Doc. No.: Date: Project: Ref. Doc.: PnP SCSI revision 0.95 Reply to: L.. Lamers To: PnP participants From: Lamers/Timm Subject: Minutes of SCSI Plug and Play Meeting - Colorado Springs, CO Agenda 1. Opening Remarks 1 2. Attendance 2 3. Approval of Agenda 2 4. Document Distribution 2 5. Key Issues 2 5.1. Default SCSI IDs 2 5.2. Scan order 3 5.3. Connector 3 5.4. Switchable Terminators 3 5.5. Document Maintenance 3 5.6. Certification 3 5.7. Reference Documents 3 6. Review of PnP SCSI revision 0.95 3 6.1. Conformance 4 6.2. Connectors 4 6.3. Cables 4 6.4. Terminators 4 6.5. Default SCSI IDs 5 6.6. System Configuration 6 6.7. Drive Number Assignment 6 6.8. Inquiry data 7 7. Action Items 7 8. Meeting Schedule 7 9. Adjournment 7 10 Steve Timm's Power Point Slides. 7 1. Opening Remarks Steve Timm, the Chair, called the meeting to order at 8:40 a.m., Friday, November 12, 1993. He thanked John Lohmeyer of NCR for hosting the meeting. As is customary, the people attending introduced themselves. A copy of the attendance list was circulated for attendance and corrections. Steve explained his role at Microsoft as a technical evangelist. His focus is on storage peripherals for current and future operating systems. He represents the storage industry to the developers within Microsoft. This meeting is intended to be a design review of PnP SCSI. SCSI is a natural fit in the industry efforts on Plug and Play. There are a number of other Plug and Play specifications being developed. PnP SCSI is intended to be platform and operating system independent. The primary goal on the first pass of PnP SCSI is to deal with the desktop market. The workstation and server segments are important and issues that relate to them will be discussed. However it may take a second generation PnP SCSI to deal with those issues. The task today is to identify the issues as they relate to the desktop market and seek consensus on resolving them. Dal Allan asked about plans for a SCSI reflector. John Lohmeyer offered to investigate the use of NCR info. 2. Attendance The following people participated in the review of the PnP SCSI specification. Voting is open to any attendee, but is restricted to one vote per company. 3M Rick Heidick Adaptec Norm Harris Adaptec Vince Bastiani Always Technology Robert Fogel AMD David Wang AMD David Skinner AMP Bob Whiteman Ancot Gary Porter AST Research Ken Grossman Bus Logic Clifford Strang Cirrus Dave Weber allas Semiconductor Loius Grantham Digital Edward A. Gardner Digital Bill Ham ENDL Dal Allan Exabyte Edward Lappin Future Domain Jeff Epstein Future Domain Mark Raymond Hewlett-Packard Jeff Williams IBM George Penokie Interphase David Lawson Methode Bob Masterson Microsoft Steve Timm Molex Joe Dambach NCR John Lohmeyer Qlogic Skip Jones Seagate Gene Milligan Seagate Gerry Holder Silicon Systems Steve FInch Tandem Computers John Mey Turbar Electronics Ray Kallio Unitrode Paul Alosi Western Digital Tak Asami Woven Electronics Doug Piper 3. Approval of Agenda The agenda was approved as noted in section 2. 4. Document Distribution Copies of the latest revision were available. 5. Key Issues 5.1. Default SCSI IDs The industry could not reach agreement on default settings of peripheral device IDs due to multiple competing requirements. 5.2. Scan order Inconsistent drive letter assignment and boot drive recognition on multi-drive systems can occur when either the host adapter is changed or another hard drive is attached. See detailed results (7.2) for an explanation. This issue will be addressed Dec. 1 by System BIOS and SCSI vendors to achieve a solution independent of SCSI bus scan order. Scan order was taken out of the spec. 5.3. Connector The consensus was to stay the 50-pin high-density connector. It was recognized that a second generation PnP SCSI may offer the 68-pin wide connector if sufficient interest is expressed. 5.4. Switchable Terminators Should peripherals containing disabled electrically switchable terminators be allowed to claim PnP SCSI compliance?. There is an issue of capacitance buildup. It was pointed out that switchable terminators even in their off state add to the node capacitance. Dave Skinner questioned whether the specification addresses the reliability of a fully configured system. The objective is to get a high probability of system operation if all devices and the host conform to Plug and Play. 5.5. Document Maintenance How will the document be maintained? It was suggested that the specification be processed as an X3T9 technical report. John Lohmeyer will follow-up on to do this. There was some discussion that a PnP SCSI reflector be set up to facilitate maintenance of the specification. There was a request from Erich Oetting that the document be organized as a technical reference document with separate sections for mechanical, electrical and logical. However, the person making the request was not willing to be the editor. Gerry Houlder requested a change in the definition of 'legacy'. change to non-scam, or remove 'between legacy devices'. Jeff requested a change to 6.2 par 3 - change "purchase" to "add" and add 'and cable'. 5.6. Certification David Skinner of AMD volunteered to work with the industry to establish PnP SCSI certification. 5.7. Reference Documents Section 2.0 list referenced documents. Gene Milligan questioned why the specific reference documents were included and why there were not documents that dealt with other than Intel architecture. It should include reference documents for other OS environments like SUN, SCSA, AT&T etc. It was agreed that these references would be added when requested by appropriate industry segments. At least one referenced document (CAM) is not x86 specific. It was requested that the Microsoft reference documents be available through MSDN. 6. Review of PnP SCSI revision 0.95 6.1. Conformance Section 4.0 lists the conformance requirements. How does the SCSI PnP specification help to ensure interoperability of SCSI devices? Can we assume that PnP compliant SCSI devices will work when integrated into a common system? Consider breaking down the test matrix on an OS by OS basis. Robert Whiteman asked if a compliance verification process would be establish prior to getting the PnP logo. Steve Timm stated that such a procedure is necessary but has not been established. Jeff Epstein stated that a specific logo for PnP SCSI should be adopted to promote the SCSI industry. Use of the logo would be based on in-house verification honor system to implement PnP compliant devices. Dave Skinner offered AMD as a host for a meeting on the conformance testing. AMD has experience in this arena with communications devices. Dave Skinner of AMD volunteered to explore setup of an industry wide PnP compliance program. The compliance checklist Appendix C needs to identify compliance criteria and test procedures to be defined. 6.2. Connectors Section 5.1 specifies the connector. Bill Ham question why the specification does not allow using the 68-pin connector. Gene Milligan noted that the 68-pin solution costs more, for cable, terminators, and pins. Steve Timm pointed out that the focus is on the desktop market, the volume is in the 50 position, and a single connector is mandatory. The workstation market may need 68 position but it is not today's problem. It needs to be not excluded in the long term. John Lohmeyer stated that a second generation PnP SCSI specification should be considered. This specification would be aimed at workstation. The method of securing must be clearly specified. John pointed out that per clause 5.2 latches are the retention mechanism. Doug Pfifer stated that clips are a significant problem because connector shells are getting bent. Should PnP specification jack-screws or clips? Since 50-pin connectors with inter-matable jack-screws are not generally available the consensus was to stay with clips. 6.3. Cables Norm Harris stated the cable assembly specification is being worked. Bill Ham noted his document X3T9.2/93-175 as being germane to the issue. 6.4. Terminators Section 5.3 specifies terminator requirements. It was agreed to add an SCSI-2 reference drawing in an appendix. This will specify that reserved lines are open. Bill Ham questioned the 4.0 voltage in section 5.3. It was noted that this requirement is at the terminator not the source. Bill also stated that the reserved lines in terminators as specified in SPI are open. Most active terminators today ground these lines per SCSI-2. Bill Ham asked if PnP was considering multi-initiator environment. No. Bill Ham requested that the 'boulay' termination be added as a requirement. Bob Whiteman stated it should be included as a reference in appendix. Bill asked for the addition of a redundant requirement in PnP SCSI. Paul Alosi noted that self-resetting devices should be used and circuit breakers should be allowed. He also noted that laptops are being developed that supply 3.3 v termpower. When these laptops are connected to an external device a higher voltage is supplied for termpower. The termpower requirement was changed from should to shall. Bob Masterson pointed out an error in section 6.1 par 2 needs the wording revised to eliminate the switchable requirement for internal only configurations. John suggested using the terminology near and far terminator instead of internal or external. Bill suggested that the system should definitely fail if the external terminator is not used. Innovation in this area is encouraged. A proposal should be developed by those desiring the value added. Mixed configuration - Ed Gardner stated that if a pcb is not at the bulkhead then having a switchable terminator at the bulkhead is difficult. The crux is that a separate terminator cost more than a switchable terminator on the adapter. 6.5. Default SCSI IDs Section 5.4.2 specifies the default SCSI IDs. Future Domain questioned the SCSI IDs assigned in table 1. Ed Gardner noted that table 1 should reflect the as configured in a system. George Penokie stated that he believes IBM position is that internal should be ID 6 and external at ID 0. George will verify it. The scan order plays into this, since the internal drive should be the boot device. The audience for table 1 was questioned. Is it an existing system with or without SCSI, as shipped or as configured, aimed at the system integrator or non-user? Norm Harris suggested eliminating the table. He noted that the scan order is only relevant in a non-PnP host adapter. In systems with SCSI internal, there are different preferences for default IDs. Steve Finch stated that the words internal/external be deleted. Consensus is table 1 is deleted. A note is added to Table 2 that IDs 5, 1, 0, are available for expansion. Change the title to delete the term 'PnP'. Mark Raymond of Future Domain agreed to work with Adaptec and NCR to come up with a proposal in conjunction with work on scan order independent drive letter assignment. 6.6. System Configuration Section 6.0 specifies system configurations. Cable specifications are needed to avoid excessive stub lengths. Adaptec agreed to make a proposal. A question was raised on the value of disabling the SCSI bus on detection of improper placement of terminators. This innovation was encouraged in conjunction with the design of automatic termination. A request to allow an external pluggable terminator in place of the switchable exit point terminator was not accepted due to significantly higher cost. 6.7. Drive Number Assignment Section 7.2 specifies the drive number assignment. This was the biggest issue coming into the meeting. Since it is clear that the industry and the installed base is divided on scan order, debating the merits of one order over another was cut off. Instead, the three goals of PnP listed below were clarified. Consensus was to replace the requirement for scan order in the specification with the three goals for PnP SCSI. The issue is more of a system level boot problem rather than a SCSI issue. Future Domain volunteered to call an industry meeting with leading BIOS, Motherboard, OS and SCSI vendors to come up with a proposal to address the stated goals of PnP. The approach will be to virtualize SCSI ID / Drive Letter mapping. This meeting is tentatively set for Dec. 1 in the Bay area. Since these issues may affect the design of PnP SCSI devices, the specification will not be rolled to 1.0 until they are resolved. Background: Today, the SCSI option ROM BIOS scans the SCSI bus using the Inquiry command to search for a boot device. A mix of option ROM's issue the Inquiry command starting with the highest or lowest SCSI ID. The first hard drive detected is attached to Interrupt 13 (address 80H by MS DOS convention). This traditional SCSI boot sequence has the industry divided on what scan order should be used. If all option ROM's scanned in the same order, then three PnP features may be readily achieved with no added cost including: a) Change host adapters with no modification to drive letter assignment, maintaining a consistent I/O model. b) Add a second hard drive without mysteriously changing the boot device. c) Enable end users to select the boot device on multi drive systems. Note - in legacy systems, you have to know the adapter scan order and set the manual device ID switches so that the boot drive will be the first hard drive scanned! It is desirable for virgin boot cycles to enable the end user to select which hard drive will be the boot device under software control. Another related issue for PnP adapters is changing drive letter assignments and the boot device on subsequent boot cycles due to dependence on Scan Order. Even with PnP SCAM, auto ID assignment is random. Therefore, the PnP SCSI spec. requires host adapters to retain the drive letter assignments in non volatile memory. Note, this is only an issue for multi-drive systems. It is interesting to note that the PnP BIOS spec. requires the BIOS to remember the boot channel in cases where there are more than one (e.g.; IDE, SCSI adapters). A solution to consistent drive letter and boot drive assignment would be to expand this requirement to record the mapping of drive letters to devices. Geometry translation is intended to promote consistency when moving devices. Jeff wants to delete annex A and change 7.1 to allow use of existing geometry. Add 'use existing geometry. Annex A gives a pseudo code example of how to determine it." Ed requested the 'dosiness' of the first two sentences be cleaned up. Mark Raymond suggested that compatibility with legacy systems and legacy devices is important. Scan order is not really the issue, but a symptom of the drive letter assignment algorithms used in the past. What is needed is a more sophisticated approach that allows mapping of SCSI devices to specific drive numbers. This would preserve the boot device and should allow a user to maintain file locations if multiple SCSI devices are present. Mark suggested deleting section 7.2 and forming a study group to define a workable solution. The SCSI reflector will be used to announce the meeting. Mark will be the champion. Gerry Houlder requested that 'excessive delays' needs a numerical values. The intent is to eliminate gross problems during boot. Norm Harris will champion the host adapter SCSI command initialization sequence. Are software configurable remembered through a power cycle? 6.8. Inquiry data Section 7.4 specifies the behavior of the INQUIRY command. After lengthy discussion, it was agreed that timing requirements for device response to the INQUIRY command are adequately covered by SCAM. Such timing has long been a subject of debate within the SCSI committee. 7. Action Items Mark Raymond - Drive letter assignment & default IDs. Norm Harris - SCSI command sequence at power on. Tom Solomon - Cable specification. John Lohmeyer to investigate the use of NCR info for distribution of PnP SCSI e-mail. Larry Lamers to prepare revision 0.96 of PnP SCSI. Larry Lamers to prepare minutes and slides used to facilitate the spec. review and results will be posted publicly to confirm issue closure. Jeff Epstein to coordinated a meeting to define a system boot cycle that consistently assigns drive letters, consistently determines the boot device and allows the end user to easily change the boot device. 8. Meeting Schedule December 1, 1993 - Santa Clara, CA hosted by Future Domain. January 6, 1994 - Colo Spgs, CO hosted by NCR. 9. Adjournment The meeting adjourned at 2:00pm. 10. Steve Timm's Power Point Slides 1. Purpose and Scope of SCSI PnP 1.1. SCSI PnP is an Industry Specification 1.2. OS and Platform independent 1.3. Address volume PC desktop market as a priority 1.4. Define the SCSI PnP Profile in terms of existing SCSI standards 1.4.1. Don't re-invent 1.5. Make computing easier for the end user 2. PnP SCSI Industry Group 2.1. Adaptec 2.2. Digital Equipment 2.3. Future Domain 2.4. Maxtor 2.5. Microsoft 2.6. NCR 3. Objectives for this Meeting 3.1. To review the content of the spec. 3.1.1. Solicit additional feedback 3.1.2. Address questions 3.1.3. Clarify issues and close where feasible 3.2. To achieve industry consensus 3.3. To finalize the definition of SCSI PnP 3.4. Outcome: Rev.1.0 public release 4. Agenda 4.1. Ground Rules 4.2. Overview of Feedback on Revision .95 4.2.1. Identify Open Issues 4.3. Specification Review 4.3.1. Discuss and Close Issues where possible 4.4. Summary of Action Plan 4.4.1. Revisit hard issues 5. Ground Rules 5.1. First Pass of Spec. Review 5.1.1. Address questions 5.1.2. Clarify issues 5.1.3. Close issues where possible 5.1.4. Defer extended discussion of any hard issues till last 5.2. Getting Consensus 5.2.1. Unanimous - no qualified objections 5.2.2. Majority vote - no serious objections 5.2.3. Proposal for resolution off line 5.3. Other? 6. Overview of Industry Feedback 6.1. Rev. .95 posted mid Oct. 6.2. Low volume of feedback so far ... 6.3. Total of five open items 6.4. More today? 7. Open Issue Summary 7.1. TERMPWR current limit (5.3): Unitrode 7.2. System Config./Termination requirements (6.0): editorial 7.3. Automatic Termination (6.3): Industry 7.4. Scan Order (7.2): Future Domain; IBM 7.5. Compliance Checklist (Appendix C): editorial 8. Specification Review 8.1. Scope 7 8.2. Reference documents 7 8.3. Glossary and Abbreviations 7 8.4. Conformance 8 9. Specification Review 9.1. Hardware 8 9.2. Connectors 8 9.3. Cables 8 9.4. Terminator requirements 8 9.5. SCSI ID assignment 9 9.6. Automatic SCSI ID assignment 9 9.7. Default SCSI ID's 9 9.8. SCAM ID assignment algorithm 10 10. TERMPWR current limit 10.1. Unitrode: Can the wording in section 5.3 be changed to include electronic circuit breakers that work just as a function of current (eg. UCC3912) 10.2. Resolution: @positive temperature coefficient device or self resetting circuit breaker@ 10.3. Table 1 - Legacy SCSI device ID Assignment 10.4. Host Adapter 10.5. External Disk Drive 10.6. Tape or R/W Optical 10.7. CD-ROM 10.8. Scanner/Printer 10.9. Docking Station Disk Drive 10.10. Internal Disk Drive 11. Table 2 - As shipped PnP SCSI Default ID 11.1. Host Adapter 11.2. Disk Drive 11.3. Tape or R/W Optical 11.4. CD-ROM 11.5. Scanner/Printer 12. Specification Review 12.1. System configurations 10 12.2. Internal configuration 10 12.3. External configuration 11 12.4. Mixed configuration 11 13. Plug and Play SCSI Overview 14. PnP SCSI Checklist 14.1. All external connectors are 50-pin high-density shielded connectors 14.2. The Host has one external connector 14.3. All Peripheral Subsytems must have two external connectors 14.4. External cable assemblies must meet SPI electrical characteristics 14.5. All SCSI peripheral devices must not include terminators 14.6. All terminators are regulated (a.k.a., active) and must meet SPI requirements 14.7. The host internal SCSI bus must be terminated correctly: 14.7.1. The internal terminator is at the end of the internal SCSI bus 14.7.2. The exit-point terminator must be disabled automatically when an external cable is attached 14.8. The host and all Peripheral Subsystems should provide Terminator Power: 14.8.1. PTC or circuit breakers must be used -- not fuses 14.9. Adding a Peripheral Subsystem requires an external terminator at the end of the bus 15. Automatic Termination 15.1. Host Adapter Circuit 15.2. Required for internal/external system configurations 15.3. Industry required to implement by 7-1-94 16. Specification Review 16.1. Software 11 16.2. Geometry translation 11 16.3. Drive number assignment 11 16.4. READ CAPACITY command 12 16.5. INQUIRY command 12 16.6. Synchronous negotiation 12 16.7. Virtual DMA services (VDS) 12 16.8. Device driver architecture 12 16.9. Software configuration 12 17. Scan Order 17.1. Future Domain: STRONGLY OBJECTS to reverse scan order drive assignment and use of the lowest priority SCSI ID as the system boot disk 17.2. IBM: Low to high scan has the potential for causing major headaches if not outright breaking things 17.3. For Plug and Play, SCAN Order is not the real issue! 18. Plug and Play Goals RE: Scan Order 18.1. To allow end users to change host adapters or add a hard drive to their system without confusing: 18.1.1. drive letter assignment 18.1.2. boot drive recognition 18.2. Enable the user to specify the boot devices explicitly or by default without confusion 18.2.1. It@s awkward to set device ID assuming a Scan order to select the boot device, given the split in Scan Order 19. Proposal for Resolution Preliminary Thinking RE: Scan Order 19.1. Make drive letters, boot device recognition independent of scan order 19.2. Virgin Boot: BIOS config. utility to enable end user to choose boot device on multiple drive systems 19.2.1. Utility provides device info for user to make informed decision 19.3. BIOS to save SCAM string on PnP systems to enable consistent drive letter assignment / boot drive recognition for subsequent boot cycles 19.4. Needs review by motherboard/BIOS vendors ... can@t solve in this forum 19.5. Short term alternatives: 19.5.1. Host Adapter to save enough information to maintain consistent boot device recognition independent of Scan Order (could extend to provide consistent drive letters) 19.5.2. Other Ideas 20. Scan Order Decision Factors 20.1. ID priority 20.2. Most hard drives ship at ID6 20.3. Majority of installed base is compatible with 20.3.1. Low to High? 20.3.2. High to Low? 21. Future Domain Position 21.1. Based on a survey of system manufacturers, peripheral manufacturers, and retailers: a high to low scan with boot disk at ID 6 is the most consistent with the industry today 21.2. Technical reasons reinforce the choice of the most important devices being at the higher ID's and the scan from hi to low 22. Summary Scan Order Proposal 22.1. OS and BIOS vendors to make drive letter assignment and boot drive recognition independent of Scan Order 22.2. Add a requirement for BIOS to query User to determine boot device 22.3. Save SCAM string for subsequent reboots 23. Specification Review 23.1. APPENDIX A - CHS mapping 13 23.2. APPENDIX B - CD-ROM SCSI command profile 15 23.3. APPENDIX C - Checklist of compliance requirements 24. Q & A 25. Follow up Action Plan