This patch achieves out of line spinlocks by creating kernel/spinlock.c
and using the _raw_* inline locking functions. Now, as much as this is
supposed to be arch agnostic, there was still a fair amount of rummaging
about in archs, mostly for the cases where the arch already has out of
line locks and i wanted to avoid the extra call, saving that extra call
also makes lock profiling easier. PPC32/64 was an example of such an arch
and i have added the necessary profile_pc() function as an example.
Size differences are with CONFIG_PREEMPT enabled since we wanted to
determine how much could be saved by moving that lot out of line too.
ppc64 = 259897 bytes:
text data bss dec hex filename
5489808 1962724 709064 8161596 7c893c vmlinux-after
5749577 1962852 709064 8421493 808075 vmlinux-before
sparc64 = 193368 bytes:
text data bss dec hex filename
3472037 633712 308920 4414669 435ccd vmlinux-after
3665285 633832 308920 4608037 465025 vmlinux-before
i386 = 524115 bytes
text data bss dec hex filename
5695619 870906 328112 6894637 69342d vmlinux-after
6221254 870634 326864 7418752 713380 vmlinux-before
x86-64 = 282446 bytes
text data bss dec hex filename
4598025 1450644 523632 6572301 64490d vmlinux-after
4881679 1449436 523632 6854747 68985b vmlinux-before
It has been compile tested (UP, SMP, PREEMPT) on i386, x86-64, sparc,
sparc64, ppc64, ppc32 and runtime tested on i386 and x86-64. I still have
to get benchmarks done (most probably i386). The patch is currently
against 2.6.9-rc1-mm1 because we'd have to back out the i386/x86_64 out of
line lock patches since this does it in a different way, i can merge later
on.
Thanks,
Zwane
> This patch achieves out of line spinlocks by creating kernel/spinlock.c
> and using the _raw_* inline locking functions.
...
> Size differences are with CONFIG_PREEMPT enabled since we wanted to
> determine how much could be saved by moving that lot out of line too.
So does this have no performance impact at all? or has that not been measured?
M.
On Fri, 3 Sep 2004, Martin J. Bligh wrote:
> > This patch achieves out of line spinlocks by creating kernel/spinlock.c
> > and using the _raw_* inline locking functions.
> ...
> > Size differences are with CONFIG_PREEMPT enabled since we wanted to
> > determine how much could be saved by moving that lot out of line too.
>
> So does this have no performance impact at all? or has that not been measured?
No performance regression on i386 with a specific benchmark (bonnie++)
http://www.zwane.ca/results/cool-locks-stp
I'm going to be running benchmarks against this version of the patch, but
still collecting which to run.
Thanks,
Zwane