- 30 Aug, 2017 4 commits
-
-
Florian Westphal authored
This reverts commit 45f119bf. Eric Dumazet says: We found at Google a significant regression caused by 45f119bf tcp: remove header prediction In typical RPC (TCP_RR), when a TCP socket receives data, we now call tcp_ack() while we used to not call it. This touches enough cache lines to cause a slowdown. so problem does not seem to be HP removal itself but the tcp_ack() call. Therefore, it might be possible to remove HP after all, provided one finds a way to elide tcp_ack for most cases. Reported-by:
Eric Dumazet <edumazet@google.com> Signed-off-by:
Florian Westphal <fw@strlen.de> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Florian Westphal authored
This change was a followup to the header prediction removal, so first revert this as a prerequisite to back out hp removal. Signed-off-by:
-
Greg KH authored
When moving the IRDA code out of net/ into drivers/staging/irda/net, the link order changes when IRDA is built into the kernel. That causes a kernel crash at boot time as netfilter isn't initialized yet. To fix this, move the init call level of the irda core to be device_initcall() as the link order keeps this being initialized at the correct time. Reported-by:
kernel test robot <fengguang.wu@intel.com> Reported-by:
Geert Uytterhoeven <geert@linux-m68k.org> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by:
-
Florian Fainelli authored
A stray return was added in the macro bcmgenet_##name##_writel where it should not, drop it. Reported-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
- 29 Aug, 2017 36 commits
-
-
Eric Dumazet authored
Florian reported UDP xmit drops that could be root caused to the too small neigh limit. Current limit is 64 KB, meaning that even a single UDP socket would hit it, since its default sk_sndbuf comes from net.core.wmem_default (~212992 bytes on 64bit arches). Once ARP/ND resolution is in progress, we should allow a little more packets to be queued, at least for one producer. Once neigh arp_queue is filled, a rogue socket should hit its sk_sndbuf limit and either block in sendmsg() or return -EAGAIN. Signed-off-by:
-
Dave Jiang authored
Since the removal of NET_DMA, dmaengine.h header file shouldn't be needed by netdevice.h anymore. Signed-off-by:
Dave Jiang <dave.jiang@intel.com> Signed-off-by:
-
Florian Fainelli authored
The GENET driver currently uses __raw_{read,write}l which means native I/O endian. This works correctly for an ARM LE kernel (default) but fails miserably on an ARM BE (BE8) kernel where registers are kept little endian, so replace uses with {read,write}l_relaxed here which is what we want because this is all performance sensitive code. Signed-off-by: -
Weilin Chang authored
Signed-off-by:
Weilin Chang <weilin.chang@cavium.com> Signed-off-by:
Felix Manlunas <felix.manlunas@cavium.com> Signed-off-by:
-
Bhumika Goyal authored
Make these const as they are not modified anywhere. Signed-off-by:
Bhumika Goyal <bhumirks@gmail.com> Signed-off-by:
-
David Ahern authored
Tariq repored local pings to linklocal address is failing: $ ifconfig ens8 ens8: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 inet 11.141.16.6 netmask 255.255.0.0 broadcast 11.141.255.255 inet6 fe80::7efe:90ff:fecb:7502 prefixlen 64 scopeid 0x20<link> ether 7c:fe:90:cb:75:02 txqueuelen 1000 (Ethernet) RX packets 12 bytes 1164 (1.1 KiB) RX errors 0 dropped 0 overruns 0 frame 0 TX packets 30 bytes 2484 (2.4 KiB) TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0 $ /bin/ping6 -c 3 fe80::7efe:90ff:fecb:7502%ens8 PING fe80::7efe:90ff:fecb:7502%ens8(fe80::7efe:90ff:fecb:7502) 56 data bytes Signed-off-by: -
Tom Lendacky authored
There is a difference in the bit position of the normal interrupt summary enable (NIE) and abnormal interrupt summary enable (AIE) between revisions of the hardware. For older revisions the NIE and AIE bits are positions 16 and 15 respectively. For newer revisions the NIE and AIE bits are positions 15 and 14. The effect in changing the bit position is that newer hardware won't receive AIE interrupts in the current version of the driver. Specifically, the driver uses this interrupt to collect statistics on when a receive buffer unavailable event occurs and to restart the driver/device when a fatal bus error occurs. Update the driver to set the interrupt enable bit based on the reported version of the hardware. Signed-off-by:
Tom Lendacky <thomas.lendacky@amd.com> Signed-off-by:
-
David S. Miller authored
Jiri Benc says: ==================== nsh: headers, GSO This adds header structs and helpers for NSH together with GSO support. Note there is no code in this patchset that actually manipulates the NSH headers. That was sent to netdev by Yi Yang ("[PATCH net-next v6 0/3] openvswitch: add NSH support"). The aim of this series is to lay the groundwork and ease the implementation for him. In addition to openvswitch, the NSH support should be added to tc (flower to match, act_nsh to push/pop NSH headers). That will come later. There's currently no plan to support NSH by other means than those two. The patch 3 in this patchset was written by Yi Yang, I took it from the aforementioned series and slightly modified it - see the note in the patch. ==================== Signed-off-by: -
Jiri Benc authored
Add a new nsh/ directory. It currently holds only GSO functions but more will come: in particular, code shared by openvswitch and tc to manipulate NSH headers. For now, assume there's no hardware support for NSH segmentation. We can always introduce netdev->nsh_features later. Signed-off-by:
Jiri Benc <jbenc@redhat.com> Signed-off-by:
-
Yi Yang authored
NSH (Network Service Header)[1] is a new protocol for service function chaining, it can be handled as a L3 protocol like IPv4 and IPv6, Eth + NSH + Inner packet or VxLAN-gpe + NSH + Inner packet are two typical use cases. This patch adds NSH header structures and helpers for NSH GSO support and Open vSwitch NSH support. [1] https://datatracker.ietf.org/doc/draft-ietf-sfc-nsh/ [Jiri: added nsh_hdr() helper and renamed the header struct to "struct nshhdr" to match the usual pattern. Removed packet type defines, these are now shared with VXLAN-GPE.] Signed-off-by:
Yi Yang <yi.y.yang@intel.com> Signed-off-by:
-
Jiri Benc authored
The values are shared between VXLAN-GPE and NSH. Originally probably by coincidence but I notified both working groups about this last year and they seem to keep the values in sync since then. Hopefully they'll get a single IANA registry for the values, too. (I asked them for that.) Factor out the code to be shared by the NSH implementation. NSH and MPLS values are added in this patch, too. For MPLS, the drafts incorrectly assign only a single value, while we have two MPLS ethertypes. I raised the problem with both groups. For now, I assume the value is for unicast. Signed-off-by:
-
Jiri Benc authored
The NSH draft says: An IEEE EtherType, 0x894F, has been allocated for NSH. Signed-off-by:
-
David S. Miller authored
Alexander Aring says: ==================== tc: act_ife: handle IEEE IFE ethertype as default this patch series will introduce the IFE ethertype which is registered by IEEE. If the netlink act_ife type netlink attribute is not given it will use this value by default now. At least it will introduce some UAPI testcases to check if the default type is used if not specified and vice versa. ==================== Signed-off-by: -
Alexander Aring authored
This patch adds a new testcase for the IFE type setting in tc. In case of user specified the type it will check if the ife is correctly configured to react on it. If it's not specified the default IFE type should be used. Signed-off-by:
Alexander Aring <aring@mojatatu.com> Acked-by:
Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by:
-
Alexander Aring authored
This patch handles a default IFE type if it's not given by user space netlink api. The default IFE type will be the registered ethertype by IEEE for IFE ForCES. Signed-off-by:
-
Alexander Aring authored
This patch adds the forces IFE lfb type according to IEEE registered ethertypes. See http://standards-oui.ieee.org/ethertype/eth.txt for more information. Since there exists the IFE subsystem it can be used there. This patch also use the correct word "ForCES" instead of "FoRCES" which is a spelling error inside the IEEE ethertype specification. Signed-off-by:
-
Florian Fainelli authored
Explain that the patch queue in patchwork should not be touched by patch submitters. Signed-off-by:
-
David S. Miller authored
Tariq Toukan says: ==================== mlx4 misc patches This patchset contains misc patches from the team to the mlx4 Core and Eth drivers. Patch 1 by Eran replaces large static allocations by dynamic ones. Patch 2 by Leon makes an explicit conversion and solves a smatch warning. In patch 3 I fix a misplaced brackets of the sizeof operation. Patch 4 by Moshe adds the ability to inform the FW regarding user mac updates. Series generated against net-next commit: 901c5d2f ARM: dts: rk3228-evb: Fix the compiling error ==================== Signed-off-by:
-
Moshe Shemesh authored
Adding support for updating the FW on new port mac, when port mac change is requested by the user. This info is required by the FW as OEM management tools require this info directly from the NIC FW. Check device capability bit to verify the FW supports user mac. If the FW does support it, use set_port command to notify the FW on the new mac. The feature is relevant only to PF port mac. Signed-off-by:
Moshe Shemesh <moshe@mellanox.com> Signed-off-by:
Tariq Toukan <tariqt@mellanox.com> Signed-off-by:
-
Tariq Toukan authored
When changing the sizeof style usage in the patch cited below, one brackets misplacement was introduced. Here we fix it. Fixes: 31975e27 ("mlx4: sizeof style usage") Signed-off-by:
-
Leon Romanovsky authored
The "lg" variable is declared as int so in all places where this variable is used as a shift operand, the output will be int too. This produces the following smatch warning: drivers/net/ethernet/mellanox/mlx4/fw.c:1532 mlx4_map_cmd() warn: should '1 << lg' be a 64 bit type? Simple declaration of "1" to be "1ULL" will fix the issue. Fixes: 225c7b1f ("IB/mlx4: Add a driver Mellanox ConnectX InfiniBand adapters") Signed-off-by:
Leon Romanovsky <leonro@mellanox.com> Signed-off-by:
-
Eran Ben Elisha authored
In order to avoid temporary large structs on the stack, allocate them dynamically. Signed-off-by:
Eran Ben Elisha <eranbe@mellanox.com> Signed-off-by:
Tal Alon <talal@mellanox.com> Signed-off-by:
Saeed Mahameed <saeedm@mellanox.com> Signed-off-by:
-
Roopa Prabhu authored
A few useful tracepoints to trace bridge forwarding database updates. Signed-off-by:
Roopa Prabhu <roopa@cumulusnetworks.com> Reviewed-by:
Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Signed-off-by:
-
David S. Miller authored
Florian Fainelli says: ==================== Endian fixes for SYSTEMPORT/SF2/MDIO While trying an ARM BE kernel for kinks, the 3 drivers below started not working and the reasons why became pretty obvious because the register space remains LE (hardwired), except for Broadcom MIPS where it follows the CPU's native endian (let's call that a feature). ==================== Signed-off-by: -
Florian Fainelli authored
The driver currently uses __raw_{read,write}l which works for all platforms supported: Broadcom MIPS LE/BE (native endian), ARM LE (native endian) but not ARM BE (registers are still LE). Switch to using the proper accessors for all platforms and explain why Broadcom MIPS BE is special here, in doing so, we introduce a couple of helper functions to abstract these differences. Signed-off-by: -
Florian Fainelli authored
RSB_SWAP0 needs to match the host CPU endian, and it needs to be set for LE and clear for BE. RSB_SWAP1 must always be cleared for SYSTEMPORT Lite. With these settings, we have the Receive Status Block always match the host endian and we do not need to perform any conversion. Since there is not necessarily a CONFIG_CPU_LITTLE_ENDIAN option defined, we test for !CONFIG_CPU_BIG_ENDIAN which is guaranteed to be set. Signed-off-by:
-
Florian Fainelli authored
The Starfigther 2 driver currently uses __raw_{read,write}l which means native I/O endian. This works correctly for an ARM LE kernel (default) but fails miserably on an ARM BE (BE8) kernel where registers are kept little endian, so replace uses with {read,write}l_relaxed here which is what we want because this is all performance sensitive code. Signed-off-by: -
Florian Fainelli authored
The SYSTEMPORT driver currently uses __raw_{read,write}l which means native I/O endian. This works correctly for an ARM LE kernel (default) but fails miserably on an ARM BE (BE8) kernel where registers are kept little endian, so replace uses with {read,write}l_relaxed here which is what we want because this is all performance sensitive code. Signed-off-by: -
David S. Miller authored
Merge tag 'wireless-drivers-next-for-davem-2017-08-28' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next Kalle Valo says: ==================== wireless-drivers-next patches for 4.14 rsi driver is getting a lot of new features lately, but as usual active development happening on iwlwifi as well as other drivers. I pulled wireless-drivers to fix multiple conflicts in iwlwifi and to make it easier further development. Major changes: ath10k * initial UBS bus support (no full support yet) * add tdls support for 10.4 firmware ath9k * add Dell Wireless 1802 wil6210 * support FW RSSI reporting rsi * support legacy power save, U-APSD, rf-kill and AP mode * RTS threshold configuration brcmfmac * support CYW4373 SDIO/USB chipset iwlwifi * some more code moved to a new directory * add new PCI ID for 7265D ==================== Signed-off-by:
-
Arvind Yadav authored
clk_div_table are not supposed to change at runtime. meson8b_dwmac structure is working with const clk_div_table. So mark the non-const structs as const. Signed-off-by:
Arvind Yadav <arvind.yadav.cs@gmail.com> Signed-off-by:
-
David S. Miller authored
Jesper Dangaard Brouer says: ==================== XDP redirect tracepoints I feel this is as far as I can take the tracepoint infrastructure to assist XDP monitoring. Tracepoints comes with a base overhead of 25 nanosec for an attached bpf_prog, and 48 nanosec for using a full perf record. This is problematic for the XDP use-case, but it is very convenient to use the existing perf infrastructure. From a performance perspective, the real solution would be to attach another bpf_prog (that understand xdp_buff), but I'm not sure we want to introduce yet another bpf attach API for this. One thing left is to standardize the possible err return codes, to a limited set, to allow easier (and faster) mapping into a bpf map. ==================== Signed-off-by: -
Jesper Dangaard Brouer authored
This tool xdp_monitor demonstrate how to use the different xdp_redirect tracepoints xdp_redirect{,_map}{,_err} from a BPF program. The default mode is to only monitor the error counters, to avoid affecting the per packet performance. Tracepoints comes with a base overhead of 25 nanosec for an attached bpf_prog, and 48 nanosec for using a full perf record (with non-matching filter). Thus, default loading the --stats mode could affect the maximum performance. This version of the tool is very simple and count all types of errors as one. It will be natural to extend this later with the different types of errors that can occur, which should help users quickly identify common mistakes. Because the TP_STRUCT was kept in sync all the tracepoints loads the same BPF code. It would also be natural to extend the map version to demonstrate how the map information could be used. Signed-off-by:Jesper Dangaard Brouer <brouer@redhat.com> Acked-by:
Alexei Starovoitov <ast@kernel.org> Signed-off-by:
-
Jesper Dangaard Brouer authored
For supporting XDP_REDIRECT, a device driver must (obviously) implement the "TX" function ndo_xdp_xmit(). An additional requirement is you cannot TX out a device, unless it also have a xdp bpf program attached. This dependency is caused by the driver code need to setup XDP resources before it can ndo_xdp_xmit. Update bpf samples xdp_redirect and xdp_redirect_map to automatically attach a dummy XDP program to the configured ifindex_out device. Use the XDP flag XDP_FLAGS_UPDATE_IF_NOEXIST on the dummy load, to avoid overriding an existing XDP prog on the device. Signed-off-by:
-
Jesper Dangaard Brouer authored
Creating as specific xdp_redirect_map variant of the xdp tracepoints allow users to write simpler/faster BPF progs that get attached to these tracepoints. Goal is to still keep the tracepoints in xdp_redirect and xdp_redirect_map similar enough, that a tool can read the top part of the TP_STRUCT and produce similar monitor statistics. Signed-off-by:
-
Jesper Dangaard Brouer authored
There is a need to separate the xdp_redirect tracepoint into two tracepoints, for separating the error case from the normal forward case. Due to the extreme speeds XDP is operating at, loading a tracepoint have a measurable impact. Single core XDP REDIRECT (ethtool tuned rx-usecs 25) can do 13.7 Mpps forwarding, but loading a simple bpf_prog at the tracepoint (with a return 0) reduce perf to 10.2 Mpps (CPU E5-1650 v4 @ 3.60GHz, driver: ixgbe) The overhead of loading a bpf-based tracepoint can be calculated to cost 25 nanosec ((1/13782002-1/10267937)*10^9 = -24.83 ns). Using perf record on the tracepoint event, with a non-matching --filter expression, the overhead is much larger. Performance drops to 8.3 Mpps, cost 48 nanosec ((1/13782002-1/8312497)*10^9 = -47.74)) Having a separate tracepoint for err cases, which should be less frequent, allow running a continuous monitor for errors while not affecting the redirect forward performance (this have also been verified by measurements). Signed-off-by:
-
Jesper Dangaard Brouer authored
Given previous patch expose the map_id, it seems natural to also report the bpf prog id. Signed-off-by:
-
