Re: [CFT][PATCH] 2.6.4-rc1 remove x86 boot page tables

[Eric W. Biederman]

hpa@xxxxxxxxx (H. Peter Anvin) writes:

> Followup to:  <m1vflp81kq.fsf@xxxxxxxxxxxxxxxxxxxxxxxxx>
> By author:    ebiederm@xxxxxxxxxxxx (Eric W. Biederman)
> In newsgroup: linux.dev.kernel
> > 
> > I think I have accounted for the sub architectures but I don't have
> > the hardware to test them.  For voyager and VISWS I actually changed
> > some code so I would appreciate a confirmation I didn't break
> > anything.  
> > 
> 
> For VISWS I think you actually need to turn paging off explicitly.

Hmm.  I will look at that.

> Also, you probably need to check that you didn't break 4G/4G,
> especially on SMP.

The patch will need a few tweaks but it should be fairly straight forward.

> I would also like to remove the magic %bx, which I did in the version
> of my patch sent to akpm and which is now in -mm (basically the SMP
> trampoline jumps to a different entrypoint instead.)  In that patch,
> %ebx is still used as a flag, but it's completely internal to head.S.

Ok I have not seen that one.  In this case there was enough common
code that it seemed reasonable to reuse it.  I managed to reduce
the code to two tests.  With just a bit of care I suspect I could
remove the tests completely.
 
> See ftp://ftp.kernel.org/pub/linux/kernel/people/hpa/earlymem-7.diff
> 
> > Thanks to HPA who got the ball started :)
> 
> :)
> 
> I definitely agree that simply using no paging until we have page
> tables is by far the cleanest approach.  I felt that is was too high
> risk for 2.6, basically because I'm a chicken, but more realistically
> because I couldn't really see the effect on all subarchitectures, and
> I didn't feel 100% confident that there wasn't anything that relied on
> memory being dual mapped; however, I'm more than happy to be proven
> wrong :)

I will try.  So far except for early_printk I have not found anything.
And that was easy to fix.
 
> Oh yes, with this change you should probably just move swapper_pg_dir
> (and empty_zero_page?) into .bss like anything else that should be
> zero after boot.

Sounds like a good idea.

Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/