On Thu, 10 Feb 2011 11:03:11 -0800
"Nicholas A. Bellinger" <nab@xxxxxxxxxxxxxxx> wrote:
> > +#define INITIAL_SRP_LIMIT 16
> > +#define DEFAULT_MAX_SECTORS 256
> > +
>
> Ok, these two hardcoded defs can be converted into context specific
> fabric dependent configfs attributes (TF_CIT_TMPL(fabric)->tfc_* below)
> using target_core_fabric_configfs.h wrappers..
I think that these values are used when you load the driver. No chance
to configure them via configfs.
> > +static char *ibmvscsis_parse_pr_out_transport_id(struct se_portal_group *se_tpg,
> > + const char *buf,
> > + u32 *out_tid_len,
> > + char **port_nexus_ptr)
> > +{
> > + return NULL;
> > +}
> > +
>
> OK, I still need to add generic SCSI_PROTOCOL_SRP handlers for PR
> TransportID processing, and get up and running with TCM_Loop SRP target
> ports. I will look at getting this resolved in LIO upstream in the near
> future, and make sure this gets converted..
I don't care much.
FYI, ibmvscsis initiators don't share a target device. So PR is
pointless in reality.
> > +static int ibmvscsis_queue_tm_rsp(struct se_cmd *se_cmd)
> > +{
> > + return 0;
> > +}
> > +
>
> The TM request/response logic still needs to be hooked up, right..?
Yeah, I'll add soon.
> > + case 0x08: /* Cache page */
> > + /* length should be 4 */
> > +#if 0
> > + if (cmd->cdb[4] != 4
> > + && cmd->cdb[4] != 0x20) {
> > + send_rsp(iue, ILLEGAL_REQUEST, 0x20);
> > + dma_free_coherent(iue->adapter->dev,
> > + MODE_SENSE_BUFFER_SIZE,
> > + mode, data_token);
> > + return FREE_IU;
> > + }
> > +#endif
> > +
>
> Can this be dropped..?
Yeah, I had better to add the proper error handling though.
> > +static int ibmvscsis_queue_data_in(struct se_cmd *se_cmd)
> > +{
> > + struct ibmvscsis_cmnd *cmd = container_of(se_cmd,
> > + struct ibmvscsis_cmnd, se_cmd);
> > + struct scsi_cmnd *sc = &cmd->sc;
> > + /*
> > + * Check for overflow residual count
> > + */
> > + if (se_cmd->se_cmd_flags & SCF_OVERFLOW_BIT)
> > + scsi_set_resid(sc, se_cmd->residual_count);
> > +
> > + sc->sdb.length = se_cmd->data_length;
> > +
> > + /*
> > + * Setup the struct se_task->task_sg[] chained SG list
> > + */
> > + if ((se_cmd->se_cmd_flags & SCF_SCSI_DATA_SG_IO_CDB) ||
> > + (se_cmd->se_cmd_flags & SCF_SCSI_CONTROL_SG_IO_CDB)) {
> > + transport_do_task_sg_chain(se_cmd);
> > +
> > + sc->sdb.table.nents = T_TASK(se_cmd)->t_tasks_sg_chained_no;
> > + sc->sdb.table.sgl = T_TASK(se_cmd)->t_tasks_sg_chained;
> > + } else if (se_cmd->se_cmd_flags & SCF_SCSI_CONTROL_NONSG_IO_CDB) {
> > + /*
> > + * Use T_TASK(se_cmd)->t_tasks_sg_bounce for control CDBs
> > + * using a contigious buffer
> > + */
> > + sg_init_table(&T_TASK(se_cmd)->t_tasks_sg_bounce, 1);
> > + sg_set_buf(&T_TASK(se_cmd)->t_tasks_sg_bounce,
> > + T_TASK(se_cmd)->t_task_buf, se_cmd->data_length);
> > +
> > + sc->sdb.table.nents = 1;
> > + sc->sdb.table.sgl = &T_TASK(se_cmd)->t_tasks_sg_bounce;
> > + }
> > + /*
> > + * Perform the SCSI READ data transfer from sc->sdb.table into
> > + * VIO LPAR memory. This will occur via libsrp in the
> > + * ibmvscsis_rdma() callback
> > + */
> > +#if 0
> > + ret = srp_transfer_data(sc, &vio_iu(iue)->srp.cmd,
> > + ibmvscsis_rdma, 1, 1);
> > + if (ret) {
> > + printk(KERN_ERR "srp_transfer_data() failed: %d, returning"
> > + " DID_ERROR\n", ret);
> > + sc->result = host_byte(DID_ERROR) | se_cmd->scsi_status;
> > + } else
> > + sc->result = host_byte(DID_OK) | se_cmd->scsi_status;
> > +#endif
>
> Mmmm, is this really supposed to be disabled..?
ibmvscsis_cmd_done calls srp_transfer_data(). I should remove the
above.
> > + /*
> > + * This will call srp_transfer_data() and post the response
> > + * to VIO via libsrp.
> > + */
> > + ibmvscsis_cmd_done(sc);
> > + return 0;
> > +}
> > +
> > + case REPORT_LUNS:
> > + sg_alloc_table(&sc->sdb.table, 1, GFP_KERNEL);
> > + pg = alloc_page(GFP_KERNEL|__GFP_ZERO);
> > + sc->sdb.length = ibmvscsis_report_luns(adapter, cmd,
> > + page_address(pg));
> > + sg_set_page(sc->sdb.table.sgl, pg, sc->sdb.length, 0);
> > + ibmvscsis_cmd_done(sc);
> > + sg_free_table(&sc->sdb.table);
> > + __free_page(pg);
> > + kfree(vsc);
> > + break;
> > + case MODE_SENSE:
> > + /* fixme: needs to use tcm */
> > + sg_alloc_table(&sc->sdb.table, 1, GFP_KERNEL);
> > + pg = alloc_page(GFP_KERNEL|__GFP_ZERO);
> > + sc->sdb.length = ibmvscsis_mode_sense(adapter,
> > + cmd, page_address(pg));
> > + sg_set_page(sc->sdb.table.sgl, pg, sc->sdb.length, 0);
> > + ibmvscsis_cmd_done(sc);
> > + sg_free_table(&sc->sdb.table);
> > + __free_page(pg);
> > + kfree(vsc);
> > + break;
>
> Just to verify again, INQUIRY and REPORT_LUNS are the only CDBs that
> have a hard requirement of being intercepted in order for VSCSI to
> function, right..?
tcm MODE_SENSE needed to be modified slightly to handle
ibmvscsis. It's a todo item.
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
