Re: [PATCH v3 1/1] iommu/sva: Invalidate KVA range on kernel TLB flush

[Ethan Zhao]

On 8/11/2025 8:57 PM, Jason Gunthorpe wrote:
> On Fri, Aug 08, 2025 at 01:15:12PM +0800, Baolu Lu wrote:
> > +static void kernel_pte_work_func(struct work_struct *work)
> > +{
> > +	struct ptdesc *ptdesc, *next;
> > +
> > +	iommu_sva_invalidate_kva_range(0, TLB_FLUSH_ALL);
> > +
> > +	guard(spinlock)(&kernel_pte_work.lock);
> > +	list_for_each_entry_safe(ptdesc, next, &kernel_pte_work.list, pt_list) {
> > +		list_del_init(&ptdesc->pt_list);
> > +		pagetable_dtor_free(ptdesc);
> > +	}
>
> Do a list_move from kernel_pte_work.list to an on-stack list head and
> then immediately release the lock. No reason to hold the spinock while
> doing frees, also no reason to do list_del_init, that memory probably
> gets zerod in pagetable_dtor_free() 
Yep，using guard(spinlock)() for scope-bound lock management sacrifices
fine-grained control over the protected area. If offers convenience at
the cost of precision.

Out of my bias, calling it sluggard(spinlock)() might be proper.

Thanks，
Ethan
> Jason