On Wed, Jul 07, 1999 at 10:58:54AM +0100, Neil Conway wrote:
>
> Now that's a different story. Are you very sure? To me, this implies
> either (a) broken hardware, (b) subtle fs corruption, (c) kernel bug.
> Dunno which to go for, but (c) seems rather unlikely :-)
>
> Can you reproduce it at will or is it just one particular pair of
> files? Can you reproduce on different filesystems?
This is not file specific nor filesystem specific.
I made some new trials today, and I succeded in isolate a little better
the problem. The pseudo corruption of my files appears when and only when
I activate the option "CONFIG_PCI_QUIRKS" and "CONFIG_PCI_OPTIMIZE".
More precisely it happens when the optimization "PCI-Memory posted write"
is switched on and it doesn't depend on the number of CPUs I'm running on.
[All experiments are done on a 2.2.10ac8 smp kernel].
When I switch this optimization off (other optimization are kept), I don't
experience the corruption problem anymore, but then I only can boot linux
with a single processor (diff between processors stepping must be the
the right explanation, as it's already been mentionned in previous posts) :
- when starting with 2 procs, the boot sequence fails just before
starting INIT on an scsi timeout.
(scsi0 is an AM53/79C974 PCscsi driver rev. 0.5)
VFS : Mounted root (ext2 filesystem) readonly.
Freeing unused kernel memory: 52k freed
scsi: aborting command due to timeout : pid 43, scsi0, channel 0, id 0,
lun 0 Read (6) 00 02 e3 10 00
SCSI host 0 abort (pid 43) timed out - resetting
SCSI bus is being reset for host 0 channel 0.
AM53C974_reset called
AM53C974 register dump:
IO base: 0xff00; CTCREG: 0x0000; CMDREG: 0x90; STATREG: 0x11; ISREG: 0x84
CFIREG: 0x80; CNTLREG1-4: 0x57; 0x40; 0x18; 0x44
DMACMD: 0xc3; DMASTC: 0x0400; DMASPA: 0x205000
DMAWBC: 0x0000; DMAWAC: 0x205400; DMASTATUS: 0x08
----------------------------------------------------------
- when starting with 1 proc, everything goes fine, except that
I receive a bunch of "eth0 : Bus master arbitration failure, status
88f3." (eth0 is PCnet/PCI 79C970 at 0xffe0) when doing file accesses.
Fabrice
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
