Adrian Bunk a écrit :
quoted text
> I tried the following patch with a full x86 .config [1]:
> 
> --- a/include/asm-x86/cache.h
> +++ b/include/asm-x86/cache.h
> -#define __read_mostly __attribute__((__section__(".data.read_mostly")))
> +/* #define __read_mostly __attribute__((__section__(".data.read_mostly"))) */
> 
> The result [2,3] was:
> 
> -rwxrwxr-x 1 bunk bunk 46607243 2007-12-13 19:50 vmlinux.old
> -rwxrwxr-x 1 bunk bunk 46598691 2007-12-13 21:55 vmlinux
> 
> It's not a surprise that the kernel can become bigger when __read_mostly 
> gets used, especially in cases where __read_mostly prevents gcc 
> optimizations.
> 
> My question is:
> Is there anywhere in the kernel a case where __read_mostly brings a 
> measurable improvement or can it be removed?

Yes, there are many cases where read_mostly brings huge improvements.

I did test your idea on a 4 CPUS server, and system time was roughly doubled, 
from 11% to 20%

Of course, you noticed that puting a __read_mostly attribute force the linker 
to reserve space for the variable. So a null variable previously in bss 
section (no space in vmlinux file) is now in .data.read_mostly. Not a big deal.

If you want, you could play some .lds games to create sort of a 
"bss.read_mostly" section to save 10000 bytes in vmlinux.

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/