- 05 Jul, 2016 40 commits
-
-
Jiri Pirko authored
Override the defaults and define the area sizes ourselves. Signed-off-by:
Jiri Pirko <jiri@mellanox.com> Reviewed-by:
Ido Schimmel <idosch@mellanox.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Jiri Pirko authored
Up until now we only used hash-based tables in the device, but we are going to use the linear table for remote routes adjacency lists. Add the configuration fields that control the size of the linear table. Signed-off-by:
-
Yotam Gigi authored
Listen to any NEIGH_UPDATE events sent and program the device accordingly. If NUD state is VALID and neighbour isn't yet offloaded, then program it into the device's table. Otherwise, just edit its parameters. If NUD state machine transitioned neighbour out of VALID state and it's present in the device's table, then remove it. Note that the device is programmed in delayed work, as the netevent notification chain is atomic and prevents us from going to sleep. Signed-off-by:
Yotam Gigi <yotamg@mellanox.com> Signed-off-by:
-
Yotam Gigi authored
As previously explained, the driver should periodically poll the device for neighbours activity according to the configured DELAY_PROBE_TIME. This will prevent active neighbours from staying in STALE state for long periods of time. During init configure the polling interval according to the DELAY_PROBE_TIME used in the default table. In addition, register a netevent notification block, so that the interval is updated whenever DELAY_PROBE_TIME changes. Using the computed interval schedule a delayed work, which will update the kernel via neigh_event_send() on any active neighbour since the last delayed work. Signed-off-by:
-
Ido Schimmel authored
When the data plane is offloaded the traffic doesn't go through the networking stack. Therefore, after first resolving a neighbour the NUD state machine will transition it from REACHABLE to STALE until it's finally deleted by the garbage collector. To prevent such situations the offloading driver should notify the NUD state machine on any neighbours that were recently used. The driver's polling interval should be set so that the NUD state machine can function as if the traffic wasn't offloaded. Currently, there are no in-tree drivers that can report confirmation for a neighbour, but only 'used' indication. Therefore, the polling interval should be set according to DELAY_FIRST_PROBE_TIME, as a neighbour will transition from REACHABLE state to DELAY (instead of STALE) if "a packet was sent within the last DELAY_FIRST_PROBE_TIME seconds" (RFC 4861). Send a netevent whenever the DELAY_FIRST_PROBE_TIME changes - either via netlink or sysctl - so that offloading drivers can correctly set their polling interval. Signed-off-by:
-
Yotam Gigi authored
The RAUHTD register allows dumping entries from the Router Unicast Host Table. Signed-off-by:
-
Yotam Gigi authored
The RAUHT register is used to configure and query the Unicast Host Table in devices that implement the Algorithmic LPM. In other words, it is used to configure neighbour entries in the device. Signed-off-by:
-
Jiri Pirko authored
We need to hold some private data for every neigh entry. It would be possible to do it using neigh_priv_len/ndo_neigh_construct/ ndo_neigh_destroy however only for the port device itself. That would not work for stacked devices like bridge/team/bond. So introduce a private neigh table. Hook onto ndos neigh_construct/destroy and add/remove table entry according to that. Signed-off-by:
-
Jiri Pirko authored
L2 upper device needs to propagate neigh_construct/destroy calls down to lower devices. Do this by defining default ndo functions and use them in team, bond, bridge and vlan. Signed-off-by:
-
Jiri Pirko authored
As the following patch will allow upper devices to follow the call down lower devices, we need to add dev here and not rely on n->dev. Signed-off-by:
-
David S. Miller authored
Florian Fainelli says: ==================== net: r6040: Misc updates Here are some various updates for the r6040 driver, mostly to make it more modern and catch up with the latest API improvements. ==================== Signed-off-by: -
Florian Fainelli authored
Bump version to 0.28 and date to 4th of July 2016. Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
Florian Fainelli authored
Update my email address in the driver and MAINTAINERS file. Signed-off-by:
-
Florian Fainelli authored
We maintain how much work we did in NAPI context, so provide that with napi_complete_done(). Signed-off-by:
-
Florian Fainelli authored
We are already in hard IRQ context, so we can use __napi_schedule_irqoff() to save a few operations. Signed-off-by:
-
Florian Fainelli authored
Kick the transmission only if this is the last SKB to transmit or the queue is not already stopped. Signed-off-by:
-
Florian Fainelli authored
Instead of taking one interrupt per packet transmitted, re-use the same NAPI context to free transmitted buffers. Since we are no longer in hard IRQ context replace dev_kfree_skb_irq() by dev_kfree_skb(). Signed-off-by:
-
Florian Fainelli authored
Pad the SKB to the minimum length of ETH_ZLEN by using skb_put_padto() and take this operation out of the critical section since there is no need to check any HW resources before doing that. Signed-off-by:
-
Florian Fainelli authored
r6040_xmit() is increasing transmit statistics during transmission while this may still fail, do this in r6040_tx() where we complete transmitted buffers instead. Signed-off-by:
-
Florian Fainelli authored
Instead of open coding our own version utilize the library provided function. Signed-off-by:
-
Wei Yongjun authored
Remove including <linux/version.h> that don't need it. Signed-off-by:
Wei Yongjun <weiyj.lk@gmail.com> Signed-off-by:
-
David S. Miller authored
Saeed Mahameed says: ==================== Mellanox 100G mlx5 ethtool ntuple steering This series adds Ethernet ethtool ntuple steering 'ethtool -N|U' and exposes two more counter sets to Ethtool statistics, RDMA vport and global flow control statistics. We start from three refactoring patches of the flow steering infrastructure - mlx5_add_flow_rule will now receive mlx5 flow spec to simplify and reduce number of parameters - All low level steering objects are now wrapped in mlx5_flow_steering structure for better encapsulation - Flow steering object will now be removed properly and generically rather than traversing on a well-known steering tree objects Patch#4 adds the infrastructure and the data structures needed for the ethtool ntuple steering, all implemented in a new file 'en_fs_ethtool.c'. Add the support for set_rxnfc ethtool callback to add/remove/replace a flow spec of ethter type L2. Patch#5 adds the support for L3/L4 flow specs and a higher priority in favor for L3/L4 rules when interleaving with L2 rules. Patch#6 adds the support for get_rxnfc ethtool callback. Patch#7,8 adds RDMA vport and global flow control statistics. Applied on top: 8186f6e3 ('net-next: mediatek: fix compile error inside mtk_poll_controller()') ==================== Signed-off-by: -
Gal Pressman authored
Just like per prio counters, the global flow counters are queried from per priority counters register. Global flow control counters are stored in priority 0 PFC counters. Signed-off-by:
Gal Pressman <galp@mellanox.com> Signed-off-by:
Saeed Mahameed <saeedm@mellanox.com> Signed-off-by:
-
Gal Pressman authored
Add the needed descriptors to expose RoCE RDMA counters. Signed-off-by:
-
Maor Gottlieb authored
Enhance the existing get_rxnfc callback: 1. Get flow rule of specific ID. 2. Get all flow rules. 3. Get number of rules. Signed-off-by:
Maor Gottlieb <maorg@mellanox.com> Signed-off-by:
-
Maor Gottlieb authored
Add support to add flow steering rules with ethtool of L3/L4 flow types (ip4/tcp4/udp4). Those rules will be in higher priority than l2 flow rules, in order to prefer more specific rules. Mask is not supported for l3/l4 flow types. Signed-off-by:
-
Maor Gottlieb authored
Implement etrhtool set_rxnfc callback to support ethtool flow spec direct steering. This patch adds only the support of ether flow type spec. L3/L4 flow specs support will be added in downstream patches. Signed-off-by:
-
Maor Gottlieb authored
Instead of explicitly cleaning up the well known parts of the steering tree, we use the generic tree structure to traverse for cleanup. No functional changes. Signed-off-by:
-
Maor Gottlieb authored
Instead of having all steering private name spaces and steering module fields flat in mlx5_core_priv, we wrap them in mlx5_flow_steering for better modularity and API exposure. Signed-off-by:
-
Maor Gottlieb authored
Reduce the set of arguments passed to mlx5_add_flow_rule by introducing flow_spec structure. Signed-off-by:
-
John Crispin authored
Commit 80673029 ("net-next: mediatek: add support for IRQ grouping") adds handling for irq 1 and 2 to the uninit function but did not remove irq 0 which is not used since irq grouping was introduced. Fix this by removing the superfluous call to free_irq(). Reported-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
John Crispin <john@phrozen.org> Acked-by:
-
Fabio Estevam authored
Since commit f786f356 ("net: ethernet: lpc_eth: use phydev from struct net_device") the 'pldat' variable became unused, so just remove it. Reported-by:
Olof's autobuilder <build@lixom.net> Signed-off-by:
Fabio Estevam <festevam@gmail.com> Signed-off-by:
-
git://git.open-mesh.org/linux-mergeDavid S. Miller authored
Simon Wunderlich says: ==================== This feature patchset includes the following changes: - Cleanup work by Markus Pargmann and Sven Eckelmann (six patches) - Initial Netlink support by Matthias Schiffer (two patches) - Throughput Meter implementation by Antonio Quartulli, a kernel-space traffic generator to estimate link speeds. This feature is useful on low-end WiFi APs where running iperf or netperf from userspace gives wrong results due to heavy userspace/kernelspace overhead. (two patches) - API clean-up work by Antonio Quartulli (one patch) ==================== Signed-off-by: -
David S. Miller authored
Jiri Pirko says: ==================== mlxsw: Implement basic FIB offload and router interfaces Introduce LPM trees management including virtual router management for HW. Implement basic FIB offloading using switchdev FIB objects. For now only support local routes and direct routes (next-hop support will be introduced in a follow-up patchset). Introduce router interfaces in patches 10-14. ==================== Signed-off-by: -
Ido Schimmel authored
As with the previously introduced L3 interfaces, listen to 'inetaddr' notifications sent for bridges devices configured on top of the port netdevs and create / destroy router interfaces (RIFs) accordingly. This also includes VLAN devices configured on top of the VLAN-aware bridge. The RIFs will be destroyed either when the last IP address is removed or when the underlying FID is is destroyed. Signed-off-by:
-
Ido Schimmel authored
Before introducing support for L3 interfaces on top of the VLAN-aware bridge we need to add some missing infrastructure. Such an interface can either be the bridge device itself or a VLAN device on top of it. In the first case the router interface (RIF) is associated with FID 1, which is created whenever the first port netdev joins the bridge. We currently assume the default PVID is 1 and that it's already created, as it seems reasonable. This can be extended in the future. However, in the second case it's entirely possible we've yet to create a matching FID. This can happen if the VLAN device was configured before making any bridge port member in the VLAN. Prevent such ordering problems by using the VLAN device's CHANGEUPPER event to configure the FID. Make the VLAN device hold a reference to the FID and prevent it from being destroyed even if none of the port netdevs is using it. Signed-off-by:
-
Ido Schimmel authored
Previous commit deprecated the vFIDs used to get traffic to the CPU ('port_vfids'). Thus, we now use the vFIDs as god intended and the artificial split is no longer needed. Rename functions and variables to reflect that. Signed-off-by: -
Ido Schimmel authored
Up until now we only supported bridged interfaces. Packets ingressing through the switch ports were either classified to FIDs (in the case of the VLAN-aware bridge) or vFIDs (in the case of VLAN-unaware bridges). The packets were then forwarded according to the FDB. Routing was done entirely in slowpath, by splitting the vFID range in two and using the lower 0.5K vFIDs as dummy bridges that simply flooded all incoming traffic to the CPU. Instead, allow packets to be routed in the device by creating router interfaces (RIFs) that will direct them to the router block. Specifically, the RIFs introduced here are Sub-port RIFs used for VLAN devices and port netdevs. Packets ingressing from the {Port / LAG ID, VID} with which the RIF was programmed with will be assigned to a special kind of FIDs called rFIDs and from there directed to the router. Create a RIF whenever the first IPv4 address was programmed on a VLAN / LAG / port netdev. Destroy it upon removal of the last IPv4 address. Receive these notifications by registering for the 'inetaddr' notification chain. A non-zero (10) priority is used for the notification block, so that RIFs will be created before routes are offloaded via FIB code. Note that another trigger for RIF destruction are CHANGEUPPER notifications causing the underlying FID's reference count to go down to zero. This can happen, for example, when a VLAN netdev with an IP address is put under bridge. While this configuration doesn't make sense it does cause the device and the kernel to get out of sync when the netdev is unbridged. We intend to address this in the future, hopefully in current cycle. Finally, Remove the lower 0.5K vFIDs, as they are deprecated by the RIFs, which will trap packets according to their DIP. Signed-off-by: -
Ido Schimmel authored
We are just about to introduce router interfaces (RIFs), but before that we need to be able update the device with the correct RIF attributes whenever they change for the netdev the RIF is backing. Two such attributes are MTU and MAC. The MAC is used both to set the source MAC of packets egressing from the RIF and also to program an FDB rule that will direct packets to the router block. Use the existing netdevice notification block and respond to CHANGEADDR and CHANGEMTU accordingly. Store both attributes in the RIF struct in case we need to revert to old attributes following a failed update. Signed-off-by:
-
Jiri Pirko authored
Add functions that iterate over lower devices and find port device. As a dependency add netdev_for_each_all_lower_dev and netdev_for_each_all_lower_dev_rcu macro with netdev_all_lower_get_next and netdev_all_lower_get_next_rcu shelpers. Also, add functions to return mlxsw struct according to lower device found and mlxsw_port struct with a reference to lower device. Signed-off-by:
-
