View Incident:
http://co-op.engr.sgi.com/BugWorks/code/bwxquery.cgi?search=Search&wlong=1&view_type=Bug&wi=800850
Status : open Priority : 2
Assigned Engineer : lord Submitter : dxm
*Modified User : lord *Modified User Domain : sgi.com
*Description :
Enabling CONFIG_HIGHMEM4GB on bruce (a 1400), then
running QA trips the following BUG() in QA 001:
kernel BUG at highmem.c:231!
Entering kdb (0xf6eec000) on processor 0 Panic: invalid operand
due to panic @ 0xc013077b
eax = 0x0000001d ebx = 0xfe268000 ecx = 0xc02b406c edx = 0x00000028
esi = 0x00000000 edi = 0xc2055790 esp = 0xf6eedda0 eip = 0xc013077b
ebp = 0xf6eeddb4 ss = 0x00000018 cs = 0x00000010 eflags = 0x00010246
.....
==========================
ADDITIONAL INFORMATION (ADD)
From: lord@xxxxxxx (BugWorks)
Date: Sep 08 2000 01:52:14PM
==========================
> I've been able to reliably reproduce Daniel's oops with highmem
> at xfs_itobp() on "sierra" running stress test 013. I've tried
> Steve's suggested check for trapping unmapped metadata pages
> in page_buf.c and his check doesn't fire.
>
> Any other ideas?
Wait for me to get around to this bug would be one .....
Another would be to change xfs_itobp a little so that it prints
out the pagebuf pointer in this case - i.e. if page_buf_offset
returns zero then print the pagebuf pointer before continuing
and dying. Make sure you have the page_buf kdbm module loaded
and then dump the pagebuf and kiobuf contents into the PV.
|