Re: [PATCH -next for tip:x86/pti] x86/tlb: drop unneeded local vars in enable_l1d_flush_for_task()

[Lukas Bulwahn]

On Tue, 29 Sep 2020, Peter Zijlstra wrote:

> On Mon, Sep 28, 2020 at 02:44:57PM +0200, Lukas Bulwahn wrote:
> > diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
> > index 6b0f4c88b07c..90515c04d90a 100644
> > --- a/arch/x86/mm/tlb.c
> > +++ b/arch/x86/mm/tlb.c
> > @@ -316,7 +316,7 @@ EXPORT_SYMBOL_GPL(leave_mm);
> >  
> >  int enable_l1d_flush_for_task(struct task_struct *tsk)
> >  {
> > -	int cpu, ret = 0, i;
> > +	int i;
> >  
> >  	/*
> >  	 * Do not enable L1D_FLUSH_OUT if
> > @@ -329,7 +329,7 @@ int enable_l1d_flush_for_task(struct task_struct *tsk)
> >  			!static_cpu_has(X86_FEATURE_FLUSH_L1D))
> >  		return -EINVAL;
> >  
> > -	cpu = get_cpu();
> > +	get_cpu();
> >  
> >  	for_each_cpu(i, &tsk->cpus_mask) {
> >  		if (cpu_data(i).smt_active == true) {
> > @@ -340,7 +340,7 @@ int enable_l1d_flush_for_task(struct task_struct *tsk)
> >  
> >  	set_ti_thread_flag(&tsk->thread_info, TIF_SPEC_L1D_FLUSH);
> >  	put_cpu();
> > -	return ret;
> > +	return 0;
> >  }
> 
> If you don't use the return value of get_cpu(), then change it over to
> preempt_{dis,en}able(), but this got me looking at the function, wtf is
> that garbage supposed to do in the first place

I also thought that preempt_{dis,en}able() would do, but thought maybe 
Balbir just considered {get,put}_cpu stylistically nicer... so I stayed 
with the functions as-is.

> 
> What do we need to disable preemption for?
>

I have no clue... not a good premise for touching the code, but I just 
wanted to make clang-analyzer happy without modifying any semantics.

Balbir, can you help out here? What was your intent?
 
> Please explain the desired semantics against sched_setaffinity().
> 

I am happy to send a proper v2 once I understand if disabling preemption 
is required and the preempt_{dis,en}able() function are preferred to the 
{get,put}_cpu functions.

Lukas