summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* net: l2tp_eth: use generic dev->stats fieldsEric Dumazet2023-10-011-22/+12
| | | | | | | | | | | Core networking has opt-in atomic variant of dev->stats, simply use DEV_STATS_INC(), DEV_STATS_ADD() and DEV_STATS_READ(). v2: removed @priv local var in l2tp_eth_dev_recv() (Simon) Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Simon Horman <horms@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
* virtio_net: avoid data-races on dev->stats fieldsEric Dumazet2023-10-011-15/+15
| | | | | | | | | | | | Use DEV_STATS_INC() and DEV_STATS_READ() which provide atomicity on paths that can be used concurrently. Reported-by: syzbot <syzkaller@googlegroups.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* net: add DEV_STATS_READ() helperEric Dumazet2023-10-012-3/+4
| | | | | | | | | | | Companion of DEV_STATS_INC() & DEV_STATS_ADD(). This is going to be used in the series. Use it in macsec_get_stats64(). Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* octeontx2-pf: Tc flower offload support for MPLSHariprasad Kelam2023-10-015-0/+187
| | | | | | | | | | | This patch extends flower offload support for MPLS protocol. Due to hardware limitation, currently driver supports lse depth up to 4. Signed-off-by: Hariprasad Kelam <hkelam@marvell.com> Signed-off-by: Sunil Goutham <sgoutham@marvell.com> Reviewed-by: Simon Horman <horms@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
* net: ethernet: xilinx: Drop kernel doc comment about return valueUwe Kleine-König2023-10-011-2/+0
| | | | | | | | | | | | | | During review of the patch that became 2e0ec0afa902 ("net: ethernet: xilinx: Convert to platform remove callback returning void") in net-next, Radhey Shyam Pandey pointed out that the change makes the documentation about the return value obsolete. The patch was applied without addressing this feedback, so here comes a fix in a separate patch. Fixes: 2e0ec0afa902 ("net: ethernet: xilinx: Convert to platform remove callback returning void") Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de> Reviewed-by: Radhey Shyam Pandey <radhey.shyam.pandey@amd.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* net: atl1c: switch to napi_consume_skb()Sieng-Piaw Liew2023-10-011-6/+7
| | | | | | | | | | | | Switch to napi_consume_skb() to take advantage of bulk free, and skb reuse through skb cache in conjunction with napi_build_skb(). When parameter 'budget' = 0, indicating non-NAPI context, dev_consume_skb_any() is called internally. Signed-off-by: Sieng-Piaw Liew <liew.s.piaw@gmail.com> Reviewed-by: Simon Horman <horms@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
* Merge branch 'sch_fq-improvements'David S. Miller2023-10-013-47/+110
|\ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Eric Dumazet says: ==================== net_sched: sch_fq: round of improvements For FQ tenth anniversary, it was time for making it faster. The FQ part (as in Fair Queue) is rather expensive, because we have to classify packets and store them in a per-flow structure, and add this per-flow structure in a hash table. Then the RR lists also add cache line misses. Most fq qdisc are almost idle. Trying to share NIC bandwidth has no benefits, thus the qdisc could behave like a FIFO. This series brings a 5 % throughput increase in intensive tcp_rr workload, and 13 % increase for (unpaced) UDP packets. v2: removed an extra label (build bot). Fix an accidental increase of stat_internal_packets counter in fast path. Added "constify qdisc_priv()" patch to allow fq_fastpath_check() first parameter to be const. typo on 'eligible' (Willem) ==================== Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com> Reviewed-by: Willem de Bruijn <willemb@google.com> Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * net_sched: sch_fq: always garbage collectEric Dumazet2023-10-011-3/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | FQ performs garbage collection at enqueue time, and only if number of flows is above a given threshold, which is hit after the qdisc has been used a bit. Since an RB-tree traversal is needed to locate a flow, it makes sense to perform gc all the time, to keep rb-trees smaller. This reduces by 50 % average storage costs in FQ, and avoids 1 cache line miss at enqueue time when fast path added in prior patch can not be used. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * net_sched: sch_fq: add fast path for mostly idle qdiscEric Dumazet2023-10-012-37/+92
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | TCQ_F_CAN_BYPASS can be used by few qdiscs. Idea is that if we queue a packet to an empty qdisc, following dequeue() would pick it immediately. FQ can not use the generic TCQ_F_CAN_BYPASS code, because some additional checks need to be performed. This patch adds a similar fast path to FQ. Most of the time, qdisc is not throttled, and many packets can avoid bringing/touching at least four cache lines, and consuming 128bytes of memory to store the state of a flow. After this patch, netperf can send UDP packets about 13 % faster, and pktgen goes 30 % faster (when FQ is in the way), on a fast NIC. TCP traffic is also improved, thanks to a reduction of cache line misses. I have measured a 5 % increase of throughput on a tcp_rr intensive workload. tc -s -d qd sh dev eth1 ... qdisc fq 8004: parent 1:2 limit 10000p flow_limit 100p buckets 1024 orphan_mask 1023 quantum 3028b initial_quantum 15140b low_rate_threshold 550Kbit refill_delay 40ms timer_slack 10us horizon 10s horizon_drop Sent 5646784384 bytes 1985161 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 flows 122 (inactive 122 throttled 0) gc 0 highprio 0 fastpath 659990 throttled 27762 latency 8.57us Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * net_sched: sch_fq: change how @inactive is trackedEric Dumazet2023-10-011-5/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Currently, when one fq qdisc has no more packets to send, it can still have some flows stored in its RR lists (q->new_flows & q->old_flows) This was a design choice, but what is a bit disturbing is that the inactive_flows counter does not include the count of empty flows in RR lists. As next patch needs to know better if there are active flows, this change makes inactive_flows exact. Before the patch, following command on an empty qdisc could have returned: lpaa17:~# tc -s -d qd sh dev eth1 | grep inactive flows 1322 (inactive 1316 throttled 0) flows 1330 (inactive 1325 throttled 0) flows 1193 (inactive 1190 throttled 0) flows 1208 (inactive 1202 throttled 0) After the patch, we now have: lpaa17:~# tc -s -d qd sh dev eth1 | grep inactive flows 1322 (inactive 1322 throttled 0) flows 1330 (inactive 1330 throttled 0) flows 1193 (inactive 1193 throttled 0) flows 1208 (inactive 1208 throttled 0) Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * net_sched: sch_fq: struct sched_data reorgEric Dumazet2023-10-011-4/+12
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | q->flows can be often modified, and q->timer_slack is read mostly. Exchange the two fields, so that cache line countaining quantum, initial_quantum, and other critical parameters stay clean (read-mostly). Move q->watchdog next to q->stat_throttled Add comments explaining how the structure is split in three different parts. pahole output before the patch: struct fq_sched_data { struct fq_flow_head new_flows; /* 0 0x10 */ struct fq_flow_head old_flows; /* 0x10 0x10 */ struct rb_root delayed; /* 0x20 0x8 */ u64 time_next_delayed_flow; /* 0x28 0x8 */ u64 ktime_cache; /* 0x30 0x8 */ unsigned long unthrottle_latency_ns; /* 0x38 0x8 */ /* --- cacheline 1 boundary (64 bytes) --- */ struct fq_flow internal __attribute__((__aligned__(64))); /* 0x40 0x80 */ /* XXX last struct has 16 bytes of padding */ /* --- cacheline 3 boundary (192 bytes) --- */ u32 quantum; /* 0xc0 0x4 */ u32 initial_quantum; /* 0xc4 0x4 */ u32 flow_refill_delay; /* 0xc8 0x4 */ u32 flow_plimit; /* 0xcc 0x4 */ unsigned long flow_max_rate; /* 0xd0 0x8 */ u64 ce_threshold; /* 0xd8 0x8 */ u64 horizon; /* 0xe0 0x8 */ u32 orphan_mask; /* 0xe8 0x4 */ u32 low_rate_threshold; /* 0xec 0x4 */ struct rb_root * fq_root; /* 0xf0 0x8 */ u8 rate_enable; /* 0xf8 0x1 */ u8 fq_trees_log; /* 0xf9 0x1 */ u8 horizon_drop; /* 0xfa 0x1 */ /* XXX 1 byte hole, try to pack */ <bad> u32 flows; /* 0xfc 0x4 */ /* --- cacheline 4 boundary (256 bytes) --- */ u32 inactive_flows; /* 0x100 0x4 */ u32 throttled_flows; /* 0x104 0x4 */ u64 stat_gc_flows; /* 0x108 0x8 */ u64 stat_internal_packets; /* 0x110 0x8 */ u64 stat_throttled; /* 0x118 0x8 */ u64 stat_ce_mark; /* 0x120 0x8 */ u64 stat_horizon_drops; /* 0x128 0x8 */ u64 stat_horizon_caps; /* 0x130 0x8 */ u64 stat_flows_plimit; /* 0x138 0x8 */ /* --- cacheline 5 boundary (320 bytes) --- */ u64 stat_pkts_too_long; /* 0x140 0x8 */ u64 stat_allocation_errors; /* 0x148 0x8 */ <bad> u32 timer_slack; /* 0x150 0x4 */ /* XXX 4 bytes hole, try to pack */ struct qdisc_watchdog watchdog; /* 0x158 0x48 */ /* size: 448, cachelines: 7, members: 34 */ /* sum members: 411, holes: 2, sum holes: 5 */ /* padding: 32 */ /* paddings: 1, sum paddings: 16 */ /* forced alignments: 1 */ }; pahole output after the patch: struct fq_sched_data { struct fq_flow_head new_flows; /* 0 0x10 */ struct fq_flow_head old_flows; /* 0x10 0x10 */ struct rb_root delayed; /* 0x20 0x8 */ u64 time_next_delayed_flow; /* 0x28 0x8 */ u64 ktime_cache; /* 0x30 0x8 */ unsigned long unthrottle_latency_ns; /* 0x38 0x8 */ /* --- cacheline 1 boundary (64 bytes) --- */ struct fq_flow internal __attribute__((__aligned__(64))); /* 0x40 0x80 */ /* XXX last struct has 16 bytes of padding */ /* --- cacheline 3 boundary (192 bytes) --- */ u32 quantum; /* 0xc0 0x4 */ u32 initial_quantum; /* 0xc4 0x4 */ u32 flow_refill_delay; /* 0xc8 0x4 */ u32 flow_plimit; /* 0xcc 0x4 */ unsigned long flow_max_rate; /* 0xd0 0x8 */ u64 ce_threshold; /* 0xd8 0x8 */ u64 horizon; /* 0xe0 0x8 */ u32 orphan_mask; /* 0xe8 0x4 */ u32 low_rate_threshold; /* 0xec 0x4 */ struct rb_root * fq_root; /* 0xf0 0x8 */ u8 rate_enable; /* 0xf8 0x1 */ u8 fq_trees_log; /* 0xf9 0x1 */ u8 horizon_drop; /* 0xfa 0x1 */ /* XXX 1 byte hole, try to pack */ <good> u32 timer_slack; /* 0xfc 0x4 */ /* --- cacheline 4 boundary (256 bytes) --- */ <good> u32 flows; /* 0x100 0x4 */ u32 inactive_flows; /* 0x104 0x4 */ u32 throttled_flows; /* 0x108 0x4 */ /* XXX 4 bytes hole, try to pack */ u64 stat_throttled; /* 0x110 0x8 */ <better> struct qdisc_watchdog watchdog; /* 0x118 0x48 */ /* --- cacheline 5 boundary (320 bytes) was 32 bytes ago --- */ u64 stat_gc_flows; /* 0x160 0x8 */ u64 stat_internal_packets; /* 0x168 0x8 */ u64 stat_ce_mark; /* 0x170 0x8 */ u64 stat_horizon_drops; /* 0x178 0x8 */ /* --- cacheline 6 boundary (384 bytes) --- */ u64 stat_horizon_caps; /* 0x180 0x8 */ u64 stat_flows_plimit; /* 0x188 0x8 */ u64 stat_pkts_too_long; /* 0x190 0x8 */ u64 stat_allocation_errors; /* 0x198 0x8 */ /* Force padding: */ u64 :64; u64 :64; u64 :64; u64 :64; /* size: 448, cachelines: 7, members: 34 */ /* sum members: 411, holes: 2, sum holes: 5 */ /* padding: 32 */ /* paddings: 1, sum paddings: 16 */ /* forced alignments: 1 */ }; Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * net_sched: constify qdisc_priv()Eric Dumazet2023-10-011-4/+4
|/ | | | | | | | In order to propagate const qualifiers, we change qdisc_priv() to accept a possibly const argument. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* Merge branch 'tcp_delack_max'David S. Miller2023-10-014-6/+23
|\ | | | | | | | | | | | | | | | | | | | | | | | | | | Eric Dumazet says: ==================== tcp: add tcp_delack_max() First patches are adding const qualifiers to four existing helpers. Third patch adds a much needed companion feature to RTAX_RTO_MIN. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
| * tcp: derive delack_max from rto_minEric Dumazet2023-10-013-2/+19
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | While BPF allows to set icsk->->icsk_delack_max and/or icsk->icsk_rto_min, we have an ip route attribute (RTAX_RTO_MIN) to be able to tune rto_min, but nothing to consequently adjust max delayed ack, which vary from 40ms to 200 ms (TCP_DELACK_{MIN|MAX}). This makes RTAX_RTO_MIN of almost no practical use, unless customers are in big trouble. Modern days datacenter communications want to set rto_min to ~5 ms, and the max delayed ack one jiffie smaller to avoid spurious retransmits. After this patch, an "rto_min 5" route attribute will effectively lower max delayed ack timers to 4 ms. Note in the following ss output, "rto:6 ... ato:4" $ ss -temoi dst XXXXXX State Recv-Q Send-Q Local Address:Port Peer Address:Port Process ESTAB 0 0 [2002:a05:6608:295::]:52950 [2002:a05:6608:297::]:41597 ino:255134 sk:1001 <-> skmem:(r0,rb1707063,t872,tb262144,f0,w0,o0,bl0,d0) ts sack cubic wscale:8,8 rto:6 rtt:0.02/0.002 ato:4 mss:4096 pmtu:4500 rcvmss:536 advmss:4096 cwnd:10 bytes_sent:54823160 bytes_acked:54823121 bytes_received:54823120 segs_out:1370582 segs_in:1370580 data_segs_out:1370579 data_segs_in:1370578 send 16.4Gbps pacing_rate 32.6Gbps delivery_rate 1.72Gbps delivered:1370579 busy:26920ms unacked:1 rcv_rtt:34.615 rcv_space:65920 rcv_ssthresh:65535 minrtt:0.015 snd_wnd:65536 While we could argue this patch fixes a bug with RTAX_RTO_MIN, I do not add a Fixes: tag, so that we can soak it a bit before asking backports to stable branches. Signed-off-by: Eric Dumazet <edumazet@google.com> Acked-by: Soheil Hassas Yeganeh <soheil@google.com> Acked-by: Neal Cardwell <ncardwell@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * tcp: constify tcp_rto_min() and tcp_rto_min_us() argumentEric Dumazet2023-10-011-2/+2
| | | | | | | | | | | | | | Make clear these functions do not change any field from TCP socket. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * net: constify sk_dst_get() and __sk_dst_get() argumentEric Dumazet2023-10-011-2/+2
|/ | | | | | | Both helpers only read fields from their socket argument. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* Merge branch '100GbE' of ↵David S. Miller2023-09-309-251/+483
|\ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue Tony Nguyen says: ==================== ice: add PTP auxiliary bus support Michal Michalik says: Auxiliary bus allows exchanging information between PFs, which allows both fixing problems and simplifying new features implementation. The auxiliary bus is enabled for all devices supported by ice driver. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
| * ice: Remove the FW shared parametersMichal Michalik2023-09-203-86/+0
| | | | | | | | | | | | | | | | | | | | The only feature using the Firmware (FW) shared parameters was the PTP clock ID. Since this ID is now shared using auxiliary buss - remove the FW shared parameters from the code. Signed-off-by: Michal Michalik <michal.michalik@intel.com> Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
| * ice: PTP: add clock domain number to auxiliary interfaceMichal Michalik2023-09-204-147/+34
| | | | | | | | | | | | | | | | | | | | | | The PHC clock id used to be moved between PFs using FW admin queue shared parameters - move the implementation to auxiliary bus. Signed-off-by: Karol Kolacinski <karol.kolacinski@intel.com> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: Michal Michalik <michal.michalik@intel.com> Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
| * ice: Use PTP auxbus for all PHYs restart in E822Michal Michalik2023-09-201-3/+21
| | | | | | | | | | | | | | | | | | | | | | | | The E822 (and other devices based on the same PHY) is having issue while setting the PHC timer - the PHY timers are drifting from the PHC. After such a set all PHYs need to be restarted and resynchronised - do it using auxiliary bus. Signed-off-by: Karol Kolacinski <karol.kolacinski@intel.com> Signed-off-by: Michal Michalik <michal.michalik@intel.com> Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
| * ice: Auxbus devices & driver for E822 TSMichal Michalik2023-09-205-17/+430
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | There is a problem in HW in E822-based devices leading to race condition. It might happen that, in order: - PF0 (which owns the PHC) requests few timestamps, - PF1 requests a timestamp, - interrupt is being triggered and both PF0 and PF1 threads are woken up, - PF0 got one timestamp, still waiting for others so not going to sleep, - PF1 gets it's timestamp, process it and go to sleep, - PF1 requests a timestamp again, - just before PF0 goes to sleep timestamp of PF1 appear, - PF0 finishes all it's timestamps and go to sleep (PF1 also sleeping). That leaves PF1 timestamp memory not read, which lead to blocking the next interrupt from arriving. Fix it by adding auxiliary devices and only one driver to handle all the timestamps for all PF's by PHC owner. In the past each PF requested it's own timestamps and process it from the start till the end which causes problem described above. Currently each PF requests the timestamps as before, but the actual reading of the completed timestamps is being done by the PTP auxiliary driver, which is registered by the PF which owns PHC. Additionally, the newly introduced auxiliary driver/devices for PTP clock owner will be used for other features in all products (including E810). Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: Karol Kolacinski <karol.kolacinski@intel.com> Signed-off-by: Michal Michalik <michal.michalik@intel.com> Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
* | pktgen: Introducing 'SHARED' flag for testing with non-shared skbLiang Chen2023-09-282-10/+66
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Currently, skbs generated by pktgen always have their reference count incremented before transmission, causing their reference count to be always greater than 1, leading to two issues: 1. Only the code paths for shared skbs can be tested. 2. In certain situations, skbs can only be released by pktgen. To enhance testing comprehensiveness, we are introducing the "SHARED" flag to indicate whether an SKB is shared. This flag is enabled by default, aligning with the current behavior. However, disabling this flag allows skbs with a reference count of 1 to be transmitted. So we can test non-shared skbs and code paths where skbs are released within the stack. Signed-off-by: Liang Chen <liangchen.linux@gmail.com> Reviewed-by: Benjamin Poirier <bpoirier@nvidia.com> Link: https://lore.kernel.org/r/20230920125658.46978-2-liangchen.linux@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
* | pktgen: Automate flag enumeration for unknown flag handlingLiang Chen2023-09-281-16/+22
| | | | | | | | | | | | | | | | | | | | | | | | When specifying an unknown flag, it will print all available flags. Currently, these flags are provided as fixed strings, which requires manual updates when flags change. Replacing it with automated flag enumeration. Signed-off-by: Liang Chen <liangchen.linux@gmail.com> Signed-off-by: Benjamin Poirier <bpoirier@nvidia.com> Link: https://lore.kernel.org/r/20230920125658.46978-1-liangchen.linux@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
* | MAINTAINERS: Add an obsolete entry for LL TEMAC driverHarini Katakam2023-09-281-0/+5
| | | | | | | | | | | | | | | | | | LL TEMAC IP is no longer supported. Hence add an entry marking the driver as obsolete. Signed-off-by: Harini Katakam <harini.katakam@amd.com> Link: https://lore.kernel.org/r/20230920115047.31345-1-harini.katakam@amd.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
* | Merge tag 'mlx5-updates-2023-09-19' of ↵Paolo Abeni2023-09-2811-208/+203
|\ \ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux Saeed Mahameed says: ==================== mlx5-updates-2023-09-19 Misc updates for mlx5 driver 1) From Erez, Add support for multicast forwarding to multi destination in bridge offloads with software steering mode (SMFS). 2) From Jianbo, Utilize the maximum aggregated link speed for police action rate. 3) From Moshe, Add a health error syndrome for pci data poisoned 4) From Shay, Enable 4 ports multiport E-switch 5) From Jiri, Trivial SF code cleanup ==================== Link: https://lore.kernel.org/r/20230920063552.296978-1-saeed@kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com>
| * | net/mlx5: Enable 4 ports multiport E-switchShay Drory2023-09-191-9/+9
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | enable_mpesw() assumed only 2 ports are available, fix this by removing that assumption and looping through the existing lag ports to enable multi-port E-switch for cards with more than 2 ports. Signed-off-by: Shay Drory <shayd@nvidia.com> Reviewed-by: Mark Bloch <mbloch@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Add a health error syndrome for pci data poisonedMoshe Shemesh2023-09-192-0/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | Add new health error syndrome to indicate that pci data poisoned error has been received while fetching device ICM data. Signed-off-by: Moshe Shemesh <moshe@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: DR, Handle multi destination action in the right orderErez Shitrit2023-09-193-3/+29
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Whenever we have few destinations from Flow-table type we need to put the one that goes to the wire to be the last one. We are using FW in order to get iterator, the FW uses RX for the first destinations and TX for the last destination, if we want the packet to be directed to the wire it should be done in the TX path and not in the RX. The code now checks if the FT is directed to the wire and if so puts it as the last destination. Signed-off-by: Erez Shitrit <erezsh@nvidia.com> Reviewed-by: Moshe Shemesh <moshe@nvidia.com> Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: DR, Add check for multi destination FTEErez Shitrit2023-09-191-0/+13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The driver should not allow rule that forward to more than one FT in TX flow unless there is a specific support from the FW. Signed-off-by: Erez Shitrit <erezsh@nvidia.com> Reviewed-by: Moshe Shemesh <moshe@nvidia.com> Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Bridge, Enable mcast in smfs steering modeErez Shitrit2023-09-192-9/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | In order to have mcast offloads the driver needs the following: It should know if that mcast comes from wire port, in addition the flow should not be marked as any specific source, that way it will give the flexibility for the driver not to be depended on the way iterator implemented in the FW. Signed-off-by: Erez Shitrit <erezsh@nvidia.com> Reviewed-by: Moshe Shemesh <moshe@nvidia.com> Reviewed-by: Vlad Buslov <vladbu@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5e: Check police action rate for matchall filterJianbo Liu2023-09-191-0/+12
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | As matchall filter uses TSAR (Transmit Scheduling Arbiter) for rate limit, the rate of police action should not be over the port's max link speed, or the maximum aggregated speed of both ports if LAG is configured. Signed-off-by: Jianbo Liu <jianbol@nvidia.com> Reviewed-by: Parav Pandit <parav@nvidia.com> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5e: Consider aggregated port speed during rate configurationJianbo Liu2023-09-191-12/+72
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | When LAG is configured, functions (PF,VF,SF) can utilize the maximum aggregated link speed for transmission. Currently the aggregated link speed is not considered. Hence, improve it to use the aggregated link speed by referring to the physical port's upper bonding device when LAG is configured. Signed-off-by: Jianbo Liu <jianbol@nvidia.com> Reviewed-by: Parav Pandit <parav@nvidia.com> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Remove redundant max_sfs check and field from struct mlx5_sf_dev_tableJiri Pirko2023-09-191-15/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | table->max_sfs is initialized in mlx5_sf_dev_table_create() and only used to check for 0 in mlx5_sf_dev_add(). mlx5_sf_dev_add() is called either from mlx5_sf_dev_state_change_handler() or mlx5_sf_dev_add_active_work(). Both ensure max SF count is not 0, using mlx5_sf_max_functions() helper before calling mlx5_sf_dev_add(). So remove the redundant check and no longer used max_sfs field. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Remove SF table reference countingJiri Pirko2023-09-191-97/+23
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Historically, the SF table reference counting was present in order to protect parallel executions of devlink ops. However, since currently this is protected with devlink instance lock, the SF table reference counting is no longer needed. Remove it entirely. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Push common deletion code into mlx5_sf_del()Jiri Pirko2023-09-191-8/+11
| | | | | | | | | | | | | | | | | | | | | | | | | | | Don't call the same functions for SF deletion on multiple places. Instead, introduce a helper mlx5_sf_del() and move the code there. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Rename mlx5_sf_deactivate_all() to mlx5_sf_del_all()Jiri Pirko2023-09-191-2/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | The function does not do deactivation, but it deletes all SFs instead. Rename accordingly. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Move state lock taking into mlx5_sf_dealloc()Jiri Pirko2023-09-191-3/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Instead of taking lock and calling mlx5_sf_dealloc(), move the lock taking into mlx5_sf_dealloc(). The other caller of mlx5_sf_dealloc() does not need it now, but will need it after a follow-up patch removing the table reference counting. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Convert SF port_indices xarray to function_ids xarrayJiri Pirko2023-09-191-18/+11
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | No need to lookup for sf by a port index. Convert the xarray to have function id as an index and optimize the remaining function id based lookup. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Use devlink port pointer to get the pointer of container SF structJiri Pirko2023-09-191-32/+12
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Benefit from the fact that struct devlink_port is eventually embedded in struct mlx5_sf and use container_of() macro to get it instead of the xarray lookup in every devlink port op. Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
| * | net/mlx5: Call mlx5_sf_id_erase() once in mlx5_sf_dealloc()Jiri Pirko2023-09-191-3/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Before every call of mlx5_sf_dealloc(), there is a call to mlx5_sf_id_erase(). So move it to the beginning of mlx5_sf_dealloc(). Also remove redundant mlx5_sf_id_erase() call from mlx5_sf_free() as it is called only from mlx5_sf_dealloc(). Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Shay Drory <shayd@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
* | | ethernet/intel: Use list_for_each_entry() helperJinjie Ruan2023-09-282-10/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Convert list_for_each() to list_for_each_entry() where applicable. No functional changed. Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Tested-by: Rafal Romanowski <rafal.romanowski@intel.com> Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://lore.kernel.org/r/20230919170409.1581074-1-anthony.l.nguyen@intel.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
* | | Merge branch 'selftests-tc-testing-parallel-tdc'Paolo Abeni2023-09-2858-1288/+2720
|\ \ \ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Pedro Tammela says: ==================== selftests/tc-testing: parallel tdc As the number of tdc tests is growing, so is our completion wall time. One of the ideas to improve this is to run tests in parallel, as they are self contained. This series allows for tests to run in parallel, in batches of 32 tests. Not all tests can run in parallel as they might conflict with each other. The code will still honor this requirement even when trying to run the tests over the worker pool. In order to make this happen we had to localize the test resources (patches 1 and 2), where instead of having all tests sharing one single namespace and veths devices each test now gets it's own local namespace and devices. Even though the tests serialize over rtnl_lock in the kernel, we measured a speedup of about 3x in a test VM. ==================== Link: https://lore.kernel.org/r/20230919135404.1778595-1-pctammela@mojatatu.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
| * | | selftests/tc-testing: update tdc documentationPedro Tammela2023-09-281-55/+10
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Update the documentation to reflect the changes made to tdc with regards to minimal requirements and test definitions expectations. Tested-by: Davide Caratti <dcaratti@redhat.com> Signed-off-by: Pedro Tammela <pctammela@mojatatu.com> Acked-by: Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
| * | | selftests/tc-testing: implement tdc parallel test runPedro Tammela2023-09-283-57/+148
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Use a Python process pool to run the tests in parallel. Not all tests can run in parallel, for instance tests that are not namespaced and tests that use netdevsim, as they can conflict with one another. The code logic will split the tests into serial and parallel. For the parallel tests, we build batches of 32 tests and queue each batch on the process pool. For the serial tests, they are queued as a whole into the process pool, which in turn executes them concurrently with the parallel tests. Even though the tests serialize on rtnl_lock in the kernel, this feature showed results with a ~3x speedup on the wall time for the entire test suite running in a VM: Before - 4m32.502s After - 1m19.202s Examples: In order to run tdc using 4 processes: ./tdc.py -J4 <...> In order to run tdc using 1 process: ./tdc.py -J1 <...> || ./tdc.py <...> Note that the kernel configuration will affect the speed of the tests, especially if such configuration slows down process creation and/or fork(). Tested-by: Davide Caratti <dcaratti@redhat.com> Signed-off-by: Pedro Tammela <pctammela@mojatatu.com> Acked-by: Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
| * | | selftests/tc-testing: update test definitions for local resourcesPedro Tammela2023-09-2851-1114/+2369
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | With resources localized on a per test basis, some tests definitions either contain redundant commands, were wrong or could be simplified. Update all of them to match the new requirements. Tested-by: Davide Caratti <dcaratti@redhat.com> Signed-off-by: Pedro Tammela <pctammela@mojatatu.com> Acked-by: Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
| * | | selftests/tc-testing: localize test resourcesPedro Tammela2023-09-285-98/+229
|/ / / | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | As of today, the current tdc architecture creates one netns and uses it to run all tests. This assumption was embedded into the nsPlugin which carried over as how the tests were written. The tdc tests are by definition self contained and can, theoretically, run in parallel. Even though in the kernel they will serialize over the rtnl lock, we should expect a significant speedup of the total wall time for the entire test suite, which is hitting close to 1100 tests at this point. A first step to achieve this goal is to remove sharing of global resources like veth/dummy interfaces and the netns. In this patch we 'localize' these resources on a per test basis. Each test gets it's own netns, VETH/dummy interfaces. The resources are spawned in the pre_suite phase, where tdc will prepare all netns and interfaces for all tests. This is done in order to avoid concurrency issues with netns / interfaces spawning and commands using them. As tdc progresses, the resources are deleted after each test finishes executing. Tests that don't use the nsPlugin still run under the root namespace, but are now required to manage any external resources like interfaces. These cannot be parallelized as their definition doesn't allow it. On the other hand, when using the nsPlugin, tests don't need to create dummy/veth interfaces as these are handled already. Tested-by: Davide Caratti <dcaratti@redhat.com> Signed-off-by: Pedro Tammela <pctammela@mojatatu.com> Acked-by: Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
* | | net: sfp: add quirk for Fiberstone GPON-ONU-34-20BIChristian Marangi2023-09-281-0/+5
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Fiberstone GPON-ONU-34-20B can operate at 2500base-X, but report 1.2GBd NRZ in their EEPROM. The module also require the ignore tx fault fixup similar to Huawei MA5671A as it gets disabled on error messages with serial redirection enabled. Signed-off-by: Christian Marangi <ansuelsmth@gmail.com> Link: https://lore.kernel.org/r/20230919124720.8210-1-ansuelsmth@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
* | | Merge branch 'mlxsw-multicast'David S. Miller2023-09-224-20/+30
|\ \ \ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Petr Machata says: ==================== mlxsw: Improve blocks selection for IPv6 multicast forwarding Amit Cohen writes: The driver configures two ACL regions during initialization, these regions are used for IPv4 and IPv6 multicast forwarding. Entries residing in these two regions match on the {SIP, DIP, VRID} key elements. Currently for IPv6 region, 9 key blocks are used. This can be improved by reducing the amount key blocks needed for the IPv6 region to 8. It is possible to use key blocks that mix subsets of the VRID element with subsets of the DIP element. To make this happen, we have to take in account the algorithm that chooses which key blocks will be used. It is lazy and not the optimal one as it is a complex task. It searches the block that contains the most elements that are required, chooses it, removes the elements that appear in the chosen block and starts again searching the block that contains the most elements. To optimize the nubmber of the blocks for IPv6 multicast forwarding, handle the following: 1. Add support for key blocks that mix subsets of the VRID element with subsets of the DIP element. 2. Prevent the algorithm from chosing another blocks for VRID. Currently, we have the block 'ipv4_4' which contains 2 sub-elements of VRID. With the existing algorithm, this block might be chosen, then 8 blocks must be chosen for SIP and DIP and we will get 9 blocks to match on {SIP, DIP, VRID}. Therefore, replace this block with a new block 'ipv4_5' that contains 1 element for VRID, this will not be chosen for IPv6 as VRID element will be broken to several sub-elements. In this way we can get 8 blocks for IPv6 multicast forwarding. This improvement was tested and indeed 8 blocks are used instead of 9. v2: - Resending without changes. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
| * | | mlxsw: Edit IPv6 key blocks to use one less block for multicast forwardingAmit Cohen2023-09-224-3/+22
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Two ACL regions that are configured by the driver during initialization are the ones used for IPv4 and IPv6 multicast forwarding. Entries residing in these two regions match on the {SIP, DIP, VRID} key elements. Currently for IPv6 region, 9 key blocks are used: * 4 for SIP - 'ipv4_1', 'ipv6_{3,4,5}' * 4 for DIP - 'ipv4_0', 'ipv6_{0,1,2/2b}' * 1 for VRID - 'ipv4_4b' This can be improved by reducing the amount key blocks needed for the IPv6 region to 8. It is possible to use key blocks that mix subsets of the VRID element with subsets of the DIP element. The following key blocks can be used: * 4 for SIP - 'ipv4_1', 'ipv6_{3,4,5}' * 1 for subset of DIP - 'ipv4_0' * 3 for the rest of DIP and subsets of VRID - 'ipv6_{0,1,2/2b}' To make this happen, add VRID sub-elements as part of existing keys - 'ipv6_{0,1,2/2b}'. Note that one of the sub-elements is called VRID_ROUTER_MSB and does not contain bit numbers like the rest, as for Spectrum < 4 this element represents bits 8-10 and for Spectrum-4 it represents bits 8-11. Breaking VRID into 3 sub-elements makes the driver use one less block in IPv6 region for multicast forwarding. The sub-elements can be filled in blocks that are used for destination IP. The algorithm in the driver that chooses which key blocks will be used is lazy and not the optimal one. It searches the block that contains the most elements that are required, chooses it, removes the elements that appear in the chosen block and starts again searching the block that contains the most elements. When key block 'ipv4_4' is defined, the algorithm might choose it, as it contains 2 sub-elements of VRID, then 8 blocks must be chosen for SIP and DIP and we get 9 blocks to match on {SIP, DIP, VRID}. That is why we had to remove key block 'ipv4_4' in a previous patch and use key block that contains one field for VRID. This improvement was tested and indeed 8 blocks are used instead of 9. Signed-off-by: Amit Cohen <amcohen@nvidia.com> Reviewed-by: Ido Schimmel <idosch@nvidia.com> Signed-off-by: Petr Machata <petrm@nvidia.com> Signed-off-by: David S. Miller <davem@davemloft.net>
| * | | mlxsw: spectrum_acl_flex_keys: Add 'ipv4_5b' flex keyAmit Cohen2023-09-221-3/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The previous patch replaced the key block 'ipv4_4' with 'ipv4_5'. The corresponding block for Spectrum-4 is 'ipv4_4b'. To be consistent, replace key block 'ipv4_4b' with 'ipv4_5b'. Signed-off-by: Amit Cohen <amcohen@nvidia.com> Reviewed-by: Ido Schimmel <idosch@nvidia.com> Signed-off-by: Petr Machata <petrm@nvidia.com> Signed-off-by: David S. Miller <davem@davemloft.net>
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-01 12:48:34 +0000