Re: [PATCH 5/8] x86/mmu: Add mm-based PASID refcounting

[Andy Lutomirski]

On Wed, Sep 29, 2021, at 10:41 AM, Luck, Tony wrote:
> On Wed, Sep 29, 2021 at 07:15:53PM +0200, Thomas Gleixner wrote:
>> On Wed, Sep 29 2021 at 09:59, Andy Lutomirski wrote:
>> > On 9/29/21 05:28, Thomas Gleixner wrote:
>> >> Looking at that patch again, none of this muck in fpu__pasid_write() is
>> >> required at all. The whole exception fixup is:
>> >> 
>> >>      if (!user_mode(regs))
>> >>               return false;
>> >> 
>> >>      if (!current->mm->pasid)
>> >>               return false;
>> >> 
>> >>      if (current->pasid_activated)
>> >>      	     return false;
>> >
>> > <-- preemption or BH here: kaboom.
>> 
>> Sigh, this had obviously to run in the early portion of #GP, i.e. before
>> enabling interrupts.
>
> Like this? Obviously with some comment about why this is being done.
>
> diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
> index a58800973aed..a848a59291e7 100644
> --- a/arch/x86/kernel/traps.c
> +++ b/arch/x86/kernel/traps.c
> @@ -536,6 +536,12 @@ DEFINE_IDTENTRY_ERRORCODE(exc_general_protection)
>  	unsigned long gp_addr;
>  	int ret;
> 
> +	if (user_mode(regs) && current->mm->pasid && !current->pasid_activated) {
> +		current->pasid_activated = 1;
> +		wrmsrl(MSR_IA32_PASID, current->mm->pasid | MSR_IA32_PASID_VALID);
> +		return;
> +	}
> +

This could do with a WARN_ON_ONCE(TIF_NEED_LOAD_FPU) imo.

Is instrumentation allowed to touch FPU state?

>  	cond_local_irq_enable(regs);
> 
>  	if (static_cpu_has(X86_FEATURE_UMIP)) {
>
> -Tony