- 20 Apr, 2022 30 commits
-
-
David S. Miller authored
Boris Sukholitko says: ==================== net/sched: flower: match on the number of vlan tags Our customers in the fiber telecom world have network configurations where they would like to control their traffic according to the number of tags appearing in the packet. For example, TR247 GPON conformance test suite specification mostly talks about untagged, single, double tagged packets and gives lax guidelines on the vlan protocol vs. number of vlan tags. This is different from the common IT networks where 802.1Q and 802.1ad protocols are usually describe single and double tagged packet. GPON configurations that we work with have arbitrary mix the above protocols and number of vlan tags in the packet. The following patch series implement number of vlans flower filter. They add num_of_vlans flower filter as an alternative to vlan ethtype protocol matching. The end result is that the following command becomes possible: tc filter add dev eth1 ingress flower \ num_of_vlans 1 vlan_prio 5 action drop Also, from our logs, we have redirect rules such that: tc filter add dev $GPON ingress flower num_of_vlans $N \ action mirred egress redirect dev $DEV where N can range from 0 to 3 and $DEV is the function of $N. Also there are rules setting skb mark based on the number of vlans: tc filter add dev $GPON ingress flower num_of_vlans $N vlan_prio \ $P action skbedit mark $M More about the patch series: - patches 1-2 remove duplicate code by introducing is_key_vlan helper. - patch 3, 4 implement num_of_vlans in the dissector and in the flower. - patch 5 uses the num_of_vlans filter to allow further matching on vlan attributes. Complementary iproute2 patches are being sent separately. Thanks, Boris. - v4: rebased to the latest net-next - v3: - more example commands in patch 3 description (request by Jamal) - patch 5 description made clearer (thanks to Jiri) - v2: - add suitable subject prefixes - more evolved patch 5 description ==================== Signed-off-by:David S. Miller <davem@davemloft.net>
-
Boris Sukholitko authored
Before this patch the existence of vlan filters was conditional on the vlan protocol being matched in the tc rule. For example, the following rule: tc filter add dev eth1 ingress flower vlan_prio 5 was illegal because vlan protocol (e.g. 802.1q) does not appear in the rule. Remove the above restriction by looking at the num_of_vlans filter to allow further matching on vlan attributes. The following rule becomes legal as a result of this commit: tc filter add dev eth1 ingress flower num_of_vlans 1 vlan_prio 5 because having num_of_vlans==1 implies that the packet is single tagged. Change is_vlan_key helper to look at the number of vlans in addition to the vlan ethertype. The outcome of this change is that outer (e.g. vlan_prio) and inner (e.g. cvlan_prio) tag vlan filters require the number of vlan tags to be greater then 0 and 1 accordingly. As a result of is_vlan_key change, the ethertype may be set to 0 when matching on the number of vlans. Update fl_set_key_vlan to avoid setting key, mask vlan_tpid for the 0 ethertype. Signed-off-by:
Boris Sukholitko <boris.sukholitko@broadcom.com> Signed-off-by:
-
Boris Sukholitko authored
These are bookkeeping parts of the new num_of_vlans filter. Defines, dump, load and set are being done here. Signed-off-by:
-
Boris Sukholitko authored
Our customers in the fiber telecom world have network configurations where they would like to control their traffic according to the number of tags appearing in the packet. For example, TR247 GPON conformance test suite specification mostly talks about untagged, single, double tagged packets and gives lax guidelines on the vlan protocol vs. number of vlan tags. This is different from the common IT networks where 802.1Q and 802.1ad protocols are usually describe single and double tagged packet. GPON configurations that we work with have arbitrary mix the above protocols and number of vlan tags in the packet. The goal is to make the following TC commands possible: tc filter add dev eth1 ingress flower \ num_of_vlans 1 vlan_prio 5 action drop From our logs, we have redirect rules such that: tc filter add dev $GPON ingress flower num_of_vlans $N \ action mirred egress redirect dev $DEV where N can range from 0 to 3 and $DEV is the function of $N. Also there are rules setting skb mark based on the number of vlans: tc filter add dev $GPON ingress flower num_of_vlans $N vlan_prio \ $P action skbedit mark $M This new dissector allows extracting the number of vlan tags existing in the packet. Signed-off-by: -
Boris Sukholitko authored
Whitespace only. Signed-off-by:
-
Boris Sukholitko authored
There are somewhat repetitive ethertype checks in fl_set_key. Refactor them into is_vlan_key helper function. To make the changes clearer, avoid touching identation levels. This is the job for the next patch in the series. Signed-off-by:
-
Haowen Bai authored
Use kzalloc rather than duplicating its implementation, which makes code simple and easy to understand. Signed-off-by:
Haowen Bai <baihaowen@meizu.com> Signed-off-by:
-
Luiz Angelo Daros de Luca authored
There is no need to add new compatible strings for each new supported chip version. The compatible string is used only to select the subdriver (rtl8365mb.c or rtl8366rb.c). Once in the subdriver, it will detect the chip model by itself, ignoring which compatible string was used. Link: https://lore.kernel.org/netdev/20220414014055.m4wbmr7tdz6hsa3m@bang-olufsen.dk/Signed-off-by:
Luiz Angelo Daros de Luca <luizluca@gmail.com> Reviewed-by:
Alvin Šipraga <alsi@bang-olufsen.dk> Reviewed-by:
Florian Fainelli <f.fainelli@gmail.com> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Acked-by:
Arınç ÜNAL <arinc.unal@arinc9.com> Signed-off-by:
-
Luiz Angelo Daros de Luca authored
Compatible strings are used to help the driver find the chip ID/version register for each chip family. After that, the driver can setup the switch accordingly. Keep only the first supported model for each family as a compatible string and reference other chip models in the description. The removed compatible strings have never been used in a released kernel. CC: devicetree@vger.kernel.org Link: https://lore.kernel.org/netdev/20220414014055.m4wbmr7tdz6hsa3m@bang-olufsen.dk/Signed-off-by:
-
David S. Miller authored
Guangbin Huang says: ==================== net: hns3: updates for -next This series includes some updates for the HNS3 ethernet driver. Change logs: V1 -> V2: - Fix failed to apply to net-next problem. ==================== Signed-off-by: -
Hao Chen authored
Remove unnecessary line wrap for hns3_set_tunable to improve function readability. Signed-off-by:
Hao Chen <chenhao288@hisilicon.com> Signed-off-by:
Guangbin Huang <huangguangbin2@huawei.com> Signed-off-by:
-
Peng Li authored
Magic values are not recommended. Signed-off-by: Peng Li<lipeng321@huawei.com> Signed-off-by:
-
Peng Li authored
This patch fixes wrong words in comments. Signed-off-by: Peng Li<lipeng321@huawei.com> Signed-off-by:
-
Peng Li authored
The param of function hclgevf_get_mbx_resp has been changed but the comments not upodated. This patch updates it. Signed-off-by: Peng Li<lipeng321@huawei.com> Signed-off-by:
-
Hao Chen authored
For the active tx spare buffer size maybe changed according to the page size, so add log to notice it. Signed-off-by:
-
Jie Wang authored
Currently, There is a low probability that pf mtu configuration fails, but the information in logs is insufficient for problem locating when the VF mtu value is illegally modified. So record the vf index and vf mtu value at the failure scenario. Signed-off-by:
Jie Wang <wangjie125@huawei.com> Signed-off-by:
-
Jian Shen authored
The struct hclge_pf_to_vf_msg is used for mailbox message from PF to VF, including both response and request. But its definition can only indicate respone, which makes the message data copy in function hclge_send_mbx_msg() unreadable. So refine it by edding a general message definition into it. Signed-off-by:
Jian Shen <shenjian15@huawei.com> Signed-off-by:
-
Hao Chen authored
Use struct hns3_ring_param to replace variable new/old_xxx and add hns3_is_ringparam_changed() to judge them if is changed to improve code readability. Signed-off-by:
-
Yufeng Mo authored
For DEVICE_VERSION_V2, the hardware does not support the CQE mode. So add capability bit for coalesce CQE mode and add parameter check for it in ethtool. Signed-off-by:
Yufeng Mo <moyufeng@huawei.com> Signed-off-by:
-
David S. Miller authored
[PATCH net-next v5 0/3] net: atlantic: Add XDP support @ 2022-04-17 10:12 Taehee Yoo 2022-04-17 10:12 ` [PATCH net-next v5 1/3] net: atlantic: Implement xdp control plane Taehee Yoo ` (2 more replies) 0 siblings, 3 replies; 4+ messages in thread From: Taehee Yoo @ 2022-04-17 10:12 UTC (permalink / raw) To: davem, kuba, pabeni, netdev, irusskikh, ast, daniel, hawk, john.fastabend, andrii, kafai, songliubraving, yhs, kpsingh, bpf Cc: ap420073 This patchset is to make atlantic to support multi-buffer XDP. The first patch implement control plane of xdp. The aq_xdp(), callback of .xdp_bpf is added. The second patch implements data plane of xdp. XDP_TX, XDP_DROP, and XDP_PASS is supported. __aq_ring_xdp_clean() is added to receive and execute xdp program. aq_nic_xmit_xdpf() is added to send packet by XDP. The third patch implements callback of .ndo_xdp_xmit. aq_xdp_xmit() is added to send redirected packets and it internally calls aq_nic_xmit_xdpf(). Memory model is MEM_TYPE_PAGE_SHARED. Order-2 page allocation is used when XDP is enabled. LRO will be disabled if XDP program doesn't supports multi buffer. AQC chip supports 32 multi-queues and 8 vectors(irq). There are two options. 1. under 8 cores and maximum 4 tx queues per core. 2. under 4 cores and maximum 8 tx queues per core. Like other drivers, these tx queues can be used only for XDP_TX, XDP_REDIRECT queue. If so, no tx_lock is needed. But this patchset doesn't use this strategy because getting hardware tx queue index cost is too high. So, tx_lock is used in the aq_nic_xmit_xdpf(). single-core, single queue, 80% cpu utilization. 32.30% [kernel] [k] aq_get_rxpages_xdp 10.44% [kernel] [k] aq_hw_read_reg <---------- here 9.86% bpf_prog_xxx_xdp_prog_tx [k] bpf_prog_xxx_xdp_prog_tx 5.51% [kernel] [k] aq_ring_rx_clean single-core, 8 queues, 100% cpu utilization, half PPS. 52.03% [kernel] [k] aq_hw_read_reg <---------- here 18.24% [kernel] [k] aq_get_rxpages_xdp 4.30% [kernel] [k] hw_atl_b0_hw_ring_rx_receive 4.24% bpf_prog_xxx_xdp_prog_tx [k] bpf_prog_xxx_xdp_prog_tx 2.79% [kernel] [k] aq_ring_rx_clean Performance result(64 Byte) 1. XDP_TX a. xdp_geieric, single core - 2.5Mpps, 100% cpu b. xdp_driver, single core - 4.5Mpps, 80% cpu c. xdp_generic, 8 core(hyper thread) - 6.3Mpps, 40% cpu d. xdp_driver, 8 core(hyper thread) - 6.3Mpps, 30% cpu 2. XDP_REDIRECT a. xdp_generic, single core - 2.3Mpps b. xdp_driver, single core - 4.5Mpps v5: - Use MEM_TYPE_PAGE_SHARED instead of MEM_TYPE_PAGE_ORDER0 - Use 2K frame size instead of 3K - Use order-2 page allocation instead of order-0 - Rename aq_get_rxpage() to aq_alloc_rxpages() - Add missing PageFree stats for ethtool - Remove aq_unset_rxpage_xdp(), introduced by v2 patch due to change of memory model - Fix wrong last parameter value of xdp_prepare_buff() - Add aq_get_rxpages_xdp() to increase page reference count v4: - Fix compile warning v3: - Change wrong PPS performance result 40% -> 80% in single core(Intel i3-12100) - Separate aq_nic_map_xdp() from aq_nic_map_skb() - Drop multi buffer packets if single buffer XDP is attached - Disable LRO when single buffer XDP is attached - Use xdp_get_{frame/buff}_len() v2: - Do not use inline in C file Taehee Yoo (3): net: atlantic: Implement xdp control plane net: atlantic: Implement xdp data plane net: atlantic: Implement .ndo_xdp_xmit handler .../net/ethernet/aquantia/atlantic/aq_cfg.h | 1 + .../ethernet/aquantia/atlantic/aq_ethtool.c | 9 + .../net/ethernet/aquantia/atlantic/aq_main.c | 87 ++++ .../net/ethernet/aquantia/atlantic/aq_main.h | 2 + .../net/ethernet/aquantia/atlantic/aq_nic.c | 136 ++++++ .../net/ethernet/aquantia/atlantic/aq_nic.h | 5 + .../net/ethernet/aquantia/atlantic/aq_ring.c | 409 ++++++++++++++++-- .../net/ethernet/aquantia/atlantic/aq_ring.h | 21 +- .../net/ethernet/aquantia/atlantic/aq_vec.c | 23 +- .../net/ethernet/aquantia/atlantic/aq_vec.h | 6 + .../aquantia/atlantic/hw_atl/hw_atl_a0.c | 6 +- .../aquantia/atlantic/hw_atl/hw_atl_b0.c | 10 +- 12 files changed, 670 insertions(+), 45 deletions(-) -- 2.17.1 ^ permalink raw reply [flat|nested] 4+ messages in thread * [PATCH net-next v5 1/3] net: atlantic: Implement xdp control plane 2022-04-17 10:12 [PATCH net-next v5 0/3] net: atlantic: Add XDP support Taehee Yoo @ 2022-04-17 10:12 ` Taehee Yoo 2022-04-17 10:12 ` [PATCH net-next v5 2/3] net: atlantic: Implement xdp data plane Taehee Yoo 2022-04-17 10:12 ` [PATCH net-next v5 3/3] net: atlantic: Implement .ndo_xdp_xmit handler Taehee Yoo 2 siblings, 0 replies; 4+ messages in thread From: Taehee Yoo @ 2022-04-17 10:12 UTC (permalink / raw) To: davem, kuba, pabeni, netdev, irusskikh, ast, daniel, hawk, john.fastabend, andrii, kafai, songliubraving, yhs, kpsingh, bpf Cc: ap420073 aq_xdp() is a xdp setup callback function for Atlantic driver. When XDP is attached or detached, the device will be restarted because it uses different headroom, tailroom, and page order value. If XDP enabled, it switches default page order value from 0 to 2. Because the default maximum frame size is still 2K and it needs additional area for headroom and tailroom. The total size(headroom + frame size + tailroom) is 2624. So, 1472Bytes will be always wasted for every frame. But when order-2 is used, these pages can be used 6 times with flip strategy. It means only about 106Bytes per frame will be wasted. Also, It supports xdp fragment feature. MTU can be 16K if xdp prog supports xdp fragment. If not, MTU can not exceed 2K - ETH_HLEN - ETH_FCS. And a static key is added and It will be used to call the xdp_clean handler in ->poll(). data plane implementation will be contained the followed patch. Signed-off-by:Taehee Yoo <ap420073@gmail.com> --- v5: - Use MEM_TYPE_PAGE_SHARED instead of MEM_TYPE_PAGE_ORDER0 - Use 2K frame size instead of 3K - Use order-2 page allocation instead of order-0 - Rename aq_get_rxpage() to aq_alloc_rxpages() v4: - No changed v3: - Disable LRO when single buffer XDP is attached v2: - No changed
-
Taehee Yoo authored
aq_xdp_xmit() is the callback function of .ndo_xdp_xmit. It internally calls aq_nic_xmit_xdpf() to send packet. Signed-off-by:
-
Taehee Yoo authored
It supports XDP_PASS, XDP_DROP and multi buffer. The new function aq_nic_xmit_xdpf() is used to send packet with xdp_frame and internally it calls aq_nic_map_xdp(). AQC chip supports 32 multi-queues and 8 vectors(irq). there are two option 1. under 8 cores and 4 tx queues per core. 2. under 4 cores and 8 tx queues per core. Like ixgbe, these tx queues can be used only for XDP_TX, XDP_REDIRECT queue. If so, no tx_lock is needed. But this patchset doesn't use this strategy because getting hardware tx queue index cost is too high. So, tx_lock is used in the aq_nic_xmit_xdpf(). single-core, single queue, 80% cpu utilization. 30.75% bpf_prog_xxx_xdp_prog_tx [k] bpf_prog_xxx_xdp_prog_tx 10.35% [kernel] [k] aq_hw_read_reg <---------- here 4.38% [kernel] [k] get_page_from_freelist single-core, 8 queues, 100% cpu utilization, half PPS. 45.56% [kernel] [k] aq_hw_read_reg <---------- here 17.58% bpf_prog_xxx_xdp_prog_tx [k] bpf_prog_xxx_xdp_prog_tx 4.72% [kernel] [k] hw_atl_b0_hw_ring_rx_receive The new function __aq_ring_xdp_clean() is a xdp rx handler and this is called only when XDP is attached. Signed-off-by:
-
Taehee Yoo authored
aq_xdp() is a xdp setup callback function for Atlantic driver. When XDP is attached or detached, the device will be restarted because it uses different headroom, tailroom, and page order value. If XDP enabled, it switches default page order value from 0 to 2. Because the default maximum frame size is still 2K and it needs additional area for headroom and tailroom. The total size(headroom + frame size + tailroom) is 2624. So, 1472Bytes will be always wasted for every frame. But when order-2 is used, these pages can be used 6 times with flip strategy. It means only about 106Bytes per frame will be wasted. Also, It supports xdp fragment feature. MTU can be 16K if xdp prog supports xdp fragment. If not, MTU can not exceed 2K - ETH_HLEN - ETH_FCS. And a static key is added and It will be used to call the xdp_clean handler in ->poll(). data plane implementation will be contained the followed patch. Signed-off-by:
-
David S. Miller authored
Vladimir Oltean says: ==================== DSA cross-chip notifier cleanups This patch set makes the following improvements: - Cross-chip notifiers pass a switch index, port index, sometimes tree index, all as integers. Sometimes we need to recover the struct dsa_port based on those integers. That recovery involves traversing a list. By passing directly a pointer to the struct dsa_port we can avoid that, and the indices passed previously can still be obtained from the passed struct dsa_port. - Resetting VLAN filtering on a switch has explicit code to make it run on a single switch, so it has no place to stay in the cross-chip notifier code. Move it out. - Changing the MTU on a user port affects only that single port, yet the code passes through the cross-chip notifier layer where all switches are notified. Avoid that. - Other related cosmetic changes in the MTU changing procedure. Apart from the slight improvement in performance given by (a) doing less work in cross-chip notifiers (b) emitting less cross-chip notifiers we also end up with about 100 less lines of code. ==================== Signed-off-by: -
Vladimir Oltean authored
A cross-chip notifier with "targeted_match=true" is one that matches only the local port of the switch that emitted it. In other words, passing through the cross-chip notifier layer serves no purpose. Eliminate this concept by calling directly ds->ops->port_change_mtu instead of emitting a targeted cross-chip notifier. This leaves the DSA_NOTIFIER_MTU event being emitted only for MTU updates on the CPU port, which need to be reflected also across all DSA links. Signed-off-by:
Vladimir Oltean <vladimir.oltean@nxp.com> Signed-off-by:
-
Vladimir Oltean authored
We can get a hold of the "ds" pointer directly from "dp", no need for the dsa_slave_priv. Signed-off-by:
-
Vladimir Oltean authored
We could retrieve the cpu_dp pointer directly from the "dp" we already have, no need to resort to dsa_to_port(ds, port). This change also removes the need for an "int port", so that is also deleted. Signed-off-by:
-
Vladimir Oltean authored
Use the more conventional iterator over user ports instead of explicitly ignoring them, and use the more conventional name "other_dp" instead of "dp_iter", for readability. Signed-off-by:
-
Vladimir Oltean authored
To determine whether a given port should react to the port targeted by the notifier, dsa_port_host_vlan_match() and dsa_port_host_address_match() look at the positioning of the switch port currently executing the notifier relative to the switch port for which the notifier was emitted. To maintain stylistic compatibility with the other match functions from switch.c, the host address and host VLAN match functions take the notifier information about targeted port, switch and tree indices as argument. However, these functions only use that information to retrieve the struct dsa_port *targeted_dp, which is an invariant for the outer loop that calls them. So it makes more sense to calculate the targeted dp only once, and pass it to them as argument. But furthermore, the targeted dp is actually known at the time the call to dsa_port_notify() is made. It is just that we decide to only save the indices of the port, switch and tree in the notifier structure, just to retrace our steps and find the dp again using dsa_switch_find() and dsa_to_port(). But both the above functions are relatively expensive, since they need to iterate through lists. It appears more straightforward to make all notifiers just pass the targeted dp inside their info structure, and have the code that needs the indices to look at info->dp->index instead of info->port, or info->dp->ds->index instead of info->sw_index, or info->dp->ds->dst->index instead of info->tree_index. For the sake of consistency, all cross-chip notifiers are converted to pass the "dp" directly. Signed-off-by:
-
Vladimir Oltean authored
In dsa_port_switchdev_unsync_attrs() there is a comment that resetting the VLAN filtering isn't done where it is expected. And since commit 108dc874 ("net: dsa: Avoid cross-chip syncing of VLAN filtering"), there is no reason to handle this in switch.c either. Therefore, move the logic to port.c, and adapt it slightly to the data structures and naming conventions from there. Signed-off-by:
-
- 19 Apr, 2022 8 commits
-
-
Paolo Abeni authored
Florent Fourcot says: ==================== rtnetlink: improve ALT_IFNAME config and fix dangerous GROUP usage First commit forbids dangerous calls when both IFNAME and GROUP are given, since it can introduce unexpected behaviour when IFNAME does not match any interface. Second patch achieves primary goal of this patchset to fix/improve IFLA_ALT_IFNAME attribute, since previous code was never working for newlink/setlink. ip-link command is probably getting interface index before, and was not using this feature. Last two patches are improving error code on corner cases. Changes in v2: * Remove ifname argument in rtnl_dev_get/do_setlink functions (simplify code) * Use a boolean to avoid condition duplication in __rtnl_newlink Changes in v3: * Simplify rtnl_dev_get signature Changes in v4: * Rename link_lookup to link_specified Changes in v5: * Re-order patches ==================== Link: https://lore.kernel.org/r/20220415165330.10497-1-florent.fourcot@wifirst.frSigned-off-by:Paolo Abeni <pabeni@redhat.com>
-
Florent Fourcot authored
A request without interface name/interface index/interface group cannot work. We should return EINVAL Signed-off-by:
Florent Fourcot <florent.fourcot@wifirst.fr> Signed-off-by:
Brian Baboch <brian.baboch@wifirst.fr> Reviewed-by:
Jakub Kicinski <kuba@kernel.org> Signed-off-by:
-
Florent Fourcot authored
If IFLA_ALT_IFNAME is set and given interface is not found, we should return ENODEV and be consistent with IFLA_IFNAME behaviour This commit extends feature of commit 76c9ac0e, "net: rtnetlink: add possibility to use alternative names as message handle" CC: Jiri Pirko <jiri@mellanox.com> Signed-off-by:
-
Florent Fourcot authored
buffer called "ifname" given in function rtnl_dev_get is always valid when called by setlink/newlink, but contains only empty string when IFLA_IFNAME is not given. So IFLA_ALT_IFNAME is always ignored This patch fixes rtnl_dev_get function with a remove of ifname argument, and move ifname copy in do_setlink when required. It extends feature of commit 76c9ac0e, "net: rtnetlink: add possibility to use alternative names as message handle"" CC: Jiri Pirko <jiri@mellanox.com> Signed-off-by:
-
Florent Fourcot authored
When the interface does not exist, and a group is given, the given parameters are being set to all interfaces of the given group. The given IFNAME/ALT_IF_NAME are being ignored in that case. That can be dangerous since a typo (or a deleted interface) can produce weird side effects for caller: Case 1: IFLA_IFNAME=valid_interface IFLA_GROUP=1 MTU=1234 Case 1 will update MTU and group of the given interface "valid_interface". Case 2: IFLA_IFNAME=doesnotexist IFLA_GROUP=1 MTU=1234 Case 2 will update MTU of all interfaces in group 1. IFLA_IFNAME is ignored in this case This behaviour is not consistent and dangerous. In order to fix this issue, we now return ENODEV when the given IFNAME does not exist. Signed-off-by:
-
Paolo Abeni authored
Tonghao Zhang says: ==================== net: sched: allow user to select txqueue From: Tonghao Zhang <xiangxia.m.yue@gmail.com> Patch 1 allow user to select txqueue in clsact hook. Patch 2 support skbhash to select txqueue. ==================== Link: https://lore.kernel.org/r/20220415164046.26636-1-xiangxia.m.yue@gmail.comSigned-off-by:
-
Tonghao Zhang authored
This patch allows users to pick queue_mapping, range from A to B. Then we can load balance packets from A to B tx queue. The range is an unsigned 16bit value in decimal format. $ tc filter ... action skbedit queue_mapping skbhash A B "skbedit queue_mapping QUEUE_MAPPING" (from "man 8 tc-skbedit") is enhanced with flags: SKBEDIT_F_TXQ_SKBHASH +----+ +----+ +----+ | P1 | | P2 | | Pn | +----+ +----+ +----+ | | | +-----------+-----------+ | | clsact/skbedit | MQ v +-----------+-----------+ | q0 | qn | qm v v v HTB/FQ FIFO ... FIFO For example: If P1 sends out packets to different Pods on other host, and we want distribute flows from qn - qm. Then we can use skb->hash as hash. setup commands: $ NETDEV=eth0 $ ip netns add n1 $ ip link add ipv1 link $NETDEV type ipvlan mode l2 $ ip link set ipv1 netns n1 $ ip netns exec n1 ifconfig ipv1 2.2.2.100/24 up $ tc qdisc add dev $NETDEV clsact $ tc filter add dev $NETDEV egress protocol ip prio 1 \ flower skip_hw src_ip 2.2.2.100 action skbedit queue_mapping skbhash 2 6 $ tc qdisc add dev $NETDEV handle 1: root mq $ tc qdisc add dev $NETDEV parent 1:1 handle 2: htb $ tc class add dev $NETDEV parent 2: classid 2:1 htb rate 100kbit $ tc class add dev $NETDEV parent 2: classid 2:2 htb rate 200kbit $ tc qdisc add dev $NETDEV parent 1:2 tbf rate 100mbit burst 100mb latency 1 $ tc qdisc add dev $NETDEV parent 1:3 pfifo $ tc qdisc add dev $NETDEV parent 1:4 pfifo $ tc qdisc add dev $NETDEV parent 1:5 pfifo $ tc qdisc add dev $NETDEV parent 1:6 pfifo $ tc qdisc add dev $NETDEV parent 1:7 pfifo $ ip netns exec n1 iperf3 -c 2.2.2.1 -i 1 -t 10 -P 10 pick txqueue from 2 - 6: $ ethtool -S $NETDEV | grep -i tx_queue_[0-9]_bytes tx_queue_0_bytes: 42 tx_queue_1_bytes: 0 tx_queue_2_bytes: 11442586444 tx_queue_3_bytes: 7383615334 tx_queue_4_bytes: 3981365579 tx_queue_5_bytes: 3983235051 tx_queue_6_bytes: 6706236461 tx_queue_7_bytes: 42 tx_queue_8_bytes: 0 tx_queue_9_bytes: 0 txqueues 2 - 6 are mapped to classid 1:3 - 1:7 $ tc -s class show dev $NETDEV ... class mq 1:3 root leaf 8002: Sent 11949133672 bytes 7929798 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 class mq 1:4 root leaf 8003: Sent 7710449050 bytes 5117279 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 class mq 1:5 root leaf 8004: Sent 4157648675 bytes 2758990 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 class mq 1:6 root leaf 8005: Sent 4159632195 bytes 2759990 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 class mq 1:7 root leaf 8006: Sent 7003169603 bytes 4646912 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 ... Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Cong Wang <xiyou.wangcong@gmail.com> Cc: Jiri Pirko <jiri@resnulli.us> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Jonathan Lemon <jonathan.lemon@gmail.com> Cc: Eric Dumazet <edumazet@google.com> Cc: Alexander Lobakin <alobakin@pm.me> Cc: Paolo Abeni <pabeni@redhat.com> Cc: Talal Ahmad <talalahmad@google.com> Cc: Kevin Hao <haokexin@gmail.com> Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org> Cc: Kees Cook <keescook@chromium.org> Cc: Kumar Kartikeya Dwivedi <memxor@gmail.com> Cc: Antoine Tenart <atenart@kernel.org> Cc: Wei Wang <weiwan@google.com> Cc: Arnd Bergmann <arnd@arndb.de> Signed-off-by:Tonghao Zhang <xiangxia.m.yue@gmail.com> Reviewed-by:
Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by:
-
Tonghao Zhang authored
This patch fixes issue: * If we install tc filters with act_skbedit in clsact hook. It doesn't work, because netdev_core_pick_tx() overwrites queue_mapping. $ tc filter ... action skbedit queue_mapping 1 And this patch is useful: * We can use FQ + EDT to implement efficient policies. Tx queues are picked by xps, ndo_select_queue of netdev driver, or skb hash in netdev_core_pick_tx(). In fact, the netdev driver, and skb hash are _not_ under control. xps uses the CPUs map to select Tx queues, but we can't figure out which task_struct of pod/containter running on this cpu in most case. We can use clsact filters to classify one pod/container traffic to one Tx queue. Why ? In containter networking environment, there are two kinds of pod/ containter/net-namespace. One kind (e.g. P1, P2), the high throughput is key in these applications. But avoid running out of network resource, the outbound traffic of these pods is limited, using or sharing one dedicated Tx queues assigned HTB/TBF/FQ Qdisc. Other kind of pods (e.g. Pn), the low latency of data access is key. And the traffic is not limited. Pods use or share other dedicated Tx queues assigned FIFO Qdisc. This choice provides two benefits. First, contention on the HTB/FQ Qdisc lock is significantly reduced since fewer CPUs contend for the same queue. More importantly, Qdisc contention can be eliminated completely if each CPU has its own FIFO Qdisc for the second kind of pods. There must be a mechanism in place to support classifying traffic based on pods/container to different Tx queues. Note that clsact is outside of Qdisc while Qdisc can run a classifier to select a sub-queue under the lock. In general recording the decision in the skb seems a little heavy handed. This patch introduces a per-CPU variable, suggested by Eric. The xmit.skip_txqueue flag is firstly cleared in __dev_queue_xmit(). - Tx Qdisc may install that skbedit actions, then xmit.skip_txqueue flag is set in qdisc->enqueue() though tx queue has been selected in netdev_tx_queue_mapping() or netdev_core_pick_tx(). That flag is cleared firstly in __dev_queue_xmit(), is useful: - Avoid picking Tx queue with netdev_tx_queue_mapping() in next netdev in such case: eth0 macvlan - eth0.3 vlan - eth0 ixgbe-phy: For example, eth0, macvlan in pod, which root Qdisc install skbedit queue_mapping, send packets to eth0.3, vlan in host. In __dev_queue_xmit() of eth0.3, clear the flag, does not select tx queue according to skb->queue_mapping because there is no filters in clsact or tx Qdisc of this netdev. Same action taked in eth0, ixgbe in Host. - Avoid picking Tx queue for next packet. If we set xmit.skip_txqueue in tx Qdisc (qdisc->enqueue()), the proper way to clear it is clearing it in __dev_queue_xmit when processing next packets. For performance reasons, use the static key. If user does not config the NET_EGRESS, the patch will not be compiled. +----+ +----+ +----+ | P1 | | P2 | | Pn | +----+ +----+ +----+ | | | +-----------+-----------+ | | clsact/skbedit | MQ v +-----------+-----------+ | q0 | q1 | qn v v v HTB/FQ HTB/FQ ... FIFO Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Cong Wang <xiyou.wangcong@gmail.com> Cc: Jiri Pirko <jiri@resnulli.us> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Cc: Jonathan Lemon <jonathan.lemon@gmail.com> Cc: Eric Dumazet <edumazet@google.com> Cc: Alexander Lobakin <alobakin@pm.me> Cc: Paolo Abeni <pabeni@redhat.com> Cc: Talal Ahmad <talalahmad@google.com> Cc: Kevin Hao <haokexin@gmail.com> Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org> Cc: Kees Cook <keescook@chromium.org> Cc: Kumar Kartikeya Dwivedi <memxor@gmail.com> Cc: Antoine Tenart <atenart@kernel.org> Cc: Wei Wang <weiwan@google.com> Cc: Arnd Bergmann <arnd@arndb.de> Suggested-by:Eric Dumazet <edumazet@google.com> Signed-off-by:
-
- 18 Apr, 2022 2 commits
-
-
Luiz Angelo Daros de Luca authored
DSA tags before IP header (categories 1 and 2) or after the payload (3) might introduce offload checksum issues. Signed-off-by:
-
David S. Miller authored
Ido Schimmel says: ==================== mlxsw: Introduce line card support for modular switch Jiri says: This patchset introduces support for modular switch systems and also introduces mlxsw support for NVIDIA Mellanox SN4800 modular switch. It contains 8 slots to accommodate line cards - replaceable PHY modules which may contain gearboxes. Currently supported line card: 16X 100GbE (QSFP28) Other line cards that are going to be supported: 8X 200GbE (QSFP56) 4X 400GbE (QSFP-DD) There may be other types of line cards added in the future. To be consistent with the port split configuration (splitter cabels), the line card entities are treated in the similar way. The nature of a line card is not "a pluggable device", but "a pluggable PHY module". A concept of "provisioning" is introduced. The user may "provision" certain slot with a line card type. Driver then creates all instances (devlink ports, netdevices, etc) related to this line card type. It does not matter if the line card is plugged-in at the time. User is able to configure netdevices, devlink ports, setup port splitters, etc. From the perspective of the switch ASIC, all is present and can be configured. The carrier of netdevices stays down if the line card is not plugged-in. Once the line card is inserted and activated, the carrier of the related netdevices is then reflecting the physical line state, same as for an ordinary fixed port. Once user does not want to use the line card related instances anymore, he can "unprovision" the slot. Driver then removes the instances. Patches 1-4 are extending devlink driver API and UAPI in order to register, show, dump, provision and activate the line card. Patches 5-17 are implementing the introduced API in mlxsw. The last patch adds a selftest for mlxsw line cards. Example: $ devlink port # No ports are listed $ devlink lc pci/0000:01:00.0: lc 1 state unprovisioned supported_types: 16x100G lc 2 state unprovisioned supported_types: 16x100G lc 3 state unprovisioned supported_types: 16x100G lc 4 state unprovisioned supported_types: 16x100G lc 5 state unprovisioned supported_types: 16x100G lc 6 state unprovisioned supported_types: 16x100G lc 7 state unprovisioned supported_types: 16x100G lc 8 state unprovisioned supported_types: 16x100G Note that driver exposes list supported line card types. Currently there is only one: "16x100G". To provision the slot #8: $ devlink lc set pci/0000:01:00.0 lc 8 type 16x100G $ devlink lc show pci/0000:01:00.0 lc 8 pci/0000:01:00.0: lc 8 state active type 16x100G supported_types: 16x100G $ devlink port pci/0000:01:00.0/0: type notset flavour cpu port 0 splittable false pci/0000:01:00.0/53: type eth netdev enp1s0nl8p1 flavour physical lc 8 port 1 splittable true lanes 4 pci/0000:01:00.0/54: type eth netdev enp1s0nl8p2 flavour physical lc 8 port 2 splittable true lanes 4 pci/0000:01:00.0/55: type eth netdev enp1s0nl8p3 flavour physical lc 8 port 3 splittable true lanes 4 pci/0000:01:00.0/56: type eth netdev enp1s0nl8p4 flavour physical lc 8 port 4 splittable true lanes 4 pci/0000:01:00.0/57: type eth netdev enp1s0nl8p5 flavour physical lc 8 port 5 splittable true lanes 4 pci/0000:01:00.0/58: type eth netdev enp1s0nl8p6 flavour physical lc 8 port 6 splittable true lanes 4 pci/0000:01:00.0/59: type eth netdev enp1s0nl8p7 flavour physical lc 8 port 7 splittable true lanes 4 pci/0000:01:00.0/60: type eth netdev enp1s0nl8p8 flavour physical lc 8 port 8 splittable true lanes 4 pci/0000:01:00.0/61: type eth netdev enp1s0nl8p9 flavour physical lc 8 port 9 splittable true lanes 4 pci/0000:01:00.0/62: type eth netdev enp1s0nl8p10 flavour physical lc 8 port 10 splittable true lanes 4 pci/0000:01:00.0/63: type eth netdev enp1s0nl8p11 flavour physical lc 8 port 11 splittable true lanes 4 pci/0000:01:00.0/64: type eth netdev enp1s0nl8p12 flavour physical lc 8 port 12 splittable true lanes 4 pci/0000:01:00.0/125: type eth netdev enp1s0nl8p13 flavour physical lc 8 port 13 splittable true lanes 4 pci/0000:01:00.0/126: type eth netdev enp1s0nl8p14 flavour physical lc 8 port 14 splittable true lanes 4 pci/0000:01:00.0/127: type eth netdev enp1s0nl8p15 flavour physical lc 8 port 15 splittable true lanes 4 pci/0000:01:00.0/128: type eth netdev enp1s0nl8p16 flavour physical lc 8 port 16 splittable true lanes 4 To uprovision the slot #8: $ devlink lc set pci/0000:01:00.0 lc 8 notype ==================== Signed-off-by:
-
