Re: [PATCH v2 1/1] iommu/sva: Invalidate KVA range on kernel TLB flush

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Jul 11, 2025 at 11:09:00AM +0800, Baolu Lu wrote:
> On 7/10/25 23:53, Peter Zijlstra wrote:
> > On Thu, Jul 10, 2025 at 03:54:32PM +0200, Peter Zijlstra wrote:
> > 
> > > > @@ -132,8 +136,15 @@ struct iommu_sva *iommu_sva_bind_device(struct device *dev, struct mm_struct *mm
> > > >   	if (ret)
> > > >   		goto out_free_domain;
> > > >   	domain->users = 1;
> > > > -	list_add(&domain->next, &mm->iommu_mm->sva_domains);
> > > > +	if (list_empty(&iommu_mm->sva_domains)) {
> > > > +		scoped_guard(spinlock_irqsave, &iommu_mms_lock) {
> > > > +			if (list_empty(&iommu_sva_mms))
> > > > +				static_branch_enable(&iommu_sva_present);
> > > > +			list_add(&iommu_mm->mm_list_elm, &iommu_sva_mms);
> > > > +		}
> > > > +	}
> > > > +	list_add(&domain->next, &iommu_mm->sva_domains);
> > > >   out:
> > > >   	refcount_set(&handle->users, 1);
> > > >   	mutex_unlock(&iommu_sva_lock);
> > > > @@ -175,6 +186,15 @@ void iommu_sva_unbind_device(struct iommu_sva *handle)
> > > >   		list_del(&domain->next);
> > > >   		iommu_domain_free(domain);
> > > >   	}
> > > > +
> > > > +	if (list_empty(&iommu_mm->sva_domains)) {
> > > > +		scoped_guard(spinlock_irqsave, &iommu_mms_lock) {
> > > > +			list_del(&iommu_mm->mm_list_elm);
> > > > +			if (list_empty(&iommu_sva_mms))
> > > > +				static_branch_disable(&iommu_sva_present);
> > > > +		}
> > > > +	}
> > > > +
> > > >   	mutex_unlock(&iommu_sva_lock);
> > > >   	kfree(handle);
> > > >   }
> > > 
> > > This seems an odd coding style choice; why the extra unneeded
> > > indentation? That is, what's wrong with:
> > > 
> > > 	if (list_empty()) {
> > > 		guard(spinlock_irqsave)(&iommu_mms_lock);
> > > 		list_del();
> > > 		if (list_empty()
> > > 			static_branch_disable();
> > > 	}
> > 
> > Well, for one, you can't do static_branch_{en,dis}able() from atomic
> > context...
> > 
> > Was this ever tested?
> 
> I conducted unit tests for vmalloc()/vfree() scenarios, and Yi performed
> fuzzing tests. We have not observed any warning messages. Perhaps
> static_branch_disable() is not triggered in the test cases?

Same with static_branch_enable(). These functions start with
cpus_read_lock(), which is percpu_down_read(), which has might_sleep().