modem freezeup

[macubergeek]

Problem: macos x is booted up more than about 15 hours. I try to connect to the internet using either Internet Connect or the dropdown modem connect menu on upper right of screen.

Result: mouse gets jerky, modem seems to freeze up and I can't do anything else in finder.

I believe this is known bug in macos x.

Has anyone else experienced this, or know when Apple will fix this?
I'm running 10.1.2

jk

[sAMANOSKe]

This is a known bug. I heard about quitting the application might help - sometimes or using PPP-Monitor (nice freeware btw.). I also read about a log that, deleted solve the problem temporarily, but i forgot ... Systemfreeze for 2 to 10 mins really sux. Apple should fix this a year ago, because lots of users go online by Modem, do they ?! - Well i do and i HATE it!!!

[Titanium Man]

Happened to me just tonight. What a pain! Sometimes it will come out of the freeze, but tonight I had to hit the reset button. So much for "multitasking" huh? Hopefully they'll fix it in 10.2

[msteiger]

As far as I can say, it ALWAYS comes out of the freeze - one time I had to wait for more than 15 mins though. If you don't have unsaved docs that you would lose it's definitely faster to reset sometimes. (so much for the "uptime fanatics")

[macubergeek]

I have a theory that the problem exists between Macos X and some of the Apple modems in the older G-4s/G-3s
When I go into Compusa and do an uptime in terminal and find one that's been up more than 20 hours....I then try to launch internet connect and try to connect. Now I know it will throw up an error message but what I'm looking for is to see if I get the same freeze up I get on my home machine. So far I've checked it twice and not found a freezeup problem. Could you guys try this if you think of it? What I figure is that Apple is sitting back not fixing it because they are trying to push everybody into buying a new box. They aren't putting alot of resources into bugfixing that benefits the older boxen.

[sAMANOSKe]

Thiz is kinda bad. Well, i hope someone discovers the bug and fix will be available soon.

maybe apple wants to push their users to buy DSL or cable ;)

[WillyT]

I've had this problem since 10.01 or 10.02. It gets worse toward the end of the month. After the monthly crontab it gets better. So right now its a 10 min wait. Saturday it will be around 5 - 10 secs. This just has to be a problem in Internet connect or pppd or pppd's kernal extention. There just aren't that many things that will grind unix to a halt but we got stuck with one.

[mervTormel]

over at the apple boards, they've got a loooong thread discussing this and hearty attempts at solving it. it is of considered opinion that there is not a consistent work around. and, as macubergeek mentions, there are missives on the apple board that suggest these are older machines and they may be neglected in the update.

http://discussions.info.apple.com/WebX?14@158.b6A0aBSUjpe^4@.ef9dcc0/210

lamentably, -mt

[mervTormel]

people,

"use verbose logging" ?

http://docs.info.apple.com/article.html?artnum=106447

what are the results of enabling verbose logging? does it just echo the modem commands and results to the log file?

[macubergeek]

ok I've gone to my friendly local mac dealer and have verified the modem kernel bug. I checked uptime from the command line 2 days x hours, on one machine running X and 18 hours on another.
Launched internet connect, put the icon for modem connect in the menubar and launched modem connect. No problem whatsoever. All I got was an error about no dial tone (neither machine was connected to a phone line).

I'm reasonably sure this is a modem bug with G-3s and early sawtooth G-4s

According to my dealer, at this time Apple was using Global Village card modems which they had alot of issues with even under 9.

Just saw a new iMac...awesome machine, very slick, and comes with two decent harmon/kardon ball speakers.
I know what my next macine will be.:D

[WillyT]

Hey its Saturday Feb 2.

Internet connect hooked me up almost instantly.

Uptime is only 18 days but that seems to be a long time to the freezing bug.

Yesterday it took about 10 mins before it would dial out.

So what happened?

The MONTHLY crontab run.
Doing login accounting:
Rotating log files:
and a kill -HUP of everything in /var/run/syslog.pid

Guess I'll wait till the problem gets bad again and try one of these on its own.:cool:

[macubergeek]

My problem is that I run a first gen sawtooth with a Global Village modem card in it. I myself don't see any time-of-the-month variance, but I'll retest tonight and see if it's any different. Thanks for the heads up.

[WillyT]

Well that didn't last long.

Up to 3 min already. At least its not 10 min or longer.

Still totally unacceptable.:mad:

[macubergeek]

WillyT What generation machine are you using? G-3 or G4?

[WillyT]

Oh. I have a G4 400MHz AGP graphics.

10G
Macintosh HD - OS 9.0.4 - OMS MIDI
SWAP - for vm

60G
Big-Bear - OS 9.2 Classic
Cinemascope - mostly .mov and .swf
Decadent - OS 10.1.2
Fat-City - downloads
Users - symbolic linked from Decadent NO changes to Netinfo
Antares - graphic and audio

Using Fink

Modified system files
using fstab and moved mount-remaining script in /etc/rc for swapfile
/etc/hostconfig - set hostname and made a copy of localhost and renamed it in Netinfo Manager
/etc/printcap - modifications for lpr

No other CRITICAL mods that I know of.

I'm not quite sure what "sawtooth" means other than /|/|/|/| waveform?

[macubergeek]

Well it's just as I suspected, your machine is very close in age to mine. According to my local Apple reseller, Apple was using an internal modem at the time that X now has problems with. I believe it came from Global Village and it had issues. X hasn't been tweaked enough yet to accommodate it. The upcoming 10.1.3 update is supposed to fix the eternal revolving beachball cursor problem. I believe this means that they've reset the process timeout soooo that should mean that if the modem has a problem initializing, we'll be able to break out of it quicker.

[Titanium Man]

For what it's worth, I have this problem too (G4 400mghz Powerbook Titanium) but it seems to get better after a restart. Go figure. But I hope an update to OS X will fix it.

[WillyT]

Well it sounds like we will have a solution RSN.

Since monthly crontab things are much better again.

For last 4 tries:
0 sec
3 min
0 sec
0 sec

Oh and just to see if I can catch it I'm running top just before I connect.

top -u -s 300

This seems to be 30.0 sec
Hope that top is actually getting some time slices do it can work.
:cool:

[mervTormel]

that top command don't look too very good. are you trying to catch this hang process?

try:

% top -ud -s5 -l36 15 > ~/toppy.out

that'll give you a scan of [ relative deltas to previous samples ] of the top 15 cpu-eaters at the end of every five seconds for 3 minutes redirected to toppy.out for later perusal.

the last arg to top is number of processes to sample

lemme know if that's a better peek at process behavior for this problem.

[WillyT]

Merv
I took about a 20 to 30 second hit this morning but had your args running top. Here is pertinent info:

This data is one second after clock stopped at 09:31:21
Processes: 54 total, 3 running, 51 sleeping... 144 threads 09:31:22
Load Avg: 1.01, 0.66, 0.46 CPU usage: 22.5% user, 20.5% sys, 57.0% idle
Networks: 0 ipkts/0K 0 opkts /0K
Disks: 0 reads/0K 0 writes/0K
VM: 0 pageins 0 pageouts
25554 Internet C 0.0% 26:32.55 0 0/0 0/0 314/40 45
Nothing else interesting.

But top skipped for 20 seconds!
Processes: 55 total, 2 running, 53 sleeping... 144 threads 09:31:42
Load Avg: 1.85, 1.25, 0.92 CPU usage: 1.3% user, 98.7% sys, 0.1% idle
Networks: 0 ipkts/0K 0 opkts /0K
Disks: 0 reads/0K 2 writes/1K
VM: 0 pageins 0 pageouts
And here is the culprit
25554 Internet C 0.0% 26:32.56 0 0/0 13/10 64/31 11
3256 pppd 0.0% 0:19.22 0 0/0 0/0 0/0 0
Almost 20 seconds all to itself bad pppd baaaaaad pppd.

Its using almost every last cycle for itself.
Now if top is 3254 and pppd is 3256 what was 3255?

As you can see from previous post my freeze times are down FOR THE MOMENT.
Uptime is 23 days.

From my previous experience with a preemptive system (15 years with Amiga computers at home) this sort of thing only happens with runnaway (or stuck in a loop) supervisor code or by forbiding multitasking. Both were considered rude and bad practice except for all out benchmark speed tests.

This looks really scarey because of the disk writes. I don't know how critical system timing is for disk access. Also don't know if this was top or something else.

[mervTormel]

---
Now if top is 3254 and pppd is 3256 what was 3255?
---

some processes are fleeting, and only exists for a flash. 3255 could have been any number of things. at first, i thought this is a non-issue. but, what if 3255 was supposed to hang around and talk to 3256, but it died. now, 3256 frets about this for awile.

can anyone replicate and verify this process before the pppd daemon? in other words, what is the process of pppd_pid minus one, two or three?

i don't know about the runaway or loop speculation. i've seen this behavior with a process just trying to read input; it blocked and wouldn't interrupt. the fact that you see it getting cpu time means it's processor resident, it may not be doing anything particularly useful or intense.

so there's a rude bug in pppd, or in the neighborhood, and there's a loose pattern. it's probably rude to apple QA, too. i wish apple would issue a patch. but i wonder if they're even working on it (artificial retirement of older models) *sigh*

---
This looks really scarey because of the disk writes.
---

do you mean the lack of disk writes? when pppd finally interrupts and releases the cpu, the update process will awaken and flush disk cache to disk. no worries. unless you crater your machine. then it's commando-s crash-cart time.

[macubergeek]

No wonder I'm seeing modem freezeup
if pppd is hogging the processor like this, I'm thinking it has a huge memory leak maybe? But what I don't understand is why this behavior dosn't seem to happen on the new machines now out? I mean could it mean that the modem driver is simply incompatible with the older modems?

[stetner]

Guys, remember, this is unix. The bug is not with pppd, but with a driver within the kernel (or the kernel itself I suppose).

The kernel will not let a process retain the CPU for a long period of time. If a process is nice'd to -20, it will run very often, looking like it is 'hogging' the cpu, but it still has to give up the cpu to the kernel every so many ticks, while the kernel makes the decision on who runs next.

But, if I make a system call, to do something like use a device (say, a modem....), the driver code is likely running as kernel code so it can access the hardware directly. Since the kernel runs the whole show, it can do what it wants, including (if it is buggy) wedging the system.

So it may be pppd making the system call that is causing problems, but it is the device driver (or possibly the kernel itself) that is allowing the system to be wedged.

Note the 98% sys time mentioned above. That is kernel time not user time.
Also note the lack of system calls for pppd which suggests to me it may be a single system call that is holding things up, because if it got a chance to make multiple system calls, other processes would probably get a go as well.

Of course none of this helps you with your freezes, sorry 8-(

Doug

[macubergeek]

very good explaination. It was my first suspicion that the kernel driver was biggy.

[Titanium Man]

I'm not so sure it's an issue with older machines. I have a 400 mghz G4 Titanium Powerbook (not even the first issue of said machine) and this problem is driving me nuts. Oddly enough, it's only been an issue within the past month or so. I have to restart my machine every day now, because if I don't I know it'll freeze up when I try to get online that night. :mad:

[sAMANOSKe]

I dunno if this is important, but i find the system is freezing not "that" long if i pause iTumes before. Can any1 confirm this? When does apple fix this - sniff. Well in fact i am flying on cable in a few weeks, sooo :)

[WillyT]

Installed 10.1.3 yesterday(this morning)

Looks like a more recent ppp.kext and pppd but not sure they are different.

Have logged on twice with no delay.

Uptime 1/2 day so who knows.

[macubergeek]

WillyT
Would you do us a favor? Leave you computer booted up and lets see how long you can go without a modem freezeup (if it every happens)?
thanks man

[WillyT]

Had a 3-5 second freeze.

The "new" ppp.kext was from 10.1.2 and only has new .info and .plist

10.1.2 also has new modem scripts.

10.1.3 does have a new AppleKeyLargo.kext

What the heck is the KeyLargo bit. I see this set in ppp log when set to verbose but this is after it dials out.

Pacifist is just excellent for seeing what gets installed.

[WillyT]

I'm a brave soul or else foolhardy.:D

Over at MacNN there is talk of a possible solution to the PPP hang bug.

I have installed a new/different AppleSCCSerial.kext on my system using info from here (http://forums.macnn.com/cgi-bin/ultimatebb.cgi?ubb=get_topic&f=46&t=004005) .

My system hung on shutdown after doing this. I hit reset and it came back up. ASP shows it is installed. I am obviously able to connect. The new file v1.2.2 is over 6 times larger than old v1.2.1.

Will be back in a few with any hang info.

(Have to load in now for a show might get back on tommorrow but for sure by Tuesday.)

[mervTormel]

Originally posted by WillyT
...What the heck is the KeyLargo bit. I see this set in ppp log when set to verbose but this is after it dials out...

FYI, KeyLargo is the code name for the I/O controller IC and it's supporting software/firmware. it supports the modem slot, among other I/Os.

http://developer.apple.com/techpubs/hardware/Developer_Notes/Macintosh_CPUs-G4/PowerMacG4/2Architecture/KeyLargo_I_O_Controller.html

[WillyT]

I have had no problems connecting since installing the AppleSCCSerial.kext

But I started getting the spinning disk pointer every time I tried to do much at all. Then I looked at the date on my vm directory. It was 5 days ago from when I installed 10.1.3. But I had shutdown and restarted just a day and a half ago.

df showed SWAP 1 mounted instead of SWAP. So I got bit by my fstab file.:rolleyes:

Restarted and all seems ok now. No spinning disk when launching etc.

I know I should have checked right after the restart. I usually do and there have been plenty of warnings about using fstab. But I was extremely pressed for time this weekend.(Sound system is loaded back on the buss ready for my next job)

[WillyT]

I have had no delays on connecting since I installed the new AppleSCCSerial.kext.

Uptime 2 days 16 hours.

On the thread at MacNN they claimed this is actually from Apple.
The pbdevelopment.plist looks like it is from a sandboxed environment.

[WillyT]

Since installing the AppleSCCSerial.kext I have not had ANY lockup type delays.

With system maintenance and POWER FAILURES I'm not getting that much solid uptime.

[darwin:~] williamt% uptime
6:47PM up 5 days, 23 hrs, 1 user, load averages: 0.92, 1.00, 0.87


With 10.1.3 and the .kext and crash reporter though I am getting nasty messages and a few unexpected quits(only when guitting apps) these all have :

Exception: EXC_BAD_ACCESS (0x0001)
Codes: KERN_INVALID_ADDRESS (0x0001) at 0x026609c8

I have no idea if this is caused by10.1.3 or the .kext .

I will happily live with this if I don't ever get the long waits when connecting.

[macubergeek]

Well since installing the new kext I have an uptime of:
7:53AM up 15 days, 15:58, 7 users, load averages: 0.75, 0.41, 0.37

The only downside is now my FaxSTF software is broken:mad

[WillyT]

Looks like its official now and much trimmer than the prerelease version.