scsi: lpfc: Correct driver deregistrations with host nvme transport

The driver's interaction with the host nvme transport has been incorrect
for a while. The driver did not wait for the unregister callbacks
(waited only 5 jiffies). Thus the driver may remove objects that may be
referenced by subsequent abort commands from the transport, and the
actual unregister callback was effectively a noop. This was especially
problematic if the driver was unloaded.

The driver now waits for the unregister callbacks, as it should, before
continuing with teardown.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This commit is contained in:
James Smart 2017-11-20 16:00:40 -08:00 committed by Martin K. Petersen
parent 3b5bde69bc
commit add9d6be3d
3 changed files with 114 additions and 6 deletions

View file

@ -134,6 +134,8 @@ struct lpfc_nodelist {
struct lpfc_scsicmd_bkt *lat_data; /* Latency data */ struct lpfc_scsicmd_bkt *lat_data; /* Latency data */
uint32_t fc4_prli_sent; uint32_t fc4_prli_sent;
uint32_t upcall_flags; uint32_t upcall_flags;
#define NLP_WAIT_FOR_UNREG 0x1
uint32_t nvme_fb_size; /* NVME target's supported byte cnt */ uint32_t nvme_fb_size; /* NVME target's supported byte cnt */
#define NVME_FB_BIT_SHIFT 9 /* PRLI Rsp first burst in 512B units. */ #define NVME_FB_BIT_SHIFT 9 /* PRLI Rsp first burst in 512B units. */
}; };

View file

@ -154,6 +154,10 @@ lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
{ {
struct lpfc_nvme_lport *lport = localport->private; struct lpfc_nvme_lport *lport = localport->private;
lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME,
"6173 localport %p delete complete\n",
lport);
/* release any threads waiting for the unreg to complete */ /* release any threads waiting for the unreg to complete */
complete(&lport->lport_unreg_done); complete(&lport->lport_unreg_done);
} }
@ -946,10 +950,19 @@ lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
freqpriv->nvme_buf = NULL; freqpriv->nvme_buf = NULL;
/* NVME targets need completion held off until the abort exchange /* NVME targets need completion held off until the abort exchange
* completes. * completes unless the NVME Rport is getting unregistered.
*/ */
if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY) ||
ndlp->upcall_flags & NLP_WAIT_FOR_UNREG) {
/* Clear the XBUSY flag to prevent double completions.
* The nvme rport is getting unregistered and there is
* no need to defer the IO.
*/
if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY)
lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
nCmd->done(nCmd); nCmd->done(nCmd);
}
spin_lock_irqsave(&phba->hbalock, flags); spin_lock_irqsave(&phba->hbalock, flags);
lpfc_ncmd->nrport = NULL; lpfc_ncmd->nrport = NULL;
@ -2234,6 +2247,47 @@ lpfc_nvme_create_localport(struct lpfc_vport *vport)
return ret; return ret;
} }
/* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
*
* The driver has to wait for the host nvme transport to callback
* indicating the localport has successfully unregistered all
* resources. Since this is an uninterruptible wait, loop every ten
* seconds and print a message indicating no progress.
*
* An uninterruptible wait is used because of the risk of transport-to-
* driver state mismatch.
*/
void
lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
struct lpfc_nvme_lport *lport)
{
#if (IS_ENABLED(CONFIG_NVME_FC))
u32 wait_tmo;
int ret;
/* Host transport has to clean up and confirm requiring an indefinite
* wait. Print a message if a 10 second wait expires and renew the
* wait. This is unexpected.
*/
wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
while (true) {
ret = wait_for_completion_timeout(&lport->lport_unreg_done,
wait_tmo);
if (unlikely(!ret)) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
"6176 Lport %p Localport %p wait "
"timed out. Renewing.\n",
lport, vport->localport);
continue;
}
break;
}
lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
"6177 Lport %p Localport %p Complete Success\n",
lport, vport->localport);
#endif
}
/** /**
* lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport. * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
* @pnvme: pointer to lpfc nvme data structure. * @pnvme: pointer to lpfc nvme data structure.
@ -2268,7 +2322,11 @@ lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
*/ */
init_completion(&lport->lport_unreg_done); init_completion(&lport->lport_unreg_done);
ret = nvme_fc_unregister_localport(localport); ret = nvme_fc_unregister_localport(localport);
wait_for_completion_timeout(&lport->lport_unreg_done, 5);
/* Wait for completion. This either blocks
* indefinitely or succeeds
*/
lpfc_nvme_lport_unreg_wait(vport, lport);
/* Regardless of the unregister upcall response, clear /* Regardless of the unregister upcall response, clear
* nvmei_support. All rports are unregistered and the * nvmei_support. All rports are unregistered and the
@ -2424,6 +2482,47 @@ lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
#endif #endif
} }
/* lpfc_nvme_rport_unreg_wait - Wait for the host to complete an rport unreg.
*
* The driver has to wait for the host nvme transport to callback
* indicating the remoteport has successfully unregistered all
* resources. Since this is an uninterruptible wait, loop every ten
* seconds and print a message indicating no progress.
*
* An uninterruptible wait is used because of the risk of transport-to-
* driver state mismatch.
*/
void
lpfc_nvme_rport_unreg_wait(struct lpfc_vport *vport,
struct lpfc_nvme_rport *rport)
{
#if (IS_ENABLED(CONFIG_NVME_FC))
u32 wait_tmo;
int ret;
/* Host transport has to clean up and confirm requiring an indefinite
* wait. Print a message if a 10 second wait expires and renew the
* wait. This is unexpected.
*/
wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
while (true) {
ret = wait_for_completion_timeout(&rport->rport_unreg_done,
wait_tmo);
if (unlikely(!ret)) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
"6174 Rport %p Remoteport %p wait "
"timed out. Renewing.\n",
rport, rport->remoteport);
continue;
}
break;
}
lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
"6175 Rport %p Remoteport %p Complete Success\n",
rport, rport->remoteport);
#endif
}
/* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport. /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
* *
* There is no notion of Devloss or rport recovery from the current * There is no notion of Devloss or rport recovery from the current
@ -2480,14 +2579,19 @@ lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
/* No concern about the role change on the nvme remoteport. /* No concern about the role change on the nvme remoteport.
* The transport will update it. * The transport will update it.
*/ */
ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG;
ret = nvme_fc_unregister_remoteport(remoteport); ret = nvme_fc_unregister_remoteport(remoteport);
if (ret != 0) { if (ret != 0)
lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
"6167 NVME unregister failed %d " "6167 NVME unregister failed %d "
"port_state x%x\n", "port_state x%x\n",
ret, remoteport->port_state); ret, remoteport->port_state);
} else
/* Wait for completion. This either blocks
* indefinitely or succeeds
*/
lpfc_nvme_rport_unreg_wait(vport, rport);
ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
} }
return; return;

View file

@ -27,6 +27,8 @@
#define LPFC_NVME_ERSP_LEN 0x20 #define LPFC_NVME_ERSP_LEN 0x20
#define LPFC_NVME_WAIT_TMO 10
struct lpfc_nvme_qhandle { struct lpfc_nvme_qhandle {
uint32_t index; /* WQ index to use */ uint32_t index; /* WQ index to use */
uint32_t qidx; /* queue index passed to create */ uint32_t qidx; /* queue index passed to create */