On Wed, Jan 20, 2021 at 09:13:20PM -0800, Jakub Kicinski wrote:
> On Thu, 21 Jan 2021 13:09:10 +0800 Kevin Hao wrote:
> > On Wed, Jan 20, 2021 at 08:59:14PM -0800, Jakub Kicinski wrote:
> > > On Thu, 21 Jan 2021 12:20:35 +0800 Kevin Hao wrote:
> > > > Hmm, why not?
> > > > buf = napi_alloc_frag(pool->rbsize + 128);
> > > > buf = PTR_ALIGN(buf, 128);
> > >
> > > I'd keep the aligning in the driver until there are more users
> > > needing this but yes, I agree, aligning the page frag buffers
> > > seems like a much better fix.
> >
> > It seems that the DPAA2 driver also need this
> > (drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c):
> > /* Prepare the HW SGT structure */
> > sgt_buf_size = priv->tx_data_offset +
> > sizeof(struct dpaa2_sg_entry) * num_dma_bufs;
> > sgt_buf = napi_alloc_frag(sgt_buf_size + DPAA2_ETH_TX_BUF_ALIGN);
> > if (unlikely(!sgt_buf)) {
> > err = -ENOMEM;
> > goto sgt_buf_alloc_failed;
> > }
> > sgt_buf = PTR_ALIGN(sgt_buf, DPAA2_ETH_TX_BUF_ALIGN);
>
> We can fix them both up as a follow up in net-next, then?
>
> Let's keep the patch small and local for the fix.OK, I will send a patch to align the buffer in the octeontx2 driver and then introduce the napi_alloc_frag_align() for the net-next. Thanks, Kevin
signature.asc
Description: PGP signature
