Re: [PATCH 20/43] x86/entry: Clean up SYSENTER_stack code

[Thomas Gleixner]

On Sat, 25 Nov 2017, Andy Lutomirski wrote:
> > On Nov 25, 2017, at 9:50 AM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> > 
> > On Sat, 25 Nov 2017, Borislav Petkov wrote:
> >>> -
> >>> +    wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_SYSENTER_stack(cpu) + 1), 0);
> >>>    wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
> >> 
> >> Right, so we have now two TSS thingies, AFAICT:
> >> 
> >>    tss = &per_cpu(cpu_tss, cpu);
> >> 
> >> which is cpu_tss and then indirectly, we have also:
> >> 
> >>    &get_cpu_entry_area((cpu))->tss
> >> 
> >> And those are two different things in my guest here:
> >> 
> >> [    0.044002] tss: 0xf5747000
> >> [    0.044706] entry area tss: 0xffef1000
> >> 
> >> What is the logic here? We carry two TSSs per CPU - one which is RO
> >> for the entry area and the other is the actual cpu_tss thing? Or am I
> >> misreading it?
> > 
> > entry area tss is a alias mapping of cpu_tss
> > 
> > +       set_percpu_fixmap_pages(get_cpu_entry_area_index(cpu, tss),
> > +                               &per_cpu(cpu_tss, cpu),
> > +                               sizeof(struct tss_struct) / PAGE_SIZE,
> > +                               PAGE_KERNEL);
> > 
> 
> Exactly.  And, in the patch I haven't emailed, the alias is RO on x86_64.

Btw, I don't think you have to worry about making it RO. The SDM is blurry
about that for 64bit. RW is a must for 32bit though.

> Maybe I should rename cpu_tss to cpu_tss_rw in that patch.

For clarity that would be nice.

Thanks,

	tglx