Re: [PATCH v5 bpf 04/11] ice: work on pre-XDP prog frag count

[Maciej Fijalkowski <maciej.fijalkowski@intel.com>]

On Wed, Jan 24, 2024 at 09:37:13AM +0100, Magnus Karlsson wrote:
> On Mon, 22 Jan 2024 at 23:16, Maciej Fijalkowski
> <maciej.fijalkowski@intel.com> wrote:
> >
> > Fix an OOM panic in XDP_DRV mode when a XDP program shrinks a
> > multi-buffer packet by 4k bytes and then redirects it to an AF_XDP
> > socket.
> >
> > Since support for handling multi-buffer frames was added to XDP, usage
> > of bpf_xdp_adjust_tail() helper within XDP program can free the page
> > that given fragment occupies and in turn decrease the fragment count
> > within skb_shared_info that is embedded in xdp_buff struct. In current
> > ice driver codebase, it can become problematic when page recycling logic
> > decides not to reuse the page. In such case, __page_frag_cache_drain()
> > is used with ice_rx_buf::pagecnt_bias that was not adjusted after
> > refcount of page was changed by XDP prog which in turn does not drain
> > the refcount to 0 and page is never freed.
> >
> > To address this, let us store the count of frags before the XDP program
> > was executed on Rx ring struct. This will be used to compare with
> > current frag count from skb_shared_info embedded in xdp_buff. A smaller
> > value in the latter indicates that XDP prog freed frag(s). Then, for
> > given delta decrement pagecnt_bias for XDP_DROP verdict.
> >
> > While at it, let us also handle the EOP frag within
> > ice_set_rx_bufs_act() to make our life easier, so all of the adjustments
> > needed to be applied against freed frags are performed in the single
> > place.
> >
> > Fixes: 2fba7dc5157b ("ice: Add support for XDP multi-buffer on Rx side")
> > Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
> > ---
> >  drivers/net/ethernet/intel/ice/ice_txrx.c     | 14 ++++++---
> >  drivers/net/ethernet/intel/ice/ice_txrx.h     |  1 +
> >  drivers/net/ethernet/intel/ice/ice_txrx_lib.h | 31 +++++++++++++------
> >  3 files changed, 32 insertions(+), 14 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c
> > index 59617f055e35..1760e81379cc 100644
> > --- a/drivers/net/ethernet/intel/ice/ice_txrx.c
> > +++ b/drivers/net/ethernet/intel/ice/ice_txrx.c
> > @@ -603,9 +603,7 @@ ice_run_xdp(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp,
> >                 ret = ICE_XDP_CONSUMED;
> >         }
> >  exit:
> > -       rx_buf->act = ret;
> > -       if (unlikely(xdp_buff_has_frags(xdp)))
> > -               ice_set_rx_bufs_act(xdp, rx_ring, ret);
> > +       ice_set_rx_bufs_act(xdp, rx_ring, ret);
> >  }
> >
> >  /**
> > @@ -893,14 +891,17 @@ ice_add_xdp_frag(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp,
> >         }
> >
> >         if (unlikely(sinfo->nr_frags == MAX_SKB_FRAGS)) {
> > -               if (unlikely(xdp_buff_has_frags(xdp)))
> > -                       ice_set_rx_bufs_act(xdp, rx_ring, ICE_XDP_CONSUMED);
> > +               ice_set_rx_bufs_act(xdp, rx_ring, ICE_XDP_CONSUMED);
> >                 return -ENOMEM;
> >         }
> >
> >         __skb_fill_page_desc_noacc(sinfo, sinfo->nr_frags++, rx_buf->page,
> >                                    rx_buf->page_offset, size);
> >         sinfo->xdp_frags_size += size;
> > +       /* remember frag count before XDP prog execution; bpf_xdp_adjust_tail()
> > +        * can pop off frags but driver has to handle it on its own
> > +        */
> > +       rx_ring->nr_frags = sinfo->nr_frags;
> >
> >         if (page_is_pfmemalloc(rx_buf->page))
> >                 xdp_buff_set_frag_pfmemalloc(xdp);
> > @@ -1251,6 +1252,7 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget)
> >
> >                 xdp->data = NULL;
> >                 rx_ring->first_desc = ntc;
> > +               rx_ring->nr_frags = 0;
> >                 continue;
> >  construct_skb:
> >                 if (likely(ice_ring_uses_build_skb(rx_ring)))
> > @@ -1266,10 +1268,12 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget)
> >                                                     ICE_XDP_CONSUMED);
> >                         xdp->data = NULL;
> >                         rx_ring->first_desc = ntc;
> > +                       rx_ring->nr_frags = 0;
> >                         break;
> >                 }
> >                 xdp->data = NULL;
> >                 rx_ring->first_desc = ntc;
> > +               rx_ring->nr_frags = 0;
> 
> Are these needed? Or asked in another way, is there some way in which
> ice_set_rx_bufs_act() can be executed before ice_add_xdp_frag()? If
> not, we could remove them.

I am afraid that if you would have fragged packet followed by non-fragged
one then ice_set_rx_bufs_act() would incorrectly go over more buffers
than it was supposed to. I think we should keep those, unless I am missing
something?

> 
> Looks good otherwise.
> 
> Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
> 
> >
> >                 stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_RXE_S);
> >                 if (unlikely(ice_test_staterr(rx_desc->wb.status_error0,
> > diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.h b/drivers/net/ethernet/intel/ice/ice_txrx.h
> > index b3379ff73674..af955b0e5dc5 100644
> > --- a/drivers/net/ethernet/intel/ice/ice_txrx.h
> > +++ b/drivers/net/ethernet/intel/ice/ice_txrx.h
> > @@ -358,6 +358,7 @@ struct ice_rx_ring {
> >         struct ice_tx_ring *xdp_ring;
> >         struct ice_rx_ring *next;       /* pointer to next ring in q_vector */
> >         struct xsk_buff_pool *xsk_pool;
> > +       u32 nr_frags;
> >         dma_addr_t dma;                 /* physical address of ring */
> >         u16 rx_buf_len;
> >         u8 dcb_tc;                      /* Traffic class of ring */
> > diff --git a/drivers/net/ethernet/intel/ice/ice_txrx_lib.h b/drivers/net/ethernet/intel/ice/ice_txrx_lib.h
> > index 762047508619..afcead4baef4 100644
> > --- a/drivers/net/ethernet/intel/ice/ice_txrx_lib.h
> > +++ b/drivers/net/ethernet/intel/ice/ice_txrx_lib.h
> > @@ -12,26 +12,39 @@
> >   * act: action to store onto Rx buffers related to XDP buffer parts
> >   *
> >   * Set action that should be taken before putting Rx buffer from first frag
> > - * to one before last. Last one is handled by caller of this function as it
> > - * is the EOP frag that is currently being processed. This function is
> > - * supposed to be called only when XDP buffer contains frags.
> > + * to the last.
> >   */
> >  static inline void
> >  ice_set_rx_bufs_act(struct xdp_buff *xdp, const struct ice_rx_ring *rx_ring,
> >                     const unsigned int act)
> >  {
> > -       const struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
> > -       u32 first = rx_ring->first_desc;
> > -       u32 nr_frags = sinfo->nr_frags;
> > +       u32 sinfo_frags = xdp_get_shared_info_from_buff(xdp)->nr_frags;
> > +       u32 nr_frags = rx_ring->nr_frags + 1;
> > +       u32 idx = rx_ring->first_desc;
> >         u32 cnt = rx_ring->count;
> >         struct ice_rx_buf *buf;
> >
> >         for (int i = 0; i < nr_frags; i++) {
> > -               buf = &rx_ring->rx_buf[first];
> > +               buf = &rx_ring->rx_buf[idx];
> >                 buf->act = act;
> >
> > -               if (++first == cnt)
> > -                       first = 0;
> > +               if (++idx == cnt)
> > +                       idx = 0;
> > +       }
> > +
> > +       /* adjust pagecnt_bias on frags freed by XDP prog */
> > +       if (sinfo_frags < rx_ring->nr_frags && act == ICE_XDP_CONSUMED) {
> > +               u32 delta = rx_ring->nr_frags - sinfo_frags;
> > +
> > +               while (delta) {
> > +                       if (idx == 0)
> > +                               idx = cnt - 1;
> > +                       else
> > +                               idx--;
> > +                       buf = &rx_ring->rx_buf[idx];
> > +                       buf->pagecnt_bias--;
> > +                       delta--;
> > +               }
> >         }
> >  }
> >
> > --
> > 2.34.1
> >
> >