Itanium kernel alignment faults - VMS

This is a discussion on Itanium kernel alignment faults - VMS ; Greetings, How can we fix kernel mode alignment faults caused by user mode code passing unaligned arguments to system services? Actually, I know how to fix them, I just don't know how to locate and identify them. According to Guy ...

+ Reply to Thread
Results 1 to 7 of 7

Thread: Itanium kernel alignment faults

  1. Itanium kernel alignment faults

    Greetings,

    How can we fix kernel mode alignment faults caused by user mode code
    passing unaligned arguments to system services? Actually, I know how
    to fix them, I just don't know how to locate and identify them.
    According to Guy Peleg's OpenVMS Technical Journal article on
    alignment faults there are 3 tools for detecting alignment faults:
    MONITOR ALIGN, SDA FLT, and debugger SET BREAK/UNALIGN. Those are
    indeed wonderful tools and they have made it straight forward to
    identify and deal with user mode alignment faults.

    Having done that successfully we are now left with lots (tens of
    thousands per second) of kernel mode alignment faults according to
    MONITOR ALIGN. And SDA FLT shows they are coming from our application
    and are located in system address space. But the debugger SET BREAK/
    UNALIGN never breaks; apparently it only works on user mode faults. I
    tried using SET BREAK/UNALIGN/SYSTEM but that was no help. Once a
    line of code is identified as calling a system service which in turn
    suffers an alignment fault then we are confident it can be
    successfully dealt with. How can we identify where in user mode code,
    i.e. a traceback, is producing a kernel alignment fault?

    Any help or pointers greatly appreciated.
    Cheers, Bill :-)

  2. Re: Itanium kernel alignment faults


    "Bill Law" wrote in message
    news:8528cb9d-c37d-4832-b40d-c47fa7c6f6a6@z66g2000hsc.googlegroups.com...
    > Greetings,
    >
    > How can we fix kernel mode alignment faults caused by user mode code
    > passing unaligned arguments to system services? Actually, I know how
    > to fix them, I just don't know how to locate and identify them.
    > According to Guy Peleg's OpenVMS Technical Journal article on
    > alignment faults there are 3 tools for detecting alignment faults:
    > MONITOR ALIGN, SDA FLT, and debugger SET BREAK/UNALIGN. Those are
    > indeed wonderful tools and they have made it straight forward to
    > identify and deal with user mode alignment faults.
    >
    > Having done that successfully we are now left with lots (tens of
    > thousands per second) of kernel mode alignment faults according to
    > MONITOR ALIGN. And SDA FLT shows they are coming from our application
    > and are located in system address space. But the debugger SET BREAK/
    > UNALIGN never breaks; apparently it only works on user mode faults. I
    > tried using SET BREAK/UNALIGN/SYSTEM but that was no help. Once a
    > line of code is identified as calling a system service which in turn
    > suffers an alignment fault then we are confident it can be
    > successfully dealt with. How can we identify where in user mode code,
    > i.e. a traceback, is producing a kernel alignment fault?
    >


    FLT START TRACE/CALLER is what you are looking for.

    Will show you the location where alignment fault happened - say
    SYS$GETDVI and the call chain that lead you in there.



    > Any help or pointers greatly appreciated.
    > Cheers, Bill :-)



    ** Posted from http://www.teranews.com **

  3. Re: Itanium kernel alignment faults

    On Jun 4, 1:29 am, "Guy Peleg"
    wrote:
    > "Bill Law" wrote in message
    >
    > news:8528cb9d-c37d-4832-b40d-c47fa7c6f6a6@z66g2000hsc.googlegroups.com...
    >
    >
    > > How can we fix kernel mode alignment faults caused by user mode code
    > > passing unaligned arguments to system services? Actually, I know how
    > > to fix them, I just don't know how to locate and identify them.

    > <--snip-->
    >
    > FLT START TRACE/CALLER is what you are looking for.
    >
    > Will show you the location where alignment fault happened - say
    > SYS$GETDVI and the call chain that lead you in there.
    >
    > > Any help or pointers greatly appreciated.
    > > Cheers, Bill :-)

    >
    > ** Posted fromhttp://www.teranews.com**


    Greetings,

    Thank you for this pointer. It appears to be undocumented on our 8.3
    system, at least it is not in the online help, but I tried it anyway.
    It immediately crashed the system. Being brave, and/or stupid, I of
    course tried it again and it again immediately crashed the system.
    Our IT department has not yet had it analyzed, but I wanted to get a
    quick note in here as a warning to others who may be considering using
    this.

    Cheers, Bill :-)

  4. Re: Itanium kernel alignment faults

    Bill Law wrote:

    > On Jun 4, 1:29 am, "Guy Peleg"
    > wrote:


    >>
    >> FLT START TRACE/CALLER is what you are looking for.
    >>

    >
    > Greetings,
    >
    > Thank you for this pointer. It appears to be undocumented on our 8.3
    > system, at least it is not in the online help,


    See :
    http://h71000.www7.hp.com/doc/82FINA...a-rezte-te.pdf
    "HP OpenVMS System Analysis Tools Manual"

    Read chapter 7 "SDA Alignment Fault Utility (FLT)".

    It seems as you have to do a "FLT LOAD" before trying "FLT START...".

    Jan-Erik.


    > but I tried it anyway.
    > It immediately crashed the system. Being brave, and/or stupid, I of
    > course tried it again and it again immediately crashed the system.
    > Our IT department has not yet had it analyzed, but I wanted to get a
    > quick note in here as a warning to others who may be considering using
    > this.
    >
    > Cheers, Bill :-)


  5. Re: Itanium kernel alignment faults


    "Bill Law" wrote in message
    news:7aa2f4be-c124-468b-816c-7a27f2be7bf0@x41g2000hsb.googlegroups.com...
    > On Jun 4, 1:29 am, "Guy Peleg"
    > wrote:
    >> "Bill Law" wrote in message
    >>
    >> news:8528cb9d-c37d-4832-b40d-c47fa7c6f6a6@z66g2000hsc.googlegroups.com...
    >>
    >>
    >> > How can we fix kernel mode alignment faults caused by user mode code
    >> > passing unaligned arguments to system services? Actually, I know how
    >> > to fix them, I just don't know how to locate and identify them.

    >> <--snip-->
    >>
    >> FLT START TRACE/CALLER is what you are looking for.
    >>
    >> Will show you the location where alignment fault happened - say
    >> SYS$GETDVI and the call chain that lead you in there.
    >>
    >> > Any help or pointers greatly appreciated.
    >> > Cheers, Bill :-)

    >>
    >> ** Posted fromhttp://www.teranews.com**

    >
    > Greetings,
    >
    > Thank you for this pointer. It appears to be undocumented on our 8.3
    > system, at least it is not in the online help, but I tried it anyway.
    > It immediately crashed the system. Being brave, and/or stupid, I of
    > course tried it again and it again immediately crashed the system.


    This is part of the risk of using undocumented tools......

    If possible, email me the output of CLUE CRASH

    > Our IT department has not yet had it analyzed, but I wanted to get a
    > quick note in here as a warning to others who may be considering using
    > this.
    >
    > Cheers, Bill :-)



    ** Posted from http://www.teranews.com **

  6. Re: Itanium kernel alignment faults

    On Jun 5, 4:59 pm, "Guy Peleg"
    wrote:
    > > It immediately crashed the system. Being brave, and/or stupid, I of
    > > course tried it again and it again immediately crashed the system.

    >
    > This is part of the risk of using undocumented tools......
    >
    > If possible, email me the output of CLUE CRASH


    Thank you for being willing to look at this, we certainly appreciate
    it.
    However, email to the email address (with REMOVE_THIS removed)
    immediately bounces with "User unknown".

    Cheers, Bill :-)

  7. Re: Itanium kernel alignment faults


    "Bill Law" wrote in message
    news:01da198a-1be2-401e-a8af-51bdbb5e3aa3@b1g2000hsg.googlegroups.com...
    > On Jun 5, 4:59 pm, "Guy Peleg"
    > wrote:
    >> > It immediately crashed the system. Being brave, and/or stupid, I of
    >> > course tried it again and it again immediately crashed the system.

    >>
    >> This is part of the risk of using undocumented tools......
    >>
    >> If possible, email me the output of CLUE CRASH

    >
    > Thank you for being willing to look at this, we certainly appreciate
    > it.
    > However, email to the email address (with REMOVE_THIS removed)
    > immediately bounces with "User unknown".


    Should be fixed now....check the addres again.

    >
    > Cheers, Bill :-)



    ** Posted from http://www.teranews.com **

+ Reply to Thread