I found the following in comments made by Alan Cox.
Note the IDE highmem fixes (scsi needs doing I suspect).
But, it may be unrelated.

 -- Bjarne

Linux 2.4.20-pre5-ac4
o       Fix error path bug in pci resource code         (Keith)
o       Fix p4-clockmod compile error                   (Adrian Bunk)
o       Align packets nicely on kaweth USB ethernet     (Oliver Neukum)
o       Further Changes file fix                        (Steven Cole)
o       TCP timestamp handling fix                      (Dave Miller)
o       Compile warning fixes                           (Niels Jensen)
o       Next batch of IDE header updates                (Andre Hedrick)
o       IDE scsi update                                 (Andre Hedrick)
        | Needs some highio cleanup yet
o       IDE DMA updates                                 (Andre Hedrick)
o       Update the IDE PCI driver layer                 (Andre Hedrick)
o       Fix pdc202xx further braindamage                (me)
o       Further icside fixes                 (Bartlomiej Zolnierkiewicz)
o       Fix ide-lib atapi DMA check          (Bartlomiej Zolnierkiewicz)
o       CMD64x rev 5/7 UDMA check fix        (Bartlomiej Zolnierkiewicz)
o       Add blk_fs_request helper                       (Jens Axboe)
o       IDE highmem fixes (scsi needs doing             (Jens Axboe)
        I suspect)
o       Longer PIO timeout for taskfile write           (Andre Hedrick)
o       Fix promise cable detect                        (Andre Hedrick)
o       Split promise into old and new drivers          (me)


On Wed, 2002-09-18 at 22:15, Bryan Whitehead wrote:
> Bjarne Thomsen wrote:
> > We have consistently encountered hard system locks
> > with all enterprise 2.4.18 and 2.4.19 kernels while
> > transferring relatively large amount of data (images)
> > either via ftp or NFS.
> 
> We heavily use NFS. The RPM's I was installing were on a NFS server.
> 
> > We did NOT encounter any problems with the 2.4.8
> > enterprise kernel from ML 8.1, so we have been using
> > that kernel with ML 8.2, until somebody finds a fix.
> > 
> > We did not have any problem with the ordinary 2.4.18
> > or 2.4.19 kernels, so the problem is only found in
> > the enterprise kernel.
> 
> Yep, same here. Only the 9.0 2.4.19 kernels have the problem.
> 
> > As an experiment I compiled one of the 2.4.18 enterprise
> > kernels without the HIGHMEM option (the only change made).
> > This kernel did not have the lock-ups, so I concluded
> > that the problem had something to do with the HIGHMEM
> > code.
> > 
> > It is actually strange to me that this problem has
> > been so little commented, considering the low prices on RAM.
> 
> We had thought it was a hardware error so we ran a bunch of tests before 
> we reported it.
> 
> > Does anybody know if this problem has been or will be fixed
> > in 2.4.20 ?
>  >
> >  Bjarne Thomsen
> >  Department of Physics and Astronomy
> >  University of Aarhus
> >
> 
> 
> Mandrake? any comment? or tips to resolve this?
> 
> 
> > On Wed, 2002-09-18 at 20:22, Bryan Whitehead wrote:
> > 
> >>Testing one of our 2GB Dual-Pentium III (1ghz) machines we get a hard 
> >>lock with kernel enterprise.
> >>kernel-enterprise-2.4.19.9mdk-1-1mdk.i586.rpm
> >>
> >>The lock up seems to be related to the scsi driver. aic7xxx. The last 
> >>message that is echoed to the syslog consol is:
> >>kernel: (scsi0:A:0:0): Locking max tag count at 64
> >>
> >>after that the machine is dead. No ping, but keyboard works enough for 
> >>Alt-SysRq stuff. Only problem is Alt-SysRq keys won't work. :( syncing 
> >>the disk results in the machine "trying" (It starts to sync the disk) 
> >>but will never finish, there is no HD activity. Then the keyboard will 
> >>not respond anymore.
> >>
> >>Trying to get a process list, or memory stats will finish off the lock 
> >>also. (as in it will become so useless that reset button is only thing 
> >>working).
> >>
> >>If my first sysrq key is to reboot it will.
> >>
> >>I ran memtest yesterday and all memory is ok. Mandrake 8.1 has no 
> >>problems on this machine.
> >>
> >>Heavy disk activity triggers the lock, like installing an RPM. :)
> >>
> >>It will always hard lock when installing the "FlashPlayer-5.0-8mdk" RPM.
> >>
> >>any suggestions? or specific sysrq keys you'd like me to try in locked 
> >>up state? (there is no oops or anything like that on the consol window)
> >>
> >>
> >>-- 
> >>Bryan Whitehead
> >>SysAdmin - JPL - Interferometry Systems and Technology
> >>Phone: 818 354 2903
> >>[EMAIL PROTECTED]
> >>
> >>
> > 
> > 
> > 
> 
> 
> -- 
> Bryan Whitehead
> SysAdmin - JPL - Interferometry Systems and Technology
> Phone: 818 354 2903
> [EMAIL PROTECTED]
> 
> 



Reply via email to