- 01 Jun, 2015 18 commits
-
-
Toshiaki Makita authored
Currently packets with non-hardware-accelerated vlan cannot be handled by GRO. This causes low performance for 802.1ad and stacked vlan, as their vlan tags are currently not stripped by hardware. This patch adds GRO support for non-hardware-accelerated vlan and improves receive performance of them. Test Environment: vlan device (.1Q) on vlan device (.1ad) on ixgbe (82599) Result: - Before $ netperf -t TCP_STREAM -H 192.168.20.2 -l 60 Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 87380 16384 16384 60.00 5233.17 Rx side CPU usage: %usr %sys %irq %soft %idle 0.27 58.03 0.00 41.70 0.00 - After $ netperf -t TCP_STREAM -H 192.168.20.2 -l 60 Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 87380 16384 16384 60.00 7586.85 Rx side CPU usage: %usr %sys %irq %soft %idle 0.50 25.83 0.00 59.53 14.14 [ Register VLAN offloads with priority 10 -DaveM ] Signed-off-by:
Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Hariprasad Shenai authored
Remove unused function cxgb4_enable_db_coalescing() and cxgb4_disable_db_coalescing() Signed-off-by:
Hariprasad Shenai <hariprasad@chelsio.com> Signed-off-by:
-
Simon Horman authored
The rocker (switch) of a rocker_port may be trivially obtained from the latter it seems cleaner not to pass the former to a function when the latter is being passed anyway. rocker_port_rx_proc() is omitted from this change as it is a hot path case. Signed-off-by:
Simon Horman <simon.horman@netronome.com> Acked-by:
Scott Feldman <sfeldma@gmail.com> Acked-by:
Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by:
-
Gabriel Krisman Bertazi authored
The driver allocates one page for each buffer on the rx ring, which is too much on architectures like ppc64 and can cause unexpected allocation failures when the system is under stress. Now, we keep a memory pool per queue, and if the architecture's PAGE_SIZE is greater than 4k, we fragment pages and assign each 4k segment to a ring element, which reduces the overall memory consumption on such architectures. This helps avoiding errors like the example below: [bnx2x_alloc_rx_sge:435(eth1)]Can't alloc sge [c00000037ffeb900] [d000000075eddeb4] .bnx2x_alloc_rx_sge+0x44/0x200 [bnx2x] [c00000037ffeb9b0] [d000000075ee0b34] .bnx2x_fill_frag_skb+0x1ac/0x460 [bnx2x] [c00000037ffebac0] [d000000075ee11f0] .bnx2x_tpa_stop+0x160/0x2e8 [bnx2x] [c00000037ffebb90] [d000000075ee1560] .bnx2x_rx_int+0x1e8/0xc30 [bnx2x] [c00000037ffebcd0] [d000000075ee2084] .bnx2x_poll+0xdc/0x3d8 [bnx2x] (unreliable) Signed-off-by:
Gabriel Krisman Bertazi <krisman@linux.vnet.ibm.com> Acked-by:
Yuval Mintz <Yuval.Mintz@qlogic.com> Reviewed-by:
Lino Sanfilippo <LinoSanfilippo@gmx.de> Signed-off-by:
-
Neil McKee authored
If new optional attribute OVS_USERSPACE_ATTR_ACTIONS is added to an OVS_ACTION_ATTR_USERSPACE action, then include the datapath actions in the upcall. This Directly associates the sampled packet with the path it takes through the virtual switch. Path information currently includes mangling, encapsulation and decapsulation actions for tunneling protocols GRE, VXLAN, Geneve, MPLS and QinQ, but this extension requires no further changes to accommodate datapath actions that may be added in the future. Adding path information enhances visibility into complex virtual networks. Signed-off-by:
Neil McKee <neil.mckee@inmon.com> Signed-off-by:
-
David S. Miller authored
When we scan a packet for GRO processing, we want to see the most common packet types in the front of the offload_base list. So add a priority field so we can handle this properly. IPv4/IPv6 get the highest priority with the implicit zero priority field. Next comes ethernet with a priority of 10, and then we have the MPLS types with a priority of 15. Suggested-by:
Eric Dumazet <eric.dumazet@gmail.com> Suggested-by:
-
Vaishali Thakkar authored
Use the timer API function setup_timer instead of structure field assignments to initialize a timer. A simplified version of the Coccinelle semantic patch that performs this transformation is as follows: @change@ expression e, func, da; @@ -init_timer (&e); +setup_timer (&e, func, da); -e.data = da; -e.function = func; Signed-off-by:
Vaishali Thakkar <vthakkar1994@gmail.com> Signed-off-by:
-
David S. Miller authored
Sowmini Varadhan says: ==================== net/rds: SOL_RDS socket option to explicitly select transport Today the underlying transport (TCP or IB) for a PF_RDS socket is implicitly selected based on the local address used to bind(2) the PF_RDS socket. This results in some non-deterministic behavior when there are un-numbered and IPoIB interfaces sharing the same IP address. It also places the constraint that the IB interface must have an IP address (and thus, IPoIB) configured on it. The non-determinism may be avoided by providing the user-space application a socket option that allows it to explicitly select the transport prior to bind(2). Patch 1 of this series provides the constant definitions needed by the application via <linux/rds.h>. Patch 2 provides the setsockopt support, and Patch 3 provides the getsockopt support. ==================== Signed-off-by: -
Sowmini Varadhan authored
The currently attached transport for a PF_RDS socket may be obtained from user space by invoking getsockopt(2) using the SO_RDS_TRANSPORT option at the SOL_RDS level. The integer optval returned will be one of the RDS_TRANS_* constants defined in linux/rds.h. Signed-off-by:
Sowmini Varadhan <sowmini.varadhan@oracle.com> Signed-off-by:
-
Sowmini Varadhan authored
An application may deterministically attach the underlying transport for a PF_RDS socket by invoking setsockopt(2) with the SO_RDS_TRANSPORT option at the SOL_RDS level. The integer argument to setsockopt must be one of the RDS_TRANS_* transport types, e.g., RDS_TRANS_TCP. The option must be specified before invoking bind(2) on the socket, and may only be used once on the socket. An attempt to set the option on a bound socket, or to invoke the option after a successful SO_RDS_TRANSPORT attachment, will return EOPNOTSUPP. Signed-off-by:
-
Sowmini Varadhan authored
User space applications that desire to explicitly select the underlying transport for a PF_RDS socket may do so by using the SO_RDS_TRANSPORT socket option at the SOL_RDS level before bind(). The integer argument provided to the socket option would be one of the RDS_TRANS_* values, e.g., RDS_TRANS_TCP. This commit exports the constant values need by such applications via <linux/rds.h> Signed-off-by:
-
Vaishali Thakkar authored
Use the timer API function setup_timer instead of structure field assignments to initialize a timer. A simplified version of the Coccinelle semantic patch that performs this transformation is as follows: @change@ expression e1, e2, e3, e4, a, b; @@ -init_timer(&e1); +setup_timer(&e1, a, b); ... when != a = e2 when != b = e3 -e1.function = a; ... when != b = e4 -e1.data = b; Signed-off-by: -
Daniel Borkmann authored
Besides others, move bpf_tail_call_proto to the remaining definitions of other protos, improve comments a bit (i.e. remove some obvious ones, where the code is already self-documenting, add objectives for others), simplify bpf_prog_array_compatible() a bit. Signed-off-by:
Daniel Borkmann <daniel@iogearbox.net> Acked-by:
Alexei Starovoitov <ast@plumgrid.com> Signed-off-by:
-
Daniel Borkmann authored
As this is already exported from tracing side via commit d9847d31 ("tracing: Allow BPF programs to call bpf_ktime_get_ns()"), we might as well want to move it to the core, so also networking users can make use of it, e.g. to measure diffs for certain flows from ingress/egress. Signed-off-by:
-
Vaishali Thakkar authored
Use the timer API function setup_timer instead of structure field assignments to initialize a timer. A simplified version of the Coccinelle semantic patch that performs this transformation is as follows: @change@ expression e1, e2, e3, e4, a, b; @@ -init_timer(&e1); +setup_timer(&e1, a, b); ... when != a = e2 when != b = e3 -e1.data = b; ... when != a = e4 -e1.function = a; Signed-off-by: -
Vaishali Thakkar authored
Use the timer API function setup_timer instead of structure field assignments to initialize a timer. A simplified version of the Coccinelle semantic patch that performs this transformation is as follows: @change@ expression e1, e2, e3, e4, a, b; @@ -init_timer(&e1); +setup_timer(&e1, a, b); ... when != a = e2 when != b = e3 -e1.data = b; ... when != a = e4 -e1.function = a; Signed-off-by: -
Vaishali Thakkar authored
Use the timer API function setup_timer instead of structure field assignments to initialize a timer. A simplified version of the Coccinelle semantic patch that performs this transformation is as follows: @change@ expression e, func, da; @@ -init_timer (&e); +setup_timer (&e, func, da); -e.data = da; -e.function = func; Signed-off-by:
-
David S. Miller authored
Merge tag 'mac80211-next-for-davem-2015-05-29' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next Johannes Berg says: ==================== As we get closer to the merge window, here are a few more things for -next: * disconnect TDLS stations on CSA to avoid issues * fix a memory leak introduced in a recent commit * switch rfkill and cfg80211 to PM ops * in an unlikely scenario, prevent a bookkeeping value to get corrupted leading to dropped packets * fix a crash in VLAN assignment * switch rfkill-gpio to more modern gpiod API * send disconnected event to userspace with proper local/remote indication ==================== Signed-off-by:
-
- 31 May, 2015 22 commits
-
-
git://git.open-mesh.org/linux-mergeDavid S. Miller authored
Antonio Quartulli says: ==================== Included changes: - checkpatch fixes - code cleanup - debugfs component is now compiled only if DEBUG_FS is selected - update copyright years - disable by default not-so-user-safe features ==================== Signed-off-by: -
Alexei Starovoitov authored
Normally the program attachment place (like sockets, qdiscs) takes care of rcu protection and calls bpf_prog_put() after a grace period. The programs stored inside prog_array may not be attached anywhere, so prog_array needs to take care of preserving rcu protection. Otherwise bpf_tail_call() will race with bpf_prog_put(). To solve that introduce bpf_prog_put_rcu() helper function and use it in 3 places where unattached program can decrement refcnt: closing program fd, deleting/replacing program in prog_array. Fixes: 04fd61ab ("bpf: allow bpf programs to tail-call other bpf programs") Reported-by:
Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by:
-
David S. Miller authored
K. Y. Srinivasan says: ==================== hv_netvsc: Implement NUMA aware memory allocation Allocate both receive buffer and send buffer from the NUMA node assigned to the primary channel. ==================== Signed-off-by: -
K. Y. Srinivasan authored
Allocate the send buffer in a NUMA aware way. Signed-off-by:
K. Y. Srinivasan <kys@microsoft.com> Signed-off-by:
-
K. Y. Srinivasan authored
Allocate the receive bufer from the NUMA node assigned to the primary channel. Signed-off-by:
-
Wang Long authored
Remove automatic variable 'err' in register_netevent_notifier() and return the result of atomic_notifier_chain_register() directly. Signed-off-by:
Wang Long <long.wanglong@huawei.com> Signed-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-nextDavid S. Miller authored
Pablo Neira Ayuso says: ==================== Netfilter updates for net-next The following patchset contains Netfilter updates for net-next, they are: 1) default CONFIG_NETFILTER_INGRESS to y for easier compile-testing of all options. 2) Allow to bind a table to net_device. This introduces the internal NFT_AF_NEEDS_DEV flag to perform a mandatory check for this binding. This is required by the next patch. 3) Add the 'netdev' table family, this new table allows you to create ingress filter basechains. This provides access to the existing nf_tables features from ingress. 4) Kill unused argument from compat_find_calc_{match,target} in ip_tables and ip6_tables, from Florian Westphal. ==================== Signed-off-by: -
David S. Miller authored
Florian Fainelli says: ==================== net: systemport: misc improvements These patches are highly inspired by changes from Petri on bcmgenet, last patch is a misc fix that I had pending for a while, but is not a candidate for 'net' at this point. ==================== Signed-off-by: -
Florian Fainelli authored
Occasionnaly we may get oversized packets from the hardware which exceed the nomimal 2KiB buffer size we allocate SKBs with. Add an early check which drops the packet to avoid invoking skb_over_panic() and move on to processing the next packet. Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
Florian Fainelli authored
Currently, bcm_sysport_desc_rx() calls bcm_sysport_rx_refill() at the end of Rx packet processing loop, after the current Rx packet has already been passed to napi_gro_receive(). However, bcm_sysport_rx_refill() might fail to allocate a new Rx skb, thus leaving a hole on the Rx queue where no valid Rx buffer exists. To eliminate this situation: 1. Rewrite bcm_sysport_rx_refill() to retain the current Rx skb on the Rx queue if a new replacement Rx skb can't be allocated and DMA-mapped. In this case, the data on the current Rx skb is effectively dropped. 2. Modify bcm_sysport_desc_rx() to call bcm_sysport_rx_refill() at the top of Rx packet processing loop, so that the new replacement Rx skb is already in place before the current Rx skb is processed. This is loosely inspired from d6707bec ("net: bcmgenet: rewrite bcmgenet_rx_refill()") Signed-off-by:
-
Florian Fainelli authored
There is a 1:1 mapping between the software maintained control block in priv->rx_cbs and the buffer address in priv->rx_bds, such that there is no need to keep computing the buffer address when refiling a control block. Signed-off-by:
-
Julia Lawall authored
Delete jump to a label on the next line, when that label is not used elsewhere. A simplified version of the semantic patch that makes this change is as follows: (http://coccinelle.lip6.fr/) // <smpl> @r@ identifier l; @@ -if (...) goto l; -l: // </smpl> Also remove the unnecessary ret variable. Signed-off-by:
Julia Lawall <Julia.Lawall@lip6.fr> Signed-off-by:
-
Arnd Bergmann authored
The thunderx ethernet driver fails to build on architectures that do not have an atomic readq() and writeq() function for 64-bit PCI bus access: drivers/net/ethernet/cavium/thunder/thunder_bgx.c: In function 'bgx_reg_read': include/asm-generic/io.h:195:23: error: implicit declaration of function 'readq' [-Werror=implicit-function-declaration] It seems impossible to get this driver to work on most 32-bit hardware, so it's better to add an explicit dependency, in order to let us keep building 'allmodconfig' kernels on all architectures. As the driver is meant for the internal hardware on an arm64 SoC, this is not a problem for usability. Allowing the build on all 64-bit architectures rather than just CONFIG_ARM64 on the other hand means that we get the benefit of build testing on x86. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
-
David S. Miller authored
Or Gerlitz says: ==================== mlx4 driver update, May 28, 2015 The 1st patch fixes an issue with a function running DPDK overriding broadcast steering rules set by other functions. Please add this one to your -stable queue. The rest of the series from Matan and Ido deals with scaling the number of IRQs that serve RoCE applications to be in par with the Ethernet driver. changes from V0: - addressed feedback from Sergei, removed extra blank line in patch #4 ==================== Signed-off-by: -
Matan Barak authored
When freeing a CQ, we need to make sure there are no asynchronous events (on the ASYNC EQ) that could relate to this CQ before freeing it. This is done by introducing synchronize_irq. Signed-off-by:
Matan Barak <matanb@mellanox.com> Signed-off-by:
Ido Shamay <idos@mellanox.com> Signed-off-by:
Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by:
-
Ido Shamay authored
Now that EQs management is in the sole responsibility of mlx4_core, the IRQ affinity hints configuration should be in its hands as well. request_irq is called only once by the first consumer (maybe mlx4_ib), so mlx4_en passes the affinity mask too late. We also need to request vectors according to the cores we want to run on. mlx4_core distribution of IRQs to cores is straight forward, EQ(i)->IRQ will set affinity hint to core i. Consumers need to request EQ vectors, according to their cores considerations (NUMA). Signed-off-by:
-
Matan Barak authored
Previously, mlx4_en allocated EQs and used them exclusively. This affected RoCE performance, as applications which are events sensitive were limited to use only the legacy EQs. Change that by introducing an EQ pool. This pool is managed by mlx4_core. EQs are assigned to ports (when there are limited number of EQs, multiple ports could be assigned to the same EQs). An exception to this rule is the ASYNC EQ which handles various events. Legacy EQs are completely removed as all EQs could be shared. When a consumer (mlx4_ib/mlx4_en) requests an EQ, it asks for EQ serving on a specific port. The core driver calculates which EQ should be assigned to that request. Because IRQs are shared between IB and Ethernet modules, their names only include the PCI device BDF address. Signed-off-by:
-
Matan Barak authored
In SRIOV, when simple (i.e - Ethernet L2 only) flow steering rules are created, always create them at MLX4_DOMAIN_NIC priority (instead of the real priority the function created them at). This is done in order to let multiple functions add broadcast/multicast rules without affecting other functions, which is necessary for DPDK in SRIOV. Signed-off-by:
-
David S. Miller authored
Merge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next Johan Hedberg says: ==================== pull request: bluetooth-next 2015-05-28 Here's a set of patches intended for 4.2. The majority of the changes are on the 802.15.4 side of things rather than Bluetooth related: - All sorts of cleanups & fixes to ieee802154 and related drivers - Rework of tx power support in ieee802154 and its drivers - Support for setting ieee802154 tx power through nl802154 - New IDs for the btusb driver - Various cleanups & smaller fixes to btusb - New btrtl driver for Realtec devices - Fix suspend/resume for Realtek devices Please let me know if there are any issues pulling. Thanks. ==================== Signed-off-by:
-
David S. Miller authored
Amir Vadai says: ==================== net/mlx5: ConnectX-4 100G Ethernet driver This patchset extends the mlx5_core driver to support Ethernet functionality. The Ethernet functionality in the mlx5 driver is integrated into the core driver and not as separated driver. The IB functionality remains in the mlx5_ib driver as before. This functionality will enable the Ethernet capability of Mellanox's new famility of cards - ConnectX-4. Due to the fact that backword compatability is being kept, existing Connect-IB cards that are using this driver are fully working with the modified driver, and no issues with current deployments should be seen. Like the ConnectX-3 cards, ConnectX-4 is a VPI (Virtual Port Interface - every port can be configured as Infiniband or Ethernet) card. Unlike previous generations, the ConnectX-4 has a separate PCI function per port. The current code has a limitation that Infiniband and Ethernet port types are mutually exclusive. When the driver is compiled with Ethernet support, the Infiniband functionality is disabled and vice versa. To control that we added the CONFIG_MLX5_CORE_EN config directive which is 'n' by default, but can be changed by the user. This limitation is short-lived and would be addressed soon. As part of this patchset, mlx5_ifc.h was heavily modified [1]. This file is now generated automatically from the device specification document. Since this patch is too big for the mail server, it might be missing in the mailing list, but could be pulled from an external git repository [2]. irq name selection is done at driver initialization and doesn't contain the interface name as part of the irq name. irq_balancer will still work thanks to an improvement introduced by Neil Horman [3] to use sysfs instead of /proc/interrupts. Patchset was applied on top of commit ed2dfd90 ("tcp/dccp: warn user for preferred ip_local_port_range") [1] - Patch 4/11 ("net/mlx5_core: HW data structs/types definitions preparation for mlx5 ehternet driver") [2] - http://git.openfabrics.org/?p=~amirv/linux.git;a=shortlog;h=refs/heads/mlx5e_v1 [3] - kernel: da8d1c8b PCI/sysfs: add per pci device msi[x] irq listing (v5) irq_balancer: 32a7757 Complete rework of how we detect and classify irqs Thanks to Achiad, Saeed, Yevheny, Or and the whole team for making this happen, Amir Changes from V4: - Removed Patch 3/12: net/mlx5_core: Add EQ renaming mechanism - Patch 12/12: net/mlx5: Extend mlx5_core to support ConnectX-4 Ethernet functionality - irq name is created on driver initialization, therefore it won't contain the network interface name in it. This won't effect irq_balancer thanks to patches introduced by Neil Horman to use sysfs instead of /proc/interrupts. Changes from V3: - PATCH 8/11: net/mlx5_core: Set/Query port MTU commands - Return value directly - no need for err. Changes from V2: - Improved changelogs and cover-letter - Added CONFIG_MLX5_EN to disable/enable the Ethernet functionality - Moved en.h and wq.[ch] into the patch with data-path related code Changes from V1: - Added patch 1/12 ("net/mlx5_core,mlx5_ib: Do not use vmap() on coherent memory") Changes from V0: - Removed V0 Patch 1/11 ("net/mlx5_core: Virtually extend work/completion queue buffers by one page") due to misuse of DMA API. Thanks Dave. - Patch 1/11 ("net/mlx5_core: Set irq affinity hints"): - Use kcalloc instead of kzalloc - Fix build error when CONFIG_CPUMASK_OFFSTACK=n. Driver loading will fail now if cpumask allocation is failing. - Using dev_to_node helper. Thanks, Ido. - Patch 3/11 ("net/mlx5_core: HW data structs/types definitions preparation for mlx5 ehternet driver") - Removed Mellanox internal comment at the head of the file. Thanks Joe - Patch 6/11 ("net/mlx5_core: Implement get/set port status") - Use direct return of function's result. Thanks Sergei. - Added Patch 8/11 ("net/mlx5_core: Set/Query port MTU commands") - Patch 9/11 ("net/mlx5: Ethernet Datapath files") - Use rq->wqe_sz instead of skb_end_offset. Thanks Ido. - Use dma_wmb() when possible instead of wmb(). Thanks Alex. - Fix checkpatch issues - Patch 10/11 ("net/mlx5: ethernet resources handling") - checkpatch issues - Added missing include - Patch 11/11 ("net/mlx5: Ethernet driver") - checkpatch issues - fixed typo - Modified use of affinity hint - Using dev_to_node helper. Thanks, Ido. - Use new hardware commands from Patch 8/11 ("net/mlx5_core: Set/Query port MTU commands") to get/set port MTU in hardware. - Removed NETIF_F_SG since hardware ring wraparound is not supported - Use dma_wmb() when possible instead of wmb(). Thanks Alex. ==================== Signed-off-by:
-
Amir Vadai authored
This is the Ethernet part of the driver for the Mellanox ConnectX(R)-4 Single/Dual-Port Adapter supporting 100Gb/s with VPI. The driver extends the existing mlx5 driver with Ethernet functionality. This patch contains the driver entry points but does not include transmit and receive (see the previous patch in the series) routines. It also adds the option MLX5_CORE_EN to Kconfig to enable/disable the Ethernet functionality. Currently, Kconfig is programmed to make Ethernet and Infiniband functionality mutally exclusive. Also changed MLX5_INFINIBAND to be depandant on MLX5_CORE instead of selecting it, since MLX5_CORE could be selected without MLX5_INFINIBAND being selected. Signed-off-by:
Amir Vadai <amirv@mellanox.com> Signed-off-by:
-
Amir Vadai authored
This patch contains the resource handling files: - flow_table.c: This file contains the code to handle the low level API to configure hardware flow table. It is separated from the flow_table_en.c, because it will be used in the future by Raw Ethernet QP in mlx5_ib too. - en_flow_table.[ch]: Ethernet flow steering handling. The flow table object contain a mapping between flow specs and TIRs. This mechanism will be used also to configure e-switch in the future, when SR-IOV support will be added. - transobj.[ch] - Low level functions to create/modify/destroy the transport objects: RQ/SQ/TIR/TIS - vport.[ch] - Handle attributes of a virtual port (vPort) in the embedded switch. Currently this switch is a passthrough, until SR-IOV support will be added. Signed-off-by:
-
