Files of death on sparc64

[James Birdsall <jwb-tech%picarefy.com@localhost>]

I have an AXi with a couple of SATA controllers on the PCI bus and four 
SATA disks. It's supposed to be my fileserver. Tonight I started moving 
another batch of files over to it and I have encountered repeated 
panics. I'm not 100% sure it's an individual file, it may just be a 
particular point in the group of files, but it doesn't matter what disk 
I'm copying them to, or whether I copy them via samba or FTP; when the 
copy reaches the critical point, I see a kernel trap or panic (typically 
"lockmgr: no context" or a data alignment error), and then on reboot I 
get an uncorrectable DMA error while it's doing the fsck. Ten-finger 
copy of the DMA error is below.

The machine has NetBSD 4.0/sparc64 installed and is running the GENERIC 
kernel off the ISO. Before 4.0 came out, I tried earlier versions, and 
they would panic in the filesystem any time I tried to FTP in a large 
file. I was pleased to install 4.0 and find that problem solved, but 
maybe it wasn't solved completely? Does anybody recognize this or have 
any ideas?

Thanks!

--James

psycho0: uncorrectable DMA error AFAR 10c4640 pa ffffffffffffffff AFSR 
210000ff00000000:210000ff00000000<P_DTE,P_DWR>
IOVA c0c80000 IOTTE 36ce0012
cpu0: kdb breakpoint at 13d0460
Stopped at netbsd:cpu_Debugger+0x4:    nop
db> machine stack
Window 0 frame64 0xe0017740 locals, ins:
ffffffffffffffff 210000ff00000000 6 10c4640 2dd9c00 fffc4000 e0018000 
e0018000
1 5 e0017930 1 138c9e0 7 e0017081=sp 1008f94=pc:netbsd:sparc_interrupt+0x1d4
Window 1 frame64 0xe0017880 locals, ins:
80000606 13c5d54 2debe80 8000 18054b8 0 f ffffffffffffffff
c818550 fffffffffffffff8 1000000 6 4 0 e0017261=sp 
13c5d4c=pc:netbsd:_bus_dmamap_upload+0x6c
Window 2 frame64 0xe0017a60 locals, ins:
c818550 33c5200 0 8 1fff ffffffffffffe000 c82ea00 13e8000
1820cb0 33c5200 0 10000 2 7fffffffffffffff e0017321=sp 
138e784=pc:netbsd:iommu_dvmamap_unload+0x24
Window 3 frame64 0xe0017b20 locals, ins:
10000 323fa00 2 1857a4c 1881400 1857a4c 0 67
2dd9c80 2dd9c00 33c5200 10000 2 d2e7dc0 e00173e1=sp 
1346fc0=pc:netbsd:pciide_dma_finish+0xc0
Window 4 frame64 0xe0017be0 locals, ins:
336f420 24 336f000 4 1880800 185e400 1 1881000
0 0 0 0 187a300 1f5a e00174a1=sp 1183608=pc:netbsd:wdcintr+0x88
Windows 5 frame64 0xe0017ca0 locals, ins:
336f000 2d8de60 0 80 336e000 e0 c8 932c000
336f228 14 12b8380 1 1896400 187a300 e0017561=sp 
1346858=pc:netbsd:pciide_pci_instr+0x58
Window 6 ... 1008f94=pc:netbsd:sparc_interrup+0x1d4
Window 7 ... 1294c74=pc:netbsd:microtime+0x14
Window 8 ... 128f968=pc:netbsd:mi_switch+0x148
Window 9 ... 128fe58=pc:netbsd:ltsleep+0x258
Window a ... 1282058=pc:netbsd:physio_wait+0x38
Window b ... 12826ac=pc:netbsd:physio+0x42c
Window c ...  12de1a0=pc:netbsd:spec_read+0x1a0
Window d ... 1230288=pc:netbsd:ufsspec_read+0x48
Window e ... 12d7de8=pc:netbsd:VOP_READ+0x28
Window f ... 12d5d24=pc:netbsd:vn_read+0xa4
Window 10 ... 12a890c=pc:netbsd:dofileread+0xac
Window 11 ... 12d1cb0=pc:netbsd:sys_pread+0xd0
Window 12 ... 13cd9e0=pc:netbsd:syscall_plain+0x120
at this point the stack trace wandered off into the weeds and the 
debugger dropped into the ROM monitor

This is a standard GENERIC kernel.

The machine has NetBSD 4.0/sparc64 installed. Before 4.0 came out, I 
tried earlier versions, and they would panic in the filesystem any time 
I tried to FTP in a large file. I was pleased to install 4.0 and find 
that problem solved, but maybe it wasn't solved completely? Does anybody 
recognize this or have any ideas?

Thanks!

--James