One last plea for correct behavior of mode select pages

Jim.Coomes at seagate.com Jim.Coomes at seagate.com
Thu Feb 15 14:02:29 PST 2001


* From the T10 Reflector (t10 at t10.org), posted by:
* Jim.Coomes at seagate.com
*
Bob,

In your message below you plead that following the "first (FCP)
PRLI" the current mode pages should be se to the saved or default
values.

Take a configuration with multiple initiators and a shared target
supporting shared mode pages. The first initiator to successfully
complete process login with the target following target power-on,
LIP-Reset, or Target Reset will indeed find the shared mode page
values set to the saved or default values. However, there is no
method for an initiator to determine that has performed the
"first" process login. There may or may not be another initiator
that has successfully completed process login and changed the
mode page values.

As the initiator can not determine it is "first", it has to be
capable of determining the mode page values are acceptable when
is is in a configuration where mode pages are being changed. The
mode pages values following any process login are indeterminate
or "unspecified". The mode page values are determined by the last
SCSI defined reset, FC reset mapped to SCSI (LIP Reset) or SCSI
Mode Select command.

See you in Dallas

Jim


---------------------- Forwarded by Jim Coomes on 02/15/2001 02:26 PM
---------------------------


Robert Snively <rsnively at Brocade.COM>@network.com on 02/13/2001 12:09:39 PM

Please respond to Robert Snively <rsnively at Brocade.COM>

Sent by:  owner-fc at network.com


To:   "Fibre Channel T11 reflector (E-mail)" <fc at network.com>, "T10
       Reflector (E-mail)" <t10 at t10.org>
cc:

Subject:  One last plea for correct behavior of mode select pages




Note my comments at the end, where I indicate  my dissatisfaction with the
agreed-upon solution.  This is likely to come  back and haunt us in no
uncertain terms.  Customers who demand this of  their vendors should be
either educated or provided with an explicitly  non-standard product.  The
standard product should have a defined current  value after a first PRLI
with shared mode pages equal to the saved or default  values, as specified
by SCSI-2, SPC, and SPC-2.  Fudging this to be vendor  specific is a really
bad idea.

I am soliciting correction of this at the  FCP-2 meeting in March.  It is
installed as shown below for now.

Bob



2.3  Consideration of mode pages,  technical


CPQ-003: PDF 33, Page 19 (Technical) Section 4.9 Clearing effects of task
management, FCP, FC-FS, and FC-AL-2 actions and Section 6.3.4 New or
repeated  PRLI

<excerpts from FCP-2> Section 4.9 Clearing effects of task management,
FCP, FC-FS, and FC-AL-2 actions Table 6 (and maybe 4) does not reflect
working  group agreement.

Table 6 - Management of mode pages during PRLI and PRLO PRLI/shared/other
initiators = use saved or default mode pages [00-342r0 requested
"unchanged,  most current values if no reset"] [discussing 00-342r0, Sep
CAP WG agreed on  "unspecified" with a note explaining what "unspecified"
means]

Table 4 - Clearing effects of link related functions (excerpt) For
PRLI(8)(13), PRLO Target mode page parameters restored from saved pages
(...)  For all initiator ports(12) = N Only for initiator port associated
with the  action = Y

Section 6.3.4 New or repeated PRLI After the completion of any new or
repeated binding PRLI, both members of all image pairs successfully
established  shall have the same state as they would have after a hard
reset or a power on  with respect to each other. All clearing actions
specified in 4.9 shall be  performed. Tasks, reservations, status, and MODE
SELECT parameters for other  image pairs are not affected. </excerpts>

According to 00-307r0 Minutes from SCSI CAP 11 Sep 2000: "Jim Coomes
reviewed  the issues between FCP-2 process login and proposed that mode
pages not change  as a result of process login (00-342r0) unless a power on
reset or target reset  also occurred. Bob Snively stated a preference for
making the behavior  unspecified. Concerns were raised about readers not
knowing what unspecified  means and it was agreed that a descriptive note
is required."

According to 00-300r3 Installation of corrections in FCP-2, item 4.115: IBM
comment from George Penokie (Editorial) "Page 14 - table 5 - The two
entries  labeled 'discard current mode pages' should be changed to 'not
specified'. There  is no reason to force the device to discard current mode
pages or do any other  action with mode pages at this point. Response:
Accepted. See 6.51.  Installation: Installed as requested. Note that
T10-00-342r0 (Clearing affects  on other initiators, by Jim Coomes) also
references this issue. At the time, the  thought was to specify the PRLI
unshared case as "unspecified", but after  further study, I have used "use
saved or default mode pages". Approved October  30, 2000."

00-342r0 requested a change in the "PRLI/shared/none logged in" row, not
the  "unshared" row. I assume that "PRLI unshared case" was meant to be
"PRLI  shared/none logged in" in the 00-300r3 resolution comment.

The issue is that an initiator probably doesn't know if other initiators
happened to be logged in or not, so it always has to check shared mode
pages  after logging in. Forcing the target to revert to saved/default
pages on either  last logout or first login doesn't help anything. Many
existing FCP targets  preserve mode pages, and a change in FCP-2 that adds
no value but makes existing  devices noncompliant should be avoided.

With non-shared pages, the initiator does know the full state of the target
with respect to its page, so full specification is helpful. The value after
logout doesn't matter, because the initiator cannot accessing the target
without  a new login. Thus specifying that pages are reset after login is
adequate.

00-342r0 suggested mentioning reset as a special case. That is covered in
table 4, so the issue can be ignored in table 6.

Table 6 should use "not specified" for the "PRLI/shared/none logged in"
case.

Table 4 should refer entirely to table 6 for the "only for initiator
associated with the action" row, rather than trying to apply a "Y" or "N"
to  them. "Saved or default" means "Y," "not specified" means "maybe", and
"maintain  current" means "N."

The first sentence of Section 6.3.4 needs this added: "except for mode page
parameters (see table 6)."

Response:

Three changes are requested by this comment:

1) Table 6 change:

Proposed change is: Table 6 should use "not specified" for the
"PRLI/shared/none logged in" case.

Action: No change is made. In the October meeting there was a strong
consensus that, if there are no active process logins with a device, it is
reset. The expected behavior after exiting from such a reset when the PRLI
is  performed is that the device is in a known and expected state, the
default or  saved state.

After further study, the working group chose to use the term "not
specified"  for this case in a vote of 7 in favor, 2 against, and 2
abstaining in the  meeting of January 15, 2001.

Installation:

Installed as voted. For the record, let me  state that I think this is a
really bad solution that will come back and haunt  us. It is likely that
profile statements will be required to correct this. I  believe that it is
a direct and non-negotiable violation of SCSI-2, 8.2.10.2 and  SPC-2,
7.11.2. Note that the clearing and management of mode pages is not
specified in SAM-2, SPI-3, or SPI-4.


2) Table 4 change:

Proposed change is: Table 4 should refer entirely to table 6 for the "only
for initiator associated with the action" row, rather than trying to apply
a "Y"  or "N" to them. "Saved or default" means "Y," "not specified" means
"maybe", and  "maintain current" means "N."

Action: Table 6 only references PRLI and PRLO, not failed discovery. Those
two cases have their "Y" and "N" indications replaced with a note
referencing  table 6, but the failed discovery cases remain unchanged.

After further study, the working group chose to replace the Y and N values
in  this location of the tables with dashes containing a super-script
reference to  note 12, which references table 6. This resolution was
approved in the meeting  of January 15, 2001.

Installation:

The change was made for the two locations associated with the two mode
select  page rows and the PRLI/PRLO column. The superscript notes were
removed from the  two mode select page row definitions in both table 4 and
5. Please review for completeness.


Section 6.3.4, first sentence change:

Proposed change is: The first sentence of Section 6.3.4 needs this added:
"except for mode page parameters (see table 6)."

Action: Accepted. The first sentence is changed accordingly. This action
was  approved in the January 15, 2001 meeting.

Installation:

Installed as requested. For the record, let  me state that I think this is
a really bad solution that will come back and  haunt us. It is likely that
profile statements will be required to correct this.  I believe that it is
a direct and non-negotiable violation of SCSI-2, 8.2.10.2  and SPC-2,
7.11.2. Note that the clearing and management of mode pages is not
specified in SAM-2, SPI-3, or SPI-4.


Bob  Snively                         e-mail:    rsnively at brocade.com
Brocade Communications  Systems     phone:  408 487 8135
1745 Technology  Drive
San Jose, CA 95110




*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list