On Sat, 15 Jun 2019 11:59:16 +0300
Ivan Khoronzhuk <[email protected]> wrote:
> >@@ -388,10 +411,12 @@ static void __xdp_return(void *data, struct
> >xdp_mem_info *mem, bool napi_direct,
> > /* mem->id is valid, checked in xdp_rxq_info_reg_mem_model() */
> > xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params);
> > page = virt_to_head_page(data);
> >- if (xa) {
> >+ if (likely(xa)) {
> > napi_direct &= !xdp_return_frame_no_direct();
> > page_pool_put_page(xa->page_pool, page, napi_direct);
>
> Interesting if it's synced with device "unregistration".
> I mean page dma unmap is bind to device that doesn't exist anymore but pages
> from pool of the device are in flight, so pool is not destroyed but what about
> device?. smth like device unreq todo list. Just to be sure, is it synched?
I looked through the code, and you are right. We are actually missing
to hold a reference count on struct device (for which we use dev->dma_ops).
It looks like we can simply do a get_device() and put_device(). Thus,
fairly simple to fix.
I'll add this fix as a separate patch, as it is a separate issue/bug
that also needs correction... Thanks for catching this!
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
