On Thu, 2018-06-21 at 14:10 -0500, Mike Christie wrote:
> >
> > @@ -3156,9 +3167,13 @@ static int translate_sense_reason(struct se_cmd *cmd, sense_reason_t reason)
> > si = &sense_info_table[(__force int)
> > TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE];
> >
> > + key = si->key;
> > if (reason == TCM_CHECK_CONDITION_UNIT_ATTENTION) {
> > - core_scsi3_ua_for_check_condition(cmd, &asc, &ascq);
> > - WARN_ON_ONCE(asc == 0);
> > + if (!core_scsi3_ua_for_check_condition(cmd, &key, &asc,
> > + &ascq)) {
> > + cmd->scsi_status = SAM_STAT_BUSY;
>
> Here is another question I had about this code path.
>
> If you hit this case do you need to undo the setting of the
> SCF_SENT_CHECK_CONDITION bit done in
> transport_send_check_condition_and_sense or does the check for that bit
> need to be fixed up.
>
> With the code addition above we have this path setting
> SCF_SENT_CHECK_CONDITION so
> iscsit_check_task_reassign_expdatasn/iscsit_task_reassign_complete_scsi_cmnd/
> transport_send_task_abort would see the CC bit set and return.
Hello Mike,
Have you noticed that patch 1/3 that was attached to my previous e-mail
moves the code that sets the SCF_SENT_CHECK_CONDITION flag inside
translate_sense_reason() and past the point where SAM_STAT_BUSY is returned?
> > * The highest priority Unit Attentions are placed at the head of the
> > @@ -244,6 +246,7 @@ void core_scsi3_ua_for_check_condition(
> > * clearing it.
> > */
> > if (dev->dev_attrib.emulate_ua_intlck_ctrl != 0) {
> > + *key = UNIT_ATTENTION;
>
>
> Is the key already set to UNIT_ATTENTION here?
Good catch - that assignment is superfluous. But I would like to keep it
because I think it makes the core_scsi3_ua_for_check_condition() easy to
read. However, that assignment makes the following entry in sense_info_table[]
superfluous:
[TCM_CHECK_CONDITION_UNIT_ATTENTION] = {
.key = UNIT_ATTENTION,
},
Bart.��.n��������+%������w��{.n����j� ����{ay� ʇڙ���f���h������_�(�階�ݢj"��������G����?���&��
