openwrt/staging/blogic.git
7 years agonet/wan: add MODULE_LICENSE for fsl_ucc_hdlc
Valentin Longchamp [Fri, 17 Feb 2017 10:31:22 +0000 (11:31 +0100)]
net/wan: add MODULE_LICENSE for fsl_ucc_hdlc

It is required to build it as a module.

Signed-off-by: Valentin Longchamp <valentin.longchamp@keymile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agords:Remove unnecessary ib_ring unalloc
Zhu Yanjun [Fri, 17 Feb 2017 09:16:22 +0000 (04:16 -0500)]
rds:Remove unnecessary ib_ring unalloc

In the function rds_ib_xmit_atomic, ib_ring is not allocated
successfully. As such, it is not necessary to unalloc it.

Cc: Joe Jin <joe.jin@oracle.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonfp: Use PCI_DEVICE_ID_NETRONOME_NFP* defines
Simon Horman [Fri, 17 Feb 2017 07:57:54 +0000 (08:57 +0100)]
nfp: Use PCI_DEVICE_ID_NETRONOME_NFP* defines

Use PCI_DEVICE_ID_NETRONOME_NFP*, defined in linux/pci_ids.h,
rather than replicating the same values in the NFP driver.

Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agopkt_sched: Remove useless qdisc_stab_lock
Gao Feng [Fri, 17 Feb 2017 06:34:19 +0000 (14:34 +0800)]
pkt_sched: Remove useless qdisc_stab_lock

The qdisc_stab_lock is used in qdisc_get_stab and qdisc_put_stab.
These two functions are invoked in qdisc_create, qdisc_change, and
qdisc_destroy which run fully under RTNL.

So it already makes sure only one could access the qdisc_stab_list at
the same time. Then it is unnecessary to use qdisc_stab_lock now.

Signed-off-by: Gao Feng <fgao@ikuai8.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agorxrpc: Change module filename to rxrpc.ko
David Howells [Fri, 17 Feb 2017 18:16:21 +0000 (18:16 +0000)]
rxrpc: Change module filename to rxrpc.ko

Change module filename from af-rxrpc.ko to rxrpc.ko so as to be consistent
with the other protocol drivers.

Also adjust the documentation to reflect this.

Further, there is no longer a standalone rxkad module, as it has been
merged into the rxrpc core, so get rid of references to that.

Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetvsc: fix typo on statistics
Simon Xiao [Fri, 17 Feb 2017 19:36:20 +0000 (11:36 -0800)]
netvsc: fix typo on statistics

Return the correct tx_errors stats in netvsc.

Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: Simon Xiao <sixiao@microsoft.com>
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agortnl: don't account unused struct ifla_port_vsi in rtnl_port_size
Daniel Borkmann [Fri, 17 Feb 2017 00:56:11 +0000 (01:56 +0100)]
rtnl: don't account unused struct ifla_port_vsi in rtnl_port_size

When allocating rtnl dump messages, struct ifla_port_vsi is never dumped,
so we can save header plus payload in rtnl_port_size(). Infact, attribute
IFLA_PORT_VSI_TYPE and struct ifla_port_vsi are not used anywhere in
the kernel. We only need to keep the nla policy should applications in
user space be filling this out. Same NLA_BINARY issue exists as was fixed
in 364d5716a7ad ("rtnetlink: ifla_vf_policy: fix misuses of NLA_BINARY")
and others, but then again IFLA_PORT_VSI_TYPE is not used anywhere, so
just add a comment that it's unused.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: qlogic: netxen: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Thu, 16 Feb 2017 22:28:01 +0000 (23:28 +0100)]
net: qlogic: netxen: 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>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hamachi: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Thu, 16 Feb 2017 21:46:14 +0000 (22:46 +0100)]
net: hamachi: 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>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: don't indicate expiry on NTF_EXT_LEARNED fdb entries
Roopa Prabhu [Thu, 16 Feb 2017 21:38:04 +0000 (13:38 -0800)]
bridge: don't indicate expiry on NTF_EXT_LEARNED fdb entries

added_by_external_learn fdb entries are added and expired by
external entities like switchdev driver or external controllers.
ageing is already disabled for such entries. Hence, don't
indicate expiry for such fdb entries.

CC: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
CC: Jiri Pirko <jiri@resnulli.us>
CC: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Tested-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bpf-misc'
David S. Miller [Fri, 17 Feb 2017 18:40:06 +0000 (13:40 -0500)]
Merge branch 'bpf-misc'

Daniel Borkmann says:

====================
Misc BPF improvements

This last series for this window adds various misc
improvements to BPF, one is to mark registered map and
prog types as __ro_after_init, another one for removing
cBPF stubs in eBPF JITs and moving the stub to the core
and last also improving JITs is to make generated images
visible to the kernel and kallsyms, so they can be
seen in traces. For details, please have a look at the
individual patches.

Thanks a lot!
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: make jited programs visible in traces
Daniel Borkmann [Thu, 16 Feb 2017 21:24:50 +0000 (22:24 +0100)]
bpf: make jited programs visible in traces

Long standing issue with JITed programs is that stack traces from
function tracing check whether a given address is kernel code
through {__,}kernel_text_address(), which checks for code in core
kernel, modules and dynamically allocated ftrace trampolines. But
what is still missing is BPF JITed programs (interpreted programs
are not an issue as __bpf_prog_run() will be attributed to them),
thus when a stack trace is triggered, the code walking the stack
won't see any of the JITed ones. The same for address correlation
done from user space via reading /proc/kallsyms. This is read by
tools like perf, but the latter is also useful for permanent live
tracing with eBPF itself in combination with stack maps when other
eBPF types are part of the callchain. See offwaketime example on
dumping stack from a map.

This work tries to tackle that issue by making the addresses and
symbols known to the kernel. The lookup from *kernel_text_address()
is implemented through a latched RB tree that can be read under
RCU in fast-path that is also shared for symbol/size/offset lookup
for a specific given address in kallsyms. The slow-path iteration
through all symbols in the seq file done via RCU list, which holds
a tiny fraction of all exported ksyms, usually below 0.1 percent.
Function symbols are exported as bpf_prog_<tag>, in order to aide
debugging and attribution. This facility is currently enabled for
root-only when bpf_jit_kallsyms is set to 1, and disabled if hardening
is active in any mode. The rationale behind this is that still a lot
of systems ship with world read permissions on kallsyms thus addresses
should not get suddenly exposed for them. If that situation gets
much better in future, we always have the option to change the
default on this. Likewise, unprivileged programs are not allowed
to add entries there either, but that is less of a concern as most
such programs types relevant in this context are for root-only anyway.
If enabled, call graphs and stack traces will then show a correct
attribution; one example is illustrated below, where the trace is
now visible in tooling such as perf script --kallsyms=/proc/kallsyms
and friends.

Before:

  7fff8166889d bpf_clone_redirect+0x80007f0020ed (/lib/modules/4.9.0-rc8+/build/vmlinux)
         f5d80 __sendmsg_nocancel+0xffff006451f1a007 (/usr/lib64/libc-2.18.so)

After:

  7fff816688b7 bpf_clone_redirect+0x80007f002107 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fffa0575728 bpf_prog_33c45a467c9e061a+0x8000600020fb (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fffa07ef1fc cls_bpf_classify+0x8000600020dc (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff81678b68 tc_classify+0x80007f002078 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8164d40b __netif_receive_skb_core+0x80007f0025fb (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8164d718 __netif_receive_skb+0x80007f002018 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8164e565 process_backlog+0x80007f002095 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8164dc71 net_rx_action+0x80007f002231 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff81767461 __softirqentry_text_start+0x80007f0020d1 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff817658ac do_softirq_own_stack+0x80007f00201c (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff810a2c20 do_softirq+0x80007f002050 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff810a2cb5 __local_bh_enable_ip+0x80007f002085 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8168d452 ip_finish_output2+0x80007f002152 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8168ea3d ip_finish_output+0x80007f00217d (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff8168f2af ip_output+0x80007f00203f (/lib/modules/4.9.0-rc8+/build/vmlinux)
  [...]
  7fff81005854 do_syscall_64+0x80007f002054 (/lib/modules/4.9.0-rc8+/build/vmlinux)
  7fff817649eb return_from_SYSCALL_64+0x80007f002000 (/lib/modules/4.9.0-rc8+/build/vmlinux)
         f5d80 __sendmsg_nocancel+0xffff01c484812007 (/usr/lib64/libc-2.18.so)

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: remove stubs for cBPF from arch code
Daniel Borkmann [Thu, 16 Feb 2017 21:24:49 +0000 (22:24 +0100)]
bpf: remove stubs for cBPF from arch code

Remove the dummy bpf_jit_compile() stubs for eBPF JITs and make
that a single __weak function in the core that can be overridden
similarly to the eBPF one. Also remove stale pr_err() mentions
of bpf_jit_compile.

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: mark all registered map/prog types as __ro_after_init
Daniel Borkmann [Thu, 16 Feb 2017 21:24:48 +0000 (22:24 +0100)]
bpf: mark all registered map/prog types as __ro_after_init

All map types and prog types are registered to the BPF core through
bpf_register_map_type() and bpf_register_prog_type() during init and
remain unchanged thereafter. As by design we don't (and never will)
have any pluggable code that can register to that at any later point
in time, lets mark all the existing bpf_{map,prog}_type_list objects
in the tree as __ro_after_init, so they can be moved to read-only
section from then onwards.

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 agobridge: vlan_tunnel: explicitly reset metadata attrs to NULL on failure
Roopa Prabhu [Thu, 16 Feb 2017 19:29:21 +0000 (11:29 -0800)]
bridge: vlan_tunnel: explicitly reset metadata attrs to NULL on failure

Fixes: efa5356b0d97 ("bridge: per vlan dst_metadata netlink support")
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: bgmac: store MAC address directly in netdev->dev_addr
Tobias Klauser [Thu, 16 Feb 2017 14:11:19 +0000 (15:11 +0100)]
net: bgmac: store MAC address directly in netdev->dev_addr

After commit 34a5102c3235 ("net: bgmac: allocate struct bgmac just once
& don't copy it") the mac_addr member of struct bgmac is no longer
necessary to pass the MAC address to bgmac_enet_probe(). Instead it can
directly be stored in netdev->dev_addr.

Also use eth_hw_addr_random() instead of eth_random_addr() in case a
random MAC is nedded. This will make sure netdev->addr_assign_type will
be properly set.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Acked-by: Jon Mason <jon.mason@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'wireless-drivers-next-for-davem-2017-02-16' of git://git.kernel.org/pub...
David S. Miller [Fri, 17 Feb 2017 18:00:45 +0000 (13:00 -0500)]
Merge tag 'wireless-drivers-next-for-davem-2017-02-16' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for 4.11

Mostly small fixes, not really any new features.

Major changes:

ath10k

* when trying older firmware versions don't confuse user with error messages

ath9k

* fix crash in AP mode (regression)
* fix relayfs crash (regression)
* fix initialisation with AR9340 and AR9550
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethoc: Use eth_hw_addr_random()
Tobias Klauser [Thu, 16 Feb 2017 12:54:32 +0000 (13:54 +0100)]
net: ethoc: Use eth_hw_addr_random()

Use eth_hw_addr_random() to set a random dev_addr and update
addr_assign_type instead of open-coding it.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'rhashtable-allocation-failure-during-insertion'
David S. Miller [Fri, 17 Feb 2017 17:28:44 +0000 (12:28 -0500)]
Merge branch 'rhashtable-allocation-failure-during-insertion'

Herbert Xu says:

====================
rhashtable: Handle table allocation failure during insertion

v2 -

Added Ack to patch 2.
Fixed RCU annotation in code path executed by rehasher by using
rht_dereference_bucket.

v1 -

This series tackles the problem of table allocation failures during
insertion.  The issue is that we cannot vmalloc during insertion.
This series deals with this by introducing nested tables.

The first two patches removes manual hash table walks which cannot
work on a nested table.

The final patch introduces nested tables.

I've tested this with test_rhashtable and it appears to work.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agorhashtable: Add nested tables
Herbert Xu [Sat, 11 Feb 2017 11:26:47 +0000 (19:26 +0800)]
rhashtable: Add nested tables

This patch adds code that handles GFP_ATOMIC kmalloc failure on
insertion.  As we cannot use vmalloc, we solve it by making our
hash table nested.  That is, we allocate single pages at each level
and reach our desired table size by nesting them.

When a nested table is created, only a single page is allocated
at the top-level.  Lower levels are allocated on demand during
insertion.  Therefore for each insertion to succeed, only two
(non-consecutive) pages are needed.

After a nested table is created, a rehash will be scheduled in
order to switch to a vmalloced table as soon as possible.  Also,
the rehash code will never rehash into a nested table.  If we
detect a nested table during a rehash, the rehash will be aborted
and a new rehash will be scheduled.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotipc: Fix tipc_sk_reinit race conditions
Herbert Xu [Sat, 11 Feb 2017 11:26:46 +0000 (19:26 +0800)]
tipc: Fix tipc_sk_reinit race conditions

There are two problems with the function tipc_sk_reinit.  Firstly
it's doing a manual walk over an rhashtable.  This is broken as
an rhashtable can be resized and if you manually walk over it
during a resize then you may miss entries.

Secondly it's missing memory barriers as previously the code used
spinlocks which provide the barriers implicitly.

This patch fixes both problems.

Fixes: 07f6c4bc048a ("tipc: convert tipc reference table to...")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogfs2: Use rhashtable walk interface in glock_hash_walk
Herbert Xu [Sat, 11 Feb 2017 11:26:45 +0000 (19:26 +0800)]
gfs2: Use rhashtable walk interface in glock_hash_walk

The function glock_hash_walk walks the rhashtable by hand.  This
is broken because if it catches the hash table in the middle of
a rehash, then it will miss entries.

This patch replaces the manual walk by using the rhashtable walk
interface.

Fixes: 88ffbf3e037e ("GFS2: Use resizable hash table for glocks")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mvneta: make mvneta_eth_tool_ops static
Jisheng Zhang [Thu, 16 Feb 2017 09:07:39 +0000 (17:07 +0800)]
net: mvneta: make mvneta_eth_tool_ops static

The mvneta_eth_tool_ops is only used internally in mvneta driver, so
make it static.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Acked-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'net-sched-reflect-hw-offload-in-classifiers'
David S. Miller [Fri, 17 Feb 2017 17:08:07 +0000 (12:08 -0500)]
Merge branch 'net-sched-reflect-hw-offload-in-classifiers'

Or Gerlitz says:

====================
net/sched: Reflect HW offload status in classifiers

Currently there is no way of querying whether a filter is
offloaded to HW or not when using "both" policy (where none
of skip_sw or skip_hw flags are set by user-space).

Added two new flags, "in hw" and "not in hw" such that user space
can determine if a filter is actually offloaded to hw. The "in hw"
UAPI semantics was chosen so it's similar to the "skip hw" flag logic.

If none of these two flags are set, this signals running
over older kernel.

As an example, add one vlan push + fwd rule, one matchall rule and one u32 rule
without any flags, and another vlan + fwd skip_sw rule, such that the different TC
classifier attempt to offload all of them -- all over mlx5 SRIOV VF rep:

flower skip_sw indev eth2_0 src_mac e4:11:22:33:44:50 dst_mac e4:1d:2d:a5:f3:9d
action vlan push id 52 action mirred egress redirect dev eth2

flower indev eth2_0 src_mac e4:11:22:33:44:50 dst_mac e4:11:22:33:44:51
action vlan push id 53 action mirred egress redirect dev eth2

u32 ht 800: flowid 800:1 match ip src 192.168.1.0/24 action drop

Since that VF rep doesn't offload matchall/u32 and can currently offload
only one vlan push rule we expect three of the rules not to be offloaded:

filter protocol ip pref 99 u32
filter protocol ip pref 99 u32 fh 800: ht divisor 1
filter protocol ip pref 99 u32 fh 800::1 order 1 key ht 800 bkt 0 flowid 800:1 not in_hw
  match c0a80100/ffffff00 at 12
action order 1: gact action drop
 random type none pass val 0
 index 8 ref 1 bind 1

filter protocol all pref 49150 matchall
filter protocol all pref 49150 matchall handle 0x1
  not in_hw
action order 1: mirred (Egress Mirror to device veth1) pipe
  index 27 ref 1 bind 1

filter protocol ip pref 49151 flower
filter protocol ip pref 49151 flower handle 0x1
  indev eth2_0
  dst_mac e4:11:22:33:44:51
  src_mac e4:11:22:33:44:50
  eth_type ipv4
  not in_hw
action order 1:  vlan push id 53 protocol 802.1Q priority 0 pipe
 index 20 ref 1 bind 1

action order 2: mirred (Egress Redirect to device eth2) stolen
  index 26 ref 1 bind 1

filter protocol ip pref 49152 flower
filter protocol ip pref 49152 flower handle 0x1
  indev eth2_0
  dst_mac e4:1d:2d:a5:f3:9d
  src_mac e4:11:22:33:44:50
  eth_type ipv4
  skip_sw
  in_hw
action order 1:  vlan push id 52 protocol 802.1Q priority 0 pipe
 index 19 ref 1 bind 1

action order 2: mirred (Egress Redirect to device eth2) stolen
  index 25 ref 1 bind 1

v3 --> v4 changes:
 - removed extra parenthesis (Dave)

v2 --> v3 changes:
 - fixed the matchall dump flags patch to do proper checks (Jakub)
 - added the same proper checks to flower where they were missing
 - that flower patch was added as #1 and hence all the other patches are offed-by-one

v1 --> v2 changes:
 - applied feedback from Jakub and Dave -- where none of the skip flags were set,
   the suggested approach didn't allow user space to distringuish between old kernel
   to a case when offloading to HW worked fine.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_bpf: Reflect HW offload status
Or Gerlitz [Thu, 16 Feb 2017 08:31:16 +0000 (10:31 +0200)]
net/sched: cls_bpf: Reflect HW offload status

BPF classifier support for the "in hw" offloading flags.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Amir Vadai <amir@vadai.me>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_u32: Reflect HW offload status
Or Gerlitz [Thu, 16 Feb 2017 08:31:15 +0000 (10:31 +0200)]
net/sched: cls_u32: Reflect HW offload status

U32 support for the "in hw" offloading flags.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Amir Vadai <amir@vadai.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_matchall: Reflect HW offloading status
Or Gerlitz [Thu, 16 Feb 2017 08:31:14 +0000 (10:31 +0200)]
net/sched: cls_matchall: Reflect HW offloading status

Matchall support for the "in hw" offloading flags.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Amir Vadai <amir@vadai.me>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_flower: Reflect HW offload status
Or Gerlitz [Thu, 16 Feb 2017 08:31:13 +0000 (10:31 +0200)]
net/sched: cls_flower: Reflect HW offload status

Flower support for the "in hw" offloading flags.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Amir Vadai <amir@vadai.me>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: Reflect HW offload status
Or Gerlitz [Thu, 16 Feb 2017 08:31:12 +0000 (10:31 +0200)]
net/sched: Reflect HW offload status

Currently there is no way of querying whether a filter is
offloaded to HW or not when using "both" policy (where none
of skip_sw or skip_hw flags are set by user-space).

Add two new flags, "in hw" and "not in hw" such that user
space can determine if a filter is actually offloaded to
hw or not. The "in hw" UAPI semantics was chosen so it's
similar to the "skip hw" flag logic.

If none of these two flags are set, this signals running
over older kernel.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Amir Vadai <amir@vadai.me>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_matchall: Dump the classifier flags
Or Gerlitz [Thu, 16 Feb 2017 08:31:11 +0000 (10:31 +0200)]
net/sched: cls_matchall: Dump the classifier flags

The classifier flags are not dumped to user-space, do that.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Yotam Gigi <yotamg@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_flower: Properly handle classifier flags dumping
Or Gerlitz [Thu, 16 Feb 2017 08:31:10 +0000 (10:31 +0200)]
net/sched: cls_flower: Properly handle classifier flags dumping

Dump the classifier flags only if non zero and make sure to check
the return status of the handler that puts them into the netlink msg.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: oki-semi: pch_gbe: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Thu, 16 Feb 2017 08:10:49 +0000 (09:10 +0100)]
net: oki-semi: pch_gbe: 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>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpsw: correct ale dev to cpsw
Ivan Khoronzhuk [Wed, 15 Feb 2017 17:45:02 +0000 (19:45 +0200)]
net: ethernet: ti: cpsw: correct ale dev to cpsw

The ale is a property of cpsw, so change dev to cpsw->dev,
aka pdev->dev, to be consistent.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: nvidia: forcedeth: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Tue, 14 Feb 2017 22:36:32 +0000 (23:36 +0100)]
net: nvidia: forcedeth: 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>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'ptp-attribute-cleanup'
David S. Miller [Fri, 17 Feb 2017 16:03:07 +0000 (11:03 -0500)]
Merge branch 'ptp-attribute-cleanup'

Dmitry Torokhov says:

====================
PTP attribute handling cleanup

PTP core was creating some attributes, such as "period" and "fifo", and the
entire "pins" attribute group, after creating class deevice, which creates
a race for userspace: uevent may arrive before all attributes are created.

This series of patches switches PTP to use is_visible() to control
visibility of attributes in a group, and device_create_with_groups() to
ensure that attributes are created before we notify userspace of a new
device.

v2:
- added Richard's acked-by to patch #1
- removed use of kmalloc_array in favor of kcalloc in patch #2 at
  Richard's request
- added a cover letter

v1:
- initial patch set
====================

Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoptp: create "pins" together with the rest of attributes
Dmitry Torokhov [Tue, 14 Feb 2017 18:23:34 +0000 (10:23 -0800)]
ptp: create "pins" together with the rest of attributes

Let's switch to using device_create_with_groups(), which will allow us to
create "pins" attribute group together with the rest of ptp device
attributes, and before userspace gets notified about ptp device creation.

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoptp: use is_visible method to hide unused attributes
Dmitry Torokhov [Tue, 14 Feb 2017 18:23:33 +0000 (10:23 -0800)]
ptp: use is_visible method to hide unused attributes

Instead of creating selected attributes after the device is created (and
after userspace potentially seen uevent), lets use attribute group
is_visible() method to control which attributes are shown. This will allow
us to create all attributes (except "pins" group, which will be taken care
of later) before userspace gets notified about new ptp class device.

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoptp: use kcalloc when allocating arrays
Dmitry Torokhov [Tue, 14 Feb 2017 18:23:32 +0000 (10:23 -0800)]
ptp: use kcalloc when allocating arrays

kcalloc is more semantically correct when allocating arrays of objects, and
overflow-safe.

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoptp: do not explicitly set drvdata in ptp_clock_register()
Dmitry Torokhov [Tue, 14 Feb 2017 18:23:31 +0000 (10:23 -0800)]
ptp: do not explicitly set drvdata in ptp_clock_register()

We do not need explicitly call dev_set_drvdata(), as it is done for us by
device_create().

Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlx4: do not fire tasklet unless necessary
Eric Dumazet [Fri, 10 Feb 2017 12:27:58 +0000 (04:27 -0800)]
mlx4: do not fire tasklet unless necessary

All rx and rx netdev interrupts are handled by respectively
by mlx4_en_rx_irq() and mlx4_en_tx_irq() which simply schedule a NAPI.

But mlx4_eq_int() also fires a tasklet to service all items that were
queued via mlx4_add_cq_to_tasklet(), but this handler was not called
unless user cqe was handled.

This is very confusing, as "mpstat -I SCPU ..." show huge number of
tasklet invocations.

This patch saves this overhead, by carefully firing the tasklet directly
from mlx4_add_cq_to_tasklet(), removing four atomic operations per IRQ.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Cc: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.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, 17 Feb 2017 02:25:49 +0000 (21:25 -0500)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
pull request (net-next): ipsec-next 2017-02-16

1) Make struct xfrm_input_afinfo const, nothing writes to it.
   From Florian Westphal.

2) Remove all places that write to the afinfo policy backend
   and make the struct const then.
   From Florian Westphal.

3) Prepare for packet consuming gro callbacks and add
   ESP GRO handlers. ESP packets can be decapsulated
   at the GRO layer then. It saves a round through
   the stack for each ESP packet.

Please note that this has a merge coflict between commit

63fca65d0863 ("net: add confirm_neigh method to dst_ops")

from net-next and

3d7d25a68ea5 ("xfrm: policy: remove garbage_collect callback")
a2817d8b279b ("xfrm: policy: remove family field")

from ipsec-next.

The conflict can be solved as it is done in linux-next.

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

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 [Fri, 17 Feb 2017 02:19:44 +0000 (21:19 -0500)]
Merge branch '10GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
10GbE Intel Wired LAN Driver Updates 2017-02-16

This series contains updates to ixgbe only.

Tony updates the driver to advertise 2.5Gb and 5.0Gb if the adapter
supports it.

Stephen Hemminger renames our dcbnl_ops since it is global to
ixgbe_dcbnl_ops to avoid namespace issues.

Mark updates the driver version based on the recent changes.

Alex has the remainder of the changes, starting with consolidating
functions that represent logical steps in the receive process so we can
later update them more easily (and align with igb).  Modify the receive
path to only synchronize the length of the frame versus the entire buffer.
Provided performance improvements by adding support for
DMA_ATTR_SKIP_CPU_SYNC and DMA_ATTR_WEAK_ORDERING.  Also made additional
performance gains by batching the page count updates instead of doing
them one at a time.  Adjusted the receive path to use 3k buffers with
8k backing them in order to support build_skb with jumbo frames.  Made
additional driver improvements by using the length of the packet instead
of the DD status to determine if a new descriptor is ready to be
processed, which cuts down on reads.  To reduce code duplication, pulled
apart the receive path into separate functions.  Added support for
providing a buffer with headroom and tailroom to allow for shared info
for NET_SKB_PAD and NET_IP_ALIGN.
====================

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 [Fri, 17 Feb 2017 00:34:01 +0000 (19:34 -0500)]
Merge git://git./linux/kernel/git/davem/net

7 years agocxgb4: Remove redundant code in t4_uld_clean_up()
Ganesh Goudar [Thu, 16 Feb 2017 06:57:15 +0000 (12:27 +0530)]
cxgb4: Remove redundant code in t4_uld_clean_up()

Remove variable rxq_info and also remove redundant assignment
to it.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocxgb4: Add new T5 and T6 pci device id's
Ganesh Goudar [Thu, 16 Feb 2017 06:55:52 +0000 (12:25 +0530)]
cxgb4: Add new T5 and T6 pci device id's

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocxgb4: Increase max number of tc u32 links
Arjun V [Thu, 16 Feb 2017 06:52:45 +0000 (12:22 +0530)]
cxgb4: Increase max number of tc u32 links

Make max number of supported tc u32 links equal to max number of filters
supported by hardware.

Signed-off-by: Arjun V <arjun@chelsio.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'media/v4.10-5' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Thu, 16 Feb 2017 18:22:41 +0000 (10:22 -0800)]
Merge tag 'media/v4.10-5' of git://git./linux/kernel/git/mchehab/linux-media

Pull media fix from Mauro Carvalho Chehab:
 "A regression fix that makes the Siano driver to work again after the
  CONFIG_VMAP_STACK change"

* tag 'media/v4.10-5' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media:
  [media] siano: make it work again with CONFIG_VMAP_STACK

7 years agovfs: fix uninitialized flags in splice_to_pipe()
Miklos Szeredi [Thu, 16 Feb 2017 16:49:02 +0000 (17:49 +0100)]
vfs: fix uninitialized flags in splice_to_pipe()

Flags (PIPE_BUF_FLAG_PACKET, PIPE_BUF_FLAG_GIFT) could remain on the
unused part of the pipe ring buffer.  Previously splice_to_pipe() left
the flags value alone, which could result in incorrect behavior.

Uninitialized flags appears to have been there from the introduction of
the splice syscall.

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Cc: <stable@vger.kernel.org> # 2.6.17+
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi...
Linus Torvalds [Thu, 16 Feb 2017 17:05:34 +0000 (09:05 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mszeredi/fuse

Pull fuse fixes from Miklos Szeredi:
 "Fix a use after free bug introduced in 4.2 and using an uninitialized
  value introduced in 4.9"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  fuse: fix uninitialized flags in pipe_buffer
  fuse: fix use after free issue in fuse_dev_do_read()

7 years agoMerge tag 'pci-v4.10-fixes-4' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
Linus Torvalds [Thu, 16 Feb 2017 17:03:37 +0000 (09:03 -0800)]
Merge tag 'pci-v4.10-fixes-4' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fix from Bjorn Helgaas:
 "Add back pcie_pme_remove() so we free the IRQ when removing PCIe port
  devices; previously the leaked IRQ caused an MSI BUG_ON"

* tag 'pci-v4.10-fixes-4' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI/PME: Restore pcie_pme_driver.remove

7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 16 Feb 2017 16:37:18 +0000 (08:37 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) In order to avoid problems in the future, make cgroup bpf overriding
    explicit using BPF_F_ALLOW_OVERRIDE. From Alexei Staovoitov.

 2) LLC sets skb->sk without proper skb->destructor and this explodes,
    fix from Eric Dumazet.

 3) Make sure when we have an ipv4 mapped source address, the
    destination is either also an ipv4 mapped address or
    ipv6_addr_any(). Fix from Jonathan T. Leighton.

 4) Avoid packet loss in fec driver by programming the multicast filter
    more intelligently. From Rui Sousa.

 5) Handle multiple threads invoking fanout_add(), fix from Eric
    Dumazet.

 6) Since we can invoke the TCP input path in process context, without
    BH being disabled, we have to accomodate that in the locking of the
    TCP probe. Also from Eric Dumazet.

 7) Fix erroneous emission of NETEVENT_DELAY_PROBE_TIME_UPDATE when we
    aren't even updating that sysctl value. From Marcus Huewe.

 8) Fix endian bugs in ibmvnic driver, from Thomas Falcon.

[ This is the second version of the pull that reverts the nested
  rhashtable changes that looked a bit too scary for this late in the
  release  - Linus ]

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (27 commits)
  rhashtable: Revert nested table changes.
  ibmvnic: Fix endian errors in error reporting output
  ibmvnic: Fix endian error when requesting device capabilities
  net: neigh: Fix netevent NETEVENT_DELAY_PROBE_TIME_UPDATE notification
  net: xilinx_emaclite: fix freezes due to unordered I/O
  net: xilinx_emaclite: fix receive buffer overflow
  bpf: kernel header files need to be copied into the tools directory
  tcp: tcp_probe: use spin_lock_bh()
  uapi: fix linux/if_pppol2tp.h userspace compilation errors
  packet: fix races in fanout_add()
  ibmvnic: Fix initial MTU settings
  net: ethernet: ti: cpsw: fix cpsw assignment in resume
  kcm: fix a null pointer dereference in kcm_sendmsg()
  net: fec: fix multicast filtering hardware setup
  ipv6: Handle IPv4-mapped src to in6addr_any dst.
  ipv6: Inhibit IPv4-mapped src address on the wire.
  net/mlx5e: Disable preemption when doing TC statistics upcall
  rhashtable: Add nested tables
  tipc: Fix tipc_sk_reinit race conditions
  gfs2: Use rhashtable walk interface in glock_hash_walk
  ...

7 years agofuse: fix uninitialized flags in pipe_buffer
Miklos Szeredi [Thu, 16 Feb 2017 14:08:20 +0000 (15:08 +0100)]
fuse: fix uninitialized flags in pipe_buffer

Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Fixes: d82718e348fe ("fuse_dev_splice_read(): switch to add_to_pipe()")
Cc: <stable@vger.kernel.org> # 4.9+
7 years agoixgbe: Don't bother clearing buffer memory for descriptor rings
Alexander Duyck [Tue, 17 Jan 2017 16:37:29 +0000 (08:37 -0800)]
ixgbe: Don't bother clearing buffer memory for descriptor rings

This patch makes it so that we don't need to bother with clearing the
memory out for the descriptor rings.  The general idea is to only free
buffers associated with buffers in use which are located between the
next_to_clean and next_to_use or next_to_alloc values.  Everything outside
of those regions can be safely ignored since they should have no buffers
associated with them.

The advantage to doing things this way is that is should speed up bring-up
and tear-down of the rings.  Specifically we can avoid the 512 or more
cycles required to memset the rings in tear-down.  In the bring-up phase we
then clear the memory as a part of initialization.  The general idea is
that the clearing in initialization can act as a prefetch of sorts for the
buffer info structures so they are in the local CPU when we go to populate
them.  This should help to improve overall time needed to perform a
suspend/resume.

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 support for build_skb
Alexander Duyck [Tue, 17 Jan 2017 16:37:13 +0000 (08:37 -0800)]
ixgbe: Add support for build_skb

This patch adds build_skb support to the Rx path.  There are several
advantages to this change.

1.  It avoids the memcpy and skb->head allocation for small packets which
    improves performance by about 5% in my tests.
2.  It avoids the memcpy, skb->head allocation, and eth_get_headlen
    for larger packets improving performance by about 10% in my tests.
3.  For VXLAN packets it allows the full header to be in skb->data which
    improves the performance by as much as 30% in some of my tests.

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 private flag to control buffer mode
Alexander Duyck [Tue, 17 Jan 2017 16:37:03 +0000 (08:37 -0800)]
ixgbe: Add private flag to control buffer mode

Since there are potential drawbacks to the new Rx allocation approach I
thought it best to add a "chicken bit" so that we can turn the feature off
if in the event that a problem is found.

It also provides a means of validating the legacy Rx path in the event that
we are forced to fall back.  At some point in the future when we are
convinced we don't need it anymore we might be able to drop the legacy-rx
flag.

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 support for padding packet
Alexander Duyck [Tue, 17 Jan 2017 16:36:54 +0000 (08:36 -0800)]
ixgbe: Add support for padding packet

This patch adds support for providing a buffer with headroom and tailroom
to allow for shared info, NET_SKB_PAD, and NET_IP_ALIGN.  With this
combined with the DMA changes we can start using build_skb to build frames
around an incoming Rx buffer instead of having to memcpy the headers.

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: Break out Rx buffer page management
Alexander Duyck [Tue, 17 Jan 2017 16:36:45 +0000 (08:36 -0800)]
ixgbe: Break out Rx buffer page management

We are going to be expanding the number of Rx paths in the driver.  Instead
of duplicating all that code I am pulling it apart into separate functions
so that we don't have so much code duplication.

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: Use length to determine if descriptor is done
Alexander Duyck [Tue, 17 Jan 2017 16:36:28 +0000 (08:36 -0800)]
ixgbe: Use length to determine if descriptor is done

This change makes it so that we use the length of the packet instead of the
DD status bit to determine if a new descriptor is ready to be processed.
The obvious advantage is that it cuts down on reads as we don't really even
need the DD bit if going from a 0 to a non-zero value on size is enough to
inform us that the packet has been completed.

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: Make use of order 1 pages and 3K buffers independent of FCoE
Alexander Duyck [Tue, 17 Jan 2017 16:36:14 +0000 (08:36 -0800)]
ixgbe: Make use of order 1 pages and 3K buffers independent of FCoE

In order to support build_skb with jumbo frames it will be necessary to use
3K buffers for the Rx path with 8K pages backing them.  This is needed on
architectures that implement 4K pages because we can't support 2K buffers
plus padding in a 4K page.

In the case of systems that support page sizes larger than 4K the 3K
attribute will only be applied to FCoE as we can fall back to using just 2K
buffers and adding the padding.

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: Update code to better handle incrementing page count
Alexander Duyck [Tue, 17 Jan 2017 16:36:03 +0000 (08:36 -0800)]
ixgbe: Update code to better handle incrementing page count

Batch the page count updates instead of doing them one at a time.  By doing
this we can improve the overall performance as the atomic increment
operations can be expensive due to the fact that on x86 they are locked
operations which can cause stalls.  By doing bulk updates we can
consolidate the stall which should help to improve the overall receive
performance.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Update driver to make use of DMA attributes in Rx path
Alexander Duyck [Tue, 17 Jan 2017 16:35:54 +0000 (08:35 -0800)]
ixgbe: Update driver to make use of DMA attributes in Rx path

This patch adds support for DMA_ATTR_SKIP_CPU_SYNC and
DMA_ATTR_WEAK_ORDERING.  By enabling both of these for the Rx path we are
able to see performance improvements on architectures that implement either
one due to the fact that page mapping and unmapping only has to sync what
is actually being used instead of the entire buffer.  In addition by
enabling the weak ordering attribute enables a performance improvement for
architectures that can associate a memory ordering with a DMA buffer such
as Sparc.

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: Only DMA sync frame length
Alexander Duyck [Tue, 17 Jan 2017 16:35:44 +0000 (08:35 -0800)]
ixgbe: Only DMA sync frame length

On some platforms, syncing a buffer for DMA is expensive. Rather than
sync the whole 2K receive buffer, only synchronise the length of the
frame, which will typically be the MTU, or a much smaller TCP ACK.

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 function for checking to see if we can reuse page
Alexander Duyck [Tue, 17 Jan 2017 16:35:34 +0000 (08:35 -0800)]
ixgbe: Add function for checking to see if we can reuse page

This patch consolidates the code for the ixgbe driver so that it is more
inline with what is already in igb.  The general idea is to just
consolidate functions that represent logical steps in the Rx process so we
can later update them more easily.

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: Update version to reflect added functionality
Mark Rustad [Mon, 12 Dec 2016 23:08:13 +0000 (15:08 -0800)]
ixgbe: Update version to reflect added functionality

Update the driver version to reflect the new devices that it
supports.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: prefix Data Center Bridge ops struct
Stephen Hemminger [Mon, 21 Nov 2016 17:52:40 +0000 (09:52 -0800)]
ixgbe: prefix Data Center Bridge ops struct

Since dcbnl_ops is global, it should be prefixed by ixgbe_

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoixgbe: Support 2.5Gb and 5Gb speed
Tony Nguyen [Fri, 11 Nov 2016 00:01:33 +0000 (16:01 -0800)]
ixgbe: Support 2.5Gb and 5Gb speed

Though not advertised through ethtool, if the link partner advertises a
2.5Gb or 5Gb connection, and the adapter supports it, allow the speed to be
used.

Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agorhashtable: Revert nested table changes.
David S. Miller [Thu, 16 Feb 2017 03:29:51 +0000 (22:29 -0500)]
rhashtable: Revert nested table changes.

This reverts commits:

6a25478077d987edc5e2f880590a2bc5fcab4441
9dbbfb0ab6680c6a85609041011484e6658e7d3c
40137906c5f55c252194ef5834130383e639536f

It's too risky to put in this late in the release
cycle.  We'll put these changes into the next merge
window instead.

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Fix endian errors in error reporting output
Thomas Falcon [Wed, 15 Feb 2017 16:33:33 +0000 (10:33 -0600)]
ibmvnic: Fix endian errors in error reporting output

Error reports received from firmware were not being converted from
big endian values, leading to bogus error codes reported on little
endian systems.

Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Fix endian error when requesting device capabilities
Thomas Falcon [Wed, 15 Feb 2017 16:32:11 +0000 (10:32 -0600)]
ibmvnic: Fix endian error when requesting device capabilities

When a vNIC client driver requests a faulty device setting, the
server returns an acceptable value for the client to request.
This 64 bit value was incorrectly being swapped as a 32 bit value,
resulting in loss of data. This patch corrects that by using
the 64 bit swap function.

Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoatm: idt77252, use setup_timer and mod_timer
Jan Koniarik [Wed, 15 Feb 2017 15:59:35 +0000 (16:59 +0100)]
atm: idt77252, use setup_timer and mod_timer

Stop accessing timer struct members directly and use setup_timer and
mod_timer helpers intended for that use. It makes the code cleaner and
will allow for easier change of the timer struct internals.

Signed-off-by: Jan Koniarik <jan.koniarik@trustica.cz>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Chas Williams <3chas3@gmail.com>
Cc: <linux-atm-general@lists.sourceforge.net>
Cc: <netdev@vger.kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: acl: Use PBS type for forward action
Jiri Pirko [Wed, 15 Feb 2017 11:09:51 +0000 (12:09 +0100)]
mlxsw: acl: Use PBS type for forward action

Current behaviour of "mirred redirect" action (forward) offload is a bit
odd. For matched packets the action forwards them to the desired
destination, but it also lets the packet duplicates to go the original
way down (bridge, router, etc). That is more like "mirred mirror".
Fix this by using PBS type which behaves exactly like "mirred redirect".
Note that PBS does not support loopback mode.

Fixes: 4cda7d8d7098 ("mlxsw: core: Introduce flexible actions support")
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 agoMerge branch 'stmmac-misc'
David S. Miller [Wed, 15 Feb 2017 18:20:57 +0000 (13:20 -0500)]
Merge branch 'stmmac-misc'

Corentin Labbe says:

====================
stmmac: misc patchs

This is a follow up of my previous stmmac serie which address some comment
done in v2.
====================

Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: invert the logic for dumping regs
LABBE Corentin [Wed, 15 Feb 2017 09:46:45 +0000 (10:46 +0100)]
net: stmmac: invert the logic for dumping regs

It is easier to follow the logic by removing the not operator

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: reduce indentation by adding a continue
LABBE Corentin [Wed, 15 Feb 2017 09:46:44 +0000 (10:46 +0100)]
net: stmmac: reduce indentation by adding a continue

As suggested by Joe Perches, replacing the "if phydev" logic permit to
reduce indentation in the for loop.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: split the stmmac_adjust_link 10/100 case
LABBE Corentin [Wed, 15 Feb 2017 09:46:43 +0000 (10:46 +0100)]
net: stmmac: split the stmmac_adjust_link 10/100 case

The 10/100 case have too many ifcase.
This patch split it for removing an if.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: run stmmac_hw_fix_mac_speed when speed is valid
LABBE Corentin [Wed, 15 Feb 2017 09:46:42 +0000 (10:46 +0100)]
net: stmmac: run stmmac_hw_fix_mac_speed when speed is valid

This patch mutualise a bit by running stmmac_hw_fix_mac_speed() after
the switch in case of valid speed.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: set speed at SPEED_UNKNOWN in case of broken speed
LABBE Corentin [Wed, 15 Feb 2017 09:46:41 +0000 (10:46 +0100)]
net: stmmac: set speed at SPEED_UNKNOWN in case of broken speed

In case of invalid speed given, stmmac_adjust_link() still record it as
current speed.
This patch modify the default case to set speed as SPEED_UNKNOWN if not
10/100/1000.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: use SPEED_UNKNOWN/DUPLEX_UNKNOWN
LABBE Corentin [Wed, 15 Feb 2017 09:46:40 +0000 (10:46 +0100)]
net: stmmac: use SPEED_UNKNOWN/DUPLEX_UNKNOWN

It is better to use DUPLEX_UNKNOWN instead of just "-1".
Using 0 for an invalid speed is bad since 0 is a valid value for speed.
So this patch replace 0 by SPEED_UNKNOWN.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: likely is useless in occasional function
LABBE Corentin [Wed, 15 Feb 2017 09:46:39 +0000 (10:46 +0100)]
net: stmmac: likely is useless in occasional function

The stmmac_adjust_link() function is called too rarely for having
likely() macros being useful.
Just remove likely annotation in it.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: remove useless parenthesis
LABBE Corentin [Wed, 15 Feb 2017 09:46:38 +0000 (10:46 +0100)]
net: stmmac: remove useless parenthesis

This patch remove some useless parenthesis.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Reviewed-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: aquantia: switch to pci_alloc_irq_vectors
Christoph Hellwig [Wed, 15 Feb 2017 07:38:47 +0000 (08:38 +0100)]
net: ethernet: aquantia: switch to pci_alloc_irq_vectors

pci_enable_msix has been long deprecated, but this driver adds a new
instance.  Convert it to pci_alloc_irq_vectors so that no new instance
of the deprecated function reaches mainline.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Pavel Belous <pavel.belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'qed-ptp'
David S. Miller [Wed, 15 Feb 2017 17:42:54 +0000 (12:42 -0500)]
Merge branch 'qed-ptp'

Yuval Mintz says:

====================
qed*: Add support for PTP

This patch series adds required changes for qed/qede drivers for
supporting the IEEE Precision Time Protocol (PTP).

Changes from previous versions:
v7: Fixed Kbuild robot warnings.

v6: Corrected broken loop iteration in previous version.
    Reduced approximation error of adjfreq.

v5: Removed two divisions from the adjust-frequency loop.
    Resulting logic would use 8 divisions [instead of 24].

v4: Remove the loop iteration for value '0' in the qed_ptp_hw_adjfreq()
    implementation.

v3: Use div_s64 for 64-bit divisions as do_div gives error for signed
    types.
    Incorporated review comments from Richard Cochran.
      - Clear timestamp resgisters as soon as timestamp is read.
      - Use shift operation in the place of 'divide by 16'.

v2: Use do_div for 64-bit divisions.
====================

Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqede: Add driver support for PTP
Sudarsana Reddy Kalluru [Wed, 15 Feb 2017 08:24:11 +0000 (10:24 +0200)]
qede: Add driver support for PTP

This patch adds the driver support for,
  - Registering the ptp clock functionality with the OS.
  - Timestamping the Rx/Tx PTP packets.
  - Ethtool callbacks related to PTP.

Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Add infrastructure for PTP support
Sudarsana Reddy Kalluru [Wed, 15 Feb 2017 08:24:10 +0000 (10:24 +0200)]
qed: Add infrastructure for PTP support

The patch adds the required qed interfaces for configuring/reading
the PTP clock on the adapter.

Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocxgb4: Update proper netdev stats for rx drops
Ganesh Goudar [Wed, 15 Feb 2017 06:15:25 +0000 (11:45 +0530)]
cxgb4: Update proper netdev stats for rx drops

Count buffer group drops or truncates as rx drops rather than
rx errors in netdev stats.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: Arjun V <arjun@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoopenvswitch: Set internal device max mtu to ETH_MAX_MTU.
Jarno Rajahalme [Wed, 15 Feb 2017 05:16:28 +0000 (21:16 -0800)]
openvswitch: Set internal device max mtu to ETH_MAX_MTU.

Commit 91572088e3fd ("net: use core MTU range checking in core net
infra") changed the openvswitch internal device to use the core net
infra for controlling the MTU range, but failed to actually set the
max_mtu as described in the commit message, which now defaults to
ETH_DATA_LEN.

This patch fixes this by setting max_mtu to ETH_MAX_MTU after
ether_setup() call.

Fixes: 91572088e3fd ("net: use core MTU range checking in core net infra")
Signed-off-by: Jarno Rajahalme <jarno@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: neigh: Fix netevent NETEVENT_DELAY_PROBE_TIME_UPDATE notification
Marcus Huewe [Wed, 15 Feb 2017 00:00:36 +0000 (01:00 +0100)]
net: neigh: Fix netevent NETEVENT_DELAY_PROBE_TIME_UPDATE notification

When setting a neigh related sysctl parameter, we always send a
NETEVENT_DELAY_PROBE_TIME_UPDATE netevent. For instance, when
executing

sysctl net.ipv6.neigh.wlp3s0.retrans_time_ms=2000

a NETEVENT_DELAY_PROBE_TIME_UPDATE netevent is generated.

This is caused by commit 2a4501ae18b5 ("neigh: Send a
notification when DELAY_PROBE_TIME changes"). According to the
commit's description, it was intended to generate such an event
when setting the "delay_first_probe_time" sysctl parameter.

In order to fix this, only generate this event when actually
setting the "delay_first_probe_time" sysctl parameter. This fix
should not have any unintended side-effects, because all but one
registered netevent callbacks check for other netevent event
types (the registered callbacks were obtained by grepping for
"register_netevent_notifier"). The only callback that uses the
NETEVENT_DELAY_PROBE_TIME_UPDATE event is
mlxsw_sp_router_netevent_event() (in
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c): in case
of this event, it only accesses the DELAY_PROBE_TIME of the
passed neigh_parms.

Fixes: 2a4501ae18b5 ("neigh: Send a notification when DELAY_PROBE_TIME changes")
Signed-off-by: Marcus Huewe <suse-tux@gmx.de>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosched: have stub for tcf_destroy_chain in case NET_CLS is not configured
Jiri Pirko [Wed, 15 Feb 2017 10:57:50 +0000 (11:57 +0100)]
sched: have stub for tcf_destroy_chain in case NET_CLS is not configured

This fixes broken build for !NET_CLS:

net/built-in.o: In function `fq_codel_destroy':
/home/sab/linux/net-next/net/sched/sch_fq_codel.c:468: undefined reference to `tcf_destroy_chain'

Fixes: cf1facda2f61 ("sched: move tcf_proto_destroy and tcf_destroy_chain helpers into cls_api")
Reported-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Tested-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: xilinx_emaclite: fix freezes due to unordered I/O
Anssi Hannula [Tue, 14 Feb 2017 17:11:45 +0000 (19:11 +0200)]
net: xilinx_emaclite: fix freezes due to unordered I/O

The xilinx_emaclite uses __raw_writel and __raw_readl for register
accesses. Those functions do not imply any kind of memory barriers and
they may be reordered.

The driver does not seem to take that into account, though, and the
driver does not satisfy the ordering requirements of the hardware.
For clear examples, see xemaclite_mdio_write() and xemaclite_mdio_read()
which try to set MDIO address before initiating the transaction.

I'm seeing system freezes with the driver with GCC 5.4 and current
Linux kernels on Zynq-7000 SoC immediately when trying to use the
interface.

In commit 123c1407af87 ("net: emaclite: Do not use microblaze and ppc
IO functions") the driver was switched from non-generic
in_be32/out_be32 (memory barriers, big endian) to
__raw_readl/__raw_writel (no memory barriers, native endian), so
apparently the device follows system endianness and the driver was
originally written with the assumption of memory barriers.

Rather than try to hunt for each case of missing barrier, just switch
the driver to use iowrite32/ioread32/iowrite32be/ioread32be depending
on endianness instead.

Tested on little-endian Zynq-7000 ARM SoC FPGA.

Signed-off-by: Anssi Hannula <anssi.hannula@bitwise.fi>
Fixes: 123c1407af87 ("net: emaclite: Do not use microblaze and ppc IO
functions")
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: xilinx_emaclite: fix receive buffer overflow
Anssi Hannula [Tue, 14 Feb 2017 17:11:44 +0000 (19:11 +0200)]
net: xilinx_emaclite: fix receive buffer overflow

xilinx_emaclite looks at the received data to try to determine the
Ethernet packet length but does not properly clamp it if
proto_type == ETH_P_IP or 1500 < proto_type <= 1518, causing a buffer
overflow and a panic via skb_panic() as the length exceeds the allocated
skb size.

Fix those cases.

Also add an additional unconditional check with WARN_ON() at the end.

Signed-off-by: Anssi Hannula <anssi.hannula@bitwise.fi>
Fixes: bb81b2ddfa19 ("net: add Xilinx emac lite device driver")
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: Rebuild bpf.o for any dependency update
Mickaël Salaün [Sat, 11 Feb 2017 22:20:23 +0000 (23:20 +0100)]
bpf: Rebuild bpf.o for any dependency update

This is needed to force a rebuild of bpf.o when one of its dependencies
(e.g. uapi/linux/bpf.h) is updated.

Add a phony target.

Signed-off-by: Mickaël Salaün <mic@digikod.net>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David S. Miller <davem@davemloft.net>
Cc: Wang Nan <wangnan0@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: Remove redundant ifdef
Mickaël Salaün [Sat, 11 Feb 2017 19:37:08 +0000 (20:37 +0100)]
bpf: Remove redundant ifdef

Remove a useless ifdef __NR_bpf as requested by Wang Nan.

Inline one-line static functions as it was in the bpf_sys.h file.

Signed-off-by: Mickaël Salaün <mic@digikod.net>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David S. Miller <davem@davemloft.net>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/r/828ab1ff-4dcf-53ff-c97b-074adb895006@huawei.com
Acked-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlx4: do not use rwlock in fast path
Eric Dumazet [Thu, 9 Feb 2017 17:10:04 +0000 (09:10 -0800)]
mlx4: do not use rwlock in fast path

Using a reader-writer lock in fast path is silly, when we can
instead use RCU or a seqlock.

For mlx4 hwstamp clock, a seqlock is the way to go, removing
two atomic operations and false sharing.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoPCI/PME: Restore pcie_pme_driver.remove
Yinghai Lu [Wed, 15 Feb 2017 05:17:48 +0000 (21:17 -0800)]
PCI/PME: Restore pcie_pme_driver.remove

In addition to making PME non-modular, d7def2040077 ("PCI/PME: Make
explicitly non-modular") removed the pcie_pme_driver .remove() method,
pcie_pme_remove().

pcie_pme_remove() freed the PME IRQ that was requested in pci_pme_probe().
The fact that we don't free the IRQ after d7def2040077 causes the following
crash when removing a PCIe port device via /sys:

  ------------[ cut here ]------------
  kernel BUG at drivers/pci/msi.c:370!
  invalid opcode: 0000 [#1] SMP
  Modules linked in:
  CPU: 1 PID: 14509 Comm: sh Tainted: G    W  4.8.0-rc1-yh-00012-gd29438d
  RIP: 0010:[<ffffffff9758bbf5>]  free_msi_irqs+0x65/0x190
  ...
  Call Trace:
   [<ffffffff9758cda4>] pci_disable_msi+0x34/0x40
   [<ffffffff97583817>] cleanup_service_irqs+0x27/0x30
   [<ffffffff97583e9a>] pcie_port_device_remove+0x2a/0x40
   [<ffffffff97584250>] pcie_portdrv_remove+0x40/0x50
   [<ffffffff97576d7b>] pci_device_remove+0x4b/0xc0
   [<ffffffff9785ebe6>] __device_release_driver+0xb6/0x150
   [<ffffffff9785eca5>] device_release_driver+0x25/0x40
   [<ffffffff975702e4>] pci_stop_bus_device+0x74/0xa0
   [<ffffffff975704ea>] pci_stop_and_remove_bus_device_locked+0x1a/0x30
   [<ffffffff97578810>] remove_store+0x50/0x70
   [<ffffffff9785a378>] dev_attr_store+0x18/0x30
   [<ffffffff97260b64>] sysfs_kf_write+0x44/0x60
   [<ffffffff9725feae>] kernfs_fop_write+0x10e/0x190
   [<ffffffff971e13f8>] __vfs_write+0x28/0x110
   [<ffffffff970b0fa4>] ? percpu_down_read+0x44/0x80
   [<ffffffff971e53a7>] ? __sb_start_write+0xa7/0xe0
   [<ffffffff971e53a7>] ? __sb_start_write+0xa7/0xe0
   [<ffffffff971e1f04>] vfs_write+0xc4/0x180
   [<ffffffff971e3089>] SyS_write+0x49/0xa0
   [<ffffffff97001a46>] do_syscall_64+0xa6/0x1b0
   [<ffffffff9819201e>] entry_SYSCALL64_slow_path+0x25/0x25
  ...
   RIP  [<ffffffff9758bbf5>] free_msi_irqs+0x65/0x190
   RSP <ffff89ad3085bc48>
  ---[ end trace f4505e1dac5b95d3 ]---
  Segmentation fault

Restore pcie_pme_remove().

[bhelgaas: changelog]
Fixes: d7def2040077 ("PCI/PME: Make explicitly non-modular")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: stable@vger.kernel.org # v4.9+
7 years agoMerge ath-next from git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/ath.git
Kalle Valo [Wed, 15 Feb 2017 14:01:04 +0000 (16:01 +0200)]
Merge ath-next from git://git./linux/kernel/git/kvalo/ath.git

ath.git patches for 4.11. Major changes:

ath10k

* when trying older firmware versions don't confuse user with error messages

ath9k

* fix crash in AP mode (regression)
* fix relayfs crash (regression)
* fix initialisation with AR9340 and AR9550

7 years agoesp: Add a software GRO codepath
Steffen Klassert [Wed, 15 Feb 2017 08:40:00 +0000 (09:40 +0100)]
esp: Add a software GRO codepath

This patch adds GRO ifrastructure and callbacks for ESP on
ipv4 and ipv6.

In case the GRO layer detects an ESP packet, the
esp{4,6}_gro_receive() function does a xfrm state lookup
and calls the xfrm input layer if it finds a matching state.
The packet will be decapsulated and reinjected it into layer 2.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
7 years agoxfrm: Extend the sec_path for IPsec offloading
Steffen Klassert [Wed, 15 Feb 2017 08:39:54 +0000 (09:39 +0100)]
xfrm: Extend the sec_path for IPsec offloading

We need to keep per packet offloading informations across
the layers. So we extend the sec_path to carry these for
the input and output offload codepath.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
7 years agofuse: fix use after free issue in fuse_dev_do_read()
Sahitya Tummala [Wed, 8 Feb 2017 15:00:56 +0000 (20:30 +0530)]
fuse: fix use after free issue in fuse_dev_do_read()

There is a potential race between fuse_dev_do_write()
and request_wait_answer() contexts as shown below:

TASK 1:
__fuse_request_send():
  |--spin_lock(&fiq->waitq.lock);
  |--queue_request();
  |--spin_unlock(&fiq->waitq.lock);
  |--request_wait_answer():
       |--if (test_bit(FR_SENT, &req->flags))
       <gets pre-empted after it is validated true>
                                   TASK 2:
                                   fuse_dev_do_write():
                                     |--clears bit FR_SENT,
                                     |--request_end():
                                        |--sets bit FR_FINISHED
                                        |--spin_lock(&fiq->waitq.lock);
                                        |--list_del_init(&req->intr_entry);
                                        |--spin_unlock(&fiq->waitq.lock);
                                        |--fuse_put_request();
       |--queue_interrupt();
       <request gets queued to interrupts list>
            |--wake_up_locked(&fiq->waitq);
       |--wait_event_freezable();
       <as FR_FINISHED is set, it returns and then
       the caller frees this request>

Now, the next fuse_dev_do_read(), see interrupts list is not empty
and then calls fuse_read_interrupt() which tries to access the request
which is already free'd and gets the below crash:

[11432.401266] Unable to handle kernel paging request at virtual address
6b6b6b6b6b6b6b6b
...
[11432.418518] Kernel BUG at ffffff80083720e0
[11432.456168] PC is at __list_del_entry+0x6c/0xc4
[11432.463573] LR is at fuse_dev_do_read+0x1ac/0x474
...
[11432.679999] [<ffffff80083720e0>] __list_del_entry+0x6c/0xc4
[11432.687794] [<ffffff80082c65e0>] fuse_dev_do_read+0x1ac/0x474
[11432.693180] [<ffffff80082c6b14>] fuse_dev_read+0x6c/0x78
[11432.699082] [<ffffff80081d5638>] __vfs_read+0xc0/0xe8
[11432.704459] [<ffffff80081d5efc>] vfs_read+0x90/0x108
[11432.709406] [<ffffff80081d67f0>] SyS_read+0x58/0x94

As FR_FINISHED bit is set before deleting the intr_entry with input
queue lock in request completion path, do the testing of this flag and
queueing atomically with the same lock in queue_interrupt().

Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Fixes: fd22d62ed0c3 ("fuse: no fc->lock for iqueue parts")
Cc: <stable@vger.kernel.org> # 4.2+
7 years agoath9k: use correct OTP register offsets for the AR9340 and AR9550
Christian Lamparter [Tue, 14 Feb 2017 19:10:30 +0000 (20:10 +0100)]
ath9k: use correct OTP register offsets for the AR9340 and AR9550

This patch fixes the OTP register definitions for the AR934x and AR9550
WMAC SoC.

Previously, the ath9k driver was unable to initialize the integrated
WMAC on an Aerohive AP121:

| ath: phy0: timeout (1000 us) on reg 0x30018: 0xbadc0ffe & 0x00000007 != 0x00000004
| ath: phy0: timeout (1000 us) on reg 0x30018: 0xbadc0ffe & 0x00000007 != 0x00000004
| ath: phy0: Unable to initialize hardware; initialization status: -5
| ath9k ar934x_wmac: failed to initialize device
| ath9k: probe of ar934x_wmac failed with error -5

It turns out that the AR9300_OTP_STATUS and AR9300_OTP_DATA
definitions contain a typo.

Cc: Gabor Juhos <juhosg@openwrt.org>
Cc: stable@vger.kernel.org
Fixes: add295a4afbdf5852d0 "ath9k: use correct OTP register offsets for AR9550"
Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: Chris Blake <chrisrblake93@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
7 years agort2500usb: don't mark register accesses as inline
Arnd Bergmann [Tue, 14 Feb 2017 21:28:33 +0000 (22:28 +0100)]
rt2500usb: don't mark register accesses as inline

When CONFIG_KASAN is set, we get a rather large stack here:

drivers/net/wireless/ralink/rt2x00/rt2500usb.c: In function 'rt2500usb_set_device_state':
drivers/net/wireless/ralink/rt2x00/rt2500usb.c:1074:1: error: the frame size of 3032 bytes is larger than 100 bytes [-Werror=frame-larger-than=]

If we don't force those functions to be inline, the compiler can figure this
out better itself and not inline the functions when doing so would be harmful,
reducing the stack size to a merge 256 bytes.

Note that there is another problem that manifests in this driver, as a result
of the typecheck() macro causing even larger stack frames.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>