- 20 Nov, 2018 40 commits
-
-
Eric Dumazet authored
There are two cases were we can avoid calling ktime_get_ns() : 1) Queue is empty. 2) Internal queue is not empty. Signed-off-by:
Eric Dumazet <edumazet@google.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Jakub Kicinski says: ==================== gred: add offload support This series adds support for GRED offload in the nfp driver. So far we have only supported the RED Qdisc offload, but we need a way to differentiate traffic types e.g. based on DSCP marking. It may seem like PRIO+RED is a good match for this job, however, (a) we don't need strict priority behaviour of PRIO, and (b) PRIO uses the legacy way of mapping ToS fields to bands, which is quite awkward and limitting. The less commonly used GRED Qdisc is a better much for the scenario, it allows multiple sets of RED parameters and queue lengths to be maintained with a single FIFO queue. This is exactly how nfp offload behaves. We use a trivial u32 classifier to assign packets to virtual queues. There is also the minor advantage that GRED can't have its child changed, therefore limitting ways in which the configuration of SW path can diverge from HW offload. Last patch of the series adds support for (G)RED in non-ECN mode, where packets are dropped instead of marked. ==================== Signed-off-by: -
Jakub Kicinski authored
Original FW only allowed us to perform ECN marking. Newer releases also support plain old drop. Add the ability to configure drop policy. This is particularly useful in combination with GRED, because different bands can have different ECN marking setting. Signed-off-by:
Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by:
John Hurley <john.hurley@netronome.com> Signed-off-by:
-
Jakub Kicinski authored
Use offload of very simple u32 filters to direct packets to GRED bands based on the DSCP marking. No u32 hashing is supported, just plain simple filters matching on ToS or Priority with appropriate mask device can support. Signed-off-by:
-
Jakub Kicinski authored
Learn how to set the DSCP map. FW uses a packed array which geometry depends on the number of supported priorities and virtual queues. Write code to assemble this map and to communicate the setting to the FW via mailbox. Signed-off-by:
-
Jakub Kicinski authored
In preparation for PRIO offload calculate how long the prio map for FW will be and make sure the configuration can be performed via the vNIC mailbox. Signed-off-by:
-
Jakub Kicinski authored
In case of egress offloads the class/flowid assigned by the filter may be very important for offloaded Qdisc selection. Provide this info to drivers. Signed-off-by:
-
Jakub Kicinski authored
Add support for GRED offload. It behaves much like RED, but can apply different parameters to different bands. GRED operates pretty much exactly like our HW/FW with a single FIFO and different RED state instances. Signed-off-by:
-
Jakub Kicinski authored
Wrap RED parameters and stats into a structure, and a 1-element array. Upcoming GRED offload will add the support for more bands. Signed-off-by:
-
Jakub Kicinski authored
Allow drivers which offload GRED to report back statistics. Since A lot of GRED stats is fairly ad hoc in nature pass to drivers the standard struct gnet_stats_basic/gnet_stats_queue pairs, and untangle the values in the core. Signed-off-by:
-
Jakub Kicinski authored
Add basic offload for the GRED Qdisc. Inform the drivers any time Qdisc or virtual queue configuration changes. Signed-off-by:
-
Jakub Kicinski authored
Add up stats for all bands for the extra ethtool statistics. Signed-off-by:
-
Jakub Kicinski authored
In PRIO-enabled FW read the statistics from per-band symbol, rather than from the standard per-PCIe-queue counters. Signed-off-by:
-
Jakub Kicinski authored
Make sure the threshold table is large enough to hold information for all bands. Signed-off-by:
-
Jakub Kicinski authored
In preparation for per-band RED offload pass band parameter to functions. For now it will always be 0. Signed-off-by:
-
Jakub Kicinski authored
In preparation for multi-band RED offload if FW is capable map the extended symbols which will allow us to set per-band parameters and read stats. Signed-off-by:
-
Yunsheng Lin authored
Before setting tm related configuration to hardware, driver needs to check the configuration provided by user is valid. Currently hclge_ieee_setets and hclge_setup_tc both implement their own checking, which has a lot in common. This patch addes hclge_dcb_common_validate to do the common checking. The checking in hclge_tm_prio_tc_info_update and hclge_tm_schd_info_update is unnecessary now, so change the return type to void, which removes the need to do error handling when one of the checking fails. Also, ets->prio_tc is indexed by user prio and ets->tc_tsa is indexed by tc num, so this patch changes them to use different index. Signed-off-by:
Yunsheng Lin <linyunsheng@huawei.com> Signed-off-by:
Tan Xiaojun <tanxiaojun@huawei.com> Signed-off-by:
-
David S. Miller authored
Ido Schimmel says: ==================== selftests: Add tests for VXLAN at an 802.1d bridge Petr says: This patchset adds several tests for VXLAN attached to an 802.1d bridge and fixes a related bug. First patch #1 fixes a bug in propagating SKB already-forwarded marks over veth to bridges, where they are irrelevant. This bug causes the vxlan_bridge_1d test suite from this patchset to fail as the packets aren't forwarded by br2. In patches #2 and #3, lib.sh is extended to support network namespaces. The use of namespaces is necessitated by VXLAN, which allows only one VXLAN device with a given VNI per namespace. Thus to host full topology on a single box for selftests, the "remote" endpoints need to be in namespaces. In patches #4-#6, lib.sh is extended in other ways to facilitate the following patches. In patches #7-#15, first the skeleton, and later the generic tests themselves are added. Patch #16 then adds another test that serves as a wrapper around the previous one, and runs it with a non-default port number. Patches #17 and #18 add mlxsw-specific tests. About those, Ido writes: The first test creates various configurations with regards to the VxLAN and bridge devices and makes sure the driver correctly forbids unsupported configuration and permits supported ones. It also verifies that the driver correctly sets the offload indication on FDB entries and the local route used for VxLAN decapsulation. The second test verifies that the driver correctly configures the singly linked list used to flood BUM traffic and that traffic is flooded as expected. ==================== Signed-off-by: -
Ido Schimmel authored
The device stores flood records in a singly linked list where each record stores up to three IPv4 addresses of remote VTEPs. The test verifies that packets are correctly flooded in various cases such as deletion of a record in the middle of the list. Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Reviewed-by:
Petr Machata <petrm@mellanox.com> Signed-off-by:
-
Ido Schimmel authored
Test various aspects of VxLAN offloading which are specific to mlxsw, such as sanitization of invalid configurations and offload indication. Signed-off-by:
-
Petr Machata authored
This simple wrapper reruns the VXLAN ping test with a port number of 8472. Signed-off-by:
-
Petr Machata authored
Test that when decapsulating from VXLAN, the values of inner and outer TOS are handled appropriately. Because VXLAN driver on its own won't produce the arbitrary TOS combinations necessary to test this feature, simply open-code a single ICMP packet and have mausezahn assemble it. Signed-off-by:
-
Petr Machata authored
Test that ECN bits in the VXLAN envelope are correctly deduced from the overlay packet. Signed-off-by:
-
Petr Machata authored
Test that TOS is inherited from the tunneled packet into the envelope as configured at the VXLAN device. Signed-off-by:
-
Petr Machata authored
This tests whether TTL of VXLAN envelope packets is properly set based on the device configuration. Signed-off-by:
-
Petr Machata authored
The ordering of the topology creation can have impact on whether a driver is successful in offloading VXLAN. Therefore add a pseudo-test that reshuffles bits of the topology, and then reruns the same suite of tests again to make sure that the new setup is supported as well. Signed-off-by:
-
Petr Machata authored
Test that when sending traffic to a learned MAC address, the traffic is forwarded accurately only to the right endpoint. Signed-off-by:
-
Petr Machata authored
Test that when sending traffic to an unlearned MAC address, the traffic is flooded to both remote VXLAN endpoints. Signed-off-by:
-
Petr Machata authored
Test end-to-end reachability between local and remote endpoints. Note that because learning is disabled on the VXLAN device, the ICMP requests will end up being flooded to all remotes. Signed-off-by:
-
Petr Machata authored
This skeleton sets up a topology with three VXLAN endpoints: one "local", possibly offloaded, and two "remote", formed using veth pairs and likely purely software bridges. The "local" endpoint is connected to host systems by a VLAN-unaware bridge. Since VXLAN tunnels must be unique per namespace, each of the "remote" endpoints is in its own namespace. H3 forms the bridge between the three domains. Signed-off-by:
-
Petr Machata authored
Such a function will be useful for counting malformed packets in the ECN decap test. To that end, introduce a common handler for handling stat-fetching, and reuse it in link_stats_tx_packets_get() and link_stats_rx_errors_get(). Signed-off-by:
-
Petr Machata authored
Make the ping routine more generic by allowing passing arbitrary ping command-line arguments. Signed-off-by:
-
Petr Machata authored
Have ping_test() recognize an optional argument with a description of the test. This is handy if there are several ping test, to make it clear which is which. Signed-off-by:
-
Petr Machata authored
In order to run a certain command inside another network namespace, it's possible to use "ip netns exec ns command". However then one can't use functions defined in lib.sh or a test suite. One option is to do "ip netns exec ns bash -c command", provided that all functions that one wishes to use (and their dependencies) are published using "export -f". That may not be practical. Therefore, introduce a helper in_ns(), which wraps a given command in a boilerplate of "ip netns exec" and "source lib.sh", thus making all library functions available. (Custom functions that a script wishes to run within a namespace still need to be exported.) Because quotes in "$@" aren't recognized in heredoc, hand-expand the array in an explicit for loop, leveraging printf %q to handle proper quoting. Signed-off-by:
-
Petr Machata authored
So far the case of NUM_NETIFS of 0 has not been interesting. However if one wishes to reuse the lib.sh routines in a setup of a separate namespace, being able to import like this is handy. Therefore replace the {1..$NUM_NETIFS} references, which cause iteration over 1 and 0, with an explicit for loop like we do in setup_wait() and tc_offload_check(), so that for NUM_NETIFS of 0 no iteration is done. Signed-off-by: -
Petr Machata authored
When a packet is trapped and the corresponding SKB marked as already-forwarded, it retains this marking even after it is forwarded across veth links into another bridge. There, since it ingresses the bridge over veth, which doesn't have offload_fwd_mark, it triggers a warning in nbp_switchdev_frame_mark(). Then nbp_switchdev_allowed_egress() decides not to allow egress from this bridge through another veth, because the SKB is already marked, and the mark (of 0) of course matches. Thus the packet is incorrectly blocked. Solve by resetting offload_fwd_mark() in skb_scrub_packet(). That function is called from tunnels and also from veth, and thus catches the cases where traffic is forwarded between bridges and transformed in a way that invalidates the marking. Signed-off-by:
-
David S. Miller authored
Sunil Goutham says: ==================== octeontx2-af: NPC MCAM support and FLR handling This patchset is a continuation to earlier submitted three patch series to add a new driver for Marvell's OcteonTX2 SOC's Resource virtualization unit (RVU) admin function driver. 1. octeontx2-af: Add RVU Admin Function driver https://www.spinics.net/lists/netdev/msg528272.html 2. octeontx2-af: NPA and NIX blocks initialization https://www.spinics.net/lists/netdev/msg529163.html 3. octeontx2-af: NPC parser and NIX blocks initialization https://www.spinics.net/lists/netdev/msg530252.html This patch series adds support for below RVU generic: - Function Level Reset irq handler When FLR is triggered for PFs, AF receives interrupt. This patchset adds logic for cleaning up of NPA, NIX and NPC block resources being used by PF. - Mailbox communication between AF and it's VFs. Unlike VFs of PF1-PFn, AF which is PF0 can communicate with it's VFs directly. Added support for the same. - AF's VFs IO configuration These VFs are mapped to use internal HW loopback channels instead of CGX LMACs. Each pair of VFs work as two of ends of hardwired interfaces. VF0's TX is VF1's Rx & viceversa. NPC block: - MCAM entry management Alloc/Free of contiguous/non-contiguous and lower/higher priority MCAM entry allocation and programming support. - MCAM counters management and map/unmap with MCAM entries - Default KEY extract profile - HW errata workarounds NIX block: - Minimum and maximum allowed packet length config - HW errata workarounds Few more changes like shift to use mutex instead of spinlock etc are done in this patchset. Changes from v2: 1 Fixed commit message of patch 'Relax resource lock into mutex' to a more unambiguous one. - Suggested by David Miller. Changes from v1: 1 Converted all mailbox message handler API names to small letters from mixed small and capital letters. - Suggested by David Miller. 2 Fixed endian issues in patch 'Add support for stripping STAG/CTAG' - Suggested by Arnd Bergmann 3 Elaborated commit message of patch 'Add FLR interrupt handler' to make it a bit more easy to understand. - Suggested by Arnd Bergmann Will fix the padding and alignment in mailbox message structure in a follow-up patch. ==================== Signed-off-by:
-
Sunil Goutham authored
Errata 35038 Software sets NIX_AF_RX_SW_SYNC[ENA] to sync (flush) in-flight packets the RX data path before configuration changes (e.g. disabling one or more RQs). Hardware clears [ENA] to indicate sync is done An issue exists whereby NIX may clear NIX_AF_RX_SW_SYNC [ENA] too early. Errata 35057 NIX may corrupt internal state when conditional clocks turn off. So turnon all clocks by default. Errata 35786 Parse nibble enable NPC configuration for KEY generation has to be identical for both Rx and Tx interfaces. Also corrected endianness configuration for NIX i.e NIX_AF_CFG[AF_BE] is bit8 and not bit1. Signed-off-by:
Sunil Goutham <sgoutham@marvell.com> Signed-off-by:
Jerin Jacob <jerinj@marvell.com> Signed-off-by:
-
Linu Cherian authored
- Add interrupt handlers for Master Enable events from PFs and Master Enable events from VFs of AF - Master Enable is required for the MSIX delivery to work - Master Enable bit trap handler doesn't have to do any anything other than clearing the TRPEND bit, since the enable/disable requirements are already taken care using mbox requests/flr handler. Signed-off-by:
Linu Cherian <lcherian@marvell.com> Signed-off-by:
-
Sunil Goutham authored
Added support to handle FLR for AF's VFs (i.e LBK VFs). Just the FLR interrupt enable/disable, handler registration etc, actual HW resource cleanup or LFs teardown logic is already there. Signed-off-by:
-
