From 743bbd93cf29f653fae0e1416a31f03231689911 Mon Sep 17 00:00:00 2001 From: Maciej Fijalkowski Date: Thu, 23 Jan 2025 16:01:16 +0100 Subject: [PATCH 1/3] ice: put Rx buffers after being done with current frame Introduce a new helper ice_put_rx_mbuf() that will go through gathered frags from current frame and will call ice_put_rx_buf() on them. Current logic that was supposed to simplify and optimize the driver where we go through a batch of all buffers processed in current NAPI instance turned out to be broken for jumbo frames and very heavy load that was coming from both multi-thread iperf and nginx/wrk pair between server and client. The delay introduced by approach that we are dropping is simply too big and we need to take the decision regarding page recycling/releasing as quick as we can. While at it, address an error path of ice_add_xdp_frag() - we were missing buffer putting from day 1 there. As a nice side effect we get rid of annoying and repetitive three-liner: xdp->data = NULL; rx_ring->first_desc = ntc; rx_ring->nr_frags = 0; by embedding it within introduced routine. Fixes: 1dc1a7e7f410 ("ice: Centrallize Rx buffer recycling") Reported-and-tested-by: Xu Du Reviewed-by: Przemek Kitszel Reviewed-by: Simon Horman Co-developed-by: Jacob Keller Signed-off-by: Jacob Keller Signed-off-by: Maciej Fijalkowski Tested-by: Chandan Kumar Rout (A Contingent Worker at Intel) Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/ice/ice_txrx.c | 79 ++++++++++++++--------- 1 file changed, 50 insertions(+), 29 deletions(-) diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c index 5d2d7736fd5f..e173d9c98988 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx.c +++ b/drivers/net/ethernet/intel/ice/ice_txrx.c @@ -1103,6 +1103,49 @@ ice_put_rx_buf(struct ice_rx_ring *rx_ring, struct ice_rx_buf *rx_buf) rx_buf->page = NULL; } +/** + * ice_put_rx_mbuf - ice_put_rx_buf() caller, for all frame frags + * @rx_ring: Rx ring with all the auxiliary data + * @xdp: XDP buffer carrying linear + frags part + * @xdp_xmit: XDP_TX/XDP_REDIRECT verdict storage + * @ntc: a current next_to_clean value to be stored at rx_ring + * + * Walk through gathered fragments and satisfy internal page + * recycle mechanism; we take here an action related to verdict + * returned by XDP program; + */ +static void ice_put_rx_mbuf(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp, + u32 *xdp_xmit, u32 ntc) +{ + u32 nr_frags = rx_ring->nr_frags + 1; + u32 idx = rx_ring->first_desc; + u32 cnt = rx_ring->count; + struct ice_rx_buf *buf; + int i; + + for (i = 0; i < nr_frags; i++) { + buf = &rx_ring->rx_buf[idx]; + + if (buf->act & (ICE_XDP_TX | ICE_XDP_REDIR)) { + ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz); + *xdp_xmit |= buf->act; + } else if (buf->act & ICE_XDP_CONSUMED) { + buf->pagecnt_bias++; + } else if (buf->act == ICE_XDP_PASS) { + ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz); + } + + ice_put_rx_buf(rx_ring, buf); + + if (++idx == cnt) + idx = 0; + } + + xdp->data = NULL; + rx_ring->first_desc = ntc; + rx_ring->nr_frags = 0; +} + /** * ice_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf * @rx_ring: Rx descriptor ring to transact packets on @@ -1120,7 +1163,6 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) unsigned int total_rx_bytes = 0, total_rx_pkts = 0; unsigned int offset = rx_ring->rx_offset; struct xdp_buff *xdp = &rx_ring->xdp; - u32 cached_ntc = rx_ring->first_desc; struct ice_tx_ring *xdp_ring = NULL; struct bpf_prog *xdp_prog = NULL; u32 ntc = rx_ring->next_to_clean; @@ -1128,7 +1170,6 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) u32 xdp_xmit = 0; u32 cached_ntu; bool failure; - u32 first; xdp_prog = READ_ONCE(rx_ring->xdp_prog); if (xdp_prog) { @@ -1190,6 +1231,7 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) xdp_prepare_buff(xdp, hard_start, offset, size, !!offset); xdp_buff_clear_frags_flag(xdp); } else if (ice_add_xdp_frag(rx_ring, xdp, rx_buf, size)) { + ice_put_rx_mbuf(rx_ring, xdp, NULL, ntc); break; } if (++ntc == cnt) @@ -1205,9 +1247,8 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) total_rx_bytes += xdp_get_buff_len(xdp); total_rx_pkts++; - xdp->data = NULL; - rx_ring->first_desc = ntc; - rx_ring->nr_frags = 0; + ice_put_rx_mbuf(rx_ring, xdp, &xdp_xmit, ntc); + continue; construct_skb: if (likely(ice_ring_uses_build_skb(rx_ring))) @@ -1221,14 +1262,11 @@ construct_skb: if (unlikely(xdp_buff_has_frags(xdp))) ice_set_rx_bufs_act(xdp, rx_ring, 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; + ice_put_rx_mbuf(rx_ring, xdp, &xdp_xmit, ntc); + + if (!skb) + break; stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_RXE_S); if (unlikely(ice_test_staterr(rx_desc->wb.status_error0, @@ -1257,23 +1295,6 @@ construct_skb: total_rx_pkts++; } - first = rx_ring->first_desc; - while (cached_ntc != first) { - struct ice_rx_buf *buf = &rx_ring->rx_buf[cached_ntc]; - - if (buf->act & (ICE_XDP_TX | ICE_XDP_REDIR)) { - ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz); - xdp_xmit |= buf->act; - } else if (buf->act & ICE_XDP_CONSUMED) { - buf->pagecnt_bias++; - } else if (buf->act == ICE_XDP_PASS) { - ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz); - } - - ice_put_rx_buf(rx_ring, buf); - if (++cached_ntc >= cnt) - cached_ntc = 0; - } rx_ring->next_to_clean = ntc; /* return up to cleaned_count buffers to hardware */ failure = ice_alloc_rx_bufs(rx_ring, ICE_RX_DESC_UNUSED(rx_ring)); From 11c4aa074d547d825b19cd8d9f288254d89d805c Mon Sep 17 00:00:00 2001 From: Maciej Fijalkowski Date: Thu, 23 Jan 2025 16:01:17 +0100 Subject: [PATCH 2/3] ice: gather page_count()'s of each frag right before XDP prog call If we store the pgcnt on few fragments while being in the middle of gathering the whole frame and we stumbled upon DD bit not being set, we terminate the NAPI Rx processing loop and come back later on. Then on next NAPI execution we work on previously stored pgcnt. Imagine that second half of page was used actively by networking stack and by the time we came back, stack is not busy with this page anymore and decremented the refcnt. The page reuse algorithm in this case should be good to reuse the page but given the old refcnt it will not do so and attempt to release the page via page_frag_cache_drain() with pagecnt_bias used as an arg. This in turn will result in negative refcnt on struct page, which was initially observed by Xu Du. Therefore, move the page count storage from ice_get_rx_buf() to a place where we are sure that whole frame has been collected, but before calling XDP program as it internally can also change the page count of fragments belonging to xdp_buff. Fixes: ac0753391195 ("ice: Store page count inside ice_rx_buf") Reported-and-tested-by: Xu Du Reviewed-by: Przemek Kitszel Reviewed-by: Simon Horman Co-developed-by: Jacob Keller Signed-off-by: Jacob Keller Signed-off-by: Maciej Fijalkowski Tested-by: Chandan Kumar Rout (A Contingent Worker at Intel) Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/ice/ice_txrx.c | 27 ++++++++++++++++++++++- 1 file changed, 26 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c index e173d9c98988..cf46bcf143b4 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx.c +++ b/drivers/net/ethernet/intel/ice/ice_txrx.c @@ -924,7 +924,6 @@ ice_get_rx_buf(struct ice_rx_ring *rx_ring, const unsigned int size, struct ice_rx_buf *rx_buf; rx_buf = &rx_ring->rx_buf[ntc]; - rx_buf->pgcnt = page_count(rx_buf->page); prefetchw(rx_buf->page); if (!size) @@ -940,6 +939,31 @@ ice_get_rx_buf(struct ice_rx_ring *rx_ring, const unsigned int size, return rx_buf; } +/** + * ice_get_pgcnts - grab page_count() for gathered fragments + * @rx_ring: Rx descriptor ring to store the page counts on + * + * This function is intended to be called right before running XDP + * program so that the page recycling mechanism will be able to take + * a correct decision regarding underlying pages; this is done in such + * way as XDP program can change the refcount of page + */ +static void ice_get_pgcnts(struct ice_rx_ring *rx_ring) +{ + u32 nr_frags = rx_ring->nr_frags + 1; + u32 idx = rx_ring->first_desc; + struct ice_rx_buf *rx_buf; + u32 cnt = rx_ring->count; + + for (int i = 0; i < nr_frags; i++) { + rx_buf = &rx_ring->rx_buf[idx]; + rx_buf->pgcnt = page_count(rx_buf->page); + + if (++idx == cnt) + idx = 0; + } +} + /** * ice_build_skb - Build skb around an existing buffer * @rx_ring: Rx descriptor ring to transact packets on @@ -1241,6 +1265,7 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) if (ice_is_non_eop(rx_ring, rx_desc)) continue; + ice_get_pgcnts(rx_ring); ice_run_xdp(rx_ring, xdp, xdp_prog, xdp_ring, rx_buf, rx_desc); if (rx_buf->act == ICE_XDP_PASS) goto construct_skb; From 468a1952df78f65c5991b7ac885c8b5b7dd87bab Mon Sep 17 00:00:00 2001 From: Maciej Fijalkowski Date: Thu, 23 Jan 2025 16:01:18 +0100 Subject: [PATCH 3/3] ice: stop storing XDP verdict within ice_rx_buf Idea behind having ice_rx_buf::act was to simplify and speed up the Rx data path by walking through buffers that were representing cleaned HW Rx descriptors. Since it caused us a major headache recently and we rolled back to old approach that 'puts' Rx buffers right after running XDP prog/creating skb, this is useless now and should be removed. Get rid of ice_rx_buf::act and related logic. We still need to take care of a corner case where XDP program releases a particular fragment. Make ice_run_xdp() to return its result and use it within ice_put_rx_mbuf(). Fixes: 2fba7dc5157b ("ice: Add support for XDP multi-buffer on Rx side") Reviewed-by: Przemek Kitszel Reviewed-by: Simon Horman Signed-off-by: Maciej Fijalkowski Tested-by: Chandan Kumar Rout (A Contingent Worker at Intel) Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/ice/ice_txrx.c | 62 +++++++++++-------- drivers/net/ethernet/intel/ice/ice_txrx.h | 1 - drivers/net/ethernet/intel/ice/ice_txrx_lib.h | 43 ------------- 3 files changed, 36 insertions(+), 70 deletions(-) diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c index cf46bcf143b4..9c9ea4c1b93b 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx.c +++ b/drivers/net/ethernet/intel/ice/ice_txrx.c @@ -527,15 +527,14 @@ err: * @xdp: xdp_buff used as input to the XDP program * @xdp_prog: XDP program to run * @xdp_ring: ring to be used for XDP_TX action - * @rx_buf: Rx buffer to store the XDP action * @eop_desc: Last descriptor in packet to read metadata from * * Returns any of ICE_XDP_{PASS, CONSUMED, TX, REDIR} */ -static void +static u32 ice_run_xdp(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp, struct bpf_prog *xdp_prog, struct ice_tx_ring *xdp_ring, - struct ice_rx_buf *rx_buf, union ice_32b_rx_flex_desc *eop_desc) + union ice_32b_rx_flex_desc *eop_desc) { unsigned int ret = ICE_XDP_PASS; u32 act; @@ -574,7 +573,7 @@ out_failure: ret = ICE_XDP_CONSUMED; } exit: - ice_set_rx_bufs_act(xdp, rx_ring, ret); + return ret; } /** @@ -860,10 +859,8 @@ ice_add_xdp_frag(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp, xdp_buff_set_frags_flag(xdp); } - if (unlikely(sinfo->nr_frags == MAX_SKB_FRAGS)) { - ice_set_rx_bufs_act(xdp, rx_ring, ICE_XDP_CONSUMED); + if (unlikely(sinfo->nr_frags == MAX_SKB_FRAGS)) return -ENOMEM; - } __skb_fill_page_desc_noacc(sinfo, sinfo->nr_frags++, rx_buf->page, rx_buf->page_offset, size); @@ -1075,12 +1072,12 @@ ice_construct_skb(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp) rx_buf->page_offset + headlen, size, xdp->frame_sz); } else { - /* buffer is unused, change the act that should be taken later - * on; data was copied onto skb's linear part so there's no + /* buffer is unused, restore biased page count in Rx buffer; + * data was copied onto skb's linear part so there's no * need for adjusting page offset and we can reuse this buffer * as-is */ - rx_buf->act = ICE_SKB_CONSUMED; + rx_buf->pagecnt_bias++; } if (unlikely(xdp_buff_has_frags(xdp))) { @@ -1133,29 +1130,34 @@ ice_put_rx_buf(struct ice_rx_ring *rx_ring, struct ice_rx_buf *rx_buf) * @xdp: XDP buffer carrying linear + frags part * @xdp_xmit: XDP_TX/XDP_REDIRECT verdict storage * @ntc: a current next_to_clean value to be stored at rx_ring + * @verdict: return code from XDP program execution * * Walk through gathered fragments and satisfy internal page * recycle mechanism; we take here an action related to verdict * returned by XDP program; */ static void ice_put_rx_mbuf(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp, - u32 *xdp_xmit, u32 ntc) + u32 *xdp_xmit, u32 ntc, u32 verdict) { u32 nr_frags = rx_ring->nr_frags + 1; u32 idx = rx_ring->first_desc; u32 cnt = rx_ring->count; + u32 post_xdp_frags = 1; struct ice_rx_buf *buf; int i; - for (i = 0; i < nr_frags; i++) { + if (unlikely(xdp_buff_has_frags(xdp))) + post_xdp_frags += xdp_get_shared_info_from_buff(xdp)->nr_frags; + + for (i = 0; i < post_xdp_frags; i++) { buf = &rx_ring->rx_buf[idx]; - if (buf->act & (ICE_XDP_TX | ICE_XDP_REDIR)) { + if (verdict & (ICE_XDP_TX | ICE_XDP_REDIR)) { ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz); - *xdp_xmit |= buf->act; - } else if (buf->act & ICE_XDP_CONSUMED) { + *xdp_xmit |= verdict; + } else if (verdict & ICE_XDP_CONSUMED) { buf->pagecnt_bias++; - } else if (buf->act == ICE_XDP_PASS) { + } else if (verdict == ICE_XDP_PASS) { ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz); } @@ -1164,6 +1166,17 @@ static void ice_put_rx_mbuf(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp, if (++idx == cnt) idx = 0; } + /* handle buffers that represented frags released by XDP prog; + * for these we keep pagecnt_bias as-is; refcount from struct page + * has been decremented within XDP prog and we do not have to increase + * the biased refcnt + */ + for (; i < nr_frags; i++) { + buf = &rx_ring->rx_buf[idx]; + ice_put_rx_buf(rx_ring, buf); + if (++idx == cnt) + idx = 0; + } xdp->data = NULL; rx_ring->first_desc = ntc; @@ -1190,9 +1203,9 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) struct ice_tx_ring *xdp_ring = NULL; struct bpf_prog *xdp_prog = NULL; u32 ntc = rx_ring->next_to_clean; + u32 cached_ntu, xdp_verdict; u32 cnt = rx_ring->count; u32 xdp_xmit = 0; - u32 cached_ntu; bool failure; xdp_prog = READ_ONCE(rx_ring->xdp_prog); @@ -1255,7 +1268,7 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) xdp_prepare_buff(xdp, hard_start, offset, size, !!offset); xdp_buff_clear_frags_flag(xdp); } else if (ice_add_xdp_frag(rx_ring, xdp, rx_buf, size)) { - ice_put_rx_mbuf(rx_ring, xdp, NULL, ntc); + ice_put_rx_mbuf(rx_ring, xdp, NULL, ntc, ICE_XDP_CONSUMED); break; } if (++ntc == cnt) @@ -1266,13 +1279,13 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) continue; ice_get_pgcnts(rx_ring); - ice_run_xdp(rx_ring, xdp, xdp_prog, xdp_ring, rx_buf, rx_desc); - if (rx_buf->act == ICE_XDP_PASS) + xdp_verdict = ice_run_xdp(rx_ring, xdp, xdp_prog, xdp_ring, rx_desc); + if (xdp_verdict == ICE_XDP_PASS) goto construct_skb; total_rx_bytes += xdp_get_buff_len(xdp); total_rx_pkts++; - ice_put_rx_mbuf(rx_ring, xdp, &xdp_xmit, ntc); + ice_put_rx_mbuf(rx_ring, xdp, &xdp_xmit, ntc, xdp_verdict); continue; construct_skb: @@ -1283,12 +1296,9 @@ construct_skb: /* exit if we failed to retrieve a buffer */ if (!skb) { rx_ring->ring_stats->rx_stats.alloc_page_failed++; - rx_buf->act = ICE_XDP_CONSUMED; - if (unlikely(xdp_buff_has_frags(xdp))) - ice_set_rx_bufs_act(xdp, rx_ring, - ICE_XDP_CONSUMED); + xdp_verdict = ICE_XDP_CONSUMED; } - ice_put_rx_mbuf(rx_ring, xdp, &xdp_xmit, ntc); + ice_put_rx_mbuf(rx_ring, xdp, &xdp_xmit, ntc, xdp_verdict); if (!skb) break; diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.h b/drivers/net/ethernet/intel/ice/ice_txrx.h index cb347c852ba9..806bce701df3 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx.h +++ b/drivers/net/ethernet/intel/ice/ice_txrx.h @@ -201,7 +201,6 @@ struct ice_rx_buf { struct page *page; unsigned int page_offset; unsigned int pgcnt; - unsigned int act; unsigned int pagecnt_bias; }; diff --git a/drivers/net/ethernet/intel/ice/ice_txrx_lib.h b/drivers/net/ethernet/intel/ice/ice_txrx_lib.h index 79f960c6680d..6cf32b404127 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx_lib.h +++ b/drivers/net/ethernet/intel/ice/ice_txrx_lib.h @@ -5,49 +5,6 @@ #define _ICE_TXRX_LIB_H_ #include "ice.h" -/** - * ice_set_rx_bufs_act - propagate Rx buffer action to frags - * @xdp: XDP buffer representing frame (linear and frags part) - * @rx_ring: Rx ring struct - * 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 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) -{ - 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[idx]; - buf->act = act; - - 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--; - } - } -} - /** * ice_test_staterr - tests bits in Rx descriptor status and error fields * @status_err_n: Rx descriptor status_error0 or status_error1 bits