- 26 Dec, 2023 19 commits
-
-
David S. Miller authored
Victor Nogueira says: ==================== net/sched: Introduce tc block ports tracking and use __context__ The "tc block" is a collection of netdevs/ports which allow qdiscs to share match-action block instances (as opposed to the traditional tc filter per netdev/port)[1]. Up to this point in the implementation, the block is unaware of its ports. This patch makes the tc block ports available to the datapath. For the datapath we provide a use case of the tc block in a mirred action in patch 3. For users can levarage mirred to do something like the following: $ tc qdisc add dev ens7 ingress_block 22 clsact $ tc qdisc add dev ens8 ingress_block 22 clsact $ tc qdisc add dev ens9 ingress_block 22 clsact $ tc filter add block 22 protocol ip pref 25 \ flower dst_ip 192.168.0.0/16 action mirred egress mirror blockid 22 In this case, if the packet arrives on ens8, it will be copied and sent to all ports in the block excluding ens8. Note that the packet is still in the pipeline at this point - meaning other actions could be added after the mirror because mirred copies/clones the skb. Example the following is valid: $ tc filter add block 22 protocol ip pref 25 flower dst_ip 192.168.0.0/16 \ action mirred egress mirror blockid 22 \ action vlan push id 123 \ action mirred egress redirect dev dummy0 redirect behavior always steals the packet from the pipeline and therefore the skb is no longer available for a subsequent action as illustrated above (in redirecting to dummy0). The behavior of redirecting to a tc block is therefore adapted to work in the same manner. So a setup as such: $ tc qdisc add dev ens7 ingress_block 22 $ tc qdisc add dev ens8 ingress_block 22 $ tc qdisc add dev ens9 ingress_block 22 $ tc filter add block 22 protocol ip pref 25 \ flower dst_ip 192.168.0.0/16 action mirred egress redirect blockid 22 for a matching packet arriving on ens7 will first send a copy/clone to ens8 (as in the "mirror" behavior) then to ens9 as in the redirect behavior above. Once this processing is done - no other actions are able to process this skb. i.e it is removed from the "pipeline". In this case, if the packet arrives on ens8, it will be copied and sent to all ports in the block excluding ens8. Patch 1 separates/exports mirror and redirect functions from act_mirred Patch 2 introduces the required infra. Patch 3 Allows mirred to blocks Subsequent patches will come with tdc test cases. __Acknowledgements__ Suggestions from Vlad Buslov and Marcelo Ricardo Leitner made this patchset better. The idea of integrating the ports into the tc block was suggested by Jiri Pirko. [1] See commit ca46abd6 ("Merge branch'net-sched-allow-qdiscs-to-share-filter-block-instances'") Changes in v2: - Remove RFC tag - Add more details in patch 0(Jiri) - When CONFIG_NET_TC_SKB_EXT is selected we have unused qdisc_cb Reported-by: kernel test robot <lkp@intel.com> (and horms@kernel.org) - Fix bad dev dereference in printk of blockcast action (Simon) Changes in v3: - Add missing xa_destroy (pointed out by Vlad) - Remove bugfix pointed by Vlad (will send in separate patch) - Removed ports from subject in patch #2 and typos (suggested by Marcelo) - Remove net_notice_ratelimited debug messages in error cases (suggested by Marcelo) - Minor changes to appease sparse's lock context warning Changes in v4: - Avoid code repetition using gotos in cast_one (suggested by Paolo) - Fix typo in cover letter (pointed out by Paolo) - Create a module description for act_blockcast (reported by Paolo and CI) Changes in v5: - Add new patch which separated mirred into mirror and redirect functions (suggested by Jiri) - Instead of repeating the code to mirror in blockcast use mirror exported function by patch1 (tcf_mirror_act) - Make Block ID into act_blockcast's parameter passed by user space instead of always getting it from SKB (suggested by Jiri) - Add tx_type parameter which will specify what transmission behaviour we want (as described earlier) Changes in v6: - Remove blockcast and make it a part of mirred (suggestd by Jiri) - Block ID is now a mirred parameter - We now allow redirecting and mirroring to either ingress or egress Changes in v7: - Remove set but not used variable in tcf_mirred_act (pointed out by Jakub) Changes in v8: - Fix uapi issues (pointed out by Jiri) - Separate last patch into 3 - two as preparations for adding block ID to mirred and one allowing mirred to block (suggested by Jiri) - Remove declaration initialisation of eg_block and in_block in qdisc_block_add_dev (suggested by Jiri) - Avoid unnecessary if guards in qdisc_block_add_dev (suggested by Jiri) - Remove unncessary block_index retrieval in __qdisc_destroy (suggested by Jiri) ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Victor Nogueira authored
So far the mirred action has dealt with syntax that handles mirror/redirection for netdev. A matching packet is redirected or mirrored to a target netdev. In this patch we enable mirred to mirror to a tc block as well. IOW, the new syntax looks as follows: ... mirred <ingress | egress> <mirror | redirect> [index INDEX] < <blockid BLOCKID> | <dev <devname>> > Examples of mirroring or redirecting to a tc block: $ tc filter add block 22 protocol ip pref 25 \ flower dst_ip 192.168.0.0/16 action mirred egress mirror blockid 22 $ tc filter add block 22 protocol ip pref 25 \ flower dst_ip 10.10.10.10/32 action mirred egress redirect blockid 22 Co-developed-by:
Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by:
Pedro Tammela <pctammela@mojatatu.com> Signed-off-by:
Victor Nogueira <victor@mojatatu.com> Signed-off-by:
-
Victor Nogueira authored
The act of replacing a device will be repeated by the init logic for the block ID in the patch that allows mirred to a block. Therefore we encapsulate this functionality in a function (tcf_mirred_replace_dev) so that we can reuse it and avoid code repetition. Co-developed-by:
-
Victor Nogueira authored
As a preparation for adding block ID to mirred, separate the part of mirred that redirect/mirrors to a dev into a specific function so that it can be called by blockcast for each dev. Also improve readability. Eg. rename use_reinsert to dont_clone and skb2 to skb_to_send. Co-developed-by:
-
Victor Nogueira authored
The datapath can now find the block of the port in which the packet arrived at. In the next patch we show a possible usage of this patch in a new version of mirred that multicasts to all ports except for the port in which the packet arrived on. Co-developed-by:
-
Victor Nogueira authored
This commit makes tc blocks track which ports have been added to them. And, with that, we'll be able to use this new information to send packets to the block's ports. Which will be done in the patch #3 of this series. Suggested-by:
Jiri Pirko <jiri@nvidia.com> Co-developed-by:
-
Denis Kirjanov authored
Since there are no more users of the macro let's finally burn it Signed-off-by:
Denis Kirjanov <dkirjanov@suse.de> Signed-off-by:
-
Denis Kirjanov authored
SOCK_DEBUG comes from the old days. Let's move logging to standard net core ratelimited logging functions Signed-off-by:
-
David S. Miller authored
Wen Gu says: ==================== net/smc: implement SMCv2.1 virtual ISM device support The fourth edition of SMCv2 adds the SMC version 2.1 feature updates for SMC-Dv2 with virtual ISM. Virtual ISM are created and supported mainly by OS or hypervisor software, comparable to IBM ISM which is based on platform firmware or hardware. With the introduction of virtual ISM, SMCv2.1 makes some updates: - Introduce feature bitmask to indicate supplemental features. - Reserve a range of CHIDs for virtual ISM. - Support extended GIDs (128 bits) in CLC handshake. So this patch set aims to implement these updates in Linux kernel. And it acts as the first part of SMC-D virtual ISM extension & loopback-ism [1]. [1] https://lore.kernel.org/netdev/1695568613-125057-1-git-send-email-guwen@linux.alibaba.com/ v8->v7: - Patch #7: v7 mistakenly changed the type of gid_ext in smc_clc_msg_accept_confirm to u64 instead of __be64 as previous versions when fixing the rebase conflicts. So fix this mistake. v7->v6: Link: https://lore.kernel.org/netdev/20231219084536.8158-1-guwen@linux.alibaba.com/ - Collect the Reviewed-by tag in v6; - Patch #3: redefine the struct smc_clc_msg_accept_confirm; - Patch #7: Because that the Patch #3 already adds '__packed' to smc_clc_msg_accept_confirm, so Patch #7 doesn't need to do the same thing. But this is a minor change, so I kept the 'Reviewed-by' tag. Other changes in previous versions but not yet acked: - Patch #1: Some minor changes in subject and fix the format issue (length exceeds 80 columns) compared to v3. - Patch #5: removes useless ini->feature_mask assignment in __smc_connect() and smc_listen_v2_check() compared to v4. - Patch #8: new added, compared to v3. v6->v5: Link: https://lore.kernel.org/netdev/1702371151-125258-1-git-send-email-guwen@linux.alibaba.com/ - Add 'Reviewed-by' label given in the previous versions: * Patch #4, #6, #9, #10 have nothing changed since v3; - Patch #2: * fix the format issue (Alignment should match open parenthesis) compared to v5; * remove useless clc->hdr.length assignment in smcr_clc_prep_confirm_accept() compared to v5; - Patch #3: new added compared to v5. - Patch #7: some minor changes like aclc_v2->aclc or clc_v2->clc compared to v5 due to the introduction of Patch #3. Since there were no major changes, I kept the 'Reviewed-by' label. Other changes in previous versions but not yet acked: - Patch #1: Some minor changes in subject and fix the format issue (length exceeds 80 columns) compared to v3. - Patch #5: removes useless ini->feature_mask assignment in __smc_connect() and smc_listen_v2_check() compared to v4. - Patch #8: new added, compared to v3. v5->v4: Link: https://lore.kernel.org/netdev/1702021259-41504-1-git-send-email-guwen@linux.alibaba.com/ - Patch #6: improve the comment of SMCD_CLC_MAX_V2_GID_ENTRIES; - Patch #4: remove useless ini->feature_mask assignment; v4->v3: https://lore.kernel.org/netdev/1701920994-73705-1-git-send-email-guwen@linux.alibaba.com/ - Patch #6: use SMCD_CLC_MAX_V2_GID_ENTRIES to indicate the max gid entries in CLC proposal and using SMC_MAX_V2_ISM_DEVS to indicate the max devices to propose; - Patch #6: use i and i+1 in smc_find_ism_v2_device_serv(); - Patch #2: replace the large if-else block in smc_clc_send_confirm_accept() with 2 subfunctions; - Fix missing byte order conversion of GID and token in CLC handshake, which is in a separate patch sending to net: https://lore.kernel.org/netdev/1701882157-87956-1-git-send-email-guwen@linux.alibaba.com/ - Patch #7: add extended GID in SMC-D lgr netlink attribute; v3->v2: https://lore.kernel.org/netdev/1701343695-122657-1-git-send-email-guwen@linux.alibaba.com/ - Rename smc_clc_fill_fce as smc_clc_fill_fce_v2x; - Remove ISM_IDENT_MASK from drivers/s390/net/ism.h; - Add explicitly assigning 'false' to ism_v2_capable in ism_dev_init(); - Remove smc_ism_set_v2_capable() helper for now, and introduce it in later loopback-ism implementation; v2->v1: - Fix sparse complaint; - Rebase to the latest net-next; ==================== Signed-off-by:
-
Wen Gu authored
The System EID (SEID) is an internal EID that is used by the SMCv2 software stack that has a predefined and constant value representing the s390 physical machine that the OS is executing on. So it should be managed by SMC stack instead of ISM driver and be consistent for all ISMv2 device (including virtual ISM devices) on s390 architecture. Suggested-by:
Alexandra Winter <wintera@linux.ibm.com> Signed-off-by:
Wen Gu <guwen@linux.alibaba.com> Reviewed-and-tested-by:
Wenjia Zhang <wenjia@linux.ibm.com> Reviewed-by:
-
Wen Gu authored
The system EID (SEID) is an internal EID used by SMC-D to represent the s390 physical machine that OS is executing on. On s390 architecture, it predefined by fixed string and part of cpuid and is enabled regardless of whether underlay device is virtual ISM or platform firmware ISM. However on non-s390 architectures where SMC-D can be used with virtual ISM devices, there is no similar information to identify physical machines, especially in virtualization scenarios. So in such cases, SEID is forcibly disabled and the user-defined UEID will be used to represent the communicable space. Signed-off-by:
-
Wen Gu authored
Virtual ISM devices introduced in SMCv2.1 requires a 128 bit extended GID vs. the existing ISM 64bit GID. So the 2nd 64 bit of extended GID should be included in SMC-D linkgroup netlink attribute as well. Signed-off-by:
-
Wen Gu authored
According to virtual ISM support feature defined by SMCv2.1, GIDs of virtual ISM device are UUIDs defined by RFC4122, which are 128-bits long. So some adaptation work is required. And note that the GIDs of existing platform firmware ISM devices still remain 64-bits long. Signed-off-by:
-
Wen Gu authored
According to virtual ISM support feature defined by SMCv2.1, CHIDs in the range 0xFF00 to 0xFFFF are reserved for use by virtual ISM devices. And two helpers are introduced to distinguish virtual ISM devices from the existing platform firmware ISM devices. Signed-off-by:
-
Wen Gu authored
This introduces virtual ISM device support feature to SMCv2.1 as the first supplemental feature. Signed-off-by:
-
Wen Gu authored
This patch adds SMCv2.x supplemental features negotiation. Supported SMCv2.x supplemental features are represented by feature_mask in FCE field. The negotiation process is as follows. Server Client Proposal(features(c-mask bits)) <----------------------------------------- Accept(features(s-mask bits)) -----------------------------------------> Confirm(features(s&c-mask bits)) <----------------------------------------- Signed-off-by: -
Wen Gu authored
The structs of CLC accept and confirm messages for SMCv1 and SMCv2 are separately defined and often casted to each other in the code, which may increase the risk of errors caused by future divergence of them. So unify them into one struct for better maintainability. Suggested-by:
-
Wen Gu authored
There is a large if-else block in smc_clc_send_confirm_accept() and it is better to split it into two sub-functions. Suggested-by:
-
Wen Gu authored
Rename some functions or variables with 'fce' in their name but used in SMCv2.1 as 'fce_v2x' for clarity. Signed-off-by:
-
- 25 Dec, 2023 1 commit
-
-
Marek Behún authored
Commit 2f3ce7a5 ("net: sfp: rework the RollBall PHY waiting code") changed the long wait before accessing RollBall / FS modules into probing for PHY every 1 second, and trying 25 times. Wei Lei reports that this does not work correctly on FS modules: when initializing, they may report values different from 0xffff in PHY ID registers for some MMDs, causing get_phy_c45_ids() to find some bogus MMD. Fix this by adding the module_t_wait member back, and setting it to 4 seconds for FS modules. Fixes: 2f3ce7a5 ("net: sfp: rework the RollBall PHY waiting code") Reported-by:
Wei Lei <quic_leiwei@quicinc.com> Signed-off-by:
Marek Behún <kabel@kernel.org> Tested-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
- 23 Dec, 2023 20 commits
-
-
David S. Miller authored
Ioana Ciornei says: ==================== dpaa2-switch: small improvements This patch set consists of a series of small improvements on the dpaa2-switch driver ranging from adding some more verbosity when encountering errors to reorganizing code to be easily extensible. Changes in v3: - 4/8: removed the fixes tag and moved it to the commit message - 5/8: specified that there is no user-visible effect - 6/8: removed the initialization of the err variable Changes in v2: - No changes to the actual diff, only rephrased some commit messages and added more information. ==================== Signed-off-by: -
Ioana Ciornei authored
In case a DPAA2 switch interface joins a bridge, the FDB used on the port will be changed to the one associated with the bridge. What this means exactly is that any VLAN installed on the port will need to be removed and then installed back so that it points to the new FDB. Once this is done, the previous FDB will become unused (no VLAN to point to it). Even though no traffic will reach this FDB, it's best to just cleanup the state of the FDB by zeroing its egress flood domain. Reviewed-by:
Simon Horman <horms@kernel.org> Signed-off-by:
Ioana Ciornei <ioana.ciornei@nxp.com> Signed-off-by:
-
Ioana Ciornei authored
Two different DPAA2 switch ports from two different DPSW instances cannot be under the same bridge. Instead of checking for this unsupported configuration in the CHANGEUPPER event, check it as early as possible in the PRECHANGEUPPER one. Reviewed-by:
-
Ioana Ciornei authored
Create separate functions, dpaa2_switch_port_prechangeupper and dpaa2_switch_port_changeupper, to be called directly when a DPSW port changes its upper device. This way we are not open-coding everything in the main event callback and we can easily extent, for example, with bond offload. Signed-off-by:
-
Ioana Ciornei authored
The DPSW object has multiple event sources multiplexed over the same IRQ. The driver has the capability to configure only some of these events to trigger the IRQ. The dpsw_get_irq_status() can clear events automatically based on the value stored in the 'status' variable passed to it. We don't want that to happen because we could get into a situation when we are clearing more events than we actually handled. Just resort to manually clearing the events that we handled. Also, since status is not used on the out path we remove its initialization to zero. This change does not have a user-visible effect because the dpaa2-switch driver enables and handles all the DPSW events which exist at the moment. Signed-off-by:
-
Ioana Ciornei authored
Commit 84cba729 ("dpaa2-switch: integrate the MAC endpoint support") added support for MAC endpoints in the dpaa2-switch driver but omitted to add the ENDPOINT_CHANGED irq to the list of interrupt sources. Fix this by extending the list of events which can raise an interrupt by extending the mask passed to the dpsw_set_irq_mask() firmware API. There is no user visible impact even without this patch since whenever a switch interface is connected/disconnected from an endpoint both events are set (LINK_CHANGED and ENDPOINT_CHANGED) and, luckily, the LINK_CHANGED event could actually raise the interrupt and thus get the MAC/PHY SW configuration started. Even with this, it's better to just not rely on undocumented firmware behavior which can change. Signed-off-by:
-
Ioana Ciornei authored
Print a netdev error when we hit a case in which a specific VLAN is already configured on the port. While at it, change the already existing netdev_warn into an _err for consistency purposes. Reviewed-by:
-
Ioana Ciornei authored
There is no restriction around the change of the MAC address on the switch ports, thus declare the interface netdevs IFF_LIVE_ADDR_CHANGE capable. Reviewed-by:
-
Ioana Ciornei authored
There is no point in updating the MAC address of a switch interface each time the link state changes, this only needs to happen in case the endpoint changes (the switch interface is [dis]connected from/to a MAC). Just move the call to dpaa2_switch_port_set_mac_addr() under DPSW_IRQ_EVENT_ENDPOINT_CHANGED. Reviewed-by:
-
David S. Miller authored
Roger Quadros says: ==================== net: ethernet: am65-cpsw: Add mqprio, frame preemption & coalescing This series adds mqprio qdisc offload in channel mode, Frame Preemption MAC merge support and RX/TX coalesing for AM65 CPSW driver. In v11 following changes were made - Fix patch "net: ethernet: ti: am65-cpsw: add mqprio qdisc offload in channel mode" by including units.h Changelog information in each patch file. ==================== Signed-off-by: -
Grygorii Strashko authored
Add SW IRQ coalescing based on hrtimers for TX and RX data path which can be enabled by ethtool commands: - RX coalescing ethtool -C eth1 rx-usecs 50 - TX coalescing can be enabled per TX queue - by default enables coalesing for TX0 ethtool -C eth1 tx-usecs 50 - configure TX0 ethtool -Q eth0 queue_mask 1 --coalesce tx-usecs 100 - configure TX1 ethtool -Q eth0 queue_mask 2 --coalesce tx-usecs 100 - configure TX0 and TX1 ethtool -Q eth0 queue_mask 3 --coalesce tx-usecs 100 --coalesce tx-usecs 100 show configuration for TX0 and TX1: ethtool -Q eth0 queue_mask 3 --show-coalesce Comparing to gro_flush_timeout and napi_defer_hard_irqs, this patch allows to enable IRQ coalesing for RX path separately. Signed-off-by:
Grygorii Strashko <grygorii.strashko@ti.com> Signed-off-by:
Roger Quadros <rogerq@kernel.org> Signed-off-by:
-
Roger Quadros authored
Add driver support for viewing / changing the MAC Merge sublayer parameters and seeing the verification state machine's current state via ethtool. As hardware does not support interrupt notification for verification events we resort to polling on link up. On link up we try a couple of times for verification success and if unsuccessful then give up. The Frame Preemption feature is described in the Technical Reference Manual [1] in section: 12.3.1.4.6.7 Intersperced Express Traffic (IET – P802.3br/D2.0) Due to Silicon Errata i2208 [2] we set limit min IET fragment size to 124 (excluding 4 bytes mCRC). [1] AM62x TRM - https://www.ti.com/lit/ug/spruiv7a/spruiv7a.pdf [2] AM62x Silicon Errata - https://www.ti.com/lit/er/sprz487c/sprz487c.pdfSigned-off-by:
Vladimir Oltean <vladimir.oltean@nxp.com> Signed-off-by:
-
Grygorii Strashko authored
This patch adds MQPRIO Qdisc offload in full 'channel' mode which allows not only setting up pri:tc mapping, but also configuring TX shapers (rate-limiting) on external port FIFOs. The MQPRIO Qdisc offload is expected to work with or without VLAN/priority tagged packets. The CPSW external Port FIFO has 8 Priority queues. The rate-limit can be set for each of these priority queues. Which Priority queue a packet is assigned to depends on PN_REG_TX_PRI_MAP register which maps header priority to switch priority. The header priority of a packet is assigned via the RX_PRI_MAP_REG which maps packet priority to header priority. The packet priority is either the VLAN priority (for VLAN tagged packets) or the thread/channel offset. For simplicity, we assign the same priority queue to all queues of a Traffic Class so it can be rate-limited correctly. Configuration example: ethtool -L eth1 tx 5 ethtool --set-priv-flags eth1 p0-rx-ptype-rrobin off tc qdisc add dev eth1 parent root handle 100: mqprio num_tc 3 \ map 0 0 1 2 0 0 0 0 0 0 0 0 0 0 0 0 \ queues 1@0 1@1 1@2 hw 1 mode channel \ shaper bw_rlimit min_rate 0 100mbit 200mbit max_rate 0 101mbit 202mbit tc qdisc replace dev eth2 handle 100: parent root mqprio num_tc 1 \ map 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 queues 1@0 hw 1 ip link add link eth1 name eth1.100 type vlan id 100 ip link set eth1.100 type vlan egress 0:0 1:1 2:2 3:3 4:4 5:5 6:6 7:7 In the above example two ports share the same TX CPPI queue 0 for low priority traffic. 3 traffic classes are defined for eth1 and mapped to: TC0 - low priority, TX CPPI queue 0 -> ext Port 1 fifo0, no rate limit TC1 - prio 2, TX CPPI queue 1 -> ext Port 1 fifo1, CIR=100Mbit/s, EIR=1Mbit/s TC2 - prio 3, TX CPPI queue 2 -> ext Port 1 fifo2, CIR=200Mbit/s, EIR=2Mbit/s Signed-off-by:
-
Roger Quadros authored
Move register definitions to header file. No functional change. Signed-off-by:
-
Roger Quadros authored
Move this code around to avoid forward declaration. No functional change. Signed-off-by:
-
Roger Quadros authored
Handle offloading commands using switch-case in am65_cpsw_setup_taprio(). Move checks to am65_cpsw_taprio_replace(). Use NL_SET_ERR_MSG_MOD for error messages. Change error message from "Failed to set cycle time extension" to "cycle time extension not supported" Signed-off-by:
-
Roger Quadros authored
We will use this Kconfig option to not only enable TAS/EST offload but also other QoS features like Multiqueue priority descriptors and MAC-Merge/Frame Preemption. TI_AM65_CPSW_QOS seems a more appropriate Kconfig option name than TI_AM65_CPSW_TAS. Signed-off-by:
-
Roger Quadros authored
Build am65-cpsw-qos only if CONFIG_TI_AM65_CPSW_TAS is enabled. Signed-off-by:
-
Vladimir Oltean authored
Some devices do not support individual 'pmac' and 'emac' stats. For such devices, resort to 'aggregate' stats. Cc: Shuah Khan <shuah@kernel.org> Signed-off-by:
-
Vladimir Oltean authored
Some devices have errata due to which they cannot report ETH_ZLEN (60) in the rx-min-frag-size. This was foreseen of course, and lldpad has logic that when we request it to advertise addFragSize 0, it will round it up to the lowest value that is _actually_ supported by the hardware. The problem is that the selftest expects lldpad to report back to us the same value as we requested. Make the selftest smarter by figuring out on its own what is a reasonable value to expect. Cc: Shuah Khan <shuah@kernel.org> Signed-off-by:
-
