Re: [RFC 00/15] x86_64: Optimize percpu accesses

[Eric W. Biederman]

Arjan van de Ven <arjan@xxxxxxxxxxxxx> writes:

> On Wed, 09 Jul 2008 13:00:19 -0700
> ebiederm@xxxxxxxxxxxx (Eric W. Biederman) wrote:
>
>> 
>> I just took a quick look at how stack_protector works on x86_64.
>> Unless there is some deep kernel magic that changes the segment
>> register to %gs from the ABI specified %fs CC_STACKPROTECTOR is
>> totally broken on x86_64.  We access our pda through %gs.
>
> and so does gcc in kernel mode.

Some gcc's in kernel mode.  The one I tested with doesn't.

>> Further -fstack-protector-all only seems to detect against buffer
>> overflows and thus corruption of the stack.  Not stack overflows.  So
>> it doesn't appear especially useful.
>
> stopping buffer overflows and other return address corruption is not
> useful? Excuse me?

Stopping buffer overflows and return address corruption is useful.  Simply
catching and panic'ing the machine when the occur is less useful.  We aren't
perfect but we have a pretty good track record of handling this with
old fashioned methods.

>> So we don't we kill the broken CONFIG_CC_STACKPROTECTOR.  Stop trying
>> to figure out how to use a zero based percpu area.
>
> So why don't we NOT do that and fix instead what you're trying to do?

So our choices are.
fix -fstack-protector to not use a hard coded offset.
fix gcc/ld to not miscompile the kernel at random times that prevents us from
booting when we add a segement with an address at 0.

-fstack-protector does not use the TLS ABI and instead uses nasty hard coded magic
and that is why it is a problem.  Otherwise we could easily support it.

>> That should allow us to make the current pda a per cpu variable, and
>> use %gs with a large offset to access the per cpu area. 
>
> and what does that gain us?

A faster more maintainable kernel.

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/