NEOHAPSIS - Peace of Mind Through Integrity and Insight

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com

Neohapsis > Archives > IBM AIX Digests> 2002-q1

From: AIX Service Mail Server (aixservaustin.ibm.com)
Date: Thu Mar 28 2002 - 02:32:26 CST

Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

APAR: IC33012 COMPID: 5724A6000 REL: 710
ABSTRACT: INCORRECT RESULT IS GENERATED WHEN A LINEAR REGRESSION MODEL

PROBLEM DESCRIPTION:
In Intelligent Miner Scoring a particular model was generating
different results than what the user was getting by using a SQL
statement. This problem occurs if a linear regression model
contains coefficients that have 0.0 as value.

PROBLEM SUMMARY:
Regression Models that include coefficients
with value zero give wrong result.

PROBLEM CONCLUSION:
Handling of coefficients with value zero
has been changed for Regression algorithm in IM Scoring code.

------

APAR: IY18738 COMPID: 5765D6100 REL: 210
ABSTRACT: LOADLEVELER NEGOTIATOR HAS MEMORY LEAK

PROBLEM DESCRIPTION:
LoadL_negotiator has a memory leak. A multi-step job with
dependencies will cause a memory leak if a step is skipped
because a dependency from an earlier step was not satisfied.

PROBLEM SUMMARY:
LoadL_negotiator has a memory leak that can cause it to run
out of memory if it is left running long enough between
restarts.

PROBLEM CONCLUSION:
The code was fixed to free memory more carefully and to
allow more memory to be used.

------

APAR: IY20321 COMPID: 5765D6100 REL: 220
ABSTRACT: ADD IN WHO CAN ISSUE FOR LL COMMAND MAN PAGES

PROBLEM DESCRIPTION:
Presently the LL commands man pages do not have "who can issue"
         info in them. The only place one can tell this is from
         the LL Using and Administrating Guide Chapter 9
         commands table.

PROBLEM SUMMARY:
All the LoadLeveler command man pages do not state who can
issue those commands

PROBLEM CONCLUSION:
All the LoadLeveler command man pages will now have a
section "Who Can Issue" to tell users who have authority to
issue those commands.

------

APAR: IY20699 COMPID: 5765D5100 REL: 311
ABSTRACT: SECURITY ISSUE

PROBLEM DESCRIPTION:
security issue

PROBLEM SUMMARY:
Security Issue

PROBLEM CONCLUSION:
Security Issue Resolved.

------

APAR: IY20797 COMPID: 5765E2600 REL: 500
ABSTRACT: SIZE_T CAN BE DEFINED TWICE: GIVING CONFLICTING DECLARATION ERR

PROBLEM DESCRIPTION:
When /usr/vacpp/include/stddef.h is included twice _SIZE_T
is undefined. This allows size_t to be incorrectly defined
twice.
The testcase below illustrates this problem.
/* test.cpp */
#include <stddef.h>
#include <stddef.h>
#ifndef _SIZE_T
#define _SIZE_T
typedef unsigned int size_t;
#endif
This compiler command will produce the following error:
$ xlC -c test.cpp
"test.cpp", line 5.22: 1540-0400 (S) "size_t" has a
conflicting declaration.
"/usr/include/stddef.h", line 67.25: 1540-0425 (I) "size_t"
is defined on line 67 of "/usr/include/stddef.h".

------

APAR: IY21157 COMPID: 5765E3200 REL: 500
ABSTRACT: #LINE PRODUCING INCORRECT OUTPUT IN DBX

PROBLEM DESCRIPTION:
Using the cc_r7 compiler, the filename that is displayed when
debugging with dbx is incorrectly displayed. The testcase is
compiled with the compile command in the PMR. In dbx, a
breakpoint is set with "stop in main" command. "run" command is
then executed. The filename that is displayed when program
stops in main shows as memPart2c.i instead of memPart2.c as
should be displayed. The testcase is explained in both the PMR
(21204,344,000) and defect 178773.

------

APAR: IY21189 COMPID: 5765D6100 REL: 220
ABSTRACT: LLSUMMARY TO BE ABLE TO READ MULTIPLE GLOBALHIST FILES

PROBLEM DESCRIPTION:
llsummary can not read multiple globalhistory files from a
given directory.

PROBLEM SUMMARY:
Only a single history file can be inputted
for llsummary to process.
New requirement is to be able to input more than
one history file for llsummary to process.

PROBLEM CONCLUSION:
llsummary now has the ability to generate a
report based on a list of history files.
The LoadLeveler document "Using and Administering,
under LoadLeveler Commands Chapter under llsummary" and
the man page for llsummary will be revised.

------

APAR: IY22052 COMPID: 5765E2600 REL: 502
ABSTRACT: INTERNAL COMPILER ERROR WITH -O

PROBLEM DESCRIPTION:
An "Internal compiler error" occurs when a testcase is compiled
with optimization (-O).
The following command illustrates the defect:
$ /usr/vacpp/bin/xlC -c -O test.C
/usr/vacpp/bin/xlC: 1501-230 Internal compiler error; please
contact your Service Representative

------

APAR: IY23192 COMPID: 5765E2600 REL: 500
ABSTRACT: EXTERN "C" LINKSPEC NOT APPLIED TO VARIABLE NAMES IN NAMESPACE

PROBLEM DESCRIPTION:
The extern "C" linkage specification is not being applied
to the names of variables that are declared inside of a
namespace.
Consider the following example:
/* globals.c */
int foo;
/* program.cpp */
int main()
{
  printf("foo: %d\n", adlib::foo);
  return 0;
}
To Build:
$ xlc -c globals.c
$ xlC globals.o program.cpp
ld: 0711-317 ERROR: Undefined symbol: adlib::foo
ld: 0711-345 Use the -bloadmap or -bnoquiet option to
obtain more information.
$ dump -tv program.o | grep foo
24 m 0x00000000 undef 1 extern
                    foo__5adlib
26 m 0x00000060 .data 1 unamex
                    foo__5adlib
.... shows that the extern "C" linkage specification was
not applied to the name of the variable foo
If the variable named foo is replaced with a function
named foo, the problem does not occur.
I've read through the section in the standard on linkage
specifications (7.5) and do not see anything that indicates
that what the customer is attempting is illegal.

PROBLEM SUMMARY:
Extern "C" names get mangled

PROBLEM CONCLUSION:
a new rule has been put in not to mangle extern "C" names

------

APAR: IY23516 COMPID: 5765E2600 REL: 502
ABSTRACT: SEGMENTATION FAULT WITH VACPP 502

PROBLEM DESCRIPTION:
segmentation fault during execution when compiled with vacpp
5.0.2 compiler. Segmentation fault doesn't occur when compiled
with 3.6.6 C++ compiler. Segmentation fault also doesn't occur
when the addactionlistener() function is removed from source.
The function is never called, but when removed the segmentation
fault does not occur.

PROBLEM CONCLUSION:
This defect stems from an attempt to initialize a
composition class in t.C's line 14: pfcBaseSession()
despite the fact that pfcSession_tk has no composition class
member.
I am not sure if this is supposed to work even though
pfcSession is a virtual base class. But the fact that this used
to work in 366 and works in IBM mode may be accidental.

------

APAR: IY24427 COMPID: 5765D6100 REL: 220
ABSTRACT: LOADL2.2 HISTORY/GLOBALHISTORY FILES 0 WRITE WHEN NFS IS FULL

PROBLEM DESCRIPTION:
When issueing llacctmrg command, If NFS mounted and File system
is Full, The GlobalHistory file will be a 0 lenght and the
History file in the Spool directory on the Schedd will be
erased to a 0 lenght. Data will be lost.

LOCAL FIX:
write locally to a JFS file system. If successful, mv to NFS
as previously desired.

PROBLEM SUMMARY:
When the NFS filesystem is full and llacctmrg command
isn't able to write its global history file there ,
the local history files in the spool directory will
go to zero length.

PROBLEM CONCLUSION:
When the NFS filesystem is full and llacctmrg command
isn't able to write its global history file there ,
the local history files in the spool directory will
not be affected.

------

APAR: IY24691 COMPID: 5765E2600 REL: 500
ABSTRACT: INTERNAL COMPILER ERROR ON NESTED PARTIAL SPECIALIZATION

PROBLEM DESCRIPTION:
Internal Compiler Error on nested partial specialization.
The testcase below fails to compile with -DBAD. The expected
output occurs when compiled with just xlC.
****TESTCASE****
struct N;
template <class T, class U>
struct L
{
    typedef T A;
    typedef U Z;
};
template <class, template <class, class> class>
struct M;
template <class T, template <class, class> class P>
struct M<L<T, N>, P> { typedef T Y; };
template <class T, class U, template <class, class> class P>
struct M<L<T, U>, P>
{
    typedef typename P<T, typename M<U, P>::Y>::Y Y;
};
template <class, class> struct R;
template <class T> struct R<L<T, N>, T> { typedef N Y; };
template <class T, class U> struct R<L<T, U>, T> { typedef U Y;
};
template <class T, class U, class V>
struct R<L<T, U>, V>
{
    typedef L<T, typename R<U, V>::Y> Y;
};
template <class T, template <class, class> class P>
struct S;
template <class T, template <class, class> class P>
struct S<L<T, N>, P> { typedef L<T, N> Y; };
template <class T, class U, template <class, class> class P>
struct S<L<T, U>, P>
{
    typedef typename M<L<T, U>, P>::Y X;
    typedef L<X, typename S<typename R<L<T, U>, X>::Y, P>::Y> Y;
};
#ifndef BAD
template <class T, class U, bool>
struct select { typedef U Y; };
template <class T, class U>
struct select<T, U, 1> { typedef T Y; };
#endif
template <class T, class U>
struct less_than
{
#ifdef BAD
    template <class, bool> struct select { typedef U Y; };
    template <class V> struct select<V, 1> { typedef T Y; };
#endif
    typedef typename select<T, U, sizeof (T) < sizeof (U)>::Y Y;
};
template <class> void foo ();
int main ()
{
    typedef
L<int, L<char, L<long double, L<short, N> > > > L;
    foo<S<L, less_than>::Y>();
}

PROBLEM CONCLUSION:
A compiler problem that resulted in
the internal compiler error was located and fixed.

------

APAR: IY24741 COMPID: 5765E2600 REL: 500
ABSTRACT: INCORRECT OUTPUT MSG WHEN USING NESTED PARTIAL SPECIALIZATION

PROBLEM DESCRIPTION:
Incorrect output msg when using nested partial specialization.
****TESTCASE*****
struct N;
template <class T, class U>
struct L
{
    typedef T A;
    typedef U Z;
};
template <class T, template <class, class> class P>
struct S;
template <class T, template <class, class> class P>
struct S<L<T, N>, P> { typedef L<T, N> Y; };
template <class T, class U, template <class, class> class P>
struct S<L<T, U>, P>
{
    template <class> struct M;
    template <class V> struct M<L<V, N> > { typedef V Y; };
    template <class V, class W>
    struct M<L<V, W> > { typedef typename P<V, typename
M<W>::Y>::Y Y; };
    template <class, class> struct R;
    template <class V> struct R<L<V, N>, V> { typedef N Y; };
    template <class V, class W> struct R<L<V, W>, V> { typedef W
Y; };
    template <class V, class W, class X>
    struct R<L<V, W>, X> { typedef L<V, typename R<W, X>::Y> Y;
};

------

APAR: IY24878 COMPID: 5765D6100 REL: 220
ABSTRACT: THE SENDER OF EMAILS ISSUED AT END OF JOBS IS NOT ALWAYS LOADL

PROBLEM DESCRIPTION:
With the migration from LL 2.1 to 2.2 email send at
the end of jobs are now send by the user ID which was
used to start LL and not, as it was in LL 2.1 always
from the loadl user.
This causes some automatic mail handling to fail (not
work as expected) on customer system. LL 2.2 needs to
be change so all email gets send by loadl user.

LOCAL FIX:
su - loadl, then start LL so the environment of loadl is used

PROBLEM SUMMARY:
The sender of emails for jobs are not always loadl.
It depends on who starts up loadl.

PROBLEM CONCLUSION:
The sender of emails for jobs will now always be loadl.

------

APAR: IY25105 COMPID: 5765E8200 REL: 230
ABSTRACT: HAGEO: GMD IS NOT ACCESSIBLE FROM 64-BITS USER SPACE

PROBLEM DESCRIPTION:
gmds not accessible from 64-bit user space

PROBLEM CONCLUSION:
Support for 64-bit applications has been added to the
GeoMirror device driver.

------

APAR: IY25323 COMPID: 5765E7200 REL: 310
ABSTRACT: NET COMMAND FIXES

PROBLEM DESCRIPTION:
-net config command can't be used to change the parameter
mmapfiles value
-net user /delete command has extra space in the output
-net config help does not show information about cache_searches
-net config parameter creates cifsConfig.new file, and this
file is not removed eventually

PROBLEM CONCLUSION:
-Adding PARM_KEY_MMAPFILES
-Removing the space from the comment
-Adding information to the net config help information
-Using system command to remove the cifsConfig.new file

------

APAR: IY25368 COMPID: 5765E2600 REL: 502
ABSTRACT: ERROR AFTER ADDING THROW() TO NON-TYPE TEMPLATE FUNCTION DECL.

PROBLEM DESCRIPTION:
After adding an exception specification throw() to template
function declaration the testcase won't not compile.The compiler
returns the following errors:
$ xlC test.cpp
"test.cpp", line 14.22: 1540-0257 (S) An object or reference of
type "A<0> (*)() cannot be initialized with an expression of
type "A<N> () throw()".
"test.cpp", line 14.20: 1540-1278 (I) The deduced type
"A<0> () throw()" does not match the specialized type "A<0> ()".
Without throw() the testcase compiles fine. ISO 14882:1998 does
not suggest any limitations to adding throw() to template
function declaration.
Testcase:
============
// compile: /usr/vacpp/bin/xlC test.cpp
///////////////////////////////////////
template <int> class A{
    // empty body
};
// Template function foo
template <int N> A<N> foo() throw(){
    return A<0>();
}
// Declare a ptr to function foo()
int main (){
    A<0>(*p_foo)() = &foo;
}

------

APAR: IY25467 COMPID: 5765E7200 REL: 310
ABSTRACT: NEED TO REDUCE THE DCE CREDENTIAL FILES

PROBLEM DESCRIPTION:
Fast Connect leaves lot of dce credential files in
/var/dce/security/creds directory with dce_auth=1

PROBLEM CONCLUSION:
logout in cifsUserProc reduces some dce credential files.

------

APAR: IY25487 COMPID: 5765E8200 REL: 230
ABSTRACT: HAGEO: NETWORK_DOWN RESULTS IN CLSTOP OF ONLY NODE UP IN CLUSTER

PROBLEM DESCRIPTION:
Customer experienced two situation in which the non-dominant
site went down due to network_down event with HAGEO running
on HAES. If bringing up the non-dominant site first and when
issuing halt -q on primary site node, a network_down event
following the node_up or node_down takeover, the
geo_network_down brought the node down through geo_node_down.

PROBLEM SUMMARY:
Customer experienced two situation in which the non-dominant
site went down due to network_down event with HAGEO running
on HAES. If bringing up the non-dominant site first and when
issuing halt -q on primary site node, a network_down event
following the node_up or node_down takeover, the
geo_network_down brought the node down through geo_node_down.
These network events were for non-geo networks and should not
have resulted in an actions by hageo.

PROBLEM CONCLUSION:
Add a test in the network_ scripts to check for a geo network.

------

APAR: IY25499 COMPID: 5765E2600 REL: 502
ABSTRACT: CMP ERR: CAN'T TYPEDEF NESTED CLASS FROM ANOTHER CLASS (HITACHI)

PROBLEM DESCRIPTION:
Compiler: VAC++ 5 (vacpp.cmp.batch 5.0.2.1)
Platform: AIX 5.1
PMR: 22991,379,000
Testcase:
=============
// Test.C
class CFoo{
  public:
    class CNested{
      public:
        static int i;
    };
};
class CBar{
  public:
    typedef CFoo::CNested CNested;
};
int CFoo::CNested::i = 0;
int main(){
    CBar::CNested::i = 1;
    return 0;
}
$ xlC test3.C
"test3.C", line 18.5: 1540-0300 (S) The "private" member "class
cannot be accessed.
The testcase should compile fine according to ISO 14882,
however the compilation fails with the error message above.
typedef should be able to define a sinonym any class name.

------

APAR: IY25502 COMPID: 5765E2600 REL: 502
ABSTRACT: SEGV FAULT AFTER FUNCTION RETURNS TO TRY BLOCK

PROBLEM DESCRIPTION:
The customer is experiencing some difficulty debugging crashes
that occur in the very elaborate exception handling in a fairly
complex C++ application. The customer sent the testcase that
demonstrates at least the part of the problem. The original
issue was actually a branch to location zero out of ptrgl during
throw processing, but at some point in the process of creating
the testcase, it turned into the segmentation violation. The
testcase is very volatile to changes especially when built with
-g (debug) option. Removing seemingly insignificant lines of cod
and unnecessary catch blocks may cause the problem not to
reproduce. The problem reproduces under very specific set of
generated instructions. Also, an empty core file is each time.

------

APAR: IY25701 COMPID: 5765D6100 REL: 220
ABSTRACT: SORT BY DISPATCH ORDER DOES NOT WORK IN LOADLEVELER 2.2

PROBLEM DESCRIPTION:
the sort of jobs by dispatch order in xloadl will not reflect
the actual dispatch order that will be used by the
scheduler.

PROBLEM SUMMARY:
The LoadLeveler xloadl gui isn't ordering the favor jobs
correctly when the sort by dispatch primary is set.

PROBLEM CONCLUSION:
The LoadLeveler xloadl gui when the sort by dispatch primary
is set will now order the favor jobs at the top of the list
sorted by q_sysprio value while the unfavorjobs will be
sorted below by q_sysprio value.

------

APAR: IY26162 COMPID: 5765E6800 REL: 300
ABSTRACT: SPMIARMD HOGS CPU

PROBLEM DESCRIPTION:
The SpmiArmd daemon which is used while measuring the
Application Response Times, uses a very high percentage of
the CPU (> 90%).Sometimes this may even cause the system
to hang.

PROBLEM CONCLUSION:
Fixed the infinite loop which was causing the SpmiArmd
to occupy very high CPU percentage

------

APAR: IY26226 COMPID: 5765E6800 REL: 300
ABSTRACT: FILTD DIES DUE TO LESSER TIMEOUT INTERVAL

PROBLEM DESCRIPTION:
The filtd daemon will terminate itself if it has not received
data_feed packets from xmservd for 10 times the data feed
interval.So if the network load is high on a system then
this timeout period may not be enough which may cause
filtd to die.

PROBLEM SUMMARY:
The filtd daemon will terminate itself if it has not received
data_feed packets from xmservd for 10 times the data feed
interval.So if the network load is high on a system then
this timeout period may not be enough which may cause
filtd to die often.

PROBLEM CONCLUSION:
Introduced a command line option '-t' to filtd so that a
suitable higher timeout value can be specified by the user.

------

APAR: IY26315 COMPID: 5765E7200 REL: 310
ABSTRACT: ALLOW 4096 PRINT SHARES (PRINTQUEUEINFO APIS)

PROBLEM DESCRIPTION:
PC-client can hang if customer uses more than 64 PrintShares
Problem typically appears when PC-client is adding a new
Windows printer-device, esp. when printing a Test Page.

PROBLEM CONCLUSION:
Increase internal PrintQueue limit from 64 to 4096.

------

APAR: IY26389 COMPID: 5765E6800 REL: 300
ABSTRACT: AGENT'S LIBRARY DOESN'T SEND CORRECT CPU VALUES TO SNMP

PROBLEM DESCRIPTION:
The values for CPU stats supplied by libSpmi to snmp daemon
are incorrect or zero.

PROBLEM CONCLUSION:
modified the code of Agent's shared library to
retrive & calculate the CPU stats correctly

------

APAR: IY26444 COMPID: 5765E2600 REL: 502
ABSTRACT: V502:TYPE SUBSTITUTION IN TEMPLATES IS NOT WORKING RIGHT.

PROBLEM DESCRIPTION:
testcase
template <class T>
class Map {
   T value;
public:
   T find(const char *key) const { return value; }
};
class Type {
public:
   class I {
   };
   class N {
   public:
      class I {
      };
   };
};
class Object {
public:
/*
* comment class I out it works
*
*/
   class I {
   public:
      virtual void op1();
   private:
      typedef bool (Type::I::*FUNC)();
      typedef Map<FUNC> FMap;
      static FMap FMap_op1;
   };
   class N {
   public:
      class I {
      public:
         virtual void op2();
      private:
         typedef bool (Type::N::I::*FUNC)();
         typedef Map<FUNC> FMap;
         static FMap FMap_op2;
      };
   };
};
void Object::N::I::op2() {
   FUNC p = FMap_op2.find("");
}
/*
*$ xlC test2.C
*"test2.C", line 48.26: 1540-0257 (S) An object
or reference of type "bool (I::*)()" cannot be
initialized with an expression of type "bool (I::*)()".
*/
What is happening here is that the return type of "find"
is "Type::N::I::*FUNC" when the template of class
"Map<Type::N::I::*FUNC>" and "Object::N::I" is instantiated
also "find" can have the return type of "Type::I::*FUNC"
when the template class "Map <Type::I::*FUNC>" "Object::I"
is instantiated.
The compiler is getting confused as to which return type to
use since (according to the error message) it seems the
compiler is not keeping enough information about the return
type.
Note:
  This also fails in dev.

PROBLEM SUMMARY:
Compiler gets confused when two classes have the same name for
a sub-class with in turn also has the same name for sub-sub
class.

------

APAR: IY26479 COMPID: 5765E7200 REL: 310
ABSTRACT: ACCESS LDB FILE IS CORRUPTED BY CIFS

PROBLEM DESCRIPTION:
User has to access a linked database when another user is
already connected to the database from another machine.Then
the first user will get error that he/she can't access
the database. The .ldb file will get corrupted.

PROBLEM CONCLUSION:
Checking if the .ldb file has any locks before trying to
delete that file

------

APAR: IY26521 COMPID: 5765D5101 REL: 111
ABSTRACT: APPARENT RESOURCE STARVATION DUE TO LOADED SWITCH. A CPU MAY BE

PROBLEM DESCRIPTION:
A node may appear to be suffering from resource starvation. Some
of the hatsd threads may have trouble running. This may result
in other symptoms, such as vsdd (ie Virtual Shared Disk )
daemon resets (ie /usr/lpp/csd/bin/rvsdd vsd will restart ),
but the problems may only happen infrequently.
If dumps with a trace running are forced at the time of the
problem, it can be seen that several threads (including the
hatsd) are waiting for a CPU to be ready. The real problem is
the if_ls code appears to be looping at interrupt level 8, as
can be seen from the trace, so other thread's on the same cpu's
run q, like hatsd, don't get a chance to run.
The switch was under a load and was processing the if_snd queue
in the off-level handler. The queue was almost full at the
time of the crash and the SLIH would have continued to run
off-level until the queue was drained. It's possible that
runnable threads with an affinity with a given run queue might
become blocked for seconds.

LOCAL FIX:
If hats needs to ensure that it's dispatched within a reasonable
period of time then it should export the environment variable
RTG_GRQ=ON (this stands for real-time global run queue) into
their environment to ensure that they are always eligible for
dispatch on a free CPU.
As an EFIX for this problem, the customer could run the
following command to ensure that all fixed priority threads
(such as hats and GIL) behave as though RT_GRQ=ON were set in
their environment:
/usr/samples/kernel/schedtune -F 1

PROBLEM SUMMARY:
Under rare circumstances, it is possible for the hats thread
to be blocked on the CPU where it is waiting to run, even
though other CPUs may be idle. In this situation, hats
exits after issuing the following message:
"Got a non-stale Proclaim message from my NS. He must have
deleted me, so I'm exiting to comply."

PROBLEM CONCLUSION:
By setting the environmental variable RT_GRQ=ON prior to
starting the hats daemon, the hats thread will be placed on
the global run queue. This will prevent the hats thread
from being blocked when CPUs are available. This will
prevent the node from appearing to be down.

------

APAR: IY26559 COMPID: 5765D5100 REL: 320
ABSTRACT: SPMON QUERY OR MONITORING CAUSES SEGMENTATION FAULT (COREDUMP)

PROBLEM DESCRIPTION:
When a customer running PSSP-3.2 executes a spmon query and/or
monitoring process, spmon generates a segmentation fault and
core dumps. The stack trace seems to core dump right at a call
to SDRFreeCh.

PROBLEM SUMMARY:
Issuing spmon to query the host responds value
of a sepcific node may result in a Segmentation fault. The
code in spmon which is used to determine the host responds
value for a node was always freeing memory, regardless of
whether the memory had been allocated in the specific code path.

PROBLEM CONCLUSION:
spmonn has been updated to only free a
certain block of memory if the block has been allocated.
This will prevent the segmentation fault from occurring when
querying the host responds value for a specific node.

------

APAR: IY26609 COMPID: 5765E7200 REL: 310
ABSTRACT: OVERWRITE A FILE WITH SAME NAME CHAGES PERMISSIONS

PROBLEM DESCRIPTION:
Overwrite a file with same file with same permissions from
another directory changes the permissions of the file.

PROBLEM CONCLUSION:
Fixed the code to keep the permissions while overwriting a
file.

------

APAR: IY26669 COMPID: 5765E2600 REL: 502
ABSTRACT: V502:INTERNAL ERROR: ASSERTION FAILURE IN "TRANSFORMERIMPL.C"

PROBLEM DESCRIPTION:
abstract
V502:Internal error: Assertion failure in "TransformerImpl.C",
Reason: Assertion failure in
"/nfs/kermit/home/nasgaard/CodeStore/kernel
/transformer/TransformerImpl.C", line 111
Callback Stack:
    CS_TraceBackUtilities::saveTraceBack
(unsigned int**,int,__sigcontext*) + 0x2c
    CS_TraceBackInformation::
CS_TraceBackInformation(void*) + 0x44
    CS_UnRecoverableError::
CS_UnRecoverableError() + 0x34
    CS_AssumeFailure::
CS_AssumeFailure(const char*,unsigned int) + 0x5c
    CS_AssumeFailure::fail(const char*,unsigned int) + 0x104
    CS_TransformerImpl::typeAnalyze(CS_Expression&) + 0x154
    CS_TransformerImpl::typeAnalyzeAndTransform
(CS_Expression&) + 0x2c
    CS_TransformerImpl::transformPostIncrementExpression
(CS_Expression&,CS_BinaryExpression::
Kind,CS_TokenLocation&) + 0x364
    CS_TransformerImpl::transformPostIncrementExpression
(CS_UnaryExpression&,CS_BinaryExpression
::Kind) + 0x15c
    CS_TransformerImpl::transform
(CS_UnaryExpression&) + 0x84
    CS_TransformerImpl::transform
(CS_Expression&) + 0x130
    CS_TransformerImpl::transformRoot
(CS_Expression&) + 0x18c
    CS_TransformerImpl::transform
(CS_ExpressionStatement&) + 0x38
    CS_TransformerImpl::transform
(CS_Statement&) + 0x160
    CS_TransformerImpl::transformStatements
(CS_LexicalBlockStatement&) + 0x34
    CS_TransformerImpl::transform
(CS_LexicalBlockStatement&) + 0x308
    CS_TransformerImpl::transform
(CS_Statement&) + 0x19c
    CS_TransformerImpl::transformStatement
(CS_Statement*&,int) + 0xc8
    CS_TransformerImpl::transform
(CS_Statement*&,int) + 0x68
    CS_FunctionBodyImpl::transform
(CS_Statement*&,CS_Declaration&,int,
CS_IncorporationState&) + 0x6c
    CS_FunctionBodyImpl::build
(CS_Statement*&,CS_Implementation::
TreeKind,int,int,CS_IncorporationState&) + 0x1220
    CS_FunctionBodyImpl::update
(CS_IncorporationState&,int) + 0x6e0
    CS_DependencyNode::processInvalidatedNode
(CS_IncorporationState&,int,int) + 0x120
    CS_Incorporator::process(int) + 0xab4
    CS_Incorporator::incorporate() + 0x4a0
    CS_DependencyGraph::incorporate
(CS_IncorporationController&,int,CS_Incorporator*) + 0x390
    CS_CodeStoreImpl::incorporate
(CS_IncorporationController&,int,
CS_Incorporator*,CS_EvaluatedProjectFileEnvironment*) + 0xd4
    main + 0x750
    __start + 0x75f37
/.../torolab.ibm.com/fs/projects/vabld/run/tuscany/502/aix/
daily/011222/bin/.orig/xlC: 1501-230
Internal compiler error; please contact you
r Service Representative
make: 1254-004 The error code from the last command is 41.

------

APAR: IY26679 COMPID: 5765D6100 REL: 220
ABSTRACT: REGATTA:LOADL JOB HUNG IN REMOVE PENDING RP

PROBLEM DESCRIPTION:
regatta:loadl job hung inremove pending rp

PROBLEM SUMMARY:
LoadLeveler job hung in remove pending RP

PROBLEM CONCLUSION:
Logic in schedd for determining when to retry trasmitting
remove order to startd is incorrect. It only increments
the retry count if the status of the job on the target
startd is already terminated. It should not retry in this
case but should increment and test the retry count in all
other cases.

------

APAR: IY26755 COMPID: 5765D5100 REL: 320
ABSTRACT: RVSD NOT FAILOVER IN A TWO NODE CLUSTER WHEN NODE FAILURE OCCURS

PROBLEM DESCRIPTION:
problem with rvsd recovery in a two node sp when one of the
nodes is powered off/failed. rvsd expected (and up until now
got) a certain order in notification from hats/hags about this
powered off/failed node. due to changes in hats/hags this
order can no longer be guaranteed.

PROBLEM SUMMARY:
RVSD recovery fails in a 2-node SP system when one of the
nodes is powered off or fails.

PROBLEM CONCLUSION:
VSD 2-node recovery was based on a particular ordering of
adapter and node failure event notification. It has been
determined that this ordering is not quaranteed so the
adapter recovery algorithm must be modified for this
special 2-node case.

------

APAR: IY26784 COMPID: 5765D9300 REL: 310
ABSTRACT: POE STATIC LINK COMMANDS DON'T WORK

PROBLEM DESCRIPTION:
The sample ld commands for creating statically linked POE
programs do not work. They result in an undefined symbol for
mt__trce.

LOCAL FIX:
Creating dynamically linked executables is the recommended
procedure. If you wanted static links so that you would not be
impacted by a change to a library, then the recommended
procedure would be to start with dynamic links. Then, before a
a library change is made, make a copy of all the used libraries
into a new directory. Next, use environment variables LIBPATH
to point to that new directory. That way, the executable would
find the preserved copies of the libraries instead of the
updated copies.

PROBLEM SUMMARY:
The example ld commands for creating static executables in
the Parallel Operating Environment Manual don't work.

PROBLEM CONCLUSION:
The poe.README file that gets installed in
/usr/lpp/ppe.poe/READMES has been updated with example ld
commands that can be used.

------

APAR: IY26828 COMPID: 5765E8200 REL: 230
ABSTRACT: HAGEO: NETWORK_DOWN RESULTS IN CLSTOP OF ONLY NODE UP IN CLUSTER

PROBLEM CONCLUSION:
Add a test in the network_ scripts to check for a geo network.

------

APAR: IY26829 COMPID: 5765E8200 REL: 230
ABSTRACT: GEO_PRESTART_GMDS DOES NOT USE CORRECT CLINFO SRC WITH HAES

PROBLEM DESCRIPTION:
When running HAES clinfo is known as "clinfoES"
to the srcmstr, however Geo_prestart_gmds attempts
to start or refresh "clinfo".

PROBLEM CONCLUSION:
Start clinfo or clinfoES based on HACMP product.

------

APAR: IY26866 COMPID: 5765C3403 REL: 430
ABSTRACT: NFSO AND HACMP CONFLICT CAUSES CONFIG_TOO_LONG ERROR

PROBLEM DESCRIPTION:
config_too_long error caused by nfso and HACMP conflict.

PROBLEM SUMMARY:
Set remote host value to 127.0.0.1 in sm_twin_reg function
if remote host value is NULL.

PROBLEM CONCLUSION:
Set remote host value to 127.0.0.1 in sm_twin_reg function
if remote host value is NULL.

------

APAR: IY26934 COMPID: 5765D5100 REL: 320
ABSTRACT: SWITCH TABLE FORCE UNLOAD FAILS RC=1 ON SP-SWITCH

PROBLEM DESCRIPTION:
Force unload of a user space job may fail unloading the switch
table. The st.log will show the messages:
swtbl_clean_table: UNLOAD_ST_LL: Request for unload for
window X was unsuccessful. Errno=1( EPERM )
This will occur on the SP-Switch only.

PROBLEM SUMMARY:
his problem occurs on the SP-Switch only, during the force
unload of user space jobs the fault service kernel extension
was not cleaning up all resources correctly.

PROBLEM CONCLUSION:
If DRAIN_ON_SWITCH_TABLE_ERROR is not set, then when a
switch table error occurs, the default action is to attempt
to force unload the job. The switch table will now be
properly cleaned up and subsequent jobs will be allowed to
run on the node.

TEMPORARY FIX:
To prevent the node from being allocated to other jobs after
switch table error occurs, you can set:
DRAIN_ON_SWITCH_TABLE_ERROR=TRUE
in the LoadL config file.

------

APAR: IY26961 COMPID: 5765E7200 REL: 310
ABSTRACT: KERBEROS LOSES CONNECTIONS.

PROBLEM DESCRIPTION:
Intermitent disconnections from Kerberos w2k client to the
server.

PROBLEM CONCLUSION:
fixed session setup code to handle the problem

------

APAR: IY26977 COMPID: 5765D6100 REL: 220
ABSTRACT: LOADLEVELER SCHEDD CRASHES WHEN REJECTING JOBS

PROBLEM DESCRIPTION:
The loadleveler schedd crashes when rejecting jobs (job steps)

PROBLEM SUMMARY:
The LoadLeveler Schedd daemon has a small window
where it can core dumps with Illegal instruction when
rejecting jobs.

PROBLEM CONCLUSION:
The LoadLeveler Schedd daemon will not coredump when
rejecting jobs.

------

APAR: IY27041 COMPID: 5765E7200 REL: 310
ABSTRACT: NET CONFIG NEEDS TO CHECK PARAMETERS' DATA

PROBLEM DESCRIPTION:
They might be able to create names that are not usable.

PROBLEM CONCLUSION:
added a new function that checks for basic characters then
calls the old more liberal function.

------

APAR: IY27066 COMPID: 5765E6900 REL: 310
ABSTRACT: NEED WALL-CLOCK TIMER FOR GANG-SCHEDULING

PROBLEM DESCRIPTION:
Desire wall-clock time (to be provided in Lyndhurst release) in
time-sharing environment. Currently, users would get incorrect
timer values (using standard AIX system calls) in time-sharing
mode.
The customer would like an efix as soon as it is available.

PROBLEM SUMMARY:
Desire wall-clock timer in time-sharing environment.
Currently, users would get incorrect timer value (using
standard AIX system calls) in time-sharing mode.

PROBLEM CONCLUSION:
A new wall clock api and a sample use program will be
provided to allow the users to get accurate wall clock
information on their jobs when they are run in a
time-sharing environment.

------

APAR: IY27143 COMPID: 5765E7200 REL: 310
ABSTRACT: CLIENTS CANN'T CONNECT TO THE SERVER.

PROBLEM DESCRIPTION:
clients cann't connect to the server if the socket is closed
and its session is not removed from the session table.

PROBLEM CONCLUSION:
detect and remove the session from the session table if
its socket is closed (TCP/IP connection between server/client
is dropped)

------

APAR: IY27187 COMPID: 5765E5400 REL: 440
ABSTRACT: BRING RESOURCE GROUP ONLINE/OFFLINE ACTIVATES OTHER RESOURCE

PROBLEM DESCRIPTION:
The customer had only one node up in the cluster and wanted to
bring online a resource group that normally belonged to the
other node (home node). When running the command to bring
this resource group online, all of the other inactive resource
groups came online but not the one requested. Customer was
running HAES.

PROBLEM SUMMARY:
The customer had only one node up in the cluster and wanted to
bring online a resource group that normally belonged to the
other node (home node). When running the command to bring
this resource group online, all of the other inactive resource
groups came online but not the one requested. Customer was
running HAES.

PROBLEM CONCLUSION:
The reconfig_resource scripts were modified to recognize
when only one resource group should be acted upon. In
addition one change was required in rgpa_utils portion of
library libclres in HAES.

------

APAR: IY27213 COMPID: 5765E7200 REL: 310
ABSTRACT: UNABLE TO BROWSE FC IN NETWORK NEIGHORHOOD

PROBLEM DESCRIPTION:
cann't see FC server on network neighborhood

PROBLEM CONCLUSION:
validate the announcing interval on LANMAN

------

APAR: IY27305 COMPID: 5765D5100 REL: 320
ABSTRACT: CLEANUP.LOGS.NODES TRIMS ALL FILES DEFINED IN /ETC/SYSLOG.CONF.

PROBLEM DESCRIPTION:
When cleanup.logs.nodes calls psyslclr, it commands it to trim
all files defined in the /etc/syslog.conf file. However, in the
comment from the /usr/lpp/ssp/bin/cleanup.logs.nodes script, it
clearly states that the code will "Delete entries from the
syslog daemon facility log file that are older than six days
old. (The SP installs this as /var/adm/SPlogs/SPdaemon.log.)"
The program is not doing what was intended.

PROBLEM SUMMARY:
cleanup.logs.nodes and cleanup.logs.ws were trimming
messages older than six days from all files listed in
/etc/syslog.conf. They should just be trimming the
messages that are older than six days from
/var/adm/SPlogs/SPdaemon.log.

PROBLEM CONCLUSION:
cleanup.logs.nodes and cleanup.logs.ws were modified to
only trim messages that are older than six days from
/var/adm/SPlogs/SPdaemon.log.

------

APAR: IY27310 COMPID: 5765E7200 REL: 310
ABSTRACT: FC CORE DUMP - BUFFER OVERFLOW / ILLEGAL INSTRUCTION

PROBLEM DESCRIPTION:
intermittent core dump in session setup for the client

PROBLEM CONCLUSION:
fixing small (a few bytes) buffer overflow

------

APAR: IY27430 COMPID: 5765E6900 REL: 310
ABSTRACT: INFORMATIVE EMAIL TO USER SCHEDD/START COMMUNICATION ERROR

PROBLEM DESCRIPTION:
Need to add more/better EMail information to the user regarding
which STartd machine the Schedd could not contact/Communication
error.

PROBLEM SUMMARY:
Added name of startd machine to the e-mail to user when a
communication error occurs between schedd and startd, so
that the communication/networking error might be better
understood if and when it occurs. If such an error occurs,
jobs that are submitted do not run, because of the
communication error.

PROBLEM CONCLUSION:
A rare communication error occurred between schedd and one
of the startd machines, causing jobs not to run. An
enhancement was made to add the name of the startd machine
which schedd could not communicate with to the e-mail sent
to the user. The problem itself was not recreateable.

------

APAR: IY27431 COMPID: 5765E6900 REL: 310
ABSTRACT: CONFIGURABLE SIGKILL FOR STARTD TO STARTER TO PROCESSES

PROBLEM DESCRIPTION:
Need to be allowed to have configurable sigkill commands for
Startd to Starter to Processes

PROBLEM SUMMARY:
When jobs are cancelled, they will not remain in RP for two
minutes due to interference by a context switch running
Gang.

PROBLEM CONCLUSION:
Changed code to avoid doing context switch when job is
cancelled. Context switch was causing job to stay in RP for
two minutes before being cancelled.

------

APAR: IY27432 COMPID: 5765E6900 REL: 310
ABSTRACT: CHANGES TO BE MADE FOR STARTD AREA AND LISWITCHADAPTER

PROBLEM DESCRIPTION:
Changes in L1Adapter to correspond to CSS/HAL fixes

PROBLEM SUMMARY:
Unable to time share jobs

PROBLEM CONCLUSION:
The necessary LoadLeveler changes will be made to switch
window
handling to work with the device driver and switch changes.

------

APAR: IY27461 COMPID: 5765D5100 REL: 320
ABSTRACT: CSS.SNAP MAY NOT CONTAIN CADD_DUMP.OUT

PROBLEM DESCRIPTION:
The css.snap script may not collect the cadd_dump.out file,
which can be helpful in debugging certain SP switch 2 problems.

PROBLEM SUMMARY:
In some cases in which the css.snap script is run
to collect data for an SP Switch 2 problem, the
trace from the device driver (cad_dump.out) will
not be collected.

PROBLEM CONCLUSION:
The css.snap script has been changed to ensure, for the
SP Switch 2, the device driver trace (cadd_dump.out) is
always collected when the css.snap command is run.

------

APAR: IY27463 COMPID: 5765C3403 REL: 430
ABSTRACT: HANG IN ADDREPAGE

PROBLEM DESCRIPTION:
The system would hang, looping in addrepage.

PROBLEM SUMMARY:
Fixed a case where addrepage could be called for working
storage pages when the repage table hash fields are zero.

PROBLEM CONCLUSION:
Fixed a case where addrepage could be called for working
storage pages when the repage table hash fields are zero.

------

APAR: IY27581 COMPID: 5765B9501 REL: 330
ABSTRACT: WAIT4RVSD PARAMETER NOT DOCUMENTED SUFFICIENT

PROBLEM DESCRIPTION:
the parameter wait4RVSD is not documented very well.
it appears quite often in sample outputs in the docs,
always set to "no" in the samples.
however, in some cases setting it to "yes" is recommended,
since otherwise it could lead to a filesystem being not
mountable, if one out of many nodes is missing.

PROBLEM SUMMARY:
The mmlsconfig command displays configuration variables that
are for problem determination and are therefore not detailed
in either the mmconfig or mmchconfig commands.

PROBLEM CONCLUSION:
For the mmlsconfig description on the man page and in the
Administration and Programming Reference was changed from
"Depending on your configuration, additional information may
be displayed for problem determination" to "Depending on
your configuration, additional information which is set by
GPFS may be displayed to assist in problem determination
when contacting the IBM Support Center.

------

APAR: IY27608 COMPID: 5765E6900 REL: 310
ABSTRACT: BAD LLQ PERFORMANCE W/ 6000+ JOBS IN QUEUE

PROBLEM DESCRIPTION:
llq is takeing close to a minute to run when it is run alone
with around 6000 jobs in the queue. It is not uncommon for
llq's to timeout when a couple invocations are made in a short
space of time.

PROBLEM SUMMARY:
LoadL 2.2 takes approximately 3 times as long to produce llq
output
when there are 6000 jobs in the queue as LoadL 1.3 did.

PROBLEM CONCLUSION:
LoadL 2.2 needs a fast path in the job query code for plain
llq's
that will skip data manipulation which is only needed for
more
detailed output. This fix in itself will not make the
performance
equal to LoadL 1.3, but will make the performance resonably
close.
Note1: because of a major design difference between 1.3 and
2.2,
  the 2.2 time from when llq is invoked until the first job
is
  displayed on the screen is still significantly longer than
it was
  on 1.3, but the last job output is reasonably similar.
Note2: only a plain llq will be enhanced. llq performance
with any
  options will remain unchanged.

------

APAR: IY27621 COMPID: 5765D5100 REL: 340
ABSTRACT: MKNIMRES CAN FAIL TO CREATE NOPROMPT RESOURCE WHEN NON-DEFAULT I

PROBLEM DESCRIPTION:
mknimres will fail to create the noprompt resource for a node
under the following conditions:
A non-default (hdisk0) install disk is specified.
There are other nodes with customized noprompt resources defined
(they have non-default install disk) whose resource name is a
supperset of the node being defined. For example a noprompt
resource "11_noprompt" exists for node 11 and it tries to create
a "1_noprompt" resource for node 1.

PROBLEM SUMMARY:
Before mknimres creates new bosinst resources (_noprompt
and _migrate) it checks to see if they already exist by
doing a simple grep on the "lsnim -l" output. But it
will get an unintentional match if there is a resource
that is a superset of the one it is checking for e.g.
1_noprompt and 11_noprompt. When this happens it will
not create the resources and setup_server will fail
later when it tries to allocate them.

PROBLEM CONCLUSION:
The grep has been changed to match the search string
specifically.

------

APAR: IY27666 COMPID: 5765D5100 REL: 320
ABSTRACT: SDR_CONFIG FAILS IF NODE 1 EXITS IN /ETC/SWITCH.INFO BUT

PROBLEM DESCRIPTION:
If node 1 exists in the /etc/switch.info file (only possible on
system of all stand-alones) but is not the first line in that
file, SDR_config will fail with the following message:
SDR_config: Msg 0016-725 SDR_config had fatal errors.
These messages will be found in SDR_config.log:
SDR_config: 0016-709 Used switch node number in
/etc/switch.info, "1 0".
SDR_config: 0016-706 Problem while attempting to read the
/etc/switch.info file.
This will cause problems when trying to add a frame or running
hmreinit. This situation can happen naturally if stand-alone
frame 1 is deleted and re-added (new nodes are always appended
to the end of switch.info).
No other node ordering seems to matter in switch.info.
Another symptom of this problem is that the Syspar_map file
will be empty due to the fatal error in SDR_config.

LOCAL FIX:
Either manually edit switch.info or delete it and let hmreinit
recreate it.

PROBLEM SUMMARY:
If /etc/switch.info contains a switch node number of 0,
but not as the first entry in the file, subsequent
invocations of SDR_config will fail after issuing the
following messages:
SDR_config: 0016-709 Used switch node number in
/etc/switch.info, "1 0".
SDR_config: 0016-706 Problem while attempting to read the
/etc/switch.info file.

PROBLEM CONCLUSION:
If /etc/switch.info contains a switch node number of 0,
but not as the first entry in the file, SDR_config
would fail because it thought the switch node number
of 0 was already in use. A check to determine if the
switch node number had been used was failing because
it was comparing 0 to a null string, which it considered
a match. The check has been modified to prevent this error.

------

APAR: IY27678 COMPID: 5765D5100 REL: 311
ABSTRACT: UPDAUTHFILES SCRIPT FAILS WHEN THE HOME DIRECTORY FOR ROOT IS

PROBLEM DESCRIPTION:
When the home directory for root on cws is not / but this /root
updauthfiles script fails to update the .rhost files. This
appears to be fixed in later releases.

LOCAL FIX:
Update the script updauthfiles to get the home directory as in
later releases and set the file names accordingly.
Here is excerpts of code in rmoh where getpuid was used to get
home directory.
$ROOTHOME = (getpwuid($<)) 7 ;
if ($ROOTHOME eq "/" ) { $ROOTHOME = ""; }
$RHOSTS = "$ROOTHOME/.rhosts";
..

PROBLEM SUMMARY:
updauthfiles is creating the .rhosts file in /, even when
the home directory of root is not /.

PROBLEM CONCLUSION:
Modified updauthfiles to create .rhosts in root's home
directory instead of in /.

------

APAR: IY27683 COMPID: 5765D5100 REL: 340
ABSTRACT: SDR_INIT IN SSP.BASIC.POST_U GIVES RC OF 1 DURING FILESET UPDATE

PROBLEM DESCRIPTION:
savebase failed in SDR_init in ssp.basic.post_u.
this happens when upgrading to ssp.basic 3.4.0.2.

LOCAL FIX:
run "SDR_init -l" manually after updating to 3.4.0.2.

PROBLEM SUMMARY:
When ssp.basic.post_u calls SDR_init during a PSSP-3.4 PTF
installation, SDR_init reports a failure running savebase.
This is due to the ODMDIR being to set to something other
than /etc/objrepos when savebase is called.

PROBLEM CONCLUSION:
The defect is fixed by calling savebase with an option to
use /etc/objrepos explicitly as the ODM directory.

------

APAR: IY27691 COMPID: 5765E7200 REL: 310
ABSTRACT: CLIENT CANNOT MODIFY THE TIME STAMPS OF DIFFERENT USER'S FILE

PROBLEM DESCRIPTION:
A user can write to a file owned by different user, but can
not modify the time stamp of the file.

PROBLEM CONCLUSION:
A parameter is provided. When the parameter is enabled,
the time stamp can be set by any user who has write
permission on that file.

------

APAR: IY27693 COMPID: 5765E7200 REL: 310
ABSTRACT: CIFS_311: MORE README UPDATES

PROBLEM DESCRIPTION:
README file needs update for latest product enhancements

PROBLEM CONCLUSION:
Update the README file.

------

APAR: IY27727 COMPID: 5765D5100 REL: 320
ABSTRACT: SRVTAB CHECKSUM FILE IS NOT GENERATED ON ACTIVE BACKUP CWS

PROBLEM DESCRIPTION:
The srvtab checksum file only gets generated when the srvtab
file gets generated. When a backup CWS is active in an HACMP
environment, the srvtab file is not regenerated; the existing
srvtab file is reused. But if the checksum file doesn't already
exist, it won't be created and setup_server will exit with
errors.

PROBLEM SUMMARY:
The code didn't take into account that a backup CWS in an
HACWS environment will not generate a new srvtab and
checksum file but will reuse the existing ones.

PROBLEM CONCLUSION:
The installation of this PTF will result in a one-time
generation of checksum files for all existing srvtab files.
Thus we will avoid the situation where there is a srvtab but
no checksum and the backup CWS is active.
320

------

APAR: IY27782 COMPID: 5765C3403 REL: 430
ABSTRACT: ADAPT LIBPERFSTAT.H TO BE C++ SAFE

PROBLEM DESCRIPTION:
Warnings compiling applications to the Perfstat
Library API.

PROBLEM CONCLUSION:
Update libperfstat.h to follow conventions

------

APAR: IY27810 COMPID: 5765E2600 REL: 502
ABSTRACT: ERROR PASSING A FUNCTION POINTER TO A TEMPLATE FUNCTION

PROBLEM DESCRIPTION:
Compiler returns error 1540-1139 when a pointer to a function
declared with throw() exception-spec is passed as an argument
to a template function which accepts a pointer of some type T.
Testcase:
=========
#include <typeinfo>
#include <iostream>
using namespace std;
template<class T> void TemplateFunction( T *p ){
    cout << typeid(*p).name() << endl;
}
void Function() throw(){
    return;
}
int main(){
    int *ptr;
    TemplateFunction( &ptr );
    TemplateFunction( &Function );
    return 0;
}
"test.C", line 17.5: 1540-1139 (S) An exception-specification
can appear only in a function or pointer declaration.

LOCAL FIX:
The testcase compiles fine if throw() is removed from function
declaration. The problem may be circumvented by modifying the
template function to accept T (*)() instead of T *; however,
it would prevent the passing of primitive pointers.

PROBLEM SUMMARY:
The type descriptor for the typeid(<expression>), and the
deduced template argument have been checked to ensure that they
do not have exception specifiers.

PROBLEM CONCLUSION:
The check for typeid(<typeid>) and specified template arguments
has been implemeted instead.

------

APAR: IY27980 COMPID: 5765E2600 REL: 500
ABSTRACT: SIGSEGV 11 IN BE : RTETRP

PROBLEM DESCRIPTION:
A segmentation violation occurs using V5 on AIX 433 and AIX 51
when compiling with optimization.
Using 3.6.6, it compiles with optimization.
It is failing in the Back End with the following segmentation
violation:
Second exception at 0x1002EF7C
EXCEPTION of type: SIGSEGV 11 in exception handler at
1002EF7C.
Regs:
00000000 2FF1F420 301B4778 00000000 00000000 00000001 FFFFFFFF
20202020
20202020 66666667 00000000 00000000 1002EDC0 000F423F 2FF21ED0
00000002
0000000F 00000000 0000000C 00000004 00000001 00000000 00000001
2FF20A1E
2FF1F480 20200000 3EDC5054 3EDC8680 300147A8 10441080 00000000
00000002
resume: cr: lr: ctr: xer: fpscr: msr:
mq:
00000000 42282242 1002EDF8 00000000 20000000 00000000 0000D0B2
DEADBEEF
Traceback:
Line ? Disp 000002E0 flush_be_msgs
Dsa:
2FF1F420
Line ? Disp 00000044
Dsa:
2FF21C80
Line ? Disp 0000023C $except
Dsa:
2FF21CE0
Line ? Disp 0000010C RTETRP
Dsa:
2FF21D70
--- End of call chain ---
Original EXCEPTION details possibly follow.
Exception of type: SIGSEGV 11 at 1002C020. Regs:
1002ACB4 2FF22260 301B4778 30033030 3016F838 00000FA0 57338844
000003E8
00000000 F0046B40 6EDE5058 000034E0 1002ACA8 00000010 2FF223F8
2FF2243C
DEADBEEF FFFFFFFF 3001F8C8 30015C38 3016D3C8 00000002 00010938
00000008
30034D9C 3EDC8EF0 30034BEC 00000004 3EDC5140 00000000 00000001
30032680
resume: cr: lr: ctr: xer: fpscr: msr:
mq:
00000000 24282844 1002ACB4 00000000 20000000 00000000 0000D0B2
DEADBEEF
Traceback:
Line ? Disp 00000058
Dsa:
2FF22260
Line ? Disp 000000DC
Dsa:
2FF222F0
--- End of call chain ---
Compile using: xlC -qnoansialias -O -V -c test.i

LOCAL FIX:
Compile without optimization.

PROBLEM SUMMARY:
same as submitters text

PROBLEM CONCLUSION:
The backend was not checking to see if
it could map a 948M file into the translators address space.
Making this check now produces a "compilation ended due to
lack of space" message.

------

APAR: IY28037 COMPID: 5765E3200 REL: 502
ABSTRACT: FOR LOOP ENDS PREMATURELY WHEN INDEX USED TO TRAVERSE LARGE ARRA

PROBLEM DESCRIPTION:
When optimizing an application that uses a for-loop index to
access elements of very large array that crosses a 4GB
boundary in memory, incorrect code is generated by the compiler
that results in the for-loop being exited unexpectedly.

LOCAL FIX:
Compile using the -qxflag=NLOOPING compiler switch which
disables the defective optimization.

PROBLEM CONCLUSION:
The backend of the compiler was emitting
32 bit comparison instructions instead of 64 bit comparison
instructions,

TEMPORARY FIX:
Specify the compiler flag -qxflags=NLOOPING

------

APAR: IY28039 COMPID: 5765C3403 REL: 430
ABSTRACT: LOAD DRIVER INSTANCE BASED ON PATHS.

PROBLEM DESCRIPTION:
Possible machine crash due to the driver being paged
out when it's private buffers are being returned.

PROBLEM CONCLUSION:
Changed the driver to make an additional kernel call to
ensure that the driver continues to be pinned in memory
if there are outstanding private buffers.

------

APAR: IY28063 COMPID: 5765D5100 REL: 320
ABSTRACT: SECURITY ISSUE

PROBLEM DESCRIPTION:
security issue

PROBLEM SUMMARY:
Security Issue

PROBLEM CONCLUSION:
Security Issue Resolved

------

APAR: IY28064 COMPID: 5765D5100 REL: 330
ABSTRACT: SECURITY ISSUE

PROBLEM DESCRIPTION:
security issue

PROBLEM SUMMARY:
Security Issue

PROBLEM CONCLUSION:
Security Issue Resolved

------

APAR: IY28065 COMPID: 5765D5100 REL: 340
ABSTRACT: SECURITY ISSUE

PROBLEM DESCRIPTION:
security issue

PROBLEM SUMMARY:
Security Issue

PROBLEM CONCLUSION:
Security Issue Resolved

------

APAR: IY28067 COMPID: 5765D5100 REL: 320
ABSTRACT: SETUP_CWS IS TRYING TO UPDATE AFS SRVTABS WHEN IT SHOULD NOT

PROBLEM DESCRIPTION:
For afs authentication, the srvtab files are only generated when
the principals are initially created. However as of IY12653
(63733 in PTF 2) the code is trying to regenerate them every
time a node is set to customize, install or migrate.

PROBLEM SUMMARY:
A change was made to the code in this release such that the
srvtab files are regenerated every time a node's bootp
response is set to customize, install, or migrate.
However, when afs authentication is being used the srvtab
file(s) can only be generated when the principals are
originally created. The result is setup_server failing
with:
afs_add_principal:0016-301 Cannot access /tmp/addprin.27490
setup_CWS: 0016-052 The add_principal command could not
add service principals to the Kerberos V4 database.
setup_server: 0016-279 Problem of internally called
command: /usr/lpp/ssp/bin/se tup_CWS; rc= 2.
setup_server: Processing incomplete (rc= 2).

PROBLEM CONCLUSION:
The code has been modified to not regenerate the
srvtab files when afs authentication is being used.

------

APAR: IY28075 COMPID: 5765E3200 REL: 500
ABSTRACT: ICE WITH -QIPA FOR C/FORTRAN CODE

PROBLEM DESCRIPTION:
Using parallel compiler for C and fortran. on combination of
C and fortran code causes and Internal compiler error when
linking the object files from both the fortran and the c
compiler.

LOCAL FIX:
Use the Fortran compiler to link

PROBLEM SUMMARY:
C front end was not build to encrypt the
code to IPA

------

APAR: IY28093 COMPID: 5765E6900 REL: 310
ABSTRACT: ADD IN WHO CAN ISSUE FOR LL COMMAND MAN PAGES

PROBLEM SUMMARY:
All the LoadLeveler command man pages do not state who can
issue those commands

PROBLEM CONCLUSION:
All the LoadLeveler command man pages will now have a
section "Who Can Issue" to tell users who have authority to
issue those commands.

------

APAR: IY28094 COMPID: 5765B9501 REL: 340
ABSTRACT: LONG WAITERS: WAITING FOR RPC REPLIES FOR TMMSGTELLSERVER

PROBLEM DESCRIPTION:
Long Waiters: waiting for RPC replies for tmMsgTellServer

PROBLEM SUMMARY:
Deadlock due to lock handling on some recent
high SMP processors.

PROBLEM CONCLUSION:
Missing isync in a locking routine caused
incorrect wait counts in processors which handle instructions
in parallel.

------

APAR: IY28095 COMPID: 5765B9501 REL: 330
ABSTRACT: LONG WAITERS: WAITING FOR RPC REPLIES FOR TMMSGTELLSERVER

PROBLEM DESCRIPTION:
Long Waiters: waiting for RPC replies for tmMsgTellServer

PROBLEM SUMMARY:
Deadlock due to lock handling on some recent
high SMP processors.

PROBLEM CONCLUSION:
Missing isync in a locking routine caused
incorrect wait counts in processors which handle instructions
in parallel.

------

APAR: IY28109 COMPID: 5765D5100 REL: 340
ABSTRACT: SWITCH TABLE FORCE UNLOAD FAILS RC=1 ON SP-SWITCH

PROBLEM SUMMARY:
his problem occurs on the SP-Switch only, during the force
unload of user space jobs the fault service kernel extension
was not cleaning up all resources correctly.

TEMPORARY FIX:
To prevent the node from being allocated to other jobs after
switch table error occurs, you can set:
DRAIN_ON_SWITCH_TABLE_ERROR=TRUE
in the LoadL config file.

------

APAR: IY28142 COMPID: 5765C3403 REL: 430
ABSTRACT: GXT3000P RENDERING PROBLEM WITH CATIA V5R6 SKETCHTRACER

PROBLEM DESCRIPTION:
Rendering problem on GXT3000P, especially when zoomed
in on a SketchTracer image. Problem only occurs with
certain images depending on height and width of base
tif file.

PROBLEM SUMMARY:
Rendering problem on GXT3000P, especially when zoomed
in on a SketchTracer image. Problem only occurs with
certain images depending on height and width of base
tif file.

PROBLEM CONCLUSION:
Correct test in Proxy path to reject a mipmapped
texture that is 2048x512 as too big. The application
will then choose a smaller size that will work fine.

------

APAR: IY28193 COMPID: 5765D5100 REL: 340
ABSTRACT: ALLOWING NODES ON AND OFF THE SWITCH

PROBLEM DESCRIPTION:
allowing nodes on and off the switch

PROBLEM SUMMARY:
Additional support required to allow nodes
at PSSP 3.1.1 and PSSP 3.2 to not be on the SP Switch2.

PROBLEM CONCLUSION:
Provided new support to allow nodes at
PSSP 3.1.1 and PSSP 3.2 to not be on the SP Switch2.

------

APAR: IY28194 COMPID: 5765D5100 REL: 340
ABSTRACT: SP SWITCHLESS FRAME FORCE NON-PARTITIONABLE

PROBLEM DESCRIPTION:
sp switchless frame force non-partitionable

PROBLEM SUMMARY:
Support required to not allow a system to be
partitionable when an SP Switchless frame exists.

PROBLEM CONCLUSION:
Provided support to not allow a system to
be partionable when an SP Switchless frame exists.

------

APAR: IY28198 COMPID: 5765E2600 REL: 502
ABSTRACT: ICE WITH -QPDF1

PROBLEM DESCRIPTION:
V502:ICE:when using -qpdf1 Regression with 4Q2001 PTF
testcase:
#include <iostream>
int main (){
std::cout<<"works"<<std::endl;
return 0;
}
compile:
xlC -qpdf1 -O3 t.C -c -v
gets an ICE.
however with
xlC -O3 t.C -c -v
does not get an ICE.
Anything before
011221 Driver is okay
This is a regression

PROBLEM SUMMARY:
V502:ICE:when using -qpdf1 Regression with 4Q2001 PTF
testcase:
#include <iostream>
int main (){
std::cout<<"works"<<std::endl;
return 0;
}
compile:
xlC -qpdf1 -O3 t.C -c -v

------

APAR: IY28199 COMPID: 5765D5100 REL: 340
ABSTRACT: SRVTAB CHECKSUM FILE IS NOT GENERATED ON ACTIVE BACKUP CWS

PROBLEM SUMMARY:
The code didn't take into account that a backup CWS in an
HACWS environment will not generate a new srvtab and
checksum file but will reuse the existing ones.

------

APAR: IY28206 COMPID: 5765C3403 REL: 430
ABSTRACT: RPC.YPPASSWDD USERNAME FIX

PROBLEM DESCRIPTION:
Changing password fails as username is trucated to three
letters.

PROBLEM SUMMARY:
***************************************************************
*USERS AFFECTED: *
* All users with the following filesets at these levels *
* bos.net.nis.server 4.3.3.76
***************************************************************
*PROBLEM DESCRIPTION: *
* Changing password fails as username is trucated to
* three letters.
***************************************************************
*RECOMMENDATION: *
* Apply apar IY28206
***************************************************************

PROBLEM CONCLUSION:
Change size of string to be copied to the proper enviromental
variable

------

APAR: IY28238 COMPID: 5765B9501 REL: 330
ABSTRACT: NFS HANGS WHEN GPFS RETURN ENOTREADY AND CONTINUES ASYNCHRONOUSL

PROBLEM DESCRIPTION:
GPFS blocks that file operation until it receives a response
from DM application. Since, they set the "dmapiEventTimeOut"
config parameter, when time expires GPFS returns ENOTREADY and
continues that ile operation asynchronously. NFS is expected to
repeatedly retry that operation.

PROBLEM SUMMARY:
NFS access for all files hangs when a data management
application using the DMAPI interface takes a long time to
respond to an event.

PROBLEM CONCLUSION:
Aviod holding a lock on a file when it is accessed via NFS
and DMAPI events are pending for this file.

------

APAR: IY28239 COMPID: 5765D5100 REL: 340
ABSTRACT: CLEANUP.LOGS.NODES TRIMS ALL FILES DEFINED IN /ETC/SYSLOG.CONF.

PROBLEM CONCLUSION:
cleanup.logs.nodes and cleanup.logs.ws were modified to
only trim messages that are older than six days from
/var/adm/SPlogs/SPdaemon.log.

------

APAR: IY28240 COMPID: 5765D5100 REL: 340
ABSTRACT: SDR_CONFIG FAILS IF NODE 1 EXITS IN /ETC/SWITCH.INFO BUT

LOCAL FIX:
Either manually edit switch.info or delete it and let hmreinit
recreate it.

------

APAR: IY28242 COMPID: 5765B9501 REL: 330
ABSTRACT: GPFS IS NOT RESTING DT_PERS FIELD AFTER REMOVING ALL

PROBLEM DESCRIPTION:
GPFS not resetting the dt_pers field after removing all dm
attributes associated with the file.

PROBLEM SUMMARY:
GPFS not resetting the dt_pers field after removing
all dm attributes associated with the file. This applies
only to applications using the libdmapi library supplied
with GPFS.

PROBLEM CONCLUSION:
Reset dt_pers flag when external attributes are removed.

------

APAR: IY28268 COMPID: 5765D5100 REL: 320
ABSTRACT: PARENT/CHILD PROCESSES TIMING PROBLEM FROM SPMON QUERY

PROBLEM DESCRIPTION:
Patent/Child processes timing problem from spmon query function

PROBLEM SUMMARY:
When spmon does a query it forks a process. But the parent
does not wait for the child to finish and exits.
If the parent is faster than the child then it will return
the user to the command line (print the prompt) before the
query's output is written. This gives the impression that
the shell prompt was never returned and that the command is
hung.

PROBLEM CONCLUSION:
The code has been modified so that on a query the parent
will now wait for the child process to finish before it
exits and returns the user to the command prompt.

------

APAR: IY28305 COMPID: 5765B9501 REL: 330
ABSTRACT: MANPAGE UPDATE

PROBLEM DESCRIPTION:
manpage update

PROBLEM SUMMARY:
When deleting nodes from a GPFS nodeset, if the user doesnot
ensure the virtual shared disks are associated with some
node remaining in the nodeset, those virtual shared disks
will be fenced from the nodeset until the GPFS daemon is
restarted.

PROBLEM CONCLUSION:
In the Administration and Programming Reference for the
mmdelnode command and the corresponding man page, this
paragraph was added to the description:
In an SP environment, when the number of nodes being deleted
from a GPFS nodeset results in the loss of quorum, virtual
shared disks associated only with the deleted nodes are
fenced from the remaining nodes in the nodeset. When the
GPFS daemon on the remaining nodes restarts in order to
achieve quorum again, the virtual shared disks will no
longer be fenced from the nodeset

------

APAR: IY28313 COMPID: 5765E2600 REL: 500
ABSTRACT: ICE IN FE INSTEAD OF ERROR MESSAGE

PROBLEM DESCRIPTION:
Simple, incorrect testcase produces ICE in FE instead of error
message. The customer realizes that the testcase is invalid,
but feels that the compiler should not be producing an ICE.
It should be giving an error message that will help the
customer to correct his code.
Reproducable on 5.0.0.0, 5.0.2.0, 4Q 2001 PTF, daily/latest
compiler build and with the BETA.
Using 366, the testcase produces the following error
message:
1540-060: (S) An expression of type "double*" is not allowed on
the left side of ".operator=".
1540-099: (W) "dit" is used but not set in function "main()".
Here is the testcase:
int main () {
double *dit;
dit.operator= (5);
}
Compile using: xlC test.cpp

PROBLEM CONCLUSION:
Development added a check to resolve
member declaration before it is dereferenced to resolve the
ICE and produce an error message.

------

APAR: IY28382 COMPID: 5765B8100 REL: 220
ABSTRACT: VARIABLES AND INPUT PARAMETERS CAN BE MODIFIED TO THE SAME NAME

PROBLEM DESCRIPTION:
Variables and Input Parameters can be modified to the same name.
DirectTalk should prevent you from having duplicate names for
variables and input parameters.

PROBLEM SUMMARY:
Variables and Input Parameters can be modified
to the same name. DirectTalk should prevent you from having
duplicate names for variables and input parameters.

PROBLEM CONCLUSION:
Rewrote the checking code correctly.

------

APAR: IY28383 COMPID: 5765B9501 REL: 330
ABSTRACT: SOME GPFS COMMANDS DON'T WORK WITH LARGE FILE SYSTEMS

PROBLEM DESCRIPTION:
GPFS commands mmgetacl, mmputacl, and mmeditacl do not work
with large file systems.

PROBLEM SUMMARY:
mmget/put/edit/delacl commands failed on large GPFS files >
2GB

PROBLEM CONCLUSION:
The problem is the stat call in tsget/put/edit/delacl. For
that to work on large files, need to add "#define
_LARGE_FILES".

------

APAR: IY28449 COMPID: 5765C3403 REL: 430
ABSTRACT: XMALLOC 'PERM_ACS' STRUCTURE

PROBLEM DESCRIPTION:
Possible crash in an HACMP setup.

PROBLEM CONCLUSION:
The device driver is to be changed to have one of it's
structures permanently pinned in memory if the driver is
being closed and there are outstanding private mbufs.

------

APAR: IY28454 COMPID: 5765D6100 REL: 220
ABSTRACT: STARTD CORE DUMPS WHEN LOG DIRECTORY DOES NOT EXIST

PROBLEM DESCRIPTION:
startd core dumps when log directory does not exist

PROBLEM SUMMARY:
LoadLeveler startd will core dump when
running interactive jobs and
the log directory does not exist.

PROBLEM CONCLUSION:
LoadLeveler startd will not core dump
when running interactive jobs even
if the log directory does not exist.

------

APAR: IY28479 COMPID: 5765D5100 REL: 330
ABSTRACT: INITIALIZATION FAILURE OCCURS IN _LAPI_INIT_FUNCTION, BUT THE

PROBLEM DESCRIPTION:
Sometimes, the _lapi_init_function fails, but it gives a zero
return code (success). If the adapter then comes backup, it will
reference un initialized fields.

PROBLEM SUMMARY:
A successful return code was returned for port
initialization when the initialization had failed. This
caused a crash when an attempt was made to use the port.

PROBLEM CONCLUSION:
Make sure a failure is returned any time a port
initialization fails.

------

APAR: IY28491 COMPID: 5765B9501 REL: 320
ABSTRACT: RECLAIMRANGETOKEN ACQUIRES SAME LOCK TWICE IN A ROW WITHOUT

PROBLEM DESCRIPTION:
ReclaimRangeToken acquires same lock twice in a row without
release, leading to deadlock.

PROBLEM SUMMARY:
ReclaimRangeToken() acquires the token lock twice and
results in deadlock

PROBLEM CONCLUSION:
Removed second call to acquire token lock.

------

APAR: IY28514 COMPID: 5765C3403 REL: 430
ABSTRACT: BLIT DISPLAYS INCORRECTLY ON GXT135P

PROBLEM DESCRIPTION:
Blit images may be displayed garbled.

PROBLEM CONCLUSION:
Correct the calculations of the source X and Y values used
for system-to-screen and screen-to-screen blits.

------

APAR: IY28524 COMPID: 5765C3403 REL: 433
ABSTRACT: ADD SUPPORT FOR A TWO BIT QERR FIELD IN THE CONTROL MODE PAGE

PROBLEM DESCRIPTION:

PROBLEM SUMMARY:
In a multi-initator environment with SPC-2 compliant drives
that set the upper QErr bit to 1, will not have that bit
set without this fix.

PROBLEM CONCLUSION:
The device driver now has support for the two bit QErr field.

------

APAR: IY28525 COMPID: 5765E2600 REL: 502
ABSTRACT: PRIORITY OF 0 PASSED TO MUNCH WHEN -G SPECIFIED WITHOUT -QMKSHRO

PROBLEM DESCRIPTION:
When one attempts to create a shared object enabled for use
with the runtime linker by specifying the -G option
but not the -qmkshrobj option on the driver command line.
The driver passes a priority of 0x0 to the munch step
instead of the default priority of 0x80000000. This results
in the shared objects being created that have their global
static variables initialized before any other shared objects
are initialized, including those in the C++ runtime.
This results in an error at runtime if the constructors in
those static objects use code, for example the iostreams
library from the C++ runtime libraries.

LOCAL FIX:
Specify the -qmkshrobj option in addition to the -G option.

PROBLEM SUMMARY:
Same as submitter's text.

PROBLEM CONCLUSION:
Pass the correct priority to munch when
the -qmkshrobj option is not specified along with -G

------

APAR: IY28529 COMPID: 5765C3403 REL: 430
ABSTRACT: ON GXT135P, 24-BIT, 1 X 1 FILLED RECTS DISPLAY INCORRECTLY

PROBLEM DESCRIPTION:
In 24-bit mode, rectangles of size 1 x 1 may be displayed in
the wrong color and at the wrong location. This can affect
higher level geometries which are decomposed into 1 X 1
rectangles before being drawn.

PROBLEM CONCLUSION:
For 24-bit, correct the calculation of the memory address
where the pixel data is to be stored and recast the pointer
to the correct type. Also for 24-bit, change the variable
which holds the pixel data to the correct type.

------

APAR: IY28536 COMPID: 5765B9501 REL: 320
ABSTRACT: FLAG NOT BEING SET IN RECLOCKSLEEPERINIT CAUSING ASSERT IN

PROBLEM DESCRIPTION:
mmds daemon asserts in kxRecLockAcquires (which is due to
to its retry_fcn!=NULL).

PROBLEM SUMMARY:
There are two lists for blocked locks (one waiting for
tokens and one waiting for the local lock). When a lock is
retried, it knows which list it will find itself on by the
existance of the retry_fcn. Acquiring a lock is a two step
process: set the retry_fcn and wait for the token, and then
clear the retry_fcn and wait for the local lock. Since the
retry_fcn was not being cleared after the first step, it
looked (incorrectly) on the acquire list when it was now on
the lock list.

PROBLEM CONCLUSION:
RecLockSleeperInit() needs to reset retry_fcn pointer to
NULL before waiting for local lock

------

APAR: IY28563 COMPID: 5765C3403 REL: 430
ABSTRACT: CALL TO LOW IN LOCKD DUE TO NULL VNODE POINTER

PROBLEM DESCRIPTION:
Machine crashes in lockd due to a call to low (jump to
address 0). This is one typical stack:
IAR: .low+0 (00000000):
LR: .vnop_lockctl+1c (001c206c)
f0611878: . nfs.ext:lm_cancel +dc (05475fa8)
f06118f8: . nfs.ext:lm_nlm4_dispatch +348 (05477e24)
f06119d8: . nfs.ext:svc_getreq +5ac (05470a84)
f0611f78: .threadentry+18 (00098894)
f0611fb8: .low+0 (00000000)
But other other functions than lm_cancel are affected.

PROBLEM CONCLUSION:
Change the order of operations so that the lock protecting the
list of lm_vnodes is held across the check for NULL vnode
pointers before returning from lm_get_vnode.

------

APAR: IY28571 COMPID: 5765E6900 REL: 310
ABSTRACT: LOADL2.2 HISTORY/GLOBALHISTORY FILES 0 WRITE WHEN NFS IS FULL

LOCAL FIX:
write locally to a JFS file system. If successful, mv to NFS
as previously desired.

PROBLEM SUMMARY:
When the NFS filesystem is full and llacctmrg command
isn't able to write its global history file there ,
the local history files in the spool directory will
go to zero length.

PROBLEM CONCLUSION:
When the NFS filesystem is full and llacctmrg command
isn't able to write its global history file there ,
the local history files in the spool directory will
not be affected.

------

APAR: IY28577 COMPID: 5765D5100 REL: 340
ABSTRACT: NODE PANIC IN VSDD:SUSVSD+6CC, RETROFIT OF DT 79627 INTO

PROBLEM DESCRIPTION:
node crashes in vsdd:susvsd+6cc.
defect 79627 for pssp 3.4 describes that in detail.
fix must be retrofitted into 3.2.

PROBLEM SUMMARY:
Certain conditions would result in suspendvsd asserting.

PROBLEM CONCLUSION:
suspendvsd was changed to assert only when it finds a
request in any vsd queues. If no requests are found in any
queues after a timeout period, suspendvsd will now complete.
Also, suspendvsd will now only close the lvm device
if the file descriptor is not null.

------

APAR: IY28609 COMPID: 5765B9501 REL: 330
ABSTRACT: SEGMENTATION IN DMCLEANUPACCESSRIGHT

PROBLEM DESCRIPTION:
segmentaion in dmCleanupAccessRight when accessing NULL sgP
pointer.

PROBLEM SUMMARY:
Customer application using DMAPI calls to GPFS caused a
segmentation violation within GPFS.

PROBLEM CONCLUSION:
Fix uninitialized pointer

------

APAR: IY28638 COMPID: 5765B9501 REL: 340
ABSTRACT: WAIT4RVSD PARAMETER NOT DOCUMENTED SUFFICIENT

PROBLEM SUMMARY:
The mmlsconfig command displays configuration variables that
are for problem determination and are therefore not detailed
in either the mmconfig or mmchconfig commands.

------

APAR: IY28644 COMPID: 5765B9501 REL: 330
ABSTRACT: SIGNAL 11 IN FORCEDONERECORDS ON RO MOUNTED FILESYSTEM

PROBLEM DESCRIPTION:
Signal 11 in forceDoneRecords on RO mounted filesystem on token
revoke. token_revoke should not call forceDoneRecords if the
filesystem logfile pointer is null, which is the case when it is
mounted RO

PROBLEM SUMMARY:
Signal 11 in forceDoneRecords on RO mounted filesystem on
token revoke.

PROBLEM CONCLUSION:
token_revoke should not call forceDoneRecords if the
filesystem logfile pointer is null, which is the case when
it is mounted RO

------

APAR: IY28689 COMPID: 5765E2600 REL: 500
ABSTRACT: BAD IMPLICIT OPERATOR= WITH MULTIPLE VIRTUAL BASE CLASSES

PROBLEM DESCRIPTION:
The default assignment operator does not work properly for a
most derived class.
The most derived class inherits from:
- two derived classes with virtual base classes
- one base class.

LOCAL FIX:
1. explicitly define operator=

PROBLEM CONCLUSION:
Development has to fix casting to
virtual base.

------

APAR: IY28690 COMPID: 5765B9501 REL: 340
ABSTRACT: MANPAGE UPDATE

PROBLEM DESCRIPTION:
manpage update

------

APAR: IY28692 COMPID: 5765E6800 REL: 300
ABSTRACT: XMSERVD LOCK CONTENTION

PROBLEM DESCRIPTION:
Other PTX consumers like topas, xmperf will display Common
Memory Locked messages very oftenly when Hot Set exception
monitoring is done.

PROBLEM CONCLUSION:
Fixed xmservd to release the obtained common memory locks
properly.

------

APAR: IY28694 COMPID: 5765C3403 REL: 430
ABSTRACT: FTP PUT TO DFS DIRECTORY FAILS WITH INVALID ARGUMENT

PROBLEM DESCRIPTION:
Unable to put or append a file in ftp if user logged in with
dce integrated login.

PROBLEM CONCLUSION:
Set the flag O_NSHARE while opening a file in store() if the
user is dce authenticated.

------

APAR: IY28695 COMPID: 5765C3403 REL: 430
ABSTRACT: POSSIBLE SYSTEM HANG RUNNING 64-BIT APPLICATIONS

PROBLEM DESCRIPTION:
Customers running 64-bit applications that use very large
global shared objects may see a system hang.

PROBLEM CONCLUSION:
Correct the behavior of the routines to increase the heap
for the shlap so that they properly update all variables
and exit.

------

APAR: IY28697 COMPID: 5765C3403 REL: 430
ABSTRACT: SYSTEM HANG WITH WLM

PROBLEM DESCRIPTION:
System hang with WLM active

PROBLEM CONCLUSION:
Ignore WLM priority penalty for boosted threads.

------

APAR: IY28698 COMPID: 5765C3403 REL: 430
ABSTRACT: BUFFER OVERFLOW IN RCP

PROBLEM DESCRIPTION:
possible remote root shell exploit

PROBLEM CONCLUSION:
changed unbounded string operators to their bounded
equivalents

------

APAR: IY28700 COMPID: 5765C3403 REL: 430
ABSTRACT: DTHELPPRINT COREDUMP WITH SOME HELP VOLUMES IN JP

PROBLEM DESCRIPTION:
dthelpprint coredumps with some help volumes in JP locales.

PROBLEM CONCLUSION:
added few lines of code to solve this defect.

------

APAR: IY28701 COMPID: 5765C3403 REL: 430
ABSTRACT: CATIA V4 GXT4000 2048X1546 RES. PIXMAP DOESN'T RENDERED ZOOMED

PROBLEM DESCRIPTION:
On GXT4000P,GXT6000P,GXT4500P and GXT6500P, a
pixmap rendered via graPHIGS on a monitor
running in 2048x1536 resolution may NOT
render.

PROBLEM CONCLUSION:
Change the precision of calculations used to convert to
DC and clip in son_drawpix.c. The calculations were using
12 bit integer math, which was resulting in an undetected
overflow at 2048x1536 resolution. This resulted in the
pixmap being completely clipped out of the rendering window.
The fix uses ordinary 32 bit integer math instead to avoid
the overflow.

------

APAR: IY28704 COMPID: 5765C3403 REL: 430
ABSTRACT: CHANGE IN VPD EXTRACTION FOR ULTRA3 PARALLEL SCSI ADAPTER

PROBLEM DESCRIPTION:
Some Ultra3 parallel SCSI adapters may not report VPD
correctly. This leads to the inability to service and
diagnose the adapter for tracking purposes.

PROBLEM CONCLUSION:
Modify the VPD extraction method so that it will work with
all versions of the adapter and correctly retrieve the VPD
data.

------

APAR: IY28706 COMPID: 5765C3403 REL: 430
ABSTRACT: SECURITY: BUFFER OVERFLOW IN RPC CODE

PROBLEM DESCRIPTION:
Possible buffer overflow condition in RPC routines.
This affects AIX 4.3 only.

PROBLEM CONCLUSION:
Correct variable data type.

------

APAR: IY28708 COMPID: 5765C3403 REL: 430
ABSTRACT: BOS.RTE.INSTALL 4.3.3.77 FAILS TO INSTALL FROM BOOT MEDIA

PROBLEM DESCRIPTION:
If using boot media that contains bos.rte.install 4.3.3.77,
then the install fails for any additional software, due to
space problems with bos.rte.install. In most cases the failure
shows up as a failure in the bosboot command.

PROBLEM CONCLUSION:
Use an upsize file to increase the space requested in /usr/lpp
directories.

------

APAR: IY28710 COMPID: 5765C3403 REL: 430
ABSTRACT: MODIFICATION OF IOCTL ROUTINES

PROBLEM DESCRIPTION:
Change will make the passthru ioctl comply with the SCSI3
standard more accurately

PROBLEM CONCLUSION:
The reserved field will not be used to place LUN value in
the ssc_passthru ioctl routine.

------

APAR: IY28711 COMPID: 5765C3403 REL: 430
ABSTRACT: ALLOCP CORE DUMPS DUE TO UPPERBOUND VALUE FROM EXTENDLV

PROBLEM DESCRIPTION:
Extendlv call to allocp fails on mirrored logical volume with
strictness = 'y'.

PROBLEM SUMMARY:
***************************************************************
*USERS AFFECTED: *
* All users with the following filesets at these levels *
* bos.rte.lvm 4.3.3.79
***************************************************************
*PROBLEM DESCRIPTION: *
***************************************************************
*RECOMMENDATION: *
* Apply apar IY28711
***************************************************************

PROBLEM CONCLUSION:
Only check validity of upperbound lv attribute if strictness
is 's' (superstrict).

------

APAR: IY28734 COMPID: 5765B8100 REL: 230
ABSTRACT: NEW_CALL RETURNS STATUS CODE 104 IF GEOTEL CS IS STARTED WITH -C

PROBLEM DESCRIPTION:
If you start the GeoTel custom server with the -c parameter to
use logical channel numbers and channel groups, calls to
New_Call and Request_Instruction will fail. These calls will
return with a status code of 104.

PROBLEM SUMMARY:
If you start the GeoTel custom server with the
-c parameter to use logical channel numbers and channel
groups, calls to New_Call and Request_Instruction will fail.
These calls will return with a status code of 104.

PROBLEM CONCLUSION:
Modified code to pass logical trunk group
and channel number when -c option specified

------

APAR: IY28737 COMPID: 5765B8100 REL: 220
ABSTRACT: DTXA ERROR 17301 OCCURS INTERMITTENTLY

PROBLEM DESCRIPTION:
DTXA error 17301 can very intermittently occur due to a timing
problem in the DTXA adapter microcode setup.
This APAR corrects the timing problem.

PROBLEM SUMMARY:
DTXA error 17301 can very intermittently occur
due to a timing problem in the DTXA adapter microcode setup.

PROBLEM CONCLUSION:
Changes to dtxahal.c (in dtxahal.rel) and
secpinit.c (in seccode.cof)

------

APAR: IY28753 COMPID: 5765B9501 REL: 320
ABSTRACT: SIGNAL 11 IN FORCEDONERECORDS ON RO MOUNTED FILESYSTEM

PROBLEM SUMMARY:
Signal 11 in forceDoneRecords on RO mounted filesystem on
token revoke.

PROBLEM CONCLUSION:
token_revoke should not call forceDoneRecords if the
filesystem logfile pointer is null, which is the case when
it is mounted RO

------

APAR: IY28790 COMPID: 5765C3403 REL: 430
ABSTRACT: FIXDATA CHANGES FOR FUTURE MAINTENANCE PACKAGES

PROBLEM DESCRIPTION:
Changes to fixdata required for future maintenance packages.

PROBLEM CONCLUSION:
Add fixdata entries for oslevel -r.

------

APAR: IY28826 COMPID: 5765B9501 REL: 330
ABSTRACT: MMFS DAEMON ASSERTING IN XATTR-VFS.C (LINE 178).

PROBLEM DESCRIPTION:
mmfs daemon asserting with the following msg ..
Failure at line 178 in file ../../../../../../../src/avs/fs/mmfs
/ts/fs/xattr-vfs.C strlen(keyP) <= keyLen.
Failure at line 178 in file ../../../../../../../src/avs/fs/mmfs
/ts/fs/xattr-vfs.C strlen(keyP) <= keyLen

PROBLEM SUMMARY:
GPFS self check logic asserted at line 178 of xattr-v.C

PROBLEM CONCLUSION:
Remove invalid assert in XAttrFind().

------

APAR: IY28829 COMPID: 5765B8100 REL: 230
ABSTRACT: THE 16TH TRUNK DOES NOT WORK ON CCS

PROBLEM DESCRIPTION:
The 16th trunk does not work on CCS

PROBLEM CONCLUSION:
Increased array dimensions of chan_seq
(table of active calls) in sl1.h

------

APAR: IY28861 COMPID: 5765D5100 REL: 330
ABSTRACT: NODE PANIC IN VSDD:SUSVSD+6CC, RETROFIT OF DT 79627 INTO

PROBLEM DESCRIPTION:
node crashes in vsdd:susvsd+6cc.
defect 79627 for pssp 3.4 describes that in detail.
fix must be retrofitted into 3.2.

PROBLEM SUMMARY:
Certain conditions would result in suspendvsd asserting.

------

APAR: IY28892 COMPID: 5765B8100 REL: 220
ABSTRACT: REMOVE 17301 FIX AS FAILS ON SOME ADAPTORS

PROBLEM DESCRIPTION:
The 17301 fix does not work on the whole range of adaptors.

PROBLEM SUMMARY:
The 17301 fix does not work on the whole range
of adaptors.

PROBLEM CONCLUSION:
Previously issued fix removed

------

APAR: IY28895 COMPID: 5765D5100 REL: 340
ABSTRACT: EXTRANEOUS CHARACTERS ON COLONY DIAG MENU

PROBLEM DESCRIPTION:
extaneous characters on colony diag menu

PROBLEM SUMMARY:
An adapter diags screen had extraneous characters displayed.
In previous releases of AIX and PSSP the positions occuppied
by these characters was automatically cleared, now we must
explicitly clear them.

PROBLEM CONCLUSION:
The diag code had to be changed to clear the extraneous
characters.

------

APAR: IY28912 COMPID: 5765C3403 REL: 430
ABSTRACT: SSA UPDATES MAY NOT INSTALL

PROBLEM DESCRIPTION:
SSA updates for devices.pci.14104500.diag and
devices.mca.8f97.diag may not install on AIX 4.3 systems.

PROBLEM CONCLUSION:
Cum devices.mca.8f97.diag and devices.pci.14104500.diag to
the 4.3.3.75 level.

------

APAR: IY28965 COMPID: 5765C3403 REL: 433
ABSTRACT: SYSTEM CRASH IN ARPRESOLVE()

PROBLEM DESCRIPTION:
Writing the new entry beyond the arp table resulting in
a system crash in arpresolve().

PROBLEM SUMMARY:
***************************************************************
*USERS AFFECTED: *
* All users with the following filesets at these levels *
* bos.net.tcp.client 4.3.3.80
***************************************************************
*PROBLEM DESCRIPTION: *
* Customer will experience system crash if the arp table
* is full at bos.net.tcp.client 4.3.3.80 level.
***************************************************************
*RECOMMENDATION: *
* Apply apar IY28965
***************************************************************

PROBLEM CONCLUSION:
Make the pointer to the new arp table to point to a right
location after finding the oldest entry in arptnew().

------

APAR: IY28980 COMPID: 5765C3403 REL: 430
ABSTRACT: ERROR LOGS SHOWS UNKNOWN ERROR REPORTED DURING ATTENUATOR TEST

PROBLEM DESCRIPTION:
Will see error reported, as unknown errors.

PROBLEM CONCLUSION:
set status_validity flag to SC_ADAPTER_ERROR, so it reports
error correctly.

------

APAR: IY28981 COMPID: 5765C3403 REL: 430
ABSTRACT: HOST WITH 6227/28 FC CARDS HUNG DURING FAILOVER

PROBLEM DESCRIPTION:
The system may hang during a failover condition.

PROBLEM CONCLUSION:
Make sure that the router monitor process' stack is pinned.

------

APAR: IY28982 COMPID: 5765C3403 REL: 430
ABSTRACT: FASTT: FAILOVER AND OPERATIONAL ENHANCEMENTS

PROBLEM DESCRIPTION:
Slow reporting of errors to the application layer on a total
failure where switches fail back and forth. EBUSY or EDEADLOCK
reported to the application on other errors. Insufficient time
allowed to recover from hardware-recoverable issues.

PROBLEM CONCLUSION:
Allow user to configure number of failover retries. Fix
reservation problems and clean up ENXIO-handling code. Allow
extra time for one retry.

------

APAR: IY28984 COMPID: 5765C3403 REL: 430
ABSTRACT: TAR COMMAND FAILS IN DIRECTORY LONGER THAN 260 CHARACTERS

PROBLEM DESCRIPTION:
Allow tar command to invoke from a directory whose length is
250 characters.

PROBLEM SUMMARY:
tar command fails to archive files when invoked from a
directory whose path length is greater than 260 characters.

PROBLEM CONCLUSION:
Code modified accordingly.

------

APAR: IY29032 COMPID: 5765C3403 REL: 430
ABSTRACT: HOT SWAP SAFTE FUNCTIONS FAIL

PROBLEM DESCRIPTION:
The diagnostic hot swap menus will show "populated" instead
of the hdisk name, ie. "hdisk0". The LED indicators will not
function properly.

PROBLEM CONCLUSION:
The two disk drive bays are located at SCSI ID's 12 and 13.
Looking at the front of the machine, ID 12 is on the left and
ID 13 is on the right.

------

APAR: IY29059 COMPID: 5765D5100 REL: 311
ABSTRACT: LATEST PSSP 3.1.1 FIXES AS OF FEBRUARY 2002.

PROBLEM DESCRIPTION:
This is the lastest PSSP ptf as of February 2002.
Order this apar to get all of the ptfs as of February 2002.

PROBLEM SUMMARY:
This is the latest PSSP ptf as of February 2002

PROBLEM CONCLUSION:
This is the latest PSSP ptf as of February
2002.

------

APAR: IY29091 COMPID: 5765D5100 REL: 330
ABSTRACT: LATEST PSSP 3.3.0 FIXES AS OF MARCH 2002

PROBLEM DESCRIPTION:
This is the latest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is a packaging apar for PSSP 3.3.0 fixes
as of March 2002

PROBLEM CONCLUSION:
This is a packaging apar for PSSP 3.3.0
fixes as of March 2002

------

APAR: IY29097 COMPID: 5765D9300 REL: 320
ABSTRACT: MAINTENANCE FOR PE R3.2

PROBLEM DESCRIPTION:
Maintenance for PE R3.2

------

APAR: IY29123 COMPID: 5765B9500 REL: 150
ABSTRACT: MAINTENANCE UPDATES

PROBLEM DESCRIPTION:
Maintenance Updates

------

APAR: IY29146 COMPID: 5765B8100 REL: 230
ABSTRACT: PARSING OF VOX.GETVDU AND VOX.GETVOX FAILS IF EXTRA ) PRESENT

PROBLEM DESCRIPTION:
Parsing of VOX.getvdu and VOX.getvox fails if extra ) present

PROBLEM SUMMARY:
Parsing of VOX.getvdu and VOX.getvox fails if
extra ) present

PROBLEM CONCLUSION:
Change VOX.getvdu and VOX.getvox to handle a
) in the name/value pairs

------

APAR: IY29154 COMPID: 5765B9500 REL: 140
ABSTRACT: MMCHFS DOES NOT CHECK FOR EXISTANCE OF SPECIAL DEVICE FILE FOR

PROBLEM DESCRIPTION:
When using mmchfs to change nodeset for fileystem there is no
check to see if the device special file for the gpfs filesystem
already exists on the nodes in the target nodeset. This becomes
a problem when the customer has a jfs filesystem that has a
device special file with the same name. The mmchfs command
completes successfully; however, if the mmchfs command is used
to bring the gpfs filesystem back to the original nodeset the
special device file that belongs to the jfs filesystem is
removed.

PROBLEM SUMMARY:
When using mmchfs to change nodeset for
filesystem there is no check to see if the device special file
for the gpfs filesystem already exists on the nodes in the
target nodeset. This becomes a problem when the customer has a
jfs filesystem that has a device special file with the same
name. The mmchfs command completes successfully; however, if
the mmchfs command is used to bring the gpfs filesystem back to
the original nodeset the special device file that belongs to the
jfs filesystem is removed.

PROBLEM CONCLUSION:
Do not remove /dev entries if they were not
created by GPFS.

------

APAR: IY29157 COMPID: 5765B9500 REL: 140
ABSTRACT: MMCHFS FAILS WITH _MOUNT_CHECK_ONLY ERROR

PROBLEM DESCRIPTION:
mmchfs fails with _MOUNT_CHECK_ONLY_error

PROBLEM SUMMARY:
Cannnot reassign a file system from a nodeset
where all nodes are unavailable.

PROBLEM CONCLUSION:
When moving a file system there is no need
for the daemon to be running anywhere in the source nodeset.

------

APAR: IY29186 COMPID: 5765D5100 REL: 340
ABSTRACT: LATEST PSSP 3.4.0 FIXES AS OF MARCH 2002

PROBLEM DESCRIPTION:
This is the lastest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is a packaging apar for PSSP 3.4.0 fixes
as of March 2002.

PROBLEM CONCLUSION:
This is a packaging apar for PSSP 3.4.0
fixes as of March 2002.

------

APAR: IY29200 COMPID: 5765B8100 REL: 230
ABSTRACT: MEMORY LEAK IN THE GRAMMER MAPPING TABLE

PROBLEM DESCRIPTION:
Memory leak in the Grammer mapping table

PROBLEM CONCLUSION:
Correct table handling.

------

APAR: IY29260 COMPID: 5765D5100 REL: 320
ABSTRACT: LATEST PSSP 3.2.0 FIXES AS OF MARCH 2002

PROBLEM DESCRIPTION:
This is the latest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is a packaging apar for PSSP 3.2.0 fixes
as of March 2002

PROBLEM CONCLUSION:
This is a packaging apar for PSSP 3.2.0
fixes as of March 2002

------

APAR: IY29367 COMPID: 5765D5100 REL: 311
ABSTRACT: LATEST PSSP 3.1.1 FIXES AS OF MARCH 2002.

PROBLEM DESCRIPTION:
This is the lastest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is the latest PSSP ptf as of March 2002

PROBLEM CONCLUSION:
This is the latest PSSP ptf as of March
2002.

------

APAR: IY29369 COMPID: 5765D5100 REL: 320
ABSTRACT: LATEST PSSP 3.2.0 FIXES AS OF MARCH 2002

PROBLEM DESCRIPTION:
This is the latest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is a packaging apar for PSSP 3.2.0 fixes
as of March 2002

PROBLEM CONCLUSION:
This is a packaging apar for PSSP 3.2.0
fixes as of March 2002

------

APAR: IY29370 COMPID: 5765D5100 REL: 330
ABSTRACT: LATEST PSSP 3.3.0 FIXES AS OF MARCH 2002

PROBLEM DESCRIPTION:
This is the latest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is a packaging apar for PSSP 3.3.0 fixes
as of March 2002

PROBLEM CONCLUSION:
This is a packaging apar for PSSP 3.3.0
fixes as of March 2002

------

APAR: IY29371 COMPID: 5765D5100 REL: 340
ABSTRACT: LATEST PSSP 3.4.0 FIXES AS OF MARCH 2002

PROBLEM DESCRIPTION:
This is the lastest PSSP ptf as of March 2002.
Order this apar to get all of the ptfs as of March 2002.

PROBLEM SUMMARY:
This is a packaging apar for PSSP 3.4.0 fixes
as of March 2002.

PROBLEM CONCLUSION:
This is a packaging apar for PSSP 3.4.0
fixes as of March 2002.

------

APAR: PQ57435 COMPID: 5765C4200 REL: 330
ABSTRACT: INCORRECT OUTPUT FROM DSRSM WHEN FIRST ROW IS ALL ZEROES.

PROBLEM DESCRIPTION:
When the first row of the original matrix is all zeroes, DSRSM
produces incorrect output.

PROBLEM SUMMARY:
When the first row of the input matrix A(stored
as AR) for DSRSM is all zeroes, the output matrix KA may
have incorrect entries until a row with non-zeroes is
encountered.

PROBLEM CONCLUSION:
DSRSM was corrected to handle this case
correctly.

------

APAR: PQ57437 COMPID: 5765C4200 REL: 330
ABSTRACT: POSSIBLE WRONG ANSWERS OR SEG FAULTS IN ESSL SMP SUBROUTINES FOR

PROBLEM DESCRIPTION:
Possible wrong answers and seg faults in SMP for large problems
in some ESSL subroutines(_GEMM, _SYMM, _HEMM, _GER, _GERC,
_GERU, _GETMO).

PROBLEM SUMMARY:
Internal ESSL subroutine DMMCRD returned
incorrect information to multiple ESSL subroutines in SMP
for large problems.

PROBLEM CONCLUSION:
DMMCRD was corrected to use 64-bit
integers for internal calculations.

------

APAR: PQ57448 COMPID: 5765C4200 REL: 330
ABSTRACT: IMPROVE DGEMM PERFORMANCE FOR SMALL SIZES ON P690

PROBLEM DESCRIPTION:
Improve performance of DGEMM on p690 for small matrix sizes.

PROBLEM SUMMARY:
Improve DGEMM performance on p690 for small
matrix sizes.

PROBLEM CONCLUSION:
DGEMM performance was improved slightly
for small matrix sizes on p690

------

APAR: PQ57481 COMPID: 5765C4200 REL: 330
ABSTRACT: IMPROVE PERFORMANCE FOR SOME LEVEL 1 BLAS ON P690

PROBLEM DESCRIPTION:
Improve performance for selected Level 1 BLAS on p690.

PROBLEM SUMMARY:
Improved performance for selected Level 1 BLAS
needed.

PROBLEM CONCLUSION:
Performance improved for codes in the
following groups: _DOT, _DOTC, _DOTU, _NORM2, _YAX, _ASUM,
_SCAL, _ROT

------

APAR: PQ57570 COMPID: 5765C4200 REL: 330
ABSTRACT: IMPROVE PERFORMANCE IN DGEMV AND DGER FOR POWER OF 2

PROBLEM DESCRIPTION:
The performance if DSPEV with power of 2 LDA running on p690
is worse with ESSL 3.3 than ESSL 3.2

PROBLEM SUMMARY:
DSPEV was shown to perform worse with power of
2 LDAs on p690 when using ESSL 3.3 than ESSL 3.2. This can
be traced to the performance of DGEMV and DGER.

PROBLEM CONCLUSION:
DGEMV and DGER were updated to improve
performance in p690 in ESSL 3.3

------

APAR: PQ57865 COMPID: 5765C4200 REL: 330
ABSTRACT: IMPROVE DTRMM PERFORMANCE FOR SPECIAL SHAPED PROBLEMS

PROBLEM DESCRIPTION:
For some non-square B matricies in DTRMM, the performance of
DTRMM is degraded. For example, when SIDE=R and M is small,
the performance could be improved.

PROBLEM SUMMARY:
DTRMM performance needs to be improved for
problems where one dimension of the rectangular matrix is
small compared to the other.

PROBLEM CONCLUSION:
DTRMM performance improved in certain
cases for non-square B matrices.

------

Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]