Re: [PATCH V7 05/18] x86/pks: Add PKS setup code

From: Thomas Gleixner
Date: Fri Nov 26 2021 - 06:40:20 EST


On Thu, Nov 25 2021 at 16:15, Thomas Gleixner wrote:
> On Tue, Aug 03 2021 at 21:32, ira weiny wrote:
> Aside of that, the function which set's up the init value is really
> bogus. As you explained in the cover letter a kernel user has to:
>
> 1) Claim an index in the enum
> 2) Add a default value to the array in that function
>
> Seriously? How is that any better than doing:
>
> #define PKS_KEY0_DEFAULT PKR_RW_ENABLE
> #define PKS_KEY1_DEFAULT PKR_ACCESS_DISABLE
> ...
> #define PKS_KEY15_DEFAULT PKR_ACCESS_DISABLE
>
> and let the compiler construct pkrs_init_value?
>
> TBH, it's not and this function has to be ripped out in case that you
> need a dynamic allocation of keys some day. So what is this buying us
> aside of horrible to read and utterly pointless code?

And as Taoyi confirmed its broken.

It surely takes a reviewer to spot that and an external engineer to run
rdmsr -a to validate that this is not working as expected, right?

Sigh...