Re: [PATCH 00/23] [v4] KAISER: unmap most of the kernel from userspace page tables

[Ingo Molnar]

* Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx> wrote:

> On 11/23/2017 10:35 PM, Ingo Molnar wrote:
> > So the pteval_t changes break the build on most non-x86 architectures (alpha, arm, 
> > arm64, etc.), because most of them don't have an asm/pgtable_types.h file.
> > 
> > pteval_t is an x86-ism.
> > 
> > So I left out the changes below.
> 
> There was a warning on the non-PAE 32-bit builds saying that there was a
> shift larger than the type.  I assumed this was because of a reference
> to _PAGE_NX, and thus we needed a change to pteval_t.
> 
> But, now that I think about it more, that doesn't make sense since
> _PAGE_NX should be #defined down to a 0 on those configs unless
> something is wrong.

If pte flags need to be passed around then the canonical way to do it is to pass 
around a pte_t, and use pte_val() on it and such.

But please investigate the warning.

One other detail: I see you fixed some of the commit titles to use standard x86 
tags - could you please also capitalize sentences? I.e.:

  - x86/mm/kaiser: allow flushing for future ASID switches
  + x86/mm/kaiser: Allow flushing for future ASID switches

Could you please also double-check whether the merges I did in the latest 
WIP.x86/mm branch are OK? Andy changed the entry stack code a bit under Kaiser, 
which created about 3 new conflicts.

The key resolutions that I did were:

        .macro interrupt func
        cld

        testb   $3, CS-ORIG_RAX(%rsp)
        jz      1f
        SWAPGS
        SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
        call    switch_to_thread_stack
1:

Plus I also dropped the extra switch_to_thread_stack call done in:

  x86/mm/kaiser: Prepare assembly for entry/exit CR3 switching

Because Andy's latest preparatory patch does it now:

  x86/entry/64: Use a percpu trampoline stack for IDT entries

Thanks,

	Ingo