Re: [PATCH] x86, AMD: Correct F15h IC aliasing issue

[Borislav Petkov <bp@amd64.org>]

On Sun, Jul 24, 2011 at 09:40:46AM -0400, Borislav Petkov wrote:
> > also, the PAGE_ALIGN() complication here looks unnecessary - can the
> > vdso 'addr' ever be not 4K aligned above?
> 
> Yeah, it can. I did some trace_printk runs and 'addr' wasn't 4K aligned
> in some cases. I think this is because of the stack randomization we do
> and we use mm->start_stack to get the vdso base address. Unfortunately,
> I can't find those traces anymore but will do some again tomorrow.

Better late than never :).

So yes, mm->start_stack comes in unaligned from setup_arg_pages() and we
use it to generate the vdso VA because the vdso is below the stack:

7fff761f9000-7fff7621a000 rw-p 00000000 00:00 0                          [stack]
7fff763ec000-7fff763ed000 r-xp 00000000 00:00 0                          [vdso]
ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0                  [vsyscall]

here's a trace_printk excerpt from vdso_addr() (gets inlined into
arch_setup_additional_pages()):

           <...>-2516  [006]   624.364274: arch_setup_additional_pages: mm->start_stack: 0x7fff7621942a, len: 4096
           <...>-2516  [006]   624.364278: arch_setup_additional_pages: randomized: 0x7fff763eb42a
           <...>-2516  [006]   624.364279: arch_setup_additional_pages: final addr: 0x7fff763eb42a

Then, both arch_get_unmapped_area{,_topdown} page-align it.

For our case, we PAGE_ALIGN it before randomizing it so that we can
control the [14:12] setting.

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551