Re: [RFCv2 00/10] Linear Address Masking enabling

[Thomas Gleixner]

On Thu, May 12 2022 at 17:08, H. J. Lu wrote:
> On Thu, May 12, 2022 at 4:35 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> > When an application asks for LAM_U57, I expect it will store tags in
>> > upper 6 bits, even if the kernel enables LAM_U48.
>>
>> The kernel does not enable LAM_U48 when the application only wants to
>> have LAM_U57, because that would restrict the address space of the
>> application to 47 bits on 5-level capable system for no reason.
>>
>> So what are you trying to tell me?
>>
> I am expecting applications to ask for LAM_U48 or LAM_U57, not just
> LAM.

That still does not tell anything.

You can as well tell me, that this will depend on the moon phase. That
would be more telling at least.

If we commit to an ABI, which we have to support forever, then we want
factual arguments, not expectations.

The fact that hardware supports 5 variants does not mean that all of
them make sense to support in the OS.

Quite the contrary, 99% of all 'flexible' hardware features are based on
bogus assumptions. The worst of these assumptions is:

      'We can handle this in software'

Sure, most of the trainwrecks hardware people provide can be 'handled'
in software, but you have to consider the price for doing so.

   The price is that we amount technical debt.

You are well aware of this as you have contributed your share of
technical debt by cramming unsupportable crap into user space projects
prematurely.

So can you please take a step back and seriously think about the
semantics and long term consequences instead of providing handwaving
expectations which are biased by uninformed wishful thinking, AKA
marketing?

Thanks,

        tglx