Subject: Re: kern/26568: Yesterday's "pciide" `irqack fix' breaks Promise 202xx controllers
To: None <bouyer@antioche.lip6.fr>
From: Paul Shupak <paul@Plectere.com>
List: netbsd-bugs
Date: 08/06/2004 17:23:49
	Manual,

	Of course right after leaving the house, the Via chipset machine
locked hard - a stack trace is copied below.  The machine was running
between two and a half to three and a half hours doing at least a
raidframe (RAID5) parity rebuild over 6 disks (4 on the south bridge,
two - one each as masters on the Promise's channels) and was on its
second "build.sh -E build" (on this machine they typically take ~1:15,
2:15 when the raid is rebuilding) - I believe the "cd src ; make -j 3
dependall" was completed.  One difference, instead of infinite looping
printf's I get a hard lock-ups with your patch.

	I know the Promise is non-standard (ever try ATAPI on one - the
company says it works for Win2K and WinXP, but not otherwise - I wonder
what non-standard tricks they play, though the final BIOS v18 recognizes
ATAPI devices, I've never seen one actually work under any OS).  Basically,
the Promise 202xx is f**cked.  (I've been meaning to remove the Promise from
this particular machine and use a CMD0680 that is already in there (one
channel using CD-RW and a DVD -- That will mean moving my 6 disks from
4 channels to three, but I doubt that will hurt:  Its just that the kernel
config has all the locators "fixed" and I need to coordinate changing the
kernel with moving the cables and changing the jumpers --  The point of
all this is I can gladly lend you, long term, a FastTrack-66 if you want
to test yourself -- contact me off list and I'll ship it to you within a
couple of week(end)s -- allow another week or so to get to France from
the USA -- It value is probably just about the same as shipping; I'd
guess about $7-$8US - It cost $20 four years ago - I certainly won't
worry about it being gone).

        Thanks for looking at this,

        paul shupak


P.S.  I can build a "-g" kernel, but don't know how to get this machine
to dump except by removing memory (two 1G ECC sticks - it is many a file
server for home directories, but is also the master DNS server for several
domains and my master KDC; so I'd rather not keep it down for very long or
remove any of its memory).

P.P.S.  It locked again while I composed this message (and while it
is again rebuilding the RAID arrays); Its been up about a half of an
hour -- I'm going back to my v1.11 pdcide.c kernel.  Sorry, it is in
"X" this time so no traceback, but the machine hasn't panic'd in 4 years
- except when two disks died simultaneously (and "raidframe" had a fit).


--------------------------------------------------------------------------------
Stack trace (copied by hand) -- ">>>" added in front of the printf(...0x1131520)
--------------------------------------------------------------------------------

cpu_Debugger(c2d54400,c2d8700a,cf38c988,c2d54400) at netbsd:cpu_Debugger+0x4
internal_command(c2d54400,cf38c994,f420,1b,0) at netbsd:internal_command+0x13c
wskb_translate(c085dfa0,2,1,1,cf38c9e4) at netbsd:wskb_translate+0x6c
wskb_input(c2d54400,2,1,1,ceddaef4) at netbsd:wskb_input+0x137
pckb_input(c2d82380,1,ce568d68,29,3) at netbsd:pckb_input+0x53
pckbportintr(c095e0e0,0,1,0,1) at netbsd:pckbportintr+0x3a
pckbcintr(c2d82280,6,c0400010,30,c2d80010) at netbsd:pckbcintr+0x94
Xintr_legacy1() at netbsd:Xintr_legacy1+0xa9
--- Interrupt ---
Xspllower(6,c2db9e00,cf38cac4,282,c2d8d01c) at netbsd:Xspllower+0xe
>>> printf(c049a261,c2d8d01c,0,1131520,0) at netbsd:printf+0x71
pdc202xx_pci_intr(c2d8d000,5,1130010,30,1130010) at netbsd:pdc202xx_pci_intr+0xa
8
Xintr_legacy5() at netbsd:Xintr_legacy5+0xa9
--- Interrupt ---
Xspllower(5,0,3,202,4) at netbsd:Xspllower+0xe
ipintr(10,c2db0030,10,18040010,c080e860) at netbsd:ipintr+0x8d
DDB lost frame for netbsd:Xsoftnet+0x41, trying 0xcf38cb78
Xsoftnet() at netbsd:Xsoftnet+0x41
--- Interrupt ---
0x3: