Re: [PATCH -v3.2 3/3] x86, AMD: Move BSP code to cpu_dev helper

[Borislav Petkov <bp@amd64.org>]

On Fri, Aug 05, 2011 at 04:07:40PM -0400, H. Peter Anvin wrote:
> On 08/05/2011 11:04 AM, Borislav Petkov wrote:
> > Move code which is run once on the BSP during boot into the cpu_dev
> > helper.
> > +static void __cpuinit bsp_init_amd(struct cpuinfo_x86 *c)
> > +{
> > +	if (static_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
> > +
> 
> You can't use static_cpu_has() here, since this code runs before
> alternatives -- it will always be false.  Furthermore, for code that
> only runs once, it is never a win to do patching.

Oh crap, this is a leftover from when run_on_bsp was struct
x86_cpuinit_ops member with no args. And I f*cked it up even then
although I went and got myself a pointer to boot_cpu_data:

+static void __cpuinit amd_run_on_bsp(void)
+{
+       struct cpuinfo_x86 *c = &boot_cpu_data;
+
+       if (static_cpu_has(X86_FEATURE_CONSTANT_TSC)) {

but forgot to use it. Good catch, will fix it tomorrow.

> Arguably bsp_init should be __init and not __cpuinit, but I don't know
> how to make that work with the machinery, and is something that can be
> fixed anyway.

Yeah, how do we do that? struct cpu_dev is __cpuinitconst,
x86_cpuinit_ops is __cpuinitdata.

We could add it to identify_boot_cpu() - there's already some per-vendor
stuff like init_amd_e400_c1e_mask() which wouldn't hurt to be behind a
vendor check. early_identify_cpu() does already the vendor check with
get_cpu_vendor() so later, in identify_cpu() we could add a run_on_bsp()
which is __init and switch/case on the ->x86_vendor inside.

Then we can collect all the run-once-on-the-BSP code in there.

Hmmm..

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551