[lkml]   [1998]   [Apr]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [SMP patch] io-apic-patch-2.1.97-A
    On Sun, Apr 19, 1998 at 12:28:32PM -0500, Robert HYATT wrote:
    > On Sun, 19 Apr 1998, Bill Broadhurst wrote:
    > >
    > > 1. Processes mysteriously die during a long (more than 2 hour)
    > > comple. The dead process can't be killed and shows 'D' in
    > > ps. If that process happens to be linked to a device, that
    > > device is continually 'busy' and only a reboot can free it.
    > > Any other process that touches the 'busy' device also dies.
    > > If this device is the root file system, the system processes
    > > will also (eventually) die, leaving the system in a "hung"
    > > state. No net access is possible and the Magic SysRq keys
    > > work (sorta) most of the time. <Magic-boot usually does.>
    > > This happens 100% of the time during a long modeling program or
    > > when re-compiling the entire X package.
    > >
    > > This symptom started at 2.1.85 and continues. It happens sooner on
    > > kernels and later on others. It happens on all my systems. I'll list
    > > them later.
    > you should check the kerneld logfile, to see if you see any sort of
    > message like "eth0: TX timed out..." I get this and my machine hangs,
    > but not "hard". No net traffic, can't start new processes nor exit
    > old ones, but somethings "sorta" work for a bit...

    No Kerneld logfile. No kerneld. No kmod. I don't use dynamic module
    loading. All modules here are loaded at boot time by insmod and
    remain loaded. Soon I'll put all the modules back into the kernel.

    > I've tracked this to high ethernet traffic blowing out the ethpro100
    > driver, getting it into a state from which it can't recover. It's been
    > there a *long* time, but got *really* bad in 96-pre1 and 96, although I
    > can't try 97 until tomorrow sometime... unless I build and boot from
    > home tonite (I hate doing this because if the boot hangs, it's a 20+
    > mile drive in to my office to unhang it..)

    Only one Intel NIC and that's on a low-traffic system as I had the same
    troubles with the driver as did you. No Ethernet hangs at all.

    > >
    > > 2. On 2.1.96/7 the system will hang *hard* during a tape backup to
    > > the SCSI tapes. This also happens 100% of the time on the units with
    > > tape drives. (Also doesn't matter whether the drive being backed up
    > > is local or on another machine on the net. All tapes are on
    > > BusLogic BT930 controllers of various vintages. I did move one to
    > > my remaining Adaptec controller but it does the same thing but much
    > > further into the tape.
    > >
    > SCSI has a definite problem in 2.1.96. I found I could not copy a large
    > file from one SCSI drive to another (large=200-500mb) without the machine
    > hanging *hard*. I backed up to 2.1.85 (the only older kernel I happened
    > to have saved in a handy place) and the SCSI copies went perfectly with
    > no problems at all. So something is "up" in 96 for certain, at least with
    > the combination of the bt958 SCSI and etherexpress Pro 100 ethernet
    > card. Sendmail would forward me a few email messages, it would hang. I'd
    > try to ftp a 20mb file from one machine to my quad processor (on a
    > switched hub which provides good thruput) and it would hang... And then
    > I found I couldn't even copy files from one drive to another reliably...

    Agreed. And it continues into 2.1.97. I just hung the system *hard*
    by copying a 19M log file.

    > > As noted, this symptom began, (I think) at 2.1.96 but I'll have to
    > > verify this as most backups have been made on a non-intel machine over
    > > the net since 2.1.88. I did a restore of about 200M of files on this
    > > machine under 2.1.95 without incident.
    > >


    Dr. Bill Broadhurst | Independent contract Engineer.
    (619)296-3710 | BIOS, Firmware, & Diagnostics. | Finger for PGP 5.0 public key.

    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to

     \ /
      Last update: 2005-03-22 13:42    [W:0.046 / U:8.680 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site