openwrt/staging/blogic.git
7 years agonet: Remove NET_CORE_BUDGET_USECS from sysctl binary interface.
David S. Miller [Fri, 21 Apr 2017 19:59:52 +0000 (15:59 -0400)]
net: Remove NET_CORE_BUDGET_USECS from sysctl binary interface.

We are not supposed to add new entries to this thing
any more.

Thanks to Eric Dumazet for noticing this.

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: LAN9303: add I2C dependency
Arnd Bergmann [Fri, 21 Apr 2017 16:22:40 +0000 (18:22 +0200)]
net: dsa: LAN9303: add I2C dependency

With CONFIG_I2C=m and NET_DSA_SMSC_LAN9303=y, we run into a link error:

drivers/base/regmap/regmap-i2c.o: In function `regmap_smbus_byte_reg_read':
regmap-i2c.c:(.text.regmap_smbus_byte_reg_read+0x18): undefined reference to `i2c_smbus_read_byte_data'
drivers/base/regmap/regmap-i2c.o: In function `regmap_smbus_byte_reg_write':
regmap-i2c.c:(.text.regmap_smbus_byte_reg_write+0x18): undefined reference to `i2c_smbus_write_byte_data'
drivers/base/regmap/regmap-i2c.o: In function `regmap_smbus_word_reg_read':
regmap-i2c.c:(.text.regmap_smbus_word_reg_read+0x18): undefined reference to `i2c_smbus_read_word_data'
drivers/base/regmap/regmap-i2c.o: In function `regmap_smbus_word_read_swapped':
regmap-i2c.c:(.text.regmap_smbus_word_read_swapped+0x18): undefined reference to `i2c_smbus_read_word_data'
drivers/base/regmap/regmap-i2c.o: In function `regmap_smbus_word_write_swapped':

This adds a Kconfig dependency to avoid the broken configuration.

Fixes: be4e119f9914 ("net: dsa: LAN9303: add I2C managed mode support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'nfc-next-4.12-1' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo...
David S. Miller [Fri, 21 Apr 2017 19:29:40 +0000 (15:29 -0400)]
Merge tag 'nfc-next-4.12-1' of git://git./linux/kernel/git/sameo/nfc-next

Samuel Ortiz says:

====================
NFC 4.12 pull request

This is the NFC pull request for 4.12. We have:

- Improvements for the pn533 command queue handling and device
  registration order.
- Removal of platform data for the pn544 and st21nfca drivers.
- Additional device tree options to support more trf7970a hardware options.
- Support for Sony's RC-S380P through the port100 driver.
- Removal of the obsolte nfcwilink driver.
- Headers inclusion cleanups (miscdevice.h, unaligned.h) for many drivers.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobonding: fix wq initialization for links created via netlink
Mahesh Bandewar [Thu, 20 Apr 2017 19:49:24 +0000 (12:49 -0700)]
bonding: fix wq initialization for links created via netlink

Earlier patch 4493b81bea ("bonding: initialize work-queues during
creation of bond") moved the work-queue initialization from bond_open()
to bond_create(). However this caused the link those are created using
netlink 'create bond option' (ip link add bondX type bond); create the
new trunk without initializing work-queues. Prior to the above mentioned
change, ndo_open was in both paths and things worked correctly. The
consequence is visible in the report shared by Joe Stringer -

I've noticed that this patch breaks bonding within namespaces if
you're not careful to perform device cleanup correctly.

Here's my repro script, you can run on any net-next with this patch
and you'll start seeing some weird behaviour:

ip netns add foo
ip li add veth0 type veth peer name veth0+ netns foo
ip li add veth1 type veth peer name veth1+ netns foo
ip netns exec foo ip li add bond0 type bond
ip netns exec foo ip li set dev veth0+ master bond0
ip netns exec foo ip li set dev veth1+ master bond0
ip netns exec foo ip addr add dev bond0 192.168.0.1/24
ip netns exec foo ip li set dev bond0 up
ip li del dev veth0
ip li del dev veth1

The second to last command segfaults, last command hangs. rtnl is now
permanently locked. It's not a problem if you take bond0 down before
deleting veths, or delete bond0 before deleting veths. If you delete
either end of the veth pair as per above, either inside or outside the
namespace, it hits this problem.

Here's some kernel logs:
[ 1221.801610] bond0: Enslaving veth0+ as an active interface with an up link
[ 1224.449581] bond0: Enslaving veth1+ as an active interface with an up link
[ 1281.193863] bond0: Releasing backup interface veth0+
[ 1281.193866] bond0: the permanent HWaddr of veth0+ -
16:bf:fb:e0:b8:43 - is still in use by bond0 - set the HWaddr of
veth0+ to a different address to avoid conflicts
[ 1281.193867] ------------[ cut here ]------------
[ 1281.193873] WARNING: CPU: 0 PID: 2024 at kernel/workqueue.c:1511
__queue_delayed_work+0x13f/0x150
[ 1281.193873] Modules linked in: bonding veth openvswitch nf_nat_ipv6
nf_nat_ipv4 nf_nat autofs4 nfsd auth_rpcgss nfs_acl binfmt_misc nfs
lockd grace sunrpc fscache ppdev vmw_balloon coretemp psmouse
serio_raw vmwgfx ttm drm_kms_helper vmw_vmci netconsole parport_pc
configfs drm i2c_piix4 fb_sys_fops syscopyarea sysfillrect sysimgblt
shpchp mac_hid nf_conntrack_ipv6 nf_defrag_ipv6 nf_conntrack_ipv4
nf_defrag_ipv4 nf_conntrack libcrc32c lp parport hid_generic usbhid
hid mptspi mptscsih e1000 mptbase ahci libahci
[ 1281.193905] CPU: 0 PID: 2024 Comm: ip Tainted: G        W
4.10.0-bisect-bond-v0.14 #37
[ 1281.193906] Hardware name: VMware, Inc. VMware Virtual
Platform/440BX Desktop Reference Platform, BIOS 6.00 09/30/2014
[ 1281.193906] Call Trace:
[ 1281.193912]  dump_stack+0x63/0x89
[ 1281.193915]  __warn+0xd1/0xf0
[ 1281.193917]  warn_slowpath_null+0x1d/0x20
[ 1281.193918]  __queue_delayed_work+0x13f/0x150
[ 1281.193920]  queue_delayed_work_on+0x27/0x40
[ 1281.193929]  bond_change_active_slave+0x25b/0x670 [bonding]
[ 1281.193932]  ? synchronize_rcu_expedited+0x27/0x30
[ 1281.193935]  __bond_release_one+0x489/0x510 [bonding]
[ 1281.193939]  ? addrconf_notify+0x1b7/0xab0
[ 1281.193942]  bond_netdev_event+0x2c5/0x2e0 [bonding]
[ 1281.193944]  ? netconsole_netdev_event+0x124/0x190 [netconsole]
[ 1281.193947]  notifier_call_chain+0x49/0x70
[ 1281.193948]  raw_notifier_call_chain+0x16/0x20
[ 1281.193950]  call_netdevice_notifiers_info+0x35/0x60
[ 1281.193951]  rollback_registered_many+0x23b/0x3e0
[ 1281.193953]  unregister_netdevice_many+0x24/0xd0
[ 1281.193955]  rtnl_delete_link+0x3c/0x50
[ 1281.193956]  rtnl_dellink+0x8d/0x1b0
[ 1281.193960]  rtnetlink_rcv_msg+0x95/0x220
[ 1281.193962]  ? __kmalloc_node_track_caller+0x35/0x280
[ 1281.193964]  ? __netlink_lookup+0xf1/0x110
[ 1281.193966]  ? rtnl_newlink+0x830/0x830
[ 1281.193967]  netlink_rcv_skb+0xa7/0xc0
[ 1281.193969]  rtnetlink_rcv+0x28/0x30
[ 1281.193970]  netlink_unicast+0x15b/0x210
[ 1281.193971]  netlink_sendmsg+0x319/0x390
[ 1281.193974]  sock_sendmsg+0x38/0x50
[ 1281.193975]  ___sys_sendmsg+0x25c/0x270
[ 1281.193978]  ? mem_cgroup_commit_charge+0x76/0xf0
[ 1281.193981]  ? page_add_new_anon_rmap+0x89/0xc0
[ 1281.193984]  ? lru_cache_add_active_or_unevictable+0x35/0xb0
[ 1281.193985]  ? __handle_mm_fault+0x4e9/0x1170
[ 1281.193987]  __sys_sendmsg+0x45/0x80
[ 1281.193989]  SyS_sendmsg+0x12/0x20
[ 1281.193991]  do_syscall_64+0x6e/0x180
[ 1281.193993]  entry_SYSCALL64_slow_path+0x25/0x25
[ 1281.193995] RIP: 0033:0x7f6ec122f5a0
[ 1281.193995] RSP: 002b:00007ffe69e89c48 EFLAGS: 00000246 ORIG_RAX:
000000000000002e
[ 1281.193997] RAX: ffffffffffffffda RBX: 00007ffe69e8dd60 RCX: 00007f6ec122f5a0
[ 1281.193997] RDX: 0000000000000000 RSI: 00007ffe69e89c90 RDI: 0000000000000003
[ 1281.193998] RBP: 00007ffe69e89c90 R08: 0000000000000000 R09: 0000000000000003
[ 1281.193999] R10: 00007ffe69e89a10 R11: 0000000000000246 R12: 0000000058f14b9f
[ 1281.193999] R13: 0000000000000000 R14: 00000000006473a0 R15: 00007ffe69e8e450
[ 1281.194001] ---[ end trace 713a77486cbfbfa3 ]---

Fixes: 4493b81bea ("bonding: initialize work-queues during creation of bond")
Reported-by: Joe Stringer <joe@ovn.org>
Tested-by: Joe Stringer <joe@ovn.org>
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Acked-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: arc_emac: switch to phy_start()/phy_stop()
Alexander Kochetkov [Thu, 20 Apr 2017 13:29:34 +0000 (16:29 +0300)]
net: arc_emac: switch to phy_start()/phy_stop()

Currently driver use phy_start_aneg() in arc_emac_open() to bring
up PHY. But phy_start() function is more appropriate for this purposes.
Besides that it call phy_start_aneg() as part of PHY startup sequence
it also can correctly bring up PHY from error and suspended states.
So the patch replace phy_start_aneg() to phy_start().

Also the patch add call to phy_stop() to arc_emac_stop() to allow
the PHY device to be fully suspended when the interface is unused.

Signed-off-by: Alexander Kochetkov <al.kochet@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec...
David S. Miller [Fri, 21 Apr 2017 19:11:28 +0000 (15:11 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
pull request (net-next): ipsec-next 2017-04-20

This adds the basic infrastructure for IPsec hardware
offloading, it creates a configuration API and adjusts
the packet path.

1) Add the needed netdev features to configure IPsec offloads.

2) Add the IPsec hardware offloading API.

3) Prepare the ESP packet path for hardware offloading.

4) Add gso handlers for esp4 and esp6, this implements
   the software fallback for GSO packets.

5) Add xfrm replay handler functions for offloading.

6) Change ESP to use a synchronous crypto algorithm on
   offloading, we don't have the option for asynchronous
   returns when we handle IPsec at layer2.

7) Add a xfrm validate function to validate_xmit_skb. This
   implements the software fallback for non GSO packets.

8) Set the inner_network and inner_transport members of
   the SKB, as well as encapsulation, to reflect the actual
   positions of these headers, and removes them only once
   encryption is done on the payload.
   From Ilan Tayari.

9) Prepare the ESP GRO codepath for hardware offloading.

10) Fix incorrect null pointer check in esp6.
    From Colin Ian King.

11) Fix for the GSO software fallback path to detect the
    fallback correctly.
    From Ilan Tayari.

Please pull or let me know if there are problems.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMAINTAINERS: Add new IPsec offloading files.
Steffen Klassert [Thu, 20 Apr 2017 08:36:32 +0000 (10:36 +0200)]
MAINTAINERS: Add new IPsec offloading files.

This adds two new files to IPsec maintenance scope:

net/ipv4/esp4_offload.c
net/ipv6/ip6_offload.c

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Fri, 21 Apr 2017 18:13:11 +0000 (14:13 -0400)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2017-04-19

This series contains updates to i40e and i40evf only, most notable being
the addition of trace points for BPF programs.

Tobias Klauser updates i40evf to use net_device stats struct instead
of a local private copy.

Preethi updates the VF driver to not enable receive checksum offload by
default for tunneled packets.

Alex fixes an issue he introduced when he converted the code over to
using the length field to determine if a descriptor was done or not.

Mitch adds the ability to dump additional information on the VFs, which
is not available through 'ip link show' using debugfs.

Scott adds trace points to the drivers so that BPF programs can be
attached for feature testing and verification.

Jingjing adds admin queue functions for Pipeline Personalization Profile
commands.

Jake does most of the heavy lifting in this series, starting with the
a reduction in the scope of the RTNL lock being held while resetting VFs
to allow multiple PFs to reset in a timely manner.  Factored out the
direct queue modification so that we are able to re-use the code.
Reduced the wait time for admin queue commands to complete, since we were
waiting a minimum of a millisecond, when in practice the admin queue
command is processed often much faster.  Cleaned up code (flag) we never
use.  Make the code to resetting all the VFs optimized for parallel
computing instead of the current way is a serialized fashion, to help
reduce the time it takes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetvsc: fix use after free on module removal
stephen hemminger [Wed, 19 Apr 2017 22:22:02 +0000 (15:22 -0700)]
netvsc: fix use after free on module removal

The NAPI data structure is embedded in the netvsc_device structure
and is freed when device is closed. There is still a reference
(in NAPI list) to this which causes a crash in netif_napi_del
when device is removed. Fix by managing NAPI instances correctly.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'tc-filter-cleanup-destroy-delete'
David S. Miller [Fri, 21 Apr 2017 17:58:16 +0000 (13:58 -0400)]
Merge branch 'tc-filter-cleanup-destroy-delete'

Cong Wang says:

====================
net_sched: clean up tc filter destroy and delete logic

The first patch fixes a potenial race condition, the second one
is pure cleanup.
====================

Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: remove useless NULL to tp->root
WANG Cong [Wed, 19 Apr 2017 21:21:22 +0000 (14:21 -0700)]
net_sched: remove useless NULL to tp->root

There is no need to NULL tp->root in ->destroy(), since tp is
going to be freed very soon, and existing readers are still
safe to read them.

For cls_route, we always init its tp->root, so it can't be NULL,
we can drop more useless code.

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet_sched: move the empty tp check from ->destroy() to ->delete()
WANG Cong [Wed, 19 Apr 2017 21:21:21 +0000 (14:21 -0700)]
net_sched: move the empty tp check from ->destroy() to ->delete()

We could have a race condition where in ->classify() path we
dereference tp->root and meanwhile a parallel ->destroy() makes it
a NULL. Daniel cured this bug in commit d936377414fa
("net, sched: respect rcu grace period on cls destruction").

This happens when ->destroy() is called for deleting a filter to
check if we are the last one in tp, this tp is still linked and
visible at that time. The root cause of this problem is the semantic
of ->destroy(), it does two things (for non-force case):

1) check if tp is empty
2) if tp is empty we could really destroy it

and its caller, if cares, needs to check its return value to see if it
is really destroyed. Therefore we can't unlink tp unless we know it is
empty.

As suggested by Daniel, we could actually move the test logic to ->delete()
so that we can safely unlink tp after ->delete() tells us the last one is
just deleted and before ->destroy().

Fixes: 1e052be69d04 ("net_sched: destroy proto tp when all filters are gone")
Cc: Roi Dayan <roid@mellanox.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: add napi_id read access to __sk_buff
Daniel Borkmann [Wed, 19 Apr 2017 21:01:17 +0000 (23:01 +0200)]
bpf: add napi_id read access to __sk_buff

Add napi_id access to __sk_buff for socket filter program types, tc
program types and other bpf_convert_ctx_access() users. Having access
to skb->napi_id is useful for per RX queue listener siloing, f.e.
in combination with SO_ATTACH_REUSEPORT_EBPF and when busy polling is
used, meaning SO_REUSEPORT enabled listeners can then select the
corresponding socket at SYN time already [1]. The skb is marked via
skb_mark_napi_id() early in the receive path (e.g., napi_gro_receive()).

Currently, sockets can only use SO_INCOMING_NAPI_ID from 6d4339028b35
("net: Introduce SO_INCOMING_NAPI_ID") as a socket option to look up
the NAPI ID associated with the queue for steering, which requires a
prior sk_mark_napi_id() after the socket was looked up.

Semantics for the __sk_buff napi_id access are similar, meaning if
skb->napi_id is < MIN_NAPI_ID (e.g. outgoing packets using sender_cpu),
then an invalid napi_id of 0 is returned to the program, otherwise a
valid non-zero napi_id.

  [1] http://netdevconf.org/2.1/slides/apr6/dumazet-BUSY-POLLING-Netdev-2.1.pdf

Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetvsc: Deal with rescinded channels correctly
K. Y. Srinivasan [Wed, 19 Apr 2017 20:53:49 +0000 (13:53 -0700)]
netvsc: Deal with rescinded channels correctly

We will not be able to send packets over a channel that has been
rescinded. Make necessary adjustments so we can properly cleanup
even when the channel is rescinded. This issue can be trigerred
in the NIC hot-remove path.

Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'ibmvnic-updates-and-bug-fixes'
David S. Miller [Fri, 21 Apr 2017 17:33:55 +0000 (13:33 -0400)]
Merge branch 'ibmvnic-updates-and-bug-fixes'

Nathan Fontenot says:

====================
ibmvnic: Updates and bug fixes

This set of patches is a series of updates to remove some unneeded
and unused code in the driver as well as bug fixes for the
ibmvnic driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Remove unused bouce buffer
Nathan Fontenot [Wed, 19 Apr 2017 17:45:22 +0000 (13:45 -0400)]
ibmvnic: Remove unused bouce buffer

The bounce buffer is not used in the ibmvnic driver, just
get rid of it.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Allocate zero-filled memory for sub crqs
Nathan Fontenot [Wed, 19 Apr 2017 17:45:16 +0000 (13:45 -0400)]
ibmvnic: Allocate zero-filled memory for sub crqs

Update the allocation of memory for the sub crq structs and their
associated pages to allocate zero-filled memory.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Disable irq prior to close
Brian King [Wed, 19 Apr 2017 17:45:10 +0000 (13:45 -0400)]
ibmvnic: Disable irq prior to close

    Add some code to call disable_irq on all the vnic interface's irqs.
    This fixes a crash observed when closing an active interface, as
    seen in the oops below when we try to access a buffer in the interrupt
    handler which we've already freed.

    Unable to handle kernel paging request for data at address 0x00000001
    Faulting instruction address: 0xd000000003886824
    Oops: Kernel access of bad area, sig: 11 [#1]
    SMP NR_CPUS=2048 NUMA pSeries
    Modules linked in: ibmvnic(OEN) rpadlpar_io(X) rpaphp(X) tcp_diag udp_diag inet_diag unix_diag af_packet_diag netlink_diag rpcsec_
    Supported: No, Unsupported modules are loaded
    CPU: 8 PID: 0 Comm: swapper/8 Tainted: G           OE   NX 4.4.49-92.11-default #1
    task: c00000007f990110 ti: c0000000fffa0000 task.ti: c00000007f9b8000
    NIP: d000000003886824 LR: d000000003886824 CTR: c0000000007eff60
    REGS: c0000000fffa3a70 TRAP: 0300   Tainted: G           OE   NX  (4.4.49-92.11-default)
    MSR: 8000000000009033 <SF,EE,ME,IR,DR,RI,LE>  CR: 22008042  XER: 20000008
    CFAR: c000000000008468 DAR: 0000000000000001 DSISR: 40000000 SOFTE: 0
    GPR00: d000000003886824 c0000000fffa3cf0 d000000003894118 0000000000000000
    GPR04: 0000000000000000 0000000000000000 c000000001249da0 0000000000000000
    GPR08: 000000000000000e 0000000000000000 c0000000ccb00000 d000000003889180
    GPR12: c0000000007eff60 c000000007af4c00 0000000000000001 c0000000010def30
    GPR16: c00000007f9b8000 c000000000b98c30 c00000007f9b8080 c000000000bab858
    GPR20: 0000000000000005 0000000000000000 c0000000ff5d7e80 c0000000f809f648
    GPR24: c0000000ff5d7ec8 0000000000000000 0000000000000000 c0000000ccb001a0
    GPR28: 000000000000000a c0000000f809f600 c0000000fd4cd900 c0000000f9cd5b00
    NIP [d000000003886824] ibmvnic_interrupt_tx+0x114/0x380 [ibmvnic]
    LR [d000000003886824] ibmvnic_interrupt_tx+0x114/0x380 [ibmvnic]
    Call Trace:
    [c0000000fffa3cf0] [d000000003886824] ibmvnic_interrupt_tx+0x114/0x380 [ibmvnic] (unreliable)
    [c0000000fffa3dd0] [c000000000132940] __handle_irq_event_percpu+0x90/0x2e0
    [c0000000fffa3e90] [c000000000132bcc] handle_irq_event_percpu+0x3c/0x90
    [c0000000fffa3ed0] [c000000000132c88] handle_irq_event+0x68/0xc0
    [c0000000fffa3f00] [c000000000137edc] handle_fasteoi_irq+0xec/0x250
    [c0000000fffa3f30] [c000000000131b04] generic_handle_irq+0x54/0x80
    [c0000000fffa3f60] [c000000000011190] __do_irq+0x80/0x1d0
    [c0000000fffa3f90] [c0000000000248d8] call_do_irq+0x14/0x24
    [c00000007f9bb9e0] [c000000000011380] do_IRQ+0xa0/0x120
    [c00000007f9bba40] [c000000000002594] hardware_interrupt_common+0x114/0x180

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Correct crq and resource releasing
Nathan Fontenot [Wed, 19 Apr 2017 17:45:04 +0000 (13:45 -0400)]
ibmvnic: Correct crq and resource releasing

We should not be releasing the crq's when calling close for the
adapter, these need to remain open to facilitate operations such
as updating the mac address. The crq's should be released in the
adpaters remove routine.

Additionally, we need to call release_reources from remove. This
corrects the scenario of trying to remove an adapter that has only
been probed.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Remove inflight list
Nathan Fontenot [Wed, 19 Apr 2017 17:44:58 +0000 (13:44 -0400)]
ibmvnic: Remove inflight list

The inflight list used to track memory that is allocated for crq that are
inflight is not needed. The one piece of the inflight list that does need
to be cleaned at module exit is the error buffer list which is already
attached to the adapter struct.

This patch removes the inflight list and moves checking the error buffer
list to ibmvnic_remove.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Do not disable IRQ after scheduling tasklet
Brian King [Wed, 19 Apr 2017 17:44:53 +0000 (13:44 -0400)]
ibmvnic: Do not disable IRQ after scheduling tasklet

Since the primary CRQ is only used for service functions and
not in the performance path, simplify the code a bit and avoid
disabling the IRQ.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Fixup atomic API usage
Brian King [Wed, 19 Apr 2017 17:44:47 +0000 (13:44 -0400)]
ibmvnic: Fixup atomic API usage

Replace a couple of modifications of an atomic followed
by a read of the atomic, which is no longer atomic, to
use atomic_XX_return variants to avoid race conditions.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Unmap longer term buffer before free
Brian King [Wed, 19 Apr 2017 17:44:41 +0000 (13:44 -0400)]
ibmvnic: Unmap longer term buffer before free

Make sure we unregister long term buffers from the adapter
prior to DMA unmapping it and freeing the buffer. Failure
to do so could result in a DMA to a now invalid address.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Fix ibmvnic_change_mac_addr struct format
Murilo Fossa Vicentini [Wed, 19 Apr 2017 17:44:35 +0000 (13:44 -0400)]
ibmvnic: Fix ibmvnic_change_mac_addr struct format

The ibmvnic_change_mac_addr struct alignment was not matching the defined
format in PAPR+, it had the reserved and return code fields swapped. As a
consequence, the CHANGE_MAC_ADDR_RSP commands were being improperly handled
and executed even when the operation wasn't successfully completed by the
system firmware.

Also changing the endianness of the debug message to make it easier to
parse the CRQ content.

Signed-off-by: Murilo Fossa Vicentini <muvic@linux.vnet.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Report errors when failing to release sub-crqs
Thomas Falcon [Wed, 19 Apr 2017 17:44:29 +0000 (13:44 -0400)]
ibmvnic: Report errors when failing to release sub-crqs

Add reporting of errors when releasing sub-crqs fails.

Signed-off-by: Thomas Falcon <tlfalcon@us.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: remove unnecessary variable assignment
Arnd Bergmann [Wed, 19 Apr 2017 17:30:59 +0000 (19:30 +0200)]
liquidio: remove unnecessary variable assignment

gcc points out an useless assignment that was added during code refactoring:

drivers/net/ethernet/cavium/liquidio/lio_ethtool.c: In function 'octnet_intrmod_callback':
drivers/net/ethernet/cavium/liquidio/lio_ethtool.c:1315:59: error: parameter 'oct_dev' set but not used [-Werror=unused-but-set-parameter]

This is harmless but can clearly be remove to avoid the warning.

Fixes: 50c0add534d2 ("liquidio: refactor interrupt moderation code")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoReplace 2 jiffies with sysctl netdev_budget_usecs to enable softirq tuning
Matthew Whitehead [Wed, 19 Apr 2017 16:37:10 +0000 (12:37 -0400)]
Replace 2 jiffies with sysctl netdev_budget_usecs to enable softirq tuning

Constants used for tuning are generally a bad idea, especially as hardware
changes over time. Replace the constant 2 jiffies with sysctl variable
netdev_budget_usecs to enable sysadmins to tune the softirq processing.
Also document the variable.

For example, a very fast machine might tune this to 1000 microseconds,
while my regression testing 486DX-25 needs it to be 4000 microseconds on
a nearly idle network to prevent time_squeeze from being incremented.

Version 2: changed jiffies to microseconds for predictable units.

Signed-off-by: Matthew Whitehead <tedheadster@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'iptunnel-policy-based-routing'
David S. Miller [Fri, 21 Apr 2017 17:21:32 +0000 (13:21 -0400)]
Merge branch 'iptunnel-policy-based-routing'

Craig Gallek says:

====================
ip_tunnel: Allow policy-based routing through tunnels

iproute2 changes to follow.  Example usage:
  ip link add gre-test type gre local 10.0.0.1 remote 10.0.0.2 fwmark 0x4
  ip -detail link show gre-test
  ...
  ip link set gre-test type gre fwmark 0
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoip_tunnel: Allow policy-based routing through tunnels
Craig Gallek [Wed, 19 Apr 2017 16:30:54 +0000 (12:30 -0400)]
ip_tunnel: Allow policy-based routing through tunnels

This feature allows the administrator to set an fwmark for
packets traversing a tunnel.  This allows the use of independent
routing tables for tunneled packets without the use of iptables.

There is no concept of per-packet routing decisions through IPv4
tunnels, so this implementation does not need to work with
per-packet route lookups as the v6 implementation may
(with IP6_TNL_F_USE_ORIG_FWMARK).

Further, since the v4 tunnel ioctls share datastructures
(which can not be trivially modified) with the kernel's internal
tunnel configuration structures, the mark attribute must be stored
in the tunnel structure itself and passed as a parameter when
creating or changing tunnel attributes.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoip6_tunnel: Allow policy-based routing through tunnels
Craig Gallek [Wed, 19 Apr 2017 16:30:53 +0000 (12:30 -0400)]
ip6_tunnel: Allow policy-based routing through tunnels

This feature allows the administrator to set an fwmark for
packets traversing a tunnel.  This allows the use of independent
routing tables for tunneled packets without the use of iptables.

Signed-off-by: Craig Gallek <kraig@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Remove redundant NULL dst check
Florian Fainelli [Thu, 20 Apr 2017 22:47:22 +0000 (15:47 -0700)]
net: dsa: Remove redundant NULL dst check

tag_lan9303.c does check for a NULL dst but that's already checked by
dsa_switch_rcv() one layer above.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Juergen Borleis <jbe@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/mlx5e: IPoIB, Fix error handling in mlx5_rdma_netdev_alloc()
Dan Carpenter [Wed, 19 Apr 2017 09:59:15 +0000 (12:59 +0300)]
net/mlx5e: IPoIB, Fix error handling in mlx5_rdma_netdev_alloc()

The labels were out of order, so it either could result in an Oops or a
leak.

Fixes: 48935bbb7ae8 ("net/mlx5e: IPoIB, Add netdevice profile skeleton")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqede: allocate enough data for ->arfs_fltr_bmap
Dan Carpenter [Wed, 19 Apr 2017 09:54:33 +0000 (12:54 +0300)]
qede: allocate enough data for ->arfs_fltr_bmap

We've got the number of longs, yes, but we should multiply by
sizeof(long) to get the number of bytes needed.

Fixes: e4917d46a653 ("qede: Add aRFS support")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp_cubic: fix typo in module param description
Chema Gonzalez [Wed, 19 Apr 2017 02:22:23 +0000 (19:22 -0700)]
tcp_cubic: fix typo in module param description

Signed-off-by: Chema Gonzalez <chemag@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoAdd Jiri Pirko as TC subsystem co-maintainer
Jamal Hadi Salim [Wed, 19 Apr 2017 01:24:00 +0000 (21:24 -0400)]
Add Jiri Pirko as TC subsystem co-maintainer

Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoAdd Cong Wang as TC subsystem co-maintainer
Jamal Hadi Salim [Wed, 19 Apr 2017 01:23:59 +0000 (21:23 -0400)]
Add Cong Wang as TC subsystem co-maintainer

Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Thu, 20 Apr 2017 20:10:31 +0000 (16:10 -0400)]
Merge branch '10GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
10GbE Intel Wired LAN Driver Updates 2017-04-18

This series contains updates to mainly ixgbe with only one ixgbevf change.

Usha adds a check to ensure the creation of number of VF's is valid based
on the traffic classes configured, all to avoid transmit hangs.

Joe Perches reduces the use of pr_cont since the output can be interleaved
by other processes.

Tony cleans up the code overwriting the KX4 config, which is configured by
the NVM.  Adds a check for MMNGC.MNG_VETO, to resolve an issue where we
were getting a link loss for the BMC when loading the driver.

Don fixes up SGMII x553 config details which were missed in earlier
implementations.  Added support for x552 XFI backplane interface support.
Cleaned up an unused define, which was causing confusion on supported
devices.

Emil fixes a link issue on KR parts by making sure the default setting is
set.  Refactors the code so that the code for allocating memory for the
list of MAC addresses that the VFs can use into its own function.  Made
some code cleans to help readability and ensure notification of SRIOV
being enabled is done upon completion.  Fixed an issue where if we failed
to allocate vfinfo in __ixgbe_enable_sriov() the driver would crash with
a NULL pointer dereference.

Philippe Reynes updates ixgbevf to use the new API for
{get|set}_link_ksettings.

Alex increases the headroom allocation when using build_skb() on a
system with 4K pages.  Fixed an issue in ixgbe_dump() where we were no
longer clearing the status bit.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ipv6: Fix UDP early demux lookup with udp_l3mdev_accept=0
subashab@codeaurora.org [Tue, 18 Apr 2017 17:39:41 +0000 (11:39 -0600)]
net: ipv6: Fix UDP early demux lookup with udp_l3mdev_accept=0

David Ahern reported that 5425077d73e0c ("net: ipv6: Add early demux
handler for UDP unicast") breaks udp_l3mdev_accept=0 since early
demux for IPv6 UDP was doing a generic socket lookup which does not
require an exact match. Fix this by making UDPv6 early demux match
connected sockets only.

v1->v2: Take reference to socket after match as suggested by Eric
v2->v3: Add comment before break

Fixes: 5425077d73e0c ("net: ipv6: Add early demux handler for UDP unicast")
Reported-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Tested-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'tcp_poll-flakes'
David S. Miller [Thu, 20 Apr 2017 19:42:11 +0000 (15:42 -0400)]
Merge branch 'tcp_poll-flakes'

Eric Dumazet says:

====================
tcp: address two poll() flakes

Some packetdrill tests are failing when host kernel is using ASAN
or other debugging infrastructure.

I was able to fix the flakes by making sure we were not
sending wakeup events too soon.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: remove poll() flakes with FastOpen
Eric Dumazet [Tue, 18 Apr 2017 16:45:52 +0000 (09:45 -0700)]
tcp: remove poll() flakes with FastOpen

When using TCP FastOpen for an active session, we send one wakeup event
from tcp_finish_connect(), right before the data eventually contained in
the received SYNACK is queued to sk->sk_receive_queue.

This means that depending on machine load or luck, poll() users
might receive POLLOUT events instead of POLLIN|POLLOUT

To fix this, we need to move the call to sk->sk_state_change()
after the (optional) call to tcp_rcv_fastopen_synack()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: remove poll() flakes when receiving RST
Eric Dumazet [Tue, 18 Apr 2017 16:45:51 +0000 (09:45 -0700)]
tcp: remove poll() flakes when receiving RST

When a RST packet is processed, we send two wakeup events to interested
polling users.

First one by a sk->sk_error_report(sk) from tcp_reset(),
followed by a sk->sk_state_change(sk) from tcp_done().

Depending on machine load and luck, poll() can either return POLLERR,
or POLLIN|POLLOUT|POLLERR|POLLHUP (this happens on 99 % of the cases)

This is probably fine, but we can avoid the confusion by reordering
things so that we have more TCP fields updated before the first wakeup.

This might even allow us to remove some barriers we added in the past.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mlxsw-flow-based-forwarding-OVS'
David S. Miller [Thu, 20 Apr 2017 19:32:32 +0000 (15:32 -0400)]
Merge branch 'mlxsw-flow-based-forwarding-OVS'

Jiri Pirko says:

====================
mlxsw: Allow flow based forwarding in OVS

This patchset does some fixes so the HW is setup correctly to do
flow-based (ACL based) forwarding for OVS-enslaved port.

The first patch is just trivial fix spotted on the way.

Patches 2-4 take care of proper FID setup which HW needs in order to
for ACL based forwarding.

The 7th patch (with dependency of patch 5 and 6) takes care of proper setup
of ports that are enslaved in OVS.

The last patch implements new FID miss trap that is used to push
packets belonging to unknown flows to kernel and userspace.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Add FID miss trap
Jiri Pirko [Tue, 18 Apr 2017 14:55:38 +0000 (16:55 +0200)]
mlxsw: spectrum: Add FID miss trap

When there is no FID set for a specific packet, the HW will drop it.
However, by default these packets are useful to be delivered to CPU as
it can inspect them and program HW accordingly. So add this trap.

This would only ever happen when port is enslaved to an OVS master.
Otherwise, packets would be dropped during VLAN / STP filtering,
before FID classification.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Allow ports to work under OVS master
Jiri Pirko [Tue, 18 Apr 2017 14:55:37 +0000 (16:55 +0200)]
mlxsw: spectrum: Allow ports to work under OVS master

>From now on, a port can become a slave of OVS master. All vlans
are enabled, STP state is set to "forwarding". It is up to the OVS
userspace daemon to setup the flows either in kernel or in HW using TC
flower offload.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: add netif_is_ovs_port helper
Jiri Pirko [Tue, 18 Apr 2017 14:55:36 +0000 (16:55 +0200)]
net: add netif_is_ovs_port helper

To find out if a netdev is an OVS port.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Teach mlxsw_sp_port_vlan_set to accept any vlan range
Jiri Pirko [Tue, 18 Apr 2017 14:55:35 +0000 (16:55 +0200)]
mlxsw: spectrum: Teach mlxsw_sp_port_vlan_set to accept any vlan range

So far, mlxsw_sp_port_vlan_set range is limited by
MLXSW_REG_SPVM_REC_MAX_COUNT. In spectrum_switchdev code this is
wrapped up by a helper function which actually does multiple calls
to FW for bigger ranges. Move the code into mlxsw_sp_port_vlan_set
and use it always. That allows caller not to care about the range.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_flower: Set dummy FID before forward action
Jiri Pirko [Tue, 18 Apr 2017 14:55:34 +0000 (16:55 +0200)]
mlxsw: spectrum_flower: Set dummy FID before forward action

HW requires the FID to be valid in order for the forward action to work.
So regardless of the current FID validity, just set the dummy FID which
would do the trick.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Add dummy FID initialization
Jiri Pirko [Tue, 18 Apr 2017 14:55:33 +0000 (16:55 +0200)]
mlxsw: spectrum: Add dummy FID initialization

For forwarding using ACL action, HW needs a valid FID to be setup. It
does not actually use it, so it can be any valid FID. So create a dummy
FID only for this purpose.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Implement action to set FID
Jiri Pirko [Tue, 18 Apr 2017 14:55:32 +0000 (16:55 +0200)]
mlxsw: spectrum: Implement action to set FID

Implement part of multipurpose Virtual Router and Forwarding Domain
Action that takes care of setting up FID. We need to use it to be able
to forward packets using ACL action when no FID is associated on RX.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Fix indent in mlxsw_sp_netdevice_port_upper_event
Jiri Pirko [Tue, 18 Apr 2017 14:55:31 +0000 (16:55 +0200)]
mlxsw: spectrum: Fix indent in mlxsw_sp_netdevice_port_upper_event

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobindings: net: stmmac: add missing note about LPI interrupt
Niklas Cassel [Tue, 18 Apr 2017 12:39:53 +0000 (14:39 +0200)]
bindings: net: stmmac: add missing note about LPI interrupt

The hardware has a LPI interrupt.
There is already code in the stmmac driver to parse and handle the
interrupt. However, this information was missing from the DT binding.

At the same time, improve the description of the existing interrupts.

Signed-off-by: Niklas Cassel <niklas.cassel@axis.com>
Acked-By: Joao Pinto <jpinto@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: remove reference to sock_filter_ext from kerneldoc comment
Tobias Klauser [Tue, 18 Apr 2017 09:27:00 +0000 (11:27 +0200)]
bpf: remove reference to sock_filter_ext from kerneldoc comment

struct sock_filter_ext didn't make it into the tree and is now called
struct bpf_insn. Reword the kerneldoc comment for bpf_convert_filter()
accordingly.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'mac80211-next-for-davem-2017-04-18' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Thu, 20 Apr 2017 17:54:40 +0000 (13:54 -0400)]
Merge tag 'mac80211-next-for-davem-2017-04-18' of git://git./linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
My last pull request has been a while, we now have:
 * connection quality monitoring with multiple thresholds
 * support for FILS shared key authentication offload
 * pre-CAC regulatory compliance - only ETSI allows this
 * sanity check for some rate confusion that hit ChromeOS
   (but nobody else uses it, evidently)
 * some documentation updates
 * lots of cleanups
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-LAN9303'
David S. Miller [Thu, 20 Apr 2017 17:48:55 +0000 (13:48 -0400)]
Merge branch 'dsa-LAN9303'

Juergen Borleis says:

====================
net: dsa: add SMSC/Microchip LAN9303 three port ethernet switch driver

The LAN9303 is a three port 10/100 ethernet switch with integrated phys
for the two external ethernet ports. The third port is an RMII/MII
interface to a host master network interface (e.g. fixed link).

While the LAN9303 device itself supports offload packet processing, this
driver does not make use of it yet. This driver just configures the device
to provide two separate network interfaces (which is the default state of
a DSA device).

Please note: the "MDIO managed mode" driver part isn't tested yet. I have
used and tested the "I2C managed mode" only.

Changes in v6:

- fix support to use the driver as a module (core, i2c and mdio)
- license info added in all parts of the driver (for module support)

Changes in v5:

- add missing include file to 'net/dsa/tag_lan9303.c'

Changes in v4:

- rebased on net-next, 'net/dsa/tag_lan9303.c' adapted to changed API

Changes in v3:

- 'ds_to_lan9303()' removed
- special PHY reg MII_LAN911X_SPECIAL_CONTROL_STATUS mapping removed
- compatible strings for I2C and MDIO are now different
- MDIO-managed-mode devicetree binding added (still compile time tested only)

Changes in v2:

- code moved to 'drivers/net/dsa'
- timeouts in completion wait loops
- macros instead of various magic numbers
- development code removed
- devicetree property names changed
- devicetree example adapted
- tried to avoid to mix 'switching' and 'forwarding'...

Comments are welcome.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: LAN9303: add MDIO managed mode support
Juergen Beisert [Tue, 18 Apr 2017 08:48:27 +0000 (10:48 +0200)]
net: dsa: LAN9303: add MDIO managed mode support

When the LAN9303 device is in MDIO manged mode, all register accesses must
be done via MDIO.

Please note: this code is compile time tested only due to the absence of such
configured hardware. It is based on a patch from Stefan Roese from 2014.

Signed-off-by: Juergen Borleis <jbe@pengutronix.de>
CC: devicetree@vger.kernel.org
CC: robh+dt@kernel.org
CC: mark.rutland@arm.com
CC: sr@denx.de
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: LAN9303: add I2C managed mode support
Juergen Beisert [Tue, 18 Apr 2017 08:48:26 +0000 (10:48 +0200)]
net: dsa: LAN9303: add I2C managed mode support

In this mode the switch device and the internal phys will be managed via
I2C interface. The MDIO interface is still supported, but for the
(emulated) CPU port only.

Signed-off-by: Juergen Borleis <jbe@pengutronix.de>
CC: devicetree@vger.kernel.org
CC: robh+dt@kernel.org
CC: mark.rutland@arm.com
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: add new DSA switch driver for the SMSC-LAN9303
Juergen Beisert [Tue, 18 Apr 2017 08:48:25 +0000 (10:48 +0200)]
net: dsa: add new DSA switch driver for the SMSC-LAN9303

The SMSC/Microchip LAN9303 is an ethernet switch device with one CPU port
and two external ethernet ports with built-in phys.

This driver uses the DSA framework, but is currently only capable of
separating the two external ports. There is no offload support yet.

Signed-off-by: Juergen Borleis <jbe@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: add support for the SMSC-LAN9303 tagging format
Juergen Beisert [Tue, 18 Apr 2017 08:48:24 +0000 (10:48 +0200)]
net: dsa: add support for the SMSC-LAN9303 tagging format

To define the outgoing port and to discover the incoming port a regular
VLAN tag is used by the LAN9303. But its VID meaning is 'special'.

This tag handler/filter depends on some hardware features which must be
enabled in the device to provide and make use of this special VLAN tag
to control the destination and the source of an ethernet packet.

Signed-off-by: Juergen Borleis <jbe@pengutronix.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/mlx4: suppress 'may be used uninitialized' warning
Greg Thelen [Tue, 18 Apr 2017 06:21:35 +0000 (23:21 -0700)]
net/mlx4: suppress 'may be used uninitialized' warning

gcc 4.8.4 complains that mlx4_SW2HW_MPT_wrapper() uses an uninitialized
'mpt' variable:
  drivers/net/ethernet/mellanox/mlx4/resource_tracker.c: In function 'mlx4_SW2HW_MPT_wrapper':
  drivers/net/ethernet/mellanox/mlx4/resource_tracker.c:2802:12: warning: 'mpt' may be used uninitialized in this function [-Wmaybe-uninitialized]
     mpt->mtt = mtt;

I think this warning is a false complaint.  mpt is only used when
mr_res_start_move_to() return zero, and in all such cases it initializes
mpt.  But apparently gcc cannot see that.

Initialize mpt to avoid the warning.

Signed-off-by: Greg Thelen <gthelen@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Thu, 20 Apr 2017 14:35:33 +0000 (10:35 -0400)]
Merge git://git./linux/kernel/git/davem/net

A function in kernel/bpf/syscall.c which got a bug fix in 'net'
was moved to kernel/bpf/verifier.c in 'net-next'.

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoi40e: use i40e_stop_rings_no_wait to implement PORT_SUSPENDED state
Jacob Keller [Thu, 13 Apr 2017 08:45:53 +0000 (04:45 -0400)]
i40e: use i40e_stop_rings_no_wait to implement PORT_SUSPENDED state

This state bit was added as a way for DCB to avoid having to wait for
the queues to disable when handling LLDP events. The logic for this was
burried deep within stop Tx and stop Rx queue code. First, let's rename
it so that it does not appear to only affect Tx when infact it modifies
both Tx and Rx flow. Second we can move it up into the i40e_stop_rings()
function, and we can simply re-use the i40e_stop_rings_no_wait() so that
we don't have to bury the implementation as deep into the call stack.

An alternative might be to remove the state bit and instead attempt to
shut down everything directly in DCP flow. This, however, is not ideal
because it creates yet another separate shutdown routine that we'd have
to maintain. In the current implementation any changes will be made to
both flows.

Change-ID: I68e1ccb901af320862bca395e9c9746f08e8b17c
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: reset all VFs in parallel when rebuilding PF
Jacob Keller [Thu, 13 Apr 2017 08:45:52 +0000 (04:45 -0400)]
i40e: reset all VFs in parallel when rebuilding PF

When there are a lot of active VFs, it can take multiple seconds to
finish resetting all of them during certain flows., which can cause some
VFs to fail to wait long enough for the reset to occur. The user might
see messages like "Never saw reset" or "Reset never finished" and the VF
driver will stop functioning properly.

The naive solution would be to simply increase the wait timer. We can
get much more clever. Notice that i40e_reset_vf is run in a serialized
fashion, and includes lots of delays.

There are two prominent delays which take most of the time. First, when
we begin resetting VFs, we have multiple 10ms delays which accrue
because we reset each VF in a serial fashion. These delays accumulate to
almost 4 seconds when handling the maximum number of VFs (128).

Secondly, there is a massive 50ms delay for each time we disable queues
on a VSI. This delay is necessary to allow HW to finish disabling queues
before we restore functionality. However, just like with the first case,
we are paying the cost for each VF, rather than disabling all VFs and
waiting once.

Both of these can be fixed, but required some previous refactoring to
handle the special case. First, we will need the
i40e_vsi_wait_queues_disabled function which was previously DCB
specific. Second, we will need to implement our own
i40e_vsi_stop_rings_no_wait function which will handle the stopping of
rings without the delays.

Finally, implement an i40e_reset_all_vfs function, which will first
start the reset of all VFs, and pay the wait cost all at once, rather
than serially waiting for each VF before we start processing then next
one. After the VF has been reset, we'll disable all the VF queues, and
then wait for them to disable. Again, we'll organize the flow such that
we pay the wait cost only once.

Finally, after we've disabled queues we'll go ahead and begin restoring
VF functionality. The result is reducing the wait time by a large factor
and ensuring that VFs do not timeout when waiting in the VF driver.

Change-ID: Ia6e8cf8d98131b78aec89db78afb8d905c9b12be
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: split some code in i40e_reset_vf into helpers
Jacob Keller [Thu, 13 Apr 2017 08:45:51 +0000 (04:45 -0400)]
i40e: split some code in i40e_reset_vf into helpers

A future patch is going to want to re-use some of the code in
i40e_reset_vf, so lets break up the beginning and ending parts into
their own helper functions. The first function will be used to
initialize the reset on a VF, while the second function will be used to
finalize the reset and restore functionality.

Change-ID: I48df808b8bf09de3c2ed8c521f57b3f0ab9e5907
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: remove I40E_FLAG_IN_NETPOLL entirely
Jacob Keller [Thu, 13 Apr 2017 08:45:50 +0000 (04:45 -0400)]
i40e: remove I40E_FLAG_IN_NETPOLL entirely

This flag was originally intended to be used to let some
driver code know when we were running from netpoll.
Ultimately this was not necessary and we never used it.
Let's remove it

Change-ID: I43b72483d91c1638071d2a7f389ab171ec5b796a
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: reduce wait time for adminq command completion
Jacob Keller [Thu, 13 Apr 2017 08:45:49 +0000 (04:45 -0400)]
i40e: reduce wait time for adminq command completion

When sending an adminq command, we wait for the command to complete in
a loop. This loop waits for an entire millisecond, when in practice the
adminq command is processed often much faster.

Change the loop to use i40e_usec_delay instead, and wait for 50 usecs
each time instead. This appears to be about the minimum time required,
based on some manual observation and testing.

The primary benefit of this change is reducing latency of various
operations in the PF driver, especially when related to having a large
number of VFs enabled.

For example, on Linux, when instantiating 128 VFs, the time to finish
the operation dropped from about 9 seconds down to under 6 seconds.
Additionally, the time it takes to finish a PF reset with 128 VFs
dropped from 5.1 seconds down to 0.7 seconds.

As the examples above show, a significant portion of the delay is wasted
waiting for admiqn operations which have already finished.

This patch shouldn't cause impact to functionality, as we still check
and keep waiting until the command does get processed. The only expected
change is an increase in CPU utilization as we now check for completion
far more times. However, in practice the commands appear to generally be
complete within the first delay window anyways.

Change-ID: If8af8388e100da0a14eaf9e1af3afadf73a958cf
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: fix CONFIG_BUSY checks in i40e_set_settings function
Jacob Keller [Thu, 13 Apr 2017 08:45:48 +0000 (04:45 -0400)]
i40e: fix CONFIG_BUSY checks in i40e_set_settings function

The check for I40E_CONFIG_BUSY state bit in the i40e_set_link_ksettings
function is fishy. First we can notice a few things about the check here.

First a similar check was introduced by commit
'c7d05ca89f8e ("i40e: driver ethtool core")'

Later a commit introducing the link settings was added by commit
'bf9c71417f72 ("i40e: Implement set_settings for ethtool")'

However, this second check was against vsi->state instead of pf->state,
and also failed to set the bit, it only checks. That indicates the locking
was not quite correct. The only other place that the state bit
in vsi->state gets used is to protect the filter list.

Since this code does not care about the mac filter list,  and seems
clear the original code should have set the pf->state bit. Fix these
issues by using pf->state correctly, and by actually setting the bit
so that we properly lock as expected.

Since these checks occur while holding the rtnl_lock(), lets also add a
timeout so that we don't potentially softlock the system.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: factor out queue control from i40e_vsi_control_(tx|rx)
Jacob Keller [Thu, 13 Apr 2017 08:45:47 +0000 (04:45 -0400)]
i40e: factor out queue control from i40e_vsi_control_(tx|rx)

A future patch will need to be able to handle controlling queues without
waiting until all VSIs are handled. Factor out the direct queue
modification so that we can easily re-use this code. The result is also
a bit easier to read since we don't embed multiple single-letter loop
counters.

Change-ID: Id923cbfa43127b1c24d8ed4f809b1012c736d9ac
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: don't hold RTNL lock while waiting for VF reset to finish
Jacob Keller [Thu, 13 Apr 2017 08:45:46 +0000 (04:45 -0400)]
i40e: don't hold RTNL lock while waiting for VF reset to finish

We made some effort to reduce the RTNL lock scope when resetting and
rebuilding the PF. Unfortunately we still held the RTNL lock during the
VF reset operation, which meant that multiple PFs could not reset in
parallel due to the global lock. For now, further reduce the scope by
not holding the RTNL lock while resetting VFs. This allows multiple PFs
to reset in a timely manner.

Change-ID: I2fbf823a0063f24dff67676cad09f0bbf83ee4ce
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: new AQ commands
Jingjing Wu [Thu, 13 Apr 2017 08:45:45 +0000 (04:45 -0400)]
i40e: new AQ commands

Add admin queue functions for Pipeline Personalization Profile AQ
commands:
 - Write Recipe Command buffer (Opcode: 0x0270)
 - Get Applied Profiles list (Opcode: 0x0271)

Change-ID: I558b4145364140f624013af48d4bbf79d21ebb0d
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Add tracepoints
Scott Peterson [Thu, 13 Apr 2017 08:45:44 +0000 (04:45 -0400)]
i40e/i40evf: Add tracepoints

This patch adds tracepoints to the i40e and i40evf drivers to which
BPF programs can be attached for feature testing and verification.
It's expected that an attached BPF program will identify and count or
log some interesting subset of traffic. The bcc-tools package is
helpful there for containing all the BPF arcana in a handy Python
wrapper. Though you can make these tracepoints log trace messages, the
messages themselves probably won't be very useful (other to verify the
tracepoint is being called while you're debugging your BPF program).

The idea here is that tracepoints have such low performance cost when
disabled that we can leave these in the upstream drivers. This may
eventually enable the instrumentation of unmodified customer systems
should the need arise to verify a NIC feature is working as expected.
In general this enables one set of feature verification tools to be
used on these drivers whether they're built with the kernel or
separately.

Users are advised against using these tracepoints for anything other
than a diagnostic tool. They have a performance impact when enabled,
and their exact placement and form may change as we see how well they
work in practice for the purposes above.

Change-ID: Id6014a7322c0e6d08068114dd20bd156f2f6435e
Signed-off-by: Scott Peterson <scott.d.peterson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: dump VF information in debugfs
Mitch Williams [Wed, 12 Apr 2017 11:16:52 +0000 (07:16 -0400)]
i40e: dump VF information in debugfs

Dump some internal state about VFs through debugfs. This provides
information not available with 'ip link show'. To use, write "dump vf
<id>" to the command file, or just "dump vf" to dump information on all
of the VFs.

Change-ID: Ibe32b7f4ae55d4358c0b903217475f708ada1ecd
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Fix support for flow director programming status
Alexander Duyck [Mon, 10 Apr 2017 09:18:43 +0000 (05:18 -0400)]
i40e: Fix support for flow director programming status

This patch fixes an issue I introduced when I converted the code over to
using the length field to determine if a descriptor was done or not. It
turns out that we are also processing programming descriptors in the Rx
path and need to have these processed even though the length field will be
0 on these packets.  What will happen with a programming descriptor is that
we will receive a descriptor that has the SPH bit set, and the header
length and packet length fields cleared.

To account for this we should be checking for the bit for split header
being set even though we aren't actually using header split. This bit is
set in the length field to indicate if a programming descriptor response is
contained in the descriptor. Since we don't support header split we don't
need to perform the extra checks of using a fixed value for the entire
length field.

In addition I am moving the function for checking if a filter is a
programming status filter into the i40e_txrx.c file since there is no
longer support for FCoE it doesn't make sense to keep this file in i40e.h.

Change-ID: I12c359c3dc70adb9d6b92b27324bb2c7f04c1a06
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Remove VF Rx csum offload for tunneled packets
alice michael [Thu, 6 Apr 2017 09:59:34 +0000 (05:59 -0400)]
i40e/i40evf: Remove VF Rx csum offload for tunneled packets

Rx checksum offload for tunneled packets was never being negotiated or
requested by VF. This capability was assumed by default and enabled in
current hardware for VF. Going forward, this feature needs to be disabled
or advanced ptypes should be negotiated with PF in the future.

Change-ID: I9e54cfa8a90e03ab6956db4412f1e337ccd2c2e0
Signed-off-by: Preethi Banala <preethi.banala@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40evf: Use net_device_stats from struct net_device
Tobias Klauser [Thu, 6 Apr 2017 06:46:28 +0000 (08:46 +0200)]
i40evf: Use net_device_stats from struct net_device

Instead of using a private copy of struct net_device_stats in
struct i40evf_adapter, use stats from struct net_device. Also remove the
now unnecessary .ndo_get_stats function.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoesp4/6: Fix GSO path for non-GSO SW-crypto packets
Ilan Tayari [Wed, 19 Apr 2017 05:41:01 +0000 (08:41 +0300)]
esp4/6: Fix GSO path for non-GSO SW-crypto packets

If esp*_offload module is loaded, outbound packets take the
GSO code path, being encapsulated at layer 3, but encrypted
in layer 2. validate_xmit_xfrm calls esp*_xmit for that.

esp*_xmit was wrongfully detecting these packets as going
through hardware crypto offload, while in fact they should
be encrypted in software, causing plaintext leakage to
the network, and also dropping at the receiver side.

Perform the encryption in esp*_xmit, if the SA doesn't have
a hardware offload_handle.

Also, align esp6 code to esp4 logic.

Fixes: fca11ebde3f0 ("esp4: Reorganize esp_output")
Fixes: 383d0350f2cc ("esp6: Reorganize esp_output")
Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
7 years agoesp6: fix incorrect null pointer check on xo
Colin Ian King [Tue, 18 Apr 2017 14:06:53 +0000 (15:06 +0100)]
esp6: fix incorrect null pointer check on xo

The check for xo being null is incorrect, currently it is checking
for non-null, it should be checking for null.

Detected with CoverityScan, CID#1429349 ("Dereference after null check")

Fixes: 7862b4058b9f ("esp: Add gso handlers for esp4 and esp6")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
7 years agosh_eth: unmap DMA buffers when freeing rings
Sergei Shtylyov [Mon, 17 Apr 2017 12:55:22 +0000 (15:55 +0300)]
sh_eth: unmap DMA buffers when freeing rings

The DMA API debugging (when enabled) causes:

WARNING: CPU: 0 PID: 1445 at lib/dma-debug.c:519 add_dma_entry+0xe0/0x12c
DMA-API: exceeded 7 overlapping mappings of cacheline 0x01b2974d

to be  printed after repeated initialization of the Ether device, e.g.
suspend/resume or 'ifconfig' up/down. This is because DMA buffers mapped
using dma_map_single() in sh_eth_ring_format() and sh_eth_start_xmit() are
never unmapped. Resolve this problem by unmapping the buffers when freeing
the descriptor  rings;  in order  to do it right, we'd have to add an extra
parameter to sh_eth_txfree() (we rename this function to sh_eth_tx_free(),
while at it).

Based on the commit a47b70ea86bd ("ravb: unmap descriptors when freeing
rings").

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoixgbe: Fix output from ixgbe_dump
Alexander Duyck [Thu, 2 Mar 2017 23:01:36 +0000 (15:01 -0800)]
ixgbe: Fix output from ixgbe_dump

I just found that when we had changed the Rx path to check for length
instead of the DD bit we introduced an issue in ixgbe_dump since we were no
longer clearing the status bits.

To correct this I am updating ixgbe_dump to look for the length bits in the
descriptor since that is what we are using in the Rx path.

Fixes: c3630cc40b4f ("ixgbe: Use length to determine if descriptor is done")
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
Linus Torvalds [Tue, 18 Apr 2017 20:56:51 +0000 (13:56 -0700)]
Merge git://git./linux/kernel/git/davem/sparc

Pull sparc fixes from David Miller:
 "Two Sparc bug fixes from Daniel Jordan and Nitin Gupta"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc:
  sparc64: Fix hugepage page table free
  sparc64: Use LOCKDEP_SMALL, not PROVE_LOCKING_SMALL

7 years agoixgbe: Add support for maximum headroom when using build_skb
Alexander Duyck [Thu, 2 Mar 2017 23:01:05 +0000 (15:01 -0800)]
ixgbe: Add support for maximum headroom when using build_skb

This patch increases the headroom allocated when using build_skb on a
system with 4K pages.  Specifically the breakdown of headroom versus cache
size is as follows:
    L1 Cache Size           Headroom
    64                      192
    64, NET_IP_ALIGN == 2   194
    128                     128
    128, NET_IP_ALIGN == 2  130
    256                     512
    256, NET_IP_ALIGN == 2  258

I stopped at supporting only a cache line size of 256 as that was the
largest cache size I could find supported in the kernel.

With this we are guaranteeing at least 128 bytes of headroom to spare in
the frame.  This should be enough for us to insert a couple of IPv6 headers
if needed which is likely enough room for anything XDP should need.

I'm leaving the padding for systems with pages larger than 4K unmodified
for now.  XDP currently isn't really setup to work on those types of
systems so we can cross that bridge when we get there.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: add check for VETO bit when configuring link for KR
Tony Nguyen [Wed, 1 Mar 2017 19:52:09 +0000 (11:52 -0800)]
ixgbe: add check for VETO bit when configuring link for KR

We did not have a check in place for MMNGC.MNG_VETO when setting up link
on X550EM_X KR devices which resulted in link loss for the BMC when
loading the driver.

This patch adds a check for ixgbe_check_reset_blocked() in setup_link()
since in that case there is no PHY reset function.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbevf: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Tue, 7 Feb 2017 15:56:33 +0000 (16:56 +0100)]
ixgbevf: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Remove unused define
Don Skidmore [Thu, 2 Feb 2017 19:38:46 +0000 (14:38 -0500)]
ixgbe: Remove unused define

Remove the Marvell 1145 PHY define as we have never had a device that
supports it and have no plan to in the future.  The existence of this
define has caused confusing on whether or not this PHY was supported
by ixgbe.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: do not use adapter->num_vfs when setting VFs via module parameter
Emil Tantilov [Fri, 20 Jan 2017 22:11:56 +0000 (14:11 -0800)]
ixgbe: do not use adapter->num_vfs when setting VFs via module parameter

Avoid setting adapter->num_vfs early in the init code path when
using the max_vfs module parameter by passing it to ixgbe_enable_sriov()
as a function parameter.

This fixes an issue where if we failed to allocate vfinfo in
__ixgbe_enable_sriov() the driver will crash with NULL pointer in
ixgbe_disable_sriov() when attempting to free the vfinfo struct based
on adapter->num_vfs. Also it cleans up the assignment of adapter->num_vfs
since now it will only be set in __ixgbe_enable_sriov() and cleared in
ixgbe_disable_sriov().

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 18 Apr 2017 20:24:42 +0000 (13:24 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) BPF tail call handling bug fixes from Daniel Borkmann.

 2) Fix allowance of too many rx queues in sfc driver, from Bert
    Kenward.

 3) Non-loopback ipv6 packets claiming src of ::1 should be dropped,
    from Florian Westphal.

 4) Statistics requests on KSZ9031 can crash, fix from Grygorii
    Strashko.

 5) TX ring handling fixes in mediatek driver, from Sean Wang.

 6) ip_ra_control can deadlock, fix lock acquisition ordering to fix,
    from Cong WANG.

 7) Fix use after free in ip_recv_error(), from Willem de Buijn.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  bpf: fix checking xdp_adjust_head on tail calls
  bpf: fix cb access in socket filter programs on tail calls
  ipv6: drop non loopback packets claiming to originate from ::1
  net: ethernet: mediatek: fix inconsistency of port number carried in TXD
  net: ethernet: mediatek: fix inconsistency between TXD and the used buffer
  net: phy: micrel: fix crash when statistic requested for KSZ9031 phy
  net: vrf: Fix setting NLM_F_EXCL flag when adding l3mdev rule
  net: thunderx: Fix set_max_bgx_per_node for 81xx rgx
  net-timestamp: avoid use-after-free in ip_recv_error
  ipv4: fix a deadlock in ip_ra_control
  sfc: limit the number of receive queues

7 years agoixgbe: return early instead of wrap block in if statement
Emil Tantilov [Fri, 20 Jan 2017 22:11:50 +0000 (14:11 -0800)]
ixgbe: return early instead of wrap block in if statement

Since we exit at the end of the block, we can save a level of
indentation by performing an early return, and make the next several
sections of code more legible, with fewer 80 character line breaks.

Also moved allocating vfinfo at the beginning and the notification
for enabling SRIOV at the end of the function when we know that it
will succeed.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: move num_vfs_macvlans allocation into separate function
Emil Tantilov [Fri, 20 Jan 2017 22:11:45 +0000 (14:11 -0800)]
ixgbe: move num_vfs_macvlans allocation into separate function

Move the code allocating memory for list of MAC addresses that
the VFs can use for MACVLAN into its own function.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agosparc64: Fix hugepage page table free
Nitin Gupta [Mon, 17 Apr 2017 22:46:41 +0000 (15:46 -0700)]
sparc64: Fix hugepage page table free

Make sure the start adderess is aligned to PMD_SIZE
boundary when freeing page table backing a hugepage
region. The issue was causing segfaults when a region
backed by 64K pages was unmapped since such a region
is in general not PMD_SIZE aligned.

Signed-off-by: Nitin Gupta <nitin.m.gupta@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosparc64: Use LOCKDEP_SMALL, not PROVE_LOCKING_SMALL
Daniel Jordan [Mon, 10 Apr 2017 15:50:52 +0000 (11:50 -0400)]
sparc64: Use LOCKDEP_SMALL, not PROVE_LOCKING_SMALL

CONFIG_PROVE_LOCKING_SMALL shrinks the memory usage of lockdep so the
kernel text, data, and bss fit in the required 32MB limit, but this
option is not set for every config that enables lockdep.

A 4.10 kernel fails to boot with the console output

    Kernel: Using 8 locked TLB entries for main kernel image.
    hypervisor_tlb_lock[2000000:0:8000000071c007c3:1]: errors with f
    Program terminated

with these config options

    CONFIG_LOCKDEP=y
    CONFIG_LOCK_STAT=y
    CONFIG_PROVE_LOCKING=n

To fix, rename CONFIG_PROVE_LOCKING_SMALL to CONFIG_LOCKDEP_SMALL, and
enable this option with CONFIG_LOCKDEP=y so we get the reduced memory
usage every time lockdep is turned on.

Tested that CONFIG_LOCKDEP_SMALL is set to 'y' if and only if
CONFIG_LOCKDEP is set to 'y'.  When other lockdep-related config options
that select CONFIG_LOCKDEP are enabled (e.g. CONFIG_LOCK_STAT or
CONFIG_PROVE_LOCKING), verified that CONFIG_LOCKDEP_SMALL is also
enabled.

Fixes: e6b5f1be7afe ("config: Adding the new config parameter CONFIG_PROVE_LOCKING_SMALL for sparc")
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Reviewed-by: Babu Moger <babu.moger@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoixgbe: add default setup_link for x550em_a MAC type
Emil Tantilov [Thu, 19 Jan 2017 23:55:12 +0000 (15:55 -0800)]
ixgbe: add default setup_link for x550em_a MAC type

Add default setting for mac->ops.setup_link on x550em_a MAC types.
This fixes a link issue on KR parts.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: list X553 backplane speeds correctly
Don Skidmore [Sat, 31 Dec 2016 02:13:18 +0000 (21:13 -0500)]
ixgbe: list X553 backplane speeds correctly

We forgot to indicate some of the supported speed on the X553
backplane.  This patch attempts to correct for that.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Add X552 XFI backplane support
Don Skidmore [Sat, 31 Dec 2016 02:07:58 +0000 (21:07 -0500)]
ixgbe: Add X552 XFI backplane support

This patch add support for X552 XFI backplane interface.  The XFI
backplane requires a custom tuned link.  HW/FW owns the link config
for XF backplane and SW must not interfere with it.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Complete support for X553 sgmii
Don Skidmore [Sat, 31 Dec 2016 02:07:14 +0000 (21:07 -0500)]
ixgbe: Complete support for X553 sgmii

The initial patches supporting X553 sgmii forgot some details.  This patch
should cover those missing spots.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Remove driver config for KX4 PHY
Tony Nguyen [Tue, 1 Nov 2016 20:58:27 +0000 (13:58 -0700)]
ixgbe: Remove driver config for KX4 PHY

The KX4 PHY is configured by the NVM.  Currently, the driver is overwriting
the config; remove the code associated with KX4 configuration.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
7 years agonet: cx89x0: move attribute declaration before struct keyword
Stefan Agner [Mon, 17 Apr 2017 20:54:34 +0000 (13:54 -0700)]
net: cx89x0: move attribute declaration before struct keyword

The attribute declaration is typically before the definition. Move
the __maybe_unused attribute declaration before the struct keyword.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoixgbe: Remove pr_cont uses
Joe Perches [Tue, 3 Jan 2017 15:28:11 +0000 (07:28 -0800)]
ixgbe: Remove pr_cont uses

As pr_cont output can be interleaved by other processes,
using pr_cont should be avoided where possible.

Miscellanea:

- Use a temporary pointer to hold the next descriptions and
  consolidate the pr_cont uses
- Use the temporary buffer to hold the 8 u32 register values and
  emit those in a single go
- Coalesce formats and logging neatening around those changes
- Fix a defective output for the rx ring entry description when
  also emitting rx_buffer_info data

This reduces overall object size a tiny bit too.

$ size drivers/net/ethernet/intel/ixgbe/*.o*
   text    data     bss     dec     hex filename
  62167     728      12   62907    f5bb drivers/net/ethernet/intel/ixgbe/ixgbe_main.o.new
  62273     728      12   63013    f625 drivers/net/ethernet/intel/ixgbe/ixgbe_main.o.old

Signed-off-by: Joe Perches <joe@perches.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agobe2net: VxLAN offload should be re-enabled when only 1 UDP port is left
Sriharsha Basavapatna [Mon, 17 Apr 2017 16:03:13 +0000 (21:33 +0530)]
be2net: VxLAN offload should be re-enabled when only 1 UDP port is left

We disable VxLAN offload when more than 1 UDP port is added to the driver,
since Skyhawk doesn't support offload with multiple ports. The existing
driver design expects the user to delete all port configurations and create
a configuration with a single UDP port for VxLAN offload to be re-enabled.
Remove this restriction by tracking the ports added and re-enabling offload
when ports get deleted and only 1 port is left.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Reviewed-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene-v2: Extend ethtool statistics
Iyappan Subramanian [Mon, 17 Apr 2017 23:47:55 +0000 (16:47 -0700)]
drivers: net: xgene-v2: Extend ethtool statistics

This patch adds extended statistics reporting to ethtool.

In summary, this patch,

   - adds ethtool.h with the statistics register definitions
   - adds 'struct xge_gstrings_extd_stats' to gather extended stats
   - modifies xge_get_strings(), get_sset_count() and
     get_ethtool_stats() accordingly
   - moves 'struct xge_gstrings_stats' to ethtool.h

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoixgbe: Avoid Tx hang by not allowing more than the number of VFs supported.
Usha Ketineni [Fri, 23 Dec 2016 18:08:14 +0000 (10:08 -0800)]
ixgbe: Avoid Tx hang by not allowing more than the number of VFs supported.

When DCB is enabled, add checks to ensure creation of number of VF's is
valid based on the traffic classes configured by the device.

Signed-off-by: Usha Ketineni <usha.k.ketineni@intel.com>
Tested-by: Ronald Bynoe <ronald.j.bynoe@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoMerge branch 'ftgmac100-batch5-features'
David S. Miller [Tue, 18 Apr 2017 18:11:10 +0000 (14:11 -0400)]
Merge branch 'ftgmac100-batch5-features'

Benjamin Herrenschmidt says:

====================
ftgmac100: Rework batch 5 - Features

This is the third spin of the fifth and last batch of
updates to the ftgmac100 driver.

This contains a few additional "features" such as:

 - Support for ethtool n-way reset
 - Multicast filtering & promisc support
 - Vlan offload
 - netpoll

And a couple of misc bits. This also adds the device-tree binding
documentation.

v2. - Addresses review comments and adds a new patch fixing a
      theorical ordering issue in my new NAPI poll implementation
    - Add a bug fix (Patch 8/9) for a potential ordering issue
      in the new NAPI poll code.

v3. - Rebase on net-next (fix conflict with an unrelated #include
      change series)
    - Update DT bindings better describing accepted phy-mode values
====================

Signed-off-by: David S. Miller <davem@davemloft.net>