sun
   microsystems              Systems Architecture Committee
_________________________________________________________________

Subject:	Multiple disk sector size support (phase I)

Submitted by:	Larry Liu

File:   	PSARC/2008/769/opinion.ascii

Date:		March 18th, 2009

Committee:	Mark Carlson, James Carlson, Glenn Skinner,
		Rick Matthews.

Product Approval Committee:

		Solaris PAC
		solaris-pac@sun.com

1.  Summary 

	This project is to enable Solaris to support disks with variable
	sector size in power of 2 starting at 512 bytes and upto 4096 bytes,
	i.e. 1 KBytes, 2KB or 4KB. The 4KB sector size disk will be released
	soon and Solaris need to support it.

2.  Decision & Precedence Information

	The project is approved as specified in reference [1].

	The project may deliver in a minor release of the ON consolidation.

3. Interfaces

The project exports the following interfaces.

       ________________________________________________________________________
      |                         Interface Exported                             |  
      |________________________________________________________________________|
      |	 Interface Name      | 	Classification   |       Comments              |  
      |______________________|___________________|_____________________________|
      | sdstrategy           |  Committed        |  To keep the stability of   |
      |	(b_lblkno in buf_t)  |			 |  buf(9S) structure and to   |
      |			     |			 |  keep backward compatibility|
      |			     |			 |  sd(7D) exports a logical   |
      |			     |			 |  block size as 512 bytes    |
      |			     |			 |  and sd(7D) is responsible  |
      |			     |			 |  for translating the logical|
      |			     |			 |  start block address in the |
      |			     |			 |  I/O request carried by     |
      |			     |      		 |  buf(9S) to physical disk   |
      |			     |			 |  block address.             |
      |______________________|___________________|_____________________________|
      |  DKIOCGMEDIAINFOEXT  |    Committed      |  A new ioctl to dkio(7I) to |
      |                      |                   |  report the physical        |
      |                      |                   |  blocksize of media.        |
      |    		     |          	 |  <sys/dkio.h>               |
      |______________________|___________________|_____________________________|
      |   rmw-type           |    Committed      |  A new [s]sd tunable in     |
      |                      |                   |  [s]sd-config-list          |
      |                      |                   |  UINT32                     |
      |______________________|___________________|_____________________________|
      | physical-block-size  |    Committed      |  A new [s]sd tunable in     |
      |                      |                   |  [s]sd-config-list.         |
      |                      |                   |  UINT32                     |
      |________________________________________________________________________|


       ________________________________________________________________________
      |                          Interfaces Imported                           |
      |________________________________________________________________________|
      |     Interface Name   |   Classification  |         Comment             |              
      |______________________|___________________|_____________________________|
      |  [s]sd-config-list   |   Committed       |   PSARC 2008/465            |
      |______________________|___________________|_____________________________|
      |  device-blksize(9P)  |   Committed       |   PSARC 2007/339            |
      |  blksize(9P)         |                   |                             |
      |________________________________________________________________________|


4.  Opinion

4.1	Extend existing ioctl

	During commitment, a question was raised about adding a new member to
	DKIOCGMEDIAINFO. If an old application issuing DKIOCGMEDIAINFO on an
	upgraded system, the kernel's copyout will corrupt memory. The project
	team decided to fix this with a new dkio(7I) ioctl DKIOCGMEDIAINFOEXT
	and leave DKIOVGMEDIAINFO unchanged.

4.2     RMW (Read Modify Write) design

	The mainly substantive issue discussed during ARC review was RMW design.
	One member asked since RMW occurs only when the application issues an
	unaligned access, and thus all modified applications will "do the right
	thing" to avoid performance degradation, why making RMW an option? Why
	not have RMW enabled all the time.

	The project team does not want to leave a hidden performance issue lurking
	and would prefer to let customers make an explicit choice to either change
	their applications or run under low performance.

	A member suggested an alternate proposal to export a "multi-state" RMW
	flag. By default, RMW is enabled as a "warning only" mode. And "return
	error on RMW" and "do not report RMW conditions" are other potential
	settings. The project team decided to take this suggestion.


5.  Minority Opinion(s)

	None.


6.  Advisory Information

	None

7.  Appendices

7.1.  Appendix A: Technical Changes Required

	None.

7.2.  Appendix B: Technical Changes Advised
	
	None.

7.3.  Appendix C: Reference Material

	Unless otherwise noted, materials are relative to the case directory
	PSARC 2008/769.
	
     1. Functional Specification.
	File: commitment.materials/design_doc

     2. PSARC Questionnaire
	File: PSARC_questionnaire.txt

     3. new ioctl description
	File: NewIOCTL.txt

     4. mail log 
	File: mail