]> git.proxmox.com Git - mirror_ubuntu-kernels.git/log
mirror_ubuntu-kernels.git
7 years agoqed: output the DPM status and WID count
Ram Amrani [Sun, 30 Apr 2017 08:49:10 +0000 (11:49 +0300)]
qed: output the DPM status and WID count

Output to the RDMA driver whether DPM mode is enabled or disabled in
the HW and if so what is the number of WIDs it supports

Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: align DPI configuration to HW requirements
Ram Amrani [Sun, 30 Apr 2017 08:49:09 +0000 (11:49 +0300)]
qed: align DPI configuration to HW requirements

When calculating doorbell BAR partitioning round up the number of
CPUs to the nearest power of 2 so the size of the DPI (per user
section) configured in the hardware will be stored properly and
not truncated.

Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: verify RoCE resource bitmaps are released
Ram Amrani [Sun, 30 Apr 2017 08:49:08 +0000 (11:49 +0300)]
qed: verify RoCE resource bitmaps are released

Add mechanism to verify RoCE resources are released prior to freeing the
bitmaps. If this is not the case, print what resources were not released.

Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: add error handling flow to TID deregistratin posting failure
Ram Amrani [Sun, 30 Apr 2017 08:49:07 +0000 (11:49 +0300)]
qed: add error handling flow to TID deregistratin posting failure

If the posting of the ramrod for the purpose of TID deregistration
fails, abort the deregistration operation without using the FW's
return code.

Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: remove unused SQ error state
Ram Amrani [Sun, 30 Apr 2017 08:49:06 +0000 (11:49 +0300)]
qed: remove unused SQ error state

The internal RoCE SQE QP state isn't being used. Instead we mark the
QP as in regular error state.

Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: configure the RoCE max message size
Ram Amrani [Sun, 30 Apr 2017 08:49:05 +0000 (11:49 +0300)]
qed: configure the RoCE max message size

Signed-off-by: Ram Amrani <Ram.Amrani@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: enhance verifier to understand stack pointer arithmetic
Yonghong Song [Sun, 30 Apr 2017 05:52:42 +0000 (22:52 -0700)]
bpf: enhance verifier to understand stack pointer arithmetic

llvm 4.0 and above generates the code like below:
....
440: (b7) r1 = 15
441: (05) goto pc+73
515: (79) r6 = *(u64 *)(r10 -152)
516: (bf) r7 = r10
517: (07) r7 += -112
518: (bf) r2 = r7
519: (0f) r2 += r1
520: (71) r1 = *(u8 *)(r8 +0)
521: (73) *(u8 *)(r2 +45) = r1
....
and the verifier complains "R2 invalid mem access 'inv'" for insn #521.
This is because verifier marks register r2 as unknown value after #519
where r2 is a stack pointer and r1 holds a constant value.

Teach verifier to recognize "stack_ptr + imm" and
"stack_ptr + reg with const val" as valid stack_ptr with new offset.

Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobenet: Use time_before_eq for time comparison
Karim Eshapa [Mon, 1 May 2017 13:58:08 +0000 (15:58 +0200)]
benet: Use time_before_eq for time comparison

Use time_before_eq for time comparison more safe and dealing
with timer wrapping to be future-proof.

Signed-off-by: Karim Eshapa <karim.eshapa@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoflower: check unused bits in MPLS fields
Benjamin LaHaise [Mon, 1 May 2017 13:58:40 +0000 (09:58 -0400)]
flower: check unused bits in MPLS fields

Since several of the the netlink attributes used to configure the flower
classifier's MPLS TC, BOS and Label fields have additional bits which are
unused, check those bits to ensure that they are actually 0 as suggested
by Jamal.

Signed-off-by: Benjamin LaHaise <benjamin.lahaise@netronome.com>
Cc: David Miller <davem@davemloft.net>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Simon Horman <simon.horman@netronome.com>
Cc: Jakub Kicinski <kubakici@wp.pl>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Mon, 1 May 2017 14:46:50 +0000 (10:46 -0400)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter/IPVS updates for net-next

The following patchset contains Netfilter updates for your net-next
tree. A large bunch of code cleanups, simplify the conntrack extension
codebase, get rid of the fake conntrack object, speed up netns by
selective synchronize_net() calls. More specifically, they are:

1) Check for ct->status bit instead of using nfct_nat() from IPVS and
   Netfilter codebase, patch from Florian Westphal.

2) Use kcalloc() wherever possible in the IPVS code, from Varsha Rao.

3) Simplify FTP IPVS helper module registration path, from Arushi Singhal.

4) Introduce nft_is_base_chain() helper function.

5) Enforce expectation limit from userspace conntrack helper,
   from Gao Feng.

6) Add nf_ct_remove_expect() helper function, from Gao Feng.

7) NAT mangle helper function return boolean, from Gao Feng.

8) ctnetlink_alloc_expect() should only work for conntrack with
   helpers, from Gao Feng.

9) Add nfnl_msg_type() helper function to nfnetlink to build the
   netlink message type.

10) Get rid of unnecessary cast on void, from simran singhal.

11) Use seq_puts()/seq_putc() instead of seq_printf() where possible,
    also from simran singhal.

12) Use list_prev_entry() from nf_tables, from simran signhal.

13) Remove unnecessary & on pointer function in the Netfilter and IPVS
    code.

14) Remove obsolete comment on set of rules per CPU in ip6_tables,
    no longer true. From Arushi Singhal.

15) Remove duplicated nf_conntrack_l4proto_udplite4, from Gao Feng.

16) Remove unnecessary nested rcu_read_lock() in
    __nf_nat_decode_session(). Code running from hooks are already
    guaranteed to run under RCU read side.

17) Remove deadcode in nf_tables_getobj(), from Aaron Conole.

18) Remove double assignment in nf_ct_l4proto_pernet_unregister_one(),
    also from Aaron.

19) Get rid of unsed __ip_set_get_netlink(), from Aaron Conole.

20) Don't propagate NF_DROP error to userspace via ctnetlink in
    __nf_nat_alloc_null_binding() function, from Gao Feng.

21) Revisit nf_ct_deliver_cached_events() to remove unnecessary checks,
    from Gao Feng.

22) Kill the fake untracked conntrack objects, use ctinfo instead to
    annotate a conntrack object is untracked, from Florian Westphal.

23) Remove nf_ct_is_untracked(), now obsolete since we have no
    conntrack template anymore, from Florian.

24) Add event mask support to nft_ct, also from Florian.

25) Move nf_conn_help structure to
    include/net/netfilter/nf_conntrack_helper.h.

26) Add a fixed 32 bytes scratchpad area for conntrack helpers.
    Thus, we don't deal with variable conntrack extensions anymore.
    Make sure userspace conntrack helper doesn't go over that size.
    Remove variable size ct extension infrastructure now this code
    got no more clients. From Florian Westphal.

27) Restore offset and length of nf_ct_ext structure to 8 bytes now
    that wraparound is not possible any longer, also from Florian.

28) Allow to get rid of unassured flows under stress in conntrack,
    this applies to DCCP, SCTP and TCP protocols, from Florian.

29) Shrink size of nf_conntrack_ecache structure, from Florian.

30) Use TCP_MAX_WSCALE instead of hardcoded 14 in TCP tracker,
    from Gao Feng.

31) Register SYNPROXY hooks on demand, from Florian Westphal.

32) Use pernet hook whenever possible, instead of global hook
    registration, from Florian Westphal.

33) Pass hook structure to ebt_register_table() to consolidate some
    infrastructure code, from Florian Westphal.

34) Use consume_skb() and return NF_STOLEN, instead of NF_DROP in the
    SYNPROXY code, to make sure device stats are not fooled, patch
    from Gao Feng.

35) Remove NF_CT_EXT_F_PREALLOC this kills quite some code that we
    don't need anymore if we just select a fixed size instead of
    expensive runtime time calculation of this. From Florian.

36) Constify nf_ct_extend_register() and nf_ct_extend_unregister(),
    from Florian.

37) Simplify nf_ct_ext_add(), this kills nf_ct_ext_create(), from
    Florian.

38) Attach NAT extension on-demand from masquerade and pptp helper
    path, from Florian.

39) Get rid of useless ip_vs_set_state_timeout(), from Aaron Conole.

40) Speed up netns by selective calls of synchronize_net(), from
    Florian Westphal.

41) Silence stack size warning gcc in 32-bit arch in snmp helper,
    from Florian.

42) Inconditionally call nf_ct_ext_destroy(), even if we have no
    extensions, to deal with the NF_NAT_MANIP_SRC case. Patch from
    Liping Zhang.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bpf-samples-skb_mode-bug-fixes'
David S. Miller [Mon, 1 May 2017 14:42:38 +0000 (10:42 -0400)]
Merge branch 'bpf-samples-skb_mode-bug-fixes'

Jesper Dangaard Brouer says:

====================
samples/bpf: two bug fixes to XDP_FLAGS_SKB_MODE attaching

Two small bugfixes for:
 commit 3993f2cb983b ("samples/bpf: Add support for SKB_MODE to xdp1 and xdp_tx_iptunnel")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: fix XDP_FLAGS_SKB_MODE detach for xdp_tx_iptunnel
Jesper Dangaard Brouer [Mon, 1 May 2017 09:26:20 +0000 (11:26 +0200)]
samples/bpf: fix XDP_FLAGS_SKB_MODE detach for xdp_tx_iptunnel

The xdp_tx_iptunnel program can be terminated in two ways, after
N-seconds or via Ctrl-C SIGINT.  The SIGINT code path does not
handle detatching the correct XDP program, in-case the program
was attached with XDP_FLAGS_SKB_MODE.

Fix this by storing the XDP flags as a global variable, which is
available for the SIGINT handler function.

Fixes: 3993f2cb983b ("samples/bpf: Add support for SKB_MODE to xdp1 and xdp_tx_iptunnel")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: fix SKB_MODE flag to be a 32-bit unsigned int
Jesper Dangaard Brouer [Mon, 1 May 2017 09:26:15 +0000 (11:26 +0200)]
samples/bpf: fix SKB_MODE flag to be a 32-bit unsigned int

The kernel side of XDP_FLAGS_SKB_MODE is unsigned, and the rtnetlink
IFLA_XDP_FLAGS is defined as NLA_U32. Thus, userspace programs under
samples/bpf/ should use the correct type.

Fixes: 3993f2cb983b ("samples/bpf: Add support for SKB_MODE to xdp1 and xdp_tx_iptunnel")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'xdp-netlink-ext-ack'
David S. Miller [Mon, 1 May 2017 14:35:48 +0000 (10:35 -0400)]
Merge branch 'xdp-netlink-ext-ack'

Jakub Kicinski says:

====================
xdp: use netlink extended ACK reporting

This series is an attempt to make XDP more user friendly by
enabling exploiting the recently added netlink extended ACK
reporting to carry messages to user space.

David Ahern's iproute2 ext ack patches for ip link are sufficient
to show the errors like this:

Error: nfp: MTU too large w/ XDP enabled

Where the message is coming directly from the driver.  There could
still be a bit of a leap for a complete novice from the message
above to the right settings, but it's a big improvement over the
standard "Invalid argument" message.

v1/non-rfc:
 - add a separate macro in patch 1;
 - add KBUILD_MODNAME as part of the message (Daniel);
 - don't print the error to logs in patch 1.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: make use of extended ack message reporting
Jakub Kicinski [Mon, 1 May 2017 04:46:48 +0000 (21:46 -0700)]
virtio_net: make use of extended ack message reporting

Try to carry error messages to the user via the netlink extended
ack message attribute.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: make use of extended ack message reporting
Jakub Kicinski [Mon, 1 May 2017 04:46:47 +0000 (21:46 -0700)]
nfp: make use of extended ack message reporting

Try to carry error messages to the user via the netlink extended
ack message attribute.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoxdp: propagate extended ack to XDP setup
Jakub Kicinski [Mon, 1 May 2017 04:46:46 +0000 (21:46 -0700)]
xdp: propagate extended ack to XDP setup

Drivers usually have a number of restrictions for running XDP
- most common being buffer sizes, LRO and number of rings.
Even though some drivers try to be helpful and print error
messages experience shows that users don't often consult
kernel logs on netlink errors.  Try to use the new extended
ack mechanism to carry the message back to user space.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetlink: add NULL-friendly helper for setting extended ACK message
Jakub Kicinski [Mon, 1 May 2017 04:46:45 +0000 (21:46 -0700)]
netlink: add NULL-friendly helper for setting extended ACK message

As we propagate extended ack reporting throughout various paths in
the kernel it may be that the same function is called with the
extended ack parameter passed as NULL.  One place where that happens
is in drivers which have a centralized reconfiguration function
called both from ndos and from ethtool_ops.  Add a new helper for
setting the error message in such conditions.

Existing helper is left as is to encourage propagating the ext act
fully wherever possible.  It also makes it clear in the code which
messages may be lost due to ext ack being NULL.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetfilter: nf_ct_ext: invoke destroy even when ext is not attached
Liping Zhang [Sat, 29 Apr 2017 13:59:49 +0000 (21:59 +0800)]
netfilter: nf_ct_ext: invoke destroy even when ext is not attached

For NF_NAT_MANIP_SRC, we will insert the ct to the nat_bysource_table,
then remove it from the nat_bysource_table via nat_extend->destroy.

But now, the nat extension is attached on demand, so if the nat extension
is not attached, we will not be notified when the ct is destroyed, i.e.
we may fail to remove ct from the nat_bysource_table.

So just keep it simple, even if the extension is not attached, we will
still invoke the related ext->destroy. And this will also preserve the
flexibility for the future extension.

Fixes: 9a08ecfe74d7 ("netfilter: don't attach a nat extension by default")
Signed-off-by: Liping Zhang <zlpnobody@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agoMerge tag 'ipvs3-for-v4.12' of http://git.kernel.org/pub/scm/linux/kernel/git/horms...
Pablo Neira Ayuso [Mon, 1 May 2017 09:45:47 +0000 (11:45 +0200)]
Merge tag 'ipvs3-for-v4.12' of http://git.kernel.org/pub/scm/linux/kernel/git/horms/ipvs-next

Simon Horman says:

====================
Third Round of IPVS Updates for v4.12

please consider these enhancements to IPVS for v4.12.
If it is too late for v4.12 then please consider them for v4.13.

* Remove unused function
* Correct comparison of unsigned value
====================

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: snmp: avoid stack size warning
Florian Westphal [Thu, 27 Apr 2017 14:39:43 +0000 (16:39 +0200)]
netfilter: snmp: avoid stack size warning

net/ipv4/netfilter/nf_nat_snmp_basic.c:1158:1: warning: the frame size
of 1160 bytes is larger than 1024 bytes

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_queue: only call synchronize_net twice if nf_queue is active
Florian Westphal [Mon, 24 Apr 2017 13:37:41 +0000 (15:37 +0200)]
netfilter: nf_queue: only call synchronize_net twice if nf_queue is active

nf_unregister_net_hook(s) can avoid a second call to synchronize_net,
provided there is no nfqueue active in that net namespace (which is
the common case).

This also gets rid of the extra arg to nf_queue_nf_hook_drop(), normally
this gets called during netns cleanup so no packets should be queued.

For the rare case of base chain being unregistered or module removal
while nfqueue is in use the extra hiccup due to the packet drops isn't
a big deal.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: nf_log: don't call synchronize_rcu in nf_log_unset
Florian Westphal [Tue, 25 Apr 2017 08:24:03 +0000 (10:24 +0200)]
netfilter: nf_log: don't call synchronize_rcu in nf_log_unset

nf_log_unregister() (which is what gets called in the logger backends
module exit paths) does a (required, module is removed) synchronize_rcu().

But nf_log_unset() is only called from pernet exit handlers. It doesn't
free any memory so there appears to be no need to call synchronize_rcu.

v2: Liping Zhang points out that nf_log_unregister() needs to be called
after pernet unregister, else rmmod would become unsafe.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonetfilter: batch synchronize_net calls during hook unregister
Florian Westphal [Mon, 24 Apr 2017 13:37:39 +0000 (15:37 +0200)]
netfilter: batch synchronize_net calls during hook unregister

synchronize_net is expensive and slows down netns cleanup a lot.

We have two APIs to unregister a hook:
nf_unregister_net_hook (which calls synchronize_net())
and
nf_unregister_net_hooks (calls nf_unregister_net_hook in a loop)

Make nf_unregister_net_hook a wapper around new helper
__nf_unregister_net_hook, which unlinks the hook but does not free it.

Then, we can call that helper in nf_unregister_net_hooks and then
call synchronize_net() only once.

Andrey Konovalov reports this change improves syzkaller fuzzing speed at
least twice.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agonet: phy: Allow BCM5481x PHYs to setup internal TX/RX clock delay
Abhishek Shah [Sun, 30 Apr 2017 05:34:21 +0000 (11:04 +0530)]
net: phy: Allow BCM5481x PHYs to setup internal TX/RX clock delay

This patch allows users to enable/disable internal TX and/or RX
clock delay for BCM5481x series PHYs so as to satisfy RGMII timing
specifications.

On a particular platform, whether TX and/or RX clock delay is required
depends on how PHY connected to the MAC IP. This requirement can be
specified through "phy-mode" property in the platform device tree.

Signed-off-by: Abhishek Shah <abhishek.shah@broadcom.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: sunhme: fix spelling mistakes: "ParityErro" -> "ParityError"
Colin Ian King [Sat, 29 Apr 2017 21:38:57 +0000 (22:38 +0100)]
net: sunhme: fix spelling mistakes: "ParityErro" -> "ParityError"

trivial fix to spelling mistakes in printk message.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnx2x: Align RX buffers
Scott Wood [Sat, 29 Apr 2017 00:17:41 +0000 (19:17 -0500)]
bnx2x: Align RX buffers

The bnx2x driver is not providing proper alignment on the receive buffers it
passes to build_skb(), causing skb_shared_info to be misaligned.
skb_shared_info contains an atomic, and while PPC normally supports
unaligned accesses, it does not support unaligned atomics.

Aligning the size of rx buffers will ensure that page_frag_alloc() returns
aligned addresses.

This can be reproduced on PPC by setting the network MTU to 1450 (or other
non-multiple-of-4) and then generating sufficient inbound network traffic
(one or two large "wget"s usually does it), producing the following oops:

Unable to handle kernel paging request for unaligned access at address 0xc00000ffc43af656
Faulting instruction address: 0xc00000000080ef8c
Oops: Kernel access of bad area, sig: 7 [#1]
SMP NR_CPUS=2048
NUMA
PowerNV
Modules linked in: vmx_crypto powernv_rng rng_core powernv_op_panel leds_powernv led_class nfsd ip_tables x_tables autofs4 xfs lpfc bnx2x mdio libcrc32c crc_t10dif crct10dif_generic crct10dif_common
CPU: 104 PID: 0 Comm: swapper/104 Not tainted 4.11.0-rc8-00088-g4c761da #2
task: c00000ffd4892400 task.stack: c00000ffd4920000
NIP: c00000000080ef8c LR: c00000000080eee8 CTR: c0000000001f8320
REGS: c00000ffffc33710 TRAP: 0600   Not tainted  (4.11.0-rc8-00088-g4c761da)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE>
  CR: 24082042  XER: 00000000
CFAR: c00000000080eea0 DAR: c00000ffc43af656 DSISR: 00000000 SOFTE: 1
GPR00: c000000000907f64 c00000ffffc33990 c000000000dd3b00 c00000ffcaf22100
GPR04: c00000ffcaf22e00 0000000000000000 0000000000000000 0000000000000000
GPR08: 0000000000b80008 c00000ffc43af636 c00000ffc43af656 0000000000000000
GPR12: c0000000001f6f00 c00000000fe1a000 000000000000049f 000000000000c51f
GPR16: 00000000ffffef33 0000000000000000 0000000000008a43 0000000000000001
GPR20: c00000ffc58a90c0 0000000000000000 000000000000dd86 0000000000000000
GPR24: c000007fd0ed10c0 00000000ffffffff 0000000000000158 000000000000014a
GPR28: c00000ffc43af010 c00000ffc9144000 c00000ffcaf22e00 c00000ffcaf22100
NIP [c00000000080ef8c] __skb_clone+0xdc/0x140
LR [c00000000080eee8] __skb_clone+0x38/0x140
Call Trace:
[c00000ffffc33990] [c00000000080fb74] skb_clone+0x74/0x110 (unreliable)
[c00000ffffc339c0] [c000000000907f64] packet_rcv+0x144/0x510
[c00000ffffc33a40] [c000000000827b64] __netif_receive_skb_core+0x5b4/0xd80
[c00000ffffc33b00] [c00000000082b2bc] netif_receive_skb_internal+0x2c/0xc0
[c00000ffffc33b40] [c00000000082c49c] napi_gro_receive+0x11c/0x260
[c00000ffffc33b80] [d000000066483d68] bnx2x_poll+0xcf8/0x17b0 [bnx2x]
[c00000ffffc33d00] [c00000000082babc] net_rx_action+0x31c/0x480
[c00000ffffc33e10] [c0000000000d5a44] __do_softirq+0x164/0x3d0
[c00000ffffc33f00] [c0000000000d60a8] irq_exit+0x108/0x120
[c00000ffffc33f20] [c000000000015b98] __do_irq+0x98/0x200
[c00000ffffc33f90] [c000000000027f14] call_do_irq+0x14/0x24
[c00000ffd4923a90] [c000000000015d94] do_IRQ+0x94/0x110
[c00000ffd4923ae0] [c000000000008d90] hardware_interrupt_common+0x150/0x160

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: bridge: Fix improper taking over HW learned FDB
Arkadi Sharshevsky [Fri, 28 Apr 2017 19:39:07 +0000 (22:39 +0300)]
net: bridge: Fix improper taking over HW learned FDB

Commit 7e26bf45e4cb ("net: bridge: allow SW learn to take over HW fdb
entries") added the ability to "take over an entry which was previously
learned via HW when it shows up from a SW port".

However, if an entry was learned via HW and then a control packet
(e.g., ARP request) was trapped to the CPU, the bridge driver will
update the entry and remove the externally learned flag, although the
entry is still present in HW. Instead, only clear the externally learned
flag in case of roaming.

Fixes: 7e26bf45e4cb ("net: bridge: allow SW learn to take over HW fdb entries")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Arkadi Sharashevsky <arkadis@mellanox.com>
Cc: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipv4: get rid of ip_ra_lock
WANG Cong [Fri, 28 Apr 2017 17:04:29 +0000 (10:04 -0700)]
ipv4: get rid of ip_ra_lock

After commit 1215e51edad1 ("ipv4: fix a deadlock in ip_ra_control")
we always take RTNL lock for ip_ra_control() which is the only place
we update the list ip_ra_chain, so the ip_ra_lock is no longer needed.

As Eric points out, BH does not need to disable either, RCU readers
don't care.

Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: bpf_load.c detect and abort if ELF maps section size is wrong
Jesper Dangaard Brouer [Fri, 28 Apr 2017 14:25:04 +0000 (16:25 +0200)]
samples/bpf: bpf_load.c detect and abort if ELF maps section size is wrong

The struct bpf_map_def was extended in commit fb30d4b71214 ("bpf: Add tests
for map-in-map") with member unsigned int inner_map_idx.  This changed the size
of the maps section in the generated ELF _kern.o files.

Unfortunately the loader in bpf_load.c does not detect or handle this.  Thus,
older _kern.o files became incompatible, and caused hard-to-debug errors
where the syscall validation rejected BPF_MAP_CREATE request.

This patch only detect the situation and aborts load_bpf_file(). It also
add code comments warning people that read this loader for inspiration
for these pitfalls.

Fixes: fb30d4b71214 ("bpf: Add tests for map-in-map")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agolwtunnel: fix error path in lwtunnel_fill_encap()
Dan Carpenter [Fri, 28 Apr 2017 13:03:48 +0000 (16:03 +0300)]
lwtunnel: fix error path in lwtunnel_fill_encap()

We recently added a check to see if nla_nest_start() fails.  There are
two issues with that.  First, if it fails then I don't think we should
call nla_nest_cancel().  Second, it's slightly convoluted but the
current code returns success but we should return -EMSGSIZE instead.

Fixes: a50fe0ffd76f ("lwtunnel: check return value of nla_nest_start")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: silence a locking static checker warning
Dan Carpenter [Fri, 28 Apr 2017 12:57:15 +0000 (15:57 +0300)]
liquidio: silence a locking static checker warning

Presumably we never hit this return, but static checkers complain that
we need to unlock so we may as well fix that.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Unlock on error in qed_vf_pf_acquire()
Dan Carpenter [Fri, 28 Apr 2017 12:56:09 +0000 (15:56 +0300)]
qed: Unlock on error in qed_vf_pf_acquire()

My static checker complains that we're holding a mutex on this error
path.  Let's goto exit instead of returning directly.

Fixes: b0bccb69eba3 ("qed: Change locking scheme for VF channel")
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 agoMerge branch 'hns-deferred-probe'
David S. Miller [Mon, 1 May 2017 02:39:25 +0000 (22:39 -0400)]
Merge branch 'hns-deferred-probe'

lipeng says:

====================
net: hns: bug fix for HNS driver

This patchset add support defered dsaf probe when mdio and
mbigen module is not insmod.

For more details, please refer to individual patch.

change log:
V4 - > V5:
1. Float on net-next;
2. Delete patch "net: hns: fixed bug that skb used after kfree"
   from this patchset;

V3 -> V4:
1. Delete redundant commit message;
2. Add Reviewed-by: Matthias Brugger <mbrugger@suse.com>;

V2 -> V3:
1. Check return value when  platform_get_irq in hns_rcb_get_cfg;

V1 -> V2:
1. Return appropriate errno in hns_mac_register_phy;
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: support deferred probe when no mdio
lipeng [Fri, 28 Apr 2017 06:49:47 +0000 (14:49 +0800)]
net: hns: support deferred probe when no mdio

In the hip06 and hip07 SoCs, phy connect to mdio bus.The mdio
module is probed with module_init, and, as such,
is not guaranteed to probe before the HNS driver. So we need
to support deferred probe.

We check for probe deferral in the mac init, so we not init DSAF
when there is no mdio, and free all resource, to later learn that
we need to defer the probe.

Signed-off-by: lipeng <lipeng321@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Reviewed-by: Matthias Brugger <mbrugger@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: support deferred probe when can not obtain irq
lipeng [Fri, 28 Apr 2017 06:49:46 +0000 (14:49 +0800)]
net: hns: support deferred probe when can not obtain irq

In the hip06 and hip07 SoCs, the interrupt lines from the
DSAF controllers are connected to mbigen hw module.
The mbigen module is probed with module_init, and, as such,
is not guaranteed to probe before the HNS driver. So we need
to support deferred probe.

Signed-off-by: lipeng <lipeng321@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Reviewed-by: Matthias Brugger <mbrugger@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'nfp-XDP_TX-optimizations'
David S. Miller [Mon, 1 May 2017 02:37:01 +0000 (22:37 -0400)]
Merge branch 'nfp-XDP_TX-optimizations'

Jakub Kicinski says:

====================
nfp: optimize XDP TX and small fixes

This series optimizes the nfp XDP TX performance a little bit.
I run quick tests on an Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz.
Single core/queue performance for both touch and drop and touch and
forward is above 20Mpps @64B packets, drop being 2Mpps faster.
I think this is max for a single queue on the low power NFPs.

There are also a few minor fixes included for code in net-next.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: provide 256 bytes of XDP headroom in all configurations
Jakub Kicinski [Fri, 28 Apr 2017 04:06:20 +0000 (21:06 -0700)]
nfp: provide 256 bytes of XDP headroom in all configurations

For legacy reasons NFP FW may be compiled to DMA packets to a constant
offset into the buffer and use the space before it for metadata.  This
ensures that packets data always start at a certain offset regardless of
the amount of preceding metadata.

If rx offset is set to 0 there may still be up to 64 bytes of metadata
but metadata will start at the beginning of the buffer, instead of:

    data_start_offset = rx_offset - meta_len

Even though we make the buffers larger to accommodate up to 64 bytes of
metadata, if there is only N bytes of metadata, we will end up with
N bytes of headroom and 64 - N bytes of tailroom.  Therefore we can't
rely on that space for XDP headroom.  Make sure we always allocate
full 256 bytes.  This, unfortunately, means we can't fit the headroom
on an u8 any more.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: don't completely refuse to work with old flashes
Jakub Kicinski [Fri, 28 Apr 2017 04:06:19 +0000 (21:06 -0700)]
nfp: don't completely refuse to work with old flashes

Right now the required Service Process ABI version is still tied
to max ID of known commands.  For new NSP commands we are adding
we are checking if NSP version is recent enough on command-by-command
basis.  The driver doesn't have to force the device to have the
very latest flash, anything newer than 0.8 should do.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: avoid reading TX queue indexes from the device
Jakub Kicinski [Fri, 28 Apr 2017 04:06:18 +0000 (21:06 -0700)]
nfp: avoid reading TX queue indexes from the device

Reading TX queue indexes from the device memory on each interrupt
is expensive.  It's doubly expensive with XDP running since we have
two TX rings to check there.  If the software indexes indicate that
the TX queue is completely empty, however, we don't need to look at
the device completion index at all.

The queuing CPU is doing a wmb() before kicking the device TX so
we should be safe to assume on the CPU handling the completions will
never see old value of the software copy of the index.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: do simple XDP TX buffer recycling
Jakub Kicinski [Fri, 28 Apr 2017 04:06:17 +0000 (21:06 -0700)]
nfp: do simple XDP TX buffer recycling

On the RX path we follow the "drop if allocation of replacement
buffer fails" rule.  With XDP we extended that to the TX action,
so if XDP prog returned TX but allocation of replacement RX buffer
failed, we will drop the packet.

To improve our XDP TX performance extend the idea of rings being
always full to XDP TX rings.  Pre-fill the XDP TX rings with RX
buffers, and when XDP prog returns TX action swap the RX buffer
with the next buffer from the TX ring.

XDP TX complete will no longer free the buffers but let them
sit on the TX ring and wait for swap with RX buffer, instead.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: drop rx_ring param from buffer allocation
Jakub Kicinski [Fri, 28 Apr 2017 04:06:16 +0000 (21:06 -0700)]
nfp: drop rx_ring param from buffer allocation

We will soon allocate RX buffers for caching on XDP TX rings.
The rx_ring parameter passed to nfp_net_rx_alloc_one() is not
actually used, remove it.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: replace -ENOTSUPP with -EOPNOTSUPP
Jakub Kicinski [Fri, 28 Apr 2017 04:06:15 +0000 (21:06 -0700)]
nfp: replace -ENOTSUPP with -EOPNOTSUPP

As Or points out in commit 423b3aecf290 ("net/mlx4: Change ENOTSUPP
to EOPNOTSUPP"), ENOTSUPP is NFS specific error.  Replace it with
EOPNOTSUPP.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio-net: use netif_tx_napi_add for tx napi
Willem de Bruijn [Fri, 28 Apr 2017 00:37:58 +0000 (20:37 -0400)]
virtio-net: use netif_tx_napi_add for tx napi

Avoid hashing the tx napi struct into napi_hash[], which is used for
busy polling receive queues.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: Initialise init_net.count to 1
David Howells [Thu, 27 Apr 2017 21:40:23 +0000 (22:40 +0100)]
net: Initialise init_net.count to 1

Initialise init_net.count to 1 for its pointer from init_nsproxy lest
someone tries to do a get_net() and a put_net() in a process in which
current->ns_proxy->net_ns points to the initial network namespace.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogeneve: fix incorrect setting of UDP checksum flag
Girish Moodalbail [Thu, 27 Apr 2017 21:11:53 +0000 (14:11 -0700)]
geneve: fix incorrect setting of UDP checksum flag

Creating a geneve link with 'udpcsum' set results in a creation of link
for which UDP checksum will NOT be computed on outbound packets, as can
be seen below.

11: gen0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN
    link/ether c2:85:27:b6:b4:15 brd ff:ff:ff:ff:ff:ff promiscuity 0
    geneve id 200 remote 192.168.13.1 dstport 6081 noudpcsum

Similarly, creating a link with 'noudpcsum' set results in a creation
of link for which UDP checksum will be computed on outbound packets.

Fixes: 9b4437a5b870 ("geneve: Unify LWT and netdev handling.")
Signed-off-by: Girish Moodalbail <girish.moodalbail@oracle.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Acked-by: Lance Richardson <lrichard@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'vxlan-disabled-ipv6'
David S. Miller [Mon, 1 May 2017 02:30:13 +0000 (22:30 -0400)]
Merge branch 'vxlan-disabled-ipv6'

Jiri Benc says:

====================
vxlan: do not error out on disabled IPv6

This patchset fixes a bug with metadata based tunnels when booted with
ipv6.disable=1.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovxlan: do not output confusing error message
Jiri Benc [Thu, 27 Apr 2017 19:24:36 +0000 (21:24 +0200)]
vxlan: do not output confusing error message

The message "Cannot bind port X, err=Y" creates only confusion. In metadata
based mode, failure of IPv6 socket creation is okay if IPv6 is disabled and
no error message should be printed. But when IPv6 tunnel was requested, such
failure is fatal. The vxlan_socket_create does not know when the error is
harmless and when it's not.

Instead of passing such information down to vxlan_socket_create, remove the
message completely. It's not useful. We propagate the error code up to the
user space and the port number comes from the user space. There's nothing in
the message that the process creating vxlan interface does not know.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovxlan: correctly handle ipv6.disable module parameter
Jiri Benc [Thu, 27 Apr 2017 19:24:35 +0000 (21:24 +0200)]
vxlan: correctly handle ipv6.disable module parameter

When IPv6 is compiled but disabled at runtime, __vxlan_sock_add returns
-EAFNOSUPPORT. For metadata based tunnels, this causes failure of the whole
operation of bringing up the tunnel.

Ignore failure of IPv6 socket creation for metadata based tunnels caused by
IPv6 not being available.

Fixes: b1be00a6c39f ("vxlan: support both IPv4 and IPv6 sockets in a single vxlan device")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnx2x: Get rid of useless temporary variable
Andy Shevchenko [Thu, 27 Apr 2017 13:37:01 +0000 (16:37 +0300)]
bnx2x: Get rid of useless temporary variable

Replace pattern

 int status;
 ...
 status = func(...);
 return status;

by

 return func(...);

No functional change intented.

Signed-off-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnx2x: Reuse bnx2x_null_format_ver()
Andy Shevchenko [Thu, 27 Apr 2017 13:37:00 +0000 (16:37 +0300)]
bnx2x: Reuse bnx2x_null_format_ver()

Reuse bnx2x_null_format_ver() in functions where it's appropriated
instead of open coded variant.

Signed-off-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnx2x: Replace custom scnprintf()
Andy Shevchenko [Thu, 27 Apr 2017 13:36:59 +0000 (16:36 +0300)]
bnx2x: Replace custom scnprintf()

Use scnprintf() when printing version instead of custom open coded variants.

Signed-off-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Acked-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'linux-can-next-for-4.12-20170427' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Mon, 1 May 2017 02:25:59 +0000 (22:25 -0400)]
Merge tag 'linux-can-next-for-4.12-20170427' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2017-04-25

this is a pull request of 1 patch for net-next/master.

This patch by Oliver Hartkopp fixes the build of the broad cast manager
with CONFIG_PROC_FS disabled.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: Fix inaccurate helper function description
Chenbo Feng [Wed, 26 Apr 2017 23:41:23 +0000 (16:41 -0700)]
bpf: Fix inaccurate helper function description

The description inside uapi/linux/bpf.h about bpf_get_socket_uid
helper function is no longer valid. It returns overflowuid rather
than 0 when failed.

Signed-off-by: Chenbo Feng <fengc@google.com>
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 agotcp: fix access to sk->sk_state in tcp_poll()
Davide Caratti [Wed, 26 Apr 2017 17:07:35 +0000 (19:07 +0200)]
tcp: fix access to sk->sk_state in tcp_poll()

avoid direct access to sk->sk_state when tcp_poll() is called on a socket
using active TCP fastopen with deferred connect. Use local variable
'state', which stores the result of sk_state_load(), like it was done in
commit 00fd38d938db ("tcp: ensure proper barriers in lockless contexts").

Fixes: 19f6d3f3c842 ("net/tcp-fastopen: Add new API support")
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Wei Wang <weiwan@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: restore skb->sk before pskb_trim() call
Eric Dumazet [Wed, 26 Apr 2017 16:09:23 +0000 (09:09 -0700)]
bpf: restore skb->sk before pskb_trim() call

While testing a fix [1] in ___pskb_trim(), addressing the WARN_ON_ONCE()
in skb_try_coalesce() reported by Andrey, I found that we had an skb
with skb->sk set but no skb->destructor.

This invalidated heuristic found in commit 158f323b9868 ("net: adjust
skb->truesize in pskb_expand_head()") and in cited patch.

Considering the BUG_ON(skb->sk) we have in skb_orphan(), we should
restrain the temporary setting to a minimal section.

[1] https://patchwork.ozlabs.org/patch/755570/
    net: adjust skb->truesize in ___pskb_trim()

Fixes: 8f917bba0042 ("bpf: pass sk to helper functions")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: macb: fix phy interrupt parsing
Alexandre Belloni [Wed, 26 Apr 2017 10:06:28 +0000 (12:06 +0200)]
net: macb: fix phy interrupt parsing

Since 83a77e9ec415, the phydev irq is explicitly set to PHY_POLL when
there is no pdata. It doesn't work on DT enabled platforms because the
phydev irq is already set by libphy before.

Fixes: 83a77e9ec415 ("net: macb: Added PCI wrapper for Platform Driver.")
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.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 [Sun, 30 Apr 2017 15:33:37 +0000 (11:33 -0400)]
Merge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2017-04-30

This series contains updates to i40e and i40evf only.

Jake provides majority of the changes in this series, starting with the
renaming of a flag to avoid confusion.  Then renamed a variable to a
more meaningful name to clarify what is actually being done and to
reduce confusion.  Amortizes the wait time when initializing or disabling
lots of VFs by using i40e_reset_all_vfs() and
i40e_vsi_stop_rings_no_wait().  Cleaned up a unnecessary delay since
pci_disable_sriov() already has its own delay, so need to add a additional
delay when removing VFs.  Avoid using the same name flags for both
vsi->state and pf->state, to make code review easier and assist future
work to use the correct state field when checking bits.  Use
DECLARE_BITMAP() to ensure that we always allocate enough space for flags.
Replace hw_disabled_flags with the new _AUTO_DISABLED flags, which are
more readable because we are not setting an *_ENABLED flag to
disable the feature.

Alex corrects a oversight where we were not reprogramming the ports
after a reset, which was causing us to lose all of the receive tunnel
offloads.

Arnd Bergmann moves the declaration of a local variable to avoid a
warning seen on architectures with larger pages about an unused variable.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sun, 30 Apr 2017 15:30:13 +0000 (11:30 -0400)]
Merge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
1GbE Intel Wired LAN Driver Updates 2017-04-30

This series contains updates to e1000e only.

Jarod Wilson fixes an issue where the workaround for 82574 & 82583
is needed for i218 as well, so set the appropriate flags.

Sasha adds support for the upcoming new i219 devices for the client
platform (CannonLake), which includes the support for 38.4MHz frequency
to support PTP on CannonLake.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoe1000e: Add Support for 38.4MHZ frequency
Sasha Neftin [Thu, 6 Apr 2017 07:27:03 +0000 (10:27 +0300)]
e1000e: Add Support for 38.4MHZ frequency

Add support for 38.4MHz frequency is required for PTP
on CannonLake. SYSTIM frequency adjustment attributes for TIMINCA are
get/set dependent on the hardware clock frequency for a different
types of adapters. 38.4MHz frequency supported by CannonLake
and active once time synchronisation mechanism was enabled
Changed abbreviation from Hz to HZ to be compliant checkpatch code style

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Reviewed-by: Raanan Avargil <raanan.avargil@intel.com>
Reviewed-by: Dima Ruinskiy <dima.ruinskiy@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoe1000e: Add Support for CannonLake
Sasha Neftin [Thu, 6 Apr 2017 07:26:47 +0000 (10:26 +0300)]
e1000e: Add Support for CannonLake

The propagation of CannonLake mac type to driver functionality

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Reviewed-by: Raanan Avargil <raanan.avargil@intel.com>
Reviewed-by: Dima Ruinskiy <dima.ruinskiy@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoe1000e: Initial Support for CannonLake
Sasha Neftin [Thu, 6 Apr 2017 07:26:32 +0000 (10:26 +0300)]
e1000e: Initial Support for CannonLake

i219 (6) and i219 (7) are the next LOM generations that will be
available on the nextIntel Client platform (CannonLake)
This patch provides the initial support for these devices

Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Reviewed-by: Raanan Avargil <raanan.avargil@intel.com>
Reviewed-by: Dima Ruinskiy <dima.ruinskiy@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoe1000e: fix PTP on e1000_pch_lpt variants
Jarod Wilson [Tue, 26 Jul 2016 18:25:35 +0000 (14:25 -0400)]
e1000e: fix PTP on e1000_pch_lpt variants

I've got reports that the Intel I-218V NIC in Intel NUC5i5RYH systems used
as a PTP slave experiences random ~10 hour clock jumps, which are resolved
if the same workaround for the 82574 and 82583 is employed, so set the
appropriate flag2 in e1000_pch_lpt_info too.

Reported-by: Rupesh Patel <rupatel@redhat.com>
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40evf: hide unused variable
Arnd Bergmann [Wed, 19 Apr 2017 17:29:48 +0000 (19:29 +0200)]
i40evf: hide unused variable

On architectures with larger pages, we get a warning about an unused variable:

drivers/net/ethernet/intel/i40evf/i40evf_main.c: In function 'i40evf_configure_rx':
drivers/net/ethernet/intel/i40evf/i40evf_main.c:690:21: error: unused variable 'netdev' [-Werror=unused-variable]

This moves the declaration into the #ifdef to avoid the warning.

Fixes: dab86afdbbd1 ("i40e/i40evf: Change the way we limit the maximum frame size for Rx")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-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 agoi40evf: allocate queues before we setup the interrupts and q_vectors
Jacob Keller [Wed, 19 Apr 2017 13:25:59 +0000 (09:25 -0400)]
i40evf: allocate queues before we setup the interrupts and q_vectors

This matches the ordering of how we free stuff during reset and remove.
It also makes logical sense because we set the interrupts based on the
number of queues. Currently this doesn't really matter in practice.
However a future patch moves the assignment of num_active_queues into
i40evf_alloc_queues, which is required by
i40evf_set_interrupt_capability.

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 agoi40evf: remove I40E_FLAG_FDIR_ATR_ENABLED
Jacob Keller [Wed, 19 Apr 2017 13:25:58 +0000 (09:25 -0400)]
i40evf: remove I40E_FLAG_FDIR_ATR_ENABLED

The flag used by the common code and PF code is I40E_FLAG_FD_ATR_ENABLED,
not *FDIR*. It turns out none of the txrx code actually shared with the
VF driver actually checks the ATR flag. This is made even more obvious
by the typo in the VF header file.

Let's just remove the flag from the VF driver since it's not needed.

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 hw_disabled_flags in favor of using separate flag bits
Jacob Keller [Wed, 19 Apr 2017 13:25:57 +0000 (09:25 -0400)]
i40e: remove hw_disabled_flags in favor of using separate flag bits

The hw_disabled_flags field was added as a way of signifying that
a feature was automatically or temporarily disabled. However, we
actually only use this for FDir features. Replace its use with new
_AUTO_DISABLED flags instead. This is more readable, because you aren't
setting an *_ENABLED flag to *disable* the feature.

Additionally, clean up a few areas where we used these bits. First, we
don't really need to set the auto-disable flag for ATR if we're fully
disabling the feature via ethtool.

Second, we should always clear the auto-disable bits in case they somehow
got set when the feature was disabled. However, avoid displaying
a message that we've re-enabled the feature.

Third, we shouldn't be re-enabling ATR in the SB ntuple add flow,
because it might have been disabled due to space constraints. Instead,
we should just wait for the fdir_check_and_reenable to be called by the
watchdog.

Overall, this change allows us to simplify some code by removing an
extra field we didn't need, and the result should make it more clear as
to what we're actually doing with these flags.

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 agoi40evf: remove needless min_t() on num_online_cpus()*2
Jacob Keller [Wed, 19 Apr 2017 13:25:56 +0000 (09:25 -0400)]
i40evf: remove needless min_t() on num_online_cpus()*2

We already set pairs to the value of adapter->num_active_queues. This
value is limited by vsi_res->num_queue_pairs and num_online_cpus(). This
means that pairs by definition is already smaller than
num_online_cpus()*2, so we don't even need to bother with this check.

Lets just remove it and update the comment.

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: use DECLARE_BITMAP for state fields
Jacob Keller [Wed, 19 Apr 2017 13:25:55 +0000 (09:25 -0400)]
i40e: use DECLARE_BITMAP for state fields

Instead of assuming our flags fit within an unsigned long, use
DECLARE_BITMAP which will ensure that we always allocate enough space.
Additionally, use __I40E_STATE_SIZE__ markers as the last element of the
enumeration so that the size of the BITMAP is compile-time assigned
rather than programmer-time assigned. This ensures that potential future
flag additions do not actually overrun the array. This is especially
important as 32bit systems would only have 32bit longs instead of 64bit
longs as we generally have assumed in the prior code.

This change also removes a dereference of the state fields throughout
the code, so it does have a bit of code churn. The conversions were
automated using sed replacements with an alternation

  s/&(vsi->back|vsi|pf)->state/\1->state/
  s/&adapter->vsi.state/adapter->vsi.state/

For debugfs, we modify the printing so that we can display chunks of the
state value on new lines. This ensures that we can print the entire set
of state values. Additionally, we now print them as 08lx to ensure that
they display nicely.

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: separate PF and VSI state flags
Jacob Keller [Fri, 21 Apr 2017 20:38:05 +0000 (13:38 -0700)]
i40e: separate PF and VSI state flags

Avoid using the same named flags for both vsi->state and pf->state. This
makes code review easier, as it is more likely that future authors will
use the correct state field when checking bits. Previous commits already
found issues with at least one check, and possibly others may be
incorrect.

This reduces confusion as it is more clear what each flag represents,
and which flags are valid for which state field.

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 unnecessary msleep() delay in i40e_free_vfs
Jacob Keller [Wed, 19 Apr 2017 13:25:53 +0000 (09:25 -0400)]
i40e: remove unnecessary msleep() delay in i40e_free_vfs

The delay was added because of a desire to ensure that the VF driver can
finish up removing. However, pci_disable_sriov already has its own
ssleep() call that will sleep for an entire second, so there is no
reason to add extra delay on top of this by using msleep here. In
practice, an msleep() won't have a huge impact on timing but there is no
real value in keeping it, so lets just simplify the code and remove it.

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: amortize wait time when disabling lots of VFs
Jacob Keller [Wed, 19 Apr 2017 13:25:52 +0000 (09:25 -0400)]
i40e: amortize wait time when disabling lots of VFs

Just as we do in i40e_reset_all_vfs, save some time when freeing VFs by
amortizing the wait time for stopping queues. We can use
i40e_vsi_stop_rings_no_wait() to begin the process of stopping all the
VF rings at once. Then, once we've started the process on each VF we can
begin waiting for the VFs to stop. This helps reduce the total wait time
by a large factor.

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: Reprogram port offloads after reset
Alexander Duyck [Wed, 19 Apr 2017 13:25:51 +0000 (09:25 -0400)]
i40e: Reprogram port offloads after reset

This patch corrects a major oversight in that we were not reprogramming the
ports after a reset.  As a result we completely lost all of the Rx tunnel
offloads on receive including Rx checksum, RSS on inner headers, and ATR.

The fix for this is pretty standard as all we needed to do is reset the
filter bits to pending for all active filters and schedule the sync event.

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 agoi40e: rename index to port to avoid confusion
Jacob Keller [Wed, 19 Apr 2017 13:25:50 +0000 (09:25 -0400)]
i40e: rename index to port to avoid confusion

The .index field of i40e_udp_port_config represents the udp port number.
Rename this variable to port so that it is more obvious.

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: make use of i40e_reset_all_vfs when initializing new VFs
Jacob Keller [Thu, 13 Apr 2017 08:45:55 +0000 (04:45 -0400)]
i40e: make use of i40e_reset_all_vfs when initializing new VFs

When allocating a large number of VFs, the driver previously used
i40e_reset_vf in a sequence. Just as when performing a normal reset,
this accumulates a large amount of delay for handling all of the VFs in
sequence. This delay is mainly due to a hardware requirement to wait
after initiating a reset on the VF.

We recently added a new function, i40e_reset_all_vfs() which can be used
to amortize the delay time, by first triggering all VF resets, then
waiting once, and finally cleaning up and allocating the VFs. This is
almost as good as truly running the resets in parallel.

In order to avoid sending a spurious reset message to a client
interface, we have a check to see whether we've assigned
pf->num_alloc_vfs yet. This was originally intended as a way to
distinguish the "initialization" case from the regular reset case.

Unfortunately, this means that we can't directly use i40e_reset_all_vfs
yet. Lets avoid this check of pf->num_alloc_vfs by replacing it with
a proper VSI state bit which we can use instead. This makes the
intention much clearer and allows us to re-use the i40e_reset_all_vfs
function directly.

Change-ID: I694279b37eb6b5a91b6670182d0c15d10244fd6e
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-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: properly spell I40E_VF_STATE_* flags
Jacob Keller [Thu, 13 Apr 2017 08:45:54 +0000 (04:45 -0400)]
i40e: properly spell I40E_VF_STATE_* flags

These flags represent the state of the VF at various times. Do not
spell them as _STAT_ which can be confusing to readers who may think
these refer to statistics.

Change-ID: I6bc092cd472e8276896a1fd7498aced2084312df
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 agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sun, 30 Apr 2017 03:16:20 +0000 (23:16 -0400)]
Merge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
10GbE Intel Wired LAN Driver Updates 2017-04-29

This series contains updates to ixgbe and ixgbevf only, most notable is
the addition of XDP support to our 10GbE drivers.

Paul fixes ixgbe to acquire the PHY semaphore before accessing PHY
registers when issuing a device reset.

John adds XDP support (yeah!) for ixgbe.

Emil fixes an issue by flushing the MACVLAN filters on VF reset to avoid
conflicts with other VFs that may end up using the same MAC address.  Also
fixed a bug where ethtool -S displayed some empty fields for ixgbevf
because it was using ixgbe_stats instead ixgbevf_stats for
IXGBEVF_QUEUE_STATS_LEN.

Tony adds the ability to specify a zero MAC address in order to clear the
VF's MAC address from the RAR table.  Also adds support for a new
1000Base-T device based on x550EM_X MAC type.  Fixed an issue where the
RSS key specified by the user would be over-written with a pre-existing
value, so change the rss_key to a pointer so we can check to see if the
key has a value set before attempting to set it.  Fixed the logic for
mailbox support for getting RETA and RSS values, which are only supported
by 82599 and x540 devices.

v2: fixed up patches #2 and #3 based on feedback from Jakub and to
    address build issues when page sizes are larger than 4k
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoixgbevf: Check for RSS key before setting value
Tony Nguyen [Thu, 13 Apr 2017 14:26:07 +0000 (07:26 -0700)]
ixgbevf: Check for RSS key before setting value

The RSS key is being repopulated every time the interface is brought up
regardless of whether there is an existing value. If the user sets the RSS
key and the interface is brought up (e.g. reset), the user specified RSS
key will be overwritten.

This patch changes the rss_key to a pointer so we can check to see if the
key has been populated and preserve it accordingly.

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: Fix errors in retrieving RETA and RSS from PF
Tony Nguyen [Thu, 13 Apr 2017 14:26:06 +0000 (07:26 -0700)]
ixgbevf: Fix errors in retrieving RETA and RSS from PF

Mailbox support for getting RETA and RSS is available for only 82599 and
x540; a previous patch reversed the logic and these adapters were
returning not supported.

Also, the NACK check in ixgbevf_get_rss_key_locked() was checking for the
command IXGBE_VF_GET_RETA instead of IXGBE_VF_GET_RSS_KEY.

This patch corrects both issues by correcting the logic and checking for
the right command.

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 agoixgbe: Check for RSS key before setting value
Tony Nguyen [Thu, 13 Apr 2017 14:26:05 +0000 (07:26 -0700)]
ixgbe: Check for RSS key before setting value

The RSS key is being repopulated every time the interface is brought up
regardless of whether there is an existing value. If the user sets the RSS
key and the interface is brought up (e.g. reset), the user specified RSS
key will be overwritten.

This patch changes the rss_key to a pointer so we can check to see if the
key has been populated and preserve it accordingly.

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 agoixgbe: Add 1000Base-T device based on X550EM_X MAC
Paul Greenwalt [Thu, 13 Apr 2017 12:07:07 +0000 (08:07 -0400)]
ixgbe: Add 1000Base-T device based on X550EM_X MAC

Add support for new 1000Base-T device based on X550EM_X MAC
type. All PHY operations are disabled as the PHY is controlled
by FW.

Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Allow setting zero MAC address for VF
Tony Nguyen [Wed, 12 Apr 2017 20:35:22 +0000 (13:35 -0700)]
ixgbe: Allow setting zero MAC address for VF

Currently, there is no logic that allows a VF's MAC address to be removed
from the RAR table.

Allow the user to specify a zero MAC address in order to clear the VF's
MAC address from the RAR table.  This functionality is also utilized by
libvirt when removing VFs.

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: fix size of queue stats length
Emil Tantilov [Fri, 31 Mar 2017 03:49:02 +0000 (20:49 -0700)]
ixgbevf: fix size of queue stats length

IXGBEVF_QUEUE_STATS_LEN is based on ixgebvf_stats, not ixgbe_stats.

This change fixes a bug where ethtool -S displayed some empty fields.

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: clean macvlan MAC filter table on VF reset
Emil Tantilov [Tue, 28 Mar 2017 18:27:54 +0000 (11:27 -0700)]
ixgbe: clean macvlan MAC filter table on VF reset

Flush the macvlan filters on VF reset to avoid conflict with other VFs that
may end up using the same MAC address.

The main change here is the call to ixgbe_set_vf_macvlan() with index 0.

Moved ixgbe_set_vf_macvlan() in front of ixgbe_vf_reset_event() to avoid
adding a prototype.

Reported-by: Sritej Kanakadandi Sritej Rama <skanakad@cisco.com>
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: delay tail write to every 'n' packets
John Fastabend [Tue, 28 Mar 2017 16:47:03 +0000 (09:47 -0700)]
ixgbe: delay tail write to every 'n' packets

Current XDP implementation hits the tail on every XDP_TX return
code. This patch changes driver behavior to only hit the tail after
packet processing is complete.

With this patch I can run XDP drop programs @ 14+Mpps and XDP_TX
programs are at ~13.5Mpps.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: add support for XDP_TX action
John Fastabend [Mon, 24 Apr 2017 10:30:18 +0000 (03:30 -0700)]
ixgbe: add support for XDP_TX action

A couple design choices were made here. First I use a new ring
pointer structure xdp_ring[] in the adapter struct instead of
pushing the newly allocated XDP TX rings into the tx_ring[]
structure. This means we have to duplicate loops around rings
in places we want to initialize both TX rings and XDP rings.
But by making it explicit it is obvious when we are using XDP
rings and when we are using TX rings. Further we don't have
to do ring arithmatic which is error prone. As a proof point
for doing this my first patches used only a single ring structure
and introduced bugs in FCoE code and macvlan code paths.

Second I am aware this is not the most optimized version of
this code possible. I want to get baseline support in using
the most readable format possible and then once this series
is included I will optimize the TX path in another series
of patches.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: add XDP support for pass and drop actions
John Fastabend [Mon, 24 Apr 2017 10:30:17 +0000 (03:30 -0700)]
ixgbe: add XDP support for pass and drop actions

Basic XDP drop support for ixgbe. Uses READ_ONCE/xchg semantics on XDP
programs instead of RCU primitives as suggested by Daniel Borkmann and
Alex Duyck.

v2: fix the build issues seen w/ XDP when page sizes are larger than 4K
    and made minor fixes based on feedback from Jakub Kicinski

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Acquire PHY semaphore before device reset
Paul Greenwalt [Mon, 13 Mar 2017 09:47:56 +0000 (05:47 -0400)]
ixgbe: Acquire PHY semaphore before device reset

A recent firmware change fixed an issue to acquire the PHY semaphore before
accessing PHY registers. This led to a case where  SW can issue a device
reset clearing the MDIO registers. This patch makes SW acquire the PHY
semaphore before issuing a device reset.

Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoMerge branch 'bpf-misc-next'
David S. Miller [Fri, 28 Apr 2017 19:48:16 +0000 (15:48 -0400)]
Merge branch 'bpf-misc-next'

Daniel Borkmann says:

====================
Misc BPF updates

This set cleans up ldimm64 leftovers from early eBPF days and
adds couple of test cases related to this to the verifier test
suite. It also cleans up the kallsyms spinlock (had same patch
also in queue) by relaxing it through switching to _bh variant.
It fixes up test_progs in relation to htons/ntohs and adds
accessor macros for the percpu tests in test_maps.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: provide a generic macro for percpu values for selftests
Daniel Borkmann [Wed, 26 Apr 2017 23:39:35 +0000 (01:39 +0200)]
bpf: provide a generic macro for percpu values for selftests

To overcome bugs as described and fixed in 89087c456fb5 ("bpf: Fix
values type used in test_maps"), provide a generic BPF_DECLARE_PERCPU()
and bpf_percpu() accessor macro for all percpu map values used in
tests.

Declaring variables works as follows (also works for structs):

  BPF_DECLARE_PERCPU(uint32_t, my_value);

They can then be accessed normally as uint32_t type through:

  bpf_percpu(my_value, <cpu_nr>)

For example:

  bpf_percpu(my_value, 0)++;

Implicitly, we make sure that the passed type is allocated and aligned
by gcc at least on a 8-byte boundary, so that it works together with
the map lookup/update syscall for percpu maps. We use it as a usage
example in test_maps, so that others are free to adapt this into their
code when necessary.

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 agobpf: fix _htons occurences in test_progs
Daniel Borkmann [Wed, 26 Apr 2017 23:39:34 +0000 (01:39 +0200)]
bpf: fix _htons occurences in test_progs

Dave reported that on sparc test_progs generates buggy swapped
eth->h_proto protocol comparisons:

  10: (15) if r3 == 0xdd86 goto pc+9
   R0=imm2,min_value=2,max_value=2 R1=pkt(id=0,off=0,r=14) R2=pkt_end R3=inv
   R4=pkt(id=0,off=14,r=14) R5=inv56 R10=fp

This is due to the unconditional ...

  #define htons __builtin_bswap16
  #define ntohs __builtin_bswap16

... in test_progs that causes this. Make use of asm/byteorder.h
and use __constant_htons() where possible and only perform the
bswap16 when on little endian in non-constant case.

Fixes: 6882804c916b ("selftests/bpf: add a test for overlapping packet range checks")
Fixes: 37821613626e ("selftests/bpf: add l4 load balancer test based on sched_cls")
Reported-by: David S. Miller <davem@davemloft.net>
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 agobpf: bpf_lock on kallsysms doesn't need to be irqsave
Hannes Frederic Sowa [Wed, 26 Apr 2017 23:39:33 +0000 (01:39 +0200)]
bpf: bpf_lock on kallsysms doesn't need to be irqsave

Hannes rightfully spotted that the bpf_lock doesn't need to be
irqsave variant. We never perform any such updates where this
would be necessary (neither right now nor in future), therefore
relax this further.

Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
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 agobpf: add various test cases to verifier selftests
Daniel Borkmann [Wed, 26 Apr 2017 23:39:32 +0000 (01:39 +0200)]
bpf: add various test cases to verifier selftests

Add several test cases around ldimm64, fp arithmetic and direct
packet access.

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 agobpf, x86_64/arm64: remove old ldimm64 artifacts from jits
Daniel Borkmann [Wed, 26 Apr 2017 23:39:31 +0000 (01:39 +0200)]
bpf, x86_64/arm64: remove old ldimm64 artifacts from jits

For both cases, the verifier is already rejecting such invalid
formed instructions. Thus, remove these artifacts from old times
and align it with ppc64, sparc64 and s390x JITs that don't have
them in the first place.

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 agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec...
David S. Miller [Fri, 28 Apr 2017 19:43:24 +0000 (15:43 -0400)]
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
pull request (net-next): ipsec-next 2017-04-28

Just one patch to fix a misplaced spin_unlock_bh in an error path.

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

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'mac80211-next-for-davem-2017-04-28' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Fri, 28 Apr 2017 18:41:15 +0000 (14:41 -0400)]
Merge tag 'mac80211-next-for-davem-2017-04-28' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next

Johannes Berg says:

====================
Another set of patches for -next:
 * API support for concurrent scheduled scan requests
 * API changes for roaming reporting
 * BSS max idle support in mac80211
 * API changes for TX status reporting in mac80211
 * API changes for RX rate reporting in mac80211
 * rewrite monitor logic to prepare for BPF filters
 * bugfix for rare devices without 2.4 GHz support
 * a bugfix for recent DFS changes
 * some further cleanups

The API changes are actually at a nice time, since it's
typically quiet just before the merge window, and trees
can be synchronized easily during it.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agorhashtable: Do not lower max_elems when max_size is zero
Herbert Xu [Fri, 28 Apr 2017 06:10:48 +0000 (14:10 +0800)]
rhashtable: Do not lower max_elems when max_size is zero

The commit 6d684e54690c ("rhashtable: Cap total number of entries
to 2^31") breaks rhashtable users that do not set max_size.  This
is because when max_size is zero max_elems is also incorrectly set
to zero instead of 2^31.

This patch fixes it by only lowering max_elems when max_size is not
zero.

Fixes: 6d684e54690c ("rhashtable: Cap total number of entries to 2^31")
Reported-by: Florian Fainelli <f.fainelli@gmail.com>
Reported-by: kernel test robot <fengguang.wu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocfg80211: add request id to cfg80211_sched_scan_*() api
Arend Van Spriel [Fri, 28 Apr 2017 12:40:28 +0000 (13:40 +0100)]
cfg80211: add request id to cfg80211_sched_scan_*() api

Have proper request id filled in the SCHED_SCAN_RESULTS and
SCHED_SCAN_STOPPED notifications toward user-space by having the
driver provide it through the api.

Reviewed-by: Hante Meuleman <hante.meuleman@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>
Reviewed-by: Franky Lin <franky.lin@broadcom.com>
Signed-off-by: Arend van Spriel <arend.vanspriel@broadcom.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
7 years agomac80211: Add support for BSS max idle period element
Avraham Stern [Wed, 26 Apr 2017 07:58:47 +0000 (10:58 +0300)]
mac80211: Add support for BSS max idle period element

Parse the BSS max idle period element and set the BSS configuration
accordingly so the driver can use this information to configure the
max idle period and to use protected management frames for keep alive
when required.

The BSS max idle period element is defined in IEEE802.11-2016,
section 9.4.2.79

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
7 years agocfg80211: unify cfg80211_roamed() and cfg80211_roamed_bss()
Avraham Stern [Wed, 26 Apr 2017 07:58:49 +0000 (10:58 +0300)]
cfg80211: unify cfg80211_roamed() and cfg80211_roamed_bss()

cfg80211_roamed() and cfg80211_roamed_bss() take the same arguments
except that cfg80211_roamed() requires the BSSID and
cfg80211_roamed_bss() requires the bss entry.

Unify the two functions by using a struct for driver initiated
roaming information so that either the BSSID or the bss entry can be
passed as an argument to the unified function.

Signed-off-by: Avraham Stern <avraham.stern@intel.com>
[modified the ath6k, brcm80211, rndis and wlan-ng drivers accordingly]
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
[modify brcmfmac to remove the useless cast, spotted by Arend]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>