| Message ID | 20230803182038.2646541-1-aleksander.lobakin@intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | page_pool: a couple of assorted optimizations | expand |
On Thu, 3 Aug 2023 20:20:32 +0200 Alexander Lobakin wrote: > That initially was a spin-off of the IAVF PP series[0], but has grown > (and shrunk) since then a bunch. In fact, it consists of three > semi-independent blocks: > > * #1-2: Compile-time optimization. Split page_pool.h into 2 headers to > not overbloat the consumers not needing complex inline helpers and > then stop including it in skbuff.h at all. The first patch is also > prereq for the whole series. > * #3: Improve cacheline locality for users of the Page Pool frag API. > * #4-6: Use direct cache recycling more aggressively, when it is safe > obviously. In addition, make sure nobody wants to use Page Pool API > with disabled interrupts. > > Patches #1 and #5 are authored by Yunsheng and Jakub respectively, with > small modifications from my side as per ML discussions. > For the perf numbers for #3-6, please see individual commit messages. Our scheming didn't help much, the series also conflicts with the net/xdp.h includes which came in via bpf-next :(
From: Jakub Kicinski <kuba@kernel.org> Date: Thu, 3 Aug 2023 18:21:21 -0700 > On Thu, 3 Aug 2023 20:20:32 +0200 Alexander Lobakin wrote: >> That initially was a spin-off of the IAVF PP series[0], but has grown >> (and shrunk) since then a bunch. In fact, it consists of three >> semi-independent blocks: >> >> * #1-2: Compile-time optimization. Split page_pool.h into 2 headers to >> not overbloat the consumers not needing complex inline helpers and >> then stop including it in skbuff.h at all. The first patch is also >> prereq for the whole series. >> * #3: Improve cacheline locality for users of the Page Pool frag API. >> * #4-6: Use direct cache recycling more aggressively, when it is safe >> obviously. In addition, make sure nobody wants to use Page Pool API >> with disabled interrupts. >> >> Patches #1 and #5 are authored by Yunsheng and Jakub respectively, with >> small modifications from my side as per ML discussions. >> For the perf numbers for #3-6, please see individual commit messages. > > Our scheming didn't help much, the series also conflicts > with the net/xdp.h includes which came in via bpf-next :( Haha that happens. I rechecked everything before sending, but well, we need a spinlock or what :D I'll send v4 in a couple hours, so that ~22 hrs will pass since v3, sounds okay? Thanks, Olek
That initially was a spin-off of the IAVF PP series[0], but has grown (and shrunk) since then a bunch. In fact, it consists of three semi-independent blocks: * #1-2: Compile-time optimization. Split page_pool.h into 2 headers to not overbloat the consumers not needing complex inline helpers and then stop including it in skbuff.h at all. The first patch is also prereq for the whole series. * #3: Improve cacheline locality for users of the Page Pool frag API. * #4-6: Use direct cache recycling more aggressively, when it is safe obviously. In addition, make sure nobody wants to use Page Pool API with disabled interrupts. Patches #1 and #5 are authored by Yunsheng and Jakub respectively, with small modifications from my side as per ML discussions. For the perf numbers for #3-6, please see individual commit messages. Also available on my GH with many more Page Pool goodies[1]. [0] https://lore.kernel.org/netdev/20230530150035.1943669-1-aleksander.lobakin@intel.com [1] https://github.com/alobakin/linux/commits/iavf-pp-frag Alexander Lobakin (4): net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h> page_pool: place frag_* fields in one cacheline net: skbuff: avoid accessing page_pool if !napi_safe when returning page net: skbuff: always try to recycle PP pages directly when in softirq Jakub Kicinski (1): page_pool: add a lockdep check for recycling in hardirq Yunsheng Lin (1): page_pool: split types and declarations from page_pool.h MAINTAINERS | 2 +- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 2 +- drivers/net/ethernet/broadcom/bnxt/bnxt_xdp.c | 2 +- drivers/net/ethernet/engleder/tsnep_main.c | 1 + drivers/net/ethernet/freescale/fec_main.c | 1 + .../net/ethernet/hisilicon/hns3/hns3_enet.c | 1 + .../net/ethernet/hisilicon/hns3/hns3_enet.h | 2 +- drivers/net/ethernet/marvell/mvneta.c | 2 +- drivers/net/ethernet/marvell/mvpp2/mvpp2.h | 2 +- .../net/ethernet/marvell/mvpp2/mvpp2_main.c | 1 + .../marvell/octeontx2/nic/otx2_common.c | 1 + .../ethernet/marvell/octeontx2/nic/otx2_pf.c | 1 + drivers/net/ethernet/mediatek/mtk_eth_soc.c | 1 + drivers/net/ethernet/mediatek/mtk_eth_soc.h | 2 +- .../ethernet/mellanox/mlx5/core/en/params.c | 1 + .../net/ethernet/mellanox/mlx5/core/en/trap.c | 1 - .../net/ethernet/mellanox/mlx5/core/en/xdp.c | 1 + .../net/ethernet/mellanox/mlx5/core/en_main.c | 2 +- .../net/ethernet/mellanox/mlx5/core/en_rx.c | 2 +- .../ethernet/mellanox/mlx5/core/en_stats.c | 2 +- .../ethernet/microchip/lan966x/lan966x_fdma.c | 1 + .../ethernet/microchip/lan966x/lan966x_main.h | 2 +- drivers/net/ethernet/socionext/netsec.c | 2 +- drivers/net/ethernet/stmicro/stmmac/stmmac.h | 2 +- .../net/ethernet/stmicro/stmmac/stmmac_main.c | 1 + drivers/net/ethernet/ti/cpsw.c | 2 +- drivers/net/ethernet/ti/cpsw_new.c | 2 +- drivers/net/ethernet/ti/cpsw_priv.c | 2 +- drivers/net/ethernet/wangxun/libwx/wx_lib.c | 2 +- drivers/net/veth.c | 2 +- drivers/net/wireless/mediatek/mt76/mac80211.c | 1 - drivers/net/wireless/mediatek/mt76/mt76.h | 1 + drivers/net/xen-netfront.c | 2 +- include/linux/lockdep.h | 7 + include/linux/skbuff.h | 5 +- .../net/{page_pool.h => page_pool/helpers.h} | 246 +----------------- include/net/page_pool/types.h | 236 +++++++++++++++++ include/trace/events/page_pool.h | 2 +- net/bpf/test_run.c | 2 +- net/core/page_pool.c | 43 +-- net/core/skbuff.c | 49 +++- net/core/xdp.c | 2 +- 42 files changed, 337 insertions(+), 307 deletions(-) rename include/net/{page_pool.h => page_pool/helpers.h} (50%) create mode 100644 include/net/page_pool/types.h --- From v2[2]: * just rebase on top of Jakub's kdoc changes landed last minute. From v1[3]: * move the "avoid calling no-op DMA sync ops" piece out of the series -- will join some other or transform into something else (Jakub et al.); * #1: restore accidentally removed path in MAINTAINERS (Yunsheng); * #1: prefer `include/net/page_pool/` over `include/net/page_pool/*` in MAINTAINERS (Yunsheng); * #2: rename page_pool_return_skb_page() to napi_pp_put_page() -- the old name seems to not make any sense for some time already (Yunsheng); * #2: guard napi_pp_put_page() with CONFIG_PAGE_POOL in skbuff.c (to not compile it when there are no users) (Yunsheng). From RFC v2[4]: * drop the dependency on the hybrid allocation series (and thus the "RFC" prefix) -- it wasn't a strict dep and it's not in the trees yet; * add [slightly reworked] Yunsheng's patch which splits page_pool.h into 2 headers -- merge conflict hell otherwise. Also fix a typo while nobody looks (Simon); * #3 (former #2): word the commitmsg a bit better, mention the main reason for the change more clearly (Ilias); * add Jakub's hardirq assertion as a prereq for the last patch; * #9 (former #7): add comment mentioning that the hardirq case is not checked due to the assertion checking it later (yes, it is illegal to use Page Pool with the interrupts disabled or when in TH) (Jakub). From RFC v1[5]: * #1: move the entire function to skbuff.c, don't try to split it (Alex); * #2-4: new; * #5: use internal flags field added in #4 and don't modify driver-defined structure (Alex, Jakub); * #6: new; * drop "add new NAPI state" as a redundant complication; * #7: replace the check for the new NAPI state to just in_softirq(), should be fine (Jakub). [2] https://lore.kernel.org/netdev/20230803164014.993838-1-aleksander.lobakin@intel.com [3] https://lore.kernel.org/netdev/20230727144336.1646454-1-aleksander.lobakin@intel.com [4] https://lore.kernel.org/netdev/20230714170853.866018-1-aleksander.lobakin@intel.com [5] https://lore.kernel.org/netdev/20230629152305.905962-1-aleksander.lobakin@intel.com