Re: Bugs on Linux 2.6.18-rc2 sg code?

[Russell King <rmk@arm.linux.org.uk>]

On Sun, Aug 20, 2006 at 01:51:33AM -0600, Fajun Chen wrote:
> I've tested mmap code in both 2.6.15.4 and 2.6.18-rc2 with
> sg_vma_nopage() from 2.6.15.4, this particular sg mmap bug seems to be
> fixed?$B!%!!#S#i#n#c#e!!#y#o#u#r!!#p#a#t#c#h!!#f#o#r!!#2!%#6!%#1#8!]#r#c#4!!#i#s!!essentially not different

This is unreadable...

> from 2.6.15.4 (the only difference is sg->offset, which is always 0
> for sg mmap or indirect IO),  this should fix the problem.  I'll test
> your patch next week to confirm it anyway.
> 
> I would like to also report a potential cache coherency issue on sg
> when running on some processors other than i386.  I ran into cache
> coherency issue on ARM XScale Iop80321.

I no longer touch SCSI, partly because the only SCSI based systems I
have are extremely (10 years) old now.  All my SCSI knowledge is based
on ancient 2.4 code and is no longer relevant.

> Existing sg code for direct IO:

I don't think direct-IO has ever been tested and debugged on ARM either.
I wouldn't know where to start with direct-IO - never used and never
looked at it.

And more importantly (for me), I'd rather not get into another squabble
with James over cache coherency ideas, so I want to stay well away from
cache coherency and block devices, thanks.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core