qla2x00t-32gbit: edif: Fix I/O timeout due to over-subscription

The current edif code does not keep track of FW IOCB resources.  This led
to IOCB queue full on error recovery (I/O timeout).  Make use of the
existing code that tracks IOCB resources to prevent over-subscription.

Link: https://lore.kernel.org/r/20220608115849.16693-2-njavali@marvell.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 63ab6cb582fa upstream ]
This commit is contained in:
Gleb Chesnokov
2022-08-08 18:47:03 +03:00
parent 09950e41b5
commit d48d24f4ef

View File

@@ -2952,6 +2952,12 @@ qla28xx_start_scsi_edif(srb_t *sp)
tot_dsds = nseg;
req_cnt = qla24xx_calc_iocbs(vha, tot_dsds);
sp->iores.res_type = RESOURCE_INI;
sp->iores.iocb_cnt = req_cnt;
if (qla_get_iocbs(sp->qpair, &sp->iores))
goto queuing_error;
if (req->cnt < (req_cnt + 2)) {
cnt = IS_SHADOW_REG_CAPABLE(ha) ? *req->out_ptr :
rd_reg_dword(req->req_q_out);
@@ -3143,6 +3149,7 @@ queuing_error:
mempool_free(sp->u.scmd.ct6_ctx, ha->ctx_mempool);
sp->u.scmd.ct6_ctx = NULL;
}
qla_put_iocbs(sp->qpair, &sp->iores);
spin_unlock_irqrestore(lock, flags);
return QLA_FUNCTION_FAILED;