GitHub/moto-9609/android_kernel_motorola_exynos9610.git
9 years agorhashtable: Per bucket locks & deferred expansion/shrinking
Thomas Graf [Fri, 2 Jan 2015 22:00:20 +0000 (23:00 +0100)]
rhashtable: Per bucket locks & deferred expansion/shrinking

Introduces an array of spinlocks to protect bucket mutations. The number
of spinlocks per CPU is configurable and selected based on the hash of
the bucket. This allows for parallel insertions and removals of entries
which do not share a lock.

The patch also defers expansion and shrinking to a worker queue which
allows insertion and removal from atomic context. Insertions and
deletions may occur in parallel to it and are only held up briefly
while the particular bucket is linked or unzipped.

Mutations of the bucket table pointer is protected by a new mutex, read
access is RCU protected.

In the event of an expansion or shrinking, the new bucket table allocated
is exposed as a so called future table as soon as the resize process
starts.  Lookups, deletions, and insertions will briefly use both tables.
The future table becomes the main table after an RCU grace period and
initial linking of the old to the new table was performed. Optimization
of the chains to make use of the new number of buckets follows only the
new table is in use.

The side effect of this is that during that RCU grace period, a bucket
traversal using any rht_for_each() variant on the main table will not see
any insertions performed during the RCU grace period which would at that
point land in the future table. The lookup will see them as it searches
both tables if needed.

Having multiple insertions and removals occur in parallel requires nelems
to become an atomic counter.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agospinlock: Add spin_lock_bh_nested()
Thomas Graf [Fri, 2 Jan 2015 22:00:19 +0000 (23:00 +0100)]
spinlock: Add spin_lock_bh_nested()

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonft_hash: Remove rhashtable_remove_pprev()
Thomas Graf [Fri, 2 Jan 2015 22:00:18 +0000 (23:00 +0100)]
nft_hash: Remove rhashtable_remove_pprev()

The removal function of nft_hash currently stores a reference to the
previous element during lookup which is used to optimize removal later
on. This was possible because a lock is held throughout calling
rhashtable_lookup() and rhashtable_remove().

With the introdution of deferred table resizing in parallel to lookups
and insertions, the nftables lock will no longer synchronize all
table mutations and the stored pprev may become invalid.

Removing this optimization makes removal slightly more expensive on
average but allows taking the resize cost out of the insert and
remove path.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Cc: netfilter-devel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorhashtable: Factor out bucket_tail() function
Thomas Graf [Fri, 2 Jan 2015 22:00:17 +0000 (23:00 +0100)]
rhashtable: Factor out bucket_tail() function

Subsequent patches will require access to the bucket tail. Access
to the tail is relatively cheap as the automatic resizing of the
table should keep the number of entries per bucket to no more
than 0.75 on average.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorhashtable: Convert bucket iterators to take table and index
Thomas Graf [Fri, 2 Jan 2015 22:00:16 +0000 (23:00 +0100)]
rhashtable: Convert bucket iterators to take table and index

This patch is in preparation to introduce per bucket spinlocks. It
extends all iterator macros to take the bucket table and bucket
index. It also introduces a new rht_dereference_bucket() to
handle protected accesses to buckets.

It introduces a barrier() to the RCU iterators to the prevent
the compiler from caching the first element.

The lockdep verifier is introduced as stub which always succeeds
and properly implement in the next patch when the locks are
introduced.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorhashtable: Use rht_obj() instead of manual offset calculation
Thomas Graf [Fri, 2 Jan 2015 22:00:15 +0000 (23:00 +0100)]
rhashtable: Use rht_obj() instead of manual offset calculation

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agorhashtable: Do hashing inside of rhashtable_lookup_compare()
Thomas Graf [Fri, 2 Jan 2015 22:00:14 +0000 (23:00 +0100)]
rhashtable: Do hashing inside of rhashtable_lookup_compare()

Hash the key inside of rhashtable_lookup_compare() like
rhashtable_lookup() does. This allows to simplify the hashing
functions and keep them private.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Cc: netfilter-devel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'timecounter-next'
David S. Miller [Fri, 2 Jan 2015 21:47:51 +0000 (16:47 -0500)]
Merge branch 'timecounter-next'

Richard Cochran says:

====================
Fixing the "Time Counter fixes and improvements"

For this series I had only tested the build with ARCH=x86 and arm, but
others like sparc64, microblaze, powerpc, and s390 will fail because
they somehow don't indirectly include clocksource.h for the drivers in
question.

This series fixes the build issues reported by:
 kbuild test robot <fengguang.wu@intel.com>
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomicroblaze: include the new timecounter header.
Richard Cochran [Fri, 2 Jan 2015 19:22:09 +0000 (20:22 +0100)]
microblaze: include the new timecounter header.

The timecounter/cyclecounter code has moved, so users need the new include.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agomlx4: include clocksource.h again
Richard Cochran [Fri, 2 Jan 2015 19:22:08 +0000 (20:22 +0100)]
mlx4: include clocksource.h again

This driver uses the function, clocksource_khz2mult, and so it really must
include clocksource.h.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoixgbe: convert to CYCLECOUNTER_MASK macro.
Richard Cochran [Fri, 2 Jan 2015 19:22:07 +0000 (20:22 +0100)]
ixgbe: convert to CYCLECOUNTER_MASK macro.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoigb: convert to CYCLECOUNTER_MASK macro.
Richard Cochran [Fri, 2 Jan 2015 19:22:06 +0000 (20:22 +0100)]
igb: convert to CYCLECOUNTER_MASK macro.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoe1000e: convert to CYCLECOUNTER_MASK macro.
Richard Cochran [Fri, 2 Jan 2015 19:22:05 +0000 (20:22 +0100)]
e1000e: convert to CYCLECOUNTER_MASK macro.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agobnx2x: convert to CYCLECOUNTER_MASK macro.
Richard Cochran [Fri, 2 Jan 2015 19:22:04 +0000 (20:22 +0100)]
bnx2x: convert to CYCLECOUNTER_MASK macro.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotimecounter: provide a macro to initialize the cyclecounter mask field.
Richard Cochran [Fri, 2 Jan 2015 19:22:03 +0000 (20:22 +0100)]
timecounter: provide a macro to initialize the cyclecounter mask field.

There is no need for users of the timecounter/cyclecounter code to include
clocksource.h just for a single macro.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMAINTAINERS: Update Open vSwitch entry.
Pravin B Shelar [Fri, 2 Jan 2015 19:18:21 +0000 (11:18 -0800)]
MAINTAINERS: Update Open vSwitch entry.

OVS development is moved to netdev mailing list. Update tree and
list in MAINTAINERS file.

Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge tag 'wireless-drivers-next-for-davem-2015-01-02' of git://git.kernel.org/pub...
David S. Miller [Fri, 2 Jan 2015 21:40:44 +0000 (16:40 -0500)]
Merge tag 'wireless-drivers-next-for-davem-2015-01-02' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Changes:

* ath9k: enable Transmit Power Control (TPC) for ar9003 chips

* rtlwifi: cleanup and updates from the vendor driver

* rsi: fix memory leak related to firmware image

* ath: parameter fix for FCC DFS pattern

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: ethernet: cisco: enic: enic_dev: Remove some unused functions
Rickard Strandqvist [Fri, 2 Jan 2015 20:29:24 +0000 (21:29 +0100)]
net: ethernet: cisco: enic: enic_dev: Remove some unused functions

Removes some functions that are not used anywhere:
enic_dev_enable2_done() enic_dev_enable2() enic_dev_deinit_done()
enic_dev_init_prov2() enic_vnic_dev_deinit()

This was partially found by using a static code analysis program called cppcheck.

Signed-off-by: Rickard Strandqvist <rickard_strandqvist@spectrumdigital.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoisdn: hisax: hfc4s8s_l1: Remove some unused functions
Rickard Strandqvist [Thu, 1 Jan 2015 19:17:22 +0000 (20:17 +0100)]
isdn: hisax: hfc4s8s_l1: Remove some unused functions

Removes some functions that are not used anywhere:
Read_hfc32() Write_hfc32() Write_hfc16()

This was partially found by using a static code analysis program called cppcheck.

Signed-off-by: Rickard Strandqvist <rickard_strandqvist@spectrumdigital.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fddi: skfp: smt.c: Remove unused function
Rickard Strandqvist [Thu, 1 Jan 2015 17:01:26 +0000 (18:01 +0100)]
net: fddi: skfp: smt.c: Remove unused function

Remove the function smt_ifconfig() that is not used anywhere.

This was partially found by using a static code analysis program called cppcheck.

Signed-off-by: Rickard Strandqvist <rickard_strandqvist@spectrumdigital.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: ethernet: chelsio: cxgb3: mc5.c: Remove some unused functions
Rickard Strandqvist [Thu, 1 Jan 2015 16:49:55 +0000 (17:49 +0100)]
net: ethernet: chelsio: cxgb3: mc5.c: Remove some unused functions

Removes some functions that are not used anywhere:
dbgi_rd_rsp3() dbgi_wr_addr3()

This was partially found by using a static code analysis program called cppcheck.

Signed-off-by: Rickard Strandqvist <rickard_strandqvist@spectrumdigital.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: skbuff: don't zero tc members when freeing skb
Florian Westphal [Wed, 31 Dec 2014 12:33:41 +0000 (13:33 +0100)]
net: skbuff: don't zero tc members when freeing skb

Not needed, only four cases:
 - kfree_skb (or one of its aliases).
   Don't need to zero, memory will be freed.
 - kfree_skb_partial and head was stolen:  memory will be freed.
 - skb_morph:  The skb header fields (including tc ones) will be
   copied over from the 'to-be-morphed' skb right after
   skb_release_head_state returns.
 - skb_segment:  Same as before, all the skb header
   fields are copied over from the original skb right away.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetoot...
David S. Miller [Fri, 2 Jan 2015 20:58:21 +0000 (15:58 -0500)]
Merge branch 'for-upstream' of git://git./linux/kernel/git/bluetooth/bluetooth-next

Johan Hedberg say:

====================
pull request: bluetooth-next 2014-12-31

Here's the first batch of bluetooth patches for 3.20.

 - Cleanups & fixes to ieee802154  drivers
 - Fix synchronization of mgmt commands with respective HCI commands
 - Add self-tests for LE pairing crypto functionality
 - Remove 'BlueFritz!' specific handling from core using a new quirk flag
 - Public address configuration support for ath3012
 - Refactor debugfs support into a dedicated file
 - Initial support for LE Data Length Extension feature from Bluetooth 4.2

Please let me know if there are any issues pulling. Thanks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogeneve: Add Geneve GRO support
Joe Stringer [Wed, 31 Dec 2014 03:10:16 +0000 (19:10 -0800)]
geneve: Add Geneve GRO support

This results in an approximately 30% increase in throughput
when handling encapsulated bulk traffic.

Signed-off-by: Joe Stringer <joestringer@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: Add Transparent Ethernet Bridging GRO support.
Jesse Gross [Wed, 31 Dec 2014 03:10:15 +0000 (19:10 -0800)]
net: Add Transparent Ethernet Bridging GRO support.

Currently the only tunnel protocol that supports GRO with encapsulated
Ethernet is VXLAN. This pulls out the Ethernet code into a proper layer
so that it can be used by other tunnel protocols such as GRE and Geneve.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/fsl: remove hardcoded clock setting from xgmac_mdio
Shaohui Xie [Tue, 30 Dec 2014 08:28:21 +0000 (16:28 +0800)]
net/fsl: remove hardcoded clock setting from xgmac_mdio

There is no need to set the clock speed in read/write which will be performed
unnecessarily for each mdio access. Init it during probe is enough.

Also, the hardcoded clock value is not a proper way for all SoCs.

Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/fsl: remove irq assignment from xgmac_mdio
Shaohui Xie [Tue, 30 Dec 2014 08:28:00 +0000 (16:28 +0800)]
net/fsl: remove irq assignment from xgmac_mdio

Which is wrong and not used, so no extra space needed by
mdiobus_alloc_size(), use mdiobus_alloc() instead.

Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet/fsl: remove reset from xgmac_mdio
Shaohui Xie [Tue, 30 Dec 2014 08:27:33 +0000 (16:27 +0800)]
net/fsl: remove reset from xgmac_mdio

Since the reset is just clock setting, individual mdio reset is
not available.

Signed-off-by: Shaohui Xie <Shaohui.Xie@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'gmac-next'
David S. Miller [Thu, 1 Jan 2015 00:14:52 +0000 (19:14 -0500)]
Merge branch 'gmac-next'

Roger Chen says:

====================
support GMAC driver for RK3288

Roger Chen (6):
  patch1: add driver for Rockchip RK3288 SoCs integrated GMAC
  patch2: define clock ID used for GMAC
  patch3: modify CRU config for Rockchip RK3288 SoCs integrated GMAC
  patch4: dts: rockchip: add gmac info for rk3288
  patch5: dts: rockchip: enable gmac on RK3288 evb board
  patch6: add document for Rockchip RK3288 GMAC

Tested on rk3288 evb board:
Execute the following command to enable ethernet,
set local IP and ping a remote host.

busybox ifconfig eth0 up
busybox ifconfig eth0 192.168.1.111
ping 192.168.1.1
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoGMAC: add document for Rockchip RK3288 GMAC
Roger Chen [Mon, 29 Dec 2014 09:44:34 +0000 (17:44 +0800)]
GMAC: add document for Rockchip RK3288 GMAC

The document descripts how to add properties for GMAC in device tree.

change since v2:

1. remove power-gpio, reset-gpio, phyirq-gpio, pmu_regulator setting
2. add "snps,reset-gpio", "snps,reset-active-low;" "snps,reset-delays-us"

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoARM: dts: rockchip: enable gmac on RK3288 evb board
Roger Chen [Mon, 29 Dec 2014 09:44:25 +0000 (17:44 +0800)]
ARM: dts: rockchip: enable gmac on RK3288 evb board

enable gmac in rk3288-evb-rk808.dts

changes since v2:
1. add fixed regulator for PHY
2. remove power-gpio, reset-gpio, phyirq-gpio, pmu_regulator setting
3. add "snps,reset-gpio", "snps,reset-active-low;" "snps,reset-delays-us"

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoARM: dts: rockchip: add gmac info for rk3288
Roger Chen [Mon, 29 Dec 2014 09:44:16 +0000 (17:44 +0800)]
ARM: dts: rockchip: add gmac info for rk3288

add gmac info in rk3288.dtsi for GMAC driver

changes since v2:
1. add drive-strength in the pinctrl settings

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoGMAC: modify CRU config for Rockchip RK3288 SoCs integrated GMAC
Roger Chen [Mon, 29 Dec 2014 09:44:07 +0000 (17:44 +0800)]
GMAC: modify CRU config for Rockchip RK3288 SoCs integrated GMAC

modify CRU config for GMAC driver

changes since v2:
1. remove SCLK_MAC_PLL

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoGMAC: define clock ID used for GMAC
Roger Chen [Mon, 29 Dec 2014 09:43:55 +0000 (17:43 +0800)]
GMAC: define clock ID used for GMAC

changes since v2:
1. remove SCLK_MAC_PLL

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoGMAC: add driver for Rockchip RK3288 SoCs integrated GMAC
Roger Chen [Mon, 29 Dec 2014 09:43:36 +0000 (17:43 +0800)]
GMAC: add driver for Rockchip RK3288 SoCs integrated GMAC

This driver is based on stmmac driver.

changes since v2:
- use tab instead of space for macros
- use HIWORD_UPDATE macro for GMAC_CLK_RX_DL_CFG and GMAC_CLK_TX_DL_CFG
- remove drive-strength setting in the driver and set it in the pinctrl settings
- use dev_err instead of pr_err
- remove clock names's macros, just use the real name of the clock
- use devm_clk_get() instead of clk_get()
- remove clk_set_parent(bsp_priv->clk_mac, bsp_priv->clk_mac_pll)
- remove gpio setting for LDO, just use regulator API
- remove phy reset using gpio in the glue layer, it has been handled in the stmmac driver
- remove handling phy interrupt (mii interrupt)

changes since v1:
- use BIT() to set register
- combine two remap_write() operations into one for the same register
- use macros for register value setting
- remove grf fail check in rk_gmac_setup() and save all the check in set_rgmii_speed()
- remove .tx_coe=1 in rk_gmac_data

Signed-off-by: Roger Chen <roger.chen@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoigb_ptp: Include clocksource.h to get CLOCKSOURCE_MASK.
David S. Miller [Wed, 31 Dec 2014 23:32:40 +0000 (18:32 -0500)]
igb_ptp: Include clocksource.h to get CLOCKSOURCE_MASK.

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoe1000e: Include clocksource.h to get CLOCKSOURCE_MASK.
David S. Miller [Wed, 31 Dec 2014 23:32:25 +0000 (18:32 -0500)]
e1000e: Include clocksource.h to get CLOCKSOURCE_MASK.

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'fib_trie-next'
David S. Miller [Wed, 31 Dec 2014 23:26:02 +0000 (18:26 -0500)]
Merge branch 'fib_trie-next'

Alexander Duyck says:

====================
fib_trie: Reduce time spent in fib_table_lookup by 35 to 75%

These patches are meant to address several performance issues I have seen
in the fib_trie implementation, and fib_table_lookup specifically.  With
these changes in place I have seen a reduction of up to 35 to 75% for the
total time spent in fib_table_lookup depending on the type of search being
performed.

On a VM running in my Corei7-4930K system with a trie of maximum depth of 7
this resulted in a reduction of over 370ns per packet in the total time to
process packets received from an ixgbe interface and route them to a dummy
interface.  This represents a failed lookup in the local trie followed by
a successful search in the main trie.

Baseline Refactor
  ixgbe->dummy routing 1.20Mpps 2.21Mpps
  ------------------------------------------------------------
  processing time per packet 835ns 453ns
  fib_table_lookup 50.1% 418ns 25.0% 113ns
  check_leaf.isra.9  7.9%  66ns    --  --
  ixgbe_clean_rx_irq  5.3%  44ns  9.8%  44ns
  ip_route_input_noref  2.9%  25ns  4.6%  21ns
  pvclock_clocksource_read  2.6%  21ns  4.6%  21ns
  ip_rcv  2.6%  22ns  4.0%  18ns

In the simple case of receiving a frame and dropping it before it can reach
the socket layer I saw a reduction of 40ns per packet.  This represents a
trip through the local trie with the correct leaf found with no need for
any backtracing.

Baseline Refactor
  ixgbe->local receive 2.65Mpps 2.96Mpps
  ------------------------------------------------------------
  processing time per packet 377ns 337ns
  fib_table_lookup 25.1%  95ns 25.8%  87ns
  ixgbe_clean_rx_irq  8.7%  33ns  9.0%  30ns
  check_leaf.isra.9  7.2%  27ns    --  --
  ip_rcv  5.7%  21ns  6.5%  22ns

These changes have resulted in several functions being inlined such as
check_leaf and fib_find_node, but due to the code simplification the
overall size of the code has been reduced.

   text    data     bss     dec     hex filename
  16932     376      16   17324    43ac net/ipv4/fib_trie.o - before
  15259     376       8   15643    3d1b net/ipv4/fib_trie.o - after

Changes since RFC:
  Replaced this_cpu_ptr with correct call to this_cpu_inc in patch 1
  Changed test for leaf_info mismatch to (key ^ n->key) & li->mask_plen in patch 10
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Add tracking value for suffix length
Alexander Duyck [Wed, 31 Dec 2014 18:57:08 +0000 (10:57 -0800)]
fib_trie: Add tracking value for suffix length

This change adds a tracking value for the maximum suffix length of all
prefixes stored in any given tnode.  With this value we can determine if we
need to backtrace or not based on if the suffix is greater than the pos
value.

By doing this we can reduce the CPU overhead for lookups in the local table
as many of the prefixes there are 32b long and have a suffix length of 0
meaning we can immediately backtrace to the root node without needing to
test any of the nodes between it and where we ended up.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Remove checks for index >= tnode_child_length from tnode_get_child
Alexander Duyck [Wed, 31 Dec 2014 18:57:02 +0000 (10:57 -0800)]
fib_trie: Remove checks for index >= tnode_child_length from tnode_get_child

For some reason the compiler doesn't seem to understand that when we are in
a loop that runs from tnode_child_length - 1 to 0 we don't expect the value
of tn->bits to change.  As such every call to tnode_get_child was rerunning
tnode_chile_length which ended up consuming quite a bit of space in the
resultant assembly code.

I have gone though and verified that in all cases where tnode_get_child
is used we are either winding though a fixed loop from tnode_child_length -
1 to 0, or are in a fastpath case where we are verifying the value by
either checking for any remaining bits after shifting index by bits and
testing for leaf, or by using tnode_child_length.

size net/ipv4/fib_trie.o
Before:
   text    data     bss     dec     hex filename
  15506     376       8   15890    3e12 net/ipv4/fib_trie.o

After:
   text    data     bss     dec     hex filename
  14827     376       8   15211    3b6b net/ipv4/fib_trie.o

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: inflate/halve nodes in a more RCU friendly way
Alexander Duyck [Wed, 31 Dec 2014 18:56:55 +0000 (10:56 -0800)]
fib_trie: inflate/halve nodes in a more RCU friendly way

This change pulls the node_set_parent functionality out of put_child_reorg
and instead leaves that to the function to take care of as well.  By doing
this we can fully construct the new cluster of tnodes and all of the
pointers out of it before we start routing pointers into it.

I am suspecting this will likely fix some concurency issues though I don't
have a good test to show as such.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Push tnode flushing down to inflate/halve
Alexander Duyck [Wed, 31 Dec 2014 18:56:49 +0000 (10:56 -0800)]
fib_trie: Push tnode flushing down to inflate/halve

This change pushes the tnode freeing down into the inflate and halve
functions.  It makes more sense here as we have a better grasp of what is
going on and when a given cluster of nodes is ready to be freed.

I believe this may address a bug in the freeing logic as well.  For some
reason if the freelist got to a certain size we would call
synchronize_rcu().  I'm assuming that what they meant to do is call
synchronize_rcu() after they had handed off that much memory via
call_rcu().  As such that is what I have updated the behavior to be.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Push assignment of child to parent down into inflate/halve
Alexander Duyck [Wed, 31 Dec 2014 18:56:43 +0000 (10:56 -0800)]
fib_trie: Push assignment of child to parent down into inflate/halve

This change makes it so that the assignment of the tnode to the parent is
handled directly within whatever function is currently handling the node be
it inflate, halve, or resize.  By doing this we can avoid some of the need
to set NULL pointers in the tree while we are resizing the subnodes.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Add functions should_inflate and should_halve
Alexander Duyck [Wed, 31 Dec 2014 18:56:37 +0000 (10:56 -0800)]
fib_trie: Add functions should_inflate and should_halve

This change pulls the logic for if we should inflate/halve the nodes out
into separate functions.  It also addresses what I believe is a bug where 1
full node is all that is needed to keep a node from ever being halved.

Simple script to reproduce the issue:
modprobe dummy; ifconfig dummy0 up
for i in `seq 0 255`; do ifconfig dummy0:$i 10.0.${i}.1/24 up; done
ifconfig dummy0:256 10.0.255.33/16 up
for i in `seq 0 254`; do ifconfig dummy0:$i down; done

Results from /proc/net/fib_triestat
Before:
Local:
Aver depth:     3.00
Max depth:      4
Leaves:         17
Prefixes:       18
Internal nodes: 11
  1: 8  2: 2  10: 1
Pointers: 1048
Null ptrs: 1021
Total size: 11  kB
After:
Local:
Aver depth:     3.41
Max depth:      5
Leaves:         17
Prefixes:       18
Internal nodes: 12
  1: 8  2: 3  3: 1
Pointers: 36
Null ptrs: 8
Total size: 3  kB

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Move resize to after inflate/halve
Alexander Duyck [Wed, 31 Dec 2014 18:56:31 +0000 (10:56 -0800)]
fib_trie: Move resize to after inflate/halve

This change consists of a cut/paste of resize to behind inflate and halve
so that I could remove the two function prototypes.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Push rcu_read_lock/unlock to callers
Alexander Duyck [Wed, 31 Dec 2014 18:56:24 +0000 (10:56 -0800)]
fib_trie: Push rcu_read_lock/unlock to callers

This change is to start cleaning up some of the rcu_read_lock/unlock
handling.  I realized while reviewing the code there are several spots that
I don't believe are being handled correctly or are masking warnings by
locally calling rcu_read_lock/unlock instead of calling them at the correct
level.

A common example is a call to fib_get_table followed by fib_table_lookup.
The rcu_read_lock/unlock ought to wrap both but there are several spots where
they were not wrapped.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Use unsigned long for anything dealing with a shift by bits
Alexander Duyck [Wed, 31 Dec 2014 18:56:18 +0000 (10:56 -0800)]
fib_trie: Use unsigned long for anything dealing with a shift by bits

This change makes it so that anything that can be shifted by, or compared
to a value shifted by bits is updated to be an unsigned long.  This is
mostly a precaution against an insanely huge address space that somehow
starts coming close to the 2^32 root node size which would require
something like 1.5 billion addresses.

I chose unsigned long instead of unsigned long long since I do not believe
it is possible to allocate a 32 bit tnode on a 32 bit system as the memory
consumed would be 16GB + 28B which exceeds the addressible space for any
one process.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Update meaning of pos to represent unchecked bits
Alexander Duyck [Wed, 31 Dec 2014 18:56:12 +0000 (10:56 -0800)]
fib_trie: Update meaning of pos to represent unchecked bits

This change moves the pos value to the other side of the "bits" field.  By
doing this it actually simplifies a significant amount of code in the trie.

For example when halving a tree we know that the bit lost exists at
oldnode->pos, and if we inflate the tree the new bit being add is at
tn->pos.  Previously to find those bits you would have to subtract pos and
bits from the keylength or start with a value of (1 << 31) and then shift
that.

There are a number of spots throughout the code that benefit from this.  In
the case of the hot-path searches the main advantage is that we can drop 2
or more operations from the search path as we no longer need to compute the
value for the index to be shifted by and can instead just use the raw pos
value.

In addition the tkey_extract_bits is now defunct and can be replaced by
get_index since the two operations were doing the same thing, but now
get_index does it much more quickly as it is only an xor and shift versus a
pair of shifts and a subtraction.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Optimize fib_table_insert
Alexander Duyck [Wed, 31 Dec 2014 18:56:06 +0000 (10:56 -0800)]
fib_trie: Optimize fib_table_insert

This patch updates the fib_table_insert function to take advantage of the
changes made to improve the performance of fib_table_lookup.  As a result
the code should be smaller and run faster then the original.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Optimize fib_find_node
Alexander Duyck [Wed, 31 Dec 2014 18:56:00 +0000 (10:56 -0800)]
fib_trie: Optimize fib_find_node

This patch makes use of the same features I made use of for
fib_table_lookup to streamline fib_find_node.  The resultant code should be
smaller and run faster than the original.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Optimize fib_table_lookup to avoid wasting time on loops/variables
Alexander Duyck [Wed, 31 Dec 2014 18:55:54 +0000 (10:55 -0800)]
fib_trie: Optimize fib_table_lookup to avoid wasting time on loops/variables

This patch is meant to reduce the complexity of fib_table_lookup by reducing
the number of variables to the bare minimum while still keeping the same if
not improved functionality versus the original.

Most of this change was started off by the desire to rid the function of
chopped_off and current_prefix_length as they actually added very little to
the function since they only applied when computing the cindex.  I was able
to replace them mostly with just a check for the prefix match.  As long as
the prefix between the key and the node being tested was the same we know
we can search the tnode fully versus just testing cindex 0.

The second portion of the change ended up being a massive reordering.
Originally the calls to check_leaf were up near the start of the loop, and
the backtracing and descending into lower levels of tnodes was later.  This
didn't make much sense as the structure of the tree means the leaves are
always the last thing to be tested.  As such I reordered things so that we
instead have a loop that will delve into the tree and only exit when we
have either found a leaf or we have exhausted the tree.  The advantage of
rearranging things like this is that we can fully inline check_leaf since
there is now only one reference to it in the function.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Merge leaf into tnode
Alexander Duyck [Wed, 31 Dec 2014 18:55:47 +0000 (10:55 -0800)]
fib_trie: Merge leaf into tnode

This change makes it so that leaf and tnode are the same struct.  As a
result there is no need for rt_trie_node anymore since everyting can be
merged into tnode.

On 32b systems this results in the leaf being 4 bytes larger, however I
don't know if that is really an issue as this and an eariler patch that
added bits & pos have increased the size from 20 to 28.  If I am not
mistaken slub/slab allocate on power of 2 sizes so 20 was likely being
rounded up to 32 anyway.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Merge tnode_free and leaf_free into node_free
Alexander Duyck [Wed, 31 Dec 2014 18:55:41 +0000 (10:55 -0800)]
fib_trie: Merge tnode_free and leaf_free into node_free

Both the leaf and the tnode had an rcu_head in them, but they had them in
slightly different places.  Since we now have them in the same spot and
know that any node with bits == 0 is a leaf and the rest are either vmalloc
or kmalloc tnodes depending on the value of bits it makes it easy to combine
the functions and reduce overhead.

In addition I have taken advantage of the rcu_head pointer to go ahead and
put together a simple linked list instead of using the tnode pointer as
this way we can merge either type of structure for freeing.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Make leaf and tnode more uniform
Alexander Duyck [Wed, 31 Dec 2014 18:55:35 +0000 (10:55 -0800)]
fib_trie: Make leaf and tnode more uniform

This change makes some fundamental changes to the way leaves and tnodes are
constructed.  The big differences are:
1.  Leaves now populate pos and bits indicating their full key size.
2.  Trie nodes now mask out their lower bits to be consistent with the leaf
3.  Both structures have been reordered so that rt_trie_node now consisists
    of a much larger region including the pos, bits, and rcu portions of
    the tnode structure.

On 32b systems this will result in the leaf being 4B larger as the pos and
bits values were added to a hole created by the key as it was only 4B in
length.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agofib_trie: Update usage stats to be percpu instead of global variables
Alexander Duyck [Wed, 31 Dec 2014 18:55:29 +0000 (10:55 -0800)]
fib_trie: Update usage stats to be percpu instead of global variables

The trie usage stats were currently being shared by all threads that were
calling fib_table_lookup.  As a result when multiple threads were
performing lookups simultaneously the trie would begin to cache bounce
between those threads.

In order to prevent this I have updated the usage stats to use a set of
percpu variables.  By doing this we should be able to avoid the cache
bouncing and still make use of these stats.

Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agogre: allow live address change
stephen hemminger [Sat, 27 Dec 2014 18:01:42 +0000 (10:01 -0800)]
gre: allow live address change

The GRE tap device supports Ethernet over GRE, but doesn't
care about the source address of the tunnel, therefore it
can be changed without bring device down.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agol2tp : multicast notification to the registered listeners
Bill Hong [Sat, 27 Dec 2014 18:12:39 +0000 (10:12 -0800)]
l2tp : multicast notification to the registered listeners

Previously l2tp module did not provide any means for the user space to
get notified when tunnels/sessions are added/modified/deleted.
This change contains the following
- create a multicast group for the listeners to register.
- notify the registered listeners when the tunnels/sessions are
  created/modified/deleted.

Signed-off-by: Bill Hong <bhong@brocade.com>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Sven-Thorsten Dietrich <sven@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotun: return proper error code from tun_do_read
Alex Gartrell [Fri, 26 Dec 2014 07:22:49 +0000 (23:22 -0800)]
tun: return proper error code from tun_do_read

Instead of -1 with EAGAIN, read on a O_NONBLOCK tun fd will return 0.  This
fixes this by properly returning the error code from __skb_recv_datagram.

Signed-off-by: Alex Gartrell <agartrell@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotun: Fixed unsigned/signed comparison
Alex Gartrell [Fri, 26 Dec 2014 07:05:03 +0000 (23:05 -0800)]
tun: Fixed unsigned/signed comparison

Validated that this was actually using the unsigned comparison with gdb.

Signed-off-by: Alex Gartrell <agartrell@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotipc: replace 0 by NULL for pointers
Fabian Frederick [Thu, 25 Dec 2014 11:05:50 +0000 (12:05 +0100)]
tipc: replace 0 by NULL for pointers

Fix sparse warning:
net/tipc/link.c:1924:40: warning: Using plain integer as NULL pointer

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'enic-next'
David S. Miller [Wed, 31 Dec 2014 18:08:54 +0000 (13:08 -0500)]
Merge branch 'enic-next'

Govindarajulu Varadarajan says:

====================
enic: Check for DMA mapping error

After dma mapping the buffers, enic does not call dma_mapping_error() to check
if mapping is successful.

This series fixes the issue by checking return value of pci_dma_mapping_error()
after pci_map_single().

This is reported by redhat here
https://bugzilla.redhat.com/show_bug.cgi?id=1145016
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoenic: add stats for dma mapping error
Govindarajulu Varadarajan [Wed, 24 Dec 2014 10:29:38 +0000 (15:59 +0530)]
enic: add stats for dma mapping error

This patch adds generic statistics for enic. As of now dma_map_error is the only
member. dma_map_erro is incremented every time dma maping error happens.

Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoenic: check dma_mapping_error
Govindarajulu Varadarajan [Wed, 24 Dec 2014 10:29:37 +0000 (15:59 +0530)]
enic: check dma_mapping_error

This patch checks for pci_dma_mapping_error() after dma mapping the data.
If the dma mapping fails we remove the previously queued frags and return
NETDEV_TX_OK.

Reported-by: Jan Stancek <jstancek@redhat.com>
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoenic: make vnic_wq_buf doubly linked
Govindarajulu Varadarajan [Wed, 24 Dec 2014 10:29:36 +0000 (15:59 +0530)]
enic: make vnic_wq_buf doubly linked

This patch makes vnic_wq_buf doubly liked list. This is needed for dma_mapping
error check, in case some frag's dma map fails, we need to move back and remove
previously queued buffers.

Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'fec-next'
David S. Miller [Wed, 31 Dec 2014 18:06:57 +0000 (13:06 -0500)]
Merge branch 'fec-next'

Fugang Duan says:

====================
net: fec: add Wake-on-LAN support

The patch series enable FEC Wake-on-LAN feature for i.MX6q/dl and i.MX6SX SOCs.
FEC HW support sleep mode, when system in suspend status with FEC all clock gate
off, magic packet can wake up system. For different SOCs, there have special SOC
GPR register to let FEC enter sleep mode or exit sleep mode, add these to platform
callback for driver' call.

Patch#1: add WOL interface supports.
Patch#2: add SOC special sleep of/off operations for driver's sleep callback.
Patch#3: add magic pattern support for devicetree.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoARM: dts: imx6qdl: enable FEC magic-packet feature
Nimrod Andy [Wed, 24 Dec 2014 09:30:41 +0000 (17:30 +0800)]
ARM: dts: imx6qdl: enable FEC magic-packet feature

Add FEC magic-packet feature support.

Signed-off-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoARM: imx: add FEC sleep mode callback function
Nimrod Andy [Wed, 24 Dec 2014 09:30:40 +0000 (17:30 +0800)]
ARM: imx: add FEC sleep mode callback function

i.MX6q/dl, i.MX6SX SOCs enet support sleep mode that magic packet can
wake up system in suspend status. For different SOCs, there have some
SOC specifical GPR register to set sleep on/off mode. So add these to
callback function for driver.

Signed-off-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fec: add Wake-on-LAN support
Nimrod Andy [Wed, 24 Dec 2014 09:30:39 +0000 (17:30 +0800)]
net: fec: add Wake-on-LAN support

Support for Wake-on-LAN using Magic Packet. ENET IP supports sleep mode
in low power status, when system enter suspend status, Magic packet can
wake up system even if all SOC clocks are gate. The patch doing below things:
- flagging the device as a wakeup source for the system, as well as
  its Wake-on-LAN interrupt
- prepare the hardware for entering WoL mode
- add standard ethtool WOL interface
- enable the ENET interrupt to wake us

Tested on i.MX6q/dl sabresd, sabreauto boards, i.MX6SX arm2 boards.

Signed-off-by: Fugang Duan <B38611@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: gianfar: add missing __iomem annotation
Kevin Hao [Wed, 24 Dec 2014 06:05:45 +0000 (14:05 +0800)]
net: gianfar: add missing __iomem annotation

Fix the following spare warning:
drivers/net/ethernet/freescale/gianfar.c:3521:60: warning: incorrect type in argument 1 (different address spaces)
drivers/net/ethernet/freescale/gianfar.c:3521:60:    expected unsigned int [noderef] <asn:2>*addr
drivers/net/ethernet/freescale/gianfar.c:3521:60:    got unsigned int [usertype] *rfbptr
drivers/net/ethernet/freescale/gianfar.c:205:16: warning: incorrect type in assignment (different address spaces)
drivers/net/ethernet/freescale/gianfar.c:205:16:    expected unsigned int [usertype] *rfbptr
drivers/net/ethernet/freescale/gianfar.c:205:16:    got unsigned int [noderef] <asn:2>*<noident>
drivers/net/ethernet/freescale/gianfar.c:2918:44: warning: incorrect type in argument 1 (different address spaces)
drivers/net/ethernet/freescale/gianfar.c:2918:44:    expected unsigned int [noderef] <asn:2>*addr
drivers/net/ethernet/freescale/gianfar.c:2918:44:    got unsigned int [usertype] *rfbptr

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: gianfar: mark the local functions static
Kevin Hao [Wed, 24 Dec 2014 06:05:44 +0000 (14:05 +0800)]
net: gianfar: mark the local functions static

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agovirtio-net: don't do header check for dodgy gso packets
Jason Wang [Wed, 24 Dec 2014 03:03:52 +0000 (11:03 +0800)]
virtio-net: don't do header check for dodgy gso packets

There's no need to do header check for virtio-net since:

- Host sets dodgy for all gso packets from guest and check the header.
- Host should be prepared for all kinds of evil packets from guest, since
  malicious guest can send any kinds of packet.

So this patch sets NETIF_F_GSO_ROBUST for virtio-net to skip the check.

Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoarm: sa1100: move irda header to linux/platform_data
Dmitry Eremin-Solenikov [Tue, 23 Dec 2014 22:14:14 +0000 (01:14 +0300)]
arm: sa1100: move irda header to linux/platform_data

In the end asm/mach/irda.h header is not used by anybody except sa1100.
Move the header to the platform data includes dir and rename it to
irda-sa11x0.h.

Signed-off-by: Dmitry Eremin-Solenikov <dbaryshkov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: sxgbe: Use setup_timer
Julia Lawall [Fri, 26 Dec 2014 14:35:46 +0000 (15:35 +0100)]
net: sxgbe: Use setup_timer

Convert a call to init_timer and accompanying intializations of
the timer's data and function fields to a call to setup_timer.

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

// <smpl>
@@
expression t,f,d;
@@

-init_timer(&t);
+setup_timer(&t,f,d);
-t.function = f;
-t.data = d;
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoksz884x: Use setup_timer
Julia Lawall [Fri, 26 Dec 2014 14:35:42 +0000 (15:35 +0100)]
ksz884x: Use setup_timer

Convert a call to init_timer and accompanying intializations of
the timer's data and function fields to a call to setup_timer.

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

// <smpl>
@@
expression t,f,d;
@@

-init_timer(&t);
+setup_timer(&t,f,d);
-t.function = f;
-t.data = d;
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoatl1e: Use setup_timer
Julia Lawall [Fri, 26 Dec 2014 14:35:35 +0000 (15:35 +0100)]
atl1e: Use setup_timer

Convert a call to init_timer and accompanying intializations of
the timer's data and function fields to a call to setup_timer.

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

// <smpl>
@@
expression t,f,d;
@@

-init_timer(&t);
+setup_timer(&t,f,d);
-t.function = f;
-t.data = d;
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoatheros: atlx: Use setup_timer
Julia Lawall [Fri, 26 Dec 2014 14:35:34 +0000 (15:35 +0100)]
atheros: atlx: Use setup_timer

Convert a call to init_timer and accompanying intializations of
the timer's data and function fields to a call to setup_timer.

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

// <smpl>
@@
expression t,f,d;
@@

-init_timer(&t);
+setup_timer(&t,f,d);
-t.function = f;
-t.data = d;
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'timecounter'
David S. Miller [Tue, 30 Dec 2014 23:30:00 +0000 (18:30 -0500)]
Merge branch 'timecounter'

Richard Cochran says:

====================
Time Counter fixes and improvements

Several PTP Hardware Clock (PHC) drivers implement the clock in
software using the timecounter/cyclecounter code. This series adds one
simple improvement and one more subtle fix to the shared timecounter
facility. Credit for this series goes to Janusz Użycki, who pointed
the issues out to me off list.

Patch #1 simply move the timecounter code into its own file. When
working on this series, it was really annoying to see half the kernel
recompile after every tweak to the timecounter stuff. There is no
reason to keep this together with the clocksource code.

Patch #2 implements an improved adjtime() method, and patches 3-10
convert all of the drivers over to the new method.

Patch #11 fixes a subtle but important issue with the timecounter WRT
frequency adjustment. As it stands now, a timecounter based PHC will
exhibit a variable frequency resolution (and variable time error)
depending on how often the clock is read.

In timecounter_read_delta(), the expression

   (delta * cc->mult) >> cc->shift;

can lose resolution from the adjusted value of 'mult'. If the value
of 'delta' is too small, then small changes in 'mult' have no effect.
However, if the delta value is large enough, then small changes in
'mult' will have an effect.

Reading the clock too often means smaller 'delta' values which in turn
will spoil the fine adjustments made to 'mult'. Up until now, this
effect did not show up in my testing. The following example explains
why.

The CPTS has an input clock of 250 MHz, and the clock source uses
mult=0x80000000 and shift=29, making the ticks to nanoseconds
conversion like this:

   ticks * 2^31
   ------------
       2^29

Imagine what happens if the clock is read every 10 milliseconds. Ten
milliseconds are about 2500000 ticks, which corresponds to about 21
bits. The product in the numerator has then 52 bits. After the shift
operation, 23 bits are preserved. This results in a frequency
adjustment resolution of about 0.1 ppm (not _too_ bad.)

A frequency resolution of 1 ppm requires 20 bits.
A frequency resolution of 1 ppb requires 30 bits.

For the 250 MHz CPTS clock, reading every 4 seconds yields a 1 ppb
resolution (which is the finest that our API allows).

However, the error can be much higher if the clock is read too often
or if time stamps occur close in time to read operations. In general
it is really not acceptable to allow the rate of clock readings to
influence the clock accuracy.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotimecounter: keep track of accumulated fractional nanoseconds
Richard Cochran [Sun, 21 Dec 2014 18:47:06 +0000 (19:47 +0100)]
timecounter: keep track of accumulated fractional nanoseconds

The current timecounter implementation will drop a variable amount
of resolution, depending on the magnitude of the time delta. In
other words, reading the clock too often or too close to a time
stamp conversion will introduce errors into the time values. This
patch fixes the issue by introducing a fractional nanosecond field
that accumulates the low order bits.

Reported-by: Janusz Użycki <j.uzycki@elproma.com.pl>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: cpts: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:47:05 +0000 (19:47 +0100)]
net: cpts: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: mlx4: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:47:04 +0000 (19:47 +0100)]
net: mlx4: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Compile tested only.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: ixgbe: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:47:03 +0000 (19:47 +0100)]
net: ixgbe: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Compile tested only.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: igb: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:47:02 +0000 (19:47 +0100)]
net: igb: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: e1000e: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:47:01 +0000 (19:47 +0100)]
net: e1000e: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: fec: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:47:00 +0000 (19:47 +0100)]
net: fec: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Compile tested only.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: bnx2x: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:46:59 +0000 (19:46 +0100)]
net: bnx2x: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Compile tested only.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agonet: xgbe: convert to timecounter adjtime.
Richard Cochran [Sun, 21 Dec 2014 18:46:58 +0000 (19:46 +0100)]
net: xgbe: convert to timecounter adjtime.

This patch changes the driver to use the new and improved method
for adjusting the offset of a timecounter.

Compile tested only.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotimecounter: provide a helper function to shift the time.
Richard Cochran [Sun, 21 Dec 2014 18:46:57 +0000 (19:46 +0100)]
timecounter: provide a helper function to shift the time.

Some PTP Hardware Clock drivers use a struct timecounter to represent
their clock. To adjust the time by a given offset, these drivers all
perform a two step read/write of their timecounter. However, it is
better and simpler just to adjust the offset in one step. This patch
introduces a little routine to help drivers implement the adjtime
method.

Suggested-by: Janusz Użycki <j.uzycki@elproma.com.pl>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agotime: move the timecounter/cyclecounter code into its own file.
Richard Cochran [Sun, 21 Dec 2014 18:46:56 +0000 (19:46 +0100)]
time: move the timecounter/cyclecounter code into its own file.

The timecounter code has almost nothing to do with the clocksource
code. Let it live in its own file. This will help isolate the
timecounter users from the clocksource users in the source tree.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 30 Dec 2014 18:45:47 +0000 (10:45 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix double SKB free in bluetooth 6lowpan layer, from Jukka Rissanen.

 2) Fix receive checksum handling in enic driver, from Govindarajulu
    Varadarajan.

 3) Fix NAPI poll list corruption in virtio_net and caif_virtio, from
    Herbert Xu.  Also, add code to detect drivers that have this mistake
    in the future.

 4) Fix doorbell endianness handling in mlx4 driver, from Amir Vadai.

 5) Don't clobber IP6CB() before xfrm6_policy_check() is called in TCP
    input path,f rom Nicolas Dichtel.

 6) Fix MPLS action validation in openvswitch, from Pravin B Shelar.

 7) Fix double SKB free in vxlan driver, also from Pravin.

 8) When we scrub a packet, which happens when we are switching the
    context of the packet (namespace, etc.), we should reset the
    secmark.  From Thomas Graf.

 9) ->ndo_gso_check() needs to do more than return true/false, it also
    has to allow the driver to clear netdev feature bits in order for
    the caller to be able to proceed properly.  From Jesse Gross.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (62 commits)
  genetlink: A genl_bind() to an out-of-range multicast group should not WARN().
  netlink/genetlink: pass network namespace to bind/unbind
  ne2k-pci: Add pci_disable_device in error handling
  bonding: change error message to debug message in __bond_release_one()
  genetlink: pass multicast bind/unbind to families
  netlink: call unbind when releasing socket
  netlink: update listeners directly when removing socket
  genetlink: pass only network namespace to genl_has_listeners()
  netlink: rename netlink_unbind() to netlink_undo_bind()
  net: Generalize ndo_gso_check to ndo_features_check
  net: incorrect use of init_completion fixup
  neigh: remove next ptr from struct neigh_table
  net: xilinx: Remove unnecessary temac_property in the driver
  net: phy: micrel: use generic config_init for KSZ8021/KSZ8031
  net/core: Handle csum for CHECKSUM_COMPLETE VXLAN forwarding
  openvswitch: fix odd_ptr_err.cocci warnings
  Bluetooth: Fix accepting connections when not using mgmt
  Bluetooth: Fix controller configuration with HCI_QUIRK_INVALID_BDADDR
  brcmfmac: Do not crash if platform data is not populated
  ipw2200: select CFG80211_WEXT
  ...

9 years agoMerge tag 'linux-kselftest-3.19-fixes' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Tue, 30 Dec 2014 18:36:59 +0000 (10:36 -0800)]
Merge tag 'linux-kselftest-3.19-fixes' of git://git./linux/kernel/git/shuah/linux-kselftest

Pull kselftest fix from Shuah Khan:
 "Fix exec test compile warnings"

* tag 'linux-kselftest-3.19-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
  selftests/exec: Use %zu to format size_t

10 years agoBluetooth: Add timing information to ECDH test case runs
Marcel Holtmann [Tue, 30 Dec 2014 08:11:21 +0000 (00:11 -0800)]
Bluetooth: Add timing information to ECDH test case runs

After successful completion of the ECDH test cases, print the time it
took to run them.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
10 years agoBluetooth: Add timing information to SMP test case runs
Marcel Holtmann [Tue, 30 Dec 2014 08:11:20 +0000 (00:11 -0800)]
Bluetooth: Add timing information to SMP test case runs

After successful completion of the SMP test cases, print the time it
took to run them.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
10 years agoBluetooth: Add LE Secure Connections tests for SMP
Johan Hedberg [Tue, 30 Dec 2014 07:50:41 +0000 (09:50 +0200)]
Bluetooth: Add LE Secure Connections tests for SMP

This patch adds SMP self-tests for the Secure Connections crypto
functions. The sample data has been taken from the core specification.

Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
10 years agoBluetooth: Add legacy SMP tests
Johan Hedberg [Tue, 30 Dec 2014 07:50:40 +0000 (09:50 +0200)]
Bluetooth: Add legacy SMP tests

This patch adds self-tests for legacy SMP crypto functions. The sample
data has been taken from the core specification.

Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
10 years agoBluetooth: Add skeleton for SMP self-tests
Johan Hedberg [Tue, 30 Dec 2014 07:50:39 +0000 (09:50 +0200)]
Bluetooth: Add skeleton for SMP self-tests

This patch adds the initial skeleton and kernel config option for SMP
self-tests.

Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
10 years agoBluetooth: Add support for ECDH test cases
Johan Hedberg [Tue, 30 Dec 2014 04:48:36 +0000 (20:48 -0800)]
Bluetooth: Add support for ECDH test cases

This patch adds the test cases for ECDH cryptographic functionality
used by Bluetooth Low Energy Secure Connections feature.

Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
10 years agoBluetooth: Add support for self testing framework
Marcel Holtmann [Tue, 30 Dec 2014 04:48:35 +0000 (20:48 -0800)]
Bluetooth: Add support for self testing framework

This add support for the Bluetooth self testing framework that allows
running certain test cases of sample data to ensure correctness of its
basic functionality.

With this patch only the basic framework will be added. It contains
the build magic that allows running this at module loading time or
at late_initcall stage when built into the kernel image.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
10 years agoBluetooth: Fix const declarations for smp_f5 and smp_f6
Johan Hedberg [Mon, 29 Dec 2014 10:04:10 +0000 (12:04 +0200)]
Bluetooth: Fix const declarations for smp_f5 and smp_f6

These SMP crypto functions should have all their input parameters
declared as const. This patch fixes the parameters that were missing the
const declaration.

Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
10 years agocc2520: fix zero perm_extended_addr address
Varka Bhadram [Fri, 26 Dec 2014 04:43:40 +0000 (10:13 +0530)]
cc2520: fix zero perm_extended_addr address

It will remove the bug of havine zero perm_extended_addr address.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
10 years agocc2520: remove 'ret' goto label
Varka Bhadram [Fri, 26 Dec 2014 04:43:39 +0000 (10:13 +0530)]
cc2520: remove 'ret' goto label

If allocation of memory fails instead of going to ret goto label
and returning from there, we can directly return -ENOMEM on failure.

Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>