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");
38 * cmd_checkout() - checks out an AFU command
39 * @afu: AFU to checkout from.
41 * Commands are checked out in a round-robin fashion. Note that since
42 * the command pool is larger than the hardware queue, the majority of
43 * times we will only loop once or twice before getting a command. The
44 * buffer and CDB within the command are initialized (zeroed) prior to
47 * Return: The checked out command or NULL when command pool is empty.
49 static struct afu_cmd
*cmd_checkout(struct afu
*afu
)
51 int k
, dec
= CXLFLASH_NUM_CMDS
;
55 k
= (afu
->cmd_couts
++ & (CXLFLASH_NUM_CMDS
- 1));
59 if (!atomic_dec_if_positive(&cmd
->free
)) {
60 pr_devel("%s: returning found index=%d cmd=%p\n",
61 __func__
, cmd
->slot
, cmd
);
62 memset(cmd
->buf
, 0, CMD_BUFSIZE
);
63 memset(cmd
->rcb
.cdb
, 0, sizeof(cmd
->rcb
.cdb
));
72 * cmd_checkin() - checks in an AFU command
73 * @cmd: AFU command to checkin.
75 * Safe to pass commands that have already been checked in. Several
76 * internal tracking fields are reset as part of the checkin. Note
77 * that these are intentionally reset prior to toggling the free bit
78 * to avoid clobbering values in the event that the command is checked
81 static void cmd_checkin(struct afu_cmd
*cmd
)
87 cmd
->sa
.host_use
[0] = 0; /* clears both completion and retry bytes */
89 if (unlikely(atomic_inc_return(&cmd
->free
) != 1)) {
90 pr_err("%s: Freeing cmd (%d) that is not in use!\n",
95 pr_devel("%s: released cmd %p index=%d\n", __func__
, cmd
, cmd
->slot
);
99 * process_cmd_err() - command error handler
100 * @cmd: AFU command that experienced the error.
101 * @scp: SCSI command associated with the AFU command in error.
103 * Translates error bits from AFU command to SCSI command results.
105 static void process_cmd_err(struct afu_cmd
*cmd
, struct scsi_cmnd
*scp
)
107 struct sisl_ioarcb
*ioarcb
;
108 struct sisl_ioasa
*ioasa
;
114 ioarcb
= &(cmd
->rcb
);
117 if (ioasa
->rc
.flags
& SISL_RC_FLAGS_UNDERRUN
) {
118 resid
= ioasa
->resid
;
119 scsi_set_resid(scp
, resid
);
120 pr_debug("%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
121 __func__
, cmd
, scp
, resid
);
124 if (ioasa
->rc
.flags
& SISL_RC_FLAGS_OVERRUN
) {
125 pr_debug("%s: cmd underrun cmd = %p scp = %p\n",
127 scp
->result
= (DID_ERROR
<< 16);
130 pr_debug("%s: cmd failed afu_rc=%d scsi_rc=%d fc_rc=%d "
131 "afu_extra=0x%X, scsi_extra=0x%X, fc_extra=0x%X\n",
132 __func__
, ioasa
->rc
.afu_rc
, ioasa
->rc
.scsi_rc
,
133 ioasa
->rc
.fc_rc
, ioasa
->afu_extra
, ioasa
->scsi_extra
,
136 if (ioasa
->rc
.scsi_rc
) {
137 /* We have a SCSI status */
138 if (ioasa
->rc
.flags
& SISL_RC_FLAGS_SENSE_VALID
) {
139 memcpy(scp
->sense_buffer
, ioasa
->sense_data
,
140 SISL_SENSE_DATA_LEN
);
141 scp
->result
= ioasa
->rc
.scsi_rc
;
143 scp
->result
= ioasa
->rc
.scsi_rc
| (DID_ERROR
<< 16);
147 * We encountered an error. Set scp->result based on nature
150 if (ioasa
->rc
.fc_rc
) {
151 /* We have an FC status */
152 switch (ioasa
->rc
.fc_rc
) {
153 case SISL_FC_RC_LINKDOWN
:
154 scp
->result
= (DID_REQUEUE
<< 16);
156 case SISL_FC_RC_RESID
:
157 /* This indicates an FCP resid underrun */
158 if (!(ioasa
->rc
.flags
& SISL_RC_FLAGS_OVERRUN
)) {
159 /* If the SISL_RC_FLAGS_OVERRUN flag was set,
160 * then we will handle this error else where.
161 * If not then we must handle it here.
162 * This is probably an AFU bug.
164 scp
->result
= (DID_ERROR
<< 16);
167 case SISL_FC_RC_RESIDERR
:
168 /* Resid mismatch between adapter and device */
169 case SISL_FC_RC_TGTABORT
:
170 case SISL_FC_RC_ABORTOK
:
171 case SISL_FC_RC_ABORTFAIL
:
172 case SISL_FC_RC_NOLOGI
:
173 case SISL_FC_RC_ABORTPEND
:
174 case SISL_FC_RC_WRABORTPEND
:
175 case SISL_FC_RC_NOEXP
:
176 case SISL_FC_RC_INUSE
:
177 scp
->result
= (DID_ERROR
<< 16);
182 if (ioasa
->rc
.afu_rc
) {
183 /* We have an AFU error */
184 switch (ioasa
->rc
.afu_rc
) {
185 case SISL_AFU_RC_NO_CHANNELS
:
186 scp
->result
= (DID_NO_CONNECT
<< 16);
188 case SISL_AFU_RC_DATA_DMA_ERR
:
189 switch (ioasa
->afu_extra
) {
190 case SISL_AFU_DMA_ERR_PAGE_IN
:
192 scp
->result
= (DID_IMM_RETRY
<< 16);
194 case SISL_AFU_DMA_ERR_INVALID_EA
:
196 scp
->result
= (DID_ERROR
<< 16);
199 case SISL_AFU_RC_OUT_OF_DATA_BUFS
:
201 scp
->result
= (DID_ALLOC_FAILURE
<< 16);
204 scp
->result
= (DID_ERROR
<< 16);
210 * cmd_complete() - command completion handler
211 * @cmd: AFU command that has completed.
213 * Prepares and submits command that has either completed or timed out to
214 * the SCSI stack. Checks AFU command back into command pool for non-internal
215 * (rcb.scp populated) commands.
217 static void cmd_complete(struct afu_cmd
*cmd
)
219 struct scsi_cmnd
*scp
;
221 struct afu
*afu
= cmd
->parent
;
222 struct cxlflash_cfg
*cfg
= afu
->parent
;
225 spin_lock_irqsave(&cmd
->slock
, lock_flags
);
226 cmd
->sa
.host_use_b
[0] |= B_DONE
;
227 spin_unlock_irqrestore(&cmd
->slock
, lock_flags
);
231 if (unlikely(cmd
->sa
.ioasc
))
232 process_cmd_err(cmd
, scp
);
234 scp
->result
= (DID_OK
<< 16);
236 cmd_is_tmf
= cmd
->cmd_tmf
;
237 cmd_checkin(cmd
); /* Don't use cmd after here */
239 pr_debug_ratelimited("%s: calling scsi_done scp=%p result=%X "
240 "ioasc=%d\n", __func__
, scp
, scp
->result
,
247 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
248 cfg
->tmf_active
= false;
249 wake_up_all_locked(&cfg
->tmf_waitq
);
250 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
253 complete(&cmd
->cevent
);
257 * context_reset() - timeout handler for AFU commands
258 * @cmd: AFU command that timed out.
260 * Sends a reset to the AFU.
262 static void context_reset(struct afu_cmd
*cmd
)
267 struct afu
*afu
= cmd
->parent
;
270 pr_debug("%s: cmd=%p\n", __func__
, cmd
);
272 spin_lock_irqsave(&cmd
->slock
, lock_flags
);
274 /* Already completed? */
275 if (cmd
->sa
.host_use_b
[0] & B_DONE
) {
276 spin_unlock_irqrestore(&cmd
->slock
, lock_flags
);
280 cmd
->sa
.host_use_b
[0] |= (B_DONE
| B_ERROR
| B_TIMEOUT
);
281 spin_unlock_irqrestore(&cmd
->slock
, lock_flags
);
284 * We really want to send this reset at all costs, so spread
285 * out wait time on successive retries for available room.
288 room
= readq_be(&afu
->host_map
->cmd_room
);
289 atomic64_set(&afu
->room
, room
);
293 } while (nretry
++ < MC_ROOM_RETRY_CNT
);
295 pr_err("%s: no cmd_room to send reset\n", __func__
);
300 writeq_be(rrin
, &afu
->host_map
->ioarrin
);
302 rrin
= readq_be(&afu
->host_map
->ioarrin
);
305 /* Double delay each time */
307 } while (nretry
++ < MC_ROOM_RETRY_CNT
);
311 * send_cmd() - sends an AFU command
312 * @afu: AFU associated with the host.
313 * @cmd: AFU command to send.
316 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
318 static int send_cmd(struct afu
*afu
, struct afu_cmd
*cmd
)
320 struct cxlflash_cfg
*cfg
= afu
->parent
;
321 struct device
*dev
= &cfg
->dev
->dev
;
328 * This routine is used by critical users such an AFU sync and to
329 * send a task management function (TMF). Thus we want to retry a
330 * bit before returning an error. To avoid the performance penalty
331 * of MMIO, we spread the update of 'room' over multiple commands.
334 newval
= atomic64_dec_if_positive(&afu
->room
);
337 room
= readq_be(&afu
->host_map
->cmd_room
);
338 atomic64_set(&afu
->room
, room
);
342 } while (nretry
++ < MC_ROOM_RETRY_CNT
);
344 dev_err(dev
, "%s: no cmd_room to send 0x%X\n",
345 __func__
, cmd
->rcb
.cdb
[0]);
348 } else if (unlikely(newval
< 0)) {
349 /* This should be rare. i.e. Only if two threads race and
350 * decrement before the MMIO read is done. In this case
351 * just benefit from the other thread having updated
354 if (nretry
++ < MC_ROOM_RETRY_CNT
) {
363 writeq_be((u64
)&cmd
->rcb
, &afu
->host_map
->ioarrin
);
365 pr_devel("%s: cmd=%p len=%d ea=%p rc=%d\n", __func__
, cmd
,
366 cmd
->rcb
.data_len
, (void *)cmd
->rcb
.data_ea
, rc
);
370 afu
->read_room
= true;
371 kref_get(&cfg
->afu
->mapcount
);
372 schedule_work(&cfg
->work_q
);
373 rc
= SCSI_MLQUEUE_HOST_BUSY
;
378 * wait_resp() - polls for a response or timeout to a sent AFU command
379 * @afu: AFU associated with the host.
380 * @cmd: AFU command that was sent.
382 static void wait_resp(struct afu
*afu
, struct afu_cmd
*cmd
)
384 ulong timeout
= msecs_to_jiffies(cmd
->rcb
.timeout
* 2 * 1000);
386 timeout
= wait_for_completion_timeout(&cmd
->cevent
, timeout
);
390 if (unlikely(cmd
->sa
.ioasc
!= 0))
391 pr_err("%s: CMD 0x%X failed, IOASC: flags 0x%X, afu_rc 0x%X, "
392 "scsi_rc 0x%X, fc_rc 0x%X\n", __func__
, cmd
->rcb
.cdb
[0],
393 cmd
->sa
.rc
.flags
, cmd
->sa
.rc
.afu_rc
, cmd
->sa
.rc
.scsi_rc
,
398 * send_tmf() - sends a Task Management Function (TMF)
399 * @afu: AFU to checkout from.
400 * @scp: SCSI command from stack.
401 * @tmfcmd: TMF command to send.
404 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
406 static int send_tmf(struct afu
*afu
, struct scsi_cmnd
*scp
, u64 tmfcmd
)
410 u32 port_sel
= scp
->device
->channel
+ 1;
412 struct Scsi_Host
*host
= scp
->device
->host
;
413 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)host
->hostdata
;
414 struct device
*dev
= &cfg
->dev
->dev
;
419 cmd
= cmd_checkout(afu
);
420 if (unlikely(!cmd
)) {
421 dev_err(dev
, "%s: could not get a free command\n", __func__
);
422 rc
= SCSI_MLQUEUE_HOST_BUSY
;
426 /* When Task Management Function is active do not send another */
427 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
429 wait_event_interruptible_lock_irq(cfg
->tmf_waitq
,
432 cfg
->tmf_active
= true;
434 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
436 cmd
->rcb
.ctx_id
= afu
->ctx_hndl
;
437 cmd
->rcb
.port_sel
= port_sel
;
438 cmd
->rcb
.lun_id
= lun_to_lunid(scp
->device
->lun
);
440 lflag
= SISL_REQ_FLAGS_TMF_CMD
;
442 cmd
->rcb
.req_flags
= (SISL_REQ_FLAGS_PORT_LUN_ID
|
443 SISL_REQ_FLAGS_SUP_UNDERRUN
| lflag
);
445 /* Stash the scp in the reserved field, for reuse during interrupt */
448 /* Copy the CDB from the cmd passed in */
449 memcpy(cmd
->rcb
.cdb
, &tmfcmd
, sizeof(tmfcmd
));
451 /* Send the command */
452 rc
= send_cmd(afu
, cmd
);
455 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
456 cfg
->tmf_active
= false;
457 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
461 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
462 to
= msecs_to_jiffies(5000);
463 to
= wait_event_interruptible_lock_irq_timeout(cfg
->tmf_waitq
,
468 cfg
->tmf_active
= false;
469 dev_err(dev
, "%s: TMF timed out!\n", __func__
);
472 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
477 static void afu_unmap(struct kref
*ref
)
479 struct afu
*afu
= container_of(ref
, struct afu
, mapcount
);
481 if (likely(afu
->afu_map
)) {
482 cxl_psa_unmap((void __iomem
*)afu
->afu_map
);
488 * cxlflash_driver_info() - information handler for this host driver
489 * @host: SCSI host associated with device.
491 * Return: A string describing the device.
493 static const char *cxlflash_driver_info(struct Scsi_Host
*host
)
495 return CXLFLASH_ADAPTER_NAME
;
499 * cxlflash_queuecommand() - sends a mid-layer request
500 * @host: SCSI host associated with device.
501 * @scp: SCSI command to send.
503 * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
505 static int cxlflash_queuecommand(struct Scsi_Host
*host
, struct scsi_cmnd
*scp
)
507 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)host
->hostdata
;
508 struct afu
*afu
= cfg
->afu
;
509 struct device
*dev
= &cfg
->dev
->dev
;
511 u32 port_sel
= scp
->device
->channel
+ 1;
513 struct scatterlist
*sg
;
519 dev_dbg_ratelimited(dev
, "%s: (scp=%p) %d/%d/%d/%llu "
520 "cdb=(%08X-%08X-%08X-%08X)\n",
521 __func__
, scp
, host
->host_no
, scp
->device
->channel
,
522 scp
->device
->id
, scp
->device
->lun
,
523 get_unaligned_be32(&((u32
*)scp
->cmnd
)[0]),
524 get_unaligned_be32(&((u32
*)scp
->cmnd
)[1]),
525 get_unaligned_be32(&((u32
*)scp
->cmnd
)[2]),
526 get_unaligned_be32(&((u32
*)scp
->cmnd
)[3]));
529 * If a Task Management Function is active, wait for it to complete
530 * before continuing with regular commands.
532 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
533 if (cfg
->tmf_active
) {
534 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
535 rc
= SCSI_MLQUEUE_HOST_BUSY
;
538 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
540 switch (cfg
->state
) {
542 dev_dbg_ratelimited(dev
, "%s: device is in reset!\n", __func__
);
543 rc
= SCSI_MLQUEUE_HOST_BUSY
;
546 dev_dbg_ratelimited(dev
, "%s: device has failed!\n", __func__
);
547 scp
->result
= (DID_NO_CONNECT
<< 16);
555 cmd
= cmd_checkout(afu
);
556 if (unlikely(!cmd
)) {
557 dev_err(dev
, "%s: could not get a free command\n", __func__
);
558 rc
= SCSI_MLQUEUE_HOST_BUSY
;
562 kref_get(&cfg
->afu
->mapcount
);
565 cmd
->rcb
.ctx_id
= afu
->ctx_hndl
;
566 cmd
->rcb
.port_sel
= port_sel
;
567 cmd
->rcb
.lun_id
= lun_to_lunid(scp
->device
->lun
);
569 if (scp
->sc_data_direction
== DMA_TO_DEVICE
)
570 lflag
= SISL_REQ_FLAGS_HOST_WRITE
;
572 lflag
= SISL_REQ_FLAGS_HOST_READ
;
574 cmd
->rcb
.req_flags
= (SISL_REQ_FLAGS_PORT_LUN_ID
|
575 SISL_REQ_FLAGS_SUP_UNDERRUN
| lflag
);
577 /* Stash the scp in the reserved field, for reuse during interrupt */
580 nseg
= scsi_dma_map(scp
);
581 if (unlikely(nseg
< 0)) {
582 dev_err(dev
, "%s: Fail DMA map! nseg=%d\n",
584 rc
= SCSI_MLQUEUE_HOST_BUSY
;
588 ncount
= scsi_sg_count(scp
);
589 scsi_for_each_sg(scp
, sg
, ncount
, i
) {
590 cmd
->rcb
.data_len
= sg_dma_len(sg
);
591 cmd
->rcb
.data_ea
= sg_dma_address(sg
);
594 /* Copy the CDB from the scsi_cmnd passed in */
595 memcpy(cmd
->rcb
.cdb
, scp
->cmnd
, sizeof(cmd
->rcb
.cdb
));
597 /* Send the command */
598 rc
= send_cmd(afu
, cmd
);
606 kref_put(&afu
->mapcount
, afu_unmap
);
607 pr_devel("%s: returning rc=%d\n", __func__
, rc
);
612 * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
613 * @cfg: Internal structure associated with the host.
615 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg
*cfg
)
617 struct pci_dev
*pdev
= cfg
->dev
;
619 if (pci_channel_offline(pdev
))
620 wait_event_timeout(cfg
->reset_waitq
,
621 !pci_channel_offline(pdev
),
622 CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT
);
626 * free_mem() - free memory associated with the AFU
627 * @cfg: Internal structure associated with the host.
629 static void free_mem(struct cxlflash_cfg
*cfg
)
633 struct afu
*afu
= cfg
->afu
;
636 for (i
= 0; i
< CXLFLASH_NUM_CMDS
; i
++) {
637 buf
= afu
->cmd
[i
].buf
;
638 if (!((u64
)buf
& (PAGE_SIZE
- 1)))
639 free_page((ulong
)buf
);
642 free_pages((ulong
)afu
, get_order(sizeof(struct afu
)));
648 * stop_afu() - stops the AFU command timers and unmaps the MMIO space
649 * @cfg: Internal structure associated with the host.
651 * Safe to call with AFU in a partially allocated/initialized state.
653 * Cleans up all state associated with the command queue, and unmaps
656 * - complete() will take care of commands we initiated (they'll be checked
657 * in as part of the cleanup that occurs after the completion)
659 * - cmd_checkin() will take care of entries that we did not initiate and that
660 * have not (and will not) complete because they are sitting on a [now stale]
663 static void stop_afu(struct cxlflash_cfg
*cfg
)
666 struct afu
*afu
= cfg
->afu
;
670 for (i
= 0; i
< CXLFLASH_NUM_CMDS
; i
++) {
672 complete(&cmd
->cevent
);
673 if (!atomic_read(&cmd
->free
))
677 if (likely(afu
->afu_map
)) {
678 cxl_psa_unmap((void __iomem
*)afu
->afu_map
);
681 kref_put(&afu
->mapcount
, afu_unmap
);
686 * term_mc() - terminates the master context
687 * @cfg: Internal structure associated with the host.
688 * @level: Depth of allocation, where to begin waterfall tear down.
690 * Safe to call with AFU/MC in partially allocated/initialized state.
692 static void term_mc(struct cxlflash_cfg
*cfg
, enum undo_level level
)
695 struct afu
*afu
= cfg
->afu
;
696 struct device
*dev
= &cfg
->dev
->dev
;
698 if (!afu
|| !cfg
->mcctx
) {
699 dev_err(dev
, "%s: returning from term_mc with NULL afu or MC\n",
706 rc
= cxl_stop_context(cfg
->mcctx
);
709 cxl_unmap_afu_irq(cfg
->mcctx
, 3, afu
);
711 cxl_unmap_afu_irq(cfg
->mcctx
, 2, afu
);
713 cxl_unmap_afu_irq(cfg
->mcctx
, 1, afu
);
715 cxl_free_afu_irqs(cfg
->mcctx
);
716 case RELEASE_CONTEXT
:
722 * term_afu() - terminates the AFU
723 * @cfg: Internal structure associated with the host.
725 * Safe to call with AFU/MC in partially allocated/initialized state.
727 static void term_afu(struct cxlflash_cfg
*cfg
)
729 term_mc(cfg
, UNDO_START
);
734 pr_debug("%s: returning\n", __func__
);
738 * cxlflash_remove() - PCI entry point to tear down host
739 * @pdev: PCI device associated with the host.
741 * Safe to use as a cleanup in partially allocated/initialized state.
743 static void cxlflash_remove(struct pci_dev
*pdev
)
745 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
748 /* If a Task Management Function is active, wait for it to complete
749 * before continuing with remove.
751 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
753 wait_event_interruptible_lock_irq(cfg
->tmf_waitq
,
756 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
758 cfg
->state
= STATE_FAILTERM
;
759 cxlflash_stop_term_user_contexts(cfg
);
761 switch (cfg
->init_state
) {
762 case INIT_STATE_SCSI
:
763 cxlflash_term_local_luns(cfg
);
764 scsi_remove_host(cfg
->host
);
767 cancel_work_sync(&cfg
->work_q
);
770 pci_release_regions(cfg
->dev
);
771 pci_disable_device(pdev
);
772 case INIT_STATE_NONE
:
774 scsi_host_put(cfg
->host
);
778 pr_debug("%s: returning\n", __func__
);
782 * alloc_mem() - allocates the AFU and its command pool
783 * @cfg: Internal structure associated with the host.
785 * A partially allocated state remains on failure.
789 * -ENOMEM on failure to allocate memory
791 static int alloc_mem(struct cxlflash_cfg
*cfg
)
796 struct device
*dev
= &cfg
->dev
->dev
;
798 /* AFU is ~12k, i.e. only one 64k page or up to four 4k pages */
799 cfg
->afu
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
800 get_order(sizeof(struct afu
)));
801 if (unlikely(!cfg
->afu
)) {
802 dev_err(dev
, "%s: cannot get %d free pages\n",
803 __func__
, get_order(sizeof(struct afu
)));
807 cfg
->afu
->parent
= cfg
;
808 cfg
->afu
->afu_map
= NULL
;
810 for (i
= 0; i
< CXLFLASH_NUM_CMDS
; buf
+= CMD_BUFSIZE
, i
++) {
811 if (!((u64
)buf
& (PAGE_SIZE
- 1))) {
812 buf
= (void *)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
813 if (unlikely(!buf
)) {
815 "%s: Allocate command buffers fail!\n",
823 cfg
->afu
->cmd
[i
].buf
= buf
;
824 atomic_set(&cfg
->afu
->cmd
[i
].free
, 1);
825 cfg
->afu
->cmd
[i
].slot
= i
;
833 * init_pci() - initializes the host as a PCI device
834 * @cfg: Internal structure associated with the host.
836 * Return: 0 on success, -errno on failure
838 static int init_pci(struct cxlflash_cfg
*cfg
)
840 struct pci_dev
*pdev
= cfg
->dev
;
843 cfg
->cxlflash_regs_pci
= pci_resource_start(pdev
, 0);
844 rc
= pci_request_regions(pdev
, CXLFLASH_NAME
);
847 "%s: Couldn't register memory range of registers\n",
852 rc
= pci_enable_device(pdev
);
853 if (rc
|| pci_channel_offline(pdev
)) {
854 if (pci_channel_offline(pdev
)) {
855 cxlflash_wait_for_pci_err_recovery(cfg
);
856 rc
= pci_enable_device(pdev
);
860 dev_err(&pdev
->dev
, "%s: Cannot enable adapter\n",
862 cxlflash_wait_for_pci_err_recovery(cfg
);
863 goto out_release_regions
;
867 rc
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(64));
869 dev_dbg(&pdev
->dev
, "%s: Failed to set 64 bit PCI DMA mask\n",
871 rc
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
875 dev_err(&pdev
->dev
, "%s: Failed to set PCI DMA mask\n",
880 pci_set_master(pdev
);
882 if (pci_channel_offline(pdev
)) {
883 cxlflash_wait_for_pci_err_recovery(cfg
);
884 if (pci_channel_offline(pdev
)) {
886 goto out_msi_disable
;
890 rc
= pci_save_state(pdev
);
892 if (rc
!= PCIBIOS_SUCCESSFUL
) {
893 dev_err(&pdev
->dev
, "%s: Failed to save PCI config space\n",
900 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
905 cxlflash_wait_for_pci_err_recovery(cfg
);
907 pci_disable_device(pdev
);
909 pci_release_regions(pdev
);
915 * init_scsi() - adds the host to the SCSI stack and kicks off host scan
916 * @cfg: Internal structure associated with the host.
918 * Return: 0 on success, -errno on failure
920 static int init_scsi(struct cxlflash_cfg
*cfg
)
922 struct pci_dev
*pdev
= cfg
->dev
;
925 rc
= scsi_add_host(cfg
->host
, &pdev
->dev
);
927 dev_err(&pdev
->dev
, "%s: scsi_add_host failed (rc=%d)\n",
932 scsi_scan_host(cfg
->host
);
935 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
940 * set_port_online() - transitions the specified host FC port to online state
941 * @fc_regs: Top of MMIO region defined for specified port.
943 * The provided MMIO region must be mapped prior to call. Online state means
944 * that the FC link layer has synced, completed the handshaking process, and
945 * is ready for login to start.
947 static void set_port_online(__be64 __iomem
*fc_regs
)
951 cmdcfg
= readq_be(&fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
952 cmdcfg
&= (~FC_MTIP_CMDCONFIG_OFFLINE
); /* clear OFF_LINE */
953 cmdcfg
|= (FC_MTIP_CMDCONFIG_ONLINE
); /* set ON_LINE */
954 writeq_be(cmdcfg
, &fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
958 * set_port_offline() - transitions the specified host FC port to offline state
959 * @fc_regs: Top of MMIO region defined for specified port.
961 * The provided MMIO region must be mapped prior to call.
963 static void set_port_offline(__be64 __iomem
*fc_regs
)
967 cmdcfg
= readq_be(&fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
968 cmdcfg
&= (~FC_MTIP_CMDCONFIG_ONLINE
); /* clear ON_LINE */
969 cmdcfg
|= (FC_MTIP_CMDCONFIG_OFFLINE
); /* set OFF_LINE */
970 writeq_be(cmdcfg
, &fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
974 * wait_port_online() - waits for the specified host FC port come online
975 * @fc_regs: Top of MMIO region defined for specified port.
976 * @delay_us: Number of microseconds to delay between reading port status.
977 * @nretry: Number of cycles to retry reading port status.
979 * The provided MMIO region must be mapped prior to call. This will timeout
980 * when the cable is not plugged in.
983 * TRUE (1) when the specified port is online
984 * FALSE (0) when the specified port fails to come online after timeout
985 * -EINVAL when @delay_us is less than 1000
987 static int wait_port_online(__be64 __iomem
*fc_regs
, u32 delay_us
, u32 nretry
)
991 if (delay_us
< 1000) {
992 pr_err("%s: invalid delay specified %d\n", __func__
, delay_us
);
997 msleep(delay_us
/ 1000);
998 status
= readq_be(&fc_regs
[FC_MTIP_STATUS
/ 8]);
999 if (status
== U64_MAX
)
1001 } while ((status
& FC_MTIP_STATUS_MASK
) != FC_MTIP_STATUS_ONLINE
&&
1004 return ((status
& FC_MTIP_STATUS_MASK
) == FC_MTIP_STATUS_ONLINE
);
1008 * wait_port_offline() - waits for the specified host FC port go offline
1009 * @fc_regs: Top of MMIO region defined for specified port.
1010 * @delay_us: Number of microseconds to delay between reading port status.
1011 * @nretry: Number of cycles to retry reading port status.
1013 * The provided MMIO region must be mapped prior to call.
1016 * TRUE (1) when the specified port is offline
1017 * FALSE (0) when the specified port fails to go offline after timeout
1018 * -EINVAL when @delay_us is less than 1000
1020 static int wait_port_offline(__be64 __iomem
*fc_regs
, u32 delay_us
, u32 nretry
)
1024 if (delay_us
< 1000) {
1025 pr_err("%s: invalid delay specified %d\n", __func__
, delay_us
);
1030 msleep(delay_us
/ 1000);
1031 status
= readq_be(&fc_regs
[FC_MTIP_STATUS
/ 8]);
1032 if (status
== U64_MAX
)
1034 } while ((status
& FC_MTIP_STATUS_MASK
) != FC_MTIP_STATUS_OFFLINE
&&
1037 return ((status
& FC_MTIP_STATUS_MASK
) == FC_MTIP_STATUS_OFFLINE
);
1041 * afu_set_wwpn() - configures the WWPN for the specified host FC port
1042 * @afu: AFU associated with the host that owns the specified FC port.
1043 * @port: Port number being configured.
1044 * @fc_regs: Top of MMIO region defined for specified port.
1045 * @wwpn: The world-wide-port-number previously discovered for port.
1047 * The provided MMIO region must be mapped prior to call. As part of the
1048 * sequence to configure the WWPN, the port is toggled offline and then back
1049 * online. This toggling action can cause this routine to delay up to a few
1050 * seconds. When configured to use the internal LUN feature of the AFU, a
1051 * failure to come online is overridden.
1054 * 0 when the WWPN is successfully written and the port comes back online
1055 * -1 when the port fails to go offline or come back up online
1057 static int afu_set_wwpn(struct afu
*afu
, int port
, __be64 __iomem
*fc_regs
,
1062 set_port_offline(fc_regs
);
1064 if (!wait_port_offline(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1065 FC_PORT_STATUS_RETRY_CNT
)) {
1066 pr_debug("%s: wait on port %d to go offline timed out\n",
1068 rc
= -1; /* but continue on to leave the port back online */
1072 writeq_be(wwpn
, &fc_regs
[FC_PNAME
/ 8]);
1074 /* Always return success after programming WWPN */
1077 set_port_online(fc_regs
);
1079 if (!wait_port_online(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1080 FC_PORT_STATUS_RETRY_CNT
)) {
1081 pr_err("%s: wait on port %d to go online timed out\n",
1085 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1091 * afu_link_reset() - resets the specified host FC port
1092 * @afu: AFU associated with the host that owns the specified FC port.
1093 * @port: Port number being configured.
1094 * @fc_regs: Top of MMIO region defined for specified port.
1096 * The provided MMIO region must be mapped prior to call. The sequence to
1097 * reset the port involves toggling it offline and then back online. This
1098 * action can cause this routine to delay up to a few seconds. An effort
1099 * is made to maintain link with the device by switching to host to use
1100 * the alternate port exclusively while the reset takes place.
1101 * failure to come online is overridden.
1103 static void afu_link_reset(struct afu
*afu
, int port
, __be64 __iomem
*fc_regs
)
1107 /* first switch the AFU to the other links, if any */
1108 port_sel
= readq_be(&afu
->afu_map
->global
.regs
.afu_port_sel
);
1109 port_sel
&= ~(1ULL << port
);
1110 writeq_be(port_sel
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1111 cxlflash_afu_sync(afu
, 0, 0, AFU_GSYNC
);
1113 set_port_offline(fc_regs
);
1114 if (!wait_port_offline(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1115 FC_PORT_STATUS_RETRY_CNT
))
1116 pr_err("%s: wait on port %d to go offline timed out\n",
1119 set_port_online(fc_regs
);
1120 if (!wait_port_online(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1121 FC_PORT_STATUS_RETRY_CNT
))
1122 pr_err("%s: wait on port %d to go online timed out\n",
1125 /* switch back to include this port */
1126 port_sel
|= (1ULL << port
);
1127 writeq_be(port_sel
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1128 cxlflash_afu_sync(afu
, 0, 0, AFU_GSYNC
);
1130 pr_debug("%s: returning port_sel=%lld\n", __func__
, port_sel
);
1134 * Asynchronous interrupt information table
1136 static const struct asyc_intr_info ainfo
[] = {
1137 {SISL_ASTATUS_FC0_OTHER
, "other error", 0, CLR_FC_ERROR
| LINK_RESET
},
1138 {SISL_ASTATUS_FC0_LOGO
, "target initiated LOGO", 0, 0},
1139 {SISL_ASTATUS_FC0_CRC_T
, "CRC threshold exceeded", 0, LINK_RESET
},
1140 {SISL_ASTATUS_FC0_LOGI_R
, "login timed out, retrying", 0, LINK_RESET
},
1141 {SISL_ASTATUS_FC0_LOGI_F
, "login failed", 0, CLR_FC_ERROR
},
1142 {SISL_ASTATUS_FC0_LOGI_S
, "login succeeded", 0, SCAN_HOST
},
1143 {SISL_ASTATUS_FC0_LINK_DN
, "link down", 0, 0},
1144 {SISL_ASTATUS_FC0_LINK_UP
, "link up", 0, 0},
1145 {SISL_ASTATUS_FC1_OTHER
, "other error", 1, CLR_FC_ERROR
| LINK_RESET
},
1146 {SISL_ASTATUS_FC1_LOGO
, "target initiated LOGO", 1, 0},
1147 {SISL_ASTATUS_FC1_CRC_T
, "CRC threshold exceeded", 1, LINK_RESET
},
1148 {SISL_ASTATUS_FC1_LOGI_R
, "login timed out, retrying", 1, LINK_RESET
},
1149 {SISL_ASTATUS_FC1_LOGI_F
, "login failed", 1, CLR_FC_ERROR
},
1150 {SISL_ASTATUS_FC1_LOGI_S
, "login succeeded", 1, SCAN_HOST
},
1151 {SISL_ASTATUS_FC1_LINK_DN
, "link down", 1, 0},
1152 {SISL_ASTATUS_FC1_LINK_UP
, "link up", 1, 0},
1153 {0x0, "", 0, 0} /* terminator */
1157 * find_ainfo() - locates and returns asynchronous interrupt information
1158 * @status: Status code set by AFU on error.
1160 * Return: The located information or NULL when the status code is invalid.
1162 static const struct asyc_intr_info
*find_ainfo(u64 status
)
1164 const struct asyc_intr_info
*info
;
1166 for (info
= &ainfo
[0]; info
->status
; info
++)
1167 if (info
->status
== status
)
1174 * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1175 * @afu: AFU associated with the host.
1177 static void afu_err_intr_init(struct afu
*afu
)
1182 /* global async interrupts: AFU clears afu_ctrl on context exit
1183 * if async interrupts were sent to that context. This prevents
1184 * the AFU form sending further async interrupts when
1186 * nobody to receive them.
1190 writeq_be(-1ULL, &afu
->afu_map
->global
.regs
.aintr_mask
);
1191 /* set LISN# to send and point to master context */
1192 reg
= ((u64
) (((afu
->ctx_hndl
<< 8) | SISL_MSI_ASYNC_ERROR
)) << 40);
1194 if (afu
->internal_lun
)
1195 reg
|= 1; /* Bit 63 indicates local lun */
1196 writeq_be(reg
, &afu
->afu_map
->global
.regs
.afu_ctrl
);
1198 writeq_be(-1ULL, &afu
->afu_map
->global
.regs
.aintr_clear
);
1199 /* unmask bits that are of interest */
1200 /* note: afu can send an interrupt after this step */
1201 writeq_be(SISL_ASTATUS_MASK
, &afu
->afu_map
->global
.regs
.aintr_mask
);
1202 /* clear again in case a bit came on after previous clear but before */
1204 writeq_be(-1ULL, &afu
->afu_map
->global
.regs
.aintr_clear
);
1206 /* Clear/Set internal lun bits */
1207 reg
= readq_be(&afu
->afu_map
->global
.fc_regs
[0][FC_CONFIG2
/ 8]);
1208 reg
&= SISL_FC_INTERNAL_MASK
;
1209 if (afu
->internal_lun
)
1210 reg
|= ((u64
)(afu
->internal_lun
- 1) << SISL_FC_INTERNAL_SHIFT
);
1211 writeq_be(reg
, &afu
->afu_map
->global
.fc_regs
[0][FC_CONFIG2
/ 8]);
1213 /* now clear FC errors */
1214 for (i
= 0; i
< NUM_FC_PORTS
; i
++) {
1215 writeq_be(0xFFFFFFFFU
,
1216 &afu
->afu_map
->global
.fc_regs
[i
][FC_ERROR
/ 8]);
1217 writeq_be(0, &afu
->afu_map
->global
.fc_regs
[i
][FC_ERRCAP
/ 8]);
1220 /* sync interrupts for master's IOARRIN write */
1221 /* note that unlike asyncs, there can be no pending sync interrupts */
1222 /* at this time (this is a fresh context and master has not written */
1223 /* IOARRIN yet), so there is nothing to clear. */
1225 /* set LISN#, it is always sent to the context that wrote IOARRIN */
1226 writeq_be(SISL_MSI_SYNC_ERROR
, &afu
->host_map
->ctx_ctrl
);
1227 writeq_be(SISL_ISTATUS_MASK
, &afu
->host_map
->intr_mask
);
1231 * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1232 * @irq: Interrupt number.
1233 * @data: Private data provided at interrupt registration, the AFU.
1235 * Return: Always return IRQ_HANDLED.
1237 static irqreturn_t
cxlflash_sync_err_irq(int irq
, void *data
)
1239 struct afu
*afu
= (struct afu
*)data
;
1243 reg
= readq_be(&afu
->host_map
->intr_status
);
1244 reg_unmasked
= (reg
& SISL_ISTATUS_UNMASK
);
1246 if (reg_unmasked
== 0UL) {
1247 pr_err("%s: %llX: spurious interrupt, intr_status %016llX\n",
1248 __func__
, (u64
)afu
, reg
);
1249 goto cxlflash_sync_err_irq_exit
;
1252 pr_err("%s: %llX: unexpected interrupt, intr_status %016llX\n",
1253 __func__
, (u64
)afu
, reg
);
1255 writeq_be(reg_unmasked
, &afu
->host_map
->intr_clear
);
1257 cxlflash_sync_err_irq_exit
:
1258 pr_debug("%s: returning rc=%d\n", __func__
, IRQ_HANDLED
);
1263 * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1264 * @irq: Interrupt number.
1265 * @data: Private data provided at interrupt registration, the AFU.
1267 * Return: Always return IRQ_HANDLED.
1269 static irqreturn_t
cxlflash_rrq_irq(int irq
, void *data
)
1271 struct afu
*afu
= (struct afu
*)data
;
1272 struct afu_cmd
*cmd
;
1273 bool toggle
= afu
->toggle
;
1275 *hrrq_start
= afu
->hrrq_start
,
1276 *hrrq_end
= afu
->hrrq_end
,
1277 *hrrq_curr
= afu
->hrrq_curr
;
1279 /* Process however many RRQ entries that are ready */
1283 if ((entry
& SISL_RESP_HANDLE_T_BIT
) != toggle
)
1286 cmd
= (struct afu_cmd
*)(entry
& ~SISL_RESP_HANDLE_T_BIT
);
1289 /* Advance to next entry or wrap and flip the toggle bit */
1290 if (hrrq_curr
< hrrq_end
)
1293 hrrq_curr
= hrrq_start
;
1294 toggle
^= SISL_RESP_HANDLE_T_BIT
;
1298 afu
->hrrq_curr
= hrrq_curr
;
1299 afu
->toggle
= toggle
;
1305 * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1306 * @irq: Interrupt number.
1307 * @data: Private data provided at interrupt registration, the AFU.
1309 * Return: Always return IRQ_HANDLED.
1311 static irqreturn_t
cxlflash_async_err_irq(int irq
, void *data
)
1313 struct afu
*afu
= (struct afu
*)data
;
1314 struct cxlflash_cfg
*cfg
= afu
->parent
;
1315 struct device
*dev
= &cfg
->dev
->dev
;
1317 const struct asyc_intr_info
*info
;
1318 struct sisl_global_map __iomem
*global
= &afu
->afu_map
->global
;
1323 reg
= readq_be(&global
->regs
.aintr_status
);
1324 reg_unmasked
= (reg
& SISL_ASTATUS_UNMASK
);
1326 if (reg_unmasked
== 0) {
1327 dev_err(dev
, "%s: spurious interrupt, aintr_status 0x%016llX\n",
1332 /* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1333 writeq_be(reg_unmasked
, &global
->regs
.aintr_clear
);
1335 /* Check each bit that is on */
1336 for (i
= 0; reg_unmasked
; i
++, reg_unmasked
= (reg_unmasked
>> 1)) {
1337 info
= find_ainfo(1ULL << i
);
1338 if (((reg_unmasked
& 0x1) == 0) || !info
)
1343 dev_err(dev
, "%s: FC Port %d -> %s, fc_status 0x%08llX\n",
1344 __func__
, port
, info
->desc
,
1345 readq_be(&global
->fc_regs
[port
][FC_STATUS
/ 8]));
1348 * Do link reset first, some OTHER errors will set FC_ERROR
1349 * again if cleared before or w/o a reset
1351 if (info
->action
& LINK_RESET
) {
1352 dev_err(dev
, "%s: FC Port %d: resetting link\n",
1354 cfg
->lr_state
= LINK_RESET_REQUIRED
;
1355 cfg
->lr_port
= port
;
1356 kref_get(&cfg
->afu
->mapcount
);
1357 schedule_work(&cfg
->work_q
);
1360 if (info
->action
& CLR_FC_ERROR
) {
1361 reg
= readq_be(&global
->fc_regs
[port
][FC_ERROR
/ 8]);
1364 * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1365 * should be the same and tracing one is sufficient.
1368 dev_err(dev
, "%s: fc %d: clearing fc_error 0x%08llX\n",
1369 __func__
, port
, reg
);
1371 writeq_be(reg
, &global
->fc_regs
[port
][FC_ERROR
/ 8]);
1372 writeq_be(0, &global
->fc_regs
[port
][FC_ERRCAP
/ 8]);
1375 if (info
->action
& SCAN_HOST
) {
1376 atomic_inc(&cfg
->scan_host_needed
);
1377 kref_get(&cfg
->afu
->mapcount
);
1378 schedule_work(&cfg
->work_q
);
1383 dev_dbg(dev
, "%s: returning IRQ_HANDLED, afu=%p\n", __func__
, afu
);
1388 * start_context() - starts the master context
1389 * @cfg: Internal structure associated with the host.
1391 * Return: A success or failure value from CXL services.
1393 static int start_context(struct cxlflash_cfg
*cfg
)
1397 rc
= cxl_start_context(cfg
->mcctx
,
1398 cfg
->afu
->work
.work_element_descriptor
,
1401 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1406 * read_vpd() - obtains the WWPNs from VPD
1407 * @cfg: Internal structure associated with the host.
1408 * @wwpn: Array of size NUM_FC_PORTS to pass back WWPNs
1410 * Return: 0 on success, -errno on failure
1412 static int read_vpd(struct cxlflash_cfg
*cfg
, u64 wwpn
[])
1414 struct pci_dev
*dev
= cfg
->parent_dev
;
1416 int ro_start
, ro_size
, i
, j
, k
;
1418 char vpd_data
[CXLFLASH_VPD_LEN
];
1419 char tmp_buf
[WWPN_BUF_LEN
] = { 0 };
1420 char *wwpn_vpd_tags
[NUM_FC_PORTS
] = { "V5", "V6" };
1422 /* Get the VPD data from the device */
1423 vpd_size
= pci_read_vpd(dev
, 0, sizeof(vpd_data
), vpd_data
);
1424 if (unlikely(vpd_size
<= 0)) {
1425 dev_err(&dev
->dev
, "%s: Unable to read VPD (size = %ld)\n",
1426 __func__
, vpd_size
);
1431 /* Get the read only section offset */
1432 ro_start
= pci_vpd_find_tag(vpd_data
, 0, vpd_size
,
1433 PCI_VPD_LRDT_RO_DATA
);
1434 if (unlikely(ro_start
< 0)) {
1435 dev_err(&dev
->dev
, "%s: VPD Read-only data not found\n",
1441 /* Get the read only section size, cap when extends beyond read VPD */
1442 ro_size
= pci_vpd_lrdt_size(&vpd_data
[ro_start
]);
1444 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
1445 if (unlikely((i
+ j
) > vpd_size
)) {
1446 pr_debug("%s: Might need to read more VPD (%d > %ld)\n",
1447 __func__
, (i
+ j
), vpd_size
);
1448 ro_size
= vpd_size
- i
;
1452 * Find the offset of the WWPN tag within the read only
1453 * VPD data and validate the found field (partials are
1454 * no good to us). Convert the ASCII data to an integer
1455 * value. Note that we must copy to a temporary buffer
1456 * because the conversion service requires that the ASCII
1457 * string be terminated.
1459 for (k
= 0; k
< NUM_FC_PORTS
; k
++) {
1461 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
1463 i
= pci_vpd_find_info_keyword(vpd_data
, i
, j
, wwpn_vpd_tags
[k
]);
1464 if (unlikely(i
< 0)) {
1465 dev_err(&dev
->dev
, "%s: Port %d WWPN not found "
1466 "in VPD\n", __func__
, k
);
1471 j
= pci_vpd_info_field_size(&vpd_data
[i
]);
1472 i
+= PCI_VPD_INFO_FLD_HDR_SIZE
;
1473 if (unlikely((i
+ j
> vpd_size
) || (j
!= WWPN_LEN
))) {
1474 dev_err(&dev
->dev
, "%s: Port %d WWPN incomplete or "
1481 memcpy(tmp_buf
, &vpd_data
[i
], WWPN_LEN
);
1482 rc
= kstrtoul(tmp_buf
, WWPN_LEN
, (ulong
*)&wwpn
[k
]);
1484 dev_err(&dev
->dev
, "%s: Fail to convert port %d WWPN "
1485 "to integer\n", __func__
, k
);
1492 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1497 * init_pcr() - initialize the provisioning and control registers
1498 * @cfg: Internal structure associated with the host.
1500 * Also sets up fast access to the mapped registers and initializes AFU
1501 * command fields that never change.
1503 static void init_pcr(struct cxlflash_cfg
*cfg
)
1505 struct afu
*afu
= cfg
->afu
;
1506 struct sisl_ctrl_map __iomem
*ctrl_map
;
1509 for (i
= 0; i
< MAX_CONTEXT
; i
++) {
1510 ctrl_map
= &afu
->afu_map
->ctrls
[i
].ctrl
;
1511 /* Disrupt any clients that could be running */
1512 /* e.g. clients that survived a master restart */
1513 writeq_be(0, &ctrl_map
->rht_start
);
1514 writeq_be(0, &ctrl_map
->rht_cnt_id
);
1515 writeq_be(0, &ctrl_map
->ctx_cap
);
1518 /* Copy frequently used fields into afu */
1519 afu
->ctx_hndl
= (u16
) cxl_process_element(cfg
->mcctx
);
1520 afu
->host_map
= &afu
->afu_map
->hosts
[afu
->ctx_hndl
].host
;
1521 afu
->ctrl_map
= &afu
->afu_map
->ctrls
[afu
->ctx_hndl
].ctrl
;
1523 /* Program the Endian Control for the master context */
1524 writeq_be(SISL_ENDIAN_CTRL
, &afu
->host_map
->endian_ctrl
);
1526 /* Initialize cmd fields that never change */
1527 for (i
= 0; i
< CXLFLASH_NUM_CMDS
; i
++) {
1528 afu
->cmd
[i
].rcb
.ctx_id
= afu
->ctx_hndl
;
1529 afu
->cmd
[i
].rcb
.msi
= SISL_MSI_RRQ_UPDATED
;
1530 afu
->cmd
[i
].rcb
.rrq
= 0x0;
1535 * init_global() - initialize AFU global registers
1536 * @cfg: Internal structure associated with the host.
1538 static int init_global(struct cxlflash_cfg
*cfg
)
1540 struct afu
*afu
= cfg
->afu
;
1541 struct device
*dev
= &cfg
->dev
->dev
;
1542 u64 wwpn
[NUM_FC_PORTS
]; /* wwpn of AFU ports */
1543 int i
= 0, num_ports
= 0;
1547 rc
= read_vpd(cfg
, &wwpn
[0]);
1549 dev_err(dev
, "%s: could not read vpd rc=%d\n", __func__
, rc
);
1553 pr_debug("%s: wwpn0=0x%llX wwpn1=0x%llX\n", __func__
, wwpn
[0], wwpn
[1]);
1555 /* Set up RRQ in AFU for master issued cmds */
1556 writeq_be((u64
) afu
->hrrq_start
, &afu
->host_map
->rrq_start
);
1557 writeq_be((u64
) afu
->hrrq_end
, &afu
->host_map
->rrq_end
);
1559 /* AFU configuration */
1560 reg
= readq_be(&afu
->afu_map
->global
.regs
.afu_config
);
1561 reg
|= SISL_AFUCONF_AR_ALL
|SISL_AFUCONF_ENDIAN
;
1562 /* enable all auto retry options and control endianness */
1563 /* leave others at default: */
1564 /* CTX_CAP write protected, mbox_r does not clear on read and */
1565 /* checker on if dual afu */
1566 writeq_be(reg
, &afu
->afu_map
->global
.regs
.afu_config
);
1568 /* Global port select: select either port */
1569 if (afu
->internal_lun
) {
1570 /* Only use port 0 */
1571 writeq_be(PORT0
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1572 num_ports
= NUM_FC_PORTS
- 1;
1574 writeq_be(BOTH_PORTS
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1575 num_ports
= NUM_FC_PORTS
;
1578 for (i
= 0; i
< num_ports
; i
++) {
1579 /* Unmask all errors (but they are still masked at AFU) */
1580 writeq_be(0, &afu
->afu_map
->global
.fc_regs
[i
][FC_ERRMSK
/ 8]);
1581 /* Clear CRC error cnt & set a threshold */
1582 (void)readq_be(&afu
->afu_map
->global
.
1583 fc_regs
[i
][FC_CNT_CRCERR
/ 8]);
1584 writeq_be(MC_CRC_THRESH
, &afu
->afu_map
->global
.fc_regs
[i
]
1585 [FC_CRC_THRESH
/ 8]);
1587 /* Set WWPNs. If already programmed, wwpn[i] is 0 */
1589 afu_set_wwpn(afu
, i
,
1590 &afu
->afu_map
->global
.fc_regs
[i
][0],
1592 dev_err(dev
, "%s: failed to set WWPN on port %d\n",
1597 /* Programming WWPN back to back causes additional
1598 * offline/online transitions and a PLOGI
1603 /* Set up master's own CTX_CAP to allow real mode, host translation */
1604 /* tables, afu cmds and read/write GSCSI cmds. */
1605 /* First, unlock ctx_cap write by reading mbox */
1606 (void)readq_be(&afu
->ctrl_map
->mbox_r
); /* unlock ctx_cap */
1607 writeq_be((SISL_CTX_CAP_REAL_MODE
| SISL_CTX_CAP_HOST_XLATE
|
1608 SISL_CTX_CAP_READ_CMD
| SISL_CTX_CAP_WRITE_CMD
|
1609 SISL_CTX_CAP_AFU_CMD
| SISL_CTX_CAP_GSCSI_CMD
),
1610 &afu
->ctrl_map
->ctx_cap
);
1611 /* Initialize heartbeat */
1612 afu
->hb
= readq_be(&afu
->afu_map
->global
.regs
.afu_hb
);
1619 * start_afu() - initializes and starts the AFU
1620 * @cfg: Internal structure associated with the host.
1622 static int start_afu(struct cxlflash_cfg
*cfg
)
1624 struct afu
*afu
= cfg
->afu
;
1625 struct afu_cmd
*cmd
;
1630 for (i
= 0; i
< CXLFLASH_NUM_CMDS
; i
++) {
1633 init_completion(&cmd
->cevent
);
1634 spin_lock_init(&cmd
->slock
);
1640 /* After an AFU reset, RRQ entries are stale, clear them */
1641 memset(&afu
->rrq_entry
, 0, sizeof(afu
->rrq_entry
));
1643 /* Initialize RRQ pointers */
1644 afu
->hrrq_start
= &afu
->rrq_entry
[0];
1645 afu
->hrrq_end
= &afu
->rrq_entry
[NUM_RRQ_ENTRY
- 1];
1646 afu
->hrrq_curr
= afu
->hrrq_start
;
1649 rc
= init_global(cfg
);
1651 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1656 * init_mc() - create and register as the master context
1657 * @cfg: Internal structure associated with the host.
1659 * Return: 0 on success, -errno on failure
1661 static int init_mc(struct cxlflash_cfg
*cfg
)
1663 struct cxl_context
*ctx
;
1664 struct device
*dev
= &cfg
->dev
->dev
;
1665 struct afu
*afu
= cfg
->afu
;
1667 enum undo_level level
;
1669 ctx
= cxl_get_context(cfg
->dev
);
1674 /* Set it up as a master with the CXL */
1675 cxl_set_master(ctx
);
1677 /* During initialization reset the AFU to start from a clean slate */
1678 rc
= cxl_afu_reset(cfg
->mcctx
);
1680 dev_err(dev
, "%s: initial AFU reset failed rc=%d\n",
1682 level
= RELEASE_CONTEXT
;
1686 rc
= cxl_allocate_afu_irqs(ctx
, 3);
1688 dev_err(dev
, "%s: call to allocate_afu_irqs failed rc=%d!\n",
1690 level
= RELEASE_CONTEXT
;
1694 rc
= cxl_map_afu_irq(ctx
, 1, cxlflash_sync_err_irq
, afu
,
1695 "SISL_MSI_SYNC_ERROR");
1696 if (unlikely(rc
<= 0)) {
1697 dev_err(dev
, "%s: IRQ 1 (SISL_MSI_SYNC_ERROR) map failed!\n",
1703 rc
= cxl_map_afu_irq(ctx
, 2, cxlflash_rrq_irq
, afu
,
1704 "SISL_MSI_RRQ_UPDATED");
1705 if (unlikely(rc
<= 0)) {
1706 dev_err(dev
, "%s: IRQ 2 (SISL_MSI_RRQ_UPDATED) map failed!\n",
1712 rc
= cxl_map_afu_irq(ctx
, 3, cxlflash_async_err_irq
, afu
,
1713 "SISL_MSI_ASYNC_ERROR");
1714 if (unlikely(rc
<= 0)) {
1715 dev_err(dev
, "%s: IRQ 3 (SISL_MSI_ASYNC_ERROR) map failed!\n",
1723 /* This performs the equivalent of the CXL_IOCTL_START_WORK.
1724 * The CXL_IOCTL_GET_PROCESS_ELEMENT is implicit in the process
1725 * element (pe) that is embedded in the context (ctx)
1727 rc
= start_context(cfg
);
1729 dev_err(dev
, "%s: start context failed rc=%d\n", __func__
, rc
);
1730 level
= UNMAP_THREE
;
1734 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1737 term_mc(cfg
, level
);
1742 * init_afu() - setup as master context and start AFU
1743 * @cfg: Internal structure associated with the host.
1745 * This routine is a higher level of control for configuring the
1746 * AFU on probe and reset paths.
1748 * Return: 0 on success, -errno on failure
1750 static int init_afu(struct cxlflash_cfg
*cfg
)
1754 struct afu
*afu
= cfg
->afu
;
1755 struct device
*dev
= &cfg
->dev
->dev
;
1757 cxl_perst_reloads_same_image(cfg
->cxl_afu
, true);
1761 dev_err(dev
, "%s: call to init_mc failed, rc=%d!\n",
1766 /* Map the entire MMIO space of the AFU */
1767 afu
->afu_map
= cxl_psa_map(cfg
->mcctx
);
1768 if (!afu
->afu_map
) {
1769 dev_err(dev
, "%s: call to cxl_psa_map failed!\n", __func__
);
1773 kref_init(&afu
->mapcount
);
1775 /* No byte reverse on reading afu_version or string will be backwards */
1776 reg
= readq(&afu
->afu_map
->global
.regs
.afu_version
);
1777 memcpy(afu
->version
, ®
, sizeof(reg
));
1778 afu
->interface_version
=
1779 readq_be(&afu
->afu_map
->global
.regs
.interface_version
);
1780 if ((afu
->interface_version
+ 1) == 0) {
1781 pr_err("Back level AFU, please upgrade. AFU version %s "
1782 "interface version 0x%llx\n", afu
->version
,
1783 afu
->interface_version
);
1788 pr_debug("%s: afu version %s, interface version 0x%llX\n", __func__
,
1789 afu
->version
, afu
->interface_version
);
1791 rc
= start_afu(cfg
);
1793 dev_err(dev
, "%s: call to start_afu failed, rc=%d!\n",
1798 afu_err_intr_init(cfg
->afu
);
1799 atomic64_set(&afu
->room
, readq_be(&afu
->host_map
->cmd_room
));
1801 /* Restore the LUN mappings */
1802 cxlflash_restore_luntable(cfg
);
1804 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1808 kref_put(&afu
->mapcount
, afu_unmap
);
1810 term_mc(cfg
, UNDO_START
);
1815 * cxlflash_afu_sync() - builds and sends an AFU sync command
1816 * @afu: AFU associated with the host.
1817 * @ctx_hndl_u: Identifies context requesting sync.
1818 * @res_hndl_u: Identifies resource requesting sync.
1819 * @mode: Type of sync to issue (lightweight, heavyweight, global).
1821 * The AFU can only take 1 sync command at a time. This routine enforces this
1822 * limitation by using a mutex to provide exclusive access to the AFU during
1823 * the sync. This design point requires calling threads to not be on interrupt
1824 * context due to the possibility of sleeping during concurrent sync operations.
1826 * AFU sync operations are only necessary and allowed when the device is
1827 * operating normally. When not operating normally, sync requests can occur as
1828 * part of cleaning up resources associated with an adapter prior to removal.
1829 * In this scenario, these requests are simply ignored (safe due to the AFU
1836 int cxlflash_afu_sync(struct afu
*afu
, ctx_hndl_t ctx_hndl_u
,
1837 res_hndl_t res_hndl_u
, u8 mode
)
1839 struct cxlflash_cfg
*cfg
= afu
->parent
;
1840 struct device
*dev
= &cfg
->dev
->dev
;
1841 struct afu_cmd
*cmd
= NULL
;
1844 static DEFINE_MUTEX(sync_active
);
1846 if (cfg
->state
!= STATE_NORMAL
) {
1847 pr_debug("%s: Sync not required! (%u)\n", __func__
, cfg
->state
);
1851 mutex_lock(&sync_active
);
1853 cmd
= cmd_checkout(afu
);
1854 if (unlikely(!cmd
)) {
1856 udelay(1000 * retry_cnt
);
1857 if (retry_cnt
< MC_RETRY_CNT
)
1859 dev_err(dev
, "%s: could not get a free command\n", __func__
);
1864 pr_debug("%s: afu=%p cmd=%p %d\n", __func__
, afu
, cmd
, ctx_hndl_u
);
1866 memset(cmd
->rcb
.cdb
, 0, sizeof(cmd
->rcb
.cdb
));
1868 cmd
->rcb
.req_flags
= SISL_REQ_FLAGS_AFU_CMD
;
1869 cmd
->rcb
.port_sel
= 0x0; /* NA */
1870 cmd
->rcb
.lun_id
= 0x0; /* NA */
1871 cmd
->rcb
.data_len
= 0x0;
1872 cmd
->rcb
.data_ea
= 0x0;
1873 cmd
->rcb
.timeout
= MC_AFU_SYNC_TIMEOUT
;
1875 cmd
->rcb
.cdb
[0] = 0xC0; /* AFU Sync */
1876 cmd
->rcb
.cdb
[1] = mode
;
1878 /* The cdb is aligned, no unaligned accessors required */
1879 *((__be16
*)&cmd
->rcb
.cdb
[2]) = cpu_to_be16(ctx_hndl_u
);
1880 *((__be32
*)&cmd
->rcb
.cdb
[4]) = cpu_to_be32(res_hndl_u
);
1882 rc
= send_cmd(afu
, cmd
);
1886 wait_resp(afu
, cmd
);
1888 /* Set on timeout */
1889 if (unlikely((cmd
->sa
.ioasc
!= 0) ||
1890 (cmd
->sa
.host_use_b
[0] & B_ERROR
)))
1893 mutex_unlock(&sync_active
);
1896 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1901 * afu_reset() - resets the AFU
1902 * @cfg: Internal structure associated with the host.
1904 * Return: 0 on success, -errno on failure
1906 static int afu_reset(struct cxlflash_cfg
*cfg
)
1909 /* Stop the context before the reset. Since the context is
1910 * no longer available restart it after the reset is complete
1917 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1922 * cxlflash_eh_device_reset_handler() - reset a single LUN
1923 * @scp: SCSI command to send.
1926 * SUCCESS as defined in scsi/scsi.h
1927 * FAILED as defined in scsi/scsi.h
1929 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd
*scp
)
1932 struct Scsi_Host
*host
= scp
->device
->host
;
1933 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)host
->hostdata
;
1934 struct afu
*afu
= cfg
->afu
;
1937 pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
1938 "cdb=(%08X-%08X-%08X-%08X)\n", __func__
, scp
,
1939 host
->host_no
, scp
->device
->channel
,
1940 scp
->device
->id
, scp
->device
->lun
,
1941 get_unaligned_be32(&((u32
*)scp
->cmnd
)[0]),
1942 get_unaligned_be32(&((u32
*)scp
->cmnd
)[1]),
1943 get_unaligned_be32(&((u32
*)scp
->cmnd
)[2]),
1944 get_unaligned_be32(&((u32
*)scp
->cmnd
)[3]));
1947 switch (cfg
->state
) {
1949 rcr
= send_tmf(afu
, scp
, TMF_LUN_RESET
);
1954 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
);
1961 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
1966 * cxlflash_eh_host_reset_handler() - reset the host adapter
1967 * @scp: SCSI command from stack identifying host.
1969 * Following a reset, the state is evaluated again in case an EEH occurred
1970 * during the reset. In such a scenario, the host reset will either yield
1971 * until the EEH recovery is complete or return success or failure based
1972 * upon the current device state.
1975 * SUCCESS as defined in scsi/scsi.h
1976 * FAILED as defined in scsi/scsi.h
1978 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd
*scp
)
1982 struct Scsi_Host
*host
= scp
->device
->host
;
1983 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)host
->hostdata
;
1985 pr_debug("%s: (scp=%p) %d/%d/%d/%llu "
1986 "cdb=(%08X-%08X-%08X-%08X)\n", __func__
, scp
,
1987 host
->host_no
, scp
->device
->channel
,
1988 scp
->device
->id
, scp
->device
->lun
,
1989 get_unaligned_be32(&((u32
*)scp
->cmnd
)[0]),
1990 get_unaligned_be32(&((u32
*)scp
->cmnd
)[1]),
1991 get_unaligned_be32(&((u32
*)scp
->cmnd
)[2]),
1992 get_unaligned_be32(&((u32
*)scp
->cmnd
)[3]));
1994 switch (cfg
->state
) {
1996 cfg
->state
= STATE_RESET
;
1997 cxlflash_mark_contexts_error(cfg
);
1998 rcr
= afu_reset(cfg
);
2001 cfg
->state
= STATE_FAILTERM
;
2003 cfg
->state
= STATE_NORMAL
;
2004 wake_up_all(&cfg
->reset_waitq
);
2008 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
);
2009 if (cfg
->state
== STATE_NORMAL
)
2017 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
2022 * cxlflash_change_queue_depth() - change the queue depth for the device
2023 * @sdev: SCSI device destined for queue depth change.
2024 * @qdepth: Requested queue depth value to set.
2026 * The requested queue depth is capped to the maximum supported value.
2028 * Return: The actual queue depth set.
2030 static int cxlflash_change_queue_depth(struct scsi_device
*sdev
, int qdepth
)
2033 if (qdepth
> CXLFLASH_MAX_CMDS_PER_LUN
)
2034 qdepth
= CXLFLASH_MAX_CMDS_PER_LUN
;
2036 scsi_change_queue_depth(sdev
, qdepth
);
2037 return sdev
->queue_depth
;
2041 * cxlflash_show_port_status() - queries and presents the current port status
2042 * @port: Desired port for status reporting.
2043 * @afu: AFU owning the specified port.
2044 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2046 * Return: The size of the ASCII string returned in @buf.
2048 static ssize_t
cxlflash_show_port_status(u32 port
, struct afu
*afu
, char *buf
)
2052 __be64 __iomem
*fc_regs
;
2054 if (port
>= NUM_FC_PORTS
)
2057 fc_regs
= &afu
->afu_map
->global
.fc_regs
[port
][0];
2058 status
= readq_be(&fc_regs
[FC_MTIP_STATUS
/ 8]);
2059 status
&= FC_MTIP_STATUS_MASK
;
2061 if (status
== FC_MTIP_STATUS_ONLINE
)
2062 disp_status
= "online";
2063 else if (status
== FC_MTIP_STATUS_OFFLINE
)
2064 disp_status
= "offline";
2066 disp_status
= "unknown";
2068 return scnprintf(buf
, PAGE_SIZE
, "%s\n", disp_status
);
2072 * port0_show() - queries and presents the current status of port 0
2073 * @dev: Generic device associated with the host owning the port.
2074 * @attr: Device attribute representing the port.
2075 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2077 * Return: The size of the ASCII string returned in @buf.
2079 static ssize_t
port0_show(struct device
*dev
,
2080 struct device_attribute
*attr
,
2083 struct Scsi_Host
*shost
= class_to_shost(dev
);
2084 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)shost
->hostdata
;
2085 struct afu
*afu
= cfg
->afu
;
2087 return cxlflash_show_port_status(0, afu
, buf
);
2091 * port1_show() - queries and presents the current status of port 1
2092 * @dev: Generic device associated with the host owning the port.
2093 * @attr: Device attribute representing the port.
2094 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2096 * Return: The size of the ASCII string returned in @buf.
2098 static ssize_t
port1_show(struct device
*dev
,
2099 struct device_attribute
*attr
,
2102 struct Scsi_Host
*shost
= class_to_shost(dev
);
2103 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)shost
->hostdata
;
2104 struct afu
*afu
= cfg
->afu
;
2106 return cxlflash_show_port_status(1, afu
, buf
);
2110 * lun_mode_show() - presents the current LUN mode of the host
2111 * @dev: Generic device associated with the host.
2112 * @attr: Device attribute representing the LUN mode.
2113 * @buf: Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2115 * Return: The size of the ASCII string returned in @buf.
2117 static ssize_t
lun_mode_show(struct device
*dev
,
2118 struct device_attribute
*attr
, char *buf
)
2120 struct Scsi_Host
*shost
= class_to_shost(dev
);
2121 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)shost
->hostdata
;
2122 struct afu
*afu
= cfg
->afu
;
2124 return scnprintf(buf
, PAGE_SIZE
, "%u\n", afu
->internal_lun
);
2128 * lun_mode_store() - sets the LUN mode of the host
2129 * @dev: Generic device associated with the host.
2130 * @attr: Device attribute representing the LUN mode.
2131 * @buf: Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2132 * @count: Length of data resizing in @buf.
2134 * The CXL Flash AFU supports a dummy LUN mode where the external
2135 * links and storage are not required. Space on the FPGA is used
2136 * to create 1 or 2 small LUNs which are presented to the system
2137 * as if they were a normal storage device. This feature is useful
2138 * during development and also provides manufacturing with a way
2139 * to test the AFU without an actual device.
2141 * 0 = external LUN[s] (default)
2142 * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2143 * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2144 * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2145 * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2147 * Return: The size of the ASCII string returned in @buf.
2149 static ssize_t
lun_mode_store(struct device
*dev
,
2150 struct device_attribute
*attr
,
2151 const char *buf
, size_t count
)
2153 struct Scsi_Host
*shost
= class_to_shost(dev
);
2154 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)shost
->hostdata
;
2155 struct afu
*afu
= cfg
->afu
;
2159 rc
= kstrtouint(buf
, 10, &lun_mode
);
2160 if (!rc
&& (lun_mode
< 5) && (lun_mode
!= afu
->internal_lun
)) {
2161 afu
->internal_lun
= lun_mode
;
2164 * When configured for internal LUN, there is only one channel,
2165 * channel number 0, else there will be 2 (default).
2167 if (afu
->internal_lun
)
2168 shost
->max_channel
= 0;
2170 shost
->max_channel
= NUM_FC_PORTS
- 1;
2173 scsi_scan_host(cfg
->host
);
2180 * ioctl_version_show() - presents the current ioctl version of the host
2181 * @dev: Generic device associated with the host.
2182 * @attr: Device attribute representing the ioctl version.
2183 * @buf: Buffer of length PAGE_SIZE to report back the ioctl version.
2185 * Return: The size of the ASCII string returned in @buf.
2187 static ssize_t
ioctl_version_show(struct device
*dev
,
2188 struct device_attribute
*attr
, char *buf
)
2190 return scnprintf(buf
, PAGE_SIZE
, "%u\n", DK_CXLFLASH_VERSION_0
);
2194 * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2195 * @port: Desired port for status reporting.
2196 * @afu: AFU owning the specified port.
2197 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2199 * Return: The size of the ASCII string returned in @buf.
2201 static ssize_t
cxlflash_show_port_lun_table(u32 port
,
2207 __be64 __iomem
*fc_port
;
2209 if (port
>= NUM_FC_PORTS
)
2212 fc_port
= &afu
->afu_map
->global
.fc_port
[port
][0];
2214 for (i
= 0; i
< CXLFLASH_NUM_VLUNS
; i
++)
2215 bytes
+= scnprintf(buf
+ bytes
, PAGE_SIZE
- bytes
,
2216 "%03d: %016llX\n", i
, readq_be(&fc_port
[i
]));
2221 * port0_lun_table_show() - presents the current LUN table of port 0
2222 * @dev: Generic device associated with the host owning the port.
2223 * @attr: Device attribute representing the port.
2224 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2226 * Return: The size of the ASCII string returned in @buf.
2228 static ssize_t
port0_lun_table_show(struct device
*dev
,
2229 struct device_attribute
*attr
,
2232 struct Scsi_Host
*shost
= class_to_shost(dev
);
2233 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)shost
->hostdata
;
2234 struct afu
*afu
= cfg
->afu
;
2236 return cxlflash_show_port_lun_table(0, afu
, buf
);
2240 * port1_lun_table_show() - presents the current LUN table of port 1
2241 * @dev: Generic device associated with the host owning the port.
2242 * @attr: Device attribute representing the port.
2243 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2245 * Return: The size of the ASCII string returned in @buf.
2247 static ssize_t
port1_lun_table_show(struct device
*dev
,
2248 struct device_attribute
*attr
,
2251 struct Scsi_Host
*shost
= class_to_shost(dev
);
2252 struct cxlflash_cfg
*cfg
= (struct cxlflash_cfg
*)shost
->hostdata
;
2253 struct afu
*afu
= cfg
->afu
;
2255 return cxlflash_show_port_lun_table(1, afu
, buf
);
2259 * mode_show() - presents the current mode of the device
2260 * @dev: Generic device associated with the device.
2261 * @attr: Device attribute representing the device mode.
2262 * @buf: Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
2264 * Return: The size of the ASCII string returned in @buf.
2266 static ssize_t
mode_show(struct device
*dev
,
2267 struct device_attribute
*attr
, char *buf
)
2269 struct scsi_device
*sdev
= to_scsi_device(dev
);
2271 return scnprintf(buf
, PAGE_SIZE
, "%s\n",
2272 sdev
->hostdata
? "superpipe" : "legacy");
2278 static DEVICE_ATTR_RO(port0
);
2279 static DEVICE_ATTR_RO(port1
);
2280 static DEVICE_ATTR_RW(lun_mode
);
2281 static DEVICE_ATTR_RO(ioctl_version
);
2282 static DEVICE_ATTR_RO(port0_lun_table
);
2283 static DEVICE_ATTR_RO(port1_lun_table
);
2285 static struct device_attribute
*cxlflash_host_attrs
[] = {
2289 &dev_attr_ioctl_version
,
2290 &dev_attr_port0_lun_table
,
2291 &dev_attr_port1_lun_table
,
2298 static DEVICE_ATTR_RO(mode
);
2300 static struct device_attribute
*cxlflash_dev_attrs
[] = {
2308 static struct scsi_host_template driver_template
= {
2309 .module
= THIS_MODULE
,
2310 .name
= CXLFLASH_ADAPTER_NAME
,
2311 .info
= cxlflash_driver_info
,
2312 .ioctl
= cxlflash_ioctl
,
2313 .proc_name
= CXLFLASH_NAME
,
2314 .queuecommand
= cxlflash_queuecommand
,
2315 .eh_device_reset_handler
= cxlflash_eh_device_reset_handler
,
2316 .eh_host_reset_handler
= cxlflash_eh_host_reset_handler
,
2317 .change_queue_depth
= cxlflash_change_queue_depth
,
2318 .cmd_per_lun
= CXLFLASH_MAX_CMDS_PER_LUN
,
2319 .can_queue
= CXLFLASH_MAX_CMDS
,
2321 .sg_tablesize
= SG_NONE
, /* No scatter gather support */
2322 .max_sectors
= CXLFLASH_MAX_SECTORS
,
2323 .use_clustering
= ENABLE_CLUSTERING
,
2324 .shost_attrs
= cxlflash_host_attrs
,
2325 .sdev_attrs
= cxlflash_dev_attrs
,
2329 * Device dependent values
2331 static struct dev_dependent_vals dev_corsa_vals
= { CXLFLASH_MAX_SECTORS
};
2332 static struct dev_dependent_vals dev_flash_gt_vals
= { CXLFLASH_MAX_SECTORS
};
2335 * PCI device binding table
2337 static struct pci_device_id cxlflash_pci_table
[] = {
2338 {PCI_VENDOR_ID_IBM
, PCI_DEVICE_ID_IBM_CORSA
,
2339 PCI_ANY_ID
, PCI_ANY_ID
, 0, 0, (kernel_ulong_t
)&dev_corsa_vals
},
2340 {PCI_VENDOR_ID_IBM
, PCI_DEVICE_ID_IBM_FLASH_GT
,
2341 PCI_ANY_ID
, PCI_ANY_ID
, 0, 0, (kernel_ulong_t
)&dev_flash_gt_vals
},
2345 MODULE_DEVICE_TABLE(pci
, cxlflash_pci_table
);
2348 * cxlflash_worker_thread() - work thread handler for the AFU
2349 * @work: Work structure contained within cxlflash associated with host.
2351 * Handles the following events:
2352 * - Link reset which cannot be performed on interrupt context due to
2353 * blocking up to a few seconds
2354 * - Read AFU command room
2357 static void cxlflash_worker_thread(struct work_struct
*work
)
2359 struct cxlflash_cfg
*cfg
= container_of(work
, struct cxlflash_cfg
,
2361 struct afu
*afu
= cfg
->afu
;
2362 struct device
*dev
= &cfg
->dev
->dev
;
2366 /* Avoid MMIO if the device has failed */
2368 if (cfg
->state
!= STATE_NORMAL
)
2371 spin_lock_irqsave(cfg
->host
->host_lock
, lock_flags
);
2373 if (cfg
->lr_state
== LINK_RESET_REQUIRED
) {
2374 port
= cfg
->lr_port
;
2376 dev_err(dev
, "%s: invalid port index %d\n",
2379 spin_unlock_irqrestore(cfg
->host
->host_lock
,
2382 /* The reset can block... */
2383 afu_link_reset(afu
, port
,
2384 &afu
->afu_map
->global
.fc_regs
[port
][0]);
2385 spin_lock_irqsave(cfg
->host
->host_lock
, lock_flags
);
2388 cfg
->lr_state
= LINK_RESET_COMPLETE
;
2391 if (afu
->read_room
) {
2392 atomic64_set(&afu
->room
, readq_be(&afu
->host_map
->cmd_room
));
2393 afu
->read_room
= false;
2396 spin_unlock_irqrestore(cfg
->host
->host_lock
, lock_flags
);
2398 if (atomic_dec_if_positive(&cfg
->scan_host_needed
) >= 0)
2399 scsi_scan_host(cfg
->host
);
2400 kref_put(&afu
->mapcount
, afu_unmap
);
2404 * cxlflash_probe() - PCI entry point to add host
2405 * @pdev: PCI device associated with the host.
2406 * @dev_id: PCI device id associated with device.
2408 * Return: 0 on success, -errno on failure
2410 static int cxlflash_probe(struct pci_dev
*pdev
,
2411 const struct pci_device_id
*dev_id
)
2413 struct Scsi_Host
*host
;
2414 struct cxlflash_cfg
*cfg
= NULL
;
2415 struct device
*phys_dev
;
2416 struct dev_dependent_vals
*ddv
;
2419 dev_dbg(&pdev
->dev
, "%s: Found CXLFLASH with IRQ: %d\n",
2420 __func__
, pdev
->irq
);
2422 ddv
= (struct dev_dependent_vals
*)dev_id
->driver_data
;
2423 driver_template
.max_sectors
= ddv
->max_sectors
;
2425 host
= scsi_host_alloc(&driver_template
, sizeof(struct cxlflash_cfg
));
2427 dev_err(&pdev
->dev
, "%s: call to scsi_host_alloc failed!\n",
2433 host
->max_id
= CXLFLASH_MAX_NUM_TARGETS_PER_BUS
;
2434 host
->max_lun
= CXLFLASH_MAX_NUM_LUNS_PER_TARGET
;
2435 host
->max_channel
= NUM_FC_PORTS
- 1;
2436 host
->unique_id
= host
->host_no
;
2437 host
->max_cmd_len
= CXLFLASH_MAX_CDB_LEN
;
2439 cfg
= (struct cxlflash_cfg
*)host
->hostdata
;
2441 rc
= alloc_mem(cfg
);
2443 dev_err(&pdev
->dev
, "%s: call to alloc_mem failed!\n",
2446 scsi_host_put(cfg
->host
);
2450 cfg
->init_state
= INIT_STATE_NONE
;
2452 cfg
->cxl_fops
= cxlflash_cxl_fops
;
2455 * The promoted LUNs move to the top of the LUN table. The rest stay
2456 * on the bottom half. The bottom half grows from the end
2457 * (index = 255), whereas the top half grows from the beginning
2460 cfg
->promote_lun_index
= 0;
2461 cfg
->last_lun_index
[0] = CXLFLASH_NUM_VLUNS
/2 - 1;
2462 cfg
->last_lun_index
[1] = CXLFLASH_NUM_VLUNS
/2 - 1;
2464 cfg
->dev_id
= (struct pci_device_id
*)dev_id
;
2466 init_waitqueue_head(&cfg
->tmf_waitq
);
2467 init_waitqueue_head(&cfg
->reset_waitq
);
2469 INIT_WORK(&cfg
->work_q
, cxlflash_worker_thread
);
2470 cfg
->lr_state
= LINK_RESET_INVALID
;
2472 spin_lock_init(&cfg
->tmf_slock
);
2473 mutex_init(&cfg
->ctx_tbl_list_mutex
);
2474 mutex_init(&cfg
->ctx_recovery_mutex
);
2475 init_rwsem(&cfg
->ioctl_rwsem
);
2476 INIT_LIST_HEAD(&cfg
->ctx_err_recovery
);
2477 INIT_LIST_HEAD(&cfg
->lluns
);
2479 pci_set_drvdata(pdev
, cfg
);
2482 * Use the special service provided to look up the physical
2483 * PCI device, since we are called on the probe of the virtual
2484 * PCI host bus (vphb)
2486 phys_dev
= cxl_get_phys_dev(pdev
);
2487 if (!dev_is_pci(phys_dev
)) {
2488 dev_err(&pdev
->dev
, "%s: not a pci dev\n", __func__
);
2492 cfg
->parent_dev
= to_pci_dev(phys_dev
);
2494 cfg
->cxl_afu
= cxl_pci_to_afu(pdev
);
2498 dev_err(&pdev
->dev
, "%s: call to init_pci "
2499 "failed rc=%d!\n", __func__
, rc
);
2502 cfg
->init_state
= INIT_STATE_PCI
;
2506 dev_err(&pdev
->dev
, "%s: call to init_afu "
2507 "failed rc=%d!\n", __func__
, rc
);
2510 cfg
->init_state
= INIT_STATE_AFU
;
2512 rc
= init_scsi(cfg
);
2514 dev_err(&pdev
->dev
, "%s: call to init_scsi "
2515 "failed rc=%d!\n", __func__
, rc
);
2518 cfg
->init_state
= INIT_STATE_SCSI
;
2521 pr_debug("%s: returning rc=%d\n", __func__
, rc
);
2525 cxlflash_remove(pdev
);
2530 * drain_ioctls() - wait until all currently executing ioctls have completed
2531 * @cfg: Internal structure associated with the host.
2533 * Obtain write access to read/write semaphore that wraps ioctl
2534 * handling to 'drain' ioctls currently executing.
2536 static void drain_ioctls(struct cxlflash_cfg
*cfg
)
2538 down_write(&cfg
->ioctl_rwsem
);
2539 up_write(&cfg
->ioctl_rwsem
);
2543 * cxlflash_pci_error_detected() - called when a PCI error is detected
2544 * @pdev: PCI device struct.
2545 * @state: PCI channel state.
2547 * When an EEH occurs during an active reset, wait until the reset is
2548 * complete and then take action based upon the device state.
2550 * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
2552 static pci_ers_result_t
cxlflash_pci_error_detected(struct pci_dev
*pdev
,
2553 pci_channel_state_t state
)
2556 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
2557 struct device
*dev
= &cfg
->dev
->dev
;
2559 dev_dbg(dev
, "%s: pdev=%p state=%u\n", __func__
, pdev
, state
);
2562 case pci_channel_io_frozen
:
2563 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
);
2564 if (cfg
->state
== STATE_FAILTERM
)
2565 return PCI_ERS_RESULT_DISCONNECT
;
2567 cfg
->state
= STATE_RESET
;
2568 scsi_block_requests(cfg
->host
);
2570 rc
= cxlflash_mark_contexts_error(cfg
);
2572 dev_err(dev
, "%s: Failed to mark user contexts!(%d)\n",
2574 term_mc(cfg
, UNDO_START
);
2576 return PCI_ERS_RESULT_NEED_RESET
;
2577 case pci_channel_io_perm_failure
:
2578 cfg
->state
= STATE_FAILTERM
;
2579 wake_up_all(&cfg
->reset_waitq
);
2580 scsi_unblock_requests(cfg
->host
);
2581 return PCI_ERS_RESULT_DISCONNECT
;
2585 return PCI_ERS_RESULT_NEED_RESET
;
2589 * cxlflash_pci_slot_reset() - called when PCI slot has been reset
2590 * @pdev: PCI device struct.
2592 * This routine is called by the pci error recovery code after the PCI
2593 * slot has been reset, just before we should resume normal operations.
2595 * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
2597 static pci_ers_result_t
cxlflash_pci_slot_reset(struct pci_dev
*pdev
)
2600 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
2601 struct device
*dev
= &cfg
->dev
->dev
;
2603 dev_dbg(dev
, "%s: pdev=%p\n", __func__
, pdev
);
2607 dev_err(dev
, "%s: EEH recovery failed! (%d)\n", __func__
, rc
);
2608 return PCI_ERS_RESULT_DISCONNECT
;
2611 return PCI_ERS_RESULT_RECOVERED
;
2615 * cxlflash_pci_resume() - called when normal operation can resume
2616 * @pdev: PCI device struct
2618 static void cxlflash_pci_resume(struct pci_dev
*pdev
)
2620 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
2621 struct device
*dev
= &cfg
->dev
->dev
;
2623 dev_dbg(dev
, "%s: pdev=%p\n", __func__
, pdev
);
2625 cfg
->state
= STATE_NORMAL
;
2626 wake_up_all(&cfg
->reset_waitq
);
2627 scsi_unblock_requests(cfg
->host
);
2630 static const struct pci_error_handlers cxlflash_err_handler
= {
2631 .error_detected
= cxlflash_pci_error_detected
,
2632 .slot_reset
= cxlflash_pci_slot_reset
,
2633 .resume
= cxlflash_pci_resume
,
2637 * PCI device structure
2639 static struct pci_driver cxlflash_driver
= {
2640 .name
= CXLFLASH_NAME
,
2641 .id_table
= cxlflash_pci_table
,
2642 .probe
= cxlflash_probe
,
2643 .remove
= cxlflash_remove
,
2644 .err_handler
= &cxlflash_err_handler
,
2648 * init_cxlflash() - module entry point
2650 * Return: 0 on success, -errno on failure
2652 static int __init
init_cxlflash(void)
2654 pr_info("%s: IBM Power CXL Flash Adapter: %s\n",
2655 __func__
, CXLFLASH_DRIVER_DATE
);
2657 cxlflash_list_init();
2659 return pci_register_driver(&cxlflash_driver
);
2663 * exit_cxlflash() - module exit point
2665 static void __exit
exit_cxlflash(void)
2667 cxlflash_term_global_luns();
2668 cxlflash_free_errpage();
2670 pci_unregister_driver(&cxlflash_driver
);
2673 module_init(init_cxlflash
);
2674 module_exit(exit_cxlflash
);