Re: [PATCH 4/8] x86/traps: Demand-populate PASID MSR via #GP

[Luck, Tony]

On Fri, Sep 24, 2021 at 03:37:22PM +0200, Peter Zijlstra wrote:
> On Thu, Sep 23, 2021 at 10:14:42AM -0700, Luck, Tony wrote:
> > On Wed, Sep 22, 2021 at 11:07:22PM +0200, Peter Zijlstra wrote:
> > > On Mon, Sep 20, 2021 at 07:23:45PM +0000, Fenghua Yu wrote:
> > > > @@ -538,6 +547,9 @@ DEFINE_IDTENTRY_ERRORCODE(exc_general_protection)
> > > >  
> > > >  	cond_local_irq_enable(regs);
> > > >  
> > > > +	if (user_mode(regs) && fixup_pasid_exception())
> > > > +		goto exit;
> > > > +
> > 
> > > So you're eating any random #GP that might or might not be PASID
> > > related. And all that witout a comment... Enlighten?
> > 
> > This is moderately well commented inside the fixup_pasid_exception()
> > function. Another copy of the comments here at the call-site seems
> > overkill.
> 
> +static bool fixup_pasid_exception(void)
> +{
> +       if (!cpu_feature_enabled(X86_FEATURE_ENQCMD))
> +               return false;
> +
> +       return __fixup_pasid_exception();
> +}
> 
> /me goes looking for comments in that function, lemme get out the
> electron microscope, because I can't seem to spot them with the naked
> eye.

If you have ctags installed then a ctrl-] on that
__fixup_pasid_exception() will take you to the function with
the comments. No electron microscope needed.

+
+/*
+ * Try to figure out if there is a PASID MSR value to propagate to the
+ * thread taking the #GP.
+ */
+bool __fixup_pasid_exception(void)
+{
+       u32 pasid;
+
+       /*
+        * This function is called only when this #GP was triggered from user
+        * space. So the mm cannot be NULL.
+        */
+       pasid = current->mm->pasid;
+
+       /* If no PASID is allocated, there is nothing to propagate. */
+       if (pasid == PASID_DISABLED)
+               return false;
+
+       /*
+        * If the current task already has a valid PASID MSR, then the #GP
+        * fault must be for some non-ENQCMD related reason.
+        */
+       if (current->has_valid_pasid)
+               return false;
+
+       /* Fix up the MSR by the PASID in the mm. */
+       fpu__pasid_write(pasid);
+       current->has_valid_pasid = 1;
+
+       return true;
+}

-Tony