SM & 100 CPU load w/ PIII CPUs (not related to mail)

Post by A.D. Fundum
Which fully copyable directories, assuming an identical relevant setup
and excluding the fill profile directories of Mozilla, Firefox and
SeaMonkey, do contain files related to SM plugins?

Plugins or add-ons? Plugins are binary and have been around forever
though getting phased out.

Post by A.D. Fundum
It's possible that the problem is introduced by a crash during the
installation of plugins (XPI packages only), so perhaps a file not
included in the profile(s) is missing or damaged. It looks like just
reinstalling won't help. Plugins were installed, but the problem still
isn't gone.

OK, add-ons. They should all install into extensions\ either in your
profile or program directory. They can create their own directories as well.
Have you tried safe-mode? eg seamonkey.exe -safe-mode or help-->restart
with addons disabled.

Post by A.D. Fundum
This is ye olde problem where the system is perfectly responsive, but
any SM-step (like pressing <ALT-X> to open a menu) may take several
minutes.

Do have an instance of plugincontainer.exe running?
Dave

A.D. Fundum

2015-12-21 19:34:17 UTC

Post by Dave Yeo
OK, add-ons. They should all install into extensions\ either
in your profile or program directory.

Sorry, add-ons. Both x:\HOME\DEFAULT\MOZILLA\EXTENSIONS and
x:\HOME\DEFAULT\MOZILLA\EXTENSIONS\{<hex-numb-er-s>} are
empty directories. I've copies the profiles of Mozilla, FF and SM (but
not
Netscape's, which is used for local browsing).

AFAICT there's no other directory called EXTENSIONS. I've used ATTRIB
* to make sure that there were no hidden files.

Post by Dave Yeo
Have you tried safe-mode?

Been there, done that. Now I've tried it again after copying all
Mozilla profiles. The CPU load still is 100% before the initial
--safe-mode screen is displayed after about one minute, and the CPU
load remains 100%.

Post by Dave Yeo
Do have an instance of plugincontainer.exe running?

Hopefully not. The good news is that the machines are "mirrors".
Typically I just have to start SM to apply an individual new setting
or to update an add-on, and that's it. I don't have to do that, but
then the mirrors still cannot be used to browse.

During the last installation I started Netscape, Mozilla, SM 1 and SM
2. Maybe SM 1 isn't needed, but I'm using it to make sure that each
browser can inherit an older profile. This worked. The performance of
SM 2 was the expected performance. During the installation of add-ons
an error occured. IIRC I installed two add-ons before restarting SM 2,
but the installation of the second add-on caused an <ALT-F4>-like
crash. That triggered the CPU load again. It's 100% ASAP after SM is
started, stays 100%, and only affects SM. After SM is closed it will
remain 100% for a short while.

I can copy Netscape's profile too, but was hoping for e.g. a corrupted
file in a non-profile directory (profile-directory: the one in
profiles.ini).

--

Dave Yeo

2015-12-22 04:52:17 UTC

A.D. Fundum wrote:
...

Post by A.D. Fundum
Hopefully not. The good news is that the machines are "mirrors".
Typically I just have to start SM to apply an individual new setting
or to update an add-on, and that's it. I don't have to do that, but
then the mirrors still cannot be used to browse.
During the last installation I started Netscape, Mozilla, SM 1 and SM
2. Maybe SM 1 isn't needed, but I'm using it to make sure that each
browser can inherit an older profile. This worked. The performance of
SM 2 was the expected performance. During the installation of add-ons
an error occured. IIRC I installed two add-ons before restarting SM 2,
but the installation of the second add-on caused an <ALT-F4>-like
crash. That triggered the CPU load again. It's 100% ASAP after SM is
started, stays 100%, and only affects SM. After SM is closed it will
remain 100% for a short while.
I can copy Netscape's profile too, but was hoping for e.g. a corrupted
file in a non-profile directory (profile-directory: the one in
profiles.ini).

OK, lets see if I understand you. You're migrating a profile from NS
4.61 to SM current by doing NS-->Mozilla-->SM1-->SM2-->SMcurrent. You
should actually use SM2.1 instead of SM2 and shouldn't need the Mozilla
and SM1 steps.
Afterwards you install 2 add-ons, get a crash and then 100%CPU, but on
only one of 2 identical machines.
You have tested safe-mode and it didn't help and you have not tested a
new profile for comparison.
You don't have an extensions directory anywhere even after installing
add-ons.
Does about:addons show any installed? Have you tried moving
addons.sqlite out of your profile?
Dave

A.D. Fundum

2015-12-22 10:50:35 UTC

You're migrating a profile from NS 4.61 to SM current by doing
NS-->Mozilla-->SM1-->SM2-->SMcurrent.

SM2 = SMcurrent, SM1 = older version of SM which should
unconditionally be able to e.g. import old NS mails. After the problem
occurs FF will be affected too, so it isn't SM-specific.

This was the historical natural order of use (without problems), so
it's just an attempt to avoid the error condition during an install of
the OS. Now it seems to be relared to add-ons. Earlier it looked like
it was related to using SM before Mozilla/NS.

You should actually use SM2.1 instead of SM2 and shouldn't need
the Mozilla and SM1 steps.

Thanks, so we can ignore all of the steps above. I've never reproduced
the problem without extra steps, but there's no reason to not agree
with you.

Afterwards you install 2 add-ons, get a crash and then 100%CPU

Yes, at least the last (and best "recorded") time.

Sometimes FF/SM has to be restarted to activate/use a new add-on, and
sometimes not. I should have restarted FF/SM after each install, but I
didn't. SM, still fast, crashed during the install of the second
add-on, before it was restarted.

but on only one of 2 identical machines.

Digital mirrors aren't identical. You can e.g. visit the /pub/incoming
directory of a Hobbes mirror to find wrongly uploaded files. The
operator of the NMSU may have deleted uploaded files, but the mirror
didn't. Setups should be identical.

You have tested safe-mode and it didn't help and you have not
tested a new profile for comparison.

Yes, no. I'm using "DELTREE" to delete profiles, and I've used both a
new profile and (at the moment) a copied profile. If a nachine wasn't
affected, the the copied profile works as expected. If a machine was
affected, then it will still be affected.

So the profile files shouldn't play a role, unless there's code which
changes the profile first, even before the --safe-mode dialog is
displayed.

When I'm talking about a copied working profile, then I'm talking
about everything under a "<passwordy>.default"-directory.

You don't have an extensions directory anywhere even after
installing add-ons.

Not anywhere.

x:\HOME\DEFAULT\Mozilla\Extensions

x:\HOME\DEFAULT\Mozilla\Firefox\Profiles\3ozcttvn.default\extensions

x:\HOME\DEFAULT\Mozilla\SeaMonkey\Profiles\1hntxk0z.default\extensions

x:\HOME\DEFAULT\Mozilla\SeaMonkey\Profiles\1hnmxy0z.default\extensions
\inspect
***@mozilla.org\chrome\inspector\content\inspector\extensions

y:\Apps\Firefox\browser\extensions

y:\Apps\SeaMonkey\extensions

y:\Apps\SeaMonkey\distribution\extensions

(working directory structure, but that shouldn't matter, I've manually
compared the first, identical directory and all other entries should
be mirrors)

The first directory is empty, an empty directory excluded. Numbers 2-4
belong to a copied, working profile. Installing SM again never helped
(entries 5-7).

Should I try to (fully) copy a working profile and to reinstall the
distributed SM-package before starting SM?

Does about:addons show any installed?

AFAICT everything works, e.g. the language pack. I'm trying
about:addons now...

--safe-mode first: after 10 seconds (IBM ThinkPad X20) the CPU-load
becomes 100%. After (a less meaningful, due to the slowlyness) 70
seconds the Safe Mode-dialog is displayed. Exit (dialog): CPU load
remains 100% during 5 seconds.

Unsafe again, about:addons and after several minutes: everything is
installed, including a dictionary and the langpack.

Have you tried moving addons.sqlite out of your profile?

I have now. No change, as expected. FF, a different profile, is
affected too. It looks like a generic file or setting is involved.

Or perhaps one generation of notebook CPUs, combined with add-ons?
Some mirrors:

IBM ThinkPad 600E (Pentium II): works
IBM 300 PL (Pentium III, desktop): works
IBM ThinkPad 600X (Pentium III): slow
IBM ThinkPad X20 (Pentium III): slow
IBM ThinkPad T23 (Pentium III-M): works
IBM ThinkCentre A30 (Pentium IV, desktop): works
IBM ThinkPad T42p (Pentium IV): works

X20 (slow) & your CPUInfo utility, FWIW:

vendor_id GenuineIntel, cpu family 6, model 8, stepping 6, flags fpu
vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx
fxsr sse, cpu MHz 609.663, model name Intel(R) Pentium(R) III
processor.

BTW, it's hard to actually reproduce the problem because a full new
install of the OS may be required. Just reinstalling SM won't help,
AFAICT. Nor does using a working profile. I didn't record the add-on
which crashed nor the order of add-ons before the install was
affected. I didn't change CONFIG.SYS nor plugins. The virtual page 803
of my install manual just reads "install all FF and SM add-ons".

--

Dave Yeo

2015-12-22 23:32:57 UTC

Post by Dave Yeo
You have tested safe-mode and it didn't help and you have not
tested a new profile for comparison.

Yes, no. I'm using "DELTREE" to delete profiles, and I've used both a
new profile and (at the moment) a copied profile. If a nachine wasn't
affected, the the copied profile works as expected. If a machine was
affected, then it will still be affected.
So the profile files shouldn't play a role, unless there's code which
changes the profile first, even before the --safe-mode dialog is
displayed.

I think there is code that checks
%MOZILLA_HOME%\Mozilla\SeaMonkey\profiles.ini very early, probably
before the safe-mode dialog. You might want to compare these, they're
pretty simple test files.

Post by A.D. Fundum
When I'm talking about a copied working profile, then I'm talking
about everything under a "<passwordy>.default"-directory.

Post by Dave Yeo
You don't have an extensions directory anywhere even after
installing add-ons.

Not anywhere.
x:\HOME\DEFAULT\Mozilla\Extensions
x:\HOME\DEFAULT\Mozilla\Firefox\Profiles\3ozcttvn.default\extensions
x:\HOME\DEFAULT\Mozilla\SeaMonkey\Profiles\1hntxk0z.default\extensions
x:\HOME\DEFAULT\Mozilla\SeaMonkey\Profiles\1hnmxy0z.default\extensions
\inspect
y:\Apps\Firefox\browser\extensions
y:\Apps\SeaMonkey\extensions
y:\Apps\SeaMonkey\distribution\extensions
(working directory structure, but that shouldn't matter, I've manually
compared the first, identical directory and all other entries should
be mirrors)
The first directory is empty, an empty directory excluded. Numbers 2-4
belong to a copied, working profile. Installing SM again never helped
(entries 5-7).
Should I try to (fully) copy a working profile and to reinstall the
distributed SM-package before starting SM?

Might be worth testing.

Post by Dave Yeo
Doesabout:addons show any installed?

AFAICT everything works, e.g. the language pack. I'm trying
about:addons now...
--safe-mode first: after 10 seconds (IBM ThinkPad X20) the CPU-load
becomes 100%. After (a less meaningful, due to the slowlyness) 70
seconds the Safe Mode-dialog is displayed. Exit (dialog): CPU load
remains 100% during 5 seconds.
Unsafe again,about:addons and after several minutes: everything is
installed, including a dictionary and the langpack.

Post by Dave Yeo
Have you tried moving addons.sqlite out of your profile?

I have now. No change, as expected. FF, a different profile, is
affected too. It looks like a generic file or setting is involved.

This is using run! or libpathstrict+beginlibpath?

Post by A.D. Fundum
Or perhaps one generation of notebook CPUs, combined with add-ons?
IBM ThinkPad 600E (Pentium II): works
IBM 300 PL (Pentium III, desktop): works
IBM ThinkPad 600X (Pentium III): slow
IBM ThinkPad X20 (Pentium III): slow
IBM ThinkPad T23 (Pentium III-M): works
IBM ThinkCentre A30 (Pentium IV, desktop): works
IBM ThinkPad T42p (Pentium IV): works
vendor_id GenuineIntel, cpu family 6, model 8, stepping 6, flags fpu
vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx
fxsr sse, cpu MHz 609.663, model name Intel(R) Pentium(R) III
processor.

Does seem to be specific to your old CPU's which have in common no SSE2
though not consistent. My oldest is a T42 with a Pentium M and it works
well.
I doubt that Mozilla tests such old CPUs and it could be related,
perhaps a subtle bug in the JavaScript JIT (Just in time compiler).
Unluckily the options to disable it are now gone. There are other places
that use SSE2 but they should have a fallback and I'd expect a simple
crash with the trp report code pointing to an xmms register.
Perhaps it is a subtle CPU or chipset bug.

Post by A.D. Fundum
BTW, it's hard to actually reproduce the problem because a full new
install of the OS may be required. Just reinstalling SM won't help,
AFAICT. Nor does using a working profile. I didn't record the add-on
which crashed nor the order of add-ons before the install was
affected. I didn't change CONFIG.SYS nor plugins. The virtual page 803
of my install manual just reads "install all FF and SM add-ons".

All I can say is weird,
Dave

A.D. Fundum

2015-12-23 04:24:06 UTC

Post by Dave Yeo
I think there is code that checks
%MOZILLA_HOME%\Mozilla\SeaMonkey\profiles.ini very early,
probably before the safe-mode dialog. You might want to compare
these, they're pretty simple test files.

One missing line (and an expected different directory name). Adding
"Default=1" didn't help. BTW, there's only one profile/user.

Post by A.D. Fundum
Should I try to (fully) copy a working profile and to reinstall the
distributed SM-package before starting SM?

Might be worth testing.

Testing... DELTREE (RM * -R) actual profile, restore working profile,
DELTREE SM dir, reinstall SM, start SM, expected result: bad luck...
Testing... Testing... --safe-mode, just to be sure... Bad luck, same
pattern.

Post by A.D. Fundum
I have now. No change, as expected. FF, a different profile, is
affected too. It looks like a generic file or setting is involved.

This is using run! or libpathstrict+beginlibpath?

No, and no changed enviroment during the (serial) install procedure of
several XPI files. FF and SM were fast, a relevant crashed occured
while installing a XPI file with FF xor SM, and then both FF are SM
are slow ("forever").

Post by A.D. Fundum
IBM ThinkPad 600E (Pentium II): works
IBM 300 PL (Pentium III, desktop): works
IBM ThinkPad 600X (Pentium III): slow
IBM ThinkPad X20 (Pentium III): slow
IBM ThinkPad T23 (Pentium III-M): works

Does seem to be specific to your old CPU's which have in
common no SSE2 though not consistent.

Yes, a forced consistency is the 500-600 MHz'ish Pentium III notebook
CPU, and restricted to one or more add-ons (no add-ons, no problem).

I've tried to produce a dump of the CPU-load graph to be able to
guesstimate where it reaches 100% and the --safe-mode dialog is being
displayed, but it's too slow. It misses the start.

--safe-mode again, I've created a dump well before the dialog would be
displayed. The horizon time scale is a 600 MHz CPU with a 100 MHz bus
speed::

Loading Image...

(0.3 KiB)

You'll know what a 100% CPU looks like, but of course it's about the
bumps before the steep cliff.

Post by Dave Yeo
Perhaps it is a subtle CPU or chipset bug.

Next time, if any, I'll try to avoid the crash by restarting SM after
each XPI-step. If the problem still occurs, then I may publish
possibly recorded steps.

I'm pretty sure that I can reproduce the problem, but as mentioned
earlier a restore seems to require a full reinstall. I've updated FF,
SM and add-on, without an accidental improvement.

My number of newer machines is limited, I'm restricted to (single
core) eCS 1.x due to NLV preferences, so a T42p is my newest one, but
I never had to re-install any other machine because of this problem.
So the CPU-theory may still be valid. I was hoping for some general
damaged file or setting, mainly because FF is affected too.

FWIW, I can decide/try to remove add-ons, but that's no mirror vs. a
broken mirror.

--

Dave Yeo

2015-12-23 18:49:33 UTC

Post by Dave Yeo
This is using run! or libpathstrict+beginlibpath?

Does seem to be specific to your old CPU's which have in
common no SSE2 though not consistent.

Yes, a forced consistency is the 500-600 MHz'ish Pentium III notebook
CPU, and restricted to one or more add-ons (no add-ons, no problem).
I've tried to produce a dump of the CPU-load graph to be able to
guesstimate where it reaches 100% and the --safe-mode dialog is being
displayed, but it's too slow. It misses the start.
--safe-mode again, I've created a dump well before the dialog would be
displayed. The horizon time scale is a 600 MHz CPU with a 100 MHz bus
http://home.uni-one.nl/m1/cpuload.png (0.3 KiB)

I see the IRQ part also goes up. I guess another possibility is a device
driver getting into a bad state. What video driver are you using?

Post by A.D. Fundum
You'll know what a 100% CPU looks like, but of course it's about the
bumps before the steep cliff.

Post by A.D. Fundum
Perhaps it is a subtle CPU or chipset bug.

Just reading on another list a discussion about whether Intels
documentation is buggy or an instruction (cvtpi2ps, maybe shifts into
MMX mode if I understand correctly) subtly changed. Of note is that the
developer considered a C2D to be ancient and had a hard time finding one
to test on, and of course that was the oldest that was tested.
It's a problem with such old machines, the software authours never test
as they always have the fastest machine they can acquire.

Post by A.D. Fundum
Next time, if any, I'll try to avoid the crash by restarting SM after
each XPI-step. If the problem still occurs, then I may publish
possibly recorded steps.

Be interesting.

Post by A.D. Fundum
I'm pretty sure that I can reproduce the problem, but as mentioned
earlier a restore seems to require a full reinstall. I've updated FF,
SM and add-on, without an accidental improvement.
My number of newer machines is limited, I'm restricted to (single
core) eCS 1.x due to NLV preferences, so a T42p is my newest one, but
I never had to re-install any other machine because of this problem.
So the CPU-theory may still be valid. I was hoping for some general
damaged file or setting, mainly because FF is affected too.
FWIW, I can decide/try to remove add-ons, but that's no mirror vs. a
broken mirror.

Dave

A.D. Fundum

2015-12-31 08:39:50 UTC

Post by Dave Yeo
I see the IRQ part also goes up. I guess another possibility is a
device driver getting into a bad state. What video driver are
you using?

IBM drivers and eCS 1.2's SNAP (works), and eCS 1.2's SNAP (fails).
According to the IRQ monitor IRQ14 (nearly always) and IRQ 11
(blinking) are active, as far as I can see, while starting FF. With SM
IRQ14 (IDE 1) blinks more often too.

As soon as the initial screen is being displaying (without being able
to use it just yet) IRQ14 is still quite active. Technically there's
enough RAM, but of course 320 MiB won't stop swapping. Nevertheless
the system remains responsive, so I can continue (most of) the other
work while SM is starting without noticing any typical Windows'ish
delays. The 600X (576 MiB? of RAM) isn't significantly faster.

The hardware (IBM ThinkPad X20) is a basic machine, so no docking
station with unusual hardware and/or strange USB devices, and so on.
MiniPCI WLAN is in use, but the other PIII (600X) has no network
adapter installed at ll.

Post by Dave Yeo
It's a problem with such old machines

Or eCS 2 (no multicore NLV for me, yet), and of course the efficient
footprint (well, playing videos excluded, any PIII should struggle
nowadays) of OS/2.

Post by Dave Yeo
the software authours never test as they always have the
fastest machine they can acquire.

Fully understood, hence the hope that a shared file was "broken". In
the past I had to use even older hardware to perform a SM Warp 4 font
test for Peter W.

FWIW: FF for WinXP is installed too (X20 only), but WinXP
(w/viruscanner) is far too slow to actually use it. I could install
about the same add-ons, but it's likely that they'll tell us to buy a
new computer. Or an old computer. I probably cannot install the same
add-ons, because FF for Windows doesn't require a langpack because its
default isn't our EN-US.

Next time I'll record what I'm doing, if I ever have to fully
reinstall eCS again. A generic CPU instruction set change is a likely
cause, and apparently there's not some broken, shared file. SM's
--safe-mode could have worked because add-ons seem to be involved, but
it doesn't. I've only experienced and reproduced it with 500-600 Intel
Pentium III notebook CPUs, but e.g. a Pentium III-Mobile CPU (IBM
ThinkPad T23) works as expected.

Thanks, and a HNY to all of you.

--

A.D. Fundum

2016-01-15 17:14:42 UTC

a Pentium III-Mobile CPU (IBM ThinkPad T23) works
as expected.

No, it doesn't. I wanted to upgrade NoScript to 2.9.0.2, but both FF
and SM were affected. I had used FF to visit a few innocent sites, but
the main change were recent upgrades of NoScript. XWLAN couldn't find
the WLAN adapter, but that's probably irrelevant. I'm using already
XPI files for upgrades, so a connection isn't required. May the second
last upgrade (to 2.9?, the version before 2.9.0.2) was fatal.

So now all of my Pentium III notebooks are affected, range 500 MHz -
1.2 GHz.

--

Dave Yeo

2016-01-16 06:53:06 UTC

a Pentium III-Mobile CPU (IBM ThinkPad T23) works
as expected.

No, it doesn't. I wanted to upgrade NoScript to 2.9.0.2, but both FF
and SM were affected. I had used FF to visit a few innocent sites, but
the main change were recent upgrades of NoScript. XWLAN couldn't find
the WLAN adapter, but that's probably irrelevant. I'm using already
XPI files for upgrades, so a connection isn't required. May the second
last upgrade (to 2.9?, the version before 2.9.0.2) was fatal.
So now all of my Pentium III notebooks are affected, range 500 MHz -
1.2 GHz.

If no-script triggered it, it's probably the JIT (JavaScript Just In
Time compiler) and a problem in the upstream code, perhaps triggering an
OS/2 bug. I doubt very much that they test on older CPUs. The JIT
translates JavaScript into machine code and is smart enough (usually) to
target your CPU producing close to the most efficient assembly code.
Complex and therefore buggy.
Unluckily PIII's are getting obsolete, even though they were a descent CPU.
Dave

A.D. Fundum

2016-01-16 12:36:22 UTC

Post by Dave Yeo
If no-script triggered it, it's probably the JIT (JavaScript Just
In Time compiler) and a problem in the upstream code,
perhaps triggering an OS/2 bug.

During the same process (of updating the machine, both in- and
excluding the use of FF and SM to import the XPI file) Java (eCS'
1.4.2) caused fatal TRAPs, while the same Java app worked on all other
machines. It seemed unrelated, and it probably is unrelated, but I'll
mention it now.

Post by Dave Yeo
Complex and therefore buggy.

So far my IBM 300PL Pentium III 500 MHz dekstop isn't affected, but
this machine is an older backup machine, without NoScipt, instead of a
mirror. This implies that the Java JIT Pentium III optimization theory
is still valid. All of my Pentium IIIs in active use are now affected,
including the last exception. Perhaps the theory is supported by the
fact that the affected system remains responsive, as if one FF/SM
thread is executing a gazzilion NOPs before it will process/update the
next item.

Next time I'll try to remove NoScript, because any theoretical time
gained by avoiding commercial scripts is far less than the time
required to do anything.

In the future a Pentium II and Pentium 4 can be affected too, but in
general a faster Pentium III was the best of both worlds: acceptable
browsing, and e.g. native DOS sound (DOSBox can be too slow and/or too
buggy). In the past one could recommend a popular IBM ThinkPad T23
(preferably with MiniPCI WLAN), but that's my last affected machine so
far. If e.g. a removed NoScript doesn't slow browsing down anymore,
then common agressive commercial scripts will.

--

A.D. Fundum

2016-01-17 10:13:03 UTC