- 27 May, 2020 21 commits
-
-
Ido Schimmel authored
IPv6 packets that need to be forwarded and have a link-local source IP are dropped by the kernel and an ICMPv6 "Destination unreachable" is sent to the sending host. As such, change the trap group of such packets so that they do not interfere with IPv6 management packets. In the future this trap will be exposed as an exception via devlink-trap. Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Reviewed-by:
Jiri Pirko <jiri@mellanox.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Ido Schimmel authored
Routed IP packets with the Router Alert option need to be trapped to the CPU as they might need to be locally delivered to raw sockets with the IP_ROUTER_ALERT / IPV6_ROUTER_ALERT socket option. Move them to the same group with other packets that might need to be trapped following route lookup. Signed-off-by:
-
Ido Schimmel authored
After the previous patch the split is no longer necessary and all the trap groups can be moved under the same enum. Signed-off-by:
-
Ido Schimmel authored
As explained in commit e6125230 ("mlxsw: spectrum_trap: Introduce dummy group with thin policer"), the purpose of the "thin" policer is to pass as less packets as possible to the CPU. The identifier of this policer is currently set according to the maximum number of used trap groups, but this is fragile: On Spectrum-1 the maximum number of policers is less than the maximum number of trap groups, which might result in an invalid policer identifier in case the number of used trap groups grows beyond the policer limit. Solve this by dynamically allocating the policer identifier. Signed-off-by:
-
Ido Schimmel authored
The number of Spectrum trap groups is not infinite, but two identifiers are occupied by SwitchX-2 specific trap groups. Free these identifiers by moving them out of the main enum. Signed-off-by:
-
Ido Schimmel authored
To align with recent recommended values. Will be configurable by future patches. Signed-off-by:
-
Ido Schimmel authored
Packets with an IPv6 link-local destination (i.e., fe80::/10) should not be forwarded and are therefore trapped to the CPU for local delivery. Since these packets are trapped for the same logical reason as packets hitting local routes, associate both traps with the same group. Signed-off-by:
-
Ido Schimmel authored
When a packet enters the device it is classified to a filtering identifier (FID) based on the ingress port and VLAN. The FID miss trap is used to trap packets for which a FID could not be found. In mlxsw this trap should only be triggered when a port is enslaved to an OVS bridge and a matching ACL rule could not be found, so as to trigger learning. These packets are therefore completely unrelated to packets hitting local routes and should be in a different group. Move them. Signed-off-by:
-
Ido Schimmel authored
Group these various IPv6 packets (e.g., router solicitations, router advertisement) together and subject them to the same policer. Signed-off-by:
-
Ido Schimmel authored
The IPv6 Neighbour Discovery (ND) group will be used for various IPv6 packets, not all of which fall under the definition of ND, so rename it to "IPV6" which is more appropriate. Signed-off-by:
-
Ido Schimmel authored
Trap groups that use the same policer settings can share the same switch case. Signed-off-by:
-
Ido Schimmel authored
Packets that are trapped via tc's trap action are currently subject to the same policer as packets hitting local routes. The latter are critical to the correct functioning of the control plane, while the former are mainly used for traffic inspection. Split the ACL trap to a separate group with its own policer. Use a higher priority for these traps than for traps using mirror action (e.g., ARP, IGMP). Otherwise, packets matching both traps will not be forwarded in hardware (because of trap action) and also not forwarded in software because they will be marked with 'offload_fwd_mark'. Signed-off-by:
-
Florian Westphal authored
We can try to coalesce skbs we take from the subflows rx queue with the tail of the mptcp rx queue. If successful, the skb head can be discarded early. We can also free the skb extensions, we do not access them after this. Signed-off-by:
Florian Westphal <fw@strlen.de> Signed-off-by:
-
Heiner Kallweit authored
Don't call netif_napi_del() manually, free_netdev() does this for us. In addition reorder calls to match reverse order of calls in probe(). Signed-off-by:
Heiner Kallweit <hkallweit1@gmail.com> Signed-off-by:
-
David S. Miller authored
Fugang Duan says: ==================== net: ethernet: fec: move GPR register offset and bit into DT The commit da722186 (net: fec: set GPR bit on suspend by DT configuration) set the GPR reigster offset and bit in driver for wol feature support. It brings trouble to enable wol feature on imx6sx/imx6ul/imx7d platforms that have multiple ethernet instances with different GPR bit for stop mode control. So the patch set is to move GPR register offset and bit define into DT, and enable imx6q/imx6dl imx6qp/imx6sx/imx6ul/imx7d stop mode support. Currently, below NXP i.MX boards support wol: - imx6q/imx6dl/imx6qp sabresd - imx6sx sabreauto - imx7d sdb imx6q/imx6dl/imx6qp sabresd board dts file miss the property "fsl,magic-packet;", so patch#4 is to add the property for stop mode support. v1 -> v2: - driver: switch back to store the quirks bitmask in driver_data - dt-bindings: rename 'gpr' property string to 'fsl,stop-mode' - imx6/7 dtsi: add imx6sx/imx6ul/imx7d ethernet stop mode property v2 -> v3: - driver: suggested by Sascha Hauer, use a struct fec_devinfo for abstracting differences between different hardware variants, it can give more freedom to describe the differences. - imx6/7 dtsi: correct one typo pointed out by Andrew. Thanks Martin, Andrew and Sascha Hauer for the review. ==================== Signed-off-by:
-
Fugang Duan authored
Enable ethernet wake-on-lan feature for imx6q/dl/qp sabresd boards since the PHY clock is supplied by external osc. Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
Fugang Duan <fugang.duan@nxp.com> Signed-off-by:
-
Fugang Duan authored
- Update the imx6qdl gpr property to define gpr register offset and bit in DT. - Add imx6sx/imx6ul/imx7d ethernet stop mode property. Reviewed-by:
-
Fugang Duan authored
- rename the 'gpr' property string to 'fsl,stop-mode'. - Update the property to define gpr register offset and bit in DT, since different instance have different gpr bit. v2: * rename 'gpr' property string to 'fsl,stop-mode'. Signed-off-by:
-
Fugang Duan authored
The commit da722186 (net: fec: set GPR bit on suspend by DT configuration) set the GPR reigster offset and bit in driver for wake on lan feature. But it introduces two issues here: - one SOC has two instances, they have different bit - different SOCs may have different offset and bit So to support wake-on-lan feature on other i.MX platforms, it should configure the GPR reigster offset and bit from DT. So the patch is to improve the commit da722186 (net: fec: set GPR bit on suspend by DT configuration) to support multiple ethernet instances on i.MX series. v2: * switch back to store the quirks bitmask in driver_data v3: * suggested by Sascha Hauer, use a struct fec_devinfo for abstracting differences between different hardware variants, it can give more freedom to describe the differences. Signed-off-by:
-
Dmitry Vyukov authored
Netlink policies are generally declared as const. This is safer and prevents potential bugs. Signed-off-by:
Dmitry Vyukov <dvyukov@google.com> Signed-off-by:
-
David S. Miller authored
Merge tag 'mac80211-next-for-net-next-2020-04-25' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next Johannes Berg says: ==================== One batch of changes, containing: * hwsim improvements from Jouni and myself, to be able to test more scenarios easily * some more HE (802.11ax) support * some initial S1G (sub 1 GHz) work for fractional MHz channels * some (action) frame registration updates to help DPP support * along with other various improvements/fixes ==================== Signed-off-by:
-
- 26 May, 2020 19 commits
-
-
David S. Miller authored
Antoine Tenart says: ==================== net: phy: mscc-miim: reduce waiting time between MDIO transactions This series aims at reducing the waiting time between MDIO transactions when using the MSCC MIIM MDIO controller. I'm not sure we need patch 4/4 and we could reasonably drop it from the series. I'm including the patch as it could help to ensure the system is functional with a non optimal configuration. We needed to improve the driver's performances as when using a PHY requiring lots of registers accesses (such as the VSC85xx family), delays would add up and ended up to be quite large which would cause issues such as: a slow initialization of the PHY, and issues when using timestamping operations (this feature will be sent quite soon to the mailing lists). ==================== Signed-off-by: -
Antoine Tenart authored
The driver uses a read polling mechanism to check the status of the MDIO bus, to know if it is ready to accept next commands. This polling mechanism uses usleep_delay() under the hood between reads which is fine as long as high resolution timers are enabled. Otherwise the delays will end up to be much longer than expected. This patch fixes this by using udelay() under the hood when CONFIG_HIGH_RES_TIMERS isn't enabled. This increases CPU usage. Signed-off-by:
Antoine Tenart <antoine.tenart@bootlin.com> Reviewed-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
Antoine Tenart authored
The MSCC MIIM MDIO driver uses a waiting logic to wait for the MDIO bus to be ready to accept next commands. It does so by polling the BUSY status bit which indicates the MDIO bus has completed all pending operations. This can take time, and the controller supports writing the next command as soon as there are no pending commands (which happens while the MDIO bus is busy completing its current command). This patch implements this improved logic by adding an helper to poll the PENDING status bit, and by adjusting where we should wait for the bus to not be busy or to not be pending. Signed-off-by:
Alexandre Belloni <alexandre.belloni@bootlin.com> Reviewed-by:
-
Antoine Tenart authored
readl_poll_timeout already returns -ETIMEDOUT if the condition isn't satisfied, there's no need to check again the condition after calling it. Remove the redundant timeout check. Signed-off-by:
-
Antoine Tenart authored
The MSCC MIIM MDIO driver uses delays to read poll a status register. I made multiple tests on a Ocelot PCS120 platform which led me to reduce those delays. The delay in between which the polling function is allowed to sleep is reduced from 100us to 50us which in almost all cases is a good value to succeed at the first retry. The overall delay is also lowered as the prior value was really way to high, 10000us is large enough. Signed-off-by:
-
Russell King authored
There is a recurring pattern throughout some of the PHY code converting a devad and regnum to our packed clause 45 representation. Rather than having this scattered around the code, let's put a common translation function in mdio.h, and provide some register accessors. Convert the phylib core, phylink, bcm87xx and cortina to use these. Signed-off-by:
Russell King <rmk+kernel@armlinux.org.uk> Reviewed-by:
-
David S. Miller authored
Guillaume Nault says: ==================== flow_dissector, cls_flower: Add support for multiple MPLS Label Stack Entries Currently, the flow dissector and the Flower classifier can only handle the first entry of an MPLS label stack. This patch series generalises the code to allow parsing and matching the Label Stack Entries that follow. Patch 1 extends the flow dissector to parse MPLS LSEs until the Bottom Of Stack bit is reached. The number of parsed LSEs is capped at FLOW_DIS_MPLS_MAX (arbitrarily set to 7). Flower and the NFP driver are updated to take into account the new layout of struct flow_dissector_key_mpls. Patch 2 extends Flower. It defines new netlink attributes, which are independent from the previous MPLS ones. Mixing the old and the new attributes in a same filter is not allowed. For backward compatibility, the old attributes are used when dumping filters that don't require the new ones. Changes since v2: * Fix compilation with the new MLX5 bareudp tunnel code. Changes since v1: * Fix compilation of NFP driver (kbuild test robot). * Fix sparse warning with entropy label (kbuild test robot). ==================== Signed-off-by: -
Guillaume Nault authored
With struct flow_dissector_key_mpls now recording the first FLOW_DIS_MPLS_MAX labels, we can extend Flower to filter on any of these LSEs independently. In order to avoid creating new netlink attributes for every possible depth, let's define a new TCA_FLOWER_KEY_MPLS_OPTS nested attribute that contains the list of LSEs to match. Each LSE is represented by another attribute, TCA_FLOWER_KEY_MPLS_OPTS_LSE, which then contains the attributes representing the depth and the MPLS fields to match at this depth (label, TTL, etc.). For each MPLS field, the mask is always set to all-ones, as this is what the original API did. We could allow user configurable masks in the future if there is demand for more flexibility. The new API also allows to only specify an LSE depth. In that case, Flower only verifies that the MPLS label stack depth is greater or equal to the provided depth (that is, an LSE exists at this depth). Filters that only match on one (or more) fields of the first LSE are dumped using the old netlink attributes, to avoid confusing user space programs that don't understand the new API. Signed-off-by:
Guillaume Nault <gnault@redhat.com> Signed-off-by:
-
Guillaume Nault authored
The current MPLS dissector only parses the first MPLS Label Stack Entry (second LSE can be parsed too, but only to set a key_id). This patch adds the possibility to parse several LSEs by making __skb_flow_dissect_mpls() return FLOW_DISSECT_RET_PROTO_AGAIN as long as the Bottom Of Stack bit hasn't been seen, up to a maximum of FLOW_DIS_MPLS_MAX entries. FLOW_DIS_MPLS_MAX is arbitrarily set to 7. This should be enough for many practical purposes, without wasting too much space. To record the parsed values, flow_dissector_key_mpls is modified to store an array of stack entries, instead of just the values of the first one. A bit field, "used_lses", is also added to keep track of the LSEs that have been set. The objective is to avoid defining a new FLOW_DISSECTOR_KEY_MPLS_XX for each level of the MPLS stack. TC flower is adapted for the new struct flow_dissector_key_mpls layout. Matching on several MPLS Label Stack Entries will be added in the next patch. The NFP and MLX5 drivers are also adapted: nfp_flower_compile_mac() and mlx5's parse_tunnel() now verify that the rule only uses the first LSE and fail if it doesn't. Finally, the behaviour of the FLOW_DISSECTOR_KEY_MPLS_ENTROPY key is slightly modified. Instead of recording the first Entropy Label, it now records the last one. This shouldn't have any consequences since there doesn't seem to have any user of FLOW_DISSECTOR_KEY_MPLS_ENTROPY in the tree. We'd probably better do a hash of all parsed MPLS labels instead (excluding reserved labels) anyway. That'd give better entropy and would probably also simplify the code. But that's not the purpose of this patch, so I'm keeping that as a future possible improvement. Signed-off-by:
-
git://git.open-mesh.org/linux-mergeDavid S. Miller authored
Simon Wunderlich says: ==================== This cleanup patchset includes the following patches: - Fix revert dynamic lockdep key changes for batman-adv, by Sven Eckelmann - use rcu_replace_pointer() where appropriate, by Antonio Quartulli - Revert "disable ethtool link speed detection when auto negotiation off", by Sven Eckelmann ==================== Signed-off-by: -
David S. Miller authored
Tuong Lien says: ==================== tipc: add some improvements This series adds some improvements to TIPC. The first patch improves the TIPC broadcast's performance with the 'Gap ACK blocks' mechanism similar to unicast before, while the others give support on tracing & statistics for broadcast links, and an alternative to carry broadcast retransmissions via unicast which might be useful in some cases. Besides, the Nagle algorithm can now automatically 'adjust' itself depending on the specific network condition a stream connection runs by the last patch. ==================== Signed-off-by: -
Tuong Lien authored
When streaming in Nagle mode, we try to bundle small messages from user as many as possible if there is one outstanding buffer, i.e. not ACK-ed by the receiving side, which helps boost up the overall throughput. So, the algorithm's effectiveness really depends on when Nagle ACK comes or what the specific network latency (RTT) is, compared to the user's message sending rate. In a bad case, the user's sending rate is low or the network latency is small, there will not be many bundles, so making a Nagle ACK or waiting for it is not meaningful. For example: a user sends its messages every 100ms and the RTT is 50ms, then for each messages, we require one Nagle ACK but then there is only one user message sent without any bundles. In a better case, even if we have a few bundles (e.g. the RTT = 300ms), but now the user sends messages in medium size, then there will not be any difference at all, that says 3 x 1000-byte data messages if bundled will still result in 3 bundles with MTU = 1500. When Nagle is ineffective, the delay in user message sending is clearly wasted instead of sending directly. Besides, adding Nagle ACKs will consume some processor load on both the sending and receiving sides. This commit adds a test on the effectiveness of the Nagle algorithm for an individual connection in the network on which it actually runs. Particularly, upon receipt of a Nagle ACK we will compare the number of bundles in the backlog queue to the number of user messages which would be sent directly without Nagle. If the ratio is good (e.g. >= 2), Nagle mode will be kept for further message sending. Otherwise, we will leave Nagle and put a 'penalty' on the connection, so it will have to spend more 'one-way' messages before being able to re-enter Nagle. In addition, the 'ack-required' bit is only set when really needed that the number of Nagle ACKs will be reduced during Nagle mode. Testing with benchmark showed that with the patch, there was not much difference in throughput for small messages since the tool continuously sends messages without a break, so Nagle would still take in effect. Acked-by:
Ying Xue <ying.xue@windriver.com> Acked-by:
Jon Maloy <jmaloy@redhat.com> Signed-off-by:
Tuong Lien <tuong.t.lien@dektech.com.au> Signed-off-by:
-
Tuong Lien authored
This commit enables dumping the statistics of a broadcast-receiver link like the traditional 'broadcast-link' one (which is for broadcast- sender). The link dumping can be triggered via netlink (e.g. the iproute2/tipc tool) by the link flag - 'TIPC_NLA_LINK_BROADCAST' as the indicator. The name of a broadcast-receiver link of a specific peer will be in the format: 'broadcast-link:<peer-id>'. For example: Link <broadcast-link:1001002> Window:50 packets RX packets:7841 fragments:2408/440 bundles:0/0 TX packets:0 fragments:0/0 bundles:0/0 RX naks:0 defs:124 dups:0 TX naks:21 acks:0 retrans:0 Congestion link:0 Send queue max:0 avg:0 In addition, the broadcast-receiver link statistics can be reset in the usual way via netlink by specifying that link name in command. Note: the 'tipc_link_name_ext()' is removed because the link name can now be retrieved simply via the 'l->name'. Acked-by:
-
Tuong Lien authored
In some environment, broadcast traffic is suppressed at high rate (i.e. a kind of bandwidth limit setting). When it is applied, TIPC broadcast can still run successfully. However, when it comes to a high load, some packets will be dropped first and TIPC tries to retransmit them but the packet retransmission is intentionally broadcast too, so making things worse and not helpful at all. This commit enables the broadcast retransmission via unicast which only retransmits packets to the specific peer that has really reported a gap i.e. not broadcasting to all nodes in the cluster, so will prevent from being suppressed, and also reduce some overheads on the other peers due to duplicates, finally improve the overall TIPC broadcast performance. Note: the functionality can be turned on/off via the sysctl file: echo 1 > /proc/sys/net/tipc/bc_retruni echo 0 > /proc/sys/net/tipc/bc_retruni Default is '0', i.e. the broadcast retransmission still works as usual. Acked-by:
-
Tuong Lien authored
In the previous commit ("tipc: add Gap ACK blocks support for broadcast link"), we have removed the following link trace events due to the code changes: - tipc_link_bc_ack - tipc_link_retrans This commit adds them back along with some minor changes to adapt to the new code. Acked-by: -
Tuong Lien authored
As achieved through commit 9195948f ("tipc: improve TIPC throughput by Gap ACK blocks"), we apply the same mechanism for the broadcast link as well. The 'Gap ACK blocks' data field in a 'PROTOCOL/STATE_MSG' will consist of two parts built for both the broadcast and unicast types: 31 16 15 0 +-------------+-------------+-------------+-------------+ | bgack_cnt | ugack_cnt | len | +-------------+-------------+-------------+-------------+ - | gap | ack | | +-------------+-------------+-------------+-------------+ > bc gacks : : : | +-------------+-------------+-------------+-------------+ - | gap | ack | | +-------------+-------------+-------------+-------------+ > uc gacks : : : | +-------------+-------------+-------------+-------------+ - which is "automatically" backward-compatible. We also increase the max number of Gap ACK blocks to 128, allowing upto 64 blocks per type (total buffer size = 516 bytes). Besides, the 'tipc_link_advance_transmq()' function is refactored which is applicable for both the unicast and broadcast cases now, so some old functions can be removed and the code is optimized. With the patch, TIPC broadcast is more robust regardless of packet loss or disorder, latency, ... in the underlying network. Its performance is boost up significantly. For example, experiment with a 5% packet loss rate results: $ time tipc-pipe --mc --rdm --data_size 123 --data_num 1500000 real 0m 42.46s user 0m 1.16s sys 0m 17.67s Without the patch: $ time tipc-pipe --mc --rdm --data_size 123 --data_num 1500000 real 8m 27.94s user 0m 0.55s sys 0m 2.38s Acked-by:
-
Yuval Basson authored
In older FW versions the completion flag was treated as the ack flag in edpm messages. Expose the FW option of setting which mode the QP is in by adding a flag to the qedr <-> qed API. Flag is added for backward compatibility with libqedr. This flag will be set by qedr after determining whether the libqedr is using the updated version. Fixes: f1093940 ("qed: Add support for QP verbs") Signed-off-by:
Yuval Basson <yuval.bason@marvell.com> Signed-off-by:
Michal Kalderon <michal.kalderon@marvell.com> Signed-off-by:
-
Eric Dumazet authored
I missed the fact that tcp_v4_err() differs from tcp_v6_err(). After commit 4d1a2d9e ("Rename skb to icmp_skb in tcp_v4_err()") the skb argument has been renamed to icmp_skb only in one function. I will in a future patch reconciliate these functions to avoid this kind of confusion. Fixes: 45af29ca ("tcp: allow traceroute -Mtcp for unpriv users") Signed-off-by:
Eric Dumazet <edumazet@google.com> Signed-off-by:
-
Sven Eckelmann authored
The commit 8c46fcd7 ("batman-adv: disable ethtool link speed detection when auto negotiation off") disabled the usage of ethtool's link_ksetting when auto negotation was enabled due to invalid values when used with tun/tap virtual net_devices. According to the patch, automatic measurements should be used for these kind of interfaces. But there are major flaws with this argumentation: * automatic measurements are not implemented * auto negotiation has nothing to do with the validity of the retrieved values The first point has to be fixed by a longer patch series. The "validity" part of the second point must be addressed in the same patch series by dropping the usage of ethtool's link_ksetting (thus always doing automatic measurements over ethernet). Drop the patch again to have more default values for various net_device types/configurations. The user can still overwrite them using the batadv_hardif's BATADV_ATTR_THROUGHPUT_OVERRIDE. Reported-by:
Matthias Schiffer <mschiffer@universe-factory.net> Signed-off-by:
Sven Eckelmann <sven@narfation.org> Signed-off-by:
Simon Wunderlich <sw@simonwunderlich.de>
-
