Hmm. I'm guessing hardware problems. Here's why:
The corruption occurs on 8-byte boundaries, and the corruption involves
setting the first four bits -- notice that the broken characters always
end in {1,3,5,7}7? The offsets are mostly 8 bytes apart, but some are
16 apart -- presumably the missing character in the middle already had
those bits set. Also, the first corrupted section is 4088 bytes, 8
less than 4096, and 1245189 is almost exactly page-aligned; the whole
section fits in one page. The second block is also approximately one
page, 4040 bytes, and is page-aligned.
And since this problem is, relatively speaking, very rare and hard for
you to trigger, I can easily believe that it would simply not have been
triggered under a different kernel.
> I don't know if this could have anything to do with this, but I
>have also been testing devel kernels, and all of them from 2.1.51 to the
>current 57 have had severe file system corruption--not of the above kind,
>but the 'EXT2-FS Error' type and hundreds of inodes corrupted and
>direcotries attached to the wrong place /usr/man/man1/ps.1.gz is really a
>direcotory holding the linux/drivers/net subdirectory of my kernel tree
>for example.
This makes it even more likely that it is hardware -- the same problem
could have wildly different results if it happens while reading an inode
in.
Doug Ledford recently posted about problems with SDRAM in PPros. Any
chance that's what you've got?
Also, try turning off DMA in the ide driver and see if that fixes the
problem. But Mark Lord would know far more about that than I would.
michaelkjohnson
"Magazines all too frequently lead to books and should be regarded by the
prudent as the heavy petting of literature." -- Fran Lebowitz