Re: [PATCH 6/6] x86: kaslr: relocate base offset at boot

[Kees Cook]

On Mon, Apr 15, 2013 at 2:46 PM, H. Peter Anvin <hpa@xxxxxxxxx> wrote:
> On 04/15/2013 02:41 PM, Kees Cook wrote:
>>>
>>> You seem to be missing something here...
>>>
>>> There are *two* mappings in 64-bit mode.  Physically, if you're going to
>>> randomize you might as well randomize over the entire range... except
>>> not too far down (on either 32 or 64 bit mode)... in particular, you
>>> don't want to drop below 16 MiB if you can avoid it.
>>>
>>> On 64 bits, there is no reason the virtual address has to be randomized
>>> the same way.
>>
>> Aren't we bound by the negative 2GB addressing due to -mcmodel=kernel?
>>
>
> Guys,
>
> Please read what I wrote.
>
> The 2 GB limit is for the *virtual* mapping.
>
> The *physical* mapping, where it lands in RAM, is completely
> independent, and if you're going to randomize the latter, there is no
> reason it has to match the former.  Instead, randomize it freely.

Ah, gotcha. I don't see much benefit in doing this as it would make
the 32-bit and 64-bit logic pretty different without much real-world
gain, IMO.

> That is different from the i386 kernel which runs at its
> physical-mapping address.
>
> Incidentally, for performance reasons please avoid locating the kernel
> below CONFIG_PHYSICAL_ADDRESS if possible.

You mean CONFIG_PHYSICAL_START? This is already done in aslr.S via
LOAD_PHYSICAL_ADDR which is calculated from the
CONFIG_PHYSICAL_ALIGN-masked CONFIG_PHYSICAL_START.

> Also make sure your code works with more than 128 e820 entries.

There should be no problem here; we're using edi to count them.

-Kees

--
Kees Cook
Chrome OS Security
--
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/