platform/kernel/linux-rpi.git
7 years agonet: stmmac: replace ENOSYS by EINVAL
LABBE Corentin [Wed, 8 Feb 2017 08:31:14 +0000 (09:31 +0100)]
net: stmmac: replace ENOSYS by EINVAL

As said by checkpatch ENOSYS means 'invalid syscall nr' and nothing
else.
This patch replace ENOSYS by the more appropriate value EINVAL.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: Use readl_poll_timeout
LABBE Corentin [Wed, 8 Feb 2017 08:31:13 +0000 (09:31 +0100)]
net: stmmac: Use readl_poll_timeout

The dwmac_dma_reset function use an open coded of readl_poll_timeout().
Replace the open coded handling with the proper function.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: replace stmmac_mdio_busy_wait by readl_poll_timeout
LABBE Corentin [Wed, 8 Feb 2017 08:31:12 +0000 (09:31 +0100)]
net: stmmac: replace stmmac_mdio_busy_wait by readl_poll_timeout

The stmmac_mdio_busy_wait() function do the same job than
readl_poll_timeout().
So is is better to replace it.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: fix some code style problem
LABBE Corentin [Wed, 8 Feb 2017 08:31:11 +0000 (09:31 +0100)]
net: stmmac: fix some code style problem

Checkpatch complains about some code style problem on stmmac_mdio.c.
This patch fix them.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: remplace asm/io.h by linux/io.h
LABBE Corentin [Wed, 8 Feb 2017 08:31:10 +0000 (09:31 +0100)]
net: stmmac: remplace asm/io.h by linux/io.h

This patch fix the checkpatch warning about asm/io.h.
Sorting all includes in the process.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: remove freesoftware address
LABBE Corentin [Wed, 8 Feb 2017 08:31:09 +0000 (09:31 +0100)]
net: stmmac: remove freesoftware address

This patch fix the checkpatch warning about free software address.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: fix some typos in comments
LABBE Corentin [Wed, 8 Feb 2017 08:31:08 +0000 (09:31 +0100)]
net: stmmac: fix some typos in comments

This patch fix some typos in comments.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: Remove the bus_setup function pointer
LABBE Corentin [Wed, 8 Feb 2017 08:31:07 +0000 (09:31 +0100)]
net: stmmac: Remove the bus_setup function pointer

The bus_setup function pointer is not used at all, this patch remove it.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: fix the typo on MAC_RNABLE_RX
LABBE Corentin [Wed, 8 Feb 2017 08:31:06 +0000 (09:31 +0100)]
net: stmmac: fix the typo on MAC_RNABLE_RX

the define MAC_RNABLE_RX have a typo, rename it to MAC_ENABLE_RX

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf, lpm: fix overflows in trie_alloc checks
Daniel Borkmann [Wed, 8 Feb 2017 00:19:43 +0000 (01:19 +0100)]
bpf, lpm: fix overflows in trie_alloc checks

Cap the maximum (total) value size and bail out if larger than KMALLOC_MAX_SIZE
as otherwise it doesn't make any sense to proceed further, since we're
guaranteed to fail to allocate elements anyway in lpm_trie_node_alloc();
likleyhood of failure is still high for large values, though, similarly
as with htab case in non-prealloc.

Next, make sure that cost vars are really u64 instead of size_t, so that we
don't overflow on 32 bit and charge only tiny map.pages against memlock while
allowing huge max_entries; cap also the max cost like we do with other map
types.

Fixes: b95a5c4db09b ("bpf: add a longest prefix match trie map implementation")
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 id info range fill size calc cleanups
Roopa Prabhu [Wed, 8 Feb 2017 00:12:00 +0000 (16:12 -0800)]
bridge: vlan tunnel id info range fill size calc cleanups

This fixes a bug and cleans up tunnelid range size
calculation code by using consistent variable names
and checks in size calculation and fill functions.

tested for a few cases of vlan-vni range mappings:
(output from patched iproute2):
$bridge vlan showtunnel
port     vid        tunid
vxlan0   100-105    1000-1005
         200        2000
         210        2100
         211-213    2100-2102
         214        2104
         216-217    2108-2109
         219        2119

Fixes: efa5356b0d97 ("bridge: per vlan dst_metadata netlink support")
Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogro_cells: move to net/core/gro_cells.c
Eric Dumazet [Tue, 7 Feb 2017 23:37:15 +0000 (15:37 -0800)]
gro_cells: move to net/core/gro_cells.c

We have many gro cells users, so lets move the code to avoid
duplication.

This creates a CONFIG_GRO_CELLS option.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mellanox: switchx2: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Tue, 7 Feb 2017 23:07:33 +0000 (00:07 +0100)]
net: mellanox: switchx2: 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>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Tested-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: qcom/emac: fix semicolon.cocci warnings
Wu Fengguang [Tue, 7 Feb 2017 19:42:05 +0000 (03:42 +0800)]
net: qcom/emac: fix semicolon.cocci warnings

drivers/net/ethernet/qualcomm/emac/emac-ethtool.c:155:49-50: Unneeded semicolon

 Remove unneeded semicolon.

Generated by: scripts/coccinelle/misc/semicolon.cocci

CC: Timur Tabi <timur@codeaurora.org>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: Add LED mode driver for Microsemi PHYs.
Raju Lakkaraju [Tue, 7 Feb 2017 13:40:26 +0000 (19:10 +0530)]
net: phy: Add LED mode driver for Microsemi PHYs.

LED Mode:
Microsemi PHY support 2 LEDs (LED[0] and LED[1]) to display different
status information that can be selected by setting LED mode.

LED Mode parameter (vsc8531, led-0-mode) and (vsc8531, led-1-mode) get
from Device Tree.

Signed-off-by: Raju Lakkaraju <Raju.Lakkaraju@microsemi.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: bcm_sf2: cleanup bcm_sf2_cfp_rule_get() a little
Dan Carpenter [Tue, 7 Feb 2017 13:15:27 +0000 (16:15 +0300)]
net: dsa: bcm_sf2: cleanup bcm_sf2_cfp_rule_get() a little

This patch doesn't affect how the code works.

My static checker complains that the mask and shift doesn't make sense
because 0xffffff << 16 goes beyond the end of 32 bits.  It should be
0xffff instead but the existing code won't cause runtime bugs.

Also the casting here is not needed and not consistent with the rest of
the code.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Tue, 7 Feb 2017 21:29:30 +0000 (16:29 -0500)]
Merge git://git./linux/kernel/git/davem/net

The conflict was an interaction between a bug fix in the
netvsc driver in 'net' and an optimization of the RX path
in 'net-next'.

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: do not dereference pointer if it's NULL
Felix Manlunas [Tue, 7 Feb 2017 20:10:58 +0000 (12:10 -0800)]
liquidio: do not dereference pointer if it's NULL

Fix smatch errors by not dereferencing iq pointer if it's NULL.

See http://marc.info/?l=kernel-janitors&m=148637299004834&w=2

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 7 Feb 2017 20:10:57 +0000 (12:10 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Load correct firmware in rtl8192ce wireless driver, from Jurij
    Smakov.

 2) Fix leak of tx_ring and tx_cq due to overwriting in mlx4 driver,
    from Martin KaFai Lau.

 3) Need to reference count PHY driver module when it is attached, from
    Mao Wenan.

 4) Don't do zero length vzalloc() in ethtool register dump, from
    Stanislaw Gruszka.

 5) Defer net_disable_timestamp() to a workqueue to get out of locking
    issues, from Eric Dumazet.

 6) We cannot drop the SKB dst when IP options refer to them, fix also
    from Eric Dumazet.

 7) Incorrect packet header offset calculations in ip6_gre, again from
    Eric Dumazet.

 8) Missing tcp_v6_restore_cb() causes use-after-free, from Eric too.

 9) tcp_splice_read() can get into an infinite loop with URG, and hey
    it's from Eric once more.

10) vnet_hdr_sz can change asynchronously, so read it once during
    decision making in macvtap and tun, from Willem de Bruijn.

11) Can't use kernel stack for DMA transfers in USB networking drivers,
    from Ben Hutchings.

12) Handle csum errors properly in UDP by calling the proper destructor,
    from Eric Dumazet.

13) For non-deterministic softirq run when scheduling NAPI from a
    workqueue in mlx4, from Benjamin Poirier.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (28 commits)
  sctp: check af before verify address in sctp_addr_id2transport
  sctp: avoid BUG_ON on sctp_wait_for_sndbuf
  mlx4: Invoke softirqs after napi_reschedule
  udp: properly cope with csum errors
  catc: Use heap buffer for memory size test
  catc: Combine failure cleanup code in catc_probe()
  rtl8150: Use heap buffers for all register access
  pegasus: Use heap buffers for all register access
  macvtap: read vnet_hdr_size once
  tun: read vnet_hdr_sz once
  tcp: avoid infinite loop in tcp_splice_read()
  hns: avoid stack overflow with CONFIG_KASAN
  ipv6: Fix IPv6 packet loss in scenarios involving roaming + snooping switches
  ipv6: tcp: add a missing tcp_v6_restore_cb()
  nl80211: Fix mesh HT operation check
  mac80211: Fix adding of mesh vendor IEs
  mac80211: Allocate a sync skcipher explicitly for FILS AEAD
  mac80211: Fix FILS AEAD protection in Association Request frame
  ip6_gre: fix ip6gre_err() invalid reads
  netlabel: out of bound access in cipso_v4_validate()
  ...

7 years agomm: fix KPF_SWAPCACHE in /proc/kpageflags
Hugh Dickins [Tue, 7 Feb 2017 19:11:16 +0000 (11:11 -0800)]
mm: fix KPF_SWAPCACHE in /proc/kpageflags

Commit 6326fec1122c ("mm: Use owner_priv bit for PageSwapCache, valid
when PageSwapBacked") aliased PG_swapcache to PG_owner_priv_1 (and
depending on PageSwapBacked being true).

As a result, the KPF_SWAPCACHE bit in '/proc/kpageflags' should now be
synthesized, instead of being shown on unrelated pages which just happen
to have PG_owner_priv_1 set.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agobridge: avoid unnecessary read of jiffies
stephen hemminger [Tue, 7 Feb 2017 16:46:46 +0000 (08:46 -0800)]
bridge: avoid unnecessary read of jiffies

Jiffies is volatile so read it once.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agospectrum: acl_tcam: Fix catchall prio value
Jiri Pirko [Tue, 7 Feb 2017 16:27:47 +0000 (17:27 +0100)]
spectrum: acl_tcam: Fix catchall prio value

This fixes an issue reported by smatch:
mlxsw_sp_acl_tcam_chunk_create() warn: impossible condition '(priority == (-1)) => (0-u32max == u64max)'

Reported-by: Or Gerlitz <ogerlitz@mellanox.com>
Reported-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Fixes: 22a677661f56 ("mlxsw: spectrum: Introduce ACL core with simple TCAM implementation")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: remove unnecessary check for vtbegin in br_fill_vlan_tinfo_range
Roopa Prabhu [Tue, 7 Feb 2017 14:43:23 +0000 (06:43 -0800)]
bridge: remove unnecessary check for vtbegin in br_fill_vlan_tinfo_range

vtbegin should not be NULL in this function, Its already checked by the
caller.

this should silence the below smatch complaint:
  net/bridge/br_netlink_tunnel.c:144 br_fill_vlan_tinfo_range()
    error: we previously assumed 'vtbegin' could be null (see line 130)

net/bridge/br_netlink_tunnel.c
   129
   130      if (vtbegin && vtend && (vtend->vid - vtbegin->vid) > 0) {
                    ^^^^^^^
Check for NULL.

Fixes: efa5356b0d97 ("bridge: per vlan dst_metadata netlink support")
Reported-By: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: drop __packed from almost all SCTP structures
Marcelo Ricardo Leitner [Tue, 7 Feb 2017 13:37:56 +0000 (11:37 -0200)]
sctp: drop __packed from almost all SCTP structures

__packed is considered harmful as it potentially generates code that
doesn't perform well and its usage should be avoided as much as
possible.

This patch drops __packed from all SCTP structures except one, which is
sctp_signed_cookie. In there it's required, as per changelog on
commit 9834a2bb4970 ("[SCTP]: Fix sctp_cookie alignment in the packet.").

After this patch, no alignment changes neither in x86 or x86_64 and
no exceptions were noticed during testing on both archs.

Code size for SCTP module also didn't change with this patch.

Cc: David Miller <davem@davemloft.net>
Cc: David Laight <David.Laight@ACULAB.COM>
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: check af before verify address in sctp_addr_id2transport
Xin Long [Tue, 7 Feb 2017 12:56:08 +0000 (20:56 +0800)]
sctp: check af before verify address in sctp_addr_id2transport

Commit 6f29a1306131 ("sctp: sctp_addr_id2transport should verify the
addr before looking up assoc") invoked sctp_verify_addr to verify the
addr.

But it didn't check af variable beforehand, once users pass an address
with family = 0 through sockopt, sctp_get_af_specific will return NULL
and NULL pointer dereference will be caused by af->sockaddr_len.

This patch is to fix it by returning NULL if af variable is NULL.

Fixes: 6f29a1306131 ("sctp: sctp_addr_id2transport should verify the addr before looking up assoc")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: tunnel: fix attribute checks in br_parse_vlan_tunnel_info
Nikolay Aleksandrov [Tue, 7 Feb 2017 11:46:46 +0000 (12:46 +0100)]
bridge: tunnel: fix attribute checks in br_parse_vlan_tunnel_info

These checks should go after the attributes have been parsed otherwise
we're using tb uninitialized.

Fixes: efa5356b0d97 ("bridge: per vlan dst_metadata netlink support")
Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: bridge: remove redundant check to see if err is set
Colin Ian King [Tue, 7 Feb 2017 10:56:38 +0000 (10:56 +0000)]
net: bridge: remove redundant check to see if err is set

The error check on err is redundant as it is being checked
previously each time it has been updated.  Remove this redundant
check.

Detected with CoverityScan, CID#140030("Logically dead code")

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosfc: fix an off by one bug
Dan Carpenter [Tue, 7 Feb 2017 07:44:31 +0000 (10:44 +0300)]
sfc: fix an off by one bug

This bug is harmless because it's just a sanity check and we always
pass valid values for "encap_type" but the test is off by one.

Fixes: 9b4108012517 ("sfc: insert catch-all filters for encapsulated traffic")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: dp83867: Recover from "port mirroring" N/A MODE4
Lukasz Majewski [Tue, 7 Feb 2017 05:20:24 +0000 (06:20 +0100)]
net: phy: dp83867: Recover from "port mirroring" N/A MODE4

The DP83867 when not properly bootstrapped - especially with LED_0 pin -
can enter N/A MODE4 for "port mirroring" feature.

To provide normal operation of the PHY, one needs not only to explicitly
disable the port mirroring feature, but as well stop some IC internal
testing (which disables RGMII communication).

To do that the STRAP_STS1 (0x006E) register must be read and RESERVED bit
11 examined. When it is set, the another RESERVED bit (11) at PHYCR
(0x0010) register must be clear to disable testing mode and enable RGMII
communication.

Thorough explanation of the problem can be found at following e2e thread:
"DP83867IR: Problem with RESERVED bits in PHY Control Register (PHYCR) -
Linux driver"

https://e2e.ti.com/support/interface/ethernet/f/903/p/571313/2096954#2096954

Signed-off-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: dp83867: Add lane swapping support in the DP83867 TI's PHY driver
Lukasz Majewski [Tue, 7 Feb 2017 05:20:23 +0000 (06:20 +0100)]
net: phy: dp83867: Add lane swapping support in the DP83867 TI's PHY driver

This patch adds support for enabling or disabling the lane swapping (called
"port mirroring" in PHY's CFG4 register) feature of the DP83867 TI's PHY
device.

One use case is when bootstrap configuration enables this feature (because
of e.g. LED_0 wrong wiring) so then one needs to disable it in software
(at u-boot/Linux).

Signed-off-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoDocumentation: devicetree: Add PHY no lane swap binding
Lukasz Majewski [Tue, 7 Feb 2017 05:21:34 +0000 (06:21 +0100)]
Documentation: devicetree: Add PHY no lane swap binding

Add the documentation to avoid PHY lane swapping. This is a boolean
entry to notify the phy device drivers that the TX/RX lanes NO need
to be swapped.
The use case for this binding mostly happens after wrong HW
configuration of PHY IC during bootstrap.

Signed-off-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'Incorrect-use-of-phy_read_status'
David S. Miller [Tue, 7 Feb 2017 18:48:23 +0000 (13:48 -0500)]
Merge branch 'Incorrect-use-of-phy_read_status'

Florian Fainelli says:

====================
net: Incorrect use of phy_read_status()

This patch series removes incorrect uses of phy_read_status() which can clobber
the PHY device link while we are executing with the state machine running.

greth was potentially another candidate, but it does funky stuff with
auto-negotation that I am still trying to understand.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Do not clobber PHY link outside of state machine
Florian Fainelli [Mon, 6 Feb 2017 23:55:23 +0000 (15:55 -0800)]
net: dsa: Do not clobber PHY link outside of state machine

Calling phy_read_status() means that we may call into
genphy_read_status() which in turn will use genphy_update_link() which
can make changes to phydev->link outside of the state machine's state
transitions. This is an invalid behavior that is now caught as of
811a919135b9 ("phy state machine: failsafe leave invalid RUNNING state")

Reported-by: Zefir Kurtisi <zefir.kurtisi@neratec.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: netcp: Do not clobber PHY link outside of state machine
Florian Fainelli [Mon, 6 Feb 2017 23:55:22 +0000 (15:55 -0800)]
net: netcp: Do not clobber PHY link outside of state machine

Calling phy_read_status() means that we may call into
genphy_read_status() which in turn will use genphy_update_link() which
can make changes to phydev->link outside of the state machine's state
transitions. This is an invalid behavior that is now caught as off
811a919135b9 ("phy state machine: failsafe leave invalid RUNNING state")

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: pxa168_eth: Do not clobber PHY link outside of state machine
Florian Fainelli [Mon, 6 Feb 2017 23:55:21 +0000 (15:55 -0800)]
net: pxa168_eth: Do not clobber PHY link outside of state machine

Calling phy_read_status() means that we may call into
genphy_read_status() which in turn will use genphy_update_link() which
can make changes to phydev->link outside of the state machine's state
transitions. This is an invalid behavior that is now caught as of
811a919135b9 ("phy state machine: failsafe leave invalid RUNNING state")

Since we don't have anything special, switch to the generic
phy_ethtool_get_link_ksettings() function now.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mv643xx_eth: Do not clobber PHY link outside of state machine
Florian Fainelli [Mon, 6 Feb 2017 23:55:20 +0000 (15:55 -0800)]
net: mv643xx_eth: Do not clobber PHY link outside of state machine

Calling phy_read_status() means that we may call into
genphy_read_status() which in turn will use genphy_update_link() which
can make changes to phydev->link outside of the state machine's state
transitions. This is an invalid behavior that is now caught as of
811a919135b9 ("phy state machine: failsafe leave invalid RUNNING state")

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'mlx5-updates-2017-01-31' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Tue, 7 Feb 2017 18:44:08 +0000 (13:44 -0500)]
Merge tag 'mlx5-updates-2017-01-31' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2017-01-31

This series includes some updates to mlx5 core and ethernet driver.

We got one patch from Or to fix some static checker warnings.

2nd patche from Dan came to add the support for 128B cache line
in the HCA, which will configures the hardware to use 128B alignment only
on systems with 128B cache lines, otherwise it will be kept as the current
default of 64B.

From me three patches to support no inline copy on TX on ConnectX-5 and
later HCAs.  Starting with two small infrastructure changes and
refactoring patches followed by two patches to add the actual support for
both xmit ndo and XDP xmit routines.
Last patch is a simple fix to return a mistakenly removed pointer from the
SQ structure, which was remove in previous submission of mlx5 4K UAR.

Saeed.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: ti: cpsw: remove netif_trans_update
Ivan Khoronzhuk [Mon, 6 Feb 2017 22:53:45 +0000 (00:53 +0200)]
net: ethernet: ti: cpsw: remove netif_trans_update

No need to update jiffies in txq->trans_start twice, it's supposed to be
done in netdev_start_xmit() and anyway is re-written. Also, no reason to
update trans time in case of an error.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bnxt_en-Add-XDP-support'
David S. Miller [Tue, 7 Feb 2017 18:31:01 +0000 (13:31 -0500)]
Merge branch 'bnxt_en-Add-XDP-support'

Michael Chan says:

====================
bnxt_en: Add XDP support.

The first 10 patches refactor the code (rx/tx code paths and ring logic)
and add the basic infrastructure to support XDP.  The 11th patch adds
basic ndo_xdp to support XDP_DROP and XDP_PASS only.  The 12th patch
completes the series with XDP_TX.

Thanks to Andy Gospodarek for testing and uncovering some bugs.

v3: Removed Kconfig option.
    Pass modified offset and length to stack for XDP_PASS.
    Improved buffer recycling scheme for XDP_TX.
    Other minor fixes.

v2: Addressed review comments from Alexei Starovoitov, Jakub Kicinski,
and David Miller:
- Added missing dma syncs.
- Added XDP headroom support.
- Added tracing in exception path.
- Clarified a parameter change.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add support for XDP_TX action.
Michael Chan [Mon, 6 Feb 2017 21:55:43 +0000 (16:55 -0500)]
bnxt_en: Add support for XDP_TX action.

Add dedicated transmit function and transmit completion handler for
XDP.  The XDP transmit logic and completion logic are different than
regular TX ring.  The TX buffer is recycled back to the RX ring when
it completes.

v3: Improved the buffer recyling scheme for XDP_TX.

v2: Add trace_xdp_exception().
    Add dma_sync.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Tested-by: Andy Gospodarek <gospo@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add basic XDP support.
Michael Chan [Mon, 6 Feb 2017 21:55:42 +0000 (16:55 -0500)]
bnxt_en: Add basic XDP support.

Add basic ndo_xdp support to setup and query program, configure the NIC
to run in rx page mode, and support XDP_PASS, XDP_DROP, XDP_ABORTED
actions only.

v3: Pass modified offset and length to stack for XDP_PASS.
    Remove Kconfig option.

v2: Added trace_xdp_exception()
    Added dma_syncs.
    Added XDP headroom support.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Tested-by: Andy Gospodarek <gospo@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Refactor tx completion path.
Michael Chan [Mon, 6 Feb 2017 21:55:41 +0000 (16:55 -0500)]
bnxt_en: Refactor tx completion path.

XDP_TX requires a different function to handle completion.  Add a
function pointer to handle tx completion logic.  Regular TX rings
will be assigned the current bnxt_tx_int() for the ->tx_int()
function pointer.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add a set of TX rings to support XDP.
Michael Chan [Mon, 6 Feb 2017 21:55:40 +0000 (16:55 -0500)]
bnxt_en: Add a set of TX rings to support XDP.

Add logic for an extra set of TX rings for XDP.  If enabled, this
set of TX rings equals the number of RX rings and shares the same
IRQ as the RX ring set.  A new field bp->tx_nr_rings_xdp is added
to keep track of these TX XDP rings.  Adjust all other relevant functions
to handle bp->tx_nr_rings_xdp.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add tx ring mapping logic.
Michael Chan [Mon, 6 Feb 2017 21:55:39 +0000 (16:55 -0500)]
bnxt_en: Add tx ring mapping logic.

To support XDP_TX, we need to add a set of dedicated TX rings, each
associated with the NAPI of an RX ring.  To assign XDP rings and regular
rings in a flexible way, we add a bp->tx_ring_map[] array to do the
remapping.  The netdev txq index is stored in the new field txq_index
so that we can retrieve the netdev txq when handling TX completions.
In this patch, before we introduce XDP_TX, the mapping is 1:1.

v2: Fixed a bug in bnxt_tx_int().

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Centralize logic to reserve rings.
Michael Chan [Mon, 6 Feb 2017 21:55:38 +0000 (16:55 -0500)]
bnxt_en: Centralize logic to reserve rings.

Currently, bnxt_setup_tc() and bnxt_set_channels() have similar and
duplicated code to check and reserve rx and tx rings.  Add a new
function bnxt_reserve_rings() to centralize the logic.  This will
make it easier to add XDP_TX support which requires allocating a
new set of TX rings.

Also, the tx ring checking logic in bnxt_setup_msix() can be removed.
The rings have been reserved before hand.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Use event bit map in RX path.
Michael Chan [Mon, 6 Feb 2017 21:55:37 +0000 (16:55 -0500)]
bnxt_en: Use event bit map in RX path.

In the current code, we have separate rx_event and agg_event parameters
to keep track of rx and aggregation events.  Combine these events into
an u8 event mask with different bits defined for different events.  This
way, it is easier to expand the logic to include XDP tx events.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add RX page mode support.
Michael Chan [Mon, 6 Feb 2017 21:55:36 +0000 (16:55 -0500)]
bnxt_en: Add RX page mode support.

This mode is to support XDP.  In this mode, each rx ring is configured
with page sized buffers for linear placement of each packet.  MTU will be
restricted to what the page sized buffers can support.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Parameterize RX buffer offsets.
Michael Chan [Mon, 6 Feb 2017 21:55:35 +0000 (16:55 -0500)]
bnxt_en: Parameterize RX buffer offsets.

Convert the global constants BNXT_RX_OFFSET and BNXT_RX_DMA_OFFSET to
device parameters.  This will make it easier to support XDP with
headroom support which requires different RX buffer offsets.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Add bp->rx_dir field for rx buffer DMA direction.
Michael Chan [Mon, 6 Feb 2017 21:55:34 +0000 (16:55 -0500)]
bnxt_en: Add bp->rx_dir field for rx buffer DMA direction.

When driver is running in XDP mode, rx buffers are DMA mapped as
DMA_BIDIRECTIONAL.  Add a field so the code will map/unmap rx buffers
according to this field.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Don't use DEFINE_DMA_UNMAP_ADDR to store DMA address in RX path.
Michael Chan [Mon, 6 Feb 2017 21:55:33 +0000 (16:55 -0500)]
bnxt_en: Don't use DEFINE_DMA_UNMAP_ADDR to store DMA address in RX path.

To support XDP_TX, we need the RX buffer's DMA address to transmit the
packet.  Convert the DMA address field to a permanent field in
bnxt_sw_rx_bd.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobnxt_en: Refactor rx SKB function.
Michael Chan [Mon, 6 Feb 2017 21:55:32 +0000 (16:55 -0500)]
bnxt_en: Refactor rx SKB function.

Minor refactoring of bnxt_rx_skb() so that it can easily be replaced by
a new function that handles packets in a single page.  Also, use a
function pointer bp->rx_skb_func() to switch to a new function when
we add the new mode in the next patch.

Add a new field data_ptr that points to the packet data in the
bnxt_sw_rx_bd structure.  The original data field is changed to void
pointer so that it can either hold the kmalloc'ed data or a page
pointer.

The last parameter of bnxt_rx_skb() which was the length parameter is
changed to include the payload offset of the packet in the upper 16 bit.
The offset is needed to support the rx page mode and is not used in
this existing function.

v3: Added a new data_ptr parameter to bp->rx_skb_func().  The caller
has the option to modify the starting address of the packet.  This
will be needed when XDP with headroom support is added.

v2: Changed the name of the last parameter to offset_and_len to make the
code more clear.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: qcom/emac: add ethool support for setting pause parameters
Timur Tabi [Mon, 6 Feb 2017 21:34:52 +0000 (15:34 -0600)]
net: qcom/emac: add ethool support for setting pause parameters

To support setting the pause parameters, the driver can no longer just
mirror the PHY.  The set_pauseparam feature allows the driver to
force the setting in the MAC, regardless of how the PHY is configured.
This means that we now need to maintain an internal state for pause
frame support, and so get_pauseparam also needs to be updated.

If the interface is already running when the setting is changed, then
the interface is reset.

Note that if the MAC is configured to enable RX pause frame support
(i.e. it transmits pause frames to throttle the other end), but the
PHY is configured to block those frames, then the feature will not work.

Also some buffer size initialization code into emac_init_adapter(),
so that it lives with similar code, including the initializtion of
pause frame support.

Signed-off-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'replace-dst_confirm'
David S. Miller [Tue, 7 Feb 2017 18:07:56 +0000 (13:07 -0500)]
Merge branch 'replace-dst_confirm'

Julian Anastasov says:

====================
net: dst_confirm replacement

This patchset addresses the problem of neighbour
confirmation where received replies from one nexthop
can cause confirmation of different nexthop when using
the same dst. Thanks to YueHaibing <yuehaibing@huawei.com>
for tracking the dst->pending_confirm problem.

Sockets can obtain cached output route. Such
routes can be to known nexthop (rt_gateway=IP) or to be
used simultaneously for different nexthop IPs by different
subnet prefixes (nh->nh_scope = RT_SCOPE_HOST, rt_gateway=0).

At first look, there are more problems:

- dst_confirm() sets flag on dst and not on dst->path,
as result, indication is lost when XFRM is used

- DNAT can change the nexthop, so the really used nexthop is
not confirmed

So, the following solution is to avoid using
dst->pending_confirm.

The current dst_confirm() usage is as follows:

Protocols confirming dst on received packets:
- TCP (1 dst per socket)
- SCTP (1 dst per transport)
- CXGB*

Protocols supporting sendmsg with MSG_CONFIRM [ | MSG_PROBE ] to
confirm neighbour:
- UDP IPv4/IPv6
- ICMPv4 PING
- RAW IPv4/IPv6
- L2TP/IPv6

MSG_CONFIRM for other purposes (fix not needed):
- CAN

Sending without locking the socket:
- UDP (when no cork)
- RAW (when hdrincl=1)

Redirects from old to new GW:
- rt6_do_redirect

The patchset includes the following changes:

1. sock: add sk_dst_pending_confirm flag

- used only by TCP with patch 4 to remember the received
indication in sk->sk_dst_pending_confirm

2. net: add dst_pending_confirm flag to skbuff

- skb->dst_pending_confirm will be used by all protocols
in following patches, via skb_{set,get}_dst_pending_confirm

3. sctp: add dst_pending_confirm flag

- SCTP uses per-transport dsts and can not use
sk->sk_dst_pending_confirm like TCP

4. tcp: replace dst_confirm with sk_dst_confirm

5. net: add confirm_neigh method to dst_ops

- IPv4 and IPv6 provision for slow neigh lookups for MSG_PROBE users.
I decided to use neigh lookup only for this case because on
MSG_PROBE the skb may pass MTU checks but it does not reach
the neigh confirmation code. This patch will be used from patch 6.

- xfrm_confirm_neigh: we use the last tunnel address, if present.
When there are only transports, the original dest address is used.

6. net: use dst_confirm_neigh for UDP, RAW, ICMP, L2TP

- dst_confirm conversion for UDP, RAW, ICMP and L2TP/IPv6

- these protocols use MSG_CONFIRM propagated by ip*_append_data
to skb->dst_pending_confirm. sk->sk_dst_pending_confirm is not
used because some sending paths do not lock the socket. For
MSG_PROBE we use the slow lookup (dst_confirm_neigh).

- there are also 2 cases that need the slow lookup:
__ip6_rt_update_pmtu and rt6_do_redirect. I hope
&ipv6_hdr(skb)->saddr is the correct nexthop address to use here.

7. net: pending_confirm is not used anymore

- I failed to understand the CXGB* code, I see dst_confirm()
calls but I'm not sure dst_neigh_output() was called. For now
I just removed the dst->pending_confirm flag and left all
dst_confirm() calls there. Any better idea?

- Now may be old function neigh_output() should be restored
instead of dst_neigh_output?
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: pending_confirm is not used anymore
Julian Anastasov [Mon, 6 Feb 2017 21:14:17 +0000 (23:14 +0200)]
net: pending_confirm is not used anymore

When same struct dst_entry can be used for many different
neighbours we can not use it for pending confirmations.
As last step, we can remove the pending_confirm flag.

Reported-by: YueHaibing <yuehaibing@huawei.com>
Fixes: 5110effee8fd ("net: Do delayed neigh confirmation.")
Fixes: f2bb4bedf35d ("ipv4: Cache output routes in fib_info nexthops.")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: use dst_confirm_neigh for UDP, RAW, ICMP, L2TP
Julian Anastasov [Mon, 6 Feb 2017 21:14:16 +0000 (23:14 +0200)]
net: use dst_confirm_neigh for UDP, RAW, ICMP, L2TP

When same struct dst_entry can be used for many different
neighbours we can not use it for pending confirmations.

The datagram protocols can use MSG_CONFIRM to confirm the
neighbour. When used with MSG_PROBE we do not reach the
code where neighbour is confirmed, so we have to do the
same slow lookup by using the dst_confirm_neigh() helper.
When MSG_PROBE is not used, ip_append_data/ip6_append_data
will set the skb flag dst_pending_confirm.

Reported-by: YueHaibing <yuehaibing@huawei.com>
Fixes: 5110effee8fd ("net: Do delayed neigh confirmation.")
Fixes: f2bb4bedf35d ("ipv4: Cache output routes in fib_info nexthops.")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: add confirm_neigh method to dst_ops
Julian Anastasov [Mon, 6 Feb 2017 21:14:15 +0000 (23:14 +0200)]
net: add confirm_neigh method to dst_ops

Add confirm_neigh method to dst_ops and use it from IPv4 and IPv6
to lookup and confirm the neighbour. Its usage via the new helper
dst_confirm_neigh() should be restricted to MSG_PROBE users for
performance reasons.

For XFRM prefer the last tunnel address, if present. With help
from Steffen Klassert.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: replace dst_confirm with sk_dst_confirm
Julian Anastasov [Mon, 6 Feb 2017 21:14:14 +0000 (23:14 +0200)]
tcp: replace dst_confirm with sk_dst_confirm

When same struct dst_entry can be used for many different
neighbours we can not use it for pending confirmations.
Use the new sk_dst_confirm() helper to propagate the
indication from received packets to sock_confirm_neigh().

Reported-by: YueHaibing <yuehaibing@huawei.com>
Fixes: 5110effee8fd ("net: Do delayed neigh confirmation.")
Fixes: f2bb4bedf35d ("ipv4: Cache output routes in fib_info nexthops.")
Tested-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: add dst_pending_confirm flag
Julian Anastasov [Mon, 6 Feb 2017 21:14:13 +0000 (23:14 +0200)]
sctp: add dst_pending_confirm flag

Add new transport flag to allow sockets to confirm neighbour.
When same struct dst_entry can be used for many different
neighbours we can not use it for pending confirmations.
The flag is propagated from transport to every packet.
It is reset when cached dst is reset.

Reported-by: YueHaibing <yuehaibing@huawei.com>
Fixes: 5110effee8fd ("net: Do delayed neigh confirmation.")
Fixes: f2bb4bedf35d ("ipv4: Cache output routes in fib_info nexthops.")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: add dst_pending_confirm flag to skbuff
Julian Anastasov [Mon, 6 Feb 2017 21:14:12 +0000 (23:14 +0200)]
net: add dst_pending_confirm flag to skbuff

Add new skbuff flag to allow protocols to confirm neighbour.
When same struct dst_entry can be used for many different
neighbours we can not use it for pending confirmations.

Add sock_confirm_neigh() helper to confirm the neighbour and
use it for IPv4, IPv6 and VRF before dst_neigh_output.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosock: add sk_dst_pending_confirm flag
Julian Anastasov [Mon, 6 Feb 2017 21:14:11 +0000 (23:14 +0200)]
sock: add sk_dst_pending_confirm flag

Add new sock flag to allow sockets to confirm neighbour.
When same struct dst_entry can be used for many different
neighbours we can not use it for pending confirmations.
As not all call paths lock the socket use full word for
the flag.

Add sk_dst_confirm as replacement for dst_confirm when
called for received packets.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: bcm7xxx: Add BCM74371 PHY ID
Florian Fainelli [Mon, 6 Feb 2017 21:01:16 +0000 (13:01 -0800)]
net: phy: bcm7xxx: Add BCM74371 PHY ID

Add the BCM74371 PHY ID to the list of supported chips. This is a 28nm
technology Gigabit PHY SoC.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoARC: [arcompact] brown paper bag bug in unaligned access delay slot fixup
Vineet Gupta [Tue, 7 Feb 2017 17:44:58 +0000 (09:44 -0800)]
ARC: [arcompact] brown paper bag bug in unaligned access delay slot fixup

Reported-by: Jo-Philipp Wich <jo@mein.io>
Fixes: 9aed02feae57bf7 ("ARC: [arcompact] handle unaligned access delay slot")
Cc: linux-kernel@vger.kernel.org
Cc: linux-snps-arc@lists.infradead.org
Cc: stable@vger.kernel.org
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agosctp: avoid BUG_ON on sctp_wait_for_sndbuf
Marcelo Ricardo Leitner [Mon, 6 Feb 2017 20:10:31 +0000 (18:10 -0200)]
sctp: avoid BUG_ON on sctp_wait_for_sndbuf

Alexander Popov reported that an application may trigger a BUG_ON in
sctp_wait_for_sndbuf if the socket tx buffer is full, a thread is
waiting on it to queue more data and meanwhile another thread peels off
the association being used by the first thread.

This patch replaces the BUG_ON call with a proper error handling. It
will return -EPIPE to the original sendmsg call, similarly to what would
have been done if the association wasn't found in the first place.

Acked-by: Alexander Popov <alex.popov@linux.com>
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlx4: Invoke softirqs after napi_reschedule
Benjamin Poirier [Mon, 6 Feb 2017 18:14:31 +0000 (10:14 -0800)]
mlx4: Invoke softirqs after napi_reschedule

mlx4 may schedule napi from a workqueue. Afterwards, softirqs are not run
in a deterministic time frame and the following message may be logged:
NOHZ: local_softirq_pending 08

The problem is the same as what was described in commit ec13ee80145c
("virtio_net: invoke softirqs after __napi_schedule") and this patch
applies the same fix to mlx4.

Fixes: 07841f9d94c1 ("net/mlx4_en: Schedule napi when RX buffers allocation fails")
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: add psample dependency for spectrum
Arnd Bergmann [Mon, 6 Feb 2017 16:26:30 +0000 (17:26 +0100)]
mlxsw: add psample dependency for spectrum

When PSAMPLE is a loadable module, spectrum must not be built-in:

drivers/net/built-in.o: In function `mlxsw_sp_rx_listener_sample_func':
spectrum.c:(.text+0xe357e): undefined reference to `psample_sample_packet'

This adds a Kconfig dependency to enforce usable configurations.

Fixes: 98d0f7b9acda ("mlxsw: spectrum: Add packet sample offloading support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Yotam Gigi <yotamg@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipv6: sr: fix non static symbol warnings
Wei Yongjun [Mon, 6 Feb 2017 16:15:05 +0000 (16:15 +0000)]
ipv6: sr: fix non static symbol warnings

Fixes the following sparse warnings:

net/ipv6/seg6_iptunnel.c:58:5: warning:
 symbol 'nla_put_srh' was not declared. Should it be static?
net/ipv6/seg6_iptunnel.c:238:5: warning:
 symbol 'seg6_input' was not declared. Should it be static?
net/ipv6/seg6_iptunnel.c:254:5: warning:
 symbol 'seg6_output' was not declared. Should it be static?

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: act_mirred: remove duplicated include from act_mirred.c
Wei Yongjun [Mon, 6 Feb 2017 16:07:13 +0000 (16:07 +0000)]
net/sched: act_mirred: remove duplicated include from act_mirred.c

Remove duplicated include.

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: wan: slic_ds26522: Remove .owner field for driver
Wei Yongjun [Mon, 6 Feb 2017 15:26:17 +0000 (15:26 +0000)]
net: wan: slic_ds26522: Remove .owner field for driver

Remove .owner field if calls are used which set it automatically.

Generated by: scripts/coccinelle/api/platform_no_drv_owner.cocci

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: wan: slic_ds26522: Use module_spi_driver to simplify the code
Wei Yongjun [Mon, 6 Feb 2017 15:25:52 +0000 (15:25 +0000)]
net: wan: slic_ds26522: Use module_spi_driver to simplify the code

module_spi_driver() makes the code simpler by eliminating
boilerplate code.

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoudp: properly cope with csum errors
Eric Dumazet [Sun, 5 Feb 2017 17:25:24 +0000 (09:25 -0800)]
udp: properly cope with csum errors

Dmitry reported that UDP sockets being destroyed would trigger the
WARN_ON(atomic_read(&sk->sk_rmem_alloc)); in inet_sock_destruct()

It turns out we do not properly destroy skb(s) that have wrong UDP
checksum.

Thanks again to syzkaller team.

Fixes : 7c13f97ffde6 ("udp: do fwd memory scheduling on dequeue")
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa2-pdata'
David S. Miller [Tue, 7 Feb 2017 15:51:48 +0000 (10:51 -0500)]
Merge branch 'dsa2-pdata'

Florian Fainelli says:

====================
net: dsa: Support for pdata in dsa2

This is not exactly new, and was sent before, although back then, I did not
have an user of the pre-declared MDIO board information, but now we do. Note
that I have additional changes queued up to have b53 register platform data for
MIPS bcm47xx and bcm63xx.

Yes I know that we should have the Orion platforms eventually be converted to
Device Tree, but until that happens, I don't want any remaining users of the
old "dsa" platform device (hence the previous DTS submissions for ARM/mvebu)
and, there will be platforms out there that most likely won't never see DT
coming their way (BCM47xx is almost 100% sure, BCM63xx maybe not in a distant
future).

We would probably want the whole series to be merged via David Miller's tree
to simplify things.

Thanks!

Changes in v5:

- dropped changes to drivers/base/ because after more than a month, we cannot
  get any answer from Greg KH

Changes in v4:

- Changed device_find_class() to device_find_in_class_name()
- Added kerneldoc above device_find_in_class_name() to explain what it does
  and the calling convention regarding device reference counts
- Changed dev_to_net_device to device_to_net_device() added comments
  about what it does and the caller conventions regarding reference counts

Changes in v3:

- Tested EPROBE_DEFER from a mockup MDIO/DSA switch driver and everything
  is fine, once the driver finally probes we have access to platform data
  as expected

- added comment above dsa_port_is_valid() that port->name is mandatory
  for platform data cases

- added an extra check in dsa_parse_member() for a NULL pdata pointer

- fixed a bunch of checkpatch errors and warnings

Changes in v2:

- Rebased against latest net-next/master

- Moved dev_find_class() to device_find_class() into drivers/base/core.c

- Moved dev_to_net_device into net/core/dev.c

- Utilize dsa_chip_data directly instead of dsa_platform_data

- Augmented dsa_chip_data to be multi-CPU port ready

Changes from last submission (few months back):

- rebased against latest net-next

- do not introduce dsa2_platform_data which was overkill and was meant to
  allow us to do exaclty the same things with platform data and Device Tree
  we use the existing dsa_platform_data instead

- properly register MDIO devices when the MDIO bus is registered and associate
  platform_data with them

- add a change to the Orion platform code to demonstrate how this can be used

Thank you
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoARM: orion: Register DSA switch as a MDIO device
Florian Fainelli [Sat, 4 Feb 2017 21:02:45 +0000 (13:02 -0800)]
ARM: orion: Register DSA switch as a MDIO device

Utilize the ability to pass board specific MDIO bus information towards a
particular MDIO device thus allowing us to provide the per-port switch layout
to the Marvell 88E6XXX switch driver.

Since we would end-up with conflicting registration paths, do not register the
"dsa" platform device anymore.

Note that the MDIO devices registered by code in net/dsa/dsa2.c does not
parse a dsa_platform_data, but directly take a dsa_chip_data (specific
to a single switch chip), so we update the different call sites to pass
this structure down to orion_ge00_switch_init().

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: Allow pre-declaration of MDIO devices
Florian Fainelli [Sat, 4 Feb 2017 21:02:44 +0000 (13:02 -0800)]
net: phy: Allow pre-declaration of MDIO devices

Allow board support code to collect pre-declarations for MDIO devices by
registering them with mdiobus_register_board_info(). SPI and I2C buses
have a similar feature, we were missing this for MDIO devices, but this
is particularly useful for e.g: MDIO-connected switches which need to
provide their port layout (often board-specific) to a MDIO Ethernet
switch driver.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Add support for platform data
Florian Fainelli [Sat, 4 Feb 2017 21:02:43 +0000 (13:02 -0800)]
net: dsa: Add support for platform data

Allow drivers to use the new DSA API with platform data. Most of the
code in net/dsa/dsa2.c does not rely so much on device_nodes and can get
the same information from platform_data instead.

We purposely do not support distributed configurations with platform
data, so drivers should be providing a pointer to a 'struct
dsa_chip_data' structure if they wish to communicate per-port layout.

Multiple CPUs port could potentially be supported and dsa_chip_data is
extended to receive up to one reference to an upstream network device
per port described by a dsa_chip_data structure.

dsa_dev_to_net_device() increments the network device's reference count,
so we intentionally call dev_put() to be consistent with the DT-enabled
path, until we have a generic notifier based solution.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Rename and export dev_to_net_device()
Florian Fainelli [Sat, 4 Feb 2017 21:02:42 +0000 (13:02 -0800)]
net: dsa: Rename and export dev_to_net_device()

In preparation for using this function in net/dsa/dsa2.c, rename the function
to make its scope DSA specific, and export it.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: Refactor remaining port setup
Andrew Lunn [Sat, 4 Feb 2017 19:15:28 +0000 (20:15 +0100)]
net: dsa: mv88e6xxx: Refactor remaining port setup

Move the remaining port configuration code which varies per device
into port.c, using ops were necessary. This makes
mv88e6xxx_6185_family() and mv88e6xxx_6095_family() unused, so remove
them.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: Implement Clause 45 access to SMI devices
Andrew Lunn [Sat, 4 Feb 2017 19:12:24 +0000 (20:12 +0100)]
net: dsa: mv88e6xxx: Implement Clause 45 access to SMI devices

The mv88e6390 MDIO bus controllers can support for clause 45 accesses.
The internal SERDES interfaces need this, and it is likely external
10GHz PHYs will be clause 45.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mv88e6390-CMODE'
David S. Miller [Tue, 7 Feb 2017 15:34:43 +0000 (10:34 -0500)]
Merge branch 'mv88e6390-CMODE'

Andrew Lunn says:

====================
Set the CMODE for mv88e6390 ports

The mv88e6390 ports 9 & 10 allow there CMODE to be set. CMODE is part
of what linux defines as phy-mode. Add the needed phy-modes to linux,
and add code which will act upon the phy-mode property to configure
the switch port.

These patches have been posted before as part of a bigger patchset
which has now been broken up. I've added the received reviewed by
tags, and added device tree documentation.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: Set the CMODE for mv88e6390 ports 9 & 10
Andrew Lunn [Sat, 4 Feb 2017 19:02:50 +0000 (20:02 +0100)]
net: dsa: mv88e6xxx: Set the CMODE for mv88e6390 ports 9 & 10

Unlike most ports, ports 9 and 10 of the 6390X family have configurable
PHY modes. Set the mode as part of adjust_link().

Ordering is important, because the SERDES interfaces connected to
ports 9 and 10 can be split and assigned to other ports. The CMODE has
to be correctly set before the SERDES interface on another port can be
configured. Such configuration is likely to be performed in
port_enable() and port_disabled(), called on slave_open() and
slave_close().

The simple case is port 9 and 10 are used for 'CPU' or 'DSA'. In this
case, the CMODE is set via a phy-mode in dsa_cpu_dsa_setup(), which is
called early in the switch setup.

When ports 9 or 10 are used as user ports, and have a fixed-phy, when
the fixed fixed-phy is attached, dsa_slave_adjust_link() is called,
which results in the adjust_link function being called, setting the
cmode. The port_enable() will for other ports will be called much
later.

When ports 9 or 10 are used as user ports and have a real phy attached
which does not use all the available SERDES interface, e.g. a 1Gbps
SGMII, there is currently no mechanism in place to set the CMODE of
the port from software. It must be hoped the stripping resistors are
correct.

At the same time, add a function to get the cmode. This will be needed
when configuring the SERDES interfaces.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: Add 2000base-x, 2500base-x and rxaui modes
Andrew Lunn [Sat, 4 Feb 2017 19:02:49 +0000 (20:02 +0100)]
net: phy: Add 2000base-x, 2500base-x and rxaui modes

The mv88e6390 ports 9 and 10 supports some additional PHY modes. Add
these modes to the PHY core so they can be used in the binding.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'net-Fix-on-stack-USB-buffers'
David S. Miller [Tue, 7 Feb 2017 15:07:03 +0000 (10:07 -0500)]
Merge branch 'net-Fix-on-stack-USB-buffers'

Ben Hutchings says:

====================
net: Fix on-stack USB buffers

Allocating USB buffers on the stack is not portable, and no longer
works on x86_64 (with VMAP_STACK enabled as per default).  This
series fixes all the instances I could find where USB networking
drivers do that.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocatc: Use heap buffer for memory size test
Ben Hutchings [Sat, 4 Feb 2017 16:57:04 +0000 (16:57 +0000)]
catc: Use heap buffer for memory size test

Allocating USB buffers on the stack is not portable, and no longer
works on x86_64 (with VMAP_STACK enabled as per default).

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocatc: Combine failure cleanup code in catc_probe()
Ben Hutchings [Sat, 4 Feb 2017 16:56:56 +0000 (16:56 +0000)]
catc: Combine failure cleanup code in catc_probe()

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agortl8150: Use heap buffers for all register access
Ben Hutchings [Sat, 4 Feb 2017 16:56:32 +0000 (16:56 +0000)]
rtl8150: Use heap buffers for all register access

Allocating USB buffers on the stack is not portable, and no longer
works on x86_64 (with VMAP_STACK enabled as per default).

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agopegasus: Use heap buffers for all register access
Ben Hutchings [Sat, 4 Feb 2017 16:56:03 +0000 (16:56 +0000)]
pegasus: Use heap buffers for all register access

Allocating USB buffers on the stack is not portable, and no longer
works on x86_64 (with VMAP_STACK enabled as per default).

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
References: https://bugs.debian.org/852556
Reported-by: Lisandro Damián Nicanor Pérez Meyer <lisandro@debian.org>
Tested-by: Lisandro Damián Nicanor Pérez Meyer <lisandro@debian.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'virtio_net-XDP-adjust_head'
David S. Miller [Tue, 7 Feb 2017 15:05:13 +0000 (10:05 -0500)]
Merge branch 'virtio_net-XDP-adjust_head'

John Fastabend says:

====================
XDP adjust head support for virtio

This series adds adjust head support for virtio. The following is my
test setup. I use qemu + virtio as follows,

./x86_64-softmmu/qemu-system-x86_64 \
  -hda /var/lib/libvirt/images/Fedora-test0.img \
  -m 4096  -enable-kvm -smp 2 -netdev tap,id=hn0,queues=4,vhost=on \
  -device virtio-net-pci,netdev=hn0,mq=on,guest_tso4=off,guest_tso6=off,guest_ecn=off,guest_ufo=off,vectors=9

In order to use XDP with virtio until LRO is supported TSO must be
turned off in the host. The important fields in the above command line
are the following,

  guest_tso4=off,guest_tso6=off,guest_ecn=off,guest_ufo=off

Also note it is possible to conusme more queues than can be supported
because when XDP is enabled for retransmit XDP attempts to use a queue
per cpu. My standard queue count is 'queues=4'.

After loading the VM I run the relevant XDP test programs in,

  ./sammples/bpf

For this series I tested xdp1, xdp2, and xdp_tx_iptunnel. I usually test
with iperf (-d option to get bidirectional traffic), ping, and pktgen.
I also have a modified xdp1 that returns XDP_PASS on any packet to ensure
the normal traffic path to the stack continues to work with XDP loaded.

It would be great to automate this soon. At the moment I do it by hand
which is starting to get tedious.

v2: original series dropped trace points after merge.
====================

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: XDP support for adjust_head
John Fastabend [Fri, 3 Feb 2017 03:16:29 +0000 (19:16 -0800)]
virtio_net: XDP support for adjust_head

Add support for XDP adjust head by allocating a 256B header region
that XDP programs can grow into. This is only enabled when a XDP
program is loaded.

In order to ensure that we do not have to unwind queue headroom push
queue setup below bpf_prog_add. It reads better to do a prog ref
unwind vs another queue setup call.

At the moment this code must do a full reset to ensure old buffers
without headroom on program add or with headroom on program removal
are not used incorrectly in the datapath. Ideally we would only
have to disable/enable the RX queues being updated but there is no
API to do this at the moment in virtio so use the big hammer. In
practice it is likely not that big of a problem as this will only
happen when XDP is enabled/disabled changing programs does not
require the reset. There is some risk that the driver may either
have an allocation failure or for some reason fail to correctly
negotiate with the underlying backend in this case the driver will
be left uninitialized. I have not seen this ever happen on my test
systems and for what its worth this same failure case can occur
from probe and other contexts in virtio framework.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: refactor freeze/restore logic into virtnet reset logic
John Fastabend [Fri, 3 Feb 2017 03:16:01 +0000 (19:16 -0800)]
virtio_net: refactor freeze/restore logic into virtnet reset logic

For XDP we will need to reset the queues to allow for buffer headroom
to be configured. In order to do this we need to essentially run the
freeze()/restore() code path. Unfortunately the locking requirements
between the freeze/restore and reset paths are different however so
we can not simply reuse the code.

This patch refactors the code path and adds a reset helper routine.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: remove duplicate queue pair binding in XDP
John Fastabend [Fri, 3 Feb 2017 03:15:32 +0000 (19:15 -0800)]
virtio_net: remove duplicate queue pair binding in XDP

Factor out qp assignment.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: factor out xdp handler for readability
John Fastabend [Fri, 3 Feb 2017 03:15:01 +0000 (19:15 -0800)]
virtio_net: factor out xdp handler for readability

At this point the do_xdp_prog is mostly if/else branches handling
the different modes of virtio_net. So remove it and handle running
the program in the per mode handlers.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: wrap rtnl_lock in test for calling with lock already held
John Fastabend [Fri, 3 Feb 2017 03:14:32 +0000 (19:14 -0800)]
virtio_net: wrap rtnl_lock in test for calling with lock already held

For XDP use case and to allow ethtool reset tests it is useful to be
able to use reset paths from contexts where rtnl lock is already
held.

This requries updating virtnet_set_queues and free_receive_bufs the
two places where rtnl_lock is taken in virtio_net. To do this we
use the following pattern,

_foo(...) { do stuff }
foo(...) { rtnl_lock(); _foo(...); rtnl_unlock()};

this allows us to use freeze()/restore() flow from both contexts.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bridge-improve-cache-utilization'
David S. Miller [Tue, 7 Feb 2017 03:53:14 +0000 (22:53 -0500)]
Merge branch 'bridge-improve-cache-utilization'

Nikolay Aleksandrov says:

====================
bridge: improve cache utilization

This is the first set which begins to deal with the bad bridge cache
access patterns. The first patch rearranges the bridge and port structs
a little so the frequently (and closely) accessed members are in the same
cache line. The second patch then moves the garbage collection to a
workqueue trying to improve system responsiveness under load (many fdbs)
and more importantly removes the need to check if the matched entry is
expired in __br_fdb_get which was a major source of false-sharing.
The third patch is a preparation for the final one which
If properly configured, i.e. ports bound to CPUs (thus updating "updated"
locally) then the bridge's HitM goes from 100% to 0%, but even without
binding we get a win because previously every lookup that iterated over
the hash chain caused false-sharing due to the first cache line being
used for both mac/vid and used/updated fields.

Some results from tests I've run:
(note that these were run in good conditions for the baseline, everything
 ran on a single NUMA node and there were only 3 fdbs)

1. baseline
100% Load HitM on the fdbs (between everyone who has done lookups and hit
                            one of the 3 hash chains of the communicating
                            src/dst fdbs)
Overall 5.06% Load HitM for the bridge, first place in the list

2. patched & ports bound to CPUs
0% Local load HitM, bridge is not even in the c2c report list
Also there's 3% consistent improvement in netperf tests.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: fdb: write to used and updated at most once per jiffy
Nikolay Aleksandrov [Sat, 4 Feb 2017 17:05:09 +0000 (18:05 +0100)]
bridge: fdb: write to used and updated at most once per jiffy

Writing once per jiffy is enough to limit the bridge's false sharing.
After this change the bridge doesn't show up in the local load HitM stats.

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: move write-heavy fdb members in their own cache line
Nikolay Aleksandrov [Sat, 4 Feb 2017 17:05:08 +0000 (18:05 +0100)]
bridge: move write-heavy fdb members in their own cache line

Fdb's used and updated fields are written to on every packet forward and
packet receive respectively. Thus if we are receiving packets from a
particular fdb, they'll cause false-sharing with everyone who has looked
it up (even if it didn't match, since mac/vid share cache line!). The
"used" field is even worse since it is updated on every packet forward
to that fdb, thus the standard config where X ports use a single gateway
results in 100% fdb false-sharing. Note that this patch does not prevent
the last scenario, but it makes it better for other bridge participants
which are not using that fdb (and are only doing lookups over it).
The point is with this move we make sure that only communicating parties
get the false-sharing, in a later patch we'll show how to avoid that too.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: move to workqueue gc
Nikolay Aleksandrov [Sat, 4 Feb 2017 17:05:07 +0000 (18:05 +0100)]
bridge: move to workqueue gc

Move the fdb garbage collector to a workqueue which fires at least 10
milliseconds apart and cleans chain by chain allowing for other tasks
to run in the meantime. When having thousands of fdbs the system is much
more responsive. Most importantly remove the need to check if the
matched entry has expired in __br_fdb_get that causes false-sharing and
is completely unnecessary if we cleanup entries, at worst we'll get 10ms
of traffic for that entry before it gets deleted.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobridge: modify bridge and port to have often accessed fields in one cache line
Nikolay Aleksandrov [Sat, 4 Feb 2017 17:05:06 +0000 (18:05 +0100)]
bridge: modify bridge and port to have often accessed fields in one cache line

Move around net_bridge so the vlan fields are in the beginning since
they're checked on every packet even if vlan filtering is disabled.
For the port move flags & vlan group to the beginning, so they're in the
same cache line with the port's state (both flags and state are checked
on each packet).

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: enable verifier to add 0 to packet ptr
William Tu [Sat, 4 Feb 2017 16:37:29 +0000 (08:37 -0800)]
bpf: enable verifier to add 0 to packet ptr

The patch fixes the case when adding a zero value to the packet
pointer.  The zero value could come from src_reg equals type
BPF_K or CONST_IMM.  The patch fixes both, otherwise the verifer
reports the following error:
  [...]
    R0=imm0,min_value=0,max_value=0
    R1=pkt(id=0,off=0,r=4)
    R2=pkt_end R3=fp-12
    R4=imm4,min_value=4,max_value=4
    R5=pkt(id=0,off=4,r=4)
  269: (bf) r2 = r0     // r2 becomes imm0
  270: (77) r2 >>= 3
  271: (bf) r4 = r1     // r4 becomes pkt ptr
  272: (0f) r4 += r2    // r4 += 0
  addition of negative constant to packet pointer is not allowed

Signed-off-by: William Tu <u9012063@gmail.com>
Signed-off-by: Mihai Budiu <mbudiu@vmware.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'read-vnet_hdr_sz-once'
David S. Miller [Tue, 7 Feb 2017 03:41:27 +0000 (22:41 -0500)]
Merge branch 'read-vnet_hdr_sz-once'

Willem de Bruijn says:

====================
read vnet_hdr_sz once

Tuntap devices allow concurrent use and update of field vnet_hdr_sz.
Read the field once to avoid TOCTOU.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomacvtap: read vnet_hdr_size once
Willem de Bruijn [Fri, 3 Feb 2017 23:20:49 +0000 (18:20 -0500)]
macvtap: read vnet_hdr_size once

When IFF_VNET_HDR is enabled, a virtio_net header must precede data.
Data length is verified to be greater than or equal to expected header
length tun->vnet_hdr_sz before copying.

Macvtap functions read the value once, but unless READ_ONCE is used,
the compiler may ignore this and read multiple times. Enforce a single
read and locally cached value to avoid updates between test and use.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Suggested-by: Eric Dumazet <edumazet@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotun: read vnet_hdr_sz once
Willem de Bruijn [Fri, 3 Feb 2017 23:20:48 +0000 (18:20 -0500)]
tun: read vnet_hdr_sz once

When IFF_VNET_HDR is enabled, a virtio_net header must precede data.
Data length is verified to be greater than or equal to expected header
length tun->vnet_hdr_sz before copying.

Read this value once and cache locally, as it can be updated between
the test and use (TOCTOU).

Signed-off-by: Willem de Bruijn <willemb@google.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
CC: Eric Dumazet <edumazet@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>