Re: [RFC PATCH 3/6] x86/cpu: Re-apply forced caps every time cpu caps are re-read

[Borislav Petkov]

On Mon, Dec 05, 2016 at 05:01:12PM -0800, Andy Lutomirski wrote:
> Calling get_cpu_cap() will reset a bunch of CPU features.  This will
> cause the system to lose track of force-set and force-cleared
> featueres in the words that are reset until the end of CPU
> initialization.  This can cause X86_FEATURE_FPU, for example, to
> change back and forth during boot and potentially confuse CPU setup.
> 
> To minimize the chance of confusion, re-apply forced caps every time
> get_cpu_cap() is called.
> 
> Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxx>
> ---
>  arch/x86/kernel/cpu/common.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
> index 347ae0a19380..24e1e4004d42 100644
> --- a/arch/x86/kernel/cpu/common.c
> +++ b/arch/x86/kernel/cpu/common.c
> @@ -758,6 +758,13 @@ void get_cpu_cap(struct cpuinfo_x86 *c)
>  		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
>  
>  	init_scattered_cpuid_features(c);
> +
> +	/*
> +	 * Clear/Set all flags overridden by options, after probe.
> +	 * This needs to happen each time we re-probe, which may happen
> +	 * several times during CPU initialization.
> +	 */
> +	apply_forced_caps(c);

I guess...

Although I have to say all that early capabilities detection has grown
a lot of cruft during the years and is nuts. And calling get_cpu_cap()
multiple times is simply plain wrong.

What we should do is read CPUID *once*, filter out caps and set our
internal representation bits and be done with it.

Stuff like setup_pku() which *sets* CPUID bits will then have to run
*before* we do the detection and that's it.

But I guess that's future work.

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.