- 03 Oct, 2017 36 commits
-
-
David S. Miller authored
Marcelo Ricardo Leitner says: ==================== Introduce SCTP Stream Schedulers This patchset introduces the SCTP Stream Schedulers are defined by https://tools.ietf.org/html/draft-ietf-tsvwg-sctp-ndata-13 It provides 3 schedulers at the moment: FCFS, Priority and Round Robin. The other 3, Round Robin per packet, Fair Capacity and Weighted Fair Capacity will be added later. More specifically, WFQ is required by WebRTC Datachannels. The draft also defines the idata chunk, allowing a usermsg to be interrupted by another piece of idata from another stream. This patchset *doesn't* include it. It will be posted later by Xin Long. Its integration with this patchset is very simple and it basically only requires a tweak in sctp_sched_dequeue_done(), to ignore datamsg boundaries. The first 5 patches are a preparation for the next ones. The most relevant patches are the 4th and 6th ones. More details are available on each patch. v2: changelog update on patch 3 ==================== Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Marcelo Ricardo Leitner authored
This patch introduces RFC Draft ndata section 3.2 Priority Based Scheduler (SCTP_SS_RR). Works by maintaining a list of enqueued streams and tracking the last one used to send data. When the datamsg is done, it switches to the next stream. See-also: https://tools.ietf.org/html/draft-ietf-tsvwg-sctp-ndata-13Tested-by:
Xin Long <lucien.xin@gmail.com> Signed-off-by:
Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Signed-off-by:
-
Marcelo Ricardo Leitner authored
This patch introduces RFC Draft ndata section 3.4 Priority Based Scheduler (SCTP_SS_PRIO). It works by having a struct sctp_stream_priority for each priority configured. This struct is then enlisted on a queue ordered per priority if, and only if, there is a stream with data queued, so that dequeueing is very straightforward: either finish current datamsg or simply dequeue from the highest priority queued, which is the next stream pointed, and that's it. If there are multiple streams assigned with the same priority and with data queued, it will do round robin amongst them while respecting datamsgs boundaries (when not using idata chunks), to be reasonably fair. We intentionally don't maintain a list of priorities nor a list of all streams with the same priority to save memory. The first would mean at least 2 other pointers per priority (which, for 1000 priorities, that can mean 16kB) and the second would also mean 2 other pointers but per stream. As SCTP supports up to 65535 streams on a given asoc, that's 1MB. This impacts when giving a priority to some stream, as we have to find out if the new priority is already being used and if we can free the old one, and also when tearing down. The new fields in struct sctp_stream_out_ext and sctp_stream are added under a union because that memory is to be shared with other schedulers. See-also: https://tools.ietf.org/html/draft-ietf-tsvwg-sctp-ndata-13Tested-by:
-
Marcelo Ricardo Leitner authored
As defined per RFC Draft ndata Section 4.3.3, named as SCTP_STREAM_SCHEDULER_VALUE. See-also: https://tools.ietf.org/html/draft-ietf-tsvwg-sctp-ndata-13Tested-by:
-
Marcelo Ricardo Leitner authored
As defined per RFC Draft ndata Section 4.3.2, named as SCTP_STREAM_SCHEDULER. See-also: https://tools.ietf.org/html/draft-ietf-tsvwg-sctp-ndata-13Tested-by:
-
Marcelo Ricardo Leitner authored
This patch introduces the hooks necessary to do stream scheduling, as per RFC Draft ndata. It also introduces the first scheduler, which is what we do today but now factored out: first come first served (FCFS). With stream scheduling now we have to track which chunk was enqueued on which stream and be able to select another other than the in front of the main outqueue. So we introduce a list on sctp_stream_out_ext structure for this purpose. We reuse sctp_chunk->transmitted_list space for the list above, as the chunk cannot belong to the two lists at the same time. By using the union in there, we can have distinct names for these moments. sctp_sched_ops are the operations expected to be implemented by each scheduler. The dequeueing is a bit particular to this implementation but it is to match how we dequeue packets today. We first dequeue and then check if it fits the packet and if not, we requeue it at head. Thus why we don't have a peek operation but have dequeue_done instead, which is called once the chunk can be safely considered as transmitted. The check removed from sctp_outq_flush is now performed by sctp_stream_outq_migrate, which is only called during assoc setup. (sctp_sendmsg() also checks for it) The only operation that is foreseen but not yet added here is a way to signalize that a new packet is starting or that the packet is done, for round robin scheduler per packet, but is intentionally left to the patch that actually implements it. Support for I-DATA chunks, also described in this RFC, with user message interleaving is straightforward as it just requires the schedulers to probe for the feature and ignore datamsg boundaries when dequeueing. See-also: https://tools.ietf.org/html/draft-ietf-tsvwg-sctp-ndata-13Tested-by:
-
Marcelo Ricardo Leitner authored
Add a helper to fetch the stream number from a given chunk. Tested-by:
-
Marcelo Ricardo Leitner authored
With the stream schedulers, sctp_stream_out will become too big to be allocated by kmalloc and as we need to allocate with BH disabled, we cannot use __vmalloc in sctp_stream_init(). This patch moves out the stats from sctp_stream_out to sctp_stream_out_ext, which will be allocated only when the application tries to sendmsg something on it. Just the introduction of sctp_stream_out_ext would already fix the issue described above by splitting the allocation in two. Moving the stats to it also reduces the pressure on the allocator as we will ask for less memory atomically when creating the socket and we will use GFP_KERNEL later. Then, for stream schedulers, we will just use sctp_stream_out_ext. Tested-by:
-
Marcelo Ricardo Leitner authored
There is 1 place allocating it and another reallocating. Move such procedures to a common function. v2: updated changelog Tested-by:
-
Marcelo Ricardo Leitner authored
There is 1 place allocating it and 2 other reallocating. Move such procedures to a common function. Tested-by:
-
Marcelo Ricardo Leitner authored
As SCTP supports up to 65535 streams, that can lead to very large allocations in sctp_stream_init(). As Xin Long noticed, systems with small amounts of memory are more prone to not have enough memory and dump warnings on dmesg initiated by user actions. Thus, silence them. Also, if the reallocation of stream->out is not necessary, skip it and keep the memory we already have. Reported-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queueDavid S. Miller authored
Jeff Kirsher says: ==================== 100GbE Intel Wired LAN Driver Updates 2017-10-03 This series contains updates to fm10k only. Jake provides majority of the changes in this series, starting with using fm10k_prepare_for_reset() if we lose PCIe link. Before we would detach the device and close the netdev, which left a lot of items still active, such as the Tx/Rx resources. This could cause problems where register reads would return potentially invalid values and would result in unknown driver behavior, so call fm10k_prepare_for_reset() much like we do for suspend/resume cycles. This will attempt to shutdown as much as possible to prevent possible issues. Then replaced the PCI specific legacy power management hooks with the new generic power management hooks for both suspend and hibernate. Introduced a workqueue item which monitors a queue of MAC and VLAN requests since a large number of MAC address or VLAN updates at once can overload the mailbox with too many messages at once. Fixed a cppcheck warning by properly declaring the min_rate and max_rate variables in the declaration and definition for .ndo_set_vf_bw, rather than using "unused" for the minimum rates. Joe Perches fixes the backward logic when using net_ratelimit(). ==================== Signed-off-by: -
Florian Westphal authored
Device alias can be set by either rtnetlink (rtnl is held) or sysfs. rtnetlink hold the rtnl mutex, sysfs acquires it for this purpose. Add an extra mutex for it and use rcu to protect concurrent accesses. This allows the sysfs path to not take rtnl and would later allow to not hold it when dumping ifalias. Based on suggestion from Eric Dumazet. Signed-off-by:
Florian Westphal <fw@strlen.de> Signed-off-by:
-
Mahesh Bandewar authored
Some NIC drivers don't have correct speed/duplex settings at the time they send NETDEV_UP notification and that messes up the bonding state. Especially 802.3ad mode which is very sensitive to these settings. In the current implementation we invoke bond_update_speed_duplex() when we receive NETDEV_UP, however, ignore the return value. If the values we get are invalid (UNKNOWN), then slave gets removed from the aggregator with speed and duplex set to UNKNOWN while link is still marked as UP. This patch fixes this scenario. Also 802.3ad mode is sensitive to these conditions while other modes are not, so making sure that it doesn't change the behavior for other modes. Signed-off-by:
Mahesh Bandewar <maheshb@google.com> Signed-off-by:
-
Dan Carpenter authored
We accidentally return success if the kmalloc_array() call fails. Fixes: 0e14c777 ("mlxsw: spectrum: Add the multicast routing hardware logic") Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Acked-by:
Yotam Gigi <yotamg@mellanox.com> Signed-off-by:
-
Dan Carpenter authored
mlxsw_afa_block_create() doesn't return error pointers, it returns NULL on error. Fixes: 0e14c777 ("mlxsw: spectrum: Add the multicast routing hardware logic") Signed-off-by:
-
Colin Ian King authored
The function mt7530_phy_write is local to the source and does not need to be in global scope, so make it static. Cleans up sparse warnings: symbol 'mt7530_phy_write' was not declared. Should it be static? Signed-off-by:
Colin Ian King <colin.king@canonical.com> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Colin Ian King authored
The functions lan9303_mdio_phy_write and lan9303_mdio_phy_read are local to the source and do not need to be in global scope, so make them static. Cleans up sparse warnings: symbol 'lan9303_mdio_phy_write' was not declared. Should it be static? symbol 'lan9303_mdio_phy_read' was not declared. Should it be static? Signed-off-by:
-
David S. Miller authored
Jiri Pirko says: ==================== mlxsw: Add support for partial multicast route offload Yotam says: Previous patchset introduced support for offloading multicast MFC routes to the Spectrum hardware. As described in that patchset, no partial offloading is supported, i.e if a route has one output interface which is not a valid offloadable device (e.g. pimreg device, dummy device, management NIC), the route is trapped to the CPU and the forwarding is done in slow-path. Add support for partial offloading of multicast routes, by letting the hardware to forward the packet to all the in-hardware devices, while the kernel ipmr module will continue forwarding to all other interfaces. Similarly to the bridge, the kernel ipmr module will forward a marked packet to an interface only if the interface has a different parent ID than the packet's ingress interfaces. The first patch introduces the offload_mr_fwd_mark skb field, which can be used by offloading drivers to indicate that a packet had already gone through multicast forwarding in hardware, similarly to the offload_fwd_mark field that indicates that a packet had already gone through L2 forwarding in hardware. Patches 2 and 3 change the ipmr module to not forward packets that had already been forwarded by the hardware, i.e. packets that are marked with offload_mr_fwd_mark and the ingress VIF shares the same parent ID with the egress VIF. Patches 4, 5, 6 and 7 add the support in the mlxsw Spectrum driver for trap and forward routes, while marking the trapped packets with the offload_mr_fwd_mark. ==================== Signed-off-by: -
Yotam Gigi authored
Add the support of trap-and-forward route action in the multicast routing offloading logic. A route will be set to trap-and-forward action if one (or more) of its output interfaces is not offload-able, i.e. does not have a valid Spectrum RIF. This way, a route with mixed output VIFs list, which contains both offload-able and un-offload-able devices can go through partial offloading in hardware, and the rest will be done in the kernel ipmr module. Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Signed-off-by:
Jiri Pirko <jiri@mellanox.com> Signed-off-by:
-
Yotam Gigi authored
In addition to the current multicast route actions, which include trap route action and a forward route action, add the trap-and-forward multicast route action, and implement it in the multicast routing hardware logic. To implement that, add a trap-and-forward ACL action as the last action in the route flexible action set. The used trap is the ACL2 trap, which marks the packets with offload_mr_forward_mark, to prevent the packet from being forwarded again by the kernel. Note: At that stage the offloading logic does not support trap-and-forward multicast routes. This patch adds the support only in the hardware logic. Signed-off-by:
-
Yotam Gigi authored
When a multicast route is configured with trap-and-forward action, the packets should be marked with skb->offload_mr_fwd_mark, in order to prevent the packets from being forwarded again by the kernel ipmr module. Due to this, it is not possible to use the already existing multicast trap (MLXSW_TRAP_ID_ACL1) as the packet should be marked differently. Add the MLXSW_TRAP_ID_ACL2 which is for trap-and-forward multicast routes, and set the offload_mr_fwd_mark skb field in its handler. Signed-off-by:
-
Yotam Gigi authored
Use trap/discard flex action to implement trap and forward. The action will later be used for multicast routing, as the multicast routing mechanism is done using ACL flexible actions in Spectrum hardware. Using that action, it will be possible to implement a trap-and-forward route. Signed-off-by:
-
Yotam Gigi authored
Change the ipmr module to not forward packets if: - The packet is marked with the offload_mr_fwd_mark, and - Both input interface and output interface share the same parent ID. This way, a packet can go through partial multicast forwarding in the hardware, where it will be forwarded only to the devices that share the same parent ID (AKA, reside inside the same hardware). The kernel will forward the packet to all other interfaces. To do this, add the ipmr_offload_forward helper, which per skb, ingress VIF and egress VIF, returns whether the forwarding was offloaded to hardware. The ipmr_queue_xmit frees the skb and does not forward it if the result is a true value. All the forwarding path code compiles out when the CONFIG_NET_SWITCHDEV is not set. Signed-off-by:
Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Signed-off-by:
-
Yotam Gigi authored
In order to allow the ipmr module to do partial multicast forwarding according to the device parent ID, add the device parent ID field to the VIF struct. This way, the forwarding path can use the parent ID field without invoking switchdev calls, which requires the RTNL lock. When a new VIF is added, set the device parent ID field in it by invoking the switchdev_port_attr_get call. Signed-off-by:
-
Yotam Gigi authored
Similarly to the offload_fwd_mark field, the offload_mr_fwd_mark field is used to allow partial offloading of MFC multicast routes. Switchdev drivers can offload MFC multicast routes to the hardware by registering to the FIB notification chain. When one of the route output interfaces is not offload-able, i.e. has different parent ID, the route cannot be fully offloaded by the hardware. Examples to non-offload-able devices are a management NIC, dummy device, pimreg device, etc. Similar problem exists in the bridge module, as one bridge can hold interfaces with different parent IDs. At the bridge, the problem is solved by the offload_fwd_mark skb field. Currently, when a route cannot go through full offload, the only solution for a switchdev driver is not to offload it at all and let the packet go through slow path. Using the offload_mr_fwd_mark field, a driver can indicate that a packet was already forwarded by hardware to all the devices with the same parent ID as the input device. Further patches in this patch-set are going to enhance ipmr to skip multicast forwarding to devices with the same parent ID if a packets is marked with that field. The reason why the already existing "offload_fwd_mark" bit cannot be used is that a switchdev driver would want to make the distinction between a packet that has already gone through L2 forwarding but did not go through multicast forwarding, and a packet that has already gone through both L2 and multicast forwarding. For example: when a packet is ingressing from a switchport enslaved to a bridge, which is configured with multicast forwarding, the following scenarios are possible: - The packet can be trapped to the CPU due to exception while multicast forwarding (for example, MTU error). In that case, it had already gone through L2 forwarding in the hardware, thus A switchdev driver would want to set the skb->offload_fwd_mark and not the skb->offload_mr_fwd_mark. - The packet can also be trapped due to a pimreg/dummy device used as one of the output interfaces. In that case, it can go through both L2 and (partial) multicast forwarding inside the hardware, thus a switchdev driver would want to set both the skb->offload_fwd_mark and skb->offload_mr_fwd_mark. Signed-off-by:
Ido Schimmel <idosch@mellaox.com> Signed-off-by:
-
Arjun Vynipadath authored
We have lost a comment for minimum mtu value set for netdevice with 'commit d894be57 ("ethernet: use net core MTU range checking in more drivers"). Updating it accordingly. Signed-off-by:
Arjun Vynipadath <arjun@chelsio.com> Signed-off-by:
Ganesh Goudar <ganeshgr@chelsio.com> Signed-off-by:
-
Jacob Keller authored
We've had support for setting both a minimum and maximum bandwidth via .ndo_set_vf_bw since commit 883a9ccb ("fm10k: Add support for SR-IOV to driver", 2014-09-20). Likely because we do not support minimum rates, the declaration mis-ordered the "unused" parameter, which causes warnings when analyzed with cppcheck. Fix this warning by properly declaring the min_rate and max_rate variables in the declaration and definition (rather than using "unused"). Also rename "rate" to max_rate so as to clarify that we only support setting the maximum rate. Signed-off-by:
Jacob Keller <jacob.e.keller@intel.com> Tested-by:
Krishneil Singh <krishneil.k.singh@intel.com> Signed-off-by:
Jeff Kirsher <jeffrey.t.kirsher@intel.com>
-
Jacob Keller authored
Don't hard code the function names in the diagnostic output when these reset related routines fail. Instead, use %s and __func__ so that future refactors don't need to change the print outs. Additionally, while we are here, add missing function header comments for the new reset_prepare and reset_done function handlers. Signed-off-by:
-
Joe Perches authored
Correct the backward logic using !net_ratelimit() Miscellanea: o Add a blank line before the error return label Signed-off-by:
Joe Perches <joe@perches.com> Tested-by:
-
Jacob Keller authored
Signed-off-by:
-
Jacob Keller authored
Now that we have a working MAC/VLAN queue for handling MAC/VLAN messages from the netdev, replace the default handler for the VF<->PF messages. This new handler is very similar to the default code, but uses the MAC/VLAN queue instead of sending the message directly. Unfortunately we can't easily re-use the default code, so we'll just replace the entire function. This ensures that a VF requesting a large number of VLANs or MAC addresses does not start a reset cycle, as explained in the commit which introduced the message queue. Signed-off-by:
Ngai-mint Kwan <ngai-mint.kwan@intel.com> Tested-by:
-
Jacob Keller authored
Under some circumstances, when dealing with a large number of MAC address or VLAN updates at once, the fm10k driver, particularly the VFs can overload the mailbox with too many messages at once. This results in a mailbox timeout, which causes the driver to initiate a reset. During the reset, we re-send all the same messages that originally caused the timeout. This results in a cycle of resets each triggering a future reset. To fix or avoid this, we introduce a workqueue item which monitors a queue of MAC and VLAN requests. These requests are queued to the end of the list, and we process as a FIFO periodically. Initially we only handle requests for the netdev, but we do handle unicast MAC addresses, multicast MAC addresses, and update VLAN requests. A future patch will add support to use this queue for handling MAC update requests from the VF<->PF mailbox. The MAC/VLAN work item will keep checking to make sure that each request does not overflow the mailbox and cause a timeout. If it might, then the work item will reschedule itself a short time later. This avoids any reset cycle, since we never send the message if the mailbox is not ready. As an alternative, we tried increasing the mailbox message FIFO, but this just delays the problem and results in needless memory waste on the system. Our new message queue is dynamically allocated so only uses as much memory as it needs. Additionally, it need not be contiguous like the Tx and Rx FIFOs. Note that this patch chose to only create a queue for MAC and VLAN messages, since these are the only messages sent in a large enough volume to cause the reset loop. Other messages are very unlikely to overflow the mailbox Tx FIFO so easily. Signed-off-by:
-
Jacob Keller authored
Replace the PCI specific legacy power management hooks with the new generic power management hooks which work properly for both suspend and hibernate. The new generic system is better and properly handles the lower level PCIe power management rather than forcing the driver to handle it. Signed-off-by:
-
Jacob Keller authored
Lets not re-invent the locking wheel. Remove our bitlock and use a proper spinlock instead. Signed-off-by:
-
Jacob Keller authored
If we lose PCIe link, such as when an unannounced PFLR event occurs, or when a device is surprise removed, we currently detach the device and close the netdev. This unfortunately leaves a lot of things still active, such as the msix_mbx_pf IRQ, and Tx/Rx resources. This can cause problems because the register reads will return potentially invalid values which may result in unknown driver behavior. Begin the process of resetting using fm10k_prepare_for_reset(), much in the same way as the suspend and resume cycle does. This will attempt to shutdown as much as possible, in order to prevent possible issues. A naive implementation for this has issues, because there are now multiple flows calling the reset logic and setting a reset bit. This would cause problems, because the "re-attach" routine might call fm10k_handle_reset() prior to the reset actually finishing. Instead, we'll add state bits to indicate which flow actually initiated the reset. For the general reset flow, we'll assume that if someone else is resetting that we do not need to handle it at all, so it does not need its own state bit. For the suspend case, we will simply issue a warning indicating that we are attempting to recover from this case when resuming. For the detached subtask, we'll simply refuse to re-attach until we've actually initiated a reset as part of that flow. Finally, we'll stop attempting to manage the mailbox subtask when we're detached, since there's nothing we can do if we don't have a PCIe address. Overall this produces a much cleaner shutdown and recovery cycle for a PCIe surprise remove event. Signed-off-by:
-
- 02 Oct, 2017 4 commits
-
-
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queueDavid S. Miller authored
Jeff Kirsher says: ==================== 40GbE Intel Wired LAN Driver Updates 2017-10-02 This series contains updates to i40e and i40evf. Shannon Nelson fixes an issue where when a machine has more CPUs than queue pairs, the counting gets a "little funky" and turns off Flow Director. So to correct it, limit the number of LAN queues initially allocated to be sure there are some left for Flow Director and other features. Lihong cleans up dead code by removing a condition check which cannot ever be true. Christophe Jaillet fixes a potential NULL pointer dereference, which could happen if kzalloc() fails. Filip corrects the reporting of supported link modes, which was incorrect for some NICs. Added support for 'ethtool -m' command, which displays information about QSFP+ modules. Mariusz adds functions to read/write the LED registers to control the LEDS, instead of accessing the registers directly whenever the LEDs need to be controlled. Jake fixes a regression where we introduced a scheduling while atomic, so introduce a separate helper function which will manage its own need for the mac_filter_hash_lock. Also cleaned up the "PF" parameter in i40e_vc_disable_vf() since it is never used and is not needed. Fixed a rare case where it is possible that a reset does not occur when i40e_vc_disable_vf() is called, so modify i40e_reset_vf() to return a bool to indicate whether it reset or not so that i40e_vc_disable_vf() can wait until a reset actually occurs. Alan adds the ability for the VF to request more or less underlying allocated queues from the PF. Fixes the incorrect method for clearing the vf_states variable with a NULL assignment, when we should be using atomic bitops since we don't actually want to clear all the flags. Fixed a resource leak, where the PF driver fails to inform clients of a VF reset because we were incorrectly checking the I40E_VF_STATE_PRE_ENABLE bit. Mitch converts i40evf_map_rings_to_vectors() to a void function since it cannot fail and allows us to clean up the checks for the function return value. Scott enables the driver(s) to pass traffic with VLAN tags using the 802.1ad Ethernet protocol. ==================== Signed-off-by: -
Scott Peterson authored
Enable i40e to pass traffic with VLAN tags using the 802.1ad ethernet protocol ID (0x88a8). This requires NIC firmware providing version 1.7 of the API. With older NIC firmware 802.1ad tagged packets will continue to be dropped. No VLAN offloads nor RSS are supported for 802.1ad VLANs. Signed-off-by:
Scott Peterson <scott.d.peterson@intel.com> Tested-by:
Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by:
-
Alan Brady authored
Currently there is a bug in which the PF driver fails to inform clients of a VF reset which then causes clients to leak resources. The bug exists because we were incorrectly checking the I40E_VF_STATE_PRE_ENABLE bit. When a VF is first init we go through a reset to initialize variables and allocate resources but we don't want to inform clients of this first reset since the client isn't fully enabled yet so we set a state bit signifying we're in a "pre-enabled" client state. During the first reset we should be clearing the bit, allowing all following resets to notify the client of the reset when the bit is not set. This patch fixes the issue by negating the 'test_and_clear_bit' check to accurately reflect the behavior we want. Signed-off-by:
Alan Brady <alan.brady@intel.com> Tested-by:
-
Alan Brady authored
Currently we inappropriately clear the vf_states variable with a null assignment. This is problematic because we should be using atomic bitops on this variable and we don't actually want to clear all the flags. We should just clear the ones we know we want to clear. Additionally remove the I40E_VF_STATE_FCOEENA bit because it is no longer being used. Signed-off-by:
-
