openwrt/staging/blogic.git
9 years agoopenvswitch: Use Geneve device.
Pravin B Shelar [Thu, 27 Aug 2015 06:46:53 +0000 (23:46 -0700)]
openvswitch: Use Geneve device.

With help of tunnel metadata mode OVS can directly use
Geneve devices to implement Geneve tunnels.
This patch removes all of the OVS specific Geneve code
and make OVS use a Geneve net_device. Basic geneve vport
is still there to handle compatibility with current
userspace application.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Reviewed-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogeneve: Add support to collect tunnel metadata.
Pravin B Shelar [Thu, 27 Aug 2015 06:46:52 +0000 (23:46 -0700)]
geneve: Add support to collect tunnel metadata.

Following patch create new tunnel flag which enable
tunnel metadata collection on given device. These devices
can be used by tunnel metadata based routing or by OVS.
Geneve Consolidation patch get rid of collect_md_tun to
simplify tunnel lookup further.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Reviewed-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogeneve: Make dst-port configurable.
Pravin B Shelar [Thu, 27 Aug 2015 06:46:51 +0000 (23:46 -0700)]
geneve: Make dst-port configurable.

Add netlink interface to configure Geneve UDP port number.
So that user can configure it for a Gevene device.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Reviewed-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotunnel: introduce udp_tun_rx_dst()
Pravin B Shelar [Thu, 27 Aug 2015 06:46:50 +0000 (23:46 -0700)]
tunnel: introduce udp_tun_rx_dst()

Introduce function udp_tun_rx_dst() to initialize tunnel dst on
receive path.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Reviewed-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogeneve: Use skb mark and protocol to lookup route.
Pravin B Shelar [Thu, 27 Aug 2015 06:46:49 +0000 (23:46 -0700)]
geneve: Use skb mark and protocol to lookup route.

On packet transmit path geneve need to lookup route. Following
patch improves route lookup using more parameters.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Reviewed-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogeneve: Initialize ethernet address in device setup.
Pravin B Shelar [Thu, 27 Aug 2015 06:46:48 +0000 (23:46 -0700)]
geneve: Initialize ethernet address in device setup.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Reviewed-by: Jesse Gross <jesse@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobridge: Add netlink support for vlan_protocol attribute
Toshiaki Makita [Thu, 27 Aug 2015 06:32:26 +0000 (15:32 +0900)]
bridge: Add netlink support for vlan_protocol attribute

This enables bridge vlan_protocol to be configured through netlink.

When CONFIG_BRIDGE_VLAN_FILTERING is disabled, kernel behaves the
same way as this feature is not implemented.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobnx2x: Add new device ids under the Qlogic vendor
Yuval Mintz [Thu, 27 Aug 2015 05:03:08 +0000 (08:03 +0300)]
bnx2x: Add new device ids under the Qlogic vendor

This adds support for 3 new PCI device combinations -
1077:16a1, 1077:16a4 and 1077:16ad.

Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agosmsc911x: Ignore error return from device_get_phy_mode()
Guenter Roeck [Thu, 27 Aug 2015 03:27:05 +0000 (20:27 -0700)]
smsc911x: Ignore error return from device_get_phy_mode()

Commit 62ee783bf1f8 ("smsc911x: Fix crash seen if neither ACPI nor OF is
configured or used") introduces an error check for the return value from
device_get_phy_mode() and bails out if there is an error. Unfortunately,
there are configurations where no phy is configured. Those configurations
now fail.

To fix the problem, accept error returns from device_get_phy_mode(),
and use the return value from device_property_read_u32() to determine
if there is a suitable firmware interface to read the configuration.

Fixes: 62ee783bf1f8 ("smsc911x: Fix crash seen if neither ACPI nor OF is configured or used")
Tested-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodevice property: Return -ENXIO if there is no suitable FW interface
Guenter Roeck [Thu, 27 Aug 2015 03:27:04 +0000 (20:27 -0700)]
device property: Return -ENXIO if there is no suitable FW interface

Return -ENXIO if device property array access functions don't find
a suitable firmware interface.

This lets drivers decide if they should use available platform data
instead.

Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Jeremy Linton <jeremy.linton@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: sched: consolidate tc_classify{,_compat}
Daniel Borkmann [Wed, 26 Aug 2015 21:00:06 +0000 (23:00 +0200)]
net: sched: consolidate tc_classify{,_compat}

For classifiers getting invoked via tc_classify(), we always need an
extra function call into tc_classify_compat(), as both are being
exported as symbols and tc_classify() itself doesn't do much except
handling of reclassifications when tp->classify() returned with
TC_ACT_RECLASSIFY.

CBQ and ATM are the only qdiscs that directly call into tc_classify_compat(),
all others use tc_classify(). When tc actions are being configured
out in the kernel, tc_classify() effectively does nothing besides
delegating.

We could spare this layer and consolidate both functions. pktgen on
single CPU constantly pushing skbs directly into the netif_receive_skb()
path with a dummy classifier on ingress qdisc attached, improves
slightly from 22.3Mpps to 23.1Mpps.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Thu, 27 Aug 2015 21:06:09 +0000 (14:06 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-08-26

This series contains updates to i40e and i40evf only.

Anjali provides a fix for i40e where the part is not receiving multicast
or VLAN tagged packets when in promiscuous mode.  This can occur when a
software bridge is created on top of the device.  Fixed the legacy and MSI
interrupt mode in the driver, which was non-existent before since we
were assuming MSIX was the only mode that the driver ran in.  Fixed the
i40evf driver, where the wrong defines were getting used in the VF
driver.

Mitch fixes a sparse warning about comparing __le16 to u16 so use
le16_to_cpu() to resolve the warning.  Also fixed a dyslexic spelling
of invalid.

Shannon adds port.crc_errors to receive CRC error counter, since it
is a receive counter.

Catherine provides a fix to move the stopping of the service task and
flow director to i40e_shutdown() instead of i40e_suspend().

Greg fixes the ethtool offline diagnostic with netqueues, which just need
to be treated the same as virtual functions when someone wants to run the
ethtool offline diagnostic test.  Also fixed up code comments for the
i40e ethtool diagnostic test function.  Cleans up redundant and unneeded
messages, since the kernel notifies all VXLAN capable registered drivers,
so no need to log this.

Neerav adds the ability to update statistics per VEB per traffic class
and dump it via ethtool.

Jingjing adds support for virtual channel offload to support receive
polling mode in the VF driver.

v2: dropped patch which added helper functions into a header, feedback from
    David Miller was to make the functions constant to reduce the driver
    footprint, so remove the patch while Anjali works on making the requested
    changes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agosmsc9194: Remove uncompilable #if 0'd use of pr_dbg
Joe Perches [Wed, 26 Aug 2015 18:49:35 +0000 (11:49 -0700)]
smsc9194: Remove uncompilable #if 0'd use of pr_dbg

No pr_dbg method exists.

While this code is #if 0'd, it'd be nicer to
use the generic hex_dump, so use it instead.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'xgene-tso'
David S. Miller [Thu, 27 Aug 2015 20:51:38 +0000 (13:51 -0700)]
Merge branch 'xgene-tso'

Iyappan Subramanian says:

====================
drivers: net: xgene: Add TSO support

Adding TSO support for 10GbE

iperf Tx data rate without TSO: 3.42 Gbps
                      with TSO: 9.41 Gbps

v2: Address review comments from v1
    - skb_linearize() if headers doesn't fit in 3 hardware buffers

v1:
* Initial version
====================

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
9 years agodrivers: net: xgene: Adding support for TSO
Iyappan Subramanian [Wed, 26 Aug 2015 18:48:06 +0000 (11:48 -0700)]
drivers: net: xgene: Adding support for TSO

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodrivers: net: xgene: Preparatory patch for TSO support
Iyappan Subramanian [Wed, 26 Aug 2015 18:48:05 +0000 (11:48 -0700)]
drivers: net: xgene: Preparatory patch for TSO support

- Rearranged descriptor writes
- Moved increment command write to xgene_enet_setup_tx_desc

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'ovs-conntrack'
David S. Miller [Thu, 27 Aug 2015 18:40:44 +0000 (11:40 -0700)]
Merge branch 'ovs-conntrack'

Joe Stringer says:

====================
OVS conntrack support

The goal of this series is to allow OVS to send packets through the Linux
kernel connection tracker, and subsequently match on fields populated by
conntrack. This functionality is enabled through a new
CONFIG_OPENVSWITCH_CONNTRACK option.

This version addresses the feedback from v5, primarily checking the behaviour
is correct with different configurations such as disabling
CONFIG_OPENVSWITCH_CONNTRACK or disabling individual conntrack features like
connlabels.

The branch below has been updated with the corresponding userspace pieces:
https://github.com/joestringer/ovs dev/ct_20150818
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Allow attaching helpers to ct action
Joe Stringer [Wed, 26 Aug 2015 18:31:53 +0000 (11:31 -0700)]
openvswitch: Allow attaching helpers to ct action

Add support for using conntrack helpers to assist protocol detection.
The new OVS_CT_ATTR_HELPER attribute of the CT action specifies a helper
to be used for this connection. If no helper is specified, then helpers
will be automatically applied as per the sysctl configuration of
net.netfilter.nf_conntrack_helper.

The helper may be specified as part of the conntrack action, eg:
ct(helper=ftp). Initial packets for related connections should be
committed to allow later packets for the flow to be considered
established.

Example ovs-ofctl flows allowing FTP connections from ports 1->2:
in_port=1,tcp,action=ct(helper=ftp,commit),2
in_port=2,tcp,ct_state=-trk,action=ct(recirc)
in_port=2,tcp,ct_state=+trk-new+est,action=1
in_port=2,tcp,ct_state=+trk+rel,action=1

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Allow matching on conntrack label
Joe Stringer [Wed, 26 Aug 2015 18:31:52 +0000 (11:31 -0700)]
openvswitch: Allow matching on conntrack label

Allow matching and setting the ct_label field. As with ct_mark, this is
populated by executing the CT action. The label field may be modified by
specifying a label and mask nested under the CT action. It is stored as
metadata attached to the connection. Label modification occurs after
lookup, and will only persist when the conntrack entry is committed by
providing the COMMIT flag to the CT action. Labels are currently fixed
to 128 bits in size.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonetfilter: connlabels: Export setting connlabel length
Joe Stringer [Wed, 26 Aug 2015 18:31:51 +0000 (11:31 -0700)]
netfilter: connlabels: Export setting connlabel length

Add functions to change connlabel length into nf_conntrack_labels.c so
they may be reused by other modules like OVS and nftables without
needing to jump through xt_match_check() hoops.

Suggested-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Florian Westphal <fw@strlen.de>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonetfilter: Always export nf_connlabels_replace()
Joe Stringer [Wed, 26 Aug 2015 18:31:50 +0000 (11:31 -0700)]
netfilter: Always export nf_connlabels_replace()

The following patches will reuse this code from OVS.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Allow matching on conntrack mark
Joe Stringer [Wed, 26 Aug 2015 18:31:49 +0000 (11:31 -0700)]
openvswitch: Allow matching on conntrack mark

Allow matching and setting the ct_mark field. As with ct_state and
ct_zone, these fields are populated when the CT action is executed. To
write to this field, a value and mask can be specified as a nested
attribute under the CT action. This data is stored with the conntrack
entry, and is executed after the lookup occurs for the CT action. The
conntrack entry itself must be committed using the COMMIT flag in the CT
action flags for this change to persist.

Signed-off-by: Justin Pettit <jpettit@nicira.com>
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Add conntrack action
Joe Stringer [Wed, 26 Aug 2015 18:31:48 +0000 (11:31 -0700)]
openvswitch: Add conntrack action

Expose the kernel connection tracker via OVS. Userspace components can
make use of the CT action to populate the connection state (ct_state)
field for a flow. This state can be subsequently matched.

Exposed connection states are OVS_CS_F_*:
- NEW (0x01) - Beginning of a new connection.
- ESTABLISHED (0x02) - Part of an existing connection.
- RELATED (0x04) - Related to an established connection.
- INVALID (0x20) - Could not track the connection for this packet.
- REPLY_DIR (0x40) - This packet is in the reply direction for the flow.
- TRACKED (0x80) - This packet has been sent through conntrack.

When the CT action is executed by itself, it will send the packet
through the connection tracker and populate the ct_state field with one
or more of the connection state flags above. The CT action will always
set the TRACKED bit.

When the COMMIT flag is passed to the conntrack action, this specifies
that information about the connection should be stored. This allows
subsequent packets for the same (or related) connections to be
correlated with this connection. Sending subsequent packets for the
connection through conntrack allows the connection tracker to consider
the packets as ESTABLISHED, RELATED, and/or REPLY_DIR.

The CT action may optionally take a zone to track the flow within. This
allows connections with the same 5-tuple to be kept logically separate
from connections in other zones. If the zone is specified, then the
"ct_zone" match field will be subsequently populated with the zone id.

IP fragments are handled by transparently assembling them as part of the
CT action. The maximum received unit (MRU) size is tracked so that
refragmentation can occur during output.

IP frag handling contributed by Andy Zhou.

Based on original design by Justin Pettit.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Justin Pettit <jpettit@nicira.com>
Signed-off-by: Andy Zhou <azhou@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agodst: Add __skb_dst_copy() variation
Joe Stringer [Wed, 26 Aug 2015 18:31:47 +0000 (11:31 -0700)]
dst: Add __skb_dst_copy() variation

This variation on skb_dst_copy() doesn't require two skbs.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoipv6: Export nf_ct_frag6_gather()
Joe Stringer [Wed, 26 Aug 2015 18:31:46 +0000 (11:31 -0700)]
ipv6: Export nf_ct_frag6_gather()

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Move MASKED* macros to datapath.h
Joe Stringer [Wed, 26 Aug 2015 18:31:45 +0000 (11:31 -0700)]
openvswitch: Move MASKED* macros to datapath.h

This will allow the ovs-conntrack code to reuse these macros.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoopenvswitch: Serialize acts with original netlink len
Joe Stringer [Wed, 26 Aug 2015 18:31:44 +0000 (11:31 -0700)]
openvswitch: Serialize acts with original netlink len

Previously, we used the kernel-internal netlink actions length to
calculate the size of messages to serialize back to userspace.
However,the sw_flow_actions may not be formatted exactly the same as the
actions on the wire, so store the original actions length when
de-serializing and re-use the original length when serializing.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobgmac: support up to 3 cores (devices) on a bus
Rafał Miłecki [Wed, 26 Aug 2015 15:53:45 +0000 (17:53 +0200)]
bgmac: support up to 3 cores (devices) on a bus

Broadcom buses may have more than 1 Ethernet device. This is used e.g.
to have few interfaces connected to different switch ports. So far we
saw chipsets with only 2 devices (e.g. BCM4706) but recent ones have
up to 3 (e.g. Netgear R8000 uses 3rd interface for most of switch
traffic, lower interfaces are for some kind of offloading).

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'wireless-drivers-next-for-davem-2015-08-26' of git://git.kernel.org/pub...
David S. Miller [Thu, 27 Aug 2015 18:25:56 +0000 (11:25 -0700)]
Merge tag 'wireless-drivers-next-for-davem-2015-08-26' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
Major changes:

iwlwifi:

* new Tx power firmware API
* bump max firmware API to 17
* fix bug in debug prints
* static checker fix
* fix unused defines
* fix command list on newest firmware

brcmfmac:

* support NVRAM loading for bcm47xx platform
* new debugfs entry for msgbuf protocol layer used with PCIe devices

ath10k:

* add spectral scan support for qca99x0
* add qca6164 support
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobpf: fix bpf_skb_set_tunnel_key() helper
Alexei Starovoitov [Wed, 26 Aug 2015 22:57:38 +0000 (15:57 -0700)]
bpf: fix bpf_skb_set_tunnel_key() helper

Make sure to indicate to tunnel driver that key.tun_id is set,
otherwise gre won't recognize the metadata.

Fixes: d3aa45ce6b94 ("bpf: add helpers to access tunnel metadata")
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoi40e/i40evf: Bump i40e to 1.3.9 and i40evf to 1.3.5
Catherine Sullivan [Fri, 10 Jul 2015 23:36:10 +0000 (19:36 -0400)]
i40e/i40evf: Bump i40e to 1.3.9 and i40evf to 1.3.5

Bump version and update the copyright year for i40evf.

Change-ID: Iddb81b9dba09f0dc57ab54937b5821ecdd721ff6
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Cache the CEE TLV status returned from firmware
Neerav Parikh [Fri, 10 Jul 2015 23:36:09 +0000 (19:36 -0400)]
i40e/i40evf: Cache the CEE TLV status returned from firmware

Store the CEE TLV status returned by firmware to allow drivers to dump that
for debug purposes.

Change-ID: Ie3c4cf8cebabee4f15e1e3fdc4fc8a68bbca40ee
Signed-off-by: Neerav Parikh <neerav.parikh@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: add VIRTCHNL_VF_OFFLOAD flag
Anjali Singhai Jain [Fri, 10 Jul 2015 23:36:08 +0000 (19:36 -0400)]
i40e/i40evf: add VIRTCHNL_VF_OFFLOAD flag

Add virtual channel offload capability to support RX polling mode in the
VF.

Change-ID: Ib643ae2a7506dfc75fc489fc207493fabefa4832
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Remove redundant and unneeded messages
Greg Rose [Fri, 10 Jul 2015 23:36:07 +0000 (19:36 -0400)]
i40e: Remove redundant and unneeded messages

The kernel notifies all VXLAN capable registered drivers, i.e. any
driver that implements ndo_add_vxlan_port(), of the addition of a
port so that the driver can track which ports are in use.  There's
no need to log this - it just fills the system log with useless and
irksome noise.

Also, when failing to init SR-IOV interfaces the driver was printing the
same message twice. Just remove the inner printk and let the outer message
catch enable as well as the other failures.

Change-ID: Id5ecb1d425c2a357ee2bc1635dab24553831dade
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40evf: Remove PF specific register definitions from the VF
Anjali Singhai Jain [Fri, 10 Jul 2015 23:36:06 +0000 (19:36 -0400)]
i40evf: Remove PF specific register definitions from the VF

There were quite a few issues when the wrong defines were getting used
in the VF driver. This patch fixes the code where PF driver registers
were getting used for VF driver, and also removes the registers that are
not being used from the VF register file.

Change-ID: If116a9730112950d006eb8ec763998fc914cc839
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Acked-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40evf: Use the correct defines to match the VF registers
Anjali Singhai Jain [Fri, 10 Jul 2015 23:36:05 +0000 (19:36 -0400)]
i40evf: Use the correct defines to match the VF registers

Use CTLN1 instead of CTLN for the VF relative register space.

Change-ID: Iefba63faf0307af55fec8dbb64f26059f7d91318
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: correct spelling error
Mitch Williams [Fri, 10 Jul 2015 23:36:04 +0000 (19:36 -0400)]
i40e: correct spelling error

Turns out that 'inavlid' is an inavlid spelling for 'invalid'.

Change-ID: Ie1fe2d0f8d1ba75ab880594875ec2e4152a76f61
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Fix comment for ethtool diagnostic link test
Greg Rose [Fri, 10 Jul 2015 23:36:03 +0000 (19:36 -0400)]
i40e: Fix comment for ethtool diagnostic link test

The existing comment is incorrect.  Add new comment to point out that the
PF reset does not affect link but if the reset is changed to a different
type that does affect link then the link test would need to be moved to
before the reset.

Change-ID: I28d786f46e9465860babdee61c1dba51016464df
Reported-by: Jeremiah Kyle <jeremiah.kyle@intel.com>
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e/i40evf: Add capability to gather VEB per TC stats
Neerav Parikh [Fri, 10 Jul 2015 23:36:02 +0000 (19:36 -0400)]
i40e/i40evf: Add capability to gather VEB per TC stats

This patch adds capability to update per VEB per TC statistics and dump
it via ethtool. It also adds a structure to hold VEB per TC statistics.
The fields can be filled by reading the GLVEBTC_* counters.

Change-ID: I28b4759b9ab6ad5a61f046a1bc9ef6b16fe31538
Signed-off-by: Neerav Parikh <neerav.parikh@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Fix ethtool offline diagnostic with netqueues
Greg Rose [Fri, 10 Jul 2015 23:36:01 +0000 (19:36 -0400)]
i40e: Fix ethtool offline diagnostic with netqueues

Treat netqueues the same way we do virtual functions when someone wants to
run the ethtool offline diagnostic test.

Change-ID: Id48d2b933f1fd0db7be06305a93c6ebe3dc821f5
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Fix legacy interrupt mode in the driver
Anjali Singhai Jain [Fri, 10 Jul 2015 23:36:00 +0000 (19:36 -0400)]
i40e: Fix legacy interrupt mode in the driver

This patch fixes the driver flow to take into account legacy interrupts.
Over time we added code that assumes MSIX is the only mode that the
driver runs in. It also enables a legacy workaround to trigger SWINT
when the TX ring has non-cache aligned descriptors pending and interrupts
are disabled.

We work with a single vector in MSI mode too, so apply the same
restrictions as Legacy.

Change-ID: I826ddff1f9bd45d2dbe11f56a3ddcef0dbf42563
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: Move function calls to i40e_shutdown instead of i40e_suspend
Catherine Sullivan [Fri, 10 Jul 2015 23:35:59 +0000 (19:35 -0400)]
i40e: Move function calls to i40e_shutdown instead of i40e_suspend

We should be stopping the service task and flow director on
shutdown not on suspension.

Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: add RX to port CRC errors label
Shannon Nelson [Fri, 10 Jul 2015 23:35:57 +0000 (19:35 -0400)]
i40e: add RX to port CRC errors label

The port.crc_errors is really an RX counter, so let's mark it as such.

Change-ID: I179afd3f8a95d45229bb4163a6aeb01f0d2d250b
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoi40e: don't degrade __le16
Mitch Williams [Fri, 10 Jul 2015 23:35:56 +0000 (19:35 -0400)]
i40e: don't degrade __le16

Sparse cries when we compare an __le16 to a u16, almost like it cares
about architectures other than x86. Weird. Use the le16_to_cpu macro to
make it stop crying.

Change-ID: Id068f4d7868a2d3df234a791a76d15938f37db35
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
9 years agoMerge branch 'act_bpf_lockless'
David S. Miller [Wed, 26 Aug 2015 18:01:45 +0000 (11:01 -0700)]
Merge branch 'act_bpf_lockless'

Alexei Starovoitov says:

====================
act_bpf: remove spinlock in fast path

v1 version had a race condition in cleanup path of bpf_prog.
I tried to fix it by adding new callback 'cleanup_rcu' to 'struct tcf_common'
and call it out of act_api cleanup path, but Daniel noticed
(thanks for the idea!) that most of the classifiers already do action cleanup
out of rcu callback.
So instead this set of patches converts tcindex and rsvp classifiers to call
tcf_exts_destroy() after rcu grace period and since action cleanup logic
in __tcf_hash_release() is only called when bind and refcnt goes to zero,
it's guaranteed that cleanup() callback is called from rcu callback.
More specifically:
patches 1 and 2 - simple fixes
patches 2 and 3 - convert tcf_exts_destroy in tcindex and rsvp to call_rcu
patch 5 - removes spin_lock from act_bpf

The cleanup of actions is now universally done after rcu grace period
and in the future we can drop (now unnecessary) call_rcu from tcf_hash_destroy()
patch 5 is using synchronize_rcu() in act_bpf replacement path, since it's
very rare and alternative of dynamically allocating 'struct tcf_bpf_cfg' just
to pass it to call_rcu looks even less appealing.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_sched: act_bpf: remove spinlock in fast path
Alexei Starovoitov [Wed, 26 Aug 2015 03:06:35 +0000 (20:06 -0700)]
net_sched: act_bpf: remove spinlock in fast path

Similar to act_gact/act_mirred, act_bpf can be lockless in packet processing
with extra care taken to free bpf programs after rcu grace period.
Replacement of existing act_bpf (very rare) is done with synchronize_rcu()
and final destruction is done from tc_action_ops->cleanup() callback that is
called from tcf_exts_destroy()->tcf_action_destroy()->__tcf_hash_release() when
bind and refcnt reach zero which is only possible when classifier is destroyed.
Previous two patches fixed the last two classifiers (tcindex and rsvp) to
call tcf_exts_destroy() from rcu callback.

Similar to gact/mirred there is a race between prog->filter and
prog->tcf_action. Meaning that the program being replaced may use
previous default action if it happened to return TC_ACT_UNSPEC.
act_mirred race betwen tcf_action and tcfm_dev is similar.
In all cases the race is harmless.
Long term we may want to improve the situation by replacing the whole
tc_action->priv as single pointer instead of updating inner fields one by one.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_sched: convert rsvp to call tcf_exts_destroy from rcu callback
Alexei Starovoitov [Wed, 26 Aug 2015 03:06:34 +0000 (20:06 -0700)]
net_sched: convert rsvp to call tcf_exts_destroy from rcu callback

Adjust destroy path of cls_rsvp to call tcf_exts_destroy() after
rcu grace period.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_sched: convert tcindex to call tcf_exts_destroy from rcu callback
Alexei Starovoitov [Wed, 26 Aug 2015 03:06:33 +0000 (20:06 -0700)]
net_sched: convert tcindex to call tcf_exts_destroy from rcu callback

Adjust destroy path of cls_tcindex to call tcf_exts_destroy() after
rcu grace period.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_sched: act_bpf: remove unnecessary copy
Alexei Starovoitov [Wed, 26 Aug 2015 03:06:32 +0000 (20:06 -0700)]
net_sched: act_bpf: remove unnecessary copy

Fix harmless typo and avoid unnecessary copy of empty 'prog' into
unused 'strcut tcf_bpf_cfg old'.

Fixes: f4eaed28c783 ("act_bpf: fix memory leaks when replacing bpf programs")
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet_sched: make tcf_hash_destroy() static
Alexei Starovoitov [Wed, 26 Aug 2015 03:06:31 +0000 (20:06 -0700)]
net_sched: make tcf_hash_destroy() static

tcf_hash_destroy() used once. Make it static.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agolib/Makefile: remove CONFIG_AVERAGE build rule
Valentin Rothberg [Wed, 26 Aug 2015 13:36:12 +0000 (15:36 +0200)]
lib/Makefile: remove CONFIG_AVERAGE build rule

The Kconfig option AVERAGE and its implementation has been removed by
commit f4e774f55fe0 ("average: remove out-of-line implementation").
Remove the dead build rule in lib/Makefile.

Signed-off-by: Valentin Rothberg <valentinrothberg@gmail.com>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge ath-next from ath.git
Kalle Valo [Wed, 26 Aug 2015 09:40:23 +0000 (12:40 +0300)]
Merge ath-next from ath.git

Major changes in ath10k:

* add spectral scan support for qca99x0
* add qca6164 support

9 years agoath10k: fix compilation warnings in wmi phyerr pull function
Raja Mani [Fri, 21 Aug 2015 13:12:36 +0000 (18:42 +0530)]
ath10k: fix compilation warnings in wmi phyerr pull function

Below compilation warnings are observed in gcc version 4.8.2.
Even though it's not seen in bit older gcc versions (for ex, 4.7.3),
It's good to fix it by changing format specifier from %d to
%zd in wmi pull phyerr functions.

wmi.c: In function 'ath10k_wmi_op_pull_phyerr_ev':
wmi.c:3567:8: warning: format '%d' expects argument of type 'int',
              but argument 4 has type 'long unsigned int' [-Wformat=]
              left_len, sizeof(*phyerr));
                        ^
wmi.c: In function 'ath10k_wmi_10_4_op_pull_phyerr_ev':
wmi.c:3612:8: warning: format '%d' expects argument of type 'int',
      but argument 4 has type 'long unsigned int' [-Wformat=]
              left_len, sizeof(*phyerr));
                        ^
Fixes: 991adf71a6cd ("ath10k: refactor phyerr event handlers")
Fixes: 2b0a2e0d7c2f ("ath10k: handle 10.4 firmware phyerr event")
Signed-off-by: Raja Mani <rmani@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add qca6164 support
Michal Kazior [Thu, 13 Aug 2015 12:32:26 +0000 (14:32 +0200)]
ath10k: add qca6164 support

This adds additional 0x0041 PCI Device ID
definition to ath10k for QCA6164 which is a 1
spatial stream sibling of the QCA6174 (which is 2
spatial stream chip).

The QCA6164 needs a dedicated board.bin file which
is different than the one used for QCA6174. If the
board.bin is wrong the device will crash early
while trying to boot firmware. The register dump
will look like this:

 ath10k_pci 0000:02:00.0: firmware register dump:
 ath10k_pci 0000:02:00.0: [00]: 0x05010000 0x000015B3 0x000A012D 0x00955B31
 ...

Note the value 0x000A012D.

Special credit goes to Alan Liu
<alanliu@qca.qualcomm.com> for providing support
help which enabled me to come up with this patch.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add spectral scan support for 10.4 fw
Raja Mani [Thu, 20 Aug 2015 04:59:05 +0000 (10:29 +0530)]
ath10k: add spectral scan support for 10.4 fw

To enable/configure spectral scan parameters in 10.4 firmware, existing
wmi spectral related functions can be reused. Link those functions in
10.4 wmi ops table.

In addition, adjust bin size (only when size is 68 bytes) before reporting
bin samples to user space. The background for this adjustment is that
qca99x0 reports bin size as 68 bytes (64 bytes + 4 bytes) in report
mode 2. First 64 bytes carries in-band tones (-32 to +31) and last 4 byte
carries band edge detection data (+32) mainly used in radar detection
purpose. Additional last 4 bytes are stripped to make bin size valid one.

This bin size adjustment will happen only for qca99x0, all other chipsets
will report proper bin sizes (64/128) without extra 4 bytes being added
at the end. The changes are validated in qca99x0 using 10.4 firmware.

Signed-off-by: Raja Mani <rmani@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: fix dma_mapping_error() handling
Michal Kazior [Wed, 19 Aug 2015 11:10:43 +0000 (13:10 +0200)]
ath10k: fix dma_mapping_error() handling

The function returns 1 when DMA mapping fails. The
driver would return bogus values and could
possibly confuse itself if DMA failed.

Fixes: 767d34fc67af ("ath10k: remove DMA mapping wrappers")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoath10k: add missing mutex unlock on failpath
Michal Kazior [Wed, 19 Aug 2015 11:08:53 +0000 (13:08 +0200)]
ath10k: add missing mutex unlock on failpath

Kernel would complain about leaving a held lock
after going back to userspace and would
subsequently deadlock.

Fixes: e04cafbc38c7 ("ath10k: fix peer limit enforcement")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
9 years agoMerge branch 'dsa-docs'
David S. Miller [Wed, 26 Aug 2015 00:01:33 +0000 (17:01 -0700)]
Merge branch 'dsa-docs'

Florian Fainelli says:

====================
Documentation: dsa

This patch series adds some documentation about DSA as a subsystem as well
as the SF2 driver since it slightly diverges from your average DSA driver ;)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoDocumentation: networking: dsa: Add Broadcom SF2 document
Florian Fainelli [Tue, 25 Aug 2015 22:33:14 +0000 (15:33 -0700)]
Documentation: networking: dsa: Add Broadcom SF2 document

Add a document describing the Broadcom Starfigther 2 switch hardware,
its specifics, and how the driver is implemented and its specifics.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoDocumentation: networking: add a DSA document
Florian Fainelli [Tue, 25 Aug 2015 22:33:13 +0000 (15:33 -0700)]
Documentation: networking: add a DSA document

Describe how the DSA subsystem works, its design principles,
limitations, and describe in details how to implement a DSA switch
driver.

Acked-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'rds-more-fixes'
David S. Miller [Tue, 25 Aug 2015 23:28:11 +0000 (16:28 -0700)]
Merge branch 'rds-more-fixes'

Santosh Shilimkar says:

====================
RDS: Few more fixes

As indicated in the earlier series [1], this is a follow-up series which
addresses few issues around the RDS FMR code. With [1] and the subject
series, now I can run many parallel threads with multiple sockets with
N x N traffic. The stress tests has survived overnight runs.

[1] https://lkml.org/lkml/2015/8/22/127
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: remove superfluous from rds_ib_alloc_fmr()
santosh.shilimkar@oracle.com [Tue, 25 Aug 2015 19:02:03 +0000 (12:02 -0700)]
RDS: remove superfluous from rds_ib_alloc_fmr()

Memory allocated for 'ibmr' uses kzalloc_node() which already
initialises the memory to zero. There is no need to do
memset() 0 on that memory.

Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: flush the FMR pool less often
santosh.shilimkar@oracle.com [Tue, 25 Aug 2015 19:02:02 +0000 (12:02 -0700)]
RDS: flush the FMR pool less often

FMR flush is an expensive and time consuming operation. Reduce the
frequency of FMR pool flush by 50% so that more FMR work gets accumulated
for more efficient flushing.

Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: push FMR pool flush work to its own worker
santosh.shilimkar@oracle.com [Tue, 25 Aug 2015 19:02:01 +0000 (12:02 -0700)]
RDS: push FMR pool flush work to its own worker

RDS FMR flush operation and also it races with connect/reconect
which happes a lot with RDS. FMR flush being on common rds_wq aggrevates
the problem. Lets push RDS FMR pool flush work to its own worker.

Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: fix fmr pool dirty_count
Wengang Wang [Tue, 25 Aug 2015 19:02:00 +0000 (12:02 -0700)]
RDS: fix fmr pool dirty_count

In rds_ib_flush_mr_pool(), dirty_count accounts the clean ones
which is wrong. This can lead to a negative dirty count value.

Lets fix it.

Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: Fix rds MR reference count in rds_rdma_unuse()
santosh.shilimkar@oracle.com [Tue, 25 Aug 2015 19:01:59 +0000 (12:01 -0700)]
RDS: Fix rds MR reference count in rds_rdma_unuse()

rds_rdma_unuse() drops the mr reference count which it hasn't
taken. Correct way of removing mr is to remove mr from the tree
and then rdma_destroy_mr() it first, then rds_mr_put() to decrement
its reference count. Whichever thread holds last reference will free
the mr via rds_mr_put()

This bug was triggering weird null pointer crashes. One if the trace
for it is captured below.

BUG: unable to handle kernel NULL pointer dereference at
0000000000000104
IP: [<ffffffffa0899471>] rds_ib_free_mr+0x31/0x130 [rds_rdma]
PGD 4366fa067 PUD 4366f9067 PMD 0
Oops: 0000 [#1] SMP

[...]

task: ffff88046da6a000 ti: ffff88046da6c000 task.ti: ffff88046da6c000
RIP: 0010:[<ffffffffa0899471>]  [<ffffffffa0899471>]
rds_ib_free_mr+0x31/0x130 [rds_rdma]
RSP: 0018:ffff88046fa43bd8  EFLAGS: 00010286
RAX: 0000000071d38b80 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff880079e7ff40
RBP: ffff88046fa43bf8 R08: 0000000000000000 R09: 0000000000000000
R10: ffff88046fa43ca8 R11: ffff88046a802ed8 R12: ffff880079e7fa40
R13: 0000000000000000 R14: ffff880079e7ff40 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffff88046fa40000(0000)
knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
CR2: 0000000000000104 CR3: 00000004366fb000 CR4: 00000000000006e0
Stack:
 ffff880079e7fa40 ffff880671d38f08 ffff880079e7ff40 0000000000000296
 ffff88046fa43c28 ffffffffa087a38b ffff880079e7fa40 ffff880671d38f10
 0000000000000000 0000000000000292 ffff88046fa43c48 ffffffffa087a3b6
Call Trace:
 <IRQ>
 [<ffffffffa087a38b>] rds_destroy_mr+0x8b/0xa0 [rds]
 [<ffffffffa087a3b6>] __rds_put_mr_final+0x16/0x30 [rds]
 [<ffffffffa087a492>] rds_rdma_unuse+0xc2/0x120 [rds]
 [<ffffffffa08766d3>] rds_recv_incoming_exthdrs+0x83/0xa0 [rds]
 [<ffffffffa0876782>] rds_recv_incoming+0x92/0x200 [rds]
 [<ffffffffa0895269>] rds_ib_process_recv+0x259/0x320 [rds_rdma]
 [<ffffffffa08962a8>] rds_ib_recv_tasklet_fn+0x1a8/0x490 [rds_rdma]
 [<ffffffff810dcd78>] ? __remove_hrtimer+0x58/0x90
 [<ffffffff810799e1>] tasklet_action+0xb1/0xc0
 [<ffffffff81079b52>] __do_softirq+0xe2/0x290
 [<ffffffff81079df6>] irq_exit+0xa6/0xb0
 [<ffffffff81613915>] do_IRQ+0x65/0xf0
 [<ffffffff816118ab>] common_interrupt+0x6b/0x6b

Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: fix the dangling reference to rds_ib_incoming_slab
santosh.shilimkar@oracle.com [Tue, 25 Aug 2015 19:01:58 +0000 (12:01 -0700)]
RDS: fix the dangling reference to rds_ib_incoming_slab

On rds_ib_frag_slab allocation failure, ensure rds_ib_incoming_slab
is not pointing to the detsroyed memory.

Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge
David S. Miller [Tue, 25 Aug 2015 23:20:38 +0000 (16:20 -0700)]
Merge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge

Antonio Quartulli says:

====================
Included changes:
- code restyling and beautification
- use int kernel types instead of C99
- update kereldoc
- prevent potential hlist double deletion of VLAN objects
- fix gw bandwidth calculation
- convert list to hlist when needed
- add lockdep_asserts calls in function with lock requirements
  described in kerneldoc
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoenic: reduce ioread in devcmd2
Govindarajulu Varadarajan [Tue, 25 Aug 2015 08:45:11 +0000 (14:15 +0530)]
enic: reduce ioread in devcmd2

posted_index is RO in firmware. We need not do ioread everytime to get
posted index. Store posted index locally.

Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agor8169: Add values missing in @get_stats64 from HW counters
Corinna Vinschen [Mon, 24 Aug 2015 10:52:39 +0000 (12:52 +0200)]
r8169: Add values missing in @get_stats64 from HW counters

The r8169 driver collects statistical information returned by
@get_stats64 by counting them in the driver itself, even though many
(but not all) of the values are already collected by tally counters
(TCs) in the NIC.  Some of these TC values are not returned by
@get_stats64.  Especially the received multicast packages are missing
from /proc/net/dev.

Rectify this by fetching the TCs and returning them from
rtl8169_get_stats64.

The counters collected in the driver obviously disappear as soon as the
driver is unloaded so after a driver is loaded the counters always start
at 0. The TCs on the other hand are only reset by a power cycle.  Without
further considerations the values collected by the driver would not match
up against the TC values.

This patch introduces a new function rtl8169_reset_counters which
resets the TCs.  Also, since rtl8169_reset_counters shares most of
its code with rtl8169_update_counters, refactor the shared code into
two new functions  rtl8169_map_counters and rtl8169_unmap_counters.

Unfortunately chip versions prior to RTL_GIGA_MAC_VER_19 don't allow
to reset the TCs programatically.  Therefore introduce an addition to
the rtl8169_private struct and a function rtl8169_init_counter_offsets
to store the TCs at first rtl_open.  Use these values as offsets in
rtl8169_get_stats64.  Propagate a failure to reset *and* update the
counters up to rtl_open and emit a warning message, if so.

Signed-off-by: Corinna Vinschen <vinschen@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agords: Fix improper gfp_t usage.
David S. Miller [Tue, 25 Aug 2015 22:54:25 +0000 (15:54 -0700)]
rds: Fix improper gfp_t usage.

>> net/rds/ib_recv.c:382:28: sparse: incorrect type in initializer (different base types)
   net/rds/ib_recv.c:382:28:    expected int [signed] can_wait
   net/rds/ib_recv.c:382:28:    got restricted gfp_t
   net/rds/ib_recv.c:828:23: sparse: cast to restricted __le64

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMAINTAINERS: update vmxnet3 driver maintainer
Shrikrishna Khare [Mon, 24 Aug 2015 21:24:11 +0000 (14:24 -0700)]
MAINTAINERS: update vmxnet3 driver maintainer

Shreyas Bhatewara would no longer maintain the vmxnet3 driver. Taking over
the role of vmxnet3 maintainer.

Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
Signed off-by: Shreyas Bhatewara <sbhatewara@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agovxlan: fix multiple inclusion of vxlan.h
Jiri Benc [Tue, 25 Aug 2015 16:36:50 +0000 (18:36 +0200)]
vxlan: fix multiple inclusion of vxlan.h

The vxlan_get_sk_family inline function was added after the last #endif,
making multiple inclusion of net/vxlan.h fail. Move it to the proper place.

Reported-by: Mark Rustad <mark.d.rustad@intel.com>
Fixes: 705cc62f6728c ("vxlan: provide access function for vxlan socket address family")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMAINTAINERS: Add VRF entry
David Ahern [Tue, 25 Aug 2015 17:26:22 +0000 (10:26 -0700)]
MAINTAINERS: Add VRF entry

Add entry for new VRF device driver.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoroute: fix a use-after-free
WANG Cong [Tue, 25 Aug 2015 17:38:53 +0000 (10:38 -0700)]
route: fix a use-after-free

This patch fixes the following crash:

 general protection fault: 0000 [#1] SMP DEBUG_PAGEALLOC
 CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.2.0-rc7+ #166
 Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
 task: ffff88010656d280 ti: ffff880106570000 task.ti: ffff880106570000
 RIP: 0010:[<ffffffff8182f91b>]  [<ffffffff8182f91b>] dst_destroy+0xa6/0xef
 RSP: 0018:ffff880107603e38  EFLAGS: 00010202
 RAX: 0000000000000001 RBX: ffff8800d225a000 RCX: ffffffff82250fd0
 RDX: 0000000000000001 RSI: ffffffff82250fd0 RDI: 6b6b6b6b6b6b6b6b
 RBP: ffff880107603e58 R08: 0000000000000001 R09: 0000000000000001
 R10: 000000000000b530 R11: ffff880107609000 R12: 0000000000000000
 R13: ffffffff82343c40 R14: 0000000000000000 R15: ffffffff8182fb4f
 FS:  0000000000000000(0000) GS:ffff880107600000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
 CR2: 00007fcabd9d3000 CR3: 00000000d7279000 CR4: 00000000000006e0
 Stack:
  ffffffff82250fd0 ffff8801077d6f00 ffffffff82253c40 ffff8800d225a000
  ffff880107603e68 ffffffff8182fb5d ffff880107603f08 ffffffff810d795e
  ffffffff810d7648 ffff880106574000 ffff88010656d280 ffff88010656d280
 Call Trace:
  <IRQ>
  [<ffffffff8182fb5d>] dst_destroy_rcu+0xe/0x1d
  [<ffffffff810d795e>] rcu_process_callbacks+0x618/0x7eb
  [<ffffffff810d7648>] ? rcu_process_callbacks+0x302/0x7eb
  [<ffffffff8182fb4f>] ? dst_gc_task+0x1eb/0x1eb
  [<ffffffff8107e11b>] __do_softirq+0x178/0x39f
  [<ffffffff8107e52e>] irq_exit+0x41/0x95
  [<ffffffff81a4f215>] smp_apic_timer_interrupt+0x34/0x40
  [<ffffffff81a4d5cd>] apic_timer_interrupt+0x6d/0x80
  <EOI>
  [<ffffffff8100b968>] ? default_idle+0x21/0x32
  [<ffffffff8100b966>] ? default_idle+0x1f/0x32
  [<ffffffff8100bf19>] arch_cpu_idle+0xf/0x11
  [<ffffffff810b0bc7>] default_idle_call+0x1f/0x21
  [<ffffffff810b0dce>] cpu_startup_entry+0x1ad/0x273
  [<ffffffff8102fe67>] start_secondary+0x135/0x156

dst is freed right before lwtstate_put(), this is not correct...

Fixes: 61adedf3e3f1 ("route: move lwtunnel state to dst_entry")
Acked-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Cong Wang <cwang@twopensource.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet-next: Fix warning while make xmldocs caused by skbuff.c
Masanari Iida [Mon, 24 Aug 2015 13:56:54 +0000 (22:56 +0900)]
net-next: Fix warning while make xmldocs caused by skbuff.c

This patch fix following warnings.

.//net/core/skbuff.c:407: warning: No description found
for parameter 'len'
.//net/core/skbuff.c:407: warning: Excess function parameter
 'length' description in '__netdev_alloc_skb'
.//net/core/skbuff.c:476: warning: No description found
 for parameter 'len'
.//net/core/skbuff.c:476: warning: Excess function parameter
'length' description in '__napi_alloc_skb'

Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoppp: implement x-netns support
Guillaume Nault [Mon, 24 Aug 2015 09:35:30 +0000 (11:35 +0200)]
ppp: implement x-netns support

Let packets move from one netns to the other at PPP encapsulation and
decapsulation time.

PPP units and channels remain in the netns in which they were
originally created. Only the net_device may move to a different
namespace. Cross netns handling is thus transparent to lower PPP
layers (PPPoE, L2TP, etc.).

PPP devices are automatically unregistered when their netns gets
removed. So read() and poll() on the unit file descriptor will
respectively receive EOF and POLLHUP. Channels aren't affected.

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: sun4i-emac: Claim emac sram
Hans de Goede [Sun, 23 Aug 2015 18:31:38 +0000 (20:31 +0200)]
net: sun4i-emac: Claim emac sram

Claim the emac sram ourselves, rather then relying on the bootloader
having mapped the sram to the emac controller during boot.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoinetpeer: remove dead code
David Ahern [Sun, 23 Aug 2015 14:21:22 +0000 (08:21 -0600)]
inetpeer: remove dead code

Remove various inlined functions not referenced in the kernel.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/mlx5e: Avoid accessing NULL pointer at ndo_select_queue
Rana Shahout [Sun, 23 Aug 2015 13:12:14 +0000 (16:12 +0300)]
net/mlx5e: Avoid accessing NULL pointer at ndo_select_queue

To avoid multiply/division operations on the data path,
we hold a {channel, tc}==>txq mapping table.
We held this mapping table inside the channel object that is
being destroyed upon some configuration operations (e.g MTU change).
So in case ndo_select_queue occurs during such a configuration operation,
it may access a NULL channel pointer, resulting in kernel panic.
To fix this issue we moved the {channel, tc}==>txq mapping table
outside the channel object so that it will be available also
during such configuration operations.

Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoah4: Fix error return in ah_input().
David S. Miller [Tue, 25 Aug 2015 20:38:50 +0000 (13:38 -0700)]
ah4: Fix error return in ah_input().

Noticed by Herbert Xu.

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoah6: fix error return code
Julia Lawall [Sun, 23 Aug 2015 00:11:21 +0000 (02:11 +0200)]
ah6: fix error return code

Return a negative error code on failure.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@@
identifier ret; expression e1,e2;
@@
(
if (\(ret < 0\|ret != 0\))
 { ... return ret; }
|
ret = 0
)
... when != ret = e1
    when != &ret
*if(...)
{
  ... when != ret = e2
      when forall
 return ret;
}
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlxsw: fix error return code
Julia Lawall [Sun, 23 Aug 2015 00:11:16 +0000 (02:11 +0200)]
mlxsw: fix error return code

Return a negative error code on failure.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@@
identifier ret; expression e1,e2;
@@
(
if (\(ret < 0\|ret != 0\))
 { ... return ret; }
|
ret = 0
)
... when != ret = e1
    when != &ret
*if(...)
{
  ... when != ret = e2
      when forall
 return ret;
}
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: davinci_emac: fix error return code
Julia Lawall [Sun, 23 Aug 2015 00:11:14 +0000 (02:11 +0200)]
net: davinci_emac: fix error return code

Propagate error code on failure.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@@
identifier ret; expression e1,e2;
@@
(
if (\(ret < 0\|ret != 0\))
 { ... return ret; }
|
ret = 0
)
... when != ret = e1
    when != &ret
*if(...)
{
  ... when != ret = e2
      when forall
 return ret;
}
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'rds-assorted-bug-fixes'
David S. Miller [Tue, 25 Aug 2015 20:35:32 +0000 (13:35 -0700)]
Merge branch 'rds-assorted-bug-fixes'

Santosh Shilimkar says:

====================
RDS: Assorted bug fixes

We would like to improve RDS upstream support and in that context, I
started playing with it.  But run into number of issues including as
basic is RDS IB RDMA doesn't work. As part of the debug, I ended up
creating the $subject series which has bunch of assorted fixes. At
least with this series I can run RDS IB RDMA and other tests
successfully.

Some of these fixes have been done by Chris Meson, Andy Grover and
Zach Brown while at Oracle. There are still more kinks with FMR and
error handling and I plan to address them in a follow up series.

Series generated against Linus's master(v4.2-rc-7) but also applies
against next-next cleanly. Its tested on Oracle hardware with IB
fabric for both bcopy as well as RDMA mode. I don't have access
to iWARP hardware so any testing help on iWARP hardware appreciated.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: check for valid cm_id before initiating connection
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:35 +0000 (15:45 -0700)]
RDS: check for valid cm_id before initiating connection

Connection could have been dropped while the route is being resolved
so check for valid cm_id before initiating the connection.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: return EMSGSIZE for oversize requests before processing/queueing
Mukesh Kacker [Sat, 22 Aug 2015 22:45:34 +0000 (15:45 -0700)]
RDS: return EMSGSIZE for oversize requests before processing/queueing

rds_send_queue_rm() allows for the "current datagram" being queued
to exceed SO_SNDBUF thresholds by checking bytes queued without
counting in length of current datagram. (Since sk_sndbuf is set
to twice requested SO_SNDBUF value as a kernel heuristic this
is usually fine!)

If this "current datagram" squeezing past the threshold is itself
many times the size of the sk_sndbuf threshold itself then even
twice the SO_SNDBUF does not save us and it gets queued but
cannot be transmitted. Threads block and deadlock and device
becomes unusable. The check for this datagram not exceeding
SNDBUF thresholds (EMSGSIZE) is not done on this datagram as
that check is only done if queueing attempt fails.
(Datagrams that follow this datagram fail queueing attempts, go
through the check and eventually trip EMSGSIZE error but zero
length datagrams silently fail!)

This fix moves the check for datagrams exceeding SNDBUF limits
before any processing or queueing is attempted and returns EMSGSIZE
early in the rds_sndmsg() code. This change also ensures that all
datagrams get checked for exceeding SNDBUF/sk_sndbuf size limits
and the large datagrams that exceed those limits do not get to
rds_send_queue_rm() code for processing.

Signed-off-by: Mukesh Kacker <mukesh.kacker@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: make sure rds_send_drop_to properly takes the m_rs_lock
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:33 +0000 (15:45 -0700)]
RDS: make sure rds_send_drop_to properly takes the m_rs_lock

rds_send_drop_to() is used during socket tear down to find all the
messages on the socket and flush them .  It can race with the
acking code unless it takes the m_rs_lock on each and every message.

This plugs a hole where we didn't take m_rs_lock on any message that
didn't have the RDS_MSG_ON_CONN set.  Taking m_rs_lock avoids
double frees and other memory corruptions as the ack code trusts
the message m_rs pointer on a socket that had actually been freed.

We must take m_rs_lock to access m_rs.  Because of lock nesting and
rs access, we also need to acquire rs_lock.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: Don't destroy the rdma id until after we're done using it
Santosh Shilimkar [Sat, 22 Aug 2015 22:45:32 +0000 (15:45 -0700)]
RDS: Don't destroy the rdma id until after we're done using it

During connection resets, we are destroying the rdma id too soon. We can't
destroy it when it is still in use. So lets move rdma_destroy_id() after
we clear the rings.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: Fix assertion level from fatal to warning
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:31 +0000 (15:45 -0700)]
RDS: Fix assertion level from fatal to warning

Fix the asserion level since its not fatal and can be hit
in normal execution paths. There is no need to take the
system down.

We keep the WARN_ON() to detect the condition if we get
here with bad pages.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: Make sure we do a signaled send for large-send
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:30 +0000 (15:45 -0700)]
RDS: Make sure we do a signaled send for large-send

WR(Work Requests )always generate a WC(Work Completion) with
signaled send. Default RDS ib code is setup for un-signaled
completion. Since RDS connction is persistent, we can end up
sending the data even after large-send when the remote end is
not active(for any reason).

By doing  a signaled send at least once per large-send,
we can at least detect the problem in work completion
handler there by avoiding sending more data to
inactive remote.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: Mark message mapped before transmit
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:29 +0000 (15:45 -0700)]
RDS: Mark message mapped before transmit

rds_send_xmit() marks the rds message map flag after
xmit_[rdma/atomic]() which is clearly wrong.  We need
to maintain the ownership between transport and rds.

Also take care of error path.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: add a sock_destruct callback debug aid
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:28 +0000 (15:45 -0700)]
RDS: add a sock_destruct callback debug aid

This helps to detect the accidental processes/apps trying to destroy
the RDS socket which they are sharing with other processes/apps.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: check for congestion updates during rds_send_xmit
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:27 +0000 (15:45 -0700)]
RDS: check for congestion updates during rds_send_xmit

Ensure we don't keep sending the data if the link is congested.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: make sure we post recv buffers
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:26 +0000 (15:45 -0700)]
RDS: make sure we post recv buffers

If we get an ENOMEM during rds_ib_recv_refill, we might never come
back and refill again later. Patch makes sure to kick krdsd into
helping out.

To achieve this we add RDS_RECV_REFILL flag and update in the refill
path based on that so that at least some therad will keep posting
receive buffers.

Since krdsd and softirq both might race for refill, we decide to
schedule on work queue based on ring_low instead of ring_empty.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: don't update ip address tables if the address hasn't changed
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:25 +0000 (15:45 -0700)]
RDS: don't update ip address tables if the address hasn't changed

If the ip address tables hasn't changed, there is no need to remove
them only to be added back again.

Lets fix it.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: destroy the ib state earlier during shutdown
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:24 +0000 (15:45 -0700)]
RDS: destroy the ib state earlier during shutdown

Destroy ib state early during shutdown. Otherwise we can get callbacks
after the QP isn't really able to handle them.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: always free recv frag as we free its ring entry
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:23 +0000 (15:45 -0700)]
RDS: always free recv frag as we free its ring entry

We were still seeing rare occurrences of the WARN_ON(recv->r_frag) which
indicates that the recv refill path was finding allocated frags in ring
entries that were marked free. These were usually followed by OOM crashes.
They only seem to be occurring in the presence of completion errors and
connection resets.

This patch ensures that we free the frag as we mark the ring entry free.
This should stop the refill path from finding allocated frags in ring
entries that were marked free.

Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoRDS: restore return value in rds_cmsg_rdma_args()
santosh.shilimkar@oracle.com [Sat, 22 Aug 2015 22:45:22 +0000 (15:45 -0700)]
RDS: restore return value in rds_cmsg_rdma_args()

In rds_cmsg_rdma_args() 'ret' is used by rds_pin_pages() which returns
number of pinned pages on success. And the same value is returned to the
caller of rds_cmsg_rdma_args() on success which is not intended.

Commit f4a3fc03c1d7 ("RDS: Clean up error handling in rds_cmsg_rdma_args")
removed the 'ret = 0' line which broke RDS RDMA mode.

Fix it by restoring the return value on rds_pin_pages() success
keeping the clean-up in place.

Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotcp: refine pacing rate determination
Eric Dumazet [Sat, 22 Aug 2015 00:38:02 +0000 (17:38 -0700)]
tcp: refine pacing rate determination

When TCP pacing was added back in linux-3.12, we chose
to apply a fixed ratio of 200 % against current rate,
to allow probing for optimal throughput even during
slow start phase, where cwnd can be doubled every other gRTT.

At Google, we found it was better applying a different ratio
while in Congestion Avoidance phase.
This ratio was set to 120 %.

We've used the normal tcp_in_slow_start() helper for a while,
then tuned the condition to select the conservative ratio
as soon as cwnd >= ssthresh/2 :

- After cwnd reduction, it is safer to ramp up more slowly,
  as we approach optimal cwnd.
- Initial ramp up (ssthresh == INFINITY) still allows doubling
  cwnd every other RTT.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>