Re: x86_64 SMP status update

[Chris Wright <chrisw@osdl.org>]

* Ian Pratt (m+Ian.Pratt@cl.cam.ac.uk) wrote:
> Chris -- thanks!
> 
> I think we'll try and get Vincent's 2.6.12 upgrade patch checked-in in
> the next day or two.

Great, it's been working well for me.  Granted, I'm usually just
crashing the domain, but... ;-)

> On the subject of x86_64 SMP, I personally believe the #1 priority is to
> get x86_64 using writeable page tables. It'll be impossible to make SMP
> work reliably until this is done.
>  
> The first step of this is just to stop set_pte using queue_l1_update and
> have it write to the pte directly. Xen is currently hard configured to
> only ever use the emulation path for x86_64 (see ptwr_do_page_fault). I
> suspect that letting it use the 'real' path will probably crash, but
> should be easy to fix up.

OK, I'll take a look.

> Ideally, we should also pull in equivalent code from this chageset for
> i386 so as to remove the static update queues altogether, and to build
> them on the stack as required:
> 
>  http://xen.bkbits.net:8080/xeno-unstable.bk/cset@1.1236.1.151
> 
> If we do more remerging with i386-xen we should just get this for free.

OK, that looks like good stuff to pull over.

> [BTW: in your SMP patch, any reason not to just merge with the i386-xen
> smpboot.c?]

Sure, one reason, laziness ;-)  Just wanted to get the thing going first.
Suppose it's worth considering.  Although I think the x86_64 smp boot
code in 2.6.12 is a bit nicer.  Food for thought...

thanks,
-chris