- 22 Jun, 2023 13 commits
-
-
Jacob Keller authored
The ice_sq_send_cmd() function is used to send messages to the control queues used to communicate with firmware, virtual functions, and even some hardware. When sending a control queue message, the driver is designed to synchronously wait for a response from the queue. Currently it waits between checks for 100 to 150 microseconds. Commit f86d6f9c ("ice: sleep, don't busy-wait, for ICE_CTL_Q_SQ_CMD_TIMEOUT") did recently change the behavior from an unnecessary delay into a sleep which is a significant improvement over the old behavior of polling using udelay. Because of the nature of PCIe transactions, the hardware won't be informed about a new message until the write to the tail register posts. This is only guaranteed to occur at the next register read. In ice_sq_send_cmd(), this happens at the ice_sq_done() call. Because of this, the driver essentially forces a minimum of one full wait time regardless of how fast the response is. For the hardware-based sideband queue, this is especially slow. It is expected that the hardware will respond within 2 or 3 microseconds, an order of magnitude faster than the 100-150 microsecond sleep. Allow such fast completions to occur without delay by introducing a small 5 microsecond delay first before entering the sleeping timeout loop. Ensure the tail write has been posted by using ice_flush(hw) first. While at it, lets also remove the ICE_CTL_Q_SQ_CMD_USEC macro as it obscures the sleep time in the inner loop. It was likely introduced to avoid "magic numbers", but in practice sleep and delay values are easier to read and understand when using actual numbers instead of a named constant. This change should allow the fast hardware based control queue messages to complete quickly without delay, while slower firmware queue response times will sleep while waiting for the response. Signed-off-by:
Jacob Keller <jacob.e.keller@intel.com> Reviewed-by:
Michal Schmidt <mschmidt@redhat.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>
-
Jakub Kicinski authored
Matthieu Baerts says: ==================== mptcp: expose more info and small improvements Patch 1-3/9 track and expose some aggregated data counters at the MPTCP level: the number of retransmissions and the bytes that have been transferred. The first patch prepares the work by moving where snd_una is updated for fallback sockets while the last patch adds some tests to cover the new code. Patch 4-6/9 introduce a new getsockopt for SOL_MPTCP: MPTCP_FULL_INFO. This new socket option allows to combine info from MPTCP_INFO, MPTCP_TCPINFO and MPTCP_SUBFLOW_ADDRS socket options into one. It can be needed to have all info in one because the path-manager can close and re-create subflows between getsockopt() and fooling the accounting. The first patch introduces a unique subflow ID to easily detect when subflows are being re-created with the same 5-tuple while the last patch adds some tests to cover the new code. Please note that patch 5/9 ("mptcp: introduce MPTCP_FULL_INFO getsockopt") can reveal a bug that were there for a bit of time, see [1]. A fix has recently been fixed to netdev for the -net tree: "mptcp: ensure listener is unhashed before updating the sk status", see [2]. There is no conflicts between the two patches but it might be better to apply this series after the one for -net and after having merged "net" into "net-next". Patch 7/9 is similar to commit 47867f0a ("selftests: mptcp: join: skip check if MIB counter not supported") recently applied in the -net tree but here it adapts the new code that is only in net-next (and it fixes a merge conflict resolution which didn't have any impact). Patch 8 and 9/9 are two simple refactoring. One to consolidate the transition to TCP_CLOSE in mptcp_do_fastclose() and avoid duplicated code. The other one reduces the scope of an argument passed to mptcp_pm_alloc_anno_list() function. Link: https://github.com/multipath-tcp/mptcp_net-next/issues/407 [1] Link: https://lore.kernel.org/netdev/20230620-upstream-net-20230620-misc-fixes-for-v6-4-v1-0-f36aa5eae8b9@tessares.net/ [2] ==================== Link: https://lore.kernel.org/r/20230620-upstream-net-next-20230620-mptcp-expose-more-info-and-misc-v1-0-62b9444bfd48@tessares.netSigned-off-by:Jakub Kicinski <kuba@kernel.org>
-
Geliang Tang authored
Pass addr parameter to mptcp_pm_alloc_anno_list() instead of entry. We can reduce the scope, e.g. in mptcp_pm_alloc_anno_list(), we only access "entry->addr", we can then restrict to the pointer to "addr" then. Signed-off-by:
Geliang Tang <geliang.tang@suse.com> Reviewed-by:
Matthieu Baerts <matthieu.baerts@tessares.net> Signed-off-by:
-
Paolo Abeni authored
The MPTCP code always set the msk state to TCP_CLOSE before calling performing the fast-close. Move such state transition in mptcp_do_fastclose() to avoid some code duplication. Signed-off-by:
Paolo Abeni <pabeni@redhat.com> Reviewed-by:
-
Matthieu Baerts authored
Selftests are supposed to run on any kernels, including the old ones not supporting all MPTCP features. One of them is the MPTCP MIB counters introduced in commit fc518953 ("mptcp: add and use MIB counter infrastructure") and more later. The MPTCP Join selftest heavily relies on these counters. If a counter is not supported by the kernel, it is not displayed when using 'nstat -z'. We can then detect that and skip the verification. A new helper (get_counter()) has been added recently in the -net tree to do the required checks and return an error if the counter is not available. This commit is similar to the one with the same title applied in the -net tree but it modifies code only present in net-next for the moment, see the Fixes commit below. While at it, we can also remove the use of ${extra_msg} variable which is never assigned in chk_rm_tx_nr() function and use 'echo' without '-n' parameter. Link: https://github.com/multipath-tcp/mptcp_net-next/issues/368 Fixes: 0639fa23 ("selftests: mptcp: add explicit check for new mibs") Signed-off-by:
-
Paolo Abeni authored
Add a testcase explicitly triggering the newly introduce MPTCP_FULL_INFO getsockopt. Link: https://github.com/multipath-tcp/mptcp_net-next/issues/388Signed-off-by:
-
Paolo Abeni authored
Some user-space applications want to monitor the subflows utilization. Dumping the per subflow tcp_info is not enough, as the PM could close and re-create the subflows under-the-hood, fooling the accounting. Even checking the src/dst addresses used by each subflow could not be enough, because new subflows could re-use the same address/port of the just closed one. This patch introduces a new socket option, allow dumping all the relevant information all-at-once (everything, everywhere...), in a consistent manner. Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/388Signed-off-by:
-
Paolo Abeni authored
The user-space need to properly account the data received/sent by individual subflows. When additional subflows are created and/or closed during the MPTCP socket lifetime, the information currently exposed via MPTCP_TCPINFO are not enough: subflows are identified only by the sequential position inside the info dumps, and that will change with the above mentioned events. To solve the above problem, this patch introduces a new subflow identifier that is unique inside the given MPTCP socket scope. The initial subflow get the id 1 and the other subflows get incremental values at join time. Link: https://github.com/multipath-tcp/mptcp_net-next/issues/388Signed-off-by:
-
Paolo Abeni authored
Update the existing sockopt test-case to do some basic checks on the newly added counters. Link: https://github.com/multipath-tcp/mptcp_net-next/issues/385Signed-off-by:
-
Paolo Abeni authored
Currently there are no data transfer counters accounting for all the subflows used by a given MPTCP socket. The user-space can compute such figures aggregating the subflow info, but that is inaccurate if any subflow is closed before the MPTCP socket itself. Add the new counters in the MPTCP socket itself and expose them via the existing diag and sockopt. While touching mptcp_diag_fill_info(), acquire the relevant locks before fetching the msk data, to ensure better data consistency Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/385Signed-off-by:
-
Paolo Abeni authored
That will avoid an unneeded conditional in both the fast-path and in the fallback case and will simplify a bit the next patch. Signed-off-by:
-
David Arinzon authored
This patch fixes a warning in the ena documentation file identified by the kernel automatic tools. The patch also adds a missing newline between sections. Signed-off-by:
David Arinzon <darinzon@amazon.com> Reported-by:
kernel test robot <lkp@intel.com> Closes: https://lore.kernel.org/oe-kbuild-all/202306171804.U7E92zoE-lkp@intel.com/Reviewed-by:
Simon Horman <simon.horman@corigine.com> Link: https://lore.kernel.org/r/20230620133544.32584-1-darinzon@amazon.comSigned-off-by:
-
Abel Wu authored
If there is no net-memcg associated with the sock, don't bother calculating its memory usage for charge. Signed-off-by:
Abel Wu <wuyun.abel@bytedance.com> Link: https://lore.kernel.org/r/20230620092712.16217-1-wuyun.abel@bytedance.comSigned-off-by:
-
- 21 Jun, 2023 27 commits
-
-
renmingshuai authored
Add the test for additional reference to chains that are explicitly created by RTM_NEWCHAIN message. The test result: 1..1 ok 1 c2b4 - soft lockup alarm will be not generated after delete the prio 0 filter of the chain This is a follow up to commit c9a82bec ("net/sched: cls_api: Fix lockup on flushing explicitly created chain"). Signed-off-by:
Mingshuai Ren <renmingshuai@huawei.com> Acked-by:
Pedro Tammela <pctammela@mojatatu.com> Acked-by:
Victor Nogueira <victor@mojatatu.com> Link: https://lore.kernel.org/r/20230620014939.2034054-1-renmingshuai@huawei.comSigned-off-by:
-
Krzysztof Kozlowski authored
The Micrel KS8851 can be attached to SPI or parallel bus and the difference is expressed in compatibles. Allow common SPI properties when this is a SPI variant and narrow the parallel memory bus properties to the second case. This fixes dtbs_check warning: qcom-msm8960-cdp.dtb: ethernet@0: Unevaluated properties are not allowed ('spi-max-frequency' was unexpected) Signed-off-by:Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Reviewed-by:
Conor Dooley <conor.dooley@microchip.com> Link: https://lore.kernel.org/r/20230619170134.65395-1-krzysztof.kozlowski@linaro.orgSigned-off-by:
-
Krzysztof Kozlowski authored
WCN3990 comes with two chains - CH0 and CH1 - where each takes VDD regulator. It seems VDD_CH1 is optional (Linux driver does not care about it), so document it to fix dtbs_check warnings like: sdm850-lenovo-yoga-c630.dtb: bluetooth: 'vddch1-supply' does not match any of the regexes: 'pinctrl-[0-9]+' Signed-off-by:
-
Ravi Gunasekaran authored
When port-to-port forwarding for interfaces in HSR node is enabled, disable promiscuous mode since L2 frame forward happens at the offloaded hardware. Signed-off-by:
Ravi Gunasekaran <r-gunasekaran@ti.com> Reviewed-by:
-
Jakub Kicinski authored
Christian Marangi says: ==================== leds: trigger: netdev: add additional modes This is a continue of [1]. It was decided to take a more gradual approach to implement LEDs support for switch and phy starting with basic support and then implementing the hw control part when we have all the prereq done. This should be the final part for the netdev trigger. I added net-next tag and added netdev mailing list since I was informed that this should be merged with netdev branch. We collect some info around and we found a good set of modes that are common in almost all the PHY and Switch. These modes are: - Modes for dedicated link speed(10, 100, 1000 mbps). Additional mode can be added later following this example. - Modes for half and full duplex. The original idea was to add hw control only modes. While the concept makes sense in practice it would results in lots of additional code and extra check to make sure we are setting correct modes. With the suggestion from Andrew it was pointed out that using the ethtool APIs we can actually get the current link speed and duplex and this effectively removed the problem of having hw control only modes since we can fallback to software. Since these modes are supported by software, we can skip providing an user for this in the LED driver to support hw control for these new modes (that will come right after this is merged) and prevent this to be another multi subsystem series. For link speed and duplex we use ethtool APIs. To call ethtool APIs, rtnl lock is needed but this can be skipped on handling netdev events as the lock is already held. [1] https://lore.kernel.org/lkml/20230216013230.22978-1-ansuelsmth@gmail.com/ ==================== Link: https://lore.kernel.org/r/20230619204700.6665-1-ansuelsmth@gmail.comSigned-off-by:
-
Christian Marangi authored
Expose hw_control status via sysfs for the netdev trigger to give userspace better understanding of the current state of the trigger and the LED. Signed-off-by:
Christian Marangi <ansuelsmth@gmail.com> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Reviewed-by:
Kalesh AP <kalesh-anakkur.purayil@broadcom.com> Acked-by:
Lee Jones <lee@kernel.org> Signed-off-by:
-
Christian Marangi authored
Add additional modes for specific link duplex. Use ethtool APIs to get the current link duplex and enable the LED accordingly. Under netdev event handler the rtnl lock is already held and is not needed to be set to access ethtool APIs. This is especially useful for PHY and Switch that supports LEDs hw control for specific link duplex. Add additional modes: - half_duplex: Turn on LED when link is half duplex - full_duplex: Turn on LED when link is full duplex Signed-off-by:
-
Christian Marangi authored
Add additional modes for specific link speed. Use ethtool APIs to get the current link speed and enable the LED accordingly. Under netdev event handler the rtnl lock is already held and is not needed to be set to access ethtool APIs. This is especially useful for PHY and Switch that supports LEDs hw control for specific link speed. (example scenario a PHY that have 2 LED connected one green and one orange where the green is turned on with 1000mbps speed and orange is turned on with 10mpbs speed) On mode set from sysfs we check if we have enabled split link speed mode and reject enabling generic link mode to prevent wrong and redundant configuration. Rework logic on the set baseline state to support these new modes to select if we need to turn on or off the LED. Add additional modes: - link_10: Turn on LED when link speed is 10mbps - link_100: Turn on LED when link speed is 100mbps - link_1000: Turn on LED when link speed is 1000mbps Signed-off-by:
-
Ivan Vecera authored
Link VF representors to parent PCI device to benefit from systemd defined naming scheme. Without this change the representor is visible as ethN. Signed-off-by:
Ivan Vecera <ivecera@redhat.com> Reviewed-by:
Michael Chan <michael.chan@broadcom.com> Link: https://lore.kernel.org/r/20230620144855.288443-1-ivecera@redhat.comSigned-off-by:
-
Jakub Kicinski authored
Petr Machata says: ==================== selftests: Preparations for out-of-order-operations patches in mlxsw The mlxsw driver currently makes the assumption that the user applies configuration in a bottom-up manner. Thus netdevices need to be added to the bridge before IP addresses are configured on that bridge or SVI added on top of it. Enslaving a netdevice to another netdevice that already has uppers is in fact forbidden by mlxsw for this reason. Despite this safety, it is rather easy to get into situations where the offloaded configuration is just plain wrong. Over the course of the following several patchsets, mlxsw code is going to be adjusted to diminish the space of wrongly offloaded configurations. Ideally the offload state will reflect the actual state, regardless of the sequence of operation used to construct that state. Several selftests build configurations that will not be offloadable in the future on some systems. The reason is that what will get offloaded is the actual configuration, not the configuration steps. For example, when a port is added to a bridge that has an IP address, that bridge will get a RIF, which it would not have with the current code. But on Nvidia Spectrum-1 machines, MAC addresses of all RIFs need to have the same prefix, which the bridge will violate. The RIF thus couldn't be created, and the enslavement is therefore canceled, because it would lead to an unoffloadable configuration. This breaks some selftests. In this patchset, adjust selftests to avoid the configurations that mlxsw would be incapable of offloading, while maintaining relevance with regards to the feature that is being tested. There are generally two cases of fixes: - Disabling IPv6 autogen on bridges that do not participate in routing, either because of the abovementioned requirement to keep the same MAC prefix on all in-HW router interfaces, or, on 802.1ad bridges, because in-HW router interfaces are not supported at all. - Setting the bridge MAC address to what it will become after the first member port is attached, so that the in-HW router interface is created with a supported MAC address. The patchset is then split thus: - Patches #1-#7 adjust generic selftests - Patches #8-#16 adjust mlxsw-specific selftests ==================== Link: https://lore.kernel.org/r/cover.1687265905.git.petrm@nvidia.comSigned-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The bridge eventually inherits MAC address from its first member, after the enslavement is acked. A number of (mainly VXLAN) selftests already work around the problem by setting the MAC address to whatever it will eventually be anyway. Do the same for this selftest. Signed-off-by:
Petr Machata <petrm@nvidia.com> Reviewed-by:
Danielle Ratson <danieller@nvidia.com> Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge (this holds for all bridges used here), the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks various aspects of VXLAN offloading and the bridges do not need to participate in routing traffic. The IP addresses or the RIFs are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridges in this selftest, thus exempting them from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks vetoing of a different aspect of the configuration and the bridge does not need to participate in routing traffic. The IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge (this holds for both bridges used here), the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks traffic prioritization and scheduling, and the bridges serve for their L2 forwarding capabilities, and do not need to participate in routing traffic. The IP addresses or the RIFs are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridges in this selftest, thus exempting them from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge (this holds for both bridges used here), the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks traffic prioritization and scheduling, and the bridges serve for their L2 forwarding capabilities, and do not need to participate in routing traffic. The IP addresses or the RIFs are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridges in this selftest, thus exempting them from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks DCB DSCP-based prioritization, and the bridge serves for its L2 forwarding capabilities, and does not need to participate in routing traffic. The IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks how many mirroring sessions a machine is capable of offloading. The IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge (this holds for all bridges used here), the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks whether a different vetoed aspect of the configuration provides an extack. The IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridges in this selftest, thus exempting them from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. The swp enslavement to the 802.1ad bridge is not allowed, because RIFs are not allowed to be created for 802.1ad bridges, but the address indicates one needs to be created. Thus the veto selftests fail already during the port enslavement. Then the attempt to create a VLAN on top of the same bridge is not vetoed, because the bridge is not related to mlxsw, and the selftest fails. Fix by disabling automatic IPv6 address generation for the bridges in this selftest, thus exempting them from the mlxsw router attention. Signed-off-by:
Amit Cohen <amcohen@nvidia.com> Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The bridge eventually inherits MAC address from its first member, after the enslavement is acked. A number of (mainly VXLAN) selftests already work around the problem by setting the MAC address to whatever it will eventually be anyway. Do the same here. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The bridge eventually inherits MAC address from its first member, after the enslavement is acked. A number of (mainly VXLAN) selftests already work around the problem by setting the MAC address to whatever it will eventually be anyway. Do the same for several mirror_gre selftests. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. These two selftests however check mirroring traffic to a gretap netdevice. The bridge here does not participate in routing traffic and the IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridges in these selftests, thus exempting them from mlxsw router attention. Since the bridges are only used for L2 forwarding, this change should not hinder usefulness of this selftest for testing SW datapath or HW datapaths in other devices. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks whether skbedit changes packet priority as appropriate. The bridge thus does not need to participate in routing traffic and the IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Since the bridge is only used for L2 forwarding, this change should not hinder usefulness of this selftest for testing SW datapath or HW datapaths in other devices. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. At the time that the front panel port is enslaved to the bridge, the bridge MAC address does not have the same prefix as other interfaces in the system. On Nvidia Spectrum-1 machines all the RIFs have to have the same 38-bit MAC address prefix. Since the bridge does not obey this limitation, the RIF cannot be created, and the enslavement attempt is vetoed on the grounds of the configuration not being offloadable. The selftest itself however checks operation of pedit on IPv4 and IPv6 dsfield and its parts. The bridge thus does not need to participate in routing traffic and the IP address or the RIF are irrelevant. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Since the bridge is only used for L2 forwarding, this change should not hinder usefulness of this selftest for testing SW datapath or HW datapaths in other devices. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. This will cause this selftest to fail spuriously. The swp enslavement to the 802.1ad bridge is not allowed, because RIFs are not allowed to be created for 802.1ad bridges, but the address indicates one needs to be created. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Signed-off-by:
-
Petr Machata authored
In a future patch, mlxsw will start adding RIFs to uppers of front panel port netdevices, if they have an IP address. This will cause this selftest to fail spuriously. The swp enslavement to the 802.1ad bridge is not allowed, because RIFs are not allowed to be created for 802.1ad bridges, but the address indicates one needs to be created. Fix by disabling automatic IPv6 address generation for the HW-offloaded bridge in this selftest, thus exempting it from mlxsw router attention. Signed-off-by:
-
Horatiu Vultur authored
Currently for each timestamp frame, the SW needs to go and read the received timestamp over the MDIO bus. But the HW has the capability to store the received nanoseconds part and the least significant two bits of the seconds in the reserved field of the PTP header. In this way we could save few MDIO transactions (actually a little more transactions because the access to the PTP registers are indirect) for each received frame. Instead of reading the rest of seconds part of the timestamp of the frame using MDIO transactions schedule PTP worker thread to read the seconds part every 500ms and then for each of the received frames use this information. Because if for example running with 512 frames per second, there is no point to read 512 times the second part. Doing all these changes will give a great CPU usage performance. Running ptp4l with logSyncInterval of -9 will give a ~60% CPU improvement. Signed-off-by:
Horatiu Vultur <horatiu.vultur@microchip.com> Acked-by:
Richard Cochran <richardcochran@gmail.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
