Re: rfc/rft: use r10 as current on x86-64

[Andi Kleen]

On Tue, Nov 22, 2005 at 11:52:04AM -0500, Benjamin LaHaise wrote:
> Hello Andi et al,
> 
> The patch below converts x86-64 to use r10 as the current pointer instead 
> of gs:pcurrent.  This results in a ~34KB savings in the code segment of 
> the kernel.  I've tested this with running a few regular applications, 
> plus a few 32 bit binaries.  If this patch is interesting, it probably 
> makes sense to merge the thread info structure into the task_struct so 
> that the assembly bits for syscall entry can be cleaned up.  Comments?

I think you could get most of the benefit by just dropping
the volatile and "memory" from read_pda(). With that gcc would
usually CSE current into a register and it would would work essentially
the same way with only minor more .text overhead, but r10 would be still
available.

Unfortunately when that's done then the kernel doesn't boot.
It's probably something silly, but i never had time to track it down.
Might want to look into that?

Looking at your patch it might be enough to make sure all users
of current after the changes in __switch_to you did use some 
other way to access it (there is unfortunately no way I know
of to make gcc flush all CSEd items without addings barriers
in the original get_current function)

-Andi
-
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/