| Author |
VSI FreeAXP Student kit and MONITOR display problem |
VolkerHalle
Member
Posts: 104
Location: Germany
Joined: 02.04.10 |
| Posted on March 05 2021 22:41 |
|
|
As first reported in comp.os.vms
https://groups.google.com/g/comp.os.vms/c/0FGKGIX83Js
there is a problem with displaying MONITOR data in OpenVMS Alpha V8.4-2L2 when running in FreeAXP 4.0.0.646 as delivered with the VSI Student Kit 2020.2. After reading the problem description, I decided to try to quickly reproduce this problem by downloading and installating the 2020.2 student kit and yes, this problem is reproducable !
Avanti Virtual Alpha version 4.0.0.646 (Apr 14 2020 14:36:57)
$ sho sys/noproc
OpenVMS V8.4-2L2 on node STUDNT 6-MAR-2021 09:51:41.17 Uptime 0 00:07:03
$
Directory SYS$COMMON:[SYSEXE]
MONITOR.EXE;1 22-MAY-2017 22:50:19.68
Image Identification Information
image name: "MONITOR"
image file identification: "X-27"
image file build identification: "XE6K-0100140000"
link date/time: 22-MAY-2017 22:50:19.45
linker identification: "A13-04"
OpenVMS Monitor Utility
SYSTEM STATISTICS
on node STUDNT
6-MAR-2021 09:47:25.05
CUR AVE MIN MAX
Interrupt State 0.49 0.39 0.16 0.66
MP Synchronization 0.00 0.00 0.00 0.00
Kernel Mode 0.16 0.06 0.00 0.16
Executive Mode 0.00 0.00 0.00 0.00
Supervisor Mode 0.00 0.00 0.00 0.00
User Mode 0.33 0.19 0.00 0.33
Compatibility Mode 0.00 0.00 0.00 0.00
Idle Time *******.** 99.34 *******.** *******.**
Process Count 10.00 10.00 10.00 10.00
Page Fault Rate 0.00 0.29 0.00 0.83
Page Read I/O Rate 0.00 0.03 0.00 0.16
Free List Size 9555.00 9555.60 9555.00 9558.00
Modified List Size 447.00 447.19 447.00 448.00
Direct I/O Rate 0.00 0.03 0.00 0.16
Buffered I/O Rate 0.49 0.53 0.33 0.83
OpenVMS Monitor Utility
DISK I/O STATISTICS
on node STUDNT
6-MAR-2021 09:49:53.72
I/O Operation Rate CUR AVE MIN MAX
STUDNT$DKA0: STUDSYS *****.** 100.65 *******.** *******.**
As reported in comp.os.vms, this problem does NOT occur, if running the same virtual disk in CHARON-AXP or on a real ES47.
I can make the associated FreeAXP Log file (OpenVMS Studentpackacke.log) available, if required.
Volker. |
 |
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
Bruce Claremont
Moderator
Posts: 623
Joined: 07.01.10 |
| Posted on March 06 2021 04:54 |
|
|
I've reproduced the problem using FreeAXP 653 and OpenVMS 8.4-2L2.
The problem does not occur under VMS 8.3 or 8.4.
We'll have a look at the issue. Thanks for the bug report.
Edited by Bruce Claremont on March 06 2021 05:02 |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
VolkerHalle
Member
Posts: 104
Location: Germany
Joined: 02.04.10 |
| Posted on March 06 2021 07:37 |
|
|
Bruce,
did you try MONITOR/RECORD SYSTEM/ALL under V8.4 and then just display that MONITOR.DAT file on V8.4-2L2 using MONITOR/INPUT=MONITOR.DAT SYSTEM/ALL - still the asterisks for Idle Time ? This would rule out any MONITOR data-collection problems.
Note that MONITOR has been changed in V8.4-2L1:
3.1.12 MONITOR Utility: MONITOR IO FREE LIST SIZE field overflow on large memory systems
V8.4-2L2 MONITOR SYSTEM/ALL has been reported to work o.k. on a real ES47, but the above 'fix' was apparently to the same MONITOR class display, so it may be related.
Idle Time of *******.** for CUR, MIN and MAX is completely bogus, AVG 99.34 - in my case - looks good.
Volker. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
VolkerHalle
Member
Posts: 104
Location: Germany
Joined: 02.04.10 |
| Posted on March 06 2021 20:48 |
|
|
Bruce,
further testing results: running MONITOR ALL/ALL shows the asterisks problem on different displays (IO,PAGE,DISK,TIMER), but it depends on load, i.e. whether the default values of 0.00 get overwritten by the measurement data. The asterisks may show up as '0.**' or '*******.**'
MONITOR TIMER is really showing lots of asterisks, as most of the data measured/shown are nearly always non-zero.
Now a little bit of speculation about the possible underlying problem:
- OpenVMS Alpha V8.4-2L2 has been completely rebuilt/recompiled for EV6 support
- the problem does not show up on a real ES47 and a CHARON-AXP/DS20, both EV6 machines
- FreeAXP is emulating an AlphaServer 400, an EV4 CPU - 'enhanced' for EV6 support
- now if the newly built OpenVMS Alpha V8.4-2L2 and the new MONITOR include some
instructions beyond EV4, there is some risk of seeing a possible 'bug' in instruction emulation
Volker. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
Bruce Claremont
Moderator
Posts: 623
Joined: 07.01.10 |
| Posted on March 07 2021 03:04 |
|
|
|
Thanks Volker! |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
VolkerHalle
Member
Posts: 104
Location: Germany
Joined: 02.04.10 |
| Posted on March 09 2021 04:20 |
|
|
Some troubleshooting results using MONITOR/RECORD TIMER:
When displaying the data, MONITOR uses the absolute counter values PMS$GQ_TQE* from the system (or from the MONITOR.DAT file), builds the differences from the values collected/recorded in the previous interval, then converts the differences to floating-point values and divides it through the collection interval length, building rate/second floating values.
These floating values are then converted to the whole part and the fractional part and stored as integers. These integer values are then output using $FAOL_S and control strings containing multiple "!7UL.!2UL"
So 'the problem' seems to be somewhere in those conversion code pathes (see MONITOR.LIS routine COMPUTE_STATS).
You would probably need the current MONITOR source listing to have a chance to further debug this. A MONITOR.EXE built with /DEBUG would certainly help here...
Volker. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
Bruce Claremont
Moderator
Posts: 623
Joined: 07.01.10 |
| Posted on March 10 2021 03:36 |
|
|
|
Thanks Volker. We'll see if VSI will assistance. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
VolkerHalle
Member
Posts: 104
Location: Germany
Joined: 02.04.10 |
| Posted on March 15 2021 22:33 |
|
|
The problem is absolutley reproducable on the new VSI 2021.1 Student Kit (OpenVMS Alpha V8.4-2L2) containing FreeAXP 4.0.0.646
When running MONITOR.EXE from OpenVMS Alpha V8.2 or E8.4 on this installation, the problem does NOT show up.
So the problem clearly is being triggered from 'some specific instruction stream' inside the MONITOR.EXE image from OpenVMS Alpha V8.4-2L2 on FreeAXP !
After creating a forced process dump from a running MONITOR TIMER operation and dumping the complete code section of the resulting MONITOR.DMP file (using SDA>exa/ins 40000:70ec0+30) , it could be shown, that the V8.4-2L2 MONITOR.EXE image uses some EV6-specific instructions:
$ sea monitor_inst.lis itof ! does V8.4-2L2 monitor use new EV6 instructions ?
MONITOR+3C128: ITOFT R25,F1
MONITOR+3C130: ITOFT R16,F0
...
$ sea monitor_inst.lis ftoi
MONITOR+3CC68: FTOIT F0,R23
MONITOR+3DEB4: FTOIS F10,R19
...
So the correct implementation of the emulation of these instructions should probably be closely examined first
Volker. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
VolkerHalle
Member
Posts: 104
Location: Germany
Joined: 02.04.10 |
| Posted on March 16 2021 02:37 |
|
|
Looks like the error is in the emulation of the FTOIS instruction
What I've done to diagnose this:
- patch the image header of a copy of MONITOR.EXE to set the EIHD$V_DEBUG flag and add/ modify the transfer addresses to first invoke $IMGSTA and then the real transfer address
- $ define/SYS/EXE monitor Dev:<dir>copy-of-monitor
- I tried DBG> set br/ins=(itoft,ftoit) and it failed:
%DEBUG-E-INTERR, debugger error in DBG$PARSE_INSTRUCTION - NYI for Alpha or session corruption
- so I set breakpoints on ALL occasions of the new EV6 instructions ITOF* and FTOI*
- run patched MONITOR with debug and a little MONITOR.DAT input file created by 3 cycles of MONITOR/RECORD TIMER - this guarantees the SAME input being processed every time !
- only 3 of all the above breakpoints were really being hit
- created a small DEBUG .COM file to examine the registers used on those 3 EV6 instructions BEFORE and AFTER the execution, the important one is this:
SET BR 10000+4A218 DO (EXA/INS @PC; EXA/S_FLOAT F10; EXA/HEX F10; STEP/INS ; EXA R26; STEP/INS; EXA/T_FLOAT F11; EXA/HEX F11)
- compared the BEFORE and AFTER values when executing these instructions on the FreeAXP and a CHARON-AXP/DS20 (sorry, I have no real EV6 machine).
... and finally found the problem:
$ type monitor_debug_ds20.log
!SHARE$MONITOR+4A218: FTOIS F10,R26
g
!break at SHARE$MONITOR+4A218
!SHARE$MONITOR+4A218: FTOIS F10,R26
!0\%F10: 0.0000000E+00
!0\%F10: 0000000800000000 <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
!stepped to SHARE$MONITOR+4A21C: ITOFT R26,F11
!0\%R26: 0000000000000040 <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
!stepped to SHARE$MONITOR+4A220: STL R26,(R18)
!0\%F11: 0.316202013338398E-321
!0\%F11: 0000000000000040
SYS$SYSROOT:[SYSMGR]MONITOR_DEBUG_FREEAXP.LOG;2
!SHARE$MONITOR+4A218: FTOIS F10,R26
g
!break at SHARE$MONITOR+4A218
!SHARE$MONITOR+4A218: FTOIS F10,R26
!0\%F10: 0.0000000E+00
!0\%F10: 0000000800000000 <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
!stepped to SHARE$MONITOR+4A21C: ITOFT R26,F11
!0\%R26: 0000000000000000 <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
!stepped to SHARE$MONITOR+4A220: STL R26,(R18)
!0\%F11: 0.000000000000000
!0\%F11: 0000000000000000
The FTOIS F10, R26 instruction delivers a different output value of R26 for the same input value of F10 when run on FreeAXP. This is worth further analysis...
Volker. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
Bruce Claremont
Moderator
Posts: 623
Joined: 07.01.10 |
| Posted on March 16 2021 05:10 |
|
|
Thanks Volker. You'll have it solved soon .
Camiel had been examining the new/modified instructions in VMS 8.4-2L2. As you've shown, the problem lies there.
The bad news is that for people running on real pre-EV6 hardware, VMS 8.4 is likely their last fully supported release. |
|
|
| Author |
RE: VSI FreeAXP Student kit and MONITOR display problem |
Bruce Claremont
Moderator
Posts: 623
Joined: 07.01.10 |
| Posted on March 22 2021 05:09 |
|
|
|
We've developed a fix, which will be available in the next release. No release date as of yet. |
|
|
