From don.cragun@sun.com Fri Mar 30 15:00:41 2007 Date: Fri, 30 Mar 2007 15:01:15 -0700 (PDT) From: Don Cragun Subject: Need help finding x86 and x64 System V ABI test suites for PSARC/2007/064 To: Paul.Neary@sun.com, Robert.Sohigian@sun.com, Tim.Marsland@sun.com Cc: Andrew.Roach@sun.com, Gary.Winiger@sun.com, Kristin.Amundsen@sun.com, Mara.Roccaforte@sun.com, Mike.Sullivan@sun.com, Tim.Sparlin@sun.com Content-transfer-encoding: 7BIT X-PMX-Version: 5.2.0.264296 Status: RO X-Lines: 26 Content-Type: TEXT/plain; charset="us-ascii" Content-Length: 1127 Paul, Robert, & Tim, I am trying to track down System V ABI test suites for x86 and x64 hardware lines for use to verify that a project recently approved by PSARC doesn't break the ABI. The PSARC case is 2007/064 (Unified POSIX and Windows Credentials for Solaris). Eric Luiggi's group has provided information for the SPARC hardware test suites (SPARC Compliance Definition version 2.4, System V gABI version 2.1, and System V SPARC psABI version 2.1). Eric's group runs these tests as part of the conformance testing we run before we ship each Solaris release. Do any of you know: 1. who runs the corresponding System V gABI and psABI test suites for x86 and x84 hardware, 2. where we can find copies of these test suites for x86 and x64 so the project team can run these tests before they integrate into the ONNV source base, and/or 3. what the correct names (including version numbers) are for the x86 and x64 gABI and psABI test suites? Any help in providing answers to these questions or pointers to people who may be able to answer these questions will be greatly appreciated. Cheers, Don From dwc@spartan.eng.sun.com Mon Apr 2 21:36:30 2007 Date: Mon, 2 Apr 2007 21:35:42 -0700 (PDT) From: Don Cragun To: Tim.Marsland@sun.com, gww@eng.sun.com Cc: Kristin.Amundsen@sun.com, Mara.Roccaforte@sun.com, Mike.Sullivan@sun.com, Paul.Neary@sun.com, Robert.Sohigian@sun.com, Tim.Sparlin@sun.com, andrew.roach@sun.com Subject: Re: Need help finding x86 and x64 System V ABI test suites for PSARC/2007/064 Content-Length: 4676 Status: RO X-Lines: 118 Gary, Although I don't doubt Andy's recollection on these issues, I will note that the System V x86 psABI that I have in my office has the same publication date on it as the original System V SPARC psABI. Given that AT&T produced an x86 psABI in the same time frame as the SPARC psABI, I'm surprised that they didn't also produce a test suite to go along with it. (Whether we acquired a copy of that test suite is a different issue since we weren't actively pursuing an x86 kernel at that time.) It was my impression that Tim and his project team created an x64 psABI even though it would have been after the AT&T breakup. I was under the impression that we created a test suite by adapting the SPARC or x86 psABI tests internally, but I could easily be wrong. I'd prefer that we wait to hear from Tim before giving up on this. Thanks, Don >From gww@eng.sun.com Mon Apr 2 16:44:00 2007 > >> Solaris never conformed (or claimed to conform) to the System V x86 ABI. >> It was produced after the SMI/ATT rift (during the SVR4.2 days). >> Similarly there is no 64-bit x64 System V ABI (as there's no such >> 'thing' to base one on). >> >> AMD tried to create an x64 ABI, but it was/is little more than calling >> conventions and register layout - not what we'd call an ABI. Tim has >> most of the details here. >> >> So I believe the answers are (but stand to be corrected): >> >> 1. Nobody >> 2. They don't exist >> 3. There are none > > Perhaps to help clarify things. 2007/064 wishes to change the > type of uid_t and gid_t from int to uint and provide for ephemeral > IDs. The project team argued that, as specified, having ephemeral > IDs in the (negative) range [0x80000000:0xfffffffe] would not cause > existing standards or applications to fail. PSARC TCRed the case > to ensure that standard tests would not fail. > > Don's request is to ensure the TCR has the correct standards tests. > The TCR at present says: > > 1. Type changes for uid_t and gid_t may not be > integrated unless all of the official generic and > processor specific ABI conformance tests for the > System V ABI pass. The list of tests is: > > A. SPARC: > > A.1. SCD version 2.4.1. > > A.2. gABI version 2.1. > > A.3. SPARC psABI version 2.1. > > B x86: > > B.1. gABI version ***. > > B.2. x86 psABI version ***. > > C. x64: > > C.1. gABI version ***. > > C.2. x64 psABI version ***. > > If the tests all pass successfully or all pass > after any newly discovered bugs are fixed, then > the type changes are accepted and may integrate. > If an unresolvable test issue is discovered, the > project team will return to ARC to discuss an > appropriate resolution, including discussing the > relative merits of leaving the types alone (i.e., > negative values for ephemeral IDs) versus docu- > menting a difference with respect to the ABI docu- > ment (and thus conforming to the language of POSIX > with respect to non-negative values). > > > From Andy's reply, perhaps the x86 and x64 parts should be removed. > I presume then that no tests are run on x86 or x64 related to > any standards. > >Gary.. >> Andy >> >> >> Don Cragun wrote: >> > Paul, Robert, & Tim, >> > I am trying to track down System V ABI test suites for x86 and >> > x64 hardware lines for use to verify that a project recently approved >> > by PSARC doesn't break the ABI. The PSARC case is 2007/064 (Unified >> > POSIX and Windows Credentials for Solaris). >> > Eric Luiggi's group has provided information for the SPARC >> > hardware test suites (SPARC Compliance Definition version 2.4, System V >> > gABI version 2.1, and System V SPARC psABI version 2.1). Eric's group >> > runs these tests as part of the conformance testing we run before we >> > ship each Solaris release. >> > Do any of you know: >> > 1. who runs the corresponding System V gABI and psABI test suites for >> > x86 and x84 hardware, >> > 2. where we can find copies of these test suites for x86 and x64 so >> > the project team can run these tests before they integrate into the >> > ONNV source base, and/or >> > 3. what the correct names (including version numbers) are for the x86 >> > and x64 gABI and psABI test suites? >> > >> > Any help in providing answers to these questions or pointers to >> > people who may be able to answer these questions will be greatly >> > appreciated. >> > >> > Cheers, >> > Don > From Tim.Marsland@sun.com Mon Apr 2 22:11:34 2007 Date: Mon, 2 Apr 2007 22:10:46 -0700 From: Tim Marsland To: Don Cragun Cc: Tim.Marsland@sun.com, gww@eng.sun.com, Kristin.Amundsen@sun.com, Mara.Roccaforte@sun.com, Mike.Sullivan@sun.com, Paul.Neary@sun.com, Robert.Sohigian@sun.com, Tim.Sparlin@sun.com, andrew.roach@sun.com Subject: Re: Need help finding x86 and x64 System V ABI test suites for PSARC/2007/064 Content-Disposition: inline User-Agent: Mutt/1.4.2.1i Status: RO X-Lines: 141 Content-Type: text/plain; charset="us-ascii" Content-Length: 6121 * Don Cragun [2007-04-02 21:35]: > Gary, > Although I don't doubt Andy's recollection on these issues, I > will note that the System V x86 psABI that I have in my office has the > same publication date on it as the original System V SPARC psABI. > Given that AT&T produced an x86 psABI in the same time frame as the > SPARC psABI, I'm surprised that they didn't also produce a test suite > to go along with it. (Whether we acquired a copy of that test suite is > a different issue since we weren't actively pursuing an x86 kernel at > that time.) It was my impression that Tim and his project team created > an x64 psABI even though it would have been after the AT&T breakup. I > was under the impression that we created a test suite by adapting the > SPARC or x86 psABI tests internally, but I could easily be wrong. > I'd prefer that we wait to hear from Tim before giving up on > this. Well, as for x64 we certainly tried to move the psABI effort along, but it was a dismal failure. In particular, no-one in the Linux world -really- wanted to work on a System V ABI; particularly when no Linux environment is even vaguely System V ABI compliant - except where it comes to calling conventions and things relating to ELF. And even that part is still in limbo -- I think Terry is still trying to get closure on the C++ side. As for tests, I thought that the test group definitely produced some additional tests for assessing 64-bit functionality on amd64, but in the absence of a psABI, I don't see how they could've created a psABI-specific test; though there were definitely conversations around pursuing that idea as part of creating the psABI specification. Bill Holler was involved in the amd64 testing effort; perhaps he can remember if any psABI-derived tests were created. tim P.S. Since I didn't keep up with the case, did we determine what the impact of C++ name mangling was on method signatures containing uid_t's in the face of changing the underlying type? That would seem to generate binary incompatibilities vs. source. > > Thanks, > Don > > >From gww@eng.sun.com Mon Apr 2 16:44:00 2007 > > > >> Solaris never conformed (or claimed to conform) to the System V x86 ABI. > >> It was produced after the SMI/ATT rift (during the SVR4.2 days). > >> Similarly there is no 64-bit x64 System V ABI (as there's no such > >> 'thing' to base one on). > >> > >> AMD tried to create an x64 ABI, but it was/is little more than calling > >> conventions and register layout - not what we'd call an ABI. Tim has > >> most of the details here. > >> > >> So I believe the answers are (but stand to be corrected): > >> > >> 1. Nobody > >> 2. They don't exist > >> 3. There are none > > > > Perhaps to help clarify things. 2007/064 wishes to change the > > type of uid_t and gid_t from int to uint and provide for ephemeral > > IDs. The project team argued that, as specified, having ephemeral > > IDs in the (negative) range [0x80000000:0xfffffffe] would not cause > > existing standards or applications to fail. PSARC TCRed the case > > to ensure that standard tests would not fail. > > > > Don's request is to ensure the TCR has the correct standards tests. > > The TCR at present says: > > > > 1. Type changes for uid_t and gid_t may not be > > integrated unless all of the official generic and > > processor specific ABI conformance tests for the > > System V ABI pass. The list of tests is: > > > > A. SPARC: > > > > A.1. SCD version 2.4.1. > > > > A.2. gABI version 2.1. > > > > A.3. SPARC psABI version 2.1. > > > > B x86: > > > > B.1. gABI version ***. > > > > B.2. x86 psABI version ***. > > > > C. x64: > > > > C.1. gABI version ***. > > > > C.2. x64 psABI version ***. > > > > If the tests all pass successfully or all pass > > after any newly discovered bugs are fixed, then > > the type changes are accepted and may integrate. > > If an unresolvable test issue is discovered, the > > project team will return to ARC to discuss an > > appropriate resolution, including discussing the > > relative merits of leaving the types alone (i.e., > > negative values for ephemeral IDs) versus docu- > > menting a difference with respect to the ABI docu- > > ment (and thus conforming to the language of POSIX > > with respect to non-negative values). > > > > > > From Andy's reply, perhaps the x86 and x64 parts should be removed. > > I presume then that no tests are run on x86 or x64 related to > > any standards. > > > >Gary.. > >> Andy > >> > >> > >> Don Cragun wrote: > >> > Paul, Robert, & Tim, > >> > I am trying to track down System V ABI test suites for x86 and > >> > x64 hardware lines for use to verify that a project recently approved > >> > by PSARC doesn't break the ABI. The PSARC case is 2007/064 (Unified > >> > POSIX and Windows Credentials for Solaris). > >> > Eric Luiggi's group has provided information for the SPARC > >> > hardware test suites (SPARC Compliance Definition version 2.4, System V > >> > gABI version 2.1, and System V SPARC psABI version 2.1). Eric's group > >> > runs these tests as part of the conformance testing we run before we > >> > ship each Solaris release. > >> > Do any of you know: > >> > 1. who runs the corresponding System V gABI and psABI test suites for > >> > x86 and x84 hardware, > >> > 2. where we can find copies of these test suites for x86 and x64 so > >> > the project team can run these tests before they integrate into the > >> > ONNV source base, and/or > >> > 3. what the correct names (including version numbers) are for the x86 > >> > and x64 gABI and psABI test suites? > >> > > >> > Any help in providing answers to these questions or pointers to > >> > people who may be able to answer these questions will be greatly > >> > appreciated. > >> > > >> > Cheers, > >> > Don From Bill.Holler@Sun.COM Mon Apr 9 16:17:56 2007 Date: Mon, 09 Apr 2007 17:17:20 -0700 From: Bill Holler Subject: Re: [Fwd: Re: Need help finding x86 and x64 System V ABI test suites for PSARC/2007/064] To: "Robert G. Sohigian" , Tim.Marsland@Sun.COM, don.cragun@Sun.COM, gww@eng.sun.com, Kristin.Amundsen@Sun.COM, Mara.Roccaforte@Sun.COM, Mike.Sullivan@Sun.COM, Paul.Neary@Sun.COM, Tim.Sparlin@Sun.COM, andrew.roach@Sun.COM Content-transfer-encoding: 7BIT X-Accept-Language: en-us, en User-Agent: Mozilla/5.0 (X11; U; SunOS sun4u; en-US; rv:1.7) Gecko/20060727 Status: RO X-Lines: 183 Content-Type: text/plain; charset="us-ascii" Content-Length: 7245 The kernel QA group did not produce X64 ABI tests. I vaguely remember discussing ABI tests in the AMD64 time frame. We thought another group might create ABI tests, but I cannot remember which group or why we thought that? The x64 ABI spec available at the time was useless for creating tests as it was mostly just calling conventions (which is tested implicitly). It does not define uid_t nor gid_t. Regards, Bill Robert G. Sohigian wrote On 04/03/07 00:54,: > Bill-- > > Any comments? > > --Robs > > -------- Original Message -------- > Subject: Re: Need help finding x86 and x64 System V ABI test suites for > PSARC/2007/064 > Date: Mon, 02 Apr 2007 22:10:46 -0700 > From: Tim Marsland > Reply-To: Tim Marsland > Organization: Software; Sun Microsystems, Inc. > To: Don Cragun > CC: Tim.Marsland@Sun.COM, gww@eng.sun.com, Kristin.Amundsen@Sun.COM, > Mara.Roccaforte@Sun.COM, Mike.Sullivan@sun.com, Paul.Neary@Sun.COM, > Robert.Sohigian@Sun.COM, Tim.Sparlin@Sun.COM, andrew.roach@sun.com > References: <200704030435.l334Zg8N029633@spartan.SFBay.Sun.COM> > > > > * Don Cragun [2007-04-02 21:35]: > >>Gary, >> Although I don't doubt Andy's recollection on these issues, I >>will note that the System V x86 psABI that I have in my office has the >>same publication date on it as the original System V SPARC psABI. >>Given that AT&T produced an x86 psABI in the same time frame as the >>SPARC psABI, I'm surprised that they didn't also produce a test suite >>to go along with it. (Whether we acquired a copy of that test suite is >>a different issue since we weren't actively pursuing an x86 kernel at >>that time.) It was my impression that Tim and his project team created >>an x64 psABI even though it would have been after the AT&T breakup. I >>was under the impression that we created a test suite by adapting the >>SPARC or x86 psABI tests internally, but I could easily be wrong. >> I'd prefer that we wait to hear from Tim before giving up on >>this. > > > Well, as for x64 we certainly tried to move the psABI effort along, but it was > a dismal failure. In particular, no-one in the Linux world -really- > wanted to work on a System V ABI; particularly when no Linux environment > is even vaguely System V ABI compliant - except where it comes to calling > conventions and things relating to ELF. And even that part is still > in limbo -- I think Terry is still trying to get closure on the C++ side. > > As for tests, I thought that the test group definitely produced some additional > tests for assessing 64-bit functionality on amd64, but in the absence of > a psABI, I don't see how they could've created a psABI-specific test; though > there were definitely conversations around pursuing that idea as part of > creating the psABI specification. Bill Holler was involved in the amd64 > testing effort; perhaps he can remember if any psABI-derived tests > were created. > > tim > > P.S. Since I didn't keep up with the case, did we determine what the impact > of C++ name mangling was on method signatures containing uid_t's > in the face of changing the underlying type? That would seem to > generate binary incompatibilities vs. source. > > >> Thanks, >> Don >> >>>From gww@eng.sun.com Mon Apr 2 16:44:00 2007 >> >>>>Solaris never conformed (or claimed to conform) to the System V x86 ABI. >>>>It was produced after the SMI/ATT rift (during the SVR4.2 days). >>>>Similarly there is no 64-bit x64 System V ABI (as there's no such >>>>'thing' to base one on). >>>> >>>>AMD tried to create an x64 ABI, but it was/is little more than calling >>>>conventions and register layout - not what we'd call an ABI. Tim has >>>>most of the details here. >>>> >>>>So I believe the answers are (but stand to be corrected): >>>> >>>>1. Nobody >>>>2. They don't exist >>>>3. There are none >>> >>> Perhaps to help clarify things. 2007/064 wishes to change the >>> type of uid_t and gid_t from int to uint and provide for ephemeral >>> IDs. The project team argued that, as specified, having ephemeral >>> IDs in the (negative) range [0x80000000:0xfffffffe] would not cause >>> existing standards or applications to fail. PSARC TCRed the case >>> to ensure that standard tests would not fail. >>> >>> Don's request is to ensure the TCR has the correct standards tests. >>> The TCR at present says: >>> >>> 1. Type changes for uid_t and gid_t may not be >>> integrated unless all of the official generic and >>> processor specific ABI conformance tests for the >>> System V ABI pass. The list of tests is: >>> >>> A. SPARC: >>> >>> A.1. SCD version 2.4.1. >>> >>> A.2. gABI version 2.1. >>> >>> A.3. SPARC psABI version 2.1. >>> >>> B x86: >>> >>> B.1. gABI version ***. >>> >>> B.2. x86 psABI version ***. >>> >>> C. x64: >>> >>> C.1. gABI version ***. >>> >>> C.2. x64 psABI version ***. >>> >>> If the tests all pass successfully or all pass >>> after any newly discovered bugs are fixed, then >>> the type changes are accepted and may integrate. >>> If an unresolvable test issue is discovered, the >>> project team will return to ARC to discuss an >>> appropriate resolution, including discussing the >>> relative merits of leaving the types alone (i.e., >>> negative values for ephemeral IDs) versus docu- >>> menting a difference with respect to the ABI docu- >>> ment (and thus conforming to the language of POSIX >>> with respect to non-negative values). >>> >>> >>> From Andy's reply, perhaps the x86 and x64 parts should be removed. >>> I presume then that no tests are run on x86 or x64 related to >>> any standards. >>> >>>Gary.. >>> >>>>Andy >>>> >>>> >>>>Don Cragun wrote: >>>> >>>>>Paul, Robert, & Tim, >>>>> I am trying to track down System V ABI test suites for x86 and >>>>>x64 hardware lines for use to verify that a project recently approved >>>>>by PSARC doesn't break the ABI. The PSARC case is 2007/064 (Unified >>>>>POSIX and Windows Credentials for Solaris). >>>>> Eric Luiggi's group has provided information for the SPARC >>>>>hardware test suites (SPARC Compliance Definition version 2.4, System V >>>>>gABI version 2.1, and System V SPARC psABI version 2.1). Eric's group >>>>>runs these tests as part of the conformance testing we run before we >>>>>ship each Solaris release. >>>>> Do any of you know: >>>>>1. who runs the corresponding System V gABI and psABI test suites for >>>>> x86 and x84 hardware, >>>>>2. where we can find copies of these test suites for x86 and x64 so >>>>> the project team can run these tests before they integrate into the >>>>> ONNV source base, and/or >>>>>3. what the correct names (including version numbers) are for the x86 >>>>> and x64 gABI and psABI test suites? >>>>> >>>>> Any help in providing answers to these questions or pointers to >>>>>people who may be able to answer these questions will be greatly >>>>>appreciated. >>>>> >>>>> Cheers, >>>>> Don > >