- 08 Nov, 2019 40 commits
-
-
Eric Dumazet authored
KCSAN reported a data-race [1] While we can use READ_ONCE() on the read sides, we need to make sure hh->hh_len is written last. [1] BUG: KCSAN: data-race in eth_header_cache / neigh_resolve_output write to 0xffff8880b9dedcb8 of 4 bytes by task 29760 on cpu 0: eth_header_cache+0xa9/0xd0 net/ethernet/eth.c:247 neigh_hh_init net/core/neighbour.c:1463 [inline] neigh_resolve_output net/core/neighbour.c:1480 [inline] neigh_resolve_output+0x415/0x470 net/core/neighbour.c:1470 neigh_output include/net/neighbour.h:511 [inline] ip6_finish_output2+0x7a2/0xec0 net/ipv6/ip6_output.c:116 __ip6_finish_output net/ipv6/ip6_output.c:142 [inline] __ip6_finish_output+0x2d7/0x330 net/ipv6/ip6_output.c:127 ip6_finish_output+0x41/0x160 net/ipv6/ip6_output.c:152 NF_HOOK_COND include/linux/netfilter.h:294 [inline] ip6_output+0xf2/0x280 net/ipv6/ip6_output.c:175 dst_output include/net/dst.h:436 [inline] NF_HOOK include/linux/netfilter.h:305 [inline] ndisc_send_skb+0x459/0x5f0 net/ipv6/ndisc.c:505 ndisc_send_ns+0x207/0x430 net/ipv6/ndisc.c:647 rt6_probe_deferred+0x98/0xf0 net/ipv6/route.c:615 process_one_work+0x3d4/0x890 kernel/workqueue.c:2269 worker_thread+0xa0/0x800 kernel/workqueue.c:2415 kthread+0x1d4/0x200 drivers/block/aoe/aoecmd.c:1253 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:352 read to 0xffff8880b9dedcb8 of 4 bytes by task 29572 on cpu 1: neigh_resolve_output net/core/neighbour.c:1479 [inline] neigh_resolve_output+0x113/0x470 net/core/neighbour.c:1470 neigh_output include/net/neighbour.h:511 [inline] ip6_finish_output2+0x7a2/0xec0 net/ipv6/ip6_output.c:116 __ip6_finish_output net/ipv6/ip6_output.c:142 [inline] __ip6_finish_output+0x2d7/0x330 net/ipv6/ip6_output.c:127 ip6_finish_output+0x41/0x160 net/ipv6/ip6_output.c:152 NF_HOOK_COND include/linux/netfilter.h:294 [inline] ip6_output+0xf2/0x280 net/ipv6/ip6_output.c:175 dst_output include/net/dst.h:436 [inline] NF_HOOK include/linux/netfilter.h:305 [inline] ndisc_send_skb+0x459/0x5f0 net/ipv6/ndisc.c:505 ndisc_send_ns+0x207/0x430 net/ipv6/ndisc.c:647 rt6_probe_deferred+0x98/0xf0 net/ipv6/route.c:615 process_one_work+0x3d4/0x890 kernel/workqueue.c:2269 worker_thread+0xa0/0x800 kernel/workqueue.c:2415 kthread+0x1d4/0x200 drivers/block/aoe/aoecmd.c:1253 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:352 Reported by Kernel Concurrency Sanitizer on: CPU: 1 PID: 29572 Comm: kworker/1:4 Not tainted 5.4.0-rc6+ #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Workqueue: events rt6_probe_deferred Signed-off-by:
Eric Dumazet <edumazet@google.com> Reported-by:
syzbot <syzkaller@googlegroups.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Eric Dumazet says: ==================== net: introduce u64_stats_t KCSAN found a data-race in per-cpu u64 stats accounting. (The stack traces are included in the 8th patch : tun: switch to u64_stats_t) This patch series first consolidate code in five patches. Then the last three patches address the data-race resolution. ==================== Signed-off-by: -
Eric Dumazet authored
In order to fix the data-race found by KCSAN, we can use the new u64_stats_t type and its accessors instead of plain u64 fields. This will still generate optimal code for both 32 and 64 bit platforms. Signed-off-by:
-
Eric Dumazet authored
In order to fix this data-race found by KCSAN [1], switch to u64_stats_t helpers. They provide all the needed annotations, without adding extra cost. [1] BUG: KCSAN: data-race in tun_get_user / tun_net_get_stats64 read to 0xffffe8ffffd8aca8 of 8 bytes by task 4882 on cpu 0: tun_net_get_stats64+0x9b/0x230 drivers/net/tun.c:1171 dev_get_stats+0x89/0x1e0 net/core/dev.c:9103 rtnl_fill_stats+0x56/0x370 net/core/rtnetlink.c:1177 rtnl_fill_ifinfo+0xd3b/0x2100 net/core/rtnetlink.c:1667 rtmsg_ifinfo_build_skb+0xb0/0x150 net/core/rtnetlink.c:3472 rtmsg_ifinfo_event.part.0+0x4e/0xb0 net/core/rtnetlink.c:3504 rtmsg_ifinfo_event net/core/rtnetlink.c:3515 [inline] rtmsg_ifinfo+0x85/0x90 net/core/rtnetlink.c:3513 __dev_notify_flags+0x18b/0x200 net/core/dev.c:7649 dev_change_flags+0xb8/0xe0 net/core/dev.c:7691 dev_ifsioc+0x201/0x6a0 net/core/dev_ioctl.c:237 dev_ioctl+0x149/0x660 net/core/dev_ioctl.c:489 sock_do_ioctl+0xdb/0x230 net/socket.c:1061 sock_ioctl+0x3a3/0x5e0 net/socket.c:1189 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:509 [inline] do_vfs_ioctl+0x991/0xc60 fs/ioctl.c:696 write to 0xffffe8ffffd8aca8 of 8 bytes by task 4883 on cpu 1: tun_get_user+0x1d94/0x2ba0 drivers/net/tun.c:2002 tun_chr_write_iter+0x79/0xd0 drivers/net/tun.c:2022 call_write_iter include/linux/fs.h:1895 [inline] new_sync_write+0x388/0x4a0 fs/read_write.c:483 __vfs_write+0xb1/0xc0 fs/read_write.c:496 __kernel_write+0xb8/0x240 fs/read_write.c:515 write_pipe_buf+0xb6/0xf0 fs/splice.c:794 splice_from_pipe_feed fs/splice.c:500 [inline] __splice_from_pipe+0x248/0x480 fs/splice.c:624 splice_from_pipe+0xbb/0x100 fs/splice.c:659 default_file_splice_write+0x45/0x90 fs/splice.c:806 do_splice_from fs/splice.c:848 [inline] direct_splice_actor+0xa0/0xc0 fs/splice.c:1020 splice_direct_to_actor+0x215/0x510 fs/splice.c:975 do_splice_direct+0x161/0x1e0 fs/splice.c:1063 do_sendfile+0x384/0x7f0 fs/read_write.c:1464 Reported by Kernel Concurrency Sanitizer on: CPU: 1 PID: 4883 Comm: syz-executor.1 Not tainted 5.4.0-rc3+ #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Signed-off-by:
-
Eric Dumazet authored
On 64bit arches, struct u64_stats_sync is empty and provides no help against load/store tearing. Using READ_ONCE()/WRITE_ONCE() would be needed. But the update side would be slightly more expensive. local64_t was defined so that we could use regular adds in a manner which is atomic wrt IRQs. However the u64_stats infra means we do not have to use local64_t on 32bit arches since the syncp provides the needed protection. Signed-off-by:
-
Eric Dumazet authored
This driver can simply use the common infrastructure instead of duplicating it. This cleanup will ease u64_stats_t adoption in a single location. Signed-off-by:
-
Eric Dumazet authored
This cleanup will ease u64_stats_t adoption in a single location. Signed-off-by:
-
Eric Dumazet authored
This cleanup will ease u64_stats_t adoption in a single location. Signed-off-by:
-
Eric Dumazet authored
No need to hand-code the exact same functions. Signed-off-by:
-
Eric Dumazet authored
Many network drivers need it and hand-coded the same function. In order to ease u64_stats_t adoption, it is time to factorize. Signed-off-by:
-
Eric Dumazet authored
Many network drivers use hand-coded implementation of the same thing, let's factorize things so that u64_stats_t adoption is done once. Signed-off-by:
-
David S. Miller authored
Florian Fainelli says: ==================== net: Demote MTU change prints to debug This patch series demotes several drivers that printed MTU change and could therefore spam the kernel console if one has a test that it's all about testing the values. Intel drivers were not also particularly consistent in how they printed the same message, so now they are. ==================== Signed-off-by: -
Florian Fainelli authored
Changing the MTU can be a frequent operation and it is already clear when (or not) a MTU change is successful, demote prints to debug prints. Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Acked-by:
Timur Tabi <timur@kernel.org> Signed-off-by:
-
Florian Fainelli authored
Changing a network device MTU can be a fairly frequent operation, and failure to change the MTU is reflected to user-space properly, both by an appropriate message as well as by looking at whether the device's MTU matches the configuration. Demote the prints to debug prints by using netdev_dbg(), making all Intel wired LAN drivers consistent, since they used a mixture of PCI device and network device prints before. Signed-off-by:
Jeff Kirsher <jeffrey.t.kirsher@intel.com> Signed-off-by:
-
Ivan Khoronzhuk authored
Update on more short variant for getting real clock in ns. Signed-off-by:
Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org> Signed-off-by:
-
David S. Miller authored
Igor Russkikh says: ==================== Aquantia Marvell atlantic driver updates 11-2019 Here is a bunch of atlantic driver new features and updates. Shortlist: - Me adding ethtool private flags for various loopback test modes, - Nikita is doing some work here on power management, implementing new PM API, He also did some checkpatch style cleanup of older driver parts. - I'm also adding a new UDP GSO offload support and flags for loopback activation - We are now Marvell, so I am changing email addresses on maintainers list. v2: styling, ip6 correct handling in udpgso ==================== Signed-off-by: -
Igor Russkikh authored
Aquantia is now part of Marvell, eventually we'll cease standalone aquantia.com domain. Thus, change the maintainers file and some other references to @marvell.com domain Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
-
Igor Russkikh authored
atlantic hardware does support UDP hardware segmentation offload. This allows user to specify one large contiguous buffer with data which then will be split automagically into multiple UDP packets of specified size. Bulk sending of large UDP streams lowers CPU usage and increases bandwidth. We did estimations both with udpgso_bench_tx test tool and with modified iperf3 measurement tool (4 streams, multithread, 200b packet size) over AQC<->AQC 10G link. Flow control is disabled to prevent RX side impact on measurements. No UDP GSO: iperf3 -c 10.0.1.2 -u -b0 -l 200 -P4 --multithread UDP GSO: iperf3 -c 10.0.1.2 -u -b0 -l 12600 --udp-lso 200 -P4 --multithread Mode CPU iperf speed Line speed Packets per second ------------------------------------------------------------- NO UDP GSO 350% 3.07 Gbps 3.8 Gbps 1,919,419 SW UDP GSO 200% 5.55 Gbps 6.4 Gbps 3,286,144 HW UDP GSO 90% 6.80 Gbps 8.4 Gbps 4,273,117 Signed-off-by:
-
Nikita Danilov authored
We now differentiate requested and negotiated flow control modes. Therefore `ethtool -A` now operates on local requested FC values, and regular link settings shows the negotiated FC settings. Signed-off-by:
Nikita Danilov <ndanilov@marvell.com> Signed-off-by:
-
Igor Russkikh authored
We are trying to follow the naming of the chip (atlantic), not company. So replace some old namings. Signed-off-by:
-
Nikita Danilov authored
Thats a pure checkpatck walkthrough the code with no functional changes. Reverse christmas tree, spacing, etc. Signed-off-by:
-
Igor Russkikh authored
Here we add a number of ethtool private flags to allow enabling various loopbacks on HW. Thats useful for verification and bringup works. Signed-off-by:
-
Nikita Danilov authored
Device FW has a separate memory area where various config fields are stored and could be used by the driver. Here we modify download/upload infrastructure to allow accessing this area. Lateron this will be used to configure various behaviours Signed-off-by:
-
Nikita Danilov authored
`ethtool -p eth0` will blink leds helping identify physical port. Signed-off-by:
-
Nikita Danilov authored
We add ethtool msglevel configuration and change some printouts to use netdev_info set of functions. Signed-off-by:
-
Nikita Danilov authored
We now implement .driver.pm callbacks, these allows driver to work correctly in hibernate usecases, especially when used in conjunction with WOL feature. Before that driver only reacted to legacy .suspend/.resume callbacks, that was a limitation in some cases. Signed-off-by:
-
Nikita Danilov authored
Wake on PHY allows to configure device to wakeup host as soon as PHY link status is changed to active. Signed-off-by:
-
Nikita Danilov authored
Here we improve FW interface structures layout and prepare these for the wake phy feature implementation. Signed-off-by:
-
David S. Miller authored
Ido Schimmel says: ==================== mlxsw: Add layer 3 devlink-trap support This patch set from Amit adds support in mlxsw for layer 3 traps that can report drops and exceptions via devlink-trap. In a similar fashion to the existing layer 2 traps, these traps can send packets to the CPU that were not routed as intended by the underlying device. The traps are divided between the two types detailed in devlink-trap documentation: drops and exceptions. Unlike drops, packets received via exception traps are also injected to the kernel's receive path, as they are required for the correct functioning of the control plane. For example, packets trapped due to TTL error must be injected to kernel's receive path for traceroute to work properly. Patch set overview: Patch #1 adds the layer 3 drop traps to devlink along with their documentation. Patch #2 adds support for layer 3 drop traps in mlxsw. Patches #3-#5 add selftests for layer 3 drop traps. Patch #6 adds the layer 3 exception traps to devlink along with their documentation. Patches #7-#9 gradually add support for layer 3 exception traps in mlxsw. Patches #10-#12 add selftests for layer 3 exception traps. ==================== Signed-off-by: -
Amit Cohen authored
Test that each supported packet trap exception is triggered under the right conditions. Signed-off-by:
Amit Cohen <amitc@mellanox.com> Acked-by:
Jiri Pirko <jiri@mellanox.com> Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Signed-off-by:
-
Amit Cohen authored
Add an option to check that packets hit the tc filter without providing the exact number of packets that should hit it. It is useful while sending many packets in background and checking that at least one of them hit the tc filter. Signed-off-by:
-
Amit Cohen authored
Add common part of all the tests - check devlink status to ensure that packets were trapped. Signed-off-by:
-
Amit Cohen authored
Add the trap IDs used to report layer 3 exceptions. Trapped packets are first reported to devlink and then injected to the kernel's receive path. All the packets have 'offload_fwd_mark' set in order to prevent them from potentially being forwarded by the bridge again. Signed-off-by:
-
Amit Cohen authored
Currently, mlxsw does not differentiate between these two cases of routes with invalid nexthops: 1. Nexthops whose nexthop device is a mlxsw upper (has a RIF), but whose neighbour could not be resolved 2. Nexthops whose nexthop device is not a mlxsw upper (e.g., management interface) Up until now this did not matter and mlxsw trapped packets for both cases using the same trap ID. However, packets that should have been routed in hardware (case 1), but incurred a problem are considered exceptions and should be reported to the user. The two cases should therefore be split between two different trap IDs. Allocate a new adjacency entry during initialization and upon the insertion of the first route with an invalid mlxsw nexthop, program this entry to discard packets. Packets hitting this entry will be reported using new trap ID - "DISCARD_ROUTER3". In the future, the entry could be written during initialization, but currently firmware requires a valid RIF, which is not available at this stage. Signed-off-by:
-
Amit Cohen authored
Currently, packets that cannot be routed in hardware (e.g., nexthop device is not upper of mlxsw), are trapped to the kernel for forwarding. Such packets are trapped using "RTR_INGRESS0" trap. This trap also traps packets that hit reject routes (e.g., "unreachable") so that the kernel will generate the appropriate ICMP error message for them. Subsequent patch will need to only report to devlink packets that hit a reject route, which is impossible as long as "RTR_INGRESS0" is overloaded like that. Solve this by using "RTR_INGRESS1" trap for packets that hit reject routes. Signed-off-by:
-
Amit Cohen authored
Add layer 3 generic packet exception traps that can report trapped packets and documentation of the traps. Unlike drop traps, these exception traps also need to inject the packet to the kernel's receive path. For example, a packet that was trapped due to unreachable neighbour need to be injected into the kernel so that it will trigger an ARP request or a neighbour solicitation message. Signed-off-by:
-
Amit Cohen authored
Test that each supported packet trap is triggered under the right conditions and that packets are indeed dropped and not forwarded. Signed-off-by:
-
Amit Cohen authored
Add proto parameter in order to enable the use of devlink_trap_cleanup() in tests that use IPv6 protocol. Signed-off-by:
-
Amit Cohen authored
l2_drops_test() is used to check that drop traps are functioning as intended. Currently it is only used in the layer 2 test, but it is also useful for the layer 3 test introduced in the subsequent patch. l2_drops_cleanup() is used to clean configurations and kill mausezahn proccess. Export the functions to the common devlink library to allow it to be re-used by future tests. Signed-off-by:
-
Amit Cohen authored
Add the trap IDs and trap group used to report layer 3 drops. Register layer 3 packet traps and associated layer 3 trap group with devlink during driver initialization. Signed-off-by:
-
