned long hrrq_flags;
1465 int num_entries = 0;
1467 spin_lock_irqsave(&hwq->hrrq_slock, hrrq_flags);
1469 if (afu_is_irqpoll_enabled(afu)) {
1470 irq_poll_sched(&hwq->irqpoll);
1471 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1475 num_entries = process_hrrq(hwq, &doneq, -1);
1476 spin_unlock_irqrestore(&hwq->hrrq_slock, hrrq_flags);
1478 if (num_entries == 0)
1481 process_cmd_doneq(&doneq);
1486 * Asynchronous interrupt information table
1489 * - Order matters here as this array is indexed by bit position.
1491 * - The checkpatch script considers the BUILD_SISL_ASTATUS_FC_PORT macro
1492 * as complex and complains due to a lack of parentheses/braces.
1494 #define ASTATUS_FC(_a, _b, _c, _d) \
1495 { SISL_ASTATUS_FC##_a##_##_b, _c, _a, (_d) }
1497 #define BUILD_SISL_ASTATUS_FC_PORT(_a) \
1498 ASTATUS_FC(_a, LINK_UP, "link up", 0), \
1499 ASTATUS_FC(_a, LINK_DN, "link down", 0), \
1500 ASTATUS_FC(_a, LOGI_S, "login succeeded", SCAN_HOST), \
1501 ASTATUS_FC(_a, LOGI_F, "login failed", CLR_FC_ERROR), \
1502 ASTATUS_FC(_a, LOGI_R, "login timed out, retrying", LINK_RESET), \
1503 ASTATUS_FC(_a, CRC_T, "CRC threshold exceeded", LINK_RESET), \
1504 ASTATUS_FC(_a, LOGO, "target initiated LOGO", 0), \
1505 ASTATUS_FC(_a, OTHER, "other error", CLR_FC_ERROR | LINK_RESET)
1507 static const struct asyc_intr_info ainfo[] = {
1508 BUILD_SISL_ASTATUS_FC_PORT(1),
1509 BUILD_SISL_ASTATUS_FC_PORT(0),
1510 BUILD_SISL_ASTATUS_FC_PORT(3),
1511 BUILD_SISL_ASTATUS_FC_PORT(2)
1515 * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1516 * @irq: Interrupt number.
1517 * @data: Private data provided at interrupt registration, the AFU.
1519 * Return: Always return IRQ_HANDLED.
1521 static irqreturn_t cxlflash_async_err_irq(int irq, void *data)
1523 struct hwq *hwq = (struct hwq *)data;
1524 struct afu *afu = hwq->afu;
1525 struct cxlflash_cfg *cfg = afu->parent;
1526 struct device *dev = &cfg->dev->dev;
1527 const struct asyc_intr_info *info;
1528 struct sisl_global_map __iomem *global = &afu->afu_map->global;
1529 __be64 __iomem *fc_port_regs;
1535 reg = readq_be(&global->regs.aintr_status);
1536 reg_unmasked = (reg & SISL_ASTATUS_UNMASK);
1538 if (unlikely(reg_unmasked == 0)) {
1539 dev_err(dev, "%s: spurious interrupt, aintr_status=%016llx\n",
1544 /* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1545 writeq_be(reg_unmasked, &global->regs.aintr_clear);
1547 /* Check each bit that is on */
1548 for_each_set_bit(bit, (ulong *)®_unmasked, BITS_PER_LONG) {
1549 if (unlikely(bit >= ARRAY_SIZE(ainfo))) {
1555 if (unlikely(info->status != 1ULL << bit)) {
1561 fc_port_regs = get_fc_port_regs(cfg, port);
1563 dev_err(dev, "%s: FC Port %d -> %s, fc_status=%016llx\n",
1564 __func__, port, info->desc,
1565 readq_be(&fc_port_regs[FC_STATUS / 8]));
1568 * Do link reset first, some OTHER errors will set FC_ERROR
1569 * again if cleared before or w/o a reset
1571 if (info->action & LINK_RESET) {
1572 dev_err(dev, "%s: FC Port %d: resetting link\n",
1574 cfg->lr_state = LINK_RESET_REQUIRED;
1575 cfg->lr_port = port;
1576 schedule_work(&cfg->work_q);
1579 if (info->action & CLR_FC_ERROR) {
1580 reg = readq_be(&fc_port_regs[FC_ERROR / 8]);
1583 * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1584 * should be the same and tracing one is sufficient.
1587 dev_err(dev, "%s: fc %d: clearing fc_error=%016llx\n",
1588 __func__, port, reg);
1590 writeq_be(reg, &fc_port_regs[FC_ERROR / 8]);
1591 writeq_be(0, &fc_port_regs[FC_ERRCAP / 8]);
1594 if (info->action & SCAN_HOST) {
1595 atomic_inc(&cfg->scan_host_needed);
1596 schedule_work(&cfg->work_q);
1605 * start_context() - starts the master context
1606 * @cfg: Internal structure associated with the host.
1607 * @index: Index of the hardware queue.
1609 * Return: A success or failure value from CXL services.
1611 static int start_context(struct cxlflash_cfg *cfg, u32 index)
1613 struct device *dev = &cfg->dev->dev;
1614 struct hwq *hwq = get_hwq(cfg->afu, index);
1617 rc = cxl_start_context(hwq->ctx,
1618 hwq->work.work_element_descriptor,
1621 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1626 * read_vpd() - obtains the WWPNs from VPD
1627 * @cfg: Internal structure associated with the host.
1628 * @wwpn: Array of size MAX_FC_PORTS to pass back WWPNs
1630 * Return: 0 on success, -errno on failure
1632 static int read_vpd(struct cxlflash_cfg *cfg, u64 wwpn[])
1634 struct device *dev = &cfg->dev->dev;
1635 struct pci_dev *pdev = cfg->dev;
1637 int ro_start, ro_size, i, j, k;
1639 char vpd_data[CXLFLASH_VPD_LEN];
1640 char tmp_buf[WWPN_BUF_LEN] = { 0 };
1641 char *wwpn_vpd_tags[MAX_FC_PORTS] = { "V5", "V6", "V7", "V8" };
1643 /* Get the VPD data from the device */
1644 vpd_size = cxl_read_adapter_vpd(pdev, vpd_data, sizeof(vpd_data));
1645 if (unlikely(vpd_size <= 0)) {
1646 dev_err(dev, "%s: Unable to read VPD (size = %ld)\n",
1647 __func__, vpd_size);
1652 /* Get the read only section offset */
1653 ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size,
1654 PCI_VPD_LRDT_RO_DATA);
1655 if (unlikely(ro_start < 0)) {
1656 dev_err(dev, "%s: VPD Read-only data not found\n", __func__);
1661 /* Get the read only section size, cap when extends beyond read VPD */
1662 ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
1664 i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1665 if (unlikely((i + j) > vpd_size)) {
1666 dev_dbg(dev, "%s: Might need to read more VPD (%d > %ld)\n",
1667 __func__, (i + j), vpd_size);
1668 ro_size = vpd_size - i;
1672 * Find the offset of the WWPN tag within the read only
1673 * VPD data and validate the found field (partials are
1674 * no good to us). Convert the ASCII data to an integer
1675 * value. Note that we must copy to a temporary buffer
1676 * because the conversion service requires that the ASCII
1677 * string be terminated.
1679 for (k = 0; k < cfg->num_fc_ports; k++) {
1681 i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
1683 i = pci_vpd_find_info_keyword(vpd_data, i, j, wwpn_vpd_tags[k]);
1684 if (unlikely(i < 0)) {
1685 dev_err(dev, "%s: Port %d WWPN not found in VPD\n",
1691 j = pci_vpd_info_field_size(&vpd_data[i]);
1692 i += PCI_VPD_INFO_FLD_HDR_SIZE;
1693 if (unlikely((i + j > vpd_size) || (j != WWPN_LEN))) {
1694 dev_err(dev, "%s: Port %d WWPN incomplete or bad VPD\n",
1700 memcpy(tmp_buf, &vpd_data[i], WWPN_LEN);
1701 rc = kstrtoul(tmp_buf, WWPN_LEN, (ulong *)&wwpn[k]);
1703 dev_err(dev, "%s: WWPN conversion failed for port %d\n",
1709 dev_dbg(dev, "%s: wwpn%d=%016llx\n", __func__, k, wwpn[k]);
1713 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1718 * init_pcr() - initialize the provisioning and control registers
1719 * @cfg: Internal structure associated with the host.
1721 * Also sets up fast access to the mapped registers and initializes AFU
1722 * command fields that never change.
1724 static void init_pcr(struct cxlflash_cfg *cfg)
1726 struct afu *afu = cfg->afu;
1727 struct sisl_ctrl_map __iomem *ctrl_map;
1731 for (i = 0; i < MAX_CONTEXT; i++) {
1732 ctrl_map = &afu->afu_map->ctrls[i].ctrl;
1733 /* Disrupt any clients that could be running */
1734 /* e.g. clients that survived a master restart */
1735 writeq_be(0, &ctrl_map->rht_start);
1736 writeq_be(0, &ctrl_map->rht_cnt_id);
1737 writeq_be(0, &ctrl_map->ctx_cap);
1740 /* Copy frequently used fields into hwq */
1741 for (i = 0; i < afu->num_hwqs; i++) {
1742 hwq = get_hwq(afu, i);
1744 hwq->ctx_hndl = (u16) cxl_process_element(hwq->ctx);
1745 hwq->host_map = &afu->afu_map->hosts[hwq->ctx_hndl].host;
1746 hwq->ctrl_map = &afu->afu_map->ctrls[hwq->ctx_hndl].ctrl;
1748 /* Program the Endian Control for the master context */
1749 writeq_be(SISL_ENDIAN_CTRL, &hwq->host_map->endian_ctrl);
1754 * init_global() - initialize AFU global registers
1755 * @cfg: Internal structure associated with the host.
1757 static int init_global(struct cxlflash_cfg *cfg)
1759 struct afu *afu = cfg->afu;
1760 struct device *dev = &cfg->dev->dev;
1762 struct sisl_host_map __iomem *hmap;
1763 __be64 __iomem *fc_port_regs;
1764 u64 wwpn[MAX_FC_PORTS]; /* wwpn of AFU ports */
1765 int i = 0, num_ports = 0;
1769 rc = read_vpd(cfg, &wwpn[0]);
1771 dev_err(dev, "%s: could not read vpd rc=%d\n", __func__, rc);
1775 /* Set up RRQ and SQ in HWQ for master issued cmds */
1776 for (i = 0; i < afu->num_hwqs; i++) {
1777 hwq = get_hwq(afu, i);
1778 hmap = hwq->host_map;
1780 writeq_be((u64) hwq->hrrq_start, &hmap->rrq_start);
1781 writeq_be((u64) hwq->hrrq_end, &hmap->rrq_end);
1783 if (afu_is_sq_cmd_mode(afu)) {
1784 writeq_be((u64)hwq->hsq_start, &hmap->sq_start);
1785 writeq_be((u64)hwq->hsq_end, &hmap->sq_end);
1789 /* AFU configuration */
1790 reg = readq_be(&afu->afu_map->global.regs.afu_config);
1791 reg |= SISL_AFUCONF_AR_ALL|SISL_AFUCONF_ENDIAN;
1792 /* enable all auto retry options and control endianness */
1793 /* leave others at default: */
1794 /* CTX_CAP write protected, mbox_r does not clear on read and */
1795 /* checker on if dual afu */
1796 writeq_be(reg, &afu->afu_map->global.regs.afu_config);
1798 /* Global port select: select either port */
1799 if (afu->internal_lun) {
1800 /* Only use port 0 */
1801 writeq_be(PORT0, &afu->afu_map->global.regs.afu_port_sel);
1804 writeq_be(PORT_MASK(cfg->num_fc_ports),
1805 &afu->afu_map->global.regs.afu_port_sel);
1806 num_ports = cfg->num_fc_ports;
1809 for (i = 0; i < num_ports; i++) {
1810 fc_port_regs = get_fc_port_regs(cfg, i);
1812 /* Unmask all errors (but they are still masked at AFU) */
1813 writeq_be(0, &fc_port_regs[FC_ERRMSK / 8]);
1814 /* Clear CRC error cnt & set a threshold */
1815 (void)readq_be(&fc_port_regs[FC_CNT_CRCERR / 8]);
1816 writeq_be(MC_CRC_THRESH, &fc_port_regs[FC_CRC_THRESH / 8]);
1818 /* Set WWPNs. If already programmed, wwpn[i] is 0 */
1820 afu_set_wwpn(afu, i, &fc_port_regs[0], wwpn[i]);
1821 /* Programming WWPN back to back causes additional
1822 * offline/online transitions and a PLOGI
1827 /* Set up master's own CTX_CAP to allow real mode, host translation */
1828 /* tables, afu cmds and read/write GSCSI cmds. */
1829 /* First, unlock ctx_cap write by reading mbox */
1830 for (i = 0; i < afu->num_hwqs; i++) {
1831 hwq = get_hwq(afu, i);
1833 (void)readq_be(&hwq->ctrl_map->mbox_r); /* unlock ctx_cap */
1834 writeq_be((SISL_CTX_CAP_REAL_MODE | SISL_CTX_CAP_HOST_XLATE |
1835 SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD |
1836 SISL_CTX_CAP_AFU_CMD | SISL_CTX_CAP_GSCSI_CMD),
1837 &hwq->ctrl_map->ctx_cap);
1841 * Determine write-same unmap support for host by evaluating the unmap
1842 * sector support bit of the context control register associated with
1843 * the primary hardware queue. Note that while this status is reflected
1844 * in a context register, the outcome can be assumed to be host-wide.
1846 hwq = get_hwq(afu, PRIMARY_HWQ);
1847 reg = readq_be(&hwq->host_map->ctx_ctrl);
1848 if (reg & SISL_CTX_CTRL_UNMAP_SECTOR)
1849 cfg->ws_unmap = true;
1851 /* Initialize heartbeat */
1852 afu->hb = readq_be(&afu->afu_map->global.regs.afu_hb);
1858 * start_afu() - initializes and starts the AFU
1859 * @cfg: Internal structure associated with the host.
1861 static int start_afu(struct cxlflash_cfg *cfg)
1863 struct afu *afu = cfg->afu;
1864 struct device *dev = &cfg->dev->dev;
1871 /* Initialize each HWQ */
1872 for (i = 0; i < afu->num_hwqs; i++) {
1873 hwq = get_hwq(afu, i);
1875 /* After an AFU reset, RRQ entries are stale, clear them */
1876 memset(&hwq->rrq_entry, 0, sizeof(hwq->rrq_entry));
1878 /* Initialize RRQ pointers */
1879 hwq->hrrq_start = &hwq->rrq_entry[0];
1880 hwq->hrrq_end = &hwq->rrq_entry[NUM_RRQ_ENTRY - 1];
1881 hwq->hrrq_curr = hwq->hrrq_start;
1884 /* Initialize spin locks */
1885 spin_lock_init(&hwq->hrrq_slock);
1886 spin_lock_init(&hwq->hsq_slock);
1889 if (afu_is_sq_cmd_mode(afu)) {
1890 memset(&hwq->sq, 0, sizeof(hwq->sq));
1891 hwq->hsq_start = &hwq->sq[0];
1892 hwq->hsq_end = &hwq->sq[NUM_SQ_ENTRY - 1];
1893 hwq->hsq_curr = hwq->hsq_start;
1895 atomic_set(&hwq->hsq_credits, NUM_SQ_ENTRY - 1);
1898 /* Initialize IRQ poll */
1899 if (afu_is_irqpoll_enabled(afu))
1900 irq_poll_init(&hwq->irqpoll, afu->irqpoll_weight,
1905 rc = init_global(cfg);
1907 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1912 * init_intr() - setup interrupt handlers for the master context
1913 * @cfg: Internal structure associated with the host.
1914 * @hwq: Hardware queue to initialize.
1916 * Return: 0 on success, -errno on failure
1918 static enum undo_level init_intr(struct cxlflash_cfg *cfg,
1921 struct device *dev = &cfg->dev->dev;
1922 struct cxl_context *ctx = hwq->ctx;
1924 enum undo_level level = UNDO_NOOP;
1925 bool is_primary_hwq = (hwq->index == PRIMARY_HWQ);
1926 int num_irqs = is_primary_hwq ? 3 : 2;
1928 rc = cxl_allocate_afu_irqs(ctx, num_irqs);
1930 dev_err(dev, "%s: allocate_afu_irqs failed rc=%d\n",
1936 rc = cxl_map_afu_irq(ctx, 1, cxlflash_sync_err_irq, hwq,
1937 "SISL_MSI_SYNC_ERROR");
1938 if (unlikely(rc <= 0)) {
1939 dev_err(dev, "%s: SISL_MSI_SYNC_ERROR map failed\n", __func__);
1944 rc = cxl_map_afu_irq(ctx, 2, cxlflash_rrq_irq, hwq,
1945 "SISL_MSI_RRQ_UPDATED");
1946 if (unlikely(rc <= 0)) {
1947 dev_err(dev, "%s: SISL_MSI_RRQ_UPDATED map failed\n", __func__);
1952 /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
1953 if (!is_primary_hwq)
1956 rc = cxl_map_afu_irq(ctx, 3, cxlflash_async_err_irq, hwq,
1957 "SISL_MSI_ASYNC_ERROR");
1958 if (unlikely(rc <= 0)) {
1959 dev_err(dev, "%s: SISL_MSI_ASYNC_ERROR map failed\n", __func__);
1968 * init_mc() - create and register as the master context
1969 * @cfg: Internal structure associated with the host.
1970 * index: HWQ Index of the master context.
1972 * Return: 0 on success, -errno on failure
1974 static int init_mc(struct cxlflash_cfg *cfg, u32 index)
1976 struct cxl_context *ctx;
1977 struct device *dev = &cfg->dev->dev;
1978 struct hwq *hwq = get_hwq(cfg->afu, index);
1980 enum undo_level level;
1982 hwq->afu = cfg->afu;
1984 INIT_LIST_HEAD(&hwq->pending_cmds);
1986 if (index == PRIMARY_HWQ)
1987 ctx = cxl_get_context(cfg->dev);
1989 ctx = cxl_dev_context_init(cfg->dev);
1990 if (unlikely(!ctx)) {
1998 /* Set it up as a master with the CXL */
1999 cxl_set_master(ctx);
2001 /* Reset AFU when initializing primary context */
2002 if (index == PRIMARY_HWQ) {
2003 rc = cxl_afu_reset(ctx);
2005 dev_err(dev, "%s: AFU reset failed rc=%d\n",
2011 level = init_intr(cfg, hwq);
2012 if (unlikely(level)) {
2013 dev_err(dev, "%s: interrupt init failed rc=%d\n", __func__, rc);
2017 /* This performs the equivalent of the CXL_IOCTL_START_WORK.
2018 * The CXL_IOCTL_GET_PROCESS_ELEMENT is implicit in the process
2019 * element (pe) that is embedded in the context (ctx)
2021 rc = start_context(cfg, index);
2023 dev_err(dev, "%s: start context failed rc=%d\n", __func__, rc);
2024 level = UNMAP_THREE;
2029 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2032 term_intr(cfg, level, index);
2033 if (index != PRIMARY_HWQ)
2034 cxl_release_context(ctx);
2041 * get_num_afu_ports() - determines and configures the number of AFU ports
2042 * @cfg: Internal structure associated with the host.
2044 * This routine determines the number of AFU ports by converting the global
2045 * port selection mask. The converted value is only valid following an AFU
2046 * reset (explicit or power-on). This routine must be invoked shortly after
2047 * mapping as other routines are dependent on the number of ports during the
2048 * initialization sequence.
2050 * To support legacy AFUs that might not have reflected an initial global
2051 * port mask (value read is 0), default to the number of ports originally
2052 * supported by the cxlflash driver (2) before hardware with other port
2053 * offerings was introduced.
2055 static void get_num_afu_ports(struct cxlflash_cfg *cfg)
2057 struct afu *afu = cfg->afu;
2058 struct device *dev = &cfg->dev->dev;
2060 int num_fc_ports = LEGACY_FC_PORTS;
2062 port_mask = readq_be(&afu->afu_map->global.regs.afu_port_sel);
2063 if (port_mask != 0ULL)
2064 num_fc_ports = min(ilog2(port_mask) + 1, MAX_FC_PORTS);
2066 dev_dbg(dev, "%s: port_mask=%016llx num_fc_ports=%d\n",
2067 __func__, port_mask, num_fc_ports);
2069 cfg->num_fc_ports = num_fc_ports;
2070 cfg->host->max_channel = PORTNUM2CHAN(num_fc_ports);
2074 * init_afu() - setup as master context and start AFU
2075 * @cfg: Internal structure associated with the host.
2077 * This routine is a higher level of control for configuring the
2078 * AFU on probe and reset paths.
2080 * Return: 0 on success, -errno on failure
2082 static int init_afu(struct cxlflash_cfg *cfg)
2086 struct afu *afu = cfg->afu;
2087 struct device *dev = &cfg->dev->dev;
2091 cxl_perst_reloads_same_image(cfg->cxl_afu, true);
2093 afu->num_hwqs = afu->desired_hwqs;
2094 for (i = 0; i < afu->num_hwqs; i++) {
2095 rc = init_mc(cfg, i);
2097 dev_err(dev, "%s: init_mc failed rc=%d index=%d\n",
2103 /* Map the entire MMIO space of the AFU using the first context */
2104 hwq = get_hwq(afu, PRIMARY_HWQ);
2105 afu->afu_map = cxl_psa_map(hwq->ctx);
2106 if (!afu->afu_map) {
2107 dev_err(dev, "%s: cxl_psa_map failed\n", __func__);
2112 /* No byte reverse on reading afu_version or string will be backwards */
2113 reg = readq(&afu->afu_map->global.regs.afu_version);
2114 memcpy(afu->version, ®, sizeof(reg));
2115 afu->interface_version =
2116 readq_be(&afu->afu_map->global.regs.interface_version);
2117 if ((afu->interface_version + 1) == 0) {
2118 dev_err(dev, "Back level AFU, please upgrade. AFU version %s "
2119 "interface version %016llx\n", afu->version,
2120 afu->interface_version);
2125 if (afu_is_sq_cmd_mode(afu)) {
2126 afu->send_cmd = send_cmd_sq;
2127 afu->context_reset = context_reset_sq;
2129 afu->send_cmd = send_cmd_ioarrin;
2130 afu->context_reset = context_reset_ioarrin;
2133 dev_dbg(dev, "%s: afu_ver=%s interface_ver=%016llx\n", __func__,
2134 afu->version, afu->interface_version);
2136 get_num_afu_ports(cfg);
2138 rc = start_afu(cfg);
2140 dev_err(dev, "%s: start_afu failed, rc=%d\n", __func__, rc);
2144 afu_err_intr_init(cfg->afu);
2145 for (i = 0; i < afu->num_hwqs; i++) {
2146 hwq = get_hwq(afu, i);
2148 hwq->room = readq_be(&hwq->host_map->cmd_room);
2151 /* Restore the LUN mappings */
2152 cxlflash_restore_luntable(cfg);
2154 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2158 for (i = afu->num_hwqs - 1; i >= 0; i--) {
2159 term_intr(cfg, UNMAP_THREE, i);
2166 * afu_reset() - resets the AFU
2167 * @cfg: Internal structure associated with the host.
2169 * Return: 0 on success, -errno on failure
2171 static int afu_reset(struct cxlflash_cfg *cfg)
2173 struct device *dev = &cfg->dev->dev;
2176 /* Stop the context before the reset. Since the context is
2177 * no longer available restart it after the reset is complete
2183 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2188 * drain_ioctls() - wait until all currently executing ioctls have completed
2189 * @cfg: Internal structure associated with the host.
2191 * Obtain write access to read/write semaphore that wraps ioctl
2192 * handling to 'drain' ioctls currently executing.
2194 static void drain_ioctls(struct cxlflash_cfg *cfg)
2196 down_write(&cfg->ioctl_rwsem);
2197 up_write(&cfg->ioctl_rwsem);
2201 * cxlflash_async_reset_host() - asynchronous host reset handler
2202 * @data: Private data provided while scheduling reset.
2203 * @cookie: Cookie that can be used for checkpointing.
2205 static void cxlflash_async_reset_host(void *data, async_cookie_t cookie)
2207 struct cxlflash_cfg *cfg = data;
2208 struct device *dev = &cfg->dev->dev;
2211 if (cfg->state != STATE_RESET) {
2212 dev_dbg(dev, "%s: Not performing a reset, state=%d\n",
2213 __func__, cfg->state);
2218 cxlflash_mark_contexts_error(cfg);
2219 rc = afu_reset(cfg);
2221 cfg->state = STATE_FAILTERM;
2223 cfg->state = STATE_NORMAL;
2224 wake_up_all(&cfg->reset_waitq);
2227 scsi_unblock_requests(cfg->host);
2231 * cxlflash_schedule_async_reset() - schedule an asynchronous host reset
2232 * @cfg: Internal structure associated with the host.
2234 static void cxlflash_schedule_async_reset(struct cxlflash_cfg *cfg)
2236 struct device *dev = &cfg->dev->dev;
2238 if (cfg->state != STATE_NORMAL) {
2239 dev_dbg(dev, "%s: Not performing reset state=%d\n",
2240 __func__, cfg->state);
2244 cfg->state = STATE_RESET;
2245 scsi_block_requests(cfg->host);
2246 cfg->async_reset_cookie = async_schedule(cxlflash_async_reset_host,
2251 * send_afu_cmd() - builds and sends an internal AFU command
2252 * @afu: AFU associated with the host.
2253 * @rcb: Pre-populated IOARCB describing command to send.
2255 * The AFU can only take one internal AFU command at a time. This limitation is
2256 * enforced by using a mutex to provide exclusive access to the AFU during the
2257 * operation. This design point requires calling threads to not be on interrupt
2258 * context due to the possibility of sleeping during concurrent AFU operations.
2260 * The command status is optionally passed back to the caller when the caller
2261 * populates the IOASA field of the IOARCB with a pointer to an IOASA structure.
2264 * 0 on success, -errno on failure
2266 static int send_afu_cmd(struct afu *afu, struct sisl_ioarcb *rcb)
2268 struct cxlflash_cfg *cfg = afu->parent;
2269 struct device *dev = &cfg->dev->dev;
2270 struct afu_cmd *cmd = NULL;
2271 struct hwq *hwq = get_hwq(afu, PRIMARY_HWQ);
2275 static DEFINE_MUTEX(sync_active);
2277 if (cfg->state != STATE_NORMAL) {
2278 dev_dbg(dev, "%s: Sync not required state=%u\n",
2279 __func__, cfg->state);
2283 mutex_lock(&sync_active);
2284 atomic_inc(&afu->cmds_active);
2285 buf = kmalloc(sizeof(*cmd) + __alignof__(*cmd) - 1, GFP_KERNEL);
2286 if (unlikely(!buf)) {
2287 dev_err(dev, "%s: no memory for command\n", __func__);
2292 cmd = (struct afu_cmd *)PTR_ALIGN(buf, __alignof__(*cmd));
2295 memset(cmd, 0, sizeof(*cmd));
2296 memcpy(&cmd->rcb, rcb, sizeof(*rcb));
2297 INIT_LIST_HEAD(&cmd->queue);
2298 init_completion(&cmd->cevent);
2300 cmd->hwq_index = hwq->index;
2301 cmd->rcb.ctx_id = hwq->ctx_hndl;
2303 dev_dbg(dev, "%s: afu=%p cmd=%p type=%02x nretry=%d\n",
2304 __func__, afu, cmd, cmd->rcb.cdb[0], nretry);
2306 rc = afu->send_cmd(afu, cmd);
2312 rc = wait_resp(afu, cmd);
2315 rc = afu->context_reset(hwq);
2317 cxlflash_schedule_async_reset(cfg);
2320 /* fall through to retry */
2324 /* fall through to exit */
2330 *rcb->ioasa = cmd->sa;
2332 atomic_dec(&afu->cmds_active);
2333 mutex_unlock(&sync_active);
2335 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2340 * cxlflash_afu_sync() - builds and sends an AFU sync command
2341 * @afu: AFU associated with the host.
2342 * @ctx: Identifies context requesting sync.
2343 * @res: Identifies resource requesting sync.
2344 * @mode: Type of sync to issue (lightweight, heavyweight, global).
2346 * AFU sync operations are only necessary and allowed when the device is
2347 * operating normally. When not operating normally, sync requests can occur as
2348 * part of cleaning up resources associated with an adapter prior to removal.
2349 * In this scenario, these requests are simply ignored (safe due to the AFU
2353 * 0 on success, -errno on failure
2355 int cxlflash_afu_sync(struct afu *afu, ctx_hndl_t ctx, res_hndl_t res, u8 mode)
2357 struct cxlflash_cfg *cfg = afu->parent;
2358 struct device *dev = &cfg->dev->dev;
2359 struct sisl_ioarcb rcb = { 0 };
2361 dev_dbg(dev, "%s: afu=%p ctx=%u res=%u mode=%u\n",
2362 __func__, afu, ctx, res, mode);
2364 rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
2365 rcb.msi = SISL_MSI_RRQ_UPDATED;
2366 rcb.timeout = MC_AFU_SYNC_TIMEOUT;
2368 rcb.cdb[0] = SISL_AFU_CMD_SYNC;
2370 put_unaligned_be16(ctx, &rcb.cdb[2]);
2371 put_unaligned_be32(res, &rcb.cdb[4]);
2373 return send_afu_cmd(afu, &rcb);
2377 * cxlflash_eh_abort_handler() - abort a SCSI command
2378 * @scp: SCSI command to abort.
2380 * CXL Flash devices do not support a single command abort. Reset the context
2381 * as per SISLite specification. Flush any pending commands in the hardware
2382 * queue before the reset.
2384 * Return: SUCCESS/FAILED as defined in scsi/scsi.h
2386 static int cxlflash_eh_abort_handler(struct scsi_cmnd *scp)
2389 struct Scsi_Host *host = scp->device->host;
2390 struct cxlflash_cfg *cfg = shost_priv(host);
2391 struct afu_cmd *cmd = sc_to_afuc(scp);
2392 struct device *dev = &cfg->dev->dev;
2393 struct afu *afu = cfg->afu;
2394 struct hwq *hwq = get_hwq(afu, cmd->hwq_index);
2396 dev_dbg(dev, "%s: (scp=%p) %d/%d/%d/%llu "
2397 "cdb=(%08x-%08x-%08x-%08x)\n", __func__, scp, host->host_no,
2398 scp->device->channel, scp->device->id, scp->device->lun,
2399 get_unaligned_be32(&((u32 *)scp->cmnd)[0]),
2400 get_unaligned_be32(&((u32 *)scp->cmnd)[1]),
2401 get_unaligned_be32(&((u32 *)scp->cmnd)[2]),
2402 get_unaligned_be32(&((u32 *)scp->cmnd)[3]));
2404 /* When the state is not normal, another reset/reload is in progress.
2405 * Return failed and the mid-layer will invoke host reset handler.
2407 if (cfg->state != STATE_NORMAL) {
2408 dev_dbg(dev, "%s: Invalid state for abort, state=%d\n",
2409 __func__, cfg->state);
2413 rc = afu->context_reset(hwq);
2420 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2425 * cxlflash_eh_device_reset_handler() - reset a single LUN
2426 * @scp: SCSI command to send.
2429 * SUCCESS as defined in scsi/scsi.h
2430 * FAILED as defined in scsi/scsi.h
2432 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd *scp)
2435 struct scsi_device *sdev = scp->device;
2436 struct Scsi_Host *host = sdev->host;
2437 struct cxlflash_cfg *cfg = shost_priv(host);
2438 struct device *dev = &cfg->dev->dev;
2441 dev_dbg(dev, "%s: %d/%d/%d/%llu\n", __func__,
2442 host->host_no, sdev->channel, sdev->id, sdev->lun);
2444 switch (cfg->state) {
2446 rcr = send_tmf(cfg, sdev, TMF_LUN_RESET);
2451 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2458 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2463 * cxlflash_eh_host_reset_handler() - reset the host adapter
2464 * @scp: SCSI command from stack identifying host.
2466 * Following a reset, the state is evaluated again in case an EEH occurred
2467 * during the reset. In such a scenario, the host reset will either yield
2468 * until the EEH recovery is complete or return success or failure based
2469 * upon the current device state.
2472 * SUCCESS as defined in scsi/scsi.h
2473 * FAILED as defined in scsi/scsi.h
2475 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd *scp)
2479 struct Scsi_Host *host = scp->device->host;
2480 struct cxlflash_cfg *cfg = shost_priv(host);
2481 struct device *dev = &cfg->dev->dev;
2483 dev_dbg(dev, "%s: %d\n", __func__, host->host_no);
2485 switch (cfg->state) {
2487 cfg->state = STATE_RESET;
2489 cxlflash_mark_contexts_error(cfg);
2490 rcr = afu_reset(cfg);
2493 cfg->state = STATE_FAILTERM;
2495 cfg->state = STATE_NORMAL;
2496 wake_up_all(&cfg->reset_waitq);
2500 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2501 if (cfg->state == STATE_NORMAL)
2509 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
2514 * cxlflash_change_queue_depth() - change the queue depth for the device
2515 * @sdev: SCSI device destined for queue depth change.
2516 * @qdepth: Requested queue depth value to set.
2518 * The requested queue depth is capped to the maximum supported value.
2520 * Return: The actual queue depth set.
2522 static int cxlflash_change_queue_depth(struct scsi_device *sdev, int qdepth)
2525 if (qdepth > CXLFLASH_MAX_CMDS_PER_LUN)
2526 qdepth = CXLFLASH_MAX_CMDS_PER_LUN;
2528 scsi_change_queue_depth(sdev, qdepth);
2529 return sdev->queue_depth;
2533 * cxlflash_show_port_status() - queries and presents the current port status
2534 * @port: Desired port for status reporting.
2535 * @cfg: Internal structure associated with the host.
2536 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2538 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2540 static ssize_t cxlflash_show_port_status(u32 port,
2541 struct cxlflash_cfg *cfg,
2544 struct device *dev = &cfg->dev->dev;
2547 __be64 __iomem *fc_port_regs;
2549 WARN_ON(port >= MAX_FC_PORTS);
2551 if (port >= cfg->num_fc_ports) {
2552 dev_info(dev, "%s: Port %d not supported on this card.\n",
2557 fc_port_regs = get_fc_port_regs(cfg, port);
2558 status = readq_be(&fc_port_regs[FC_MTIP_STATUS / 8]);
2559 status &= FC_MTIP_STATUS_MASK;
2561 if (status == FC_MTIP_STATUS_ONLINE)
2562 disp_status = "online";
2563 else if (status == FC_MTIP_STATUS_OFFLINE)
2564 disp_status = "offline";
2566 disp_status = "unknown";
2568 return scnprintf(buf, PAGE_SIZE, "%s\n", disp_status);
2572 * port0_show() - queries and presents the current status of port 0
2573 * @dev: Generic device associated with the host owning the port.
2574 * @attr: Device attribute representing the port.
2575 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2577 * Return: The size of the ASCII string returned in @buf.
2579 static ssize_t port0_show(struct device *dev,
2580 struct device_attribute *attr,
2583 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2585 return cxlflash_show_port_status(0, cfg, buf);
2589 * port1_show() - queries and presents the current status of port 1
2590 * @dev: Generic device associated with the host owning the port.
2591 * @attr: Device attribute representing the port.
2592 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2594 * Return: The size of the ASCII string returned in @buf.
2596 static ssize_t port1_show(struct device *dev,
2597 struct device_attribute *attr,
2600 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2602 return cxlflash_show_port_status(1, cfg, buf);
2606 * port2_show() - queries and presents the current status of port 2
2607 * @dev: Generic device associated with the host owning the port.
2608 * @attr: Device attribute representing the port.
2609 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2611 * Return: The size of the ASCII string returned in @buf.
2613 static ssize_t port2_show(struct device *dev,
2614 struct device_attribute *attr,
2617 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2619 return cxlflash_show_port_status(2, cfg, buf);
2623 * port3_show() - queries and presents the current status of port 3
2624 * @dev: Generic device associated with the host owning the port.
2625 * @attr: Device attribute representing the port.
2626 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2628 * Return: The size of the ASCII string returned in @buf.
2630 static ssize_t port3_show(struct device *dev,
2631 struct device_attribute *attr,
2634 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2636 return cxlflash_show_port_status(3, cfg, buf);
2640 * lun_mode_show() - presents the current LUN mode of the host
2641 * @dev: Generic device associated with the host.
2642 * @attr: Device attribute representing the LUN mode.
2643 * @buf: Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2645 * Return: The size of the ASCII string returned in @buf.
2647 static ssize_t lun_mode_show(struct device *dev,
2648 struct device_attribute *attr, char *buf)
2650 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2651 struct afu *afu = cfg->afu;
2653 return scnprintf(buf, PAGE_SIZE, "%u\n", afu->internal_lun);
2657 * lun_mode_store() - sets the LUN mode of the host
2658 * @dev: Generic device associated with the host.
2659 * @attr: Device attribute representing the LUN mode.
2660 * @buf: Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2661 * @count: Length of data resizing in @buf.
2663 * The CXL Flash AFU supports a dummy LUN mode where the external
2664 * links and storage are not required. Space on the FPGA is used
2665 * to create 1 or 2 small LUNs which are presented to the system
2666 * as if they were a normal storage device. This feature is useful
2667 * during development and also provides manufacturing with a way
2668 * to test the AFU without an actual device.
2670 * 0 = external LUN[s] (default)
2671 * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2672 * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2673 * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2674 * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2676 * Return: The size of the ASCII string returned in @buf.
2678 static ssize_t lun_mode_store(struct device *dev,
2679 struct device_attribute *attr,
2680 const char *buf, size_t count)
2682 struct Scsi_Host *shost = class_to_shost(dev);
2683 struct cxlflash_cfg *cfg = shost_priv(shost);
2684 struct afu *afu = cfg->afu;
2688 rc = kstrtouint(buf, 10, &lun_mode);
2689 if (!rc && (lun_mode < 5) && (lun_mode != afu->internal_lun)) {
2690 afu->internal_lun = lun_mode;
2693 * When configured for internal LUN, there is only one channel,
2694 * channel number 0, else there will be one less than the number
2695 * of fc ports for this card.
2697 if (afu->internal_lun)
2698 shost->max_channel = 0;
2700 shost->max_channel = PORTNUM2CHAN(cfg->num_fc_ports);
2703 scsi_scan_host(cfg->host);
2710 * ioctl_version_show() - presents the current ioctl version of the host
2711 * @dev: Generic device associated with the host.
2712 * @attr: Device attribute representing the ioctl version.
2713 * @buf: Buffer of length PAGE_SIZE to report back the ioctl version.
2715 * Return: The size of the ASCII string returned in @buf.
2717 static ssize_t ioctl_version_show(struct device *dev,
2718 struct device_attribute *attr, char *buf)
2722 bytes = scnprintf(buf, PAGE_SIZE,
2723 "disk: %u\n", DK_CXLFLASH_VERSION_0);
2724 bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2725 "host: %u\n", HT_CXLFLASH_VERSION_0);
2731 * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2732 * @port: Desired port for status reporting.
2733 * @cfg: Internal structure associated with the host.
2734 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2736 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2738 static ssize_t cxlflash_show_port_lun_table(u32 port,
2739 struct cxlflash_cfg *cfg,
2742 struct device *dev = &cfg->dev->dev;
2743 __be64 __iomem *fc_port_luns;
2747 WARN_ON(port >= MAX_FC_PORTS);
2749 if (port >= cfg->num_fc_ports) {
2750 dev_info(dev, "%s: Port %d not supported on this card.\n",
2755 fc_port_luns = get_fc_port_luns(cfg, port);
2757 for (i = 0; i < CXLFLASH_NUM_VLUNS; i++)
2758 bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
2760 i, readq_be(&fc_port_luns[i]));
2765 * port0_lun_table_show() - presents the current LUN table of port 0
2766 * @dev: Generic device associated with the host owning the port.
2767 * @attr: Device attribute representing the port.
2768 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2770 * Return: The size of the ASCII string returned in @buf.
2772 static ssize_t port0_lun_table_show(struct device *dev,
2773 struct device_attribute *attr,
2776 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2778 return cxlflash_show_port_lun_table(0, cfg, buf);
2782 * port1_lun_table_show() - presents the current LUN table of port 1
2783 * @dev: Generic device associated with the host owning the port.
2784 * @attr: Device attribute representing the port.
2785 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2787 * Return: The size of the ASCII string returned in @buf.
2789 static ssize_t port1_lun_table_show(struct device *dev,
2790 struct device_attribute *attr,
2793 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2795 return cxlflash_show_port_lun_table(1, cfg, buf);
2799 * port2_lun_table_show() - presents the current LUN table of port 2
2800 * @dev: Generic device associated with the host owning the port.
2801 * @attr: Device attribute representing the port.
2802 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2804 * Return: The size of the ASCII string returned in @buf.
2806 static ssize_t port2_lun_table_show(struct device *dev,
2807 struct device_attribute *attr,
2810 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2812 return cxlflash_show_port_lun_table(2, cfg, buf);
2816 * port3_lun_table_show() - presents the current LUN table of port 3
2817 * @dev: Generic device associated with the host owning the port.
2818 * @attr: Device attribute representing the port.
2819 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2821 * Return: The size of the ASCII string returned in @buf.
2823 static ssize_t port3_lun_table_show(struct device *dev,
2824 struct device_attribute *attr,
2827 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2829 return cxlflash_show_port_lun_table(3, cfg, buf);
2833 * irqpoll_weight_show() - presents the current IRQ poll weight for the host
2834 * @dev: Generic device associated with the host.
2835 * @attr: Device attribute representing the IRQ poll weight.
2836 * @buf: Buffer of length PAGE_SIZE to report back the current IRQ poll
2839 * An IRQ poll weight of 0 indicates polling is disabled.
2841 * Return: The size of the ASCII string returned in @buf.
2843 static ssize_t irqpoll_weight_show(struct device *dev,
2844 struct device_attribute *attr, char *buf)
2846 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2847 struct afu *afu = cfg->afu;
2849 return scnprintf(buf, PAGE_SIZE, "%u\n", afu->irqpoll_weight);
2853 * irqpoll_weight_store() - sets the current IRQ poll weight for the host
2854 * @dev: Generic device associated with the host.
2855 * @attr: Device attribute representing the IRQ poll weight.
2856 * @buf: Buffer of length PAGE_SIZE containing the desired IRQ poll
2858 * @count: Length of data resizing in @buf.
2860 * An IRQ poll weight of 0 indicates polling is disabled.
2862 * Return: The size of the ASCII string returned in @buf.
2864 static ssize_t irqpoll_weight_store(struct device *dev,
2865 struct device_attribute *attr,
2866 const char *buf, size_t count)
2868 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2869 struct device *cfgdev = &cfg->dev->dev;
2870 struct afu *afu = cfg->afu;
2875 rc = kstrtouint(buf, 10, &weight);
2881 "Invalid IRQ poll weight. It must be 256 or less.\n");
2885 if (weight == afu->irqpoll_weight) {
2887 "Current IRQ poll weight has the same weight.\n");
2891 if (afu_is_irqpoll_enabled(afu)) {
2892 for (i = 0; i < afu->num_hwqs; i++) {
2893 hwq = get_hwq(afu, i);
2895 irq_poll_disable(&hwq->irqpoll);
2899 afu->irqpoll_weight = weight;
2902 for (i = 0; i < afu->num_hwqs; i++) {
2903 hwq = get_hwq(afu, i);
2905 irq_poll_init(&hwq->irqpoll, weight, cxlflash_irqpoll);
2913 * num_hwqs_show() - presents the number of hardware queues for the host
2914 * @dev: Generic device associated with the host.
2915 * @attr: Device attribute representing the number of hardware queues.
2916 * @buf: Buffer of length PAGE_SIZE to report back the number of hardware
2919 * Return: The size of the ASCII string returned in @buf.
2921 static ssize_t num_hwqs_show(struct device *dev,
2922 struct device_attribute *attr, char *buf)
2924 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2925 struct afu *afu = cfg->afu;
2927 return scnprintf(buf, PAGE_SIZE, "%u\n", afu->num_hwqs);
2931 * num_hwqs_store() - sets the number of hardware queues for the host
2932 * @dev: Generic device associated with the host.
2933 * @attr: Device attribute representing the number of hardware queues.
2934 * @buf: Buffer of length PAGE_SIZE containing the number of hardware
2936 * @count: Length of data resizing in @buf.
2938 * n > 0: num_hwqs = n
2939 * n = 0: num_hwqs = num_online_cpus()
2940 * n < 0: num_online_cpus() / abs(n)
2942 * Return: The size of the ASCII string returned in @buf.
2944 static ssize_t num_hwqs_store(struct device *dev,
2945 struct device_attribute *attr,
2946 const char *buf, size_t count)
2948 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
2949 struct afu *afu = cfg->afu;
2951 int nhwqs, num_hwqs;
2953 rc = kstrtoint(buf, 10, &nhwqs);
2959 else if (nhwqs == 0)
2960 num_hwqs = num_online_cpus();
2962 num_hwqs = num_online_cpus() / abs(nhwqs);
2964 afu->desired_hwqs = min(num_hwqs, CXLFLASH_MAX_HWQS);
2965 WARN_ON_ONCE(afu->desired_hwqs == 0);
2968 switch (cfg->state) {
2970 cfg->state = STATE_RESET;
2972 cxlflash_mark_contexts_error(cfg);
2973 rc = afu_reset(cfg);
2975 cfg->state = STATE_FAILTERM;
2977 cfg->state = STATE_NORMAL;
2978 wake_up_all(&cfg->reset_waitq);
2981 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET);
2982 if (cfg->state == STATE_NORMAL)
2985 /* Ideally should not happen */
2986 dev_err(dev, "%s: Device is not ready, state=%d\n",
2987 __func__, cfg->state);
2994 static const char *hwq_mode_name[MAX_HWQ_MODE] = { "rr", "tag", "cpu" };
2997 * hwq_mode_show() - presents the HWQ steering mode for the host
2998 * @dev: Generic device associated with the host.
2999 * @attr: Device attribute representing the HWQ steering mode.
3000 * @buf: Buffer of length PAGE_SIZE to report back the HWQ steering mode
3001 * as a character string.
3003 * Return: The size of the ASCII string returned in @buf.
3005 static ssize_t hwq_mode_show(struct device *dev,
3006 struct device_attribute *attr, char *buf)
3008 struct cxlflash_cfg *cfg = shost_priv(class_to_shost(dev));
3009 struct afu *afu = cfg->afu;
3011 return scnprintf(buf, PAGE_SIZE, "%s\n", hwq_mode_name[afu->hwq_mode]);
3015 * hwq_mode_store() - sets the HWQ steering mode for the host
3016 * @dev: Generic device associated with the host.
3017 * @attr: Device attribute representing the HWQ steering mode.
3018 * @buf: Buffer of length PAGE_SIZE containing the HWQ steering mode
3019 * as a character string.
3020 * @count: Length of data resizing in @buf.
3023 * tag = Block MQ Tagging
3024 * cpu = CPU Affinity
3026 * Return: The size of the ASCII string returned in @buf.
3028 static ssize_t hwq_mode_store(struct device *dev,
3029 struct device_attribute *attr,
3030 const char *buf, size_t count)
3032 struct Scsi_Host *shost = class_to_shost(dev);
3033 struct cxlflash_cfg *cfg = shost_priv(shost);
3034 struct device *cfgdev = &cfg->dev->dev;
3035 struct afu *afu = cfg->afu;
3037 u32 mode = MAX_HWQ_MODE;
3039 for (i = 0; i < MAX_HWQ_MODE; i++) {
3040 if (!strncmp(hwq_mode_name[i], buf, strlen(hwq_mode_name[i]))) {
3046 if (mode >= MAX_HWQ_MODE) {
3047 dev_info(cfgdev, "Invalid HWQ steering mode.\n");
3051 if ((mode == HWQ_MODE_TAG) && !shost_use_blk_mq(shost)) {
3052 dev_info(cfgdev, "SCSI-MQ is not enabled, use a different "
3053 "HWQ steering mode.\n");
3057 afu->hwq_mode = mode;
3063 * mode_show() - presents the current mode of the device
3064 * @dev: Generic device associated with the device.
3065 * @attr: Device attribute representing the device mode.
3066 * @buf: Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
3068 * Return: The size of the ASCII string returned in @buf.
3070 static ssize_t mode_show(struct device *dev,
3071 struct device_attribute *attr, char *buf)
3073 struct scsi_device *sdev = to_scsi_device(dev);
3075 return scnprintf(buf, PAGE_SIZE, "%s\n",
3076 sdev->hostdata ? "superpipe" : "legacy");
3082 static DEVICE_ATTR_RO(port0);
3083 static DEVICE_ATTR_RO(port1);
3084 static DEVICE_ATTR_RO(port2);
3085 static DEVICE_ATTR_RO(port3);
3086 static DEVICE_ATTR_RW(lun_mode);
3087 static DEVICE_ATTR_RO(ioctl_version);
3088 static DEVICE_ATTR_RO(port0_lun_table);
3089 static DEVICE_ATTR_RO(port1_lun_table);
3090 static DEVICE_ATTR_RO(port2_lun_table);
3091 static DEVICE_ATTR_RO(port3_lun_table);
3092 static DEVICE_ATTR_RW(irqpoll_weight);
3093 static DEVICE_ATTR_RW(num_hwqs);
3094 static DEVICE_ATTR_RW(hwq_mode);
3096 static struct device_attribute *cxlflash_host_attrs[] = {
3102 &dev_attr_ioctl_version,
3103 &dev_attr_port0_lun_table,
3104 &dev_attr_port1_lun_table,
3105 &dev_attr_port2_lun_table,
3106 &dev_attr_port3_lun_table,
3107 &dev_attr_irqpoll_weight,
3116 static DEVICE_ATTR_RO(mode);
3118 static struct device_attribute *cxlflash_dev_attrs[] = {
3126 static struct scsi_host_template driver_template = {
3127 .module = THIS_MODULE,
3128 .name = CXLFLASH_ADAPTER_NAME,
3129 .info = cxlflash_driver_info,
3130 .ioctl = cxlflash_ioctl,
3131 .proc_name = CXLFLASH_NAME,
3132 .queuecommand = cxlflash_queuecommand,
3133 .eh_abort_handler = cxlflash_eh_abort_handler,
3134 .eh_device_reset_handler = cxlflash_eh_device_reset_handler,
3135 .eh_host_reset_handler = cxlflash_eh_host_reset_handler,
3136 .change_queue_depth = cxlflash_change_queue_depth,
3137 .cmd_per_lun = CXLFLASH_MAX_CMDS_PER_LUN,
3138 .can_queue = CXLFLASH_MAX_CMDS,
3139 .cmd_size = sizeof(struct afu_cmd) + __alignof__(struct afu_cmd) - 1,
3141 .sg_tablesize = 1, /* No scatter gather support */
3142 .max_sectors = CXLFLASH_MAX_SECTORS,
3143 .use_clustering = ENABLE_CLUSTERING,
3144 .shost_attrs = cxlflash_host_attrs,
3145 .sdev_attrs = cxlflash_dev_attrs,
3149 * Device dependent values
3151 static struct dev_dependent_vals dev_corsa_vals = { CXLFLASH_MAX_SECTORS,
3153 static struct dev_dependent_vals dev_flash_gt_vals = { CXLFLASH_MAX_SECTORS,
3154 CXLFLASH_NOTIFY_SHUTDOWN };
3155 static struct dev_dependent_vals dev_briard_vals = { CXLFLASH_MAX_SECTORS,
3156 CXLFLASH_NOTIFY_SHUTDOWN };
3159 * PCI device binding table
3161 static struct pci_device_id cxlflash_pci_table[] = {
3162 {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_CORSA,
3163 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_corsa_vals},
3164 {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_FLASH_GT,
3165 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_flash_gt_vals},
3166 {PCI_VENDOR_ID_IBM, PCI_DEVICE_ID_IBM_BRIARD,
3167 PCI_ANY_ID, PCI_ANY_ID, 0, 0, (kernel_ulong_t)&dev_briard_vals},
3171 MODULE_DEVICE_TABLE(pci, cxlflash_pci_table);
3174 * cxlflash_worker_thread() - work thread handler for the AFU
3175 * @work: Work structure contained within cxlflash associated with host.
3177 * Handles the following events:
3178 * - Link reset which cannot be performed on interrupt context due to
3179 * blocking up to a few seconds
3182 static void cxlflash_worker_thread(struct work_struct *work)
3184 struct cxlflash_cfg *cfg = container_of(work, struct cxlflash_cfg,
3186 struct afu *afu = cfg->afu;
3187 struct device *dev = &cfg->dev->dev;
3188 __be64 __iomem *fc_port_regs;
3192 /* Avoid MMIO if the device has failed */
3194 if (cfg->state != STATE_NORMAL)
3197 spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3199 if (cfg->lr_state == LINK_RESET_REQUIRED) {
3200 port = cfg->lr_port;
3202 dev_err(dev, "%s: invalid port index %d\n",
3205 spin_unlock_irqrestore(cfg->host->host_lock,
3208 /* The reset can block... */
3209 fc_port_regs = get_fc_port_regs(cfg, port);
3210 afu_link_reset(afu, port, fc_port_regs);
3211 spin_lock_irqsave(cfg->host->host_lock, lock_flags);
3214 cfg->lr_state = LINK_RESET_COMPLETE;
3217 spin_unlock_irqrestore(cfg->host->host_lock, lock_flags);
3219 if (atomic_dec_if_positive(&cfg->scan_host_needed) >= 0)
3220 scsi_scan_host(cfg->host);
3224 * cxlflash_chr_open() - character device open handler
3225 * @inode: Device inode associated with this character device.
3226 * @file: File pointer for this device.
3228 * Only users with admin privileges are allowed to open the character device.
3230 * Return: 0 on success, -errno on failure
3232 static int cxlflash_chr_open(struct inode *inode, struct file *file)
3234 struct cxlflash_cfg *cfg;
3236 if (!capable(CAP_SYS_ADMIN))
3239 cfg = container_of(inode->i_cdev, struct cxlflash_cfg, cdev);
3240 file->private_data = cfg;
3246 * decode_hioctl() - translates encoded host ioctl to easily identifiable string
3247 * @cmd: The host ioctl command to decode.
3249 * Return: A string identifying the decoded host ioctl.
3251 static char *decode_hioctl(int cmd)
3254 case HT_CXLFLASH_LUN_PROVISION:
3255 return __stringify_1(HT_CXLFLASH_LUN_PROVISION);
3262 * cxlflash_lun_provision() - host LUN provisioning handler
3263 * @cfg: Internal structure associated with the host.
3264 * @arg: Kernel copy of userspace ioctl data structure.
3266 * Return: 0 on success, -errno on failure
3268 static int cxlflash_lun_provision(struct cxlflash_cfg *cfg,
3269 struct ht_cxlflash_lun_provision *lunprov)
3271 struct afu *afu = cfg->afu;
3272 struct device *dev = &cfg->dev->dev;
3273 struct sisl_ioarcb rcb;
3274 struct sisl_ioasa asa;
3275 __be64 __iomem *fc_port_regs;
3276 u16 port = lunprov->port;
3277 u16 scmd = lunprov->hdr.subcmd;
3284 if (!afu_is_lun_provision(afu)) {
3289 if (port >= cfg->num_fc_ports) {
3295 case HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN:
3296 type = SISL_AFU_LUN_PROVISION_CREATE;
3297 size = lunprov->size;
3300 case HT_CXLFLASH_LUN_PROVISION_SUBCMD_DELETE_LUN:
3301 type = SISL_AFU_LUN_PROVISION_DELETE;
3303 lun_id = lunprov->lun_id;
3305 case HT_CXLFLASH_LUN_PROVISION_SUBCMD_QUERY_PORT:
3306 fc_port_regs = get_fc_port_regs(cfg, port);
3308 reg = readq_be(&fc_port_regs[FC_MAX_NUM_LUNS / 8]);
3309 lunprov->max_num_luns = reg;
3310 reg = readq_be(&fc_port_regs[FC_CUR_NUM_LUNS / 8]);
3311 lunprov->cur_num_luns = reg;
3312 reg = readq_be(&fc_port_regs[FC_MAX_CAP_PORT / 8]);
3313 lunprov->max_cap_port = reg;
3314 reg = readq_be(&fc_port_regs[FC_CUR_CAP_PORT / 8]);
3315 lunprov->cur_cap_port = reg;
3323 memset(&rcb, 0, sizeof(rcb));
3324 memset(&asa, 0, sizeof(asa));
3325 rcb.req_flags = SISL_REQ_FLAGS_AFU_CMD;
3326 rcb.lun_id = lun_id;
3327 rcb.msi = SISL_MSI_RRQ_UPDATED;
3328 rcb.timeout = MC_LUN_PROV_TIMEOUT;
3331 rcb.cdb[0] = SISL_AFU_CMD_LUN_PROVISION;
3334 put_unaligned_be64(size, &rcb.cdb[8]);
3336 rc = send_afu_cmd(afu, &rcb);
3338 dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3339 __func__, rc, asa.ioasc, asa.afu_extra);
3343 if (scmd == HT_CXLFLASH_LUN_PROVISION_SUBCMD_CREATE_LUN) {
3344 lunprov->lun_id = (u64)asa.lunid_hi << 32 | asa.lunid_lo;
3345 memcpy(lunprov->wwid, asa.wwid, sizeof(lunprov->wwid));
3348 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3353 * cxlflash_afu_debug() - host AFU debug handler
3354 * @cfg: Internal structure associated with the host.
3355 * @arg: Kernel copy of userspace ioctl data structure.
3357 * For debug requests requiring a data buffer, always provide an aligned
3358 * (cache line) buffer to the AFU to appease any alignment requirements.
3360 * Return: 0 on success, -errno on failure
3362 static int cxlflash_afu_debug(struct cxlflash_cfg *cfg,
3363 struct ht_cxlflash_afu_debug *afu_dbg)
3365 struct afu *afu = cfg->afu;
3366 struct device *dev = &cfg->dev->dev;
3367 struct sisl_ioarcb rcb;
3368 struct sisl_ioasa asa;
3371 void __user *ubuf = (__force void __user *)afu_dbg->data_ea;
3372 u16 req_flags = SISL_REQ_FLAGS_AFU_CMD;
3373 u32 ulen = afu_dbg->data_len;
3374 bool is_write = afu_dbg->hdr.flags & HT_CXLFLASH_HOST_WRITE;
3377 if (!afu_is_afu_debug(afu)) {
3383 req_flags |= SISL_REQ_FLAGS_SUP_UNDERRUN;
3385 if (ulen > HT_CXLFLASH_AFU_DEBUG_MAX_DATA_LEN) {
3390 if (unlikely(!access_ok(is_write ? VERIFY_READ : VERIFY_WRITE,
3396 buf = kmalloc(ulen + cache_line_size() - 1, GFP_KERNEL);
3397 if (unlikely(!buf)) {
3402 kbuf = PTR_ALIGN(buf, cache_line_size());
3405 req_flags |= SISL_REQ_FLAGS_HOST_WRITE;
3407 if (copy_from_user(kbuf, ubuf, ulen)) {
3414 memset(&rcb, 0, sizeof(rcb));
3415 memset(&asa, 0, sizeof(asa));
3417 rcb.req_flags = req_flags;
3418 rcb.msi = SISL_MSI_RRQ_UPDATED;
3419 rcb.timeout = MC_AFU_DEBUG_TIMEOUT;
3423 rcb.data_len = ulen;
3424 rcb.data_ea = (uintptr_t)kbuf;
3427 rcb.cdb[0] = SISL_AFU_CMD_DEBUG;
3428 memcpy(&rcb.cdb[4], afu_dbg->afu_subcmd,
3429 HT_CXLFLASH_AFU_DEBUG_SUBCMD_LEN);
3431 rc = send_afu_cmd(afu, &rcb);
3433 dev_err(dev, "%s: send_afu_cmd failed rc=%d asc=%08x afux=%x\n",
3434 __func__, rc, asa.ioasc, asa.afu_extra);
3438 if (ulen && !is_write) {
3439 if (copy_to_user(ubuf, kbuf, ulen))
3444 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3449 * cxlflash_chr_ioctl() - character device IOCTL handler
3450 * @file: File pointer for this device.
3451 * @cmd: IOCTL command.
3452 * @arg: Userspace ioctl data structure.
3454 * A read/write semaphore is used to implement a 'drain' of currently
3455 * running ioctls. The read semaphore is taken at the beginning of each
3456 * ioctl thread and released upon concluding execution. Additionally the
3457 * semaphore should be released and then reacquired in any ioctl execution
3458 * path which will wait for an event to occur that is outside the scope of
3459 * the ioctl (i.e. an adapter reset). To drain the ioctls currently running,
3460 * a thread simply needs to acquire the write semaphore.
3462 * Return: 0 on success, -errno on failure
3464 static long cxlflash_chr_ioctl(struct file *file, unsigned int cmd,
3467 typedef int (*hioctl) (struct cxlflash_cfg *, void *);
3469 struct cxlflash_cfg *cfg = file->private_data;
3470 struct device *dev = &cfg->dev->dev;
3471 char buf[sizeof(union cxlflash_ht_ioctls)];
3472 void __user *uarg = (void __user *)arg;
3473 struct ht_cxlflash_hdr *hdr;
3475 bool known_ioctl = false;
3478 hioctl do_ioctl = NULL;
3480 static const struct {
3483 } ioctl_tbl[] = { /* NOTE: order matters here */
3484 { sizeof(struct ht_cxlflash_lun_provision),
3485 (hioctl)cxlflash_lun_provision },
3486 { sizeof(struct ht_cxlflash_afu_debug),
3487 (hioctl)cxlflash_afu_debug },
3490 /* Hold read semaphore so we can drain if needed */
3491 down_read(&cfg->ioctl_rwsem);
3493 dev_dbg(dev, "%s: cmd=%u idx=%d tbl_size=%lu\n",
3494 __func__, cmd, idx, sizeof(ioctl_tbl));
3497 case HT_CXLFLASH_LUN_PROVISION:
3498 case HT_CXLFLASH_AFU_DEBUG:
3500 idx = _IOC_NR(HT_CXLFLASH_LUN_PROVISION) - _IOC_NR(cmd);
3501 size = ioctl_tbl[idx].size;
3502 do_ioctl = ioctl_tbl[idx].ioctl;
3504 if (likely(do_ioctl))
3513 if (unlikely(copy_from_user(&buf, uarg, size))) {
3514 dev_err(dev, "%s: copy_from_user() fail "
3515 "size=%lu cmd=%d (%s) uarg=%p\n",
3516 __func__, size, cmd, decode_hioctl(cmd), uarg);
3521 hdr = (struct ht_cxlflash_hdr *)&buf;
3522 if (hdr->version != HT_CXLFLASH_VERSION_0) {
3523 dev_dbg(dev, "%s: Version %u not supported for %s\n",
3524 __func__, hdr->version, decode_hioctl(cmd));
3529 if (hdr->rsvd[0] || hdr->rsvd[1] || hdr->return_flags) {
3530 dev_dbg(dev, "%s: Reserved/rflags populated\n", __func__);
3535 rc = do_ioctl(cfg, (void *)&buf);
3537 if (unlikely(copy_to_user(uarg, &buf, size))) {
3538 dev_err(dev, "%s: copy_to_user() fail "
3539 "size=%lu cmd=%d (%s) uarg=%p\n",
3540 __func__, size, cmd, decode_hioctl(cmd), uarg);
3544 /* fall through to exit */
3547 up_read(&cfg->ioctl_rwsem);
3548 if (unlikely(rc && known_ioctl))
3549 dev_err(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3550 __func__, decode_hioctl(cmd), cmd, rc);
3552 dev_dbg(dev, "%s: ioctl %s (%08X) returned rc=%d\n",
3553 __func__, decode_hioctl(cmd), cmd, rc);
3558 * Character device file operations
3560 static const struct file_operations cxlflash_chr_fops = {
3561 .owner = THIS_MODULE,
3562 .open = cxlflash_chr_open,
3563 .unlocked_ioctl = cxlflash_chr_ioctl,
3564 .compat_ioctl = cxlflash_chr_ioctl,
3568 * init_chrdev() - initialize the character device for the host
3569 * @cfg: Internal structure associated with the host.
3571 * Return: 0 on success, -errno on failure
3573 static int init_chrdev(struct cxlflash_cfg *cfg)
3575 struct device *dev = &cfg->dev->dev;
3576 struct device *char_dev;
3581 minor = cxlflash_get_minor();
3582 if (unlikely(minor < 0)) {
3583 dev_err(dev, "%s: Exhausted allowed adapters\n", __func__);
3588 devno = MKDEV(cxlflash_major, minor);
3589 cdev_init(&cfg->cdev, &cxlflash_chr_fops);
3591 rc = cdev_add(&cfg->cdev, devno, 1);
3593 dev_err(dev, "%s: cdev_add failed rc=%d\n", __func__, rc);
3597 char_dev = device_create(cxlflash_class, NULL, devno,
3598 NULL, "cxlflash%d", minor);
3599 if (IS_ERR(char_dev)) {
3600 rc = PTR_ERR(char_dev);
3601 dev_err(dev, "%s: device_create failed rc=%d\n",
3606 cfg->chardev = char_dev;
3608 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3611 cdev_del(&cfg->cdev);
3613 cxlflash_put_minor(minor);
3618 * cxlflash_probe() - PCI entry point to add host
3619 * @pdev: PCI device associated with the host.
3620 * @dev_id: PCI device id associated with device.
3622 * The device will initially start out in a 'probing' state and
3623 * transition to the 'normal' state at the end of a successful
3624 * probe. Should an EEH event occur during probe, the notification
3625 * thread (error_detected()) will wait until the probe handler
3626 * is nearly complete. At that time, the device will be moved to
3627 * a 'probed' state and the EEH thread woken up to drive the slot
3628 * reset and recovery (device moves to 'normal' state). Meanwhile,
3629 * the probe will be allowed to exit successfully.
3631 * Return: 0 on success, -errno on failure
3633 static int cxlflash_probe(struct pci_dev *pdev,
3634 const struct pci_device_id *dev_id)
3636 struct Scsi_Host *host;
3637 struct cxlflash_cfg *cfg = NULL;
3638 struct device *dev = &pdev->dev;
3639 struct dev_dependent_vals *ddv;
3643 dev_dbg(&pdev->dev, "%s: Found CXLFLASH with IRQ: %d\n",
3644 __func__, pdev->irq);
3646 ddv = (struct dev_dependent_vals *)dev_id->driver_data;
3647 driver_template.max_sectors = ddv->max_sectors;
3649 host = scsi_host_alloc(&driver_template, sizeof(struct cxlflash_cfg));
3651 dev_err(dev, "%s: scsi_host_alloc failed\n", __func__);
3656 host->max_id = CXLFLASH_MAX_NUM_TARGETS_PER_BUS;
3657 host->max_lun = CXLFLASH_MAX_NUM_LUNS_PER_TARGET;
3658 host->unique_id = host->host_no;
3659 host->max_cmd_len = CXLFLASH_MAX_CDB_LEN;
3661 cfg = shost_priv(host);
3662 cfg->state = STATE_PROBING;
3664 rc = alloc_mem(cfg);
3666 dev_err(dev, "%s: alloc_mem failed\n", __func__);
3668 scsi_host_put(cfg->host);
3672 cfg->init_state = INIT_STATE_NONE;
3674 cfg->cxl_fops = cxlflash_cxl_fops;
3677 * Promoted LUNs move to the top of the LUN table. The rest stay on
3678 * the bottom half. The bottom half grows from the end (index = 255),
3679 * whereas the top half grows from the beginning (index = 0).
3681 * Initialize the last LUN index for all possible ports.
3683 cfg->promote_lun_index = 0;
3685 for (k = 0; k < MAX_FC_PORTS; k++)
3686 cfg->last_lun_index[k] = CXLFLASH_NUM_VLUNS/2 - 1;
3688 cfg->dev_id = (struct pci_device_id *)dev_id;
3690 init_waitqueue_head(&cfg->tmf_waitq);
3691 init_waitqueue_head(&cfg->reset_waitq);
3693 INIT_WORK(&cfg->work_q, cxlflash_worker_thread);
3694 cfg->lr_state = LINK_RESET_INVALID;
3696 spin_lock_init(&cfg->tmf_slock);
3697 mutex_init(&cfg->ctx_tbl_list_mutex);
3698 mutex_init(&cfg->ctx_recovery_mutex);
3699 init_rwsem(&cfg->ioctl_rwsem);
3700 INIT_LIST_HEAD(&cfg->ctx_err_recovery);
3701 INIT_LIST_HEAD(&cfg->lluns);
3703 pci_set_drvdata(pdev, cfg);
3705 cfg->cxl_afu = cxl_pci_to_afu(pdev);
3709 dev_err(dev, "%s: init_pci failed rc=%d\n", __func__, rc);
3712 cfg->init_state = INIT_STATE_PCI;
3715 if (rc && !wq_has_sleeper(&cfg->reset_waitq)) {
3716 dev_err(dev, "%s: init_afu failed rc=%d\n", __func__, rc);
3719 cfg->init_state = INIT_STATE_AFU;
3721 rc = init_scsi(cfg);
3723 dev_err(dev, "%s: init_scsi failed rc=%d\n", __func__, rc);
3726 cfg->init_state = INIT_STATE_SCSI;
3728 rc = init_chrdev(cfg);
3730 dev_err(dev, "%s: init_chrdev failed rc=%d\n", __func__, rc);
3733 cfg->init_state = INIT_STATE_CDEV;
3735 if (wq_has_sleeper(&cfg->reset_waitq)) {
3736 cfg->state = STATE_PROBED;
3737 wake_up_all(&cfg->reset_waitq);
3739 cfg->state = STATE_NORMAL;
3741 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
3745 cfg->state = STATE_PROBED;
3746 cxlflash_remove(pdev);
3751 * cxlflash_pci_error_detected() - called when a PCI error is detected
3752 * @pdev: PCI device struct.
3753 * @state: PCI channel state.
3755 * When an EEH occurs during an active reset, wait until the reset is
3756 * complete and then take action based upon the device state.
3758 * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
3760 static pci_ers_result_t cxlflash_pci_error_detected(struct pci_dev *pdev,
3761 pci_channel_state_t state)
3764 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3765 struct device *dev = &cfg->dev->dev;
3767 dev_dbg(dev, "%s: pdev=%p state=%u\n", __func__, pdev, state);
3770 case pci_channel_io_frozen:
3771 wait_event(cfg->reset_waitq, cfg->state != STATE_RESET &&
3772 cfg->state != STATE_PROBING);
3773 if (cfg->state == STATE_FAILTERM)
3774 return PCI_ERS_RESULT_DISCONNECT;
3776 cfg->state = STATE_RESET;
3777 scsi_block_requests(cfg->host);
3779 rc = cxlflash_mark_contexts_error(cfg);
3781 dev_err(dev, "%s: Failed to mark user contexts rc=%d\n",
3784 return PCI_ERS_RESULT_NEED_RESET;
3785 case pci_channel_io_perm_failure:
3786 cfg->state = STATE_FAILTERM;
3787 wake_up_all(&cfg->reset_waitq);
3788 scsi_unblock_requests(cfg->host);
3789 return PCI_ERS_RESULT_DISCONNECT;
3793 return PCI_ERS_RESULT_NEED_RESET;
3797 * cxlflash_pci_slot_reset() - called when PCI slot has been reset
3798 * @pdev: PCI device struct.
3800 * This routine is called by the pci error recovery code after the PCI
3801 * slot has been reset, just before we should resume normal operations.
3803 * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
3805 static pci_ers_result_t cxlflash_pci_slot_reset(struct pci_dev *pdev)
3808 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3809 struct device *dev = &cfg->dev->dev;
3811 dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3815 dev_err(dev, "%s: EEH recovery failed rc=%d\n", __func__, rc);
3816 return PCI_ERS_RESULT_DISCONNECT;
3819 return PCI_ERS_RESULT_RECOVERED;
3823 * cxlflash_pci_resume() - called when normal operation can resume
3824 * @pdev: PCI device struct
3826 static void cxlflash_pci_resume(struct pci_dev *pdev)
3828 struct cxlflash_cfg *cfg = pci_get_drvdata(pdev);
3829 struct device *dev = &cfg->dev->dev;
3831 dev_dbg(dev, "%s: pdev=%p\n", __func__, pdev);
3833 cfg->state = STATE_NORMAL;
3834 wake_up_all(&cfg->reset_waitq);
3835 scsi_unblock_requests(cfg->host);
3839 * cxlflash_devnode() - provides devtmpfs for devices in the cxlflash class
3840 * @dev: Character device.
3841 * @mode: Mode that can be used to verify access.
3843 * Return: Allocated string describing the devtmpfs structure.
3845 static char *cxlflash_devnode(struct device *dev, umode_t *mode)
3847 return kasprintf(GFP_KERNEL, "cxlflash/%s", dev_name(dev));
3851 * cxlflash_class_init() - create character device class
3853 * Return: 0 on success, -errno on failure
3855 static int cxlflash_class_init(void)
3860 rc = alloc_chrdev_region(&devno, 0, CXLFLASH_MAX_ADAPTERS, "cxlflash");
3862 pr_err("%s: alloc_chrdev_region failed rc=%d\n", __func__, rc);
3866 cxlflash_major = MAJOR(devno);
3868 cxlflash_class = class_create(THIS_MODULE, "cxlflash");
3869 if (IS_ERR(cxlflash_class)) {
3870 rc = PTR_ERR(cxlflash_class);
3871 pr_err("%s: class_create failed rc=%d\n", __func__, rc);
3875 cxlflash_class->devnode = cxlflash_devnode;
3877 pr_debug("%s: returning rc=%d\n", __func__, rc);
3880 unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3885 * cxlflash_class_exit() - destroy character device class
3887 static void cxlflash_class_exit(void)
3889 dev_t devno = MKDEV(cxlflash_major, 0);
3891 class_destroy(cxlflash_class);
3892 unregister_chrdev_region(devno, CXLFLASH_MAX_ADAPTERS);
3895 static const struct pci_error_handlers cxlflash_err_handler = {
3896 .error_detected = cxlflash_pci_error_detected,
3897 .slot_reset = cxlflash_pci_slot_reset,
3898 .resume = cxlflash_pci_resume,
3902 * PCI device structure
3904 static struct pci_driver cxlflash_driver = {
3905 .name = CXLFLASH_NAME,
3906 .id_table = cxlflash_pci_table,
3907 .probe = cxlflash_probe,
3908 .remove = cxlflash_remove,
3909 .shutdown = cxlflash_remove,
3910 .err_handler = &cxlflash_err_handler,
3914 * init_cxlflash() - module entry point
3916 * Return: 0 on success, -errno on failure
3918 static int __init init_cxlflash(void)
3923 cxlflash_list_init();
3924 rc = cxlflash_class_init();
3928 rc = pci_register_driver(&cxlflash_driver);
3932 pr_debug("%s: returning rc=%d\n", __func__, rc);
3935 cxlflash_class_exit();
3940 * exit_cxlflash() - module exit point
3942 static void __exit exit_cxlflash(void)
3944 cxlflash_term_global_luns();
3945 cxlflash_free_errpage();
3947 pci_unregister_driver(&cxlflash_driver);
3948 cxlflash_class_exit();
3951 module_init(init_cxlflash);
3952 module_exit(exit_cxlflash);