data corruption with current (maybe sis chipset related?)

[Terry Lambert]

Don Lewis wrote:
> On  9 May, Terry Lambert wrote:
> > Chris BeHanna wrote:
> >>     And, why aren't Bosko's patches in the tree?
> >
> > I don't know.  I do know that they increased the minimum memory
> > requirements by 4M (part of Bosko's approach to a fix requires
> > linking the kernel with a base address aligned on a 4M boundary).
> 
> How hard would it be to make this a compile time option?  Small memory
> machines are unlikely to want to use 4 MB pages anyway.  In other words
> a configuration option that would disable 4 MB pages and put the kernel
> at its current location when set one way, and would enable 4MB pages and
> relocate the kernel on a 4 MB boundary when set the other way.

The KVA space size is a compile time option, and it changes
the base address, so changing the relocation address is just
a matter of a small amount of additional code; however, the
locore.s and machdep.c and any other changes necessary to
use the patch are much harder to make compile-time variant.
Particularly the assembly code.


> I really dislike our default configuration of a little bit more speed at
> the expense of data integrity.  If that's what I really wanted, I could
> probably get even more speed by overclocking.

Or running Linux.  Or using background fsck.  8-) 8-).

-- Terry