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 * process_cmd_err() - command error handler
39 * @cmd: AFU command that experienced the error.
40 * @scp: SCSI command associated with the AFU command in error.
42 * Translates error bits from AFU command to SCSI command results.
44 static void process_cmd_err(struct afu_cmd
*cmd
, struct scsi_cmnd
*scp
)
46 struct afu
*afu
= cmd
->parent
;
47 struct cxlflash_cfg
*cfg
= afu
->parent
;
48 struct device
*dev
= &cfg
->dev
->dev
;
49 struct sisl_ioarcb
*ioarcb
;
50 struct sisl_ioasa
*ioasa
;
59 if (ioasa
->rc
.flags
& SISL_RC_FLAGS_UNDERRUN
) {
61 scsi_set_resid(scp
, resid
);
62 dev_dbg(dev
, "%s: cmd underrun cmd = %p scp = %p, resid = %d\n",
63 __func__
, cmd
, scp
, resid
);
66 if (ioasa
->rc
.flags
& SISL_RC_FLAGS_OVERRUN
) {
67 dev_dbg(dev
, "%s: cmd underrun cmd = %p scp = %p\n",
69 scp
->result
= (DID_ERROR
<< 16);
72 dev_dbg(dev
, "%s: cmd failed afu_rc=%02x scsi_rc=%02x fc_rc=%02x "
73 "afu_extra=%02x scsi_extra=%02x fc_extra=%02x\n", __func__
,
74 ioasa
->rc
.afu_rc
, ioasa
->rc
.scsi_rc
, ioasa
->rc
.fc_rc
,
75 ioasa
->afu_extra
, ioasa
->scsi_extra
, ioasa
->fc_extra
);
77 if (ioasa
->rc
.scsi_rc
) {
78 /* We have a SCSI status */
79 if (ioasa
->rc
.flags
& SISL_RC_FLAGS_SENSE_VALID
) {
80 memcpy(scp
->sense_buffer
, ioasa
->sense_data
,
82 scp
->result
= ioasa
->rc
.scsi_rc
;
84 scp
->result
= ioasa
->rc
.scsi_rc
| (DID_ERROR
<< 16);
88 * We encountered an error. Set scp->result based on nature
91 if (ioasa
->rc
.fc_rc
) {
92 /* We have an FC status */
93 switch (ioasa
->rc
.fc_rc
) {
94 case SISL_FC_RC_LINKDOWN
:
95 scp
->result
= (DID_REQUEUE
<< 16);
97 case SISL_FC_RC_RESID
:
98 /* This indicates an FCP resid underrun */
99 if (!(ioasa
->rc
.flags
& SISL_RC_FLAGS_OVERRUN
)) {
100 /* If the SISL_RC_FLAGS_OVERRUN flag was set,
101 * then we will handle this error else where.
102 * If not then we must handle it here.
103 * This is probably an AFU bug.
105 scp
->result
= (DID_ERROR
<< 16);
108 case SISL_FC_RC_RESIDERR
:
109 /* Resid mismatch between adapter and device */
110 case SISL_FC_RC_TGTABORT
:
111 case SISL_FC_RC_ABORTOK
:
112 case SISL_FC_RC_ABORTFAIL
:
113 case SISL_FC_RC_NOLOGI
:
114 case SISL_FC_RC_ABORTPEND
:
115 case SISL_FC_RC_WRABORTPEND
:
116 case SISL_FC_RC_NOEXP
:
117 case SISL_FC_RC_INUSE
:
118 scp
->result
= (DID_ERROR
<< 16);
123 if (ioasa
->rc
.afu_rc
) {
124 /* We have an AFU error */
125 switch (ioasa
->rc
.afu_rc
) {
126 case SISL_AFU_RC_NO_CHANNELS
:
127 scp
->result
= (DID_NO_CONNECT
<< 16);
129 case SISL_AFU_RC_DATA_DMA_ERR
:
130 switch (ioasa
->afu_extra
) {
131 case SISL_AFU_DMA_ERR_PAGE_IN
:
133 scp
->result
= (DID_IMM_RETRY
<< 16);
135 case SISL_AFU_DMA_ERR_INVALID_EA
:
137 scp
->result
= (DID_ERROR
<< 16);
140 case SISL_AFU_RC_OUT_OF_DATA_BUFS
:
142 scp
->result
= (DID_ALLOC_FAILURE
<< 16);
145 scp
->result
= (DID_ERROR
<< 16);
151 * cmd_complete() - command completion handler
152 * @cmd: AFU command that has completed.
154 * Prepares and submits command that has either completed or timed out to
155 * the SCSI stack. Checks AFU command back into command pool for non-internal
156 * (cmd->scp populated) commands.
158 static void cmd_complete(struct afu_cmd
*cmd
)
160 struct scsi_cmnd
*scp
;
162 struct afu
*afu
= cmd
->parent
;
163 struct cxlflash_cfg
*cfg
= afu
->parent
;
164 struct device
*dev
= &cfg
->dev
->dev
;
169 if (unlikely(cmd
->sa
.ioasc
))
170 process_cmd_err(cmd
, scp
);
172 scp
->result
= (DID_OK
<< 16);
174 cmd_is_tmf
= cmd
->cmd_tmf
;
176 dev_dbg_ratelimited(dev
, "%s:scp=%p result=%08x ioasc=%08x\n",
177 __func__
, scp
, scp
->result
, cmd
->sa
.ioasc
);
182 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
183 cfg
->tmf_active
= false;
184 wake_up_all_locked(&cfg
->tmf_waitq
);
185 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
188 complete(&cmd
->cevent
);
192 * context_reset() - reset context via specified register
193 * @hwq: Hardware queue owning the context to be reset.
194 * @reset_reg: MMIO register to perform reset.
196 * Return: 0 on success, -errno on failure
198 static int context_reset(struct hwq
*hwq
, __be64 __iomem
*reset_reg
)
200 struct cxlflash_cfg
*cfg
= hwq
->afu
->parent
;
201 struct device
*dev
= &cfg
->dev
->dev
;
206 dev_dbg(dev
, "%s: hwq=%p\n", __func__
, hwq
);
208 writeq_be(val
, reset_reg
);
210 val
= readq_be(reset_reg
);
211 if ((val
& 0x1) == 0x0) {
216 /* Double delay each time */
218 } while (nretry
++ < MC_ROOM_RETRY_CNT
);
220 dev_dbg(dev
, "%s: returning rc=%d, val=%016llx nretry=%d\n",
221 __func__
, rc
, val
, nretry
);
226 * context_reset_ioarrin() - reset context via IOARRIN register
227 * @hwq: Hardware queue owning the context to be reset.
229 * Return: 0 on success, -errno on failure
231 static int context_reset_ioarrin(struct hwq
*hwq
)
233 return context_reset(hwq
, &hwq
->host_map
->ioarrin
);
237 * context_reset_sq() - reset context via SQ_CONTEXT_RESET register
238 * @hwq: Hardware queue owning the context to be reset.
240 * Return: 0 on success, -errno on failure
242 static int context_reset_sq(struct hwq
*hwq
)
244 return context_reset(hwq
, &hwq
->host_map
->sq_ctx_reset
);
248 * send_cmd_ioarrin() - sends an AFU command via IOARRIN register
249 * @afu: AFU associated with the host.
250 * @cmd: AFU command to send.
253 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
255 static int send_cmd_ioarrin(struct afu
*afu
, struct afu_cmd
*cmd
)
257 struct cxlflash_cfg
*cfg
= afu
->parent
;
258 struct device
*dev
= &cfg
->dev
->dev
;
259 struct hwq
*hwq
= get_hwq(afu
, cmd
->hwq_index
);
265 * To avoid the performance penalty of MMIO, spread the update of
266 * 'room' over multiple commands.
268 spin_lock_irqsave(&hwq
->hsq_slock
, lock_flags
);
269 if (--hwq
->room
< 0) {
270 room
= readq_be(&hwq
->host_map
->cmd_room
);
272 dev_dbg_ratelimited(dev
, "%s: no cmd_room to send "
273 "0x%02X, room=0x%016llX\n",
274 __func__
, cmd
->rcb
.cdb
[0], room
);
276 rc
= SCSI_MLQUEUE_HOST_BUSY
;
279 hwq
->room
= room
- 1;
282 writeq_be((u64
)&cmd
->rcb
, &hwq
->host_map
->ioarrin
);
284 spin_unlock_irqrestore(&hwq
->hsq_slock
, lock_flags
);
285 dev_dbg(dev
, "%s: cmd=%p len=%u ea=%016llx rc=%d\n", __func__
,
286 cmd
, cmd
->rcb
.data_len
, cmd
->rcb
.data_ea
, rc
);
291 * send_cmd_sq() - sends an AFU command via SQ ring
292 * @afu: AFU associated with the host.
293 * @cmd: AFU command to send.
296 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
298 static int send_cmd_sq(struct afu
*afu
, struct afu_cmd
*cmd
)
300 struct cxlflash_cfg
*cfg
= afu
->parent
;
301 struct device
*dev
= &cfg
->dev
->dev
;
302 struct hwq
*hwq
= get_hwq(afu
, cmd
->hwq_index
);
307 newval
= atomic_dec_if_positive(&hwq
->hsq_credits
);
309 rc
= SCSI_MLQUEUE_HOST_BUSY
;
313 cmd
->rcb
.ioasa
= &cmd
->sa
;
315 spin_lock_irqsave(&hwq
->hsq_slock
, lock_flags
);
317 *hwq
->hsq_curr
= cmd
->rcb
;
318 if (hwq
->hsq_curr
< hwq
->hsq_end
)
321 hwq
->hsq_curr
= hwq
->hsq_start
;
322 writeq_be((u64
)hwq
->hsq_curr
, &hwq
->host_map
->sq_tail
);
324 spin_unlock_irqrestore(&hwq
->hsq_slock
, lock_flags
);
326 dev_dbg(dev
, "%s: cmd=%p len=%u ea=%016llx ioasa=%p rc=%d curr=%p "
327 "head=%016llx tail=%016llx\n", __func__
, cmd
, cmd
->rcb
.data_len
,
328 cmd
->rcb
.data_ea
, cmd
->rcb
.ioasa
, rc
, hwq
->hsq_curr
,
329 readq_be(&hwq
->host_map
->sq_head
),
330 readq_be(&hwq
->host_map
->sq_tail
));
335 * wait_resp() - polls for a response or timeout to a sent AFU command
336 * @afu: AFU associated with the host.
337 * @cmd: AFU command that was sent.
339 * Return: 0 on success, -errno on failure
341 static int wait_resp(struct afu
*afu
, struct afu_cmd
*cmd
)
343 struct cxlflash_cfg
*cfg
= afu
->parent
;
344 struct device
*dev
= &cfg
->dev
->dev
;
346 ulong timeout
= msecs_to_jiffies(cmd
->rcb
.timeout
* 2 * 1000);
348 timeout
= wait_for_completion_timeout(&cmd
->cevent
, timeout
);
352 if (unlikely(cmd
->sa
.ioasc
!= 0)) {
353 dev_err(dev
, "%s: cmd %02x failed, ioasc=%08x\n",
354 __func__
, cmd
->rcb
.cdb
[0], cmd
->sa
.ioasc
);
362 * cmd_to_target_hwq() - selects a target hardware queue for a SCSI command
363 * @host: SCSI host associated with device.
364 * @scp: SCSI command to send.
365 * @afu: SCSI command to send.
367 * Hashes a command based upon the hardware queue mode.
369 * Return: Trusted index of target hardware queue
371 static u32
cmd_to_target_hwq(struct Scsi_Host
*host
, struct scsi_cmnd
*scp
,
377 if (afu
->num_hwqs
== 1)
380 switch (afu
->hwq_mode
) {
382 hwq
= afu
->hwq_rr_count
++ % afu
->num_hwqs
;
385 tag
= blk_mq_unique_tag(scp
->request
);
386 hwq
= blk_mq_unique_tag_to_hwq(tag
);
389 hwq
= smp_processor_id() % afu
->num_hwqs
;
399 * send_tmf() - sends a Task Management Function (TMF)
400 * @afu: AFU to checkout from.
401 * @scp: SCSI command from stack.
402 * @tmfcmd: TMF command to send.
405 * 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
407 static int send_tmf(struct afu
*afu
, struct scsi_cmnd
*scp
, u64 tmfcmd
)
409 struct Scsi_Host
*host
= scp
->device
->host
;
410 struct cxlflash_cfg
*cfg
= shost_priv(host
);
411 struct afu_cmd
*cmd
= sc_to_afucz(scp
);
412 struct device
*dev
= &cfg
->dev
->dev
;
413 int hwq_index
= cmd_to_target_hwq(host
, scp
, afu
);
414 struct hwq
*hwq
= get_hwq(afu
, hwq_index
);
419 /* When Task Management Function is active do not send another */
420 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
422 wait_event_interruptible_lock_irq(cfg
->tmf_waitq
,
425 cfg
->tmf_active
= true;
426 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
431 cmd
->hwq_index
= hwq_index
;
433 cmd
->rcb
.ctx_id
= hwq
->ctx_hndl
;
434 cmd
->rcb
.msi
= SISL_MSI_RRQ_UPDATED
;
435 cmd
->rcb
.port_sel
= CHAN2PORTMASK(scp
->device
->channel
);
436 cmd
->rcb
.lun_id
= lun_to_lunid(scp
->device
->lun
);
437 cmd
->rcb
.req_flags
= (SISL_REQ_FLAGS_PORT_LUN_ID
|
438 SISL_REQ_FLAGS_SUP_UNDERRUN
|
439 SISL_REQ_FLAGS_TMF_CMD
);
440 memcpy(cmd
->rcb
.cdb
, &tmfcmd
, sizeof(tmfcmd
));
442 rc
= afu
->send_cmd(afu
, cmd
);
444 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
445 cfg
->tmf_active
= false;
446 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
450 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
451 to
= msecs_to_jiffies(5000);
452 to
= wait_event_interruptible_lock_irq_timeout(cfg
->tmf_waitq
,
457 cfg
->tmf_active
= false;
458 dev_err(dev
, "%s: TMF timed out\n", __func__
);
461 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
467 * cxlflash_driver_info() - information handler for this host driver
468 * @host: SCSI host associated with device.
470 * Return: A string describing the device.
472 static const char *cxlflash_driver_info(struct Scsi_Host
*host
)
474 return CXLFLASH_ADAPTER_NAME
;
478 * cxlflash_queuecommand() - sends a mid-layer request
479 * @host: SCSI host associated with device.
480 * @scp: SCSI command to send.
482 * Return: 0 on success, SCSI_MLQUEUE_HOST_BUSY on failure
484 static int cxlflash_queuecommand(struct Scsi_Host
*host
, struct scsi_cmnd
*scp
)
486 struct cxlflash_cfg
*cfg
= shost_priv(host
);
487 struct afu
*afu
= cfg
->afu
;
488 struct device
*dev
= &cfg
->dev
->dev
;
489 struct afu_cmd
*cmd
= sc_to_afucz(scp
);
490 struct scatterlist
*sg
= scsi_sglist(scp
);
491 int hwq_index
= cmd_to_target_hwq(host
, scp
, afu
);
492 struct hwq
*hwq
= get_hwq(afu
, hwq_index
);
493 u16 req_flags
= SISL_REQ_FLAGS_SUP_UNDERRUN
;
497 dev_dbg_ratelimited(dev
, "%s: (scp=%p) %d/%d/%d/%llu "
498 "cdb=(%08x-%08x-%08x-%08x)\n",
499 __func__
, scp
, host
->host_no
, scp
->device
->channel
,
500 scp
->device
->id
, scp
->device
->lun
,
501 get_unaligned_be32(&((u32
*)scp
->cmnd
)[0]),
502 get_unaligned_be32(&((u32
*)scp
->cmnd
)[1]),
503 get_unaligned_be32(&((u32
*)scp
->cmnd
)[2]),
504 get_unaligned_be32(&((u32
*)scp
->cmnd
)[3]));
507 * If a Task Management Function is active, wait for it to complete
508 * before continuing with regular commands.
510 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
511 if (cfg
->tmf_active
) {
512 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
513 rc
= SCSI_MLQUEUE_HOST_BUSY
;
516 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
518 switch (cfg
->state
) {
522 dev_dbg_ratelimited(dev
, "%s: device is in reset\n", __func__
);
523 rc
= SCSI_MLQUEUE_HOST_BUSY
;
526 dev_dbg_ratelimited(dev
, "%s: device has failed\n", __func__
);
527 scp
->result
= (DID_NO_CONNECT
<< 16);
536 cmd
->rcb
.data_len
= sg
->length
;
537 cmd
->rcb
.data_ea
= (uintptr_t)sg_virt(sg
);
542 cmd
->hwq_index
= hwq_index
;
544 cmd
->rcb
.ctx_id
= hwq
->ctx_hndl
;
545 cmd
->rcb
.msi
= SISL_MSI_RRQ_UPDATED
;
546 cmd
->rcb
.port_sel
= CHAN2PORTMASK(scp
->device
->channel
);
547 cmd
->rcb
.lun_id
= lun_to_lunid(scp
->device
->lun
);
549 if (scp
->sc_data_direction
== DMA_TO_DEVICE
)
550 req_flags
|= SISL_REQ_FLAGS_HOST_WRITE
;
552 cmd
->rcb
.req_flags
= req_flags
;
553 memcpy(cmd
->rcb
.cdb
, scp
->cmnd
, sizeof(cmd
->rcb
.cdb
));
555 rc
= afu
->send_cmd(afu
, cmd
);
561 * cxlflash_wait_for_pci_err_recovery() - wait for error recovery during probe
562 * @cfg: Internal structure associated with the host.
564 static void cxlflash_wait_for_pci_err_recovery(struct cxlflash_cfg
*cfg
)
566 struct pci_dev
*pdev
= cfg
->dev
;
568 if (pci_channel_offline(pdev
))
569 wait_event_timeout(cfg
->reset_waitq
,
570 !pci_channel_offline(pdev
),
571 CXLFLASH_PCI_ERROR_RECOVERY_TIMEOUT
);
575 * free_mem() - free memory associated with the AFU
576 * @cfg: Internal structure associated with the host.
578 static void free_mem(struct cxlflash_cfg
*cfg
)
580 struct afu
*afu
= cfg
->afu
;
583 free_pages((ulong
)afu
, get_order(sizeof(struct afu
)));
589 * cxlflash_reset_sync() - synchronizing point for asynchronous resets
590 * @cfg: Internal structure associated with the host.
592 static void cxlflash_reset_sync(struct cxlflash_cfg
*cfg
)
594 if (cfg
->async_reset_cookie
== 0)
597 /* Wait until all async calls prior to this cookie have completed */
598 async_synchronize_cookie(cfg
->async_reset_cookie
+ 1);
599 cfg
->async_reset_cookie
= 0;
603 * stop_afu() - stops the AFU command timers and unmaps the MMIO space
604 * @cfg: Internal structure associated with the host.
606 * Safe to call with AFU in a partially allocated/initialized state.
608 * Cancels scheduled worker threads, waits for any active internal AFU
609 * commands to timeout, disables IRQ polling and then unmaps the MMIO space.
611 static void stop_afu(struct cxlflash_cfg
*cfg
)
613 struct afu
*afu
= cfg
->afu
;
617 cancel_work_sync(&cfg
->work_q
);
618 if (!current_is_async())
619 cxlflash_reset_sync(cfg
);
622 while (atomic_read(&afu
->cmds_active
))
625 if (afu_is_irqpoll_enabled(afu
)) {
626 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
627 hwq
= get_hwq(afu
, i
);
629 irq_poll_disable(&hwq
->irqpoll
);
633 if (likely(afu
->afu_map
)) {
634 cxl_psa_unmap((void __iomem
*)afu
->afu_map
);
641 * term_intr() - disables all AFU interrupts
642 * @cfg: Internal structure associated with the host.
643 * @level: Depth of allocation, where to begin waterfall tear down.
644 * @index: Index of the hardware queue.
646 * Safe to call with AFU/MC in partially allocated/initialized state.
648 static void term_intr(struct cxlflash_cfg
*cfg
, enum undo_level level
,
651 struct afu
*afu
= cfg
->afu
;
652 struct device
*dev
= &cfg
->dev
->dev
;
656 dev_err(dev
, "%s: returning with NULL afu\n", __func__
);
660 hwq
= get_hwq(afu
, index
);
663 dev_err(dev
, "%s: returning with NULL MC\n", __func__
);
669 /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
670 if (index
== PRIMARY_HWQ
)
671 cxl_unmap_afu_irq(hwq
->ctx
, 3, hwq
);
673 cxl_unmap_afu_irq(hwq
->ctx
, 2, hwq
);
675 cxl_unmap_afu_irq(hwq
->ctx
, 1, hwq
);
677 cxl_free_afu_irqs(hwq
->ctx
);
680 /* No action required */
686 * term_mc() - terminates the master context
687 * @cfg: Internal structure associated with the host.
688 * @index: Index of the hardware queue.
690 * Safe to call with AFU/MC in partially allocated/initialized state.
692 static void term_mc(struct cxlflash_cfg
*cfg
, u32 index
)
694 struct afu
*afu
= cfg
->afu
;
695 struct device
*dev
= &cfg
->dev
->dev
;
699 dev_err(dev
, "%s: returning with NULL afu\n", __func__
);
703 hwq
= get_hwq(afu
, index
);
706 dev_err(dev
, "%s: returning with NULL MC\n", __func__
);
710 WARN_ON(cxl_stop_context(hwq
->ctx
));
711 if (index
!= PRIMARY_HWQ
)
712 WARN_ON(cxl_release_context(hwq
->ctx
));
717 * term_afu() - terminates the AFU
718 * @cfg: Internal structure associated with the host.
720 * Safe to call with AFU/MC in partially allocated/initialized state.
722 static void term_afu(struct cxlflash_cfg
*cfg
)
724 struct device
*dev
= &cfg
->dev
->dev
;
728 * Tear down is carefully orchestrated to ensure
729 * no interrupts can come in when the problem state
732 * 1) Disable all AFU interrupts for each master
733 * 2) Unmap the problem state area
734 * 3) Stop each master context
736 for (k
= cfg
->afu
->num_hwqs
- 1; k
>= 0; k
--)
737 term_intr(cfg
, UNMAP_THREE
, k
);
742 for (k
= cfg
->afu
->num_hwqs
- 1; k
>= 0; k
--)
745 dev_dbg(dev
, "%s: returning\n", __func__
);
749 * notify_shutdown() - notifies device of pending shutdown
750 * @cfg: Internal structure associated with the host.
751 * @wait: Whether to wait for shutdown processing to complete.
753 * This function will notify the AFU that the adapter is being shutdown
754 * and will wait for shutdown processing to complete if wait is true.
755 * This notification should flush pending I/Os to the device and halt
756 * further I/Os until the next AFU reset is issued and device restarted.
758 static void notify_shutdown(struct cxlflash_cfg
*cfg
, bool wait
)
760 struct afu
*afu
= cfg
->afu
;
761 struct device
*dev
= &cfg
->dev
->dev
;
762 struct dev_dependent_vals
*ddv
;
763 __be64 __iomem
*fc_port_regs
;
765 int i
, retry_cnt
= 0;
767 ddv
= (struct dev_dependent_vals
*)cfg
->dev_id
->driver_data
;
768 if (!(ddv
->flags
& CXLFLASH_NOTIFY_SHUTDOWN
))
771 if (!afu
|| !afu
->afu_map
) {
772 dev_dbg(dev
, "%s: Problem state area not mapped\n", __func__
);
777 for (i
= 0; i
< cfg
->num_fc_ports
; i
++) {
778 fc_port_regs
= get_fc_port_regs(cfg
, i
);
780 reg
= readq_be(&fc_port_regs
[FC_CONFIG2
/ 8]);
781 reg
|= SISL_FC_SHUTDOWN_NORMAL
;
782 writeq_be(reg
, &fc_port_regs
[FC_CONFIG2
/ 8]);
788 /* Wait up to 1.5 seconds for shutdown processing to complete */
789 for (i
= 0; i
< cfg
->num_fc_ports
; i
++) {
790 fc_port_regs
= get_fc_port_regs(cfg
, i
);
794 status
= readq_be(&fc_port_regs
[FC_STATUS
/ 8]);
795 if (status
& SISL_STATUS_SHUTDOWN_COMPLETE
)
797 if (++retry_cnt
>= MC_RETRY_CNT
) {
798 dev_dbg(dev
, "%s: port %d shutdown processing "
799 "not yet completed\n", __func__
, i
);
802 msleep(100 * retry_cnt
);
808 * cxlflash_remove() - PCI entry point to tear down host
809 * @pdev: PCI device associated with the host.
811 * Safe to use as a cleanup in partially allocated/initialized state. Note that
812 * the reset_waitq is flushed as part of the stop/termination of user contexts.
814 static void cxlflash_remove(struct pci_dev
*pdev
)
816 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
817 struct device
*dev
= &pdev
->dev
;
820 if (!pci_is_enabled(pdev
)) {
821 dev_dbg(dev
, "%s: Device is disabled\n", __func__
);
825 /* If a Task Management Function is active, wait for it to complete
826 * before continuing with remove.
828 spin_lock_irqsave(&cfg
->tmf_slock
, lock_flags
);
830 wait_event_interruptible_lock_irq(cfg
->tmf_waitq
,
833 spin_unlock_irqrestore(&cfg
->tmf_slock
, lock_flags
);
835 /* Notify AFU and wait for shutdown processing to complete */
836 notify_shutdown(cfg
, true);
838 cfg
->state
= STATE_FAILTERM
;
839 cxlflash_stop_term_user_contexts(cfg
);
841 switch (cfg
->init_state
) {
842 case INIT_STATE_SCSI
:
843 cxlflash_term_local_luns(cfg
);
844 scsi_remove_host(cfg
->host
);
848 pci_disable_device(pdev
);
849 case INIT_STATE_NONE
:
851 scsi_host_put(cfg
->host
);
855 dev_dbg(dev
, "%s: returning\n", __func__
);
859 * alloc_mem() - allocates the AFU and its command pool
860 * @cfg: Internal structure associated with the host.
862 * A partially allocated state remains on failure.
866 * -ENOMEM on failure to allocate memory
868 static int alloc_mem(struct cxlflash_cfg
*cfg
)
871 struct device
*dev
= &cfg
->dev
->dev
;
873 /* AFU is ~28k, i.e. only one 64k page or up to seven 4k pages */
874 cfg
->afu
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
875 get_order(sizeof(struct afu
)));
876 if (unlikely(!cfg
->afu
)) {
877 dev_err(dev
, "%s: cannot get %d free pages\n",
878 __func__
, get_order(sizeof(struct afu
)));
882 cfg
->afu
->parent
= cfg
;
883 cfg
->afu
->desired_hwqs
= CXLFLASH_DEF_HWQS
;
884 cfg
->afu
->afu_map
= NULL
;
890 * init_pci() - initializes the host as a PCI device
891 * @cfg: Internal structure associated with the host.
893 * Return: 0 on success, -errno on failure
895 static int init_pci(struct cxlflash_cfg
*cfg
)
897 struct pci_dev
*pdev
= cfg
->dev
;
898 struct device
*dev
= &cfg
->dev
->dev
;
901 rc
= pci_enable_device(pdev
);
902 if (rc
|| pci_channel_offline(pdev
)) {
903 if (pci_channel_offline(pdev
)) {
904 cxlflash_wait_for_pci_err_recovery(cfg
);
905 rc
= pci_enable_device(pdev
);
909 dev_err(dev
, "%s: Cannot enable adapter\n", __func__
);
910 cxlflash_wait_for_pci_err_recovery(cfg
);
916 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
921 * init_scsi() - adds the host to the SCSI stack and kicks off host scan
922 * @cfg: Internal structure associated with the host.
924 * Return: 0 on success, -errno on failure
926 static int init_scsi(struct cxlflash_cfg
*cfg
)
928 struct pci_dev
*pdev
= cfg
->dev
;
929 struct device
*dev
= &cfg
->dev
->dev
;
932 rc
= scsi_add_host(cfg
->host
, &pdev
->dev
);
934 dev_err(dev
, "%s: scsi_add_host failed rc=%d\n", __func__
, rc
);
938 scsi_scan_host(cfg
->host
);
941 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
946 * set_port_online() - transitions the specified host FC port to online state
947 * @fc_regs: Top of MMIO region defined for specified port.
949 * The provided MMIO region must be mapped prior to call. Online state means
950 * that the FC link layer has synced, completed the handshaking process, and
951 * is ready for login to start.
953 static void set_port_online(__be64 __iomem
*fc_regs
)
957 cmdcfg
= readq_be(&fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
958 cmdcfg
&= (~FC_MTIP_CMDCONFIG_OFFLINE
); /* clear OFF_LINE */
959 cmdcfg
|= (FC_MTIP_CMDCONFIG_ONLINE
); /* set ON_LINE */
960 writeq_be(cmdcfg
, &fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
964 * set_port_offline() - transitions the specified host FC port to offline state
965 * @fc_regs: Top of MMIO region defined for specified port.
967 * The provided MMIO region must be mapped prior to call.
969 static void set_port_offline(__be64 __iomem
*fc_regs
)
973 cmdcfg
= readq_be(&fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
974 cmdcfg
&= (~FC_MTIP_CMDCONFIG_ONLINE
); /* clear ON_LINE */
975 cmdcfg
|= (FC_MTIP_CMDCONFIG_OFFLINE
); /* set OFF_LINE */
976 writeq_be(cmdcfg
, &fc_regs
[FC_MTIP_CMDCONFIG
/ 8]);
980 * wait_port_online() - waits for the specified host FC port come online
981 * @fc_regs: Top of MMIO region defined for specified port.
982 * @delay_us: Number of microseconds to delay between reading port status.
983 * @nretry: Number of cycles to retry reading port status.
985 * The provided MMIO region must be mapped prior to call. This will timeout
986 * when the cable is not plugged in.
989 * TRUE (1) when the specified port is online
990 * FALSE (0) when the specified port fails to come online after timeout
992 static bool wait_port_online(__be64 __iomem
*fc_regs
, u32 delay_us
, u32 nretry
)
996 WARN_ON(delay_us
< 1000);
999 msleep(delay_us
/ 1000);
1000 status
= readq_be(&fc_regs
[FC_MTIP_STATUS
/ 8]);
1001 if (status
== U64_MAX
)
1003 } while ((status
& FC_MTIP_STATUS_MASK
) != FC_MTIP_STATUS_ONLINE
&&
1006 return ((status
& FC_MTIP_STATUS_MASK
) == FC_MTIP_STATUS_ONLINE
);
1010 * wait_port_offline() - waits for the specified host FC port go offline
1011 * @fc_regs: Top of MMIO region defined for specified port.
1012 * @delay_us: Number of microseconds to delay between reading port status.
1013 * @nretry: Number of cycles to retry reading port status.
1015 * The provided MMIO region must be mapped prior to call.
1018 * TRUE (1) when the specified port is offline
1019 * FALSE (0) when the specified port fails to go offline after timeout
1021 static bool wait_port_offline(__be64 __iomem
*fc_regs
, u32 delay_us
, u32 nretry
)
1025 WARN_ON(delay_us
< 1000);
1028 msleep(delay_us
/ 1000);
1029 status
= readq_be(&fc_regs
[FC_MTIP_STATUS
/ 8]);
1030 if (status
== U64_MAX
)
1032 } while ((status
& FC_MTIP_STATUS_MASK
) != FC_MTIP_STATUS_OFFLINE
&&
1035 return ((status
& FC_MTIP_STATUS_MASK
) == FC_MTIP_STATUS_OFFLINE
);
1039 * afu_set_wwpn() - configures the WWPN for the specified host FC port
1040 * @afu: AFU associated with the host that owns the specified FC port.
1041 * @port: Port number being configured.
1042 * @fc_regs: Top of MMIO region defined for specified port.
1043 * @wwpn: The world-wide-port-number previously discovered for port.
1045 * The provided MMIO region must be mapped prior to call. As part of the
1046 * sequence to configure the WWPN, the port is toggled offline and then back
1047 * online. This toggling action can cause this routine to delay up to a few
1048 * seconds. When configured to use the internal LUN feature of the AFU, a
1049 * failure to come online is overridden.
1051 static void afu_set_wwpn(struct afu
*afu
, int port
, __be64 __iomem
*fc_regs
,
1054 struct cxlflash_cfg
*cfg
= afu
->parent
;
1055 struct device
*dev
= &cfg
->dev
->dev
;
1057 set_port_offline(fc_regs
);
1058 if (!wait_port_offline(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1059 FC_PORT_STATUS_RETRY_CNT
)) {
1060 dev_dbg(dev
, "%s: wait on port %d to go offline timed out\n",
1064 writeq_be(wwpn
, &fc_regs
[FC_PNAME
/ 8]);
1066 set_port_online(fc_regs
);
1067 if (!wait_port_online(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1068 FC_PORT_STATUS_RETRY_CNT
)) {
1069 dev_dbg(dev
, "%s: wait on port %d to go online timed out\n",
1075 * afu_link_reset() - resets the specified host FC port
1076 * @afu: AFU associated with the host that owns the specified FC port.
1077 * @port: Port number being configured.
1078 * @fc_regs: Top of MMIO region defined for specified port.
1080 * The provided MMIO region must be mapped prior to call. The sequence to
1081 * reset the port involves toggling it offline and then back online. This
1082 * action can cause this routine to delay up to a few seconds. An effort
1083 * is made to maintain link with the device by switching to host to use
1084 * the alternate port exclusively while the reset takes place.
1085 * failure to come online is overridden.
1087 static void afu_link_reset(struct afu
*afu
, int port
, __be64 __iomem
*fc_regs
)
1089 struct cxlflash_cfg
*cfg
= afu
->parent
;
1090 struct device
*dev
= &cfg
->dev
->dev
;
1093 /* first switch the AFU to the other links, if any */
1094 port_sel
= readq_be(&afu
->afu_map
->global
.regs
.afu_port_sel
);
1095 port_sel
&= ~(1ULL << port
);
1096 writeq_be(port_sel
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1097 cxlflash_afu_sync(afu
, 0, 0, AFU_GSYNC
);
1099 set_port_offline(fc_regs
);
1100 if (!wait_port_offline(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1101 FC_PORT_STATUS_RETRY_CNT
))
1102 dev_err(dev
, "%s: wait on port %d to go offline timed out\n",
1105 set_port_online(fc_regs
);
1106 if (!wait_port_online(fc_regs
, FC_PORT_STATUS_RETRY_INTERVAL_US
,
1107 FC_PORT_STATUS_RETRY_CNT
))
1108 dev_err(dev
, "%s: wait on port %d to go online timed out\n",
1111 /* switch back to include this port */
1112 port_sel
|= (1ULL << port
);
1113 writeq_be(port_sel
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1114 cxlflash_afu_sync(afu
, 0, 0, AFU_GSYNC
);
1116 dev_dbg(dev
, "%s: returning port_sel=%016llx\n", __func__
, port_sel
);
1120 * afu_err_intr_init() - clears and initializes the AFU for error interrupts
1121 * @afu: AFU associated with the host.
1123 static void afu_err_intr_init(struct afu
*afu
)
1125 struct cxlflash_cfg
*cfg
= afu
->parent
;
1126 __be64 __iomem
*fc_port_regs
;
1128 struct hwq
*hwq
= get_hwq(afu
, PRIMARY_HWQ
);
1131 /* global async interrupts: AFU clears afu_ctrl on context exit
1132 * if async interrupts were sent to that context. This prevents
1133 * the AFU form sending further async interrupts when
1135 * nobody to receive them.
1139 writeq_be(-1ULL, &afu
->afu_map
->global
.regs
.aintr_mask
);
1140 /* set LISN# to send and point to primary master context */
1141 reg
= ((u64
) (((hwq
->ctx_hndl
<< 8) | SISL_MSI_ASYNC_ERROR
)) << 40);
1143 if (afu
->internal_lun
)
1144 reg
|= 1; /* Bit 63 indicates local lun */
1145 writeq_be(reg
, &afu
->afu_map
->global
.regs
.afu_ctrl
);
1147 writeq_be(-1ULL, &afu
->afu_map
->global
.regs
.aintr_clear
);
1148 /* unmask bits that are of interest */
1149 /* note: afu can send an interrupt after this step */
1150 writeq_be(SISL_ASTATUS_MASK
, &afu
->afu_map
->global
.regs
.aintr_mask
);
1151 /* clear again in case a bit came on after previous clear but before */
1153 writeq_be(-1ULL, &afu
->afu_map
->global
.regs
.aintr_clear
);
1155 /* Clear/Set internal lun bits */
1156 fc_port_regs
= get_fc_port_regs(cfg
, 0);
1157 reg
= readq_be(&fc_port_regs
[FC_CONFIG2
/ 8]);
1158 reg
&= SISL_FC_INTERNAL_MASK
;
1159 if (afu
->internal_lun
)
1160 reg
|= ((u64
)(afu
->internal_lun
- 1) << SISL_FC_INTERNAL_SHIFT
);
1161 writeq_be(reg
, &fc_port_regs
[FC_CONFIG2
/ 8]);
1163 /* now clear FC errors */
1164 for (i
= 0; i
< cfg
->num_fc_ports
; i
++) {
1165 fc_port_regs
= get_fc_port_regs(cfg
, i
);
1167 writeq_be(0xFFFFFFFFU
, &fc_port_regs
[FC_ERROR
/ 8]);
1168 writeq_be(0, &fc_port_regs
[FC_ERRCAP
/ 8]);
1171 /* sync interrupts for master's IOARRIN write */
1172 /* note that unlike asyncs, there can be no pending sync interrupts */
1173 /* at this time (this is a fresh context and master has not written */
1174 /* IOARRIN yet), so there is nothing to clear. */
1176 /* set LISN#, it is always sent to the context that wrote IOARRIN */
1177 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
1178 hwq
= get_hwq(afu
, i
);
1180 writeq_be(SISL_MSI_SYNC_ERROR
, &hwq
->host_map
->ctx_ctrl
);
1181 writeq_be(SISL_ISTATUS_MASK
, &hwq
->host_map
->intr_mask
);
1186 * cxlflash_sync_err_irq() - interrupt handler for synchronous errors
1187 * @irq: Interrupt number.
1188 * @data: Private data provided at interrupt registration, the AFU.
1190 * Return: Always return IRQ_HANDLED.
1192 static irqreturn_t
cxlflash_sync_err_irq(int irq
, void *data
)
1194 struct hwq
*hwq
= (struct hwq
*)data
;
1195 struct cxlflash_cfg
*cfg
= hwq
->afu
->parent
;
1196 struct device
*dev
= &cfg
->dev
->dev
;
1200 reg
= readq_be(&hwq
->host_map
->intr_status
);
1201 reg_unmasked
= (reg
& SISL_ISTATUS_UNMASK
);
1203 if (reg_unmasked
== 0UL) {
1204 dev_err(dev
, "%s: spurious interrupt, intr_status=%016llx\n",
1206 goto cxlflash_sync_err_irq_exit
;
1209 dev_err(dev
, "%s: unexpected interrupt, intr_status=%016llx\n",
1212 writeq_be(reg_unmasked
, &hwq
->host_map
->intr_clear
);
1214 cxlflash_sync_err_irq_exit
:
1219 * process_hrrq() - process the read-response queue
1220 * @afu: AFU associated with the host.
1221 * @doneq: Queue of commands harvested from the RRQ.
1222 * @budget: Threshold of RRQ entries to process.
1224 * This routine must be called holding the disabled RRQ spin lock.
1226 * Return: The number of entries processed.
1228 static int process_hrrq(struct hwq
*hwq
, struct list_head
*doneq
, int budget
)
1230 struct afu
*afu
= hwq
->afu
;
1231 struct afu_cmd
*cmd
;
1232 struct sisl_ioasa
*ioasa
;
1233 struct sisl_ioarcb
*ioarcb
;
1234 bool toggle
= hwq
->toggle
;
1237 *hrrq_start
= hwq
->hrrq_start
,
1238 *hrrq_end
= hwq
->hrrq_end
,
1239 *hrrq_curr
= hwq
->hrrq_curr
;
1241 /* Process ready RRQ entries up to the specified budget (if any) */
1245 if ((entry
& SISL_RESP_HANDLE_T_BIT
) != toggle
)
1248 entry
&= ~SISL_RESP_HANDLE_T_BIT
;
1250 if (afu_is_sq_cmd_mode(afu
)) {
1251 ioasa
= (struct sisl_ioasa
*)entry
;
1252 cmd
= container_of(ioasa
, struct afu_cmd
, sa
);
1254 ioarcb
= (struct sisl_ioarcb
*)entry
;
1255 cmd
= container_of(ioarcb
, struct afu_cmd
, rcb
);
1258 list_add_tail(&cmd
->queue
, doneq
);
1260 /* Advance to next entry or wrap and flip the toggle bit */
1261 if (hrrq_curr
< hrrq_end
)
1264 hrrq_curr
= hrrq_start
;
1265 toggle
^= SISL_RESP_HANDLE_T_BIT
;
1268 atomic_inc(&hwq
->hsq_credits
);
1271 if (budget
> 0 && num_hrrq
>= budget
)
1275 hwq
->hrrq_curr
= hrrq_curr
;
1276 hwq
->toggle
= toggle
;
1282 * process_cmd_doneq() - process a queue of harvested RRQ commands
1283 * @doneq: Queue of completed commands.
1285 * Note that upon return the queue can no longer be trusted.
1287 static void process_cmd_doneq(struct list_head
*doneq
)
1289 struct afu_cmd
*cmd
, *tmp
;
1291 WARN_ON(list_empty(doneq
));
1293 list_for_each_entry_safe(cmd
, tmp
, doneq
, queue
)
1298 * cxlflash_irqpoll() - process a queue of harvested RRQ commands
1299 * @irqpoll: IRQ poll structure associated with queue to poll.
1300 * @budget: Threshold of RRQ entries to process per poll.
1302 * Return: The number of entries processed.
1304 static int cxlflash_irqpoll(struct irq_poll
*irqpoll
, int budget
)
1306 struct hwq
*hwq
= container_of(irqpoll
, struct hwq
, irqpoll
);
1307 unsigned long hrrq_flags
;
1309 int num_entries
= 0;
1311 spin_lock_irqsave(&hwq
->hrrq_slock
, hrrq_flags
);
1313 num_entries
= process_hrrq(hwq
, &doneq
, budget
);
1314 if (num_entries
< budget
)
1315 irq_poll_complete(irqpoll
);
1317 spin_unlock_irqrestore(&hwq
->hrrq_slock
, hrrq_flags
);
1319 process_cmd_doneq(&doneq
);
1324 * cxlflash_rrq_irq() - interrupt handler for read-response queue (normal path)
1325 * @irq: Interrupt number.
1326 * @data: Private data provided at interrupt registration, the AFU.
1328 * Return: IRQ_HANDLED or IRQ_NONE when no ready entries found.
1330 static irqreturn_t
cxlflash_rrq_irq(int irq
, void *data
)
1332 struct hwq
*hwq
= (struct hwq
*)data
;
1333 struct afu
*afu
= hwq
->afu
;
1334 unsigned long hrrq_flags
;
1336 int num_entries
= 0;
1338 spin_lock_irqsave(&hwq
->hrrq_slock
, hrrq_flags
);
1340 if (afu_is_irqpoll_enabled(afu
)) {
1341 irq_poll_sched(&hwq
->irqpoll
);
1342 spin_unlock_irqrestore(&hwq
->hrrq_slock
, hrrq_flags
);
1346 num_entries
= process_hrrq(hwq
, &doneq
, -1);
1347 spin_unlock_irqrestore(&hwq
->hrrq_slock
, hrrq_flags
);
1349 if (num_entries
== 0)
1352 process_cmd_doneq(&doneq
);
1357 * Asynchronous interrupt information table
1360 * - Order matters here as this array is indexed by bit position.
1362 * - The checkpatch script considers the BUILD_SISL_ASTATUS_FC_PORT macro
1363 * as complex and complains due to a lack of parentheses/braces.
1365 #define ASTATUS_FC(_a, _b, _c, _d) \
1366 { SISL_ASTATUS_FC##_a##_##_b, _c, _a, (_d) }
1368 #define BUILD_SISL_ASTATUS_FC_PORT(_a) \
1369 ASTATUS_FC(_a, LINK_UP, "link up", 0), \
1370 ASTATUS_FC(_a, LINK_DN, "link down", 0), \
1371 ASTATUS_FC(_a, LOGI_S, "login succeeded", SCAN_HOST), \
1372 ASTATUS_FC(_a, LOGI_F, "login failed", CLR_FC_ERROR), \
1373 ASTATUS_FC(_a, LOGI_R, "login timed out, retrying", LINK_RESET), \
1374 ASTATUS_FC(_a, CRC_T, "CRC threshold exceeded", LINK_RESET), \
1375 ASTATUS_FC(_a, LOGO, "target initiated LOGO", 0), \
1376 ASTATUS_FC(_a, OTHER, "other error", CLR_FC_ERROR | LINK_RESET)
1378 static const struct asyc_intr_info ainfo
[] = {
1379 BUILD_SISL_ASTATUS_FC_PORT(1),
1380 BUILD_SISL_ASTATUS_FC_PORT(0),
1381 BUILD_SISL_ASTATUS_FC_PORT(3),
1382 BUILD_SISL_ASTATUS_FC_PORT(2)
1386 * cxlflash_async_err_irq() - interrupt handler for asynchronous errors
1387 * @irq: Interrupt number.
1388 * @data: Private data provided at interrupt registration, the AFU.
1390 * Return: Always return IRQ_HANDLED.
1392 static irqreturn_t
cxlflash_async_err_irq(int irq
, void *data
)
1394 struct hwq
*hwq
= (struct hwq
*)data
;
1395 struct afu
*afu
= hwq
->afu
;
1396 struct cxlflash_cfg
*cfg
= afu
->parent
;
1397 struct device
*dev
= &cfg
->dev
->dev
;
1398 const struct asyc_intr_info
*info
;
1399 struct sisl_global_map __iomem
*global
= &afu
->afu_map
->global
;
1400 __be64 __iomem
*fc_port_regs
;
1406 reg
= readq_be(&global
->regs
.aintr_status
);
1407 reg_unmasked
= (reg
& SISL_ASTATUS_UNMASK
);
1409 if (unlikely(reg_unmasked
== 0)) {
1410 dev_err(dev
, "%s: spurious interrupt, aintr_status=%016llx\n",
1415 /* FYI, it is 'okay' to clear AFU status before FC_ERROR */
1416 writeq_be(reg_unmasked
, &global
->regs
.aintr_clear
);
1418 /* Check each bit that is on */
1419 for_each_set_bit(bit
, (ulong
*)®_unmasked
, BITS_PER_LONG
) {
1420 if (unlikely(bit
>= ARRAY_SIZE(ainfo
))) {
1426 if (unlikely(info
->status
!= 1ULL << bit
)) {
1432 fc_port_regs
= get_fc_port_regs(cfg
, port
);
1434 dev_err(dev
, "%s: FC Port %d -> %s, fc_status=%016llx\n",
1435 __func__
, port
, info
->desc
,
1436 readq_be(&fc_port_regs
[FC_STATUS
/ 8]));
1439 * Do link reset first, some OTHER errors will set FC_ERROR
1440 * again if cleared before or w/o a reset
1442 if (info
->action
& LINK_RESET
) {
1443 dev_err(dev
, "%s: FC Port %d: resetting link\n",
1445 cfg
->lr_state
= LINK_RESET_REQUIRED
;
1446 cfg
->lr_port
= port
;
1447 schedule_work(&cfg
->work_q
);
1450 if (info
->action
& CLR_FC_ERROR
) {
1451 reg
= readq_be(&fc_port_regs
[FC_ERROR
/ 8]);
1454 * Since all errors are unmasked, FC_ERROR and FC_ERRCAP
1455 * should be the same and tracing one is sufficient.
1458 dev_err(dev
, "%s: fc %d: clearing fc_error=%016llx\n",
1459 __func__
, port
, reg
);
1461 writeq_be(reg
, &fc_port_regs
[FC_ERROR
/ 8]);
1462 writeq_be(0, &fc_port_regs
[FC_ERRCAP
/ 8]);
1465 if (info
->action
& SCAN_HOST
) {
1466 atomic_inc(&cfg
->scan_host_needed
);
1467 schedule_work(&cfg
->work_q
);
1476 * start_context() - starts the master context
1477 * @cfg: Internal structure associated with the host.
1478 * @index: Index of the hardware queue.
1480 * Return: A success or failure value from CXL services.
1482 static int start_context(struct cxlflash_cfg
*cfg
, u32 index
)
1484 struct device
*dev
= &cfg
->dev
->dev
;
1485 struct hwq
*hwq
= get_hwq(cfg
->afu
, index
);
1488 rc
= cxl_start_context(hwq
->ctx
,
1489 hwq
->work
.work_element_descriptor
,
1492 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
1497 * read_vpd() - obtains the WWPNs from VPD
1498 * @cfg: Internal structure associated with the host.
1499 * @wwpn: Array of size MAX_FC_PORTS to pass back WWPNs
1501 * Return: 0 on success, -errno on failure
1503 static int read_vpd(struct cxlflash_cfg
*cfg
, u64 wwpn
[])
1505 struct device
*dev
= &cfg
->dev
->dev
;
1506 struct pci_dev
*pdev
= cfg
->dev
;
1508 int ro_start
, ro_size
, i
, j
, k
;
1510 char vpd_data
[CXLFLASH_VPD_LEN
];
1511 char tmp_buf
[WWPN_BUF_LEN
] = { 0 };
1512 char *wwpn_vpd_tags
[MAX_FC_PORTS
] = { "V5", "V6", "V7", "V8" };
1514 /* Get the VPD data from the device */
1515 vpd_size
= cxl_read_adapter_vpd(pdev
, vpd_data
, sizeof(vpd_data
));
1516 if (unlikely(vpd_size
<= 0)) {
1517 dev_err(dev
, "%s: Unable to read VPD (size = %ld)\n",
1518 __func__
, vpd_size
);
1523 /* Get the read only section offset */
1524 ro_start
= pci_vpd_find_tag(vpd_data
, 0, vpd_size
,
1525 PCI_VPD_LRDT_RO_DATA
);
1526 if (unlikely(ro_start
< 0)) {
1527 dev_err(dev
, "%s: VPD Read-only data not found\n", __func__
);
1532 /* Get the read only section size, cap when extends beyond read VPD */
1533 ro_size
= pci_vpd_lrdt_size(&vpd_data
[ro_start
]);
1535 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
1536 if (unlikely((i
+ j
) > vpd_size
)) {
1537 dev_dbg(dev
, "%s: Might need to read more VPD (%d > %ld)\n",
1538 __func__
, (i
+ j
), vpd_size
);
1539 ro_size
= vpd_size
- i
;
1543 * Find the offset of the WWPN tag within the read only
1544 * VPD data and validate the found field (partials are
1545 * no good to us). Convert the ASCII data to an integer
1546 * value. Note that we must copy to a temporary buffer
1547 * because the conversion service requires that the ASCII
1548 * string be terminated.
1550 for (k
= 0; k
< cfg
->num_fc_ports
; k
++) {
1552 i
= ro_start
+ PCI_VPD_LRDT_TAG_SIZE
;
1554 i
= pci_vpd_find_info_keyword(vpd_data
, i
, j
, wwpn_vpd_tags
[k
]);
1555 if (unlikely(i
< 0)) {
1556 dev_err(dev
, "%s: Port %d WWPN not found in VPD\n",
1562 j
= pci_vpd_info_field_size(&vpd_data
[i
]);
1563 i
+= PCI_VPD_INFO_FLD_HDR_SIZE
;
1564 if (unlikely((i
+ j
> vpd_size
) || (j
!= WWPN_LEN
))) {
1565 dev_err(dev
, "%s: Port %d WWPN incomplete or bad VPD\n",
1571 memcpy(tmp_buf
, &vpd_data
[i
], WWPN_LEN
);
1572 rc
= kstrtoul(tmp_buf
, WWPN_LEN
, (ulong
*)&wwpn
[k
]);
1574 dev_err(dev
, "%s: WWPN conversion failed for port %d\n",
1580 dev_dbg(dev
, "%s: wwpn%d=%016llx\n", __func__
, k
, wwpn
[k
]);
1584 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
1589 * init_pcr() - initialize the provisioning and control registers
1590 * @cfg: Internal structure associated with the host.
1592 * Also sets up fast access to the mapped registers and initializes AFU
1593 * command fields that never change.
1595 static void init_pcr(struct cxlflash_cfg
*cfg
)
1597 struct afu
*afu
= cfg
->afu
;
1598 struct sisl_ctrl_map __iomem
*ctrl_map
;
1602 for (i
= 0; i
< MAX_CONTEXT
; i
++) {
1603 ctrl_map
= &afu
->afu_map
->ctrls
[i
].ctrl
;
1604 /* Disrupt any clients that could be running */
1605 /* e.g. clients that survived a master restart */
1606 writeq_be(0, &ctrl_map
->rht_start
);
1607 writeq_be(0, &ctrl_map
->rht_cnt_id
);
1608 writeq_be(0, &ctrl_map
->ctx_cap
);
1611 /* Copy frequently used fields into hwq */
1612 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
1613 hwq
= get_hwq(afu
, i
);
1615 hwq
->ctx_hndl
= (u16
) cxl_process_element(hwq
->ctx
);
1616 hwq
->host_map
= &afu
->afu_map
->hosts
[hwq
->ctx_hndl
].host
;
1617 hwq
->ctrl_map
= &afu
->afu_map
->ctrls
[hwq
->ctx_hndl
].ctrl
;
1619 /* Program the Endian Control for the master context */
1620 writeq_be(SISL_ENDIAN_CTRL
, &hwq
->host_map
->endian_ctrl
);
1625 * init_global() - initialize AFU global registers
1626 * @cfg: Internal structure associated with the host.
1628 static int init_global(struct cxlflash_cfg
*cfg
)
1630 struct afu
*afu
= cfg
->afu
;
1631 struct device
*dev
= &cfg
->dev
->dev
;
1633 struct sisl_host_map __iomem
*hmap
;
1634 __be64 __iomem
*fc_port_regs
;
1635 u64 wwpn
[MAX_FC_PORTS
]; /* wwpn of AFU ports */
1636 int i
= 0, num_ports
= 0;
1640 rc
= read_vpd(cfg
, &wwpn
[0]);
1642 dev_err(dev
, "%s: could not read vpd rc=%d\n", __func__
, rc
);
1646 /* Set up RRQ and SQ in HWQ for master issued cmds */
1647 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
1648 hwq
= get_hwq(afu
, i
);
1649 hmap
= hwq
->host_map
;
1651 writeq_be((u64
) hwq
->hrrq_start
, &hmap
->rrq_start
);
1652 writeq_be((u64
) hwq
->hrrq_end
, &hmap
->rrq_end
);
1654 if (afu_is_sq_cmd_mode(afu
)) {
1655 writeq_be((u64
)hwq
->hsq_start
, &hmap
->sq_start
);
1656 writeq_be((u64
)hwq
->hsq_end
, &hmap
->sq_end
);
1660 /* AFU configuration */
1661 reg
= readq_be(&afu
->afu_map
->global
.regs
.afu_config
);
1662 reg
|= SISL_AFUCONF_AR_ALL
|SISL_AFUCONF_ENDIAN
;
1663 /* enable all auto retry options and control endianness */
1664 /* leave others at default: */
1665 /* CTX_CAP write protected, mbox_r does not clear on read and */
1666 /* checker on if dual afu */
1667 writeq_be(reg
, &afu
->afu_map
->global
.regs
.afu_config
);
1669 /* Global port select: select either port */
1670 if (afu
->internal_lun
) {
1671 /* Only use port 0 */
1672 writeq_be(PORT0
, &afu
->afu_map
->global
.regs
.afu_port_sel
);
1675 writeq_be(PORT_MASK(cfg
->num_fc_ports
),
1676 &afu
->afu_map
->global
.regs
.afu_port_sel
);
1677 num_ports
= cfg
->num_fc_ports
;
1680 for (i
= 0; i
< num_ports
; i
++) {
1681 fc_port_regs
= get_fc_port_regs(cfg
, i
);
1683 /* Unmask all errors (but they are still masked at AFU) */
1684 writeq_be(0, &fc_port_regs
[FC_ERRMSK
/ 8]);
1685 /* Clear CRC error cnt & set a threshold */
1686 (void)readq_be(&fc_port_regs
[FC_CNT_CRCERR
/ 8]);
1687 writeq_be(MC_CRC_THRESH
, &fc_port_regs
[FC_CRC_THRESH
/ 8]);
1689 /* Set WWPNs. If already programmed, wwpn[i] is 0 */
1691 afu_set_wwpn(afu
, i
, &fc_port_regs
[0], wwpn
[i
]);
1692 /* Programming WWPN back to back causes additional
1693 * offline/online transitions and a PLOGI
1698 /* Set up master's own CTX_CAP to allow real mode, host translation */
1699 /* tables, afu cmds and read/write GSCSI cmds. */
1700 /* First, unlock ctx_cap write by reading mbox */
1701 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
1702 hwq
= get_hwq(afu
, i
);
1704 (void)readq_be(&hwq
->ctrl_map
->mbox_r
); /* unlock ctx_cap */
1705 writeq_be((SISL_CTX_CAP_REAL_MODE
| SISL_CTX_CAP_HOST_XLATE
|
1706 SISL_CTX_CAP_READ_CMD
| SISL_CTX_CAP_WRITE_CMD
|
1707 SISL_CTX_CAP_AFU_CMD
| SISL_CTX_CAP_GSCSI_CMD
),
1708 &hwq
->ctrl_map
->ctx_cap
);
1710 /* Initialize heartbeat */
1711 afu
->hb
= readq_be(&afu
->afu_map
->global
.regs
.afu_hb
);
1717 * start_afu() - initializes and starts the AFU
1718 * @cfg: Internal structure associated with the host.
1720 static int start_afu(struct cxlflash_cfg
*cfg
)
1722 struct afu
*afu
= cfg
->afu
;
1723 struct device
*dev
= &cfg
->dev
->dev
;
1730 /* Initialize each HWQ */
1731 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
1732 hwq
= get_hwq(afu
, i
);
1734 /* After an AFU reset, RRQ entries are stale, clear them */
1735 memset(&hwq
->rrq_entry
, 0, sizeof(hwq
->rrq_entry
));
1737 /* Initialize RRQ pointers */
1738 hwq
->hrrq_start
= &hwq
->rrq_entry
[0];
1739 hwq
->hrrq_end
= &hwq
->rrq_entry
[NUM_RRQ_ENTRY
- 1];
1740 hwq
->hrrq_curr
= hwq
->hrrq_start
;
1743 /* Initialize spin locks */
1744 spin_lock_init(&hwq
->hrrq_slock
);
1745 spin_lock_init(&hwq
->hsq_slock
);
1748 if (afu_is_sq_cmd_mode(afu
)) {
1749 memset(&hwq
->sq
, 0, sizeof(hwq
->sq
));
1750 hwq
->hsq_start
= &hwq
->sq
[0];
1751 hwq
->hsq_end
= &hwq
->sq
[NUM_SQ_ENTRY
- 1];
1752 hwq
->hsq_curr
= hwq
->hsq_start
;
1754 atomic_set(&hwq
->hsq_credits
, NUM_SQ_ENTRY
- 1);
1757 /* Initialize IRQ poll */
1758 if (afu_is_irqpoll_enabled(afu
))
1759 irq_poll_init(&hwq
->irqpoll
, afu
->irqpoll_weight
,
1764 rc
= init_global(cfg
);
1766 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
1771 * init_intr() - setup interrupt handlers for the master context
1772 * @cfg: Internal structure associated with the host.
1773 * @hwq: Hardware queue to initialize.
1775 * Return: 0 on success, -errno on failure
1777 static enum undo_level
init_intr(struct cxlflash_cfg
*cfg
,
1780 struct device
*dev
= &cfg
->dev
->dev
;
1781 struct cxl_context
*ctx
= hwq
->ctx
;
1783 enum undo_level level
= UNDO_NOOP
;
1784 bool is_primary_hwq
= (hwq
->index
== PRIMARY_HWQ
);
1785 int num_irqs
= is_primary_hwq
? 3 : 2;
1787 rc
= cxl_allocate_afu_irqs(ctx
, num_irqs
);
1789 dev_err(dev
, "%s: allocate_afu_irqs failed rc=%d\n",
1795 rc
= cxl_map_afu_irq(ctx
, 1, cxlflash_sync_err_irq
, hwq
,
1796 "SISL_MSI_SYNC_ERROR");
1797 if (unlikely(rc
<= 0)) {
1798 dev_err(dev
, "%s: SISL_MSI_SYNC_ERROR map failed\n", __func__
);
1803 rc
= cxl_map_afu_irq(ctx
, 2, cxlflash_rrq_irq
, hwq
,
1804 "SISL_MSI_RRQ_UPDATED");
1805 if (unlikely(rc
<= 0)) {
1806 dev_err(dev
, "%s: SISL_MSI_RRQ_UPDATED map failed\n", __func__
);
1811 /* SISL_MSI_ASYNC_ERROR is setup only for the primary HWQ */
1812 if (!is_primary_hwq
)
1815 rc
= cxl_map_afu_irq(ctx
, 3, cxlflash_async_err_irq
, hwq
,
1816 "SISL_MSI_ASYNC_ERROR");
1817 if (unlikely(rc
<= 0)) {
1818 dev_err(dev
, "%s: SISL_MSI_ASYNC_ERROR map failed\n", __func__
);
1827 * init_mc() - create and register as the master context
1828 * @cfg: Internal structure associated with the host.
1829 * index: HWQ Index of the master context.
1831 * Return: 0 on success, -errno on failure
1833 static int init_mc(struct cxlflash_cfg
*cfg
, u32 index
)
1835 struct cxl_context
*ctx
;
1836 struct device
*dev
= &cfg
->dev
->dev
;
1837 struct hwq
*hwq
= get_hwq(cfg
->afu
, index
);
1839 enum undo_level level
;
1841 hwq
->afu
= cfg
->afu
;
1844 if (index
== PRIMARY_HWQ
)
1845 ctx
= cxl_get_context(cfg
->dev
);
1847 ctx
= cxl_dev_context_init(cfg
->dev
);
1848 if (unlikely(!ctx
)) {
1856 /* Set it up as a master with the CXL */
1857 cxl_set_master(ctx
);
1859 /* Reset AFU when initializing primary context */
1860 if (index
== PRIMARY_HWQ
) {
1861 rc
= cxl_afu_reset(ctx
);
1863 dev_err(dev
, "%s: AFU reset failed rc=%d\n",
1869 level
= init_intr(cfg
, hwq
);
1870 if (unlikely(level
)) {
1871 dev_err(dev
, "%s: interrupt init failed rc=%d\n", __func__
, rc
);
1875 /* This performs the equivalent of the CXL_IOCTL_START_WORK.
1876 * The CXL_IOCTL_GET_PROCESS_ELEMENT is implicit in the process
1877 * element (pe) that is embedded in the context (ctx)
1879 rc
= start_context(cfg
, index
);
1881 dev_err(dev
, "%s: start context failed rc=%d\n", __func__
, rc
);
1882 level
= UNMAP_THREE
;
1887 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
1890 term_intr(cfg
, level
, index
);
1891 if (index
!= PRIMARY_HWQ
)
1892 cxl_release_context(ctx
);
1899 * get_num_afu_ports() - determines and configures the number of AFU ports
1900 * @cfg: Internal structure associated with the host.
1902 * This routine determines the number of AFU ports by converting the global
1903 * port selection mask. The converted value is only valid following an AFU
1904 * reset (explicit or power-on). This routine must be invoked shortly after
1905 * mapping as other routines are dependent on the number of ports during the
1906 * initialization sequence.
1908 * To support legacy AFUs that might not have reflected an initial global
1909 * port mask (value read is 0), default to the number of ports originally
1910 * supported by the cxlflash driver (2) before hardware with other port
1911 * offerings was introduced.
1913 static void get_num_afu_ports(struct cxlflash_cfg
*cfg
)
1915 struct afu
*afu
= cfg
->afu
;
1916 struct device
*dev
= &cfg
->dev
->dev
;
1918 int num_fc_ports
= LEGACY_FC_PORTS
;
1920 port_mask
= readq_be(&afu
->afu_map
->global
.regs
.afu_port_sel
);
1921 if (port_mask
!= 0ULL)
1922 num_fc_ports
= min(ilog2(port_mask
) + 1, MAX_FC_PORTS
);
1924 dev_dbg(dev
, "%s: port_mask=%016llx num_fc_ports=%d\n",
1925 __func__
, port_mask
, num_fc_ports
);
1927 cfg
->num_fc_ports
= num_fc_ports
;
1928 cfg
->host
->max_channel
= PORTNUM2CHAN(num_fc_ports
);
1932 * init_afu() - setup as master context and start AFU
1933 * @cfg: Internal structure associated with the host.
1935 * This routine is a higher level of control for configuring the
1936 * AFU on probe and reset paths.
1938 * Return: 0 on success, -errno on failure
1940 static int init_afu(struct cxlflash_cfg
*cfg
)
1944 struct afu
*afu
= cfg
->afu
;
1945 struct device
*dev
= &cfg
->dev
->dev
;
1949 cxl_perst_reloads_same_image(cfg
->cxl_afu
, true);
1951 afu
->num_hwqs
= afu
->desired_hwqs
;
1952 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
1953 rc
= init_mc(cfg
, i
);
1955 dev_err(dev
, "%s: init_mc failed rc=%d index=%d\n",
1961 /* Map the entire MMIO space of the AFU using the first context */
1962 hwq
= get_hwq(afu
, PRIMARY_HWQ
);
1963 afu
->afu_map
= cxl_psa_map(hwq
->ctx
);
1964 if (!afu
->afu_map
) {
1965 dev_err(dev
, "%s: cxl_psa_map failed\n", __func__
);
1970 /* No byte reverse on reading afu_version or string will be backwards */
1971 reg
= readq(&afu
->afu_map
->global
.regs
.afu_version
);
1972 memcpy(afu
->version
, ®
, sizeof(reg
));
1973 afu
->interface_version
=
1974 readq_be(&afu
->afu_map
->global
.regs
.interface_version
);
1975 if ((afu
->interface_version
+ 1) == 0) {
1976 dev_err(dev
, "Back level AFU, please upgrade. AFU version %s "
1977 "interface version %016llx\n", afu
->version
,
1978 afu
->interface_version
);
1983 if (afu_is_sq_cmd_mode(afu
)) {
1984 afu
->send_cmd
= send_cmd_sq
;
1985 afu
->context_reset
= context_reset_sq
;
1987 afu
->send_cmd
= send_cmd_ioarrin
;
1988 afu
->context_reset
= context_reset_ioarrin
;
1991 dev_dbg(dev
, "%s: afu_ver=%s interface_ver=%016llx\n", __func__
,
1992 afu
->version
, afu
->interface_version
);
1994 get_num_afu_ports(cfg
);
1996 rc
= start_afu(cfg
);
1998 dev_err(dev
, "%s: start_afu failed, rc=%d\n", __func__
, rc
);
2002 afu_err_intr_init(cfg
->afu
);
2003 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
2004 hwq
= get_hwq(afu
, i
);
2006 hwq
->room
= readq_be(&hwq
->host_map
->cmd_room
);
2009 /* Restore the LUN mappings */
2010 cxlflash_restore_luntable(cfg
);
2012 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
2016 for (i
= afu
->num_hwqs
- 1; i
>= 0; i
--) {
2017 term_intr(cfg
, UNMAP_THREE
, i
);
2024 * afu_reset() - resets the AFU
2025 * @cfg: Internal structure associated with the host.
2027 * Return: 0 on success, -errno on failure
2029 static int afu_reset(struct cxlflash_cfg
*cfg
)
2031 struct device
*dev
= &cfg
->dev
->dev
;
2034 /* Stop the context before the reset. Since the context is
2035 * no longer available restart it after the reset is complete
2041 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
2046 * drain_ioctls() - wait until all currently executing ioctls have completed
2047 * @cfg: Internal structure associated with the host.
2049 * Obtain write access to read/write semaphore that wraps ioctl
2050 * handling to 'drain' ioctls currently executing.
2052 static void drain_ioctls(struct cxlflash_cfg
*cfg
)
2054 down_write(&cfg
->ioctl_rwsem
);
2055 up_write(&cfg
->ioctl_rwsem
);
2059 * cxlflash_async_reset_host() - asynchronous host reset handler
2060 * @data: Private data provided while scheduling reset.
2061 * @cookie: Cookie that can be used for checkpointing.
2063 static void cxlflash_async_reset_host(void *data
, async_cookie_t cookie
)
2065 struct cxlflash_cfg
*cfg
= data
;
2066 struct device
*dev
= &cfg
->dev
->dev
;
2069 if (cfg
->state
!= STATE_RESET
) {
2070 dev_dbg(dev
, "%s: Not performing a reset, state=%d\n",
2071 __func__
, cfg
->state
);
2076 cxlflash_mark_contexts_error(cfg
);
2077 rc
= afu_reset(cfg
);
2079 cfg
->state
= STATE_FAILTERM
;
2081 cfg
->state
= STATE_NORMAL
;
2082 wake_up_all(&cfg
->reset_waitq
);
2085 scsi_unblock_requests(cfg
->host
);
2089 * cxlflash_schedule_async_reset() - schedule an asynchronous host reset
2090 * @cfg: Internal structure associated with the host.
2092 static void cxlflash_schedule_async_reset(struct cxlflash_cfg
*cfg
)
2094 struct device
*dev
= &cfg
->dev
->dev
;
2096 if (cfg
->state
!= STATE_NORMAL
) {
2097 dev_dbg(dev
, "%s: Not performing reset state=%d\n",
2098 __func__
, cfg
->state
);
2102 cfg
->state
= STATE_RESET
;
2103 scsi_block_requests(cfg
->host
);
2104 cfg
->async_reset_cookie
= async_schedule(cxlflash_async_reset_host
,
2109 * cxlflash_afu_sync() - builds and sends an AFU sync command
2110 * @afu: AFU associated with the host.
2111 * @ctx_hndl_u: Identifies context requesting sync.
2112 * @res_hndl_u: Identifies resource requesting sync.
2113 * @mode: Type of sync to issue (lightweight, heavyweight, global).
2115 * The AFU can only take 1 sync command at a time. This routine enforces this
2116 * limitation by using a mutex to provide exclusive access to the AFU during
2117 * the sync. This design point requires calling threads to not be on interrupt
2118 * context due to the possibility of sleeping during concurrent sync operations.
2120 * AFU sync operations are only necessary and allowed when the device is
2121 * operating normally. When not operating normally, sync requests can occur as
2122 * part of cleaning up resources associated with an adapter prior to removal.
2123 * In this scenario, these requests are simply ignored (safe due to the AFU
2127 * 0 on success, -errno on failure
2129 int cxlflash_afu_sync(struct afu
*afu
, ctx_hndl_t ctx_hndl_u
,
2130 res_hndl_t res_hndl_u
, u8 mode
)
2132 struct cxlflash_cfg
*cfg
= afu
->parent
;
2133 struct device
*dev
= &cfg
->dev
->dev
;
2134 struct afu_cmd
*cmd
= NULL
;
2135 struct hwq
*hwq
= get_hwq(afu
, PRIMARY_HWQ
);
2139 static DEFINE_MUTEX(sync_active
);
2141 if (cfg
->state
!= STATE_NORMAL
) {
2142 dev_dbg(dev
, "%s: Sync not required state=%u\n",
2143 __func__
, cfg
->state
);
2147 mutex_lock(&sync_active
);
2148 atomic_inc(&afu
->cmds_active
);
2149 buf
= kzalloc(sizeof(*cmd
) + __alignof__(*cmd
) - 1, GFP_KERNEL
);
2150 if (unlikely(!buf
)) {
2151 dev_err(dev
, "%s: no memory for command\n", __func__
);
2156 cmd
= (struct afu_cmd
*)PTR_ALIGN(buf
, __alignof__(*cmd
));
2159 init_completion(&cmd
->cevent
);
2161 cmd
->hwq_index
= hwq
->index
;
2163 dev_dbg(dev
, "%s: afu=%p cmd=%p ctx=%d nretry=%d\n",
2164 __func__
, afu
, cmd
, ctx_hndl_u
, nretry
);
2166 cmd
->rcb
.req_flags
= SISL_REQ_FLAGS_AFU_CMD
;
2167 cmd
->rcb
.ctx_id
= hwq
->ctx_hndl
;
2168 cmd
->rcb
.msi
= SISL_MSI_RRQ_UPDATED
;
2169 cmd
->rcb
.timeout
= MC_AFU_SYNC_TIMEOUT
;
2171 cmd
->rcb
.cdb
[0] = 0xC0; /* AFU Sync */
2172 cmd
->rcb
.cdb
[1] = mode
;
2174 /* The cdb is aligned, no unaligned accessors required */
2175 *((__be16
*)&cmd
->rcb
.cdb
[2]) = cpu_to_be16(ctx_hndl_u
);
2176 *((__be32
*)&cmd
->rcb
.cdb
[4]) = cpu_to_be32(res_hndl_u
);
2178 rc
= afu
->send_cmd(afu
, cmd
);
2184 rc
= wait_resp(afu
, cmd
);
2185 if (rc
== -ETIMEDOUT
) {
2186 rc
= afu
->context_reset(hwq
);
2187 if (!rc
&& ++nretry
< 2)
2189 cxlflash_schedule_async_reset(cfg
);
2193 atomic_dec(&afu
->cmds_active
);
2194 mutex_unlock(&sync_active
);
2196 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
2201 * cxlflash_eh_device_reset_handler() - reset a single LUN
2202 * @scp: SCSI command to send.
2205 * SUCCESS as defined in scsi/scsi.h
2206 * FAILED as defined in scsi/scsi.h
2208 static int cxlflash_eh_device_reset_handler(struct scsi_cmnd
*scp
)
2211 struct Scsi_Host
*host
= scp
->device
->host
;
2212 struct cxlflash_cfg
*cfg
= shost_priv(host
);
2213 struct device
*dev
= &cfg
->dev
->dev
;
2214 struct afu
*afu
= cfg
->afu
;
2217 dev_dbg(dev
, "%s: (scp=%p) %d/%d/%d/%llu "
2218 "cdb=(%08x-%08x-%08x-%08x)\n", __func__
, scp
, host
->host_no
,
2219 scp
->device
->channel
, scp
->device
->id
, scp
->device
->lun
,
2220 get_unaligned_be32(&((u32
*)scp
->cmnd
)[0]),
2221 get_unaligned_be32(&((u32
*)scp
->cmnd
)[1]),
2222 get_unaligned_be32(&((u32
*)scp
->cmnd
)[2]),
2223 get_unaligned_be32(&((u32
*)scp
->cmnd
)[3]));
2226 switch (cfg
->state
) {
2228 rcr
= send_tmf(afu
, scp
, TMF_LUN_RESET
);
2233 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
);
2240 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
2245 * cxlflash_eh_host_reset_handler() - reset the host adapter
2246 * @scp: SCSI command from stack identifying host.
2248 * Following a reset, the state is evaluated again in case an EEH occurred
2249 * during the reset. In such a scenario, the host reset will either yield
2250 * until the EEH recovery is complete or return success or failure based
2251 * upon the current device state.
2254 * SUCCESS as defined in scsi/scsi.h
2255 * FAILED as defined in scsi/scsi.h
2257 static int cxlflash_eh_host_reset_handler(struct scsi_cmnd
*scp
)
2261 struct Scsi_Host
*host
= scp
->device
->host
;
2262 struct cxlflash_cfg
*cfg
= shost_priv(host
);
2263 struct device
*dev
= &cfg
->dev
->dev
;
2265 dev_dbg(dev
, "%s: (scp=%p) %d/%d/%d/%llu "
2266 "cdb=(%08x-%08x-%08x-%08x)\n", __func__
, scp
, host
->host_no
,
2267 scp
->device
->channel
, scp
->device
->id
, scp
->device
->lun
,
2268 get_unaligned_be32(&((u32
*)scp
->cmnd
)[0]),
2269 get_unaligned_be32(&((u32
*)scp
->cmnd
)[1]),
2270 get_unaligned_be32(&((u32
*)scp
->cmnd
)[2]),
2271 get_unaligned_be32(&((u32
*)scp
->cmnd
)[3]));
2273 switch (cfg
->state
) {
2275 cfg
->state
= STATE_RESET
;
2277 cxlflash_mark_contexts_error(cfg
);
2278 rcr
= afu_reset(cfg
);
2281 cfg
->state
= STATE_FAILTERM
;
2283 cfg
->state
= STATE_NORMAL
;
2284 wake_up_all(&cfg
->reset_waitq
);
2288 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
);
2289 if (cfg
->state
== STATE_NORMAL
)
2297 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
2302 * cxlflash_change_queue_depth() - change the queue depth for the device
2303 * @sdev: SCSI device destined for queue depth change.
2304 * @qdepth: Requested queue depth value to set.
2306 * The requested queue depth is capped to the maximum supported value.
2308 * Return: The actual queue depth set.
2310 static int cxlflash_change_queue_depth(struct scsi_device
*sdev
, int qdepth
)
2313 if (qdepth
> CXLFLASH_MAX_CMDS_PER_LUN
)
2314 qdepth
= CXLFLASH_MAX_CMDS_PER_LUN
;
2316 scsi_change_queue_depth(sdev
, qdepth
);
2317 return sdev
->queue_depth
;
2321 * cxlflash_show_port_status() - queries and presents the current port status
2322 * @port: Desired port for status reporting.
2323 * @cfg: Internal structure associated with the host.
2324 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2326 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2328 static ssize_t
cxlflash_show_port_status(u32 port
,
2329 struct cxlflash_cfg
*cfg
,
2332 struct device
*dev
= &cfg
->dev
->dev
;
2335 __be64 __iomem
*fc_port_regs
;
2337 WARN_ON(port
>= MAX_FC_PORTS
);
2339 if (port
>= cfg
->num_fc_ports
) {
2340 dev_info(dev
, "%s: Port %d not supported on this card.\n",
2345 fc_port_regs
= get_fc_port_regs(cfg
, port
);
2346 status
= readq_be(&fc_port_regs
[FC_MTIP_STATUS
/ 8]);
2347 status
&= FC_MTIP_STATUS_MASK
;
2349 if (status
== FC_MTIP_STATUS_ONLINE
)
2350 disp_status
= "online";
2351 else if (status
== FC_MTIP_STATUS_OFFLINE
)
2352 disp_status
= "offline";
2354 disp_status
= "unknown";
2356 return scnprintf(buf
, PAGE_SIZE
, "%s\n", disp_status
);
2360 * port0_show() - queries and presents the current status of port 0
2361 * @dev: Generic device associated with the host owning the port.
2362 * @attr: Device attribute representing the port.
2363 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2365 * Return: The size of the ASCII string returned in @buf.
2367 static ssize_t
port0_show(struct device
*dev
,
2368 struct device_attribute
*attr
,
2371 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2373 return cxlflash_show_port_status(0, cfg
, buf
);
2377 * port1_show() - queries and presents the current status of port 1
2378 * @dev: Generic device associated with the host owning the port.
2379 * @attr: Device attribute representing the port.
2380 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2382 * Return: The size of the ASCII string returned in @buf.
2384 static ssize_t
port1_show(struct device
*dev
,
2385 struct device_attribute
*attr
,
2388 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2390 return cxlflash_show_port_status(1, cfg
, buf
);
2394 * port2_show() - queries and presents the current status of port 2
2395 * @dev: Generic device associated with the host owning the port.
2396 * @attr: Device attribute representing the port.
2397 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2399 * Return: The size of the ASCII string returned in @buf.
2401 static ssize_t
port2_show(struct device
*dev
,
2402 struct device_attribute
*attr
,
2405 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2407 return cxlflash_show_port_status(2, cfg
, buf
);
2411 * port3_show() - queries and presents the current status of port 3
2412 * @dev: Generic device associated with the host owning the port.
2413 * @attr: Device attribute representing the port.
2414 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2416 * Return: The size of the ASCII string returned in @buf.
2418 static ssize_t
port3_show(struct device
*dev
,
2419 struct device_attribute
*attr
,
2422 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2424 return cxlflash_show_port_status(3, cfg
, buf
);
2428 * lun_mode_show() - presents the current LUN mode of the host
2429 * @dev: Generic device associated with the host.
2430 * @attr: Device attribute representing the LUN mode.
2431 * @buf: Buffer of length PAGE_SIZE to report back the LUN mode in ASCII.
2433 * Return: The size of the ASCII string returned in @buf.
2435 static ssize_t
lun_mode_show(struct device
*dev
,
2436 struct device_attribute
*attr
, char *buf
)
2438 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2439 struct afu
*afu
= cfg
->afu
;
2441 return scnprintf(buf
, PAGE_SIZE
, "%u\n", afu
->internal_lun
);
2445 * lun_mode_store() - sets the LUN mode of the host
2446 * @dev: Generic device associated with the host.
2447 * @attr: Device attribute representing the LUN mode.
2448 * @buf: Buffer of length PAGE_SIZE containing the LUN mode in ASCII.
2449 * @count: Length of data resizing in @buf.
2451 * The CXL Flash AFU supports a dummy LUN mode where the external
2452 * links and storage are not required. Space on the FPGA is used
2453 * to create 1 or 2 small LUNs which are presented to the system
2454 * as if they were a normal storage device. This feature is useful
2455 * during development and also provides manufacturing with a way
2456 * to test the AFU without an actual device.
2458 * 0 = external LUN[s] (default)
2459 * 1 = internal LUN (1 x 64K, 512B blocks, id 0)
2460 * 2 = internal LUN (1 x 64K, 4K blocks, id 0)
2461 * 3 = internal LUN (2 x 32K, 512B blocks, ids 0,1)
2462 * 4 = internal LUN (2 x 32K, 4K blocks, ids 0,1)
2464 * Return: The size of the ASCII string returned in @buf.
2466 static ssize_t
lun_mode_store(struct device
*dev
,
2467 struct device_attribute
*attr
,
2468 const char *buf
, size_t count
)
2470 struct Scsi_Host
*shost
= class_to_shost(dev
);
2471 struct cxlflash_cfg
*cfg
= shost_priv(shost
);
2472 struct afu
*afu
= cfg
->afu
;
2476 rc
= kstrtouint(buf
, 10, &lun_mode
);
2477 if (!rc
&& (lun_mode
< 5) && (lun_mode
!= afu
->internal_lun
)) {
2478 afu
->internal_lun
= lun_mode
;
2481 * When configured for internal LUN, there is only one channel,
2482 * channel number 0, else there will be one less than the number
2483 * of fc ports for this card.
2485 if (afu
->internal_lun
)
2486 shost
->max_channel
= 0;
2488 shost
->max_channel
= PORTNUM2CHAN(cfg
->num_fc_ports
);
2491 scsi_scan_host(cfg
->host
);
2498 * ioctl_version_show() - presents the current ioctl version of the host
2499 * @dev: Generic device associated with the host.
2500 * @attr: Device attribute representing the ioctl version.
2501 * @buf: Buffer of length PAGE_SIZE to report back the ioctl version.
2503 * Return: The size of the ASCII string returned in @buf.
2505 static ssize_t
ioctl_version_show(struct device
*dev
,
2506 struct device_attribute
*attr
, char *buf
)
2508 return scnprintf(buf
, PAGE_SIZE
, "%u\n", DK_CXLFLASH_VERSION_0
);
2512 * cxlflash_show_port_lun_table() - queries and presents the port LUN table
2513 * @port: Desired port for status reporting.
2514 * @cfg: Internal structure associated with the host.
2515 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2517 * Return: The size of the ASCII string returned in @buf or -EINVAL.
2519 static ssize_t
cxlflash_show_port_lun_table(u32 port
,
2520 struct cxlflash_cfg
*cfg
,
2523 struct device
*dev
= &cfg
->dev
->dev
;
2524 __be64 __iomem
*fc_port_luns
;
2528 WARN_ON(port
>= MAX_FC_PORTS
);
2530 if (port
>= cfg
->num_fc_ports
) {
2531 dev_info(dev
, "%s: Port %d not supported on this card.\n",
2536 fc_port_luns
= get_fc_port_luns(cfg
, port
);
2538 for (i
= 0; i
< CXLFLASH_NUM_VLUNS
; i
++)
2539 bytes
+= scnprintf(buf
+ bytes
, PAGE_SIZE
- bytes
,
2541 i
, readq_be(&fc_port_luns
[i
]));
2546 * port0_lun_table_show() - presents the current LUN table of port 0
2547 * @dev: Generic device associated with the host owning the port.
2548 * @attr: Device attribute representing the port.
2549 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2551 * Return: The size of the ASCII string returned in @buf.
2553 static ssize_t
port0_lun_table_show(struct device
*dev
,
2554 struct device_attribute
*attr
,
2557 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2559 return cxlflash_show_port_lun_table(0, cfg
, buf
);
2563 * port1_lun_table_show() - presents the current LUN table of port 1
2564 * @dev: Generic device associated with the host owning the port.
2565 * @attr: Device attribute representing the port.
2566 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2568 * Return: The size of the ASCII string returned in @buf.
2570 static ssize_t
port1_lun_table_show(struct device
*dev
,
2571 struct device_attribute
*attr
,
2574 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2576 return cxlflash_show_port_lun_table(1, cfg
, buf
);
2580 * port2_lun_table_show() - presents the current LUN table of port 2
2581 * @dev: Generic device associated with the host owning the port.
2582 * @attr: Device attribute representing the port.
2583 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2585 * Return: The size of the ASCII string returned in @buf.
2587 static ssize_t
port2_lun_table_show(struct device
*dev
,
2588 struct device_attribute
*attr
,
2591 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2593 return cxlflash_show_port_lun_table(2, cfg
, buf
);
2597 * port3_lun_table_show() - presents the current LUN table of port 3
2598 * @dev: Generic device associated with the host owning the port.
2599 * @attr: Device attribute representing the port.
2600 * @buf: Buffer of length PAGE_SIZE to report back port status in ASCII.
2602 * Return: The size of the ASCII string returned in @buf.
2604 static ssize_t
port3_lun_table_show(struct device
*dev
,
2605 struct device_attribute
*attr
,
2608 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2610 return cxlflash_show_port_lun_table(3, cfg
, buf
);
2614 * irqpoll_weight_show() - presents the current IRQ poll weight for the host
2615 * @dev: Generic device associated with the host.
2616 * @attr: Device attribute representing the IRQ poll weight.
2617 * @buf: Buffer of length PAGE_SIZE to report back the current IRQ poll
2620 * An IRQ poll weight of 0 indicates polling is disabled.
2622 * Return: The size of the ASCII string returned in @buf.
2624 static ssize_t
irqpoll_weight_show(struct device
*dev
,
2625 struct device_attribute
*attr
, char *buf
)
2627 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2628 struct afu
*afu
= cfg
->afu
;
2630 return scnprintf(buf
, PAGE_SIZE
, "%u\n", afu
->irqpoll_weight
);
2634 * irqpoll_weight_store() - sets the current IRQ poll weight for the host
2635 * @dev: Generic device associated with the host.
2636 * @attr: Device attribute representing the IRQ poll weight.
2637 * @buf: Buffer of length PAGE_SIZE containing the desired IRQ poll
2639 * @count: Length of data resizing in @buf.
2641 * An IRQ poll weight of 0 indicates polling is disabled.
2643 * Return: The size of the ASCII string returned in @buf.
2645 static ssize_t
irqpoll_weight_store(struct device
*dev
,
2646 struct device_attribute
*attr
,
2647 const char *buf
, size_t count
)
2649 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2650 struct device
*cfgdev
= &cfg
->dev
->dev
;
2651 struct afu
*afu
= cfg
->afu
;
2656 rc
= kstrtouint(buf
, 10, &weight
);
2662 "Invalid IRQ poll weight. It must be 256 or less.\n");
2666 if (weight
== afu
->irqpoll_weight
) {
2668 "Current IRQ poll weight has the same weight.\n");
2672 if (afu_is_irqpoll_enabled(afu
)) {
2673 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
2674 hwq
= get_hwq(afu
, i
);
2676 irq_poll_disable(&hwq
->irqpoll
);
2680 afu
->irqpoll_weight
= weight
;
2683 for (i
= 0; i
< afu
->num_hwqs
; i
++) {
2684 hwq
= get_hwq(afu
, i
);
2686 irq_poll_init(&hwq
->irqpoll
, weight
, cxlflash_irqpoll
);
2694 * num_hwqs_show() - presents the number of hardware queues for the host
2695 * @dev: Generic device associated with the host.
2696 * @attr: Device attribute representing the number of hardware queues.
2697 * @buf: Buffer of length PAGE_SIZE to report back the number of hardware
2700 * Return: The size of the ASCII string returned in @buf.
2702 static ssize_t
num_hwqs_show(struct device
*dev
,
2703 struct device_attribute
*attr
, char *buf
)
2705 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2706 struct afu
*afu
= cfg
->afu
;
2708 return scnprintf(buf
, PAGE_SIZE
, "%u\n", afu
->num_hwqs
);
2712 * num_hwqs_store() - sets the number of hardware queues for the host
2713 * @dev: Generic device associated with the host.
2714 * @attr: Device attribute representing the number of hardware queues.
2715 * @buf: Buffer of length PAGE_SIZE containing the number of hardware
2717 * @count: Length of data resizing in @buf.
2719 * n > 0: num_hwqs = n
2720 * n = 0: num_hwqs = num_online_cpus()
2721 * n < 0: num_online_cpus() / abs(n)
2723 * Return: The size of the ASCII string returned in @buf.
2725 static ssize_t
num_hwqs_store(struct device
*dev
,
2726 struct device_attribute
*attr
,
2727 const char *buf
, size_t count
)
2729 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2730 struct afu
*afu
= cfg
->afu
;
2732 int nhwqs
, num_hwqs
;
2734 rc
= kstrtoint(buf
, 10, &nhwqs
);
2740 else if (nhwqs
== 0)
2741 num_hwqs
= num_online_cpus();
2743 num_hwqs
= num_online_cpus() / abs(nhwqs
);
2745 afu
->desired_hwqs
= min(num_hwqs
, CXLFLASH_MAX_HWQS
);
2746 WARN_ON_ONCE(afu
->desired_hwqs
== 0);
2749 switch (cfg
->state
) {
2751 cfg
->state
= STATE_RESET
;
2753 cxlflash_mark_contexts_error(cfg
);
2754 rc
= afu_reset(cfg
);
2756 cfg
->state
= STATE_FAILTERM
;
2758 cfg
->state
= STATE_NORMAL
;
2759 wake_up_all(&cfg
->reset_waitq
);
2762 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
);
2763 if (cfg
->state
== STATE_NORMAL
)
2766 /* Ideally should not happen */
2767 dev_err(dev
, "%s: Device is not ready, state=%d\n",
2768 __func__
, cfg
->state
);
2775 static const char *hwq_mode_name
[MAX_HWQ_MODE
] = { "rr", "tag", "cpu" };
2778 * hwq_mode_show() - presents the HWQ steering mode for the host
2779 * @dev: Generic device associated with the host.
2780 * @attr: Device attribute representing the HWQ steering mode.
2781 * @buf: Buffer of length PAGE_SIZE to report back the HWQ steering mode
2782 * as a character string.
2784 * Return: The size of the ASCII string returned in @buf.
2786 static ssize_t
hwq_mode_show(struct device
*dev
,
2787 struct device_attribute
*attr
, char *buf
)
2789 struct cxlflash_cfg
*cfg
= shost_priv(class_to_shost(dev
));
2790 struct afu
*afu
= cfg
->afu
;
2792 return scnprintf(buf
, PAGE_SIZE
, "%s\n", hwq_mode_name
[afu
->hwq_mode
]);
2796 * hwq_mode_store() - sets the HWQ steering mode for the host
2797 * @dev: Generic device associated with the host.
2798 * @attr: Device attribute representing the HWQ steering mode.
2799 * @buf: Buffer of length PAGE_SIZE containing the HWQ steering mode
2800 * as a character string.
2801 * @count: Length of data resizing in @buf.
2804 * tag = Block MQ Tagging
2805 * cpu = CPU Affinity
2807 * Return: The size of the ASCII string returned in @buf.
2809 static ssize_t
hwq_mode_store(struct device
*dev
,
2810 struct device_attribute
*attr
,
2811 const char *buf
, size_t count
)
2813 struct Scsi_Host
*shost
= class_to_shost(dev
);
2814 struct cxlflash_cfg
*cfg
= shost_priv(shost
);
2815 struct device
*cfgdev
= &cfg
->dev
->dev
;
2816 struct afu
*afu
= cfg
->afu
;
2818 u32 mode
= MAX_HWQ_MODE
;
2820 for (i
= 0; i
< MAX_HWQ_MODE
; i
++) {
2821 if (!strncmp(hwq_mode_name
[i
], buf
, strlen(hwq_mode_name
[i
]))) {
2827 if (mode
>= MAX_HWQ_MODE
) {
2828 dev_info(cfgdev
, "Invalid HWQ steering mode.\n");
2832 if ((mode
== HWQ_MODE_TAG
) && !shost_use_blk_mq(shost
)) {
2833 dev_info(cfgdev
, "SCSI-MQ is not enabled, use a different "
2834 "HWQ steering mode.\n");
2838 afu
->hwq_mode
= mode
;
2844 * mode_show() - presents the current mode of the device
2845 * @dev: Generic device associated with the device.
2846 * @attr: Device attribute representing the device mode.
2847 * @buf: Buffer of length PAGE_SIZE to report back the dev mode in ASCII.
2849 * Return: The size of the ASCII string returned in @buf.
2851 static ssize_t
mode_show(struct device
*dev
,
2852 struct device_attribute
*attr
, char *buf
)
2854 struct scsi_device
*sdev
= to_scsi_device(dev
);
2856 return scnprintf(buf
, PAGE_SIZE
, "%s\n",
2857 sdev
->hostdata
? "superpipe" : "legacy");
2863 static DEVICE_ATTR_RO(port0
);
2864 static DEVICE_ATTR_RO(port1
);
2865 static DEVICE_ATTR_RO(port2
);
2866 static DEVICE_ATTR_RO(port3
);
2867 static DEVICE_ATTR_RW(lun_mode
);
2868 static DEVICE_ATTR_RO(ioctl_version
);
2869 static DEVICE_ATTR_RO(port0_lun_table
);
2870 static DEVICE_ATTR_RO(port1_lun_table
);
2871 static DEVICE_ATTR_RO(port2_lun_table
);
2872 static DEVICE_ATTR_RO(port3_lun_table
);
2873 static DEVICE_ATTR_RW(irqpoll_weight
);
2874 static DEVICE_ATTR_RW(num_hwqs
);
2875 static DEVICE_ATTR_RW(hwq_mode
);
2877 static struct device_attribute
*cxlflash_host_attrs
[] = {
2883 &dev_attr_ioctl_version
,
2884 &dev_attr_port0_lun_table
,
2885 &dev_attr_port1_lun_table
,
2886 &dev_attr_port2_lun_table
,
2887 &dev_attr_port3_lun_table
,
2888 &dev_attr_irqpoll_weight
,
2897 static DEVICE_ATTR_RO(mode
);
2899 static struct device_attribute
*cxlflash_dev_attrs
[] = {
2907 static struct scsi_host_template driver_template
= {
2908 .module
= THIS_MODULE
,
2909 .name
= CXLFLASH_ADAPTER_NAME
,
2910 .info
= cxlflash_driver_info
,
2911 .ioctl
= cxlflash_ioctl
,
2912 .proc_name
= CXLFLASH_NAME
,
2913 .queuecommand
= cxlflash_queuecommand
,
2914 .eh_device_reset_handler
= cxlflash_eh_device_reset_handler
,
2915 .eh_host_reset_handler
= cxlflash_eh_host_reset_handler
,
2916 .change_queue_depth
= cxlflash_change_queue_depth
,
2917 .cmd_per_lun
= CXLFLASH_MAX_CMDS_PER_LUN
,
2918 .can_queue
= CXLFLASH_MAX_CMDS
,
2919 .cmd_size
= sizeof(struct afu_cmd
) + __alignof__(struct afu_cmd
) - 1,
2921 .sg_tablesize
= 1, /* No scatter gather support */
2922 .max_sectors
= CXLFLASH_MAX_SECTORS
,
2923 .use_clustering
= ENABLE_CLUSTERING
,
2924 .shost_attrs
= cxlflash_host_attrs
,
2925 .sdev_attrs
= cxlflash_dev_attrs
,
2929 * Device dependent values
2931 static struct dev_dependent_vals dev_corsa_vals
= { CXLFLASH_MAX_SECTORS
,
2933 static struct dev_dependent_vals dev_flash_gt_vals
= { CXLFLASH_MAX_SECTORS
,
2934 CXLFLASH_NOTIFY_SHUTDOWN
};
2935 static struct dev_dependent_vals dev_briard_vals
= { CXLFLASH_MAX_SECTORS
,
2936 CXLFLASH_NOTIFY_SHUTDOWN
};
2939 * PCI device binding table
2941 static struct pci_device_id cxlflash_pci_table
[] = {
2942 {PCI_VENDOR_ID_IBM
, PCI_DEVICE_ID_IBM_CORSA
,
2943 PCI_ANY_ID
, PCI_ANY_ID
, 0, 0, (kernel_ulong_t
)&dev_corsa_vals
},
2944 {PCI_VENDOR_ID_IBM
, PCI_DEVICE_ID_IBM_FLASH_GT
,
2945 PCI_ANY_ID
, PCI_ANY_ID
, 0, 0, (kernel_ulong_t
)&dev_flash_gt_vals
},
2946 {PCI_VENDOR_ID_IBM
, PCI_DEVICE_ID_IBM_BRIARD
,
2947 PCI_ANY_ID
, PCI_ANY_ID
, 0, 0, (kernel_ulong_t
)&dev_briard_vals
},
2951 MODULE_DEVICE_TABLE(pci
, cxlflash_pci_table
);
2954 * cxlflash_worker_thread() - work thread handler for the AFU
2955 * @work: Work structure contained within cxlflash associated with host.
2957 * Handles the following events:
2958 * - Link reset which cannot be performed on interrupt context due to
2959 * blocking up to a few seconds
2962 static void cxlflash_worker_thread(struct work_struct
*work
)
2964 struct cxlflash_cfg
*cfg
= container_of(work
, struct cxlflash_cfg
,
2966 struct afu
*afu
= cfg
->afu
;
2967 struct device
*dev
= &cfg
->dev
->dev
;
2968 __be64 __iomem
*fc_port_regs
;
2972 /* Avoid MMIO if the device has failed */
2974 if (cfg
->state
!= STATE_NORMAL
)
2977 spin_lock_irqsave(cfg
->host
->host_lock
, lock_flags
);
2979 if (cfg
->lr_state
== LINK_RESET_REQUIRED
) {
2980 port
= cfg
->lr_port
;
2982 dev_err(dev
, "%s: invalid port index %d\n",
2985 spin_unlock_irqrestore(cfg
->host
->host_lock
,
2988 /* The reset can block... */
2989 fc_port_regs
= get_fc_port_regs(cfg
, port
);
2990 afu_link_reset(afu
, port
, fc_port_regs
);
2991 spin_lock_irqsave(cfg
->host
->host_lock
, lock_flags
);
2994 cfg
->lr_state
= LINK_RESET_COMPLETE
;
2997 spin_unlock_irqrestore(cfg
->host
->host_lock
, lock_flags
);
2999 if (atomic_dec_if_positive(&cfg
->scan_host_needed
) >= 0)
3000 scsi_scan_host(cfg
->host
);
3004 * cxlflash_probe() - PCI entry point to add host
3005 * @pdev: PCI device associated with the host.
3006 * @dev_id: PCI device id associated with device.
3008 * The device will initially start out in a 'probing' state and
3009 * transition to the 'normal' state at the end of a successful
3010 * probe. Should an EEH event occur during probe, the notification
3011 * thread (error_detected()) will wait until the probe handler
3012 * is nearly complete. At that time, the device will be moved to
3013 * a 'probed' state and the EEH thread woken up to drive the slot
3014 * reset and recovery (device moves to 'normal' state). Meanwhile,
3015 * the probe will be allowed to exit successfully.
3017 * Return: 0 on success, -errno on failure
3019 static int cxlflash_probe(struct pci_dev
*pdev
,
3020 const struct pci_device_id
*dev_id
)
3022 struct Scsi_Host
*host
;
3023 struct cxlflash_cfg
*cfg
= NULL
;
3024 struct device
*dev
= &pdev
->dev
;
3025 struct dev_dependent_vals
*ddv
;
3029 dev_dbg(&pdev
->dev
, "%s: Found CXLFLASH with IRQ: %d\n",
3030 __func__
, pdev
->irq
);
3032 ddv
= (struct dev_dependent_vals
*)dev_id
->driver_data
;
3033 driver_template
.max_sectors
= ddv
->max_sectors
;
3035 host
= scsi_host_alloc(&driver_template
, sizeof(struct cxlflash_cfg
));
3037 dev_err(dev
, "%s: scsi_host_alloc failed\n", __func__
);
3042 host
->max_id
= CXLFLASH_MAX_NUM_TARGETS_PER_BUS
;
3043 host
->max_lun
= CXLFLASH_MAX_NUM_LUNS_PER_TARGET
;
3044 host
->unique_id
= host
->host_no
;
3045 host
->max_cmd_len
= CXLFLASH_MAX_CDB_LEN
;
3047 cfg
= shost_priv(host
);
3049 rc
= alloc_mem(cfg
);
3051 dev_err(dev
, "%s: alloc_mem failed\n", __func__
);
3053 scsi_host_put(cfg
->host
);
3057 cfg
->init_state
= INIT_STATE_NONE
;
3059 cfg
->cxl_fops
= cxlflash_cxl_fops
;
3062 * Promoted LUNs move to the top of the LUN table. The rest stay on
3063 * the bottom half. The bottom half grows from the end (index = 255),
3064 * whereas the top half grows from the beginning (index = 0).
3066 * Initialize the last LUN index for all possible ports.
3068 cfg
->promote_lun_index
= 0;
3070 for (k
= 0; k
< MAX_FC_PORTS
; k
++)
3071 cfg
->last_lun_index
[k
] = CXLFLASH_NUM_VLUNS
/2 - 1;
3073 cfg
->dev_id
= (struct pci_device_id
*)dev_id
;
3075 init_waitqueue_head(&cfg
->tmf_waitq
);
3076 init_waitqueue_head(&cfg
->reset_waitq
);
3078 INIT_WORK(&cfg
->work_q
, cxlflash_worker_thread
);
3079 cfg
->lr_state
= LINK_RESET_INVALID
;
3081 spin_lock_init(&cfg
->tmf_slock
);
3082 mutex_init(&cfg
->ctx_tbl_list_mutex
);
3083 mutex_init(&cfg
->ctx_recovery_mutex
);
3084 init_rwsem(&cfg
->ioctl_rwsem
);
3085 INIT_LIST_HEAD(&cfg
->ctx_err_recovery
);
3086 INIT_LIST_HEAD(&cfg
->lluns
);
3088 pci_set_drvdata(pdev
, cfg
);
3090 cfg
->cxl_afu
= cxl_pci_to_afu(pdev
);
3094 dev_err(dev
, "%s: init_pci failed rc=%d\n", __func__
, rc
);
3097 cfg
->init_state
= INIT_STATE_PCI
;
3100 if (rc
&& !wq_has_sleeper(&cfg
->reset_waitq
)) {
3101 dev_err(dev
, "%s: init_afu failed rc=%d\n", __func__
, rc
);
3104 cfg
->init_state
= INIT_STATE_AFU
;
3106 rc
= init_scsi(cfg
);
3108 dev_err(dev
, "%s: init_scsi failed rc=%d\n", __func__
, rc
);
3111 cfg
->init_state
= INIT_STATE_SCSI
;
3113 if (wq_has_sleeper(&cfg
->reset_waitq
)) {
3114 cfg
->state
= STATE_PROBED
;
3115 wake_up_all(&cfg
->reset_waitq
);
3117 cfg
->state
= STATE_NORMAL
;
3119 dev_dbg(dev
, "%s: returning rc=%d\n", __func__
, rc
);
3123 cxlflash_remove(pdev
);
3128 * cxlflash_pci_error_detected() - called when a PCI error is detected
3129 * @pdev: PCI device struct.
3130 * @state: PCI channel state.
3132 * When an EEH occurs during an active reset, wait until the reset is
3133 * complete and then take action based upon the device state.
3135 * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT
3137 static pci_ers_result_t
cxlflash_pci_error_detected(struct pci_dev
*pdev
,
3138 pci_channel_state_t state
)
3141 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
3142 struct device
*dev
= &cfg
->dev
->dev
;
3144 dev_dbg(dev
, "%s: pdev=%p state=%u\n", __func__
, pdev
, state
);
3147 case pci_channel_io_frozen
:
3148 wait_event(cfg
->reset_waitq
, cfg
->state
!= STATE_RESET
&&
3149 cfg
->state
!= STATE_PROBING
);
3150 if (cfg
->state
== STATE_FAILTERM
)
3151 return PCI_ERS_RESULT_DISCONNECT
;
3153 cfg
->state
= STATE_RESET
;
3154 scsi_block_requests(cfg
->host
);
3156 rc
= cxlflash_mark_contexts_error(cfg
);
3158 dev_err(dev
, "%s: Failed to mark user contexts rc=%d\n",
3161 return PCI_ERS_RESULT_NEED_RESET
;
3162 case pci_channel_io_perm_failure
:
3163 cfg
->state
= STATE_FAILTERM
;
3164 wake_up_all(&cfg
->reset_waitq
);
3165 scsi_unblock_requests(cfg
->host
);
3166 return PCI_ERS_RESULT_DISCONNECT
;
3170 return PCI_ERS_RESULT_NEED_RESET
;
3174 * cxlflash_pci_slot_reset() - called when PCI slot has been reset
3175 * @pdev: PCI device struct.
3177 * This routine is called by the pci error recovery code after the PCI
3178 * slot has been reset, just before we should resume normal operations.
3180 * Return: PCI_ERS_RESULT_RECOVERED or PCI_ERS_RESULT_DISCONNECT
3182 static pci_ers_result_t
cxlflash_pci_slot_reset(struct pci_dev
*pdev
)
3185 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
3186 struct device
*dev
= &cfg
->dev
->dev
;
3188 dev_dbg(dev
, "%s: pdev=%p\n", __func__
, pdev
);
3192 dev_err(dev
, "%s: EEH recovery failed rc=%d\n", __func__
, rc
);
3193 return PCI_ERS_RESULT_DISCONNECT
;
3196 return PCI_ERS_RESULT_RECOVERED
;
3200 * cxlflash_pci_resume() - called when normal operation can resume
3201 * @pdev: PCI device struct
3203 static void cxlflash_pci_resume(struct pci_dev
*pdev
)
3205 struct cxlflash_cfg
*cfg
= pci_get_drvdata(pdev
);
3206 struct device
*dev
= &cfg
->dev
->dev
;
3208 dev_dbg(dev
, "%s: pdev=%p\n", __func__
, pdev
);
3210 cfg
->state
= STATE_NORMAL
;
3211 wake_up_all(&cfg
->reset_waitq
);
3212 scsi_unblock_requests(cfg
->host
);
3215 static const struct pci_error_handlers cxlflash_err_handler
= {
3216 .error_detected
= cxlflash_pci_error_detected
,
3217 .slot_reset
= cxlflash_pci_slot_reset
,
3218 .resume
= cxlflash_pci_resume
,
3222 * PCI device structure
3224 static struct pci_driver cxlflash_driver
= {
3225 .name
= CXLFLASH_NAME
,
3226 .id_table
= cxlflash_pci_table
,
3227 .probe
= cxlflash_probe
,
3228 .remove
= cxlflash_remove
,
3229 .shutdown
= cxlflash_remove
,
3230 .err_handler
= &cxlflash_err_handler
,
3234 * init_cxlflash() - module entry point
3236 * Return: 0 on success, -errno on failure
3238 static int __init
init_cxlflash(void)
3241 cxlflash_list_init();
3243 return pci_register_driver(&cxlflash_driver
);
3247 * exit_cxlflash() - module exit point
3249 static void __exit
exit_cxlflash(void)
3251 cxlflash_term_global_luns();
3252 cxlflash_free_errpage();
3254 pci_unregister_driver(&cxlflash_driver
);
3257 module_init(init_cxlflash
);
3258 module_exit(exit_cxlflash
);