openwrt/staging/blogic.git
10 years agoe100: Convert uses of __constant_<foo> to <foo>
Joe Perches [Thu, 13 Mar 2014 05:19:09 +0000 (05:19 +0000)]
e100: Convert uses of __constant_<foo> to <foo>

The use of __constant_<foo> has been unnecessary for quite awhile now.

Make these uses consistent with the rest of the kernel.

Signed-off-by: Joe Perches <joe@perches.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoigb: enable VLAN stripping for VMs with i350
Stefan Assmann [Wed, 11 Dec 2013 22:10:12 +0000 (22:10 +0000)]
igb: enable VLAN stripping for VMs with i350

For i350 VLAN stripping for VMs is not enabled in the VMOLR register but in
the DVMOLR register. Making the changes accordingly. It's not necessary to
unset the E1000_VMOLR_STRVLAN bit on i350 as the hardware will simply ignore
it.

Without this change if a VLAN is configured for a VF assigned to a guest
via (i.e.)
ip link set p1p1 vf 0 vlan 10
the VLAN tag will not be stripped from packets going into the VM. Which they
should be because the VM itself is not aware of the VLAN at all.

Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoqeth: Fix IP version detection for VLAN traffic
Stefan Raspl [Wed, 19 Mar 2014 06:58:02 +0000 (07:58 +0100)]
qeth: Fix IP version detection for VLAN traffic

The current code would always return 0 for VLAN-encapsulated IP traffic.
One notable side effect was that VLAN traffic would never get prioritized
on OSD and OSX devices when priority queueing modes prio_queueing_tos or
prio_queueing_prec were enabled.

Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqeth: Removed unused parameter
Stefan Raspl [Wed, 19 Mar 2014 06:58:01 +0000 (07:58 +0100)]
qeth: Removed unused parameter

Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoqeth: make qeth_query_card_info_cb() static
Heiko Carstens [Wed, 19 Mar 2014 06:58:00 +0000 (07:58 +0100)]
qeth: make qeth_query_card_info_cb() static

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoaf_iucv: recvmsg problem for SOCK_STREAM sockets
Ursula Braun [Wed, 19 Mar 2014 06:57:59 +0000 (07:57 +0100)]
af_iucv: recvmsg problem for SOCK_STREAM sockets

Commit f9c41a62bba3f3f7ef3541b2a025e3371bcbba97 introduced
a problem for SOCK_STREAM sockets, when only part of the
incoming iucv message is received by user space. In this
case the remaining data of the iucv message is lost.
This patch makes sure an incompletely received iucv message
is queued back to the receive queue.

Signed-off-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Reported-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Thu, 20 Mar 2014 03:55:48 +0000 (23:55 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/net-next

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates

This series contains updates to i40e, i40evf, e1000e, ixgbe and ixgbevf.

Mitch adds support for the VF link state ndo which allows the PF driver
to control the virtual link state of the VF devices.  Added
support for viewing and modifying RSS hash options and RSS hash look-up
table programming through ethtool for i40evf.  Fixed complaint about
the use of min() where min_t() should be used in i40evf.

Anjali adds support for ethtool -k option for NTUPLE control for i40e.

Elizabeth cleans up and refactors i40e_open() to separate out the VSI
code into its own i40e_vsi_open().

Jesse enables the hardware feature head write back to avoid updating the
descriptor ring by marking each descriptor with a DD bit and instead
writes a memory location with an update to where the driver should clean
up to in i40e and i40evf.  Reduces context descriptors for i40e/i40evf
since we do not need context descriptors for every packet, only for
TSO or timesync.

Dan Carpenter fixes a potential array underflow in i40e_vc_process_vf_msg().

Dave fixes an e1000e hardware unit hang where the check for pending Tx work
when link is lost was mistakenly moved to be done only when link is first
detected to be lost.  Fixed a problem with poor network performance on
certain silicon in e1000e when configured for 100M HDX performance.

Carolyn adds register defines needed for time sync functions and the code
to call the updated defines.

Jacob adds the ixgbe function for writing PCI config word and checks
whether the adapter has been removed first.

Mark adds the bit __IXGBEVF_REMOVING to indicate that the module is being
removed because the __IXGBEVF_DOWN bit had been overloaded for this
purpose, but leads to trouble.  ixgbevf_down function can now prevent
multiple executions by doing test_and_set_bit on __IXGBEVF_DOWN.

v2:
- dropped patch Mitch's patch "i40evf: Support RSS option in ethtool"
  based on feedback from Ben Hutchings so that Mitch can re-work the
  patch solution

v3:
- removed unnecessary parenthesis in patch 1 based on feedback from David
  Miller
- changed a macro to get the next queue to a function in patch 2 based on
  feedback from David Miller
- added blank lines after variable declaration and code in two functions
  in patch 6 based on feedback from David Miller
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbevf: Protect ixgbevf_down with __IXGBEVF_DOWN bit
Mark Rustad [Tue, 4 Mar 2014 03:02:18 +0000 (03:02 +0000)]
ixgbevf: Protect ixgbevf_down with __IXGBEVF_DOWN bit

The ixgbevf_down function can now prevent multiple executions by
doing test_and_set_bit on __IXGBEVF_DOWN. This did not work before
introduction of the __IXGBEVF_REMOVING bit, because of overloading
of __IXGBEVF_DOWN. Also add smp_mb__before_clear_bit call before
clearing the __IXGBEVF_DOWN bit.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoixgbevf: Indicate removal state explicitly
Mark Rustad [Tue, 4 Mar 2014 03:02:13 +0000 (03:02 +0000)]
ixgbevf: Indicate removal state explicitly

Add a bit, __IXGBEVF_REMOVING, to indicate that the module is being
removed. The __IXGBEVF_DOWN bit had been overloaded for this purpose,
but that leads to trouble. A few places now check both __IXGBEVF_DOWN
and __IXGBEVF_REMOVING.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoixgbe: add ixgbe_write_pci_cfg_word with ixgbe_removed check
Jacob Keller [Sat, 22 Feb 2014 01:23:53 +0000 (01:23 +0000)]
ixgbe: add ixgbe_write_pci_cfg_word with ixgbe_removed check

Inline with the current use for ixgbe_read_pci_cfg_word, create a
similar function for writing PCI config, which checks whether the
adapter has been removed first, if Live Error Recovery has been enabled.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoigb: Add register defines needed for time sync functions
Carolyn Wyborny [Tue, 11 Mar 2014 06:15:37 +0000 (06:15 +0000)]
igb: Add register defines needed for time sync functions

This patch adds defines needed for implementing the auxiliary time sync
functions and also changes code to call the updated defines instead of
the old.

Reported-by: Richard Cochran <ricahrdcochran@gmail.com>
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoe1000e: Fix Explicitly set Transmit Control Register
David Ertman [Mon, 13 Jan 2014 23:19:27 +0000 (23:19 +0000)]
e1000e: Fix Explicitly set Transmit Control Register

This patch causes the TCTL to be explicitly set to fix a problem with
poor network performance (throughput) on certain silicon when configured
for 100M HDX performance.

Cc: Todd Fujinaka <todd.fujinaka@intel.com>
Signed-off-by: Dave Ertman <davidx.m.ertman@intel.com>
Acked-by: Bruce W. Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoe1000e: Fix Hardware Unit Hang
David Ertman [Wed, 8 Jan 2014 01:07:55 +0000 (01:07 +0000)]
e1000e: Fix Hardware Unit Hang

The check for pending Tx work when link is lost was mistakenly moved to be
done only when link is first detected to be lost.  It turns out there is a
small window of opportunity for additional Tx work to get queued up shortly
after link is dropped.

Move the check back to the place it was before in the watchdog task.  Put in
additional debug information for other reset paths and a final catch-all for
false hangs in the scheduled function that prints out the hardware hang
message.

Signed-off-by: Dave Ertman <davidx.m.ertman@intel.com>
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40e/i40evf: Bump build versions
Catherine Sullivan [Fri, 14 Feb 2014 02:14:42 +0000 (02:14 +0000)]
i40e/i40evf: Bump build versions

Bump to version 0.3.36 for i40e and 0.9.16 for i40evf.

Change-ID: I7b4ff97b32d2825181803c03c316381a7608a618
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40e: potential array underflow in i40e_vc_process_vf_msg()
Dan Carpenter [Wed, 15 Jan 2014 06:43:39 +0000 (06:43 +0000)]
i40e: potential array underflow in i40e_vc_process_vf_msg()

If "vf_id" is smaller than hw->func_caps.vf_base_id then it leads to
an array underflow of the pf->vf[] array.  This is unlikely to happen
unless the hardware is bad, but it's a small change and it silences a
static checker warning.

Fixes: 7efa84b7abc1 ('i40e: support VFs on PFs other than 0')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40e/i40evf: reduce context descriptors
Jesse Brandeburg [Fri, 14 Feb 2014 02:14:41 +0000 (02:14 +0000)]
i40e/i40evf: reduce context descriptors

We don't need context descriptors for every packet, only tso
or timesync.  This fixes a bug in the driver where it would
always add a context even if all the passed in values
to the context descriptor function were 0/default values.

Change-ID: I0101d2b893380707b5c2de61aab3e16d4310e9a1
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40e/i40evf: enable hardware feature head write back
Jesse Brandeburg [Fri, 14 Feb 2014 02:14:40 +0000 (02:14 +0000)]
i40e/i40evf: enable hardware feature head write back

The hardware supports a feature to avoid updating the descriptor
ring by marking each descriptor with a DD bit, and instead
writes a memory location with an update to where the driver
should clean up to.  Enable this feature.

Change-ID: I5da4e0681f0b581a6401c950a81808792267fe57
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40e: Refactor and cleanup i40e_open(), adding i40e_vsi_open()
Elizabeth Kappler [Sat, 15 Feb 2014 07:41:38 +0000 (07:41 +0000)]
i40e: Refactor and cleanup i40e_open(), adding i40e_vsi_open()

This patch cleans up and moves a portion of i40e_open to i40e_vsi_open,
in order to have a shorter vsi_open function that does only that.

Change-ID: I1c418dda94dcfc0eb7d4386a70c330692ef5ecc9
Signed-off-by: Elizabeth Kappler <elizabeth.m.kappler@intel.com>
Signed-off-by: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40e: Patch to enable Ethtool/netdev feature flag for NTUPLE control
Anjali Singhai Jain [Fri, 14 Feb 2014 02:14:38 +0000 (02:14 +0000)]
i40e: Patch to enable Ethtool/netdev feature flag for NTUPLE control

This enables option '-k/-K' in ethtool for NTUPLE control.
NTUPLE control requires a reset, to take effect. When the feature is
turned off, the SW list of stored FD SB filters gets cleaned up.

Change-ID: I9d564b67a10d4afa11de3b320d601c3d2e6edc1f
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40evf: use min_t
Mitch Williams [Tue, 11 Feb 2014 08:27:52 +0000 (08:27 +0000)]
i40evf: use min_t

Checkpatch complained in an earlier patch about using min(), but that
change would have been completely unrelated to the point of that patch.
So fix it here.

Change-ID: I2cd87b39cfd406850d283b88f259757a6bcd14cd
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agoi40evf: correctly program RSS HLUT table
Mitch Williams [Tue, 11 Feb 2014 08:27:50 +0000 (08:27 +0000)]
i40evf: correctly program RSS HLUT table

The HLUT programming loop in in i40evf_configure_rss was a) overly-
complicated, and b) just plain broken. Most of the entries ended up being
not written at all, so most of the flows ended up at queue zero.

Refactor the HLUT programming loop to simply walk through the registers
and write four values to each one, incrementing through the number of
available queues.

Change-ID: I75766179bc67e4e997187794f3144e28c83fd00d
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agomicrel: fix masking off LED bits
Sergei Shtylyov [Tue, 18 Mar 2014 23:58:16 +0000 (02:58 +0300)]
micrel: fix masking off LED bits

Commit 20d8435a1cff (phy: micrel: add of configuration for LED mode) made the
obvious mistake when masking off  the LED mode bits: forgot to do a logical NOT
to the mask with which it ANDs the register value, so that unrelated bits are
cleared instead.

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Ben Dooks <ben.dooks@codethink.co.uk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: support VF link state ndo
Mitch Williams [Tue, 11 Feb 2014 08:27:49 +0000 (08:27 +0000)]
i40e: support VF link state ndo

This netdev op allows the PF driver to control the virtual link state of
the VF devices. This can be used to deny naughty VF drivers access to
the wire, or to allow VFs (regardless of temperament) to communicate
with each other over the device's internal switch even though external
link is down.

Add the actual ndo function, and modify vc_notify_link_state to check
the link status of each VF before sending a message in the case when
physical link changes state.

Change-ID: Ib5a6924da78c540789f21d26b5e8086d71c29384
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
10 years agonetfilter: Add missing vmalloc.h include to nft_hash.c
David S. Miller [Wed, 19 Mar 2014 03:12:02 +0000 (23:12 -0400)]
netfilter: Add missing vmalloc.h include to nft_hash.c

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoieee802154: properly unshare skbs in ieee802154 *_rcv functions
Phoebe Buckheister [Mon, 17 Mar 2014 17:30:19 +0000 (18:30 +0100)]
ieee802154: properly unshare skbs in ieee802154 *_rcv functions

ieee802154 sockets do not properly unshare received skbs, which leads to
panics (at least) when they are used in conjunction with 6lowpan, so
run skb_share_check on received skbs.
6lowpan also contains a use-after-free, which is trivially fixed by
replacing the inlined skb_share_check with the explicit call.

Signed-off-by: Phoebe Buckheister <phoebe.buckheister@itwm.fraunhofer.de>
Tested-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge tag 'linux-can-next-for-3.15-20140317' of git://gitorious.org/linux-can/linux...
David S. Miller [Tue, 18 Mar 2014 19:20:00 +0000 (15:20 -0400)]
Merge tag 'linux-can-next-for-3.15-20140317' of git://gitorious.org/linux-can/linux-can-next

linux-can-next-for-3.15-20140317

Marc Kleine-Budde says:

====================
this is a pull request of three patches for net-next/master.

It consists of a patch by Oliver Hartkopp, which unifies the MTU
settings for CAN interfaces. A patch by Christopher R. Baker populates
netdev::dev_id for udev discrimination for multi interface CAN devices.
Alexander Shiyan contributes a patch for the mcp251x driver which fixes
the regulators operation if CONFIG_REGULATOR is not enabled.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovia: fix a punctuation typo
wangweidong [Mon, 17 Mar 2014 07:52:17 +0000 (15:52 +0800)]
via: fix a punctuation typo

In generic, after an assignment, we use ';' instead of ','.
Although, it won't hurt.

Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agodoc: update driver TX algorithm in timestamping.txt
Jakub Kicinski [Sun, 16 Mar 2014 19:32:48 +0000 (20:32 +0100)]
doc: update driver TX algorithm in timestamping.txt

Since cd4d8fdad1f1 ("net: kernel panic in dev_hard_start_xmit:
remove faulty software TX time stamping") dev_hard_start_xmit()
will not provide software timestamps. It's a responsibility of
the drivers to call skb_tx_timestamp() at the right time.

Cc: linux-doc@vger.kernel.org
Signed-off-by: Jakub Kicinski <kubakici@wp.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: ratelimit pr_warn()s in 802.3ad mode
Veaceslav Falico [Sun, 16 Mar 2014 16:55:03 +0000 (17:55 +0100)]
bonding: ratelimit pr_warn()s in 802.3ad mode

Only ratelimit the ones that might spam, omiting the ones from
enslave/deslave.

CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: sched: use no more than one page in struct fw_head
Eric Dumazet [Tue, 18 Mar 2014 03:20:49 +0000 (20:20 -0700)]
net: sched: use no more than one page in struct fw_head

In commit b4e9b520ca5d ("[NET_SCHED]: Add mask support to fwmark
classifier") Patrick added an u32 field in fw_head, making it slightly
bigger than one page.

Lets use 256 slots to make fw_hash() more straight forward, and move
@mask to the beginning of the structure as we often use a small number
of skb->mark. @mask and first hash buckets share the same cache line.

This brings back the memory usage to less than 4000 bytes, and permits
John to add a rcu_head at the end of the structure later without any
worry.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Thomas Graf <tgraf@suug.ch>
Cc: John Fastabend <john.fastabend@gmail.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec...
David S. Miller [Tue, 18 Mar 2014 18:09:07 +0000 (14:09 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/klassert/ipsec-next

Steffen Klassert says:

====================
One patch to rename a newly introduced struct. The rest is
the rework of the IPsec virtual tunnel interface for ipv6 to
support inter address family tunneling and namespace crossing.

1) Rename the newly introduced struct xfrm_filter to avoid a
   conflict with iproute2. From Nicolas Dichtel.

2) Introduce xfrm_input_afinfo to access the address family
   dependent tunnel callback functions properly.

3) Add and use a IPsec protocol multiplexer for ipv6.

4) Remove dst_entry caching. vti can lookup multiple different
   dst entries, dependent of the configured xfrm states. Therefore
   it does not make to cache a dst_entry.

5) Remove caching of flow informations. vti6 does not use the the
   tunnel endpoint addresses to do route and xfrm lookups.

6) Update the vti6 to use its own receive hook.

7) Remove the now unused xfrm_tunnel_notifier. This was used from vti
   and is replaced by the IPsec protocol multiplexer hooks.

8) Support inter address family tunneling for vti6.

9) Check if the tunnel endpoints of the xfrm state and the vti interface
   are matching and return an error otherwise.

10) Enable namespace crossing for vti devices.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet/i40e: Avoid double setting of NETIF_F_SG for the HW encapsulation feature mask
Or Gerlitz [Tue, 18 Mar 2014 08:36:45 +0000 (10:36 +0200)]
net/i40e: Avoid double setting of NETIF_F_SG for the HW encapsulation feature mask

The networking core does it for the driver during registration time.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: Rename i40e_ptype_lookup i40evf_ptype_lookup
Eric W Biederman [Tue, 18 Mar 2014 07:26:50 +0000 (00:26 -0700)]
i40evf: Rename i40e_ptype_lookup i40evf_ptype_lookup

When compiling the i40e and the i40evf driver into the same kernel I get:
LD      drivers/net/ethernet/intel/built-in.o
drivers/net/ethernet/intel/i40evf/built-in.o:(.data+0x300): multiple definition of `i40e_ptype_lookup'
drivers/net/ethernet/intel/i40e/built-in.o:(.data+0x780): first defined here
make[3]: *** [drivers/net/ethernet/intel/built-in.o] Error 1
make[2]: *** [drivers/net/ethernet/intel] Error 2
make[1]: *** [drivers/net/ethernet/] Error 2
make: *** [sub-make] Error 2

Fix this by renaming the i40evf version of this structure from
i40e_ptype_lookup to i40evf_ptype_lookup.

This build failure was introduced in:
  commit 206812b5fccb808d1194344eaa942f68f59b2630
  Author: Jesse Brandeburg <jesse.brandeburg@intel.com>
  i40e/i40evf: i40e implementation for skb_set_hash

Cc: Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: Catherine Sullivan <catherine.sullivan@intel.com>
Signed-off-by: Eric W Biederman <ebiederm@xmission.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoe1000e: fix the build error when PM is disabled
Kevin Hao [Tue, 18 Mar 2014 07:26:49 +0000 (00:26 -0700)]
e1000e: fix the build error when PM is disabled

The commit 2800209994f8 (e1000e: Refactor PM flows) changed the
SET_SYSTEM_SLEEP_PM_OPS to open-coded assignment, but forgot to
protect them with CONFIG_PM_SLEEP. Then cause the following build
error when PM is disabled:
drivers/net/ethernet/intel/e1000e/netdev.c:7079:13:
error: 'e1000e_pm_suspend' undeclared here (not in a function)
  .suspend = e1000e_pm_suspend,
             ^
drivers/net/ethernet/intel/e1000e/netdev.c:7080:13:
error: 'e1000e_pm_resume' undeclared here (not in a function)
  .resume  = e1000e_pm_resume,
             ^
drivers/net/ethernet/intel/e1000e/netdev.c:7082:11:
error: 'e1000e_pm_thaw' undeclared here (not in a function)
  .thaw  = e1000e_pm_thaw,
           ^
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoigb: remove references to long gone command line parameters
Fernando Luis Vazquez Cao [Tue, 18 Mar 2014 07:26:48 +0000 (00:26 -0700)]
igb: remove references to long gone command line parameters

Command line parameters QueuePairs, Node, EEE, DMAC and InterruptThrottleRate
do not exist these days. Remove all references to them in the Documentation
folder and update code comments.

Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'altera_tse'
David S. Miller [Tue, 18 Mar 2014 01:37:25 +0000 (21:37 -0400)]
Merge branch 'altera_tse'

Vince Bridgers says:

====================
Altera Triple Speed Ethernet (TSE) Driver

This is the version 6 submission for the Altera Triple Speed Ethernet (TSE)
driver. All comments received during the version 2, 3, 4, and 5 submissions
have been accepted. Please find the change log and a description of the
submission below.

If you find the submission acceptable, please consider this patch set for
inclusion into the Linux kernel.

V6: Address comments from V5 review
    - add call to skb_tx_timestamp in the drivers transmit path
    - correct use of unsigned int where it was cast to pointer. Use types
      appropriate for intended and correct use to let the compiler warn us
      when type usage is incorrect.
    - use correct semantics for pointer arithmetic in same code path

V5: Address comments from V4 review
    - Add descriptions of statistics to driver documentation. The statstics
      supported by the driver/controller map to IEEE and RFC statistics, and
      the names and mappings are described in the user documentation.
    - Change "unsigned int" to u32 in device structure definitions
    - Change used of netdev_warn to netif_warn in altera_sgdma.c
    - Change stat name rx_fifo_drops to ether_drops to match the event
      actually counted by the hardware.

V4: Address comments from V3 review
    - Change statistics names in ethtool module to follow common use in
      other ethernet drivers.
    - remove an unnecessary case in ethtool module
    - change logging to use netdev_* where possible instead of dev_*
    - remove logging for OOM errors since those are already logged

V3: Address comments from V2 review
    - Reorder patch submission so that net/ethernet Makefile and Kconfig
      are committed last, thus not breaking bisect
    - Use of_get_mac_address instead of of_get_property
    - Change supplemental and hash configuration bindings to boolean/empty,
      and more meaningful names
    - Add check for failure from calls to of_phy_connect and
      connect_local_phy
    - Correct code to find mdio child node
    - Update bindings document
    - Remove cast to u64 when not necessary
    - add use of const for statistics strings

V2: Address comments from initial RFC review.
    - The driver files were broken up by major sections of functionality.
      These include MSGDMA, SGDMA, Misc, and Main.
    - Add patch for MAINTAINERS file, add the maintainer for this submission
    - Use 32-bit lower/upper physical address accessor functions so the driver
      is 64-bit ready.
    - Use standard bindings where applicable. Especially phy-addr, and change
      "altr,rx-fifo-depth" to "rx-fifo-depth" and "altr,tx-fifo-depth" to
      "tx-fifo-depth".
    - Add use of max-frame-size property
    - Update bindings documents accordingly
    - Correct interrupt handler to use budget parameter in the convential way
    - Use macros consistently to define bit fields across files
    - Correct include exclusion macro in altera_msgdmahw.h (typo)
    - Remove use of barriers, these were not necessary since the DMA APIs
      ensure memory & buffer consistency
    - Remove use of netif_carrier_off in driver
    - move probing of phy from the open function to the probe function
    - use of_get_phy_mode instead of custom function
    - Use the .data field in the device structure to obtain a pointer
      to SGDMA or MSGDMA device specific properties and functions.
    - remove custom function to access devicetree since Altera specific
      bindings requiring it's use have been deprecated in favor of
      standard bindings.

The Altera TSE is a 10/100/1000 Mbps Ethernet soft IP component that can be
configured and synthesized using Quartus, and programmed into Altera FPGAs.
Two types of soft DMA IP components are supported by this driver - the Altera
SGDMA and the MSGDMA. The MSGDMA DMA component is preferred over the SGDMA,
since the SGDMA will be deprecated in favor of the MSGDMA. Software supporting
both is provided for customers still using the SGDMA and to demonstrate how
multiple types of DMA engines may be supported by the TSE driver in the event
customers wish to develop their own custom soft DMA engine for particular
applications.

The design has been tested on Altera's Cyclone 4, 5, and Cyclone 5 SOC
development kits using an ARM A9 processor and an Altera NIOS2 processor.
Differences in CPU/DMA coherency management and address alignment are
addressed by proper use of driver APIs and semantics.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: ethernet: Change Ethernet Makefile and Kconfig for Altera TSE driver
Vince Bridgers [Mon, 17 Mar 2014 22:52:41 +0000 (17:52 -0500)]
net: ethernet: Change Ethernet Makefile and Kconfig for Altera TSE driver

This patch changes the Ethernet Makefile and Kconfig files to add the Altera
Ethernet driver component.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMAINTAINERS: Add entry for Altera Triple Speed Ethernet Driver
Vince Bridgers [Mon, 17 Mar 2014 22:52:40 +0000 (17:52 -0500)]
MAINTAINERS: Add entry for Altera Triple Speed Ethernet Driver

Add a MAINTAINERS entry covering the Altera Triple Speed
Ethernet Driver, with support for the MSGDMA and SGDMA
soft DMA IP components.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAltera TSE: Add Altera Ethernet Driver Makefile and Kconfig
Vince Bridgers [Mon, 17 Mar 2014 22:52:39 +0000 (17:52 -0500)]
Altera TSE: Add Altera Ethernet Driver Makefile and Kconfig

This patch adds the Altera Triple Speed Ethernet Makfile and
Kconfig file.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAltera TSE: Add main and header file for Altera Ethernet Driver
Vince Bridgers [Mon, 17 Mar 2014 22:52:38 +0000 (17:52 -0500)]
Altera TSE: Add main and header file for Altera Ethernet Driver

This patch adds the main driver and header file for the Altera Triple
Speed Ethernet driver.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAltera TSE: Add Miscellaneous Files for Altera Ethernet Driver
Vince Bridgers [Mon, 17 Mar 2014 22:52:37 +0000 (17:52 -0500)]
Altera TSE: Add Miscellaneous Files for Altera Ethernet Driver

This patch adds miscellaneous files for the Altera Ethernet Driver,
including ethtool support.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAltera TSE: Add Altera Ethernet Driver SGDMA file components
Vince Bridgers [Mon, 17 Mar 2014 22:52:36 +0000 (17:52 -0500)]
Altera TSE: Add Altera Ethernet Driver SGDMA file components

This patch adds the SGDMA soft IP support for the Altera Triple
Speed Ethernet driver.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoAltera TSE: Add Altera Ethernet Driver MSGDMA File Components
Vince Bridgers [Mon, 17 Mar 2014 22:52:35 +0000 (17:52 -0500)]
Altera TSE: Add Altera Ethernet Driver MSGDMA File Components

This patch adds the MSGDMA soft IP support for the Altera Triple
Speed Ethernet driver.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoDocumentation: networking: Add Altera Ethernet (TSE) Documentation
Vince Bridgers [Mon, 17 Mar 2014 22:52:34 +0000 (17:52 -0500)]
Documentation: networking: Add Altera Ethernet (TSE) Documentation

This patch adds a bindings description for the Altera Triple Speed Ethernet
(TSE) driver. The bindings support the legacy SGDMA soft IP as well as the
preferred MSGDMA soft IP. The TSE can be configured and synthesized in soft
logic using Altera's Quartus toolchain. Please consult the bindings document
for supported options.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agodts: Add bindings for the Altera Triple Speed Ethernet driver
Vince Bridgers [Mon, 17 Mar 2014 22:52:33 +0000 (17:52 -0500)]
dts: Add bindings for the Altera Triple Speed Ethernet driver

This patch adds a bindings description for the Altera Triple Speed Ethernet
(TSE) driver. The bindings support the legacy SGDMA soft IP as well as the
preferred MSGDMA soft IP. The TSE can be configured and synthesized in soft
logic using Altera's Quartus toolchain. Please consult the bindings document
for supported options.

Signed-off-by: Vince Bridgers <vbridgers2013@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetfilter: conntrack: Fix UP builds
Eric Dumazet [Mon, 17 Mar 2014 20:37:53 +0000 (13:37 -0700)]
netfilter: conntrack: Fix UP builds

ARRAY_SIZE(nf_conntrack_locks) is undefined if spinlock_t is an
empty structure. Replace it by CONNTRACK_LOCKS

Fixes: 93bb0ceb75be ("netfilter: conntrack: remove central spinlock nf_conntrack_lock")
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'at86rf230'
David S. Miller [Mon, 17 Mar 2014 20:10:43 +0000 (16:10 -0400)]
Merge branch 'at86rf230'

Alexander Aring says:

====================
at86rf230: various fixes and devicetree support

this patch series fix some bugs with the at86rf231 chip and cleaup some code.
Also add devicetree support for the at86rf230 driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoat86rf230: add support for devicetree
Alexander Aring [Sat, 15 Mar 2014 08:29:07 +0000 (09:29 +0100)]
at86rf230: add support for devicetree

This patch adds devicetree support for the at86rf230 driver.

Possible gpios to configure are "reset-gpio" and "sleep-gpio".
Also add support to configure the "irq-type" for the irq polarity
register.

Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoat86rf230: make reset pin optionally
Alexander Aring [Sat, 15 Mar 2014 08:29:06 +0000 (09:29 +0100)]
at86rf230: make reset pin optionally

This patch make the reset pin optionally. Some devices like the atben
from qi-hardware don't have a reset pin externally. The usually way is
to turn power off/on for the atben device to initiate a device reset.

Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoat86rf230: change reset timings
Alexander Aring [Sat, 15 Mar 2014 08:29:05 +0000 (09:29 +0100)]
at86rf230: change reset timings

While checkpatch another patch I got a:

"WARNING: msleep < 20ms can sleep for up to 20ms"

The datasheet of at86rf231 and at86rf212 says a minimum delay for reset
pulse width and spi access latency after reset is 625 nanoseconds.

This patch removes the 1 milliseconds sleep and replace it with a 1
microseconds udelay which should be also okay for the reset pulse width.

To change the state from RESET -> TRX_OFF the at86rf230 device needs 120
microseconds, this is a worst case of all at86rf* chips.

Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoat86rf230: move locking state in xmit
Alexander Aring [Sat, 15 Mar 2014 08:29:04 +0000 (09:29 +0100)]
at86rf230: move locking state in xmit

There is no need to lock the clearing of IRQ_TRX_END in status.

Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoat86rf230: fix unexpected state change
Alexander Aring [Sat, 15 Mar 2014 08:29:03 +0000 (09:29 +0100)]
at86rf230: fix unexpected state change

This patch fix a unexpected state change for the at86rf231 chip.
We can't change into STATE_FORCE_TX_ON while the chip is in one of
SLEEP, P_ON, RESET, TRX_OFF, and all *_NOCLK states.

In this case we are in the TRX_OFF state. See datasheet [1] page 71 for
more information.

Without this patch you will get the following message on a at86rf231 device:

[   20.065218] unexpected state change: 8, asked for 4
[   20.070527] ------------[ cut here ]------------
[   20.075414] WARNING: CPU: 0 PID: 160 at net/mac802154/ieee802154_dev.c:43 mac802154_slave_open+0x70/0xb8()
[   20.085594] Modules linked in: autofs4
[   20.089667] CPU: 0 PID: 160 Comm: ifconfig Not tainted 3.14.0-20140108-1-00993-g905c192 #162
[   20.098612] [<c00127b8>] (unwind_backtrace) from [<c0010b1c>] (show_stack+0x10/0x14)
[   20.106819] [<c0010b1c>] (show_stack) from [<c0033838>] (warn_slowpath_common+0x60/0x80)
[   20.115311] [<c0033838>] (warn_slowpath_common) from [<c00338e8>] (warn_slowpath_null+0x18/0x20)
[   20.124590] [<c00338e8>] (warn_slowpath_null) from [<c057b7e8>] (mac802154_slave_open+0x70/0xb8)
[   20.133880] [<c057b7e8>] (mac802154_slave_open) from [<c0488a58>] (__dev_open+0xa8/0x108)
[   20.142553] [<c0488a58>] (__dev_open) from [<c0488cb0>] (__dev_change_flags+0x8c/0x148)
[   20.151051] [<c0488cb0>] (__dev_change_flags) from [<c0488d84>] (dev_change_flags+0x18/0x48)
[   20.159968] [<c0488d84>] (dev_change_flags) from [<c04e2e9c>] (devinet_ioctl+0x2b0/0x63c)
[   20.168623] [<c04e2e9c>] (devinet_ioctl) from [<c04712e4>] (sock_ioctl+0x23c/0x29c)
[   20.176727] [<c04712e4>] (sock_ioctl) from [<c00e3cb8>] (do_vfs_ioctl+0x4a8/0x578)
[   20.184671] [<c00e3cb8>] (do_vfs_ioctl) from [<c00e3dd4>] (SyS_ioctl+0x4c/0x78)
[   20.192402] [<c00e3dd4>] (SyS_ioctl) from [<c000da00>] (ret_fast_syscall+0x0/0x48)
[   20.200392] ---[ end trace 9a34542f4ea08e47 ]---

This patch was tested on at86rf231 and at86rf212.

[1] http://www.atmel.com/images/doc8111.pdf

Signed-off-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'sh_eth'
David S. Miller [Mon, 17 Mar 2014 20:06:48 +0000 (16:06 -0400)]
Merge branch 'sh_eth'

Sergei Shtylyov says:

====================
Beautify 'sh_eth' driver's messages

This patchset converts te driver to using netdev_*() and netif_*() to print out
its messages whenever possible.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosh_eth: fold netif_msg_*() and netdev_*() calls into netif_*() invocations
Sergei Shtylyov [Sat, 15 Mar 2014 00:30:59 +0000 (03:30 +0300)]
sh_eth: fold netif_msg_*() and netdev_*() calls into netif_*() invocations

Now that we call netdev_*() under netif_msg_*() checks, we can fold these into
netif_*() macro invocations.

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosh_eth: convert dev_*() to netdev_*() calls
Sergei Shtylyov [Sat, 15 Mar 2014 00:29:14 +0000 (03:29 +0300)]
sh_eth: convert dev_*() to netdev_*() calls

Convert dev_*(&ndev->dev, ...) to netdev_*(ndev, ...) calls since they are a bit
shorter and at the same time give more information on a device.

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosh_eth: convert pr_*() to netdev_*() calls
Sergei Shtylyov [Sat, 15 Mar 2014 00:27:54 +0000 (03:27 +0300)]
sh_eth: convert pr_*() to netdev_*() calls

Convert pr_*() to netdev_*() calls as the latter provide info on a device.

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosh_eth: exit probe with unknown register layout
Sergei Shtylyov [Sat, 15 Mar 2014 00:11:24 +0000 (03:11 +0300)]
sh_eth: exit probe with unknown register layout

Exit the driver's probe() method when the register layout is unknown as the
driver would cause kernel oops in this case anyway.

While at it, move the corresponding error message printout and convert it from
pr_err() to dev_err().

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'netpoll-next'
David S. Miller [Mon, 17 Mar 2014 19:48:53 +0000 (15:48 -0400)]
Merge branch 'netpoll-next'

Eric W. Biederman says:

====================
netpoll: Cleanup received packet processing

This is the long-winded, careful, and polite version of removing the netpoll
receive packet processing.

First I untangle the code in small steps.  Then I modify the code to not
force reception and dropping of packets when we are transmiting a packet
with netpoll.  Finally I move all of the packet reception under
CONFIG_NETPOLL_TRAP and delete CONFIG_NETPOLL_TRAP.

If someone wants to do a stable backport of these patches, it would
require backporting the first 18 patches that handle the budget == 0 in
the networking drivers, and the first 6 of these patches.

If anyone wants to resurrect netpoll packet reception someday it should
just be a matter of reverting the last patch.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Remove dead packet receive code (CONFIG_NETPOLL_TRAP)
Eric W. Biederman [Sat, 15 Mar 2014 03:51:52 +0000 (20:51 -0700)]
netpoll: Remove dead packet receive code (CONFIG_NETPOLL_TRAP)

The netpoll packet receive code only becomes active if the netpoll
rx_skb_hook is implemented, and there is not a single implementation
of the netpoll rx_skb_hook in the kernel.

All of the out of tree implementations I have found all call
netpoll_poll which was removed from the kernel in 2011, so this
change should not add any additional breakage.

There are problems with the netpoll packet receive code.  __netpoll_rx
does not call dev_kfree_skb_irq or dev_kfree_skb_any in hard irq
context.  netpoll_neigh_reply leaks every skb it receives.  Reception
of packets does not work successfully on stacked devices (aka bonding,
team, bridge, and vlans).

Given that the netpoll packet receive code is buggy, there are no
out of tree users that will be merged soon, and the code has
not been used for in tree for a decade let's just remove it.

Reverting this commit can server as a starting point for anyone
who wants to resurrect netpoll packet reception support.

Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Move all receive processing under CONFIG_NETPOLL_TRAP
Eric W. Biederman [Sat, 15 Mar 2014 03:50:58 +0000 (20:50 -0700)]
netpoll: Move all receive processing under CONFIG_NETPOLL_TRAP

Make rx_skb_hook, and rx in struct netpoll depend on
CONFIG_NETPOLL_TRAP Make rx_lock, rx_np, and neigh_tx in struct
netpoll_info depend on CONFIG_NETPOLL_TRAP

Make the functions netpoll_rx_on, netpoll_rx, and netpoll_receive_skb
no-ops when CONFIG_NETPOLL_TRAP is not set.

Only build netpoll_neigh_reply, checksum_udp service_neigh_queue,
pkt_is_ns, and __netpoll_rx when CONFIG_NETPOLL_TRAP is defined.

Add helper functions netpoll_trap_setup, netpoll_trap_setup_info,
netpoll_trap_cleanup, and netpoll_trap_cleanup_info that initialize
and cleanup the struct netpoll and struct netpoll_info receive
specific fields when CONFIG_NETPOLL_TRAP is enabled and do nothing
otherwise.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Consolidate neigh_tx processing in service_neigh_queue
Eric W. Biederman [Sat, 15 Mar 2014 03:50:25 +0000 (20:50 -0700)]
netpoll: Consolidate neigh_tx processing in service_neigh_queue

Move the bond slave device neigh_tx handling into service_neigh_queue.

In connection with neigh_tx processing remove unnecessary tests of
a NULL netpoll_info.  As the netpoll_poll_dev has already used
and thus verified the existince of the netpoll_info.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Move netpoll_trap under CONFIG_NETPOLL_TRAP
Eric W. Biederman [Sat, 15 Mar 2014 03:49:43 +0000 (20:49 -0700)]
netpoll: Move netpoll_trap under CONFIG_NETPOLL_TRAP

Now that we no longer need to receive packets to safely drain the
network drivers receive queue move netpoll_trap and netpoll_set_trap
under CONFIG_NETPOLL_TRAP

Making netpoll_trap and netpoll_set_trap noop inline functions
when CONFIG_NETPOLL_TRAP is not set.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Don't drop all received packets.
Eric W. Biederman [Sat, 15 Mar 2014 03:48:28 +0000 (20:48 -0700)]
netpoll: Don't drop all received packets.

Change the strategy of netpoll from dropping all packets received
during netpoll_poll_dev to calling napi poll with a budget of 0
(to avoid processing drivers rx queue), and to ignore packets received
with netif_rx (those will safely be placed on the backlog queue).

All of the netpoll supporting drivers have been reviewed to ensure
either thay use netif_rx or that a budget of 0 is supported by their
napi poll routine and that a budget of 0 will not process the drivers
rx queues.

Not dropping packets makes NETPOLL_RX_DROP unnecesary so it is removed.

npinfo->rx_flags is removed  as rx_flags with just the NETPOLL_RX_ENABLED
flag becomes just a redundant mirror of list_empty(&npinfo->rx_np).

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Add netpoll_rx_processing
Eric W. Biederman [Sat, 15 Mar 2014 03:47:49 +0000 (20:47 -0700)]
netpoll: Add netpoll_rx_processing

Add a helper netpoll_rx_processing that reports when netpoll has
receive side processing to perform.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Warn if more packets are processed than are budgeted
Eric W. Biederman [Sat, 15 Mar 2014 03:47:15 +0000 (20:47 -0700)]
netpoll: Warn if more packets are processed than are budgeted

There is already a warning for this case in the normal netpoll path,
but put a copy here in case how netpoll calls the poll functions
causes a differenet result.

netpoll will shortly call the napi poll routine with a budget 0 to
avoid any rx packets being processed.  As nothing does that today
we may encounter drivers that have problems so a netpoll specific
warning seems desirable.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Visit all napi handlers in poll_napi
Eric W. Biederman [Sat, 15 Mar 2014 03:45:51 +0000 (20:45 -0700)]
netpoll: Visit all napi handlers in poll_napi

In poll_napi loop through all of the napi handlers even when the
budget falls to 0 to ensure that we process all of the tx_queues, and
so that we continue to call into drivers when our initial budget is 0.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: Pass budget into poll_napi
Eric W. Biederman [Sat, 15 Mar 2014 03:45:17 +0000 (20:45 -0700)]
netpoll: Pass budget into poll_napi

This moves the control logic to the top level in netpoll_poll_dev
instead of having it dispersed throughout netpoll_poll_dev,
poll_napi and poll_one_napi.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetpoll: move setting of NETPOLL_RX_DROP into netpoll_poll_dev
Eric W. Biederman [Sat, 15 Mar 2014 03:44:37 +0000 (20:44 -0700)]
netpoll: move setting of NETPOLL_RX_DROP into netpoll_poll_dev

Today netpoll depends on setting NETPOLL_RX_DROP before networking
drivers receive packets in interrupt context so that the packets can
be dropped.  Move this setting into netpoll_poll_dev from
poll_one_napi so that if ndo_poll_controller happens to receive
packets we will drop the packets on the floor instead of letting the
packets bounce through the networking stack and potentially cause problems.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Mon, 17 Mar 2014 19:06:24 +0000 (15:06 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

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

The following patchset contains Netfilter/IPVS updates for net-next,
most relevantly they are:

* cleanup to remove double semicolon from stephen hemminger.

* calm down sparse warning in xt_ipcomp, from Fan Du.

* nf_ct_labels support for nf_tables, from Florian Westphal.

* new macros to simplify rcu dereferences in the scope of nfnetlink
  and nf_tables, from Patrick McHardy.

* Accept queue and drop (including reason for drop) to verdict
  parsing in nf_tables, also from Patrick.

* Remove unused random seed initialization in nfnetlink_log, from
  Florian Westphal.

* Allow to attach user-specific information to nf_tables rules, useful
  to attach user comments to rule, from me.

* Return errors in ipset according to the manpage documentation, from
  Jozsef Kadlecsik.

* Fix coccinelle warnings related to incorrect bool type usage for ipset,
  from Fengguang Wu.

* Add hash:ip,mark set type to ipset, from Vytas Dauksa.

* Fix message for each spotted by ipset for each netns that is created,
  from Ilia Mirkin.

* Add forceadd option to ipset, which evicts a random entry from the set
  if it becomes full, from Josh Hunt.

* Minor IPVS cleanups and fixes from Andi Kleen and Tingwei Liu.

* Improve conntrack scalability by removing a central spinlock, original
  work from Eric Dumazet. Jesper Dangaard Brouer took them over to address
  remaining issues. Several patches to prepare this change come in first
  place.

* Rework nft_hash to resolve bugs (leaking chain, missing rcu synchronization
  on element removal, etc. from Patrick McHardy.

* Restore context in the rule deletion path, as we now release rule objects
  synchronously, from Patrick McHardy. This gets back event notification for
  anonymous sets.

* Fix NAT family validation in nft_nat, also from Patrick.

* Improve scalability of xt_connlimit by using an array of spinlocks and
  by introducing a rb-tree of hashtables for faster lookup of accounted
  objects per network. This patch was preceded by several patches and
  refactorizations to accomodate this change including the use of kmem_cache,
  from Florian Westphal.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonetfilter: connlimit: use rbtree for per-host conntrack obj storage
Florian Westphal [Wed, 12 Mar 2014 22:49:51 +0000 (23:49 +0100)]
netfilter: connlimit: use rbtree for per-host conntrack obj storage

With current match design every invocation of the connlimit_match
function means we have to perform (number_of_conntracks % 256) lookups
in the conntrack table [ to perform GC/delete stale entries ].
This is also the reason why ____nf_conntrack_find() in perf top has
> 20% cpu time per core.

This patch changes the storage to rbtree which cuts down the number of
ct objects that need testing.

When looking up a new tuple, we only test the connections of the host
objects we visit while searching for the wanted host/network (or
the leaf we need to insert at).

The slot count is reduced to 32.  Increasing slot count doesn't
speed up things much because of rbtree nature.

before patch (50kpps rx, 10kpps tx):
+  20.95%  ksoftirqd/0  [nf_conntrack] [k] ____nf_conntrack_find
+  20.50%  ksoftirqd/1  [nf_conntrack] [k] ____nf_conntrack_find
+  20.27%  ksoftirqd/2  [nf_conntrack] [k] ____nf_conntrack_find
+   5.76%  ksoftirqd/1  [nf_conntrack] [k] hash_conntrack_raw
+   5.39%  ksoftirqd/2  [nf_conntrack] [k] hash_conntrack_raw
+   5.35%  ksoftirqd/0  [nf_conntrack] [k] hash_conntrack_raw

after (90kpps, 51kpps tx):
+  17.24%       swapper  [nf_conntrack]    [k] ____nf_conntrack_find
+   6.60%   ksoftirqd/2  [nf_conntrack]    [k] ____nf_conntrack_find
+   2.73%       swapper  [nf_conntrack]    [k] hash_conntrack_raw
+   2.36%       swapper  [xt_connlimit]    [k] count_tree

Obvious disadvantages to previous version are the increase in code
complexity and the increased memory cost.

Partially based on Eric Dumazets fq scheduler.

Reviewed-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
10 years agonetfilter: connlimit: make same_source_net signed
Florian Westphal [Wed, 12 Mar 2014 22:49:50 +0000 (23:49 +0100)]
netfilter: connlimit: make same_source_net signed

currently returns 1 if they're the same.  Make it work like mem/strcmp
so it can be used as rbtree search function.

Reviewed-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
10 years agonetfilter: connlimit: use keyed locks
Florian Westphal [Wed, 12 Mar 2014 22:49:49 +0000 (23:49 +0100)]
netfilter: connlimit: use keyed locks

connlimit currently suffers from spinlock contention, example for
4-core system with rps enabled:

+  20.84%   ksoftirqd/2  [kernel.kallsyms] [k] _raw_spin_lock_bh
+  20.76%   ksoftirqd/1  [kernel.kallsyms] [k] _raw_spin_lock_bh
+  20.42%   ksoftirqd/0  [kernel.kallsyms] [k] _raw_spin_lock_bh
+   6.07%   ksoftirqd/2  [nf_conntrack]    [k] ____nf_conntrack_find
+   6.07%   ksoftirqd/1  [nf_conntrack]    [k] ____nf_conntrack_find
+   5.97%   ksoftirqd/0  [nf_conntrack]    [k] ____nf_conntrack_find
+   2.47%   ksoftirqd/2  [nf_conntrack]    [k] hash_conntrack_raw
+   2.45%   ksoftirqd/0  [nf_conntrack]    [k] hash_conntrack_raw
+   2.44%   ksoftirqd/1  [nf_conntrack]    [k] hash_conntrack_raw

May allow parallel lookup/insert/delete if the entry is hashed to
another slot.  With patch:

+  20.95%  ksoftirqd/0  [nf_conntrack] [k] ____nf_conntrack_find
+  20.50%  ksoftirqd/1  [nf_conntrack] [k] ____nf_conntrack_find
+  20.27%  ksoftirqd/2  [nf_conntrack] [k] ____nf_conntrack_find
+   5.76%  ksoftirqd/1  [nf_conntrack] [k] hash_conntrack_raw
+   5.39%  ksoftirqd/2  [nf_conntrack] [k] hash_conntrack_raw
+   5.35%  ksoftirqd/0  [nf_conntrack] [k] hash_conntrack_raw
+   2.00%  ksoftirqd/1  [kernel.kallsyms] [k] __rcu_read_unlock

Improved rx processing rate from ~35kpps to ~50 kpps.

Reviewed-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
10 years agocan: mcp251x: Fix regulators operation without CONFIG_REGULATOR
Alexander Shiyan [Fri, 14 Mar 2014 08:46:20 +0000 (12:46 +0400)]
can: mcp251x: Fix regulators operation without CONFIG_REGULATOR

If CONFIG_REGULATOR is not set, devm_regulator_get() returns NULL,
so use IS_ERR_OR_NULL() macro for checks.

Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
10 years agocan: populate netdev::dev_id for udev discrimination
Christopher R. Baker [Sat, 8 Mar 2014 16:00:20 +0000 (11:00 -0500)]
can: populate netdev::dev_id for udev discrimination

My objective is to be able to totally discriminate CAN ports on multi-port
cards via udev so as to rename them to semantically interesting/unique names
for my system (e.g., "ecuCAN" and "auxCAN" instead of "can0" and "can1").

The following patch assigns the dev_id field to match the channel number on all
multi-channel devices. I can only test my two-port Peak PCI card, but it works
as expected: ATTRS{dev_id} now expresses the port number and my udev rules now
unambiguously pick out and rename my individual CAN ports.

Signed-off-by: Christopher R. Baker <cbaker@rec.ri.cmu.edu>
Tested-by: Oliver Hartkopp <socketcan@hartkopp.net> [PEAK PCAN-USB pro and EMS PCMCIA]
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
10 years agocan: Unify MTU settings for CAN interfaces
Oliver Hartkopp [Fri, 7 Mar 2014 08:23:41 +0000 (09:23 +0100)]
can: Unify MTU settings for CAN interfaces

CAN interfaces only support MTU values of 16 (CAN 2.0) and 72 (CAN FD).
Setting the MTU to other values is pointless but it does not really hurt.
With the introduction of the CAN FD support in drivers/net/can a new
function to switch the MTU for CAN FD has been introduced.

This patch makes use of this can_change_mtu() function to check for correct
MTU settings also in legacy CAN (2.0) devices.

Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
10 years agoMerge branch 'napi_budget_zero'
David S. Miller [Sat, 15 Mar 2014 02:53:35 +0000 (22:53 -0400)]
Merge branch 'napi_budget_zero'

Eric W. Biederman says:

====================
Don't receive packets when the napi budget == 0

After reading through all 120 drivers supporting netpoll I have found 16
more that process at least received packet when the napi budget == 0.

Processing more packets than your budget has always been a bug but
we haven't cared before so it looks like these drivers slipped through,
and need fixes.

As netpoll will shortly be using a budget of 0 to get the tx queue
processing with the rx queue processing we now care.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:11:22 +0000 (18:11 -0700)]
sfc: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agovxge: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:10:50 +0000 (18:10 -0700)]
vxge: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotc35815: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:10:14 +0000 (18:10 -0700)]
tc35815: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotilepro: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:09:01 +0000 (18:09 -0700)]
tilepro: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotilegx: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:08:21 +0000 (18:08 -0700)]
tilegx: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agos2io: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:06:26 +0000 (18:06 -0700)]
s2io: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agomlx4: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:05:58 +0000 (18:05 -0700)]
mlx4: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosky2: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:05:26 +0000 (18:05 -0700)]
sky2: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoibmveth: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:03:50 +0000 (18:03 -0700)]
ibmveth: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agofs_enet: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:03:23 +0000 (18:03 -0700)]
fs_enet: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoenic: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:02:08 +0000 (18:02 -0700)]
enic: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoamd8111e: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:01:11 +0000 (18:01 -0700)]
amd8111e: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoixgbe: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:00:41 +0000 (18:00 -0700)]
ixgbe: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoigb: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 01:00:06 +0000 (18:00 -0700)]
igb: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 00:59:10 +0000 (17:59 -0700)]
i40e: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Don't receive packets when the napi budget == 0
Eric W. Biederman [Sat, 15 Mar 2014 00:57:59 +0000 (17:57 -0700)]
bnx2x: Don't receive packets when the napi budget == 0

Processing any incoming packets with a with a napi budget of 0
is incorrect driver behavior.

This matters as netpoll will shortly call drivers with a budget of 0
to avoid receive packet processing happening in hard irq context.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'cxgb4-next'
David S. Miller [Sat, 15 Mar 2014 02:44:19 +0000 (22:44 -0400)]
Merge branch 'cxgb4-next'

Hariprasad Shenai says:

====================
Doorbell drop Avoidance Bug fix for iw_cxgb4

This patch series provides fixes for Chelsio T4/T5 adapters
related to DB Drop avoidance and other small fix related to keepalive on
iw-cxgb4.

The patches series is created against David Miller's 'net-next' tree.
And includes patches on cxgb4 and iw_cxgb4 driver.

We would like to request this patch series to get merged via David Miller's
'net-next' tree.

We have included all the maintainers of respective drivers. Kindly review the
change and let us know in case of any review comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agocxgb4/iw_cxgb4: Doorbell Drop Avoidance Bug Fixes
Steve Wise [Fri, 14 Mar 2014 16:22:08 +0000 (21:52 +0530)]
cxgb4/iw_cxgb4: Doorbell Drop Avoidance Bug Fixes

The current logic suffers from a slow response time to disable user DB
usage, and also fails to avoid DB FIFO drops under heavy load. This commit
fixes these deficiencies and makes the avoidance logic more optimal.
This is done by more efficiently notifying the ULDs of potential DB
problems, and implements a smoother flow control algorithm in iw_cxgb4,
which is the ULD that puts the most load on the DB fifo.

Design:

cxgb4:

Direct ULD callback from the DB FULL/DROP interrupt handler.  This allows
the ULD to stop doing user DB writes as quickly as possible.

While user DB usage is disabled, the LLD will accumulate DB write events
for its queues.  Then once DB usage is reenabled, a single DB write is
done for each queue with its accumulated write count.  This reduces the
load put on the DB fifo when reenabling.

iw_cxgb4:

Instead of marking each qp to indicate DB writes are disabled, we create
a device-global status page that each user process maps.  This allows
iw_cxgb4 to only set this single bit to disable all DB writes for all
user QPs vs traversing the idr of all the active QPs.  If the libcxgb4
doesn't support this, then we fall back to the old approach of marking
each QP.  Thus we allow the new driver to work with an older libcxgb4.

When the LLD upcalls iw_cxgb4 indicating DB FULL, we disable all DB writes
via the status page and transition the DB state to STOPPED.  As user
processes see that DB writes are disabled, they call into iw_cxgb4
to submit their DB write events.  Since the DB state is in STOPPED,
the QP trying to write gets enqueued on a new DB "flow control" list.
As subsequent DB writes are submitted for this flow controlled QP, the
amount of writes are accumulated for each QP on the flow control list.
So all the user QPs that are actively ringing the DB get put on this
list and the number of writes they request are accumulated.

When the LLD upcalls iw_cxgb4 indicating DB EMPTY, which is in a workq
context, we change the DB state to FLOW_CONTROL, and begin resuming all
the QPs that are on the flow control list.  This logic runs on until
the flow control list is empty or we exit FLOW_CONTROL mode (due to
a DB DROP upcall, for example).  QPs are removed from this list, and
their accumulated DB write counts written to the DB FIFO.  Sets of QPs,
called chunks in the code, are removed at one time. The chunk size is 64.
So 64 QPs are resumed at a time, and before the next chunk is resumed, the
logic waits (blocks) for the DB FIFO to drain.  This prevents resuming to
quickly and overflowing the FIFO.  Once the flow control list is empty,
the db state transitions back to NORMAL and user QPs are again allowed
to write directly to the user DB register.

The algorithm is designed such that if the DB write load is high enough,
then all the DB writes get submitted by the kernel using this flow
controlled approach to avoid DB drops.  As the load lightens though, we
resume to normal DB writes directly by user applications.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agocxgb4/iw_cxgb4: Treat CPL_ERR_KEEPALV_NEG_ADVICE as negative advice
Steve Wise [Fri, 14 Mar 2014 16:22:07 +0000 (21:52 +0530)]
cxgb4/iw_cxgb4: Treat CPL_ERR_KEEPALV_NEG_ADVICE as negative advice

Based on original work by Anand Priyadarshee <anandp@chelsio.com>.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: Replace u64_stats_fetch_begin_bh to u64_stats_fetch_begin_irq
Eric W. Biederman [Fri, 14 Mar 2014 04:26:42 +0000 (21:26 -0700)]
net: Replace u64_stats_fetch_begin_bh to u64_stats_fetch_begin_irq

Replace the bh safe variant with the hard irq safe variant.

We need a hard irq safe variant to deal with netpoll transmitting
packets from hard irq context, and we need it in most if not all of
the places using the bh safe variant.

Except on 32bit uni-processor the code is exactly the same so don't
bother with a bh variant, just have a hard irq safe variant that
everyone can use.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Sat, 15 Mar 2014 02:31:55 +0000 (22:31 -0400)]
Merge git://git./linux/kernel/git/davem/net

Conflicts:
drivers/net/usb/r8152.c
drivers/net/xen-netback/netback.c

Both the r8152 and netback conflicts were simple overlapping
changes.

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'alb_learning'
David S. Miller [Sat, 15 Mar 2014 02:21:04 +0000 (22:21 -0400)]
Merge branch 'alb_learning'

Veaceslav Falico says:

====================
bonding: use correct ether type for alb

There have been reports that, while using the ETH_P_LOOP ether type
(0x0060), the ether type is treated as its packet length.

To avoid that and to not break already existing apps - add new ether type
ETH_P_LOOPBACK that contains the correct id - 0x9000.
====================

Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
10 years agobonding: use the correct ether type for alb
Veaceslav Falico [Thu, 13 Mar 2014 11:41:58 +0000 (12:41 +0100)]
bonding: use the correct ether type for alb

Currently it's using the wrong ETH_P_LOOP type, which is sometimes treated
as packet length instead of ether type (because it's 0x0060).

Use the new ETH_P_LOOPBACK type.

CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoether: add loopback type ETH_P_LOOPBACK
Veaceslav Falico [Thu, 13 Mar 2014 11:41:57 +0000 (12:41 +0100)]
ether: add loopback type ETH_P_LOOPBACK

Per IEEE 802.3*, the correct packet type for loopback 0x9000. There's
already one ETH_P_LOOP 0x0060, which has been there for ages, however it's
plainly wrong as anything that small is considered a length field.

We can't remove it because legacy, so add a new type which corresponds to
the correct id.

http://www.iana.org/assignments/ieee-802-numbers/ieee-802-numbers.xhtml

CC: "David S. Miller" <davem@davemloft.net>
CC: Stefan Richter <stefanr@s5r6.in-berlin.de>
CC: Simon Wunderlich <sw@simonwunderlich.de>
CC: Neil Jerram <Neil.Jerram@metaswitch.com>
CC: Simon Horman <horms@verge.net.au>
CC: Arvid Brodin <Arvid.Brodin@xdin.com>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>