brcmfmac: Only handle p2p_stop_device if vif is valid

In some situations it is possible that vif has been removed while
cfg80211 invokes the p2p_stop_device handler. This will result in
crash.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Deleting of p2p device is leaking memory.

When a p2p device gets deleted, the memory for the vif is not being
released. This is solved by reorganizing the cleanup path and
properly freeing the memory.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Workaround in change vif for wpa_supplicant support.

Different wpa_supplicants have different behavior and expectations
regarding the change_virtual_intf behavior. This patch implements
a workaround for the different versions and possible brcmfmac
configuration.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Make p2pon module param always available.

p2pon module param is currently under define BRCMDBG. Though it is
a needed option for older versions of the wpa_supplicant which do not
support the P2P_DEVICE interface.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Fix bug in flowring management.

The hash index stored in the flowrings is of type u16 but gets
stored in u8. This can result in incorrect indexing and possibly
result in crashes. This patch fixes the type.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Inform p2p module about p2pon through API

When the p2pon module param is used then p2p attach will initialize
p2p device iface in the firmware, but it is doing that by checking
data. It is cleaner to pass the p2pon information to p2p by API.
This information is also needed for other patch.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Only assign primary netdev to if2bss array.

The if2bss allows for translation of ifidx to bssidx which has a 1:n
relation. Therefor only the first (primary) netdev should be
assigned in this array. This fixes the p2pon=1 module param usage.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Fix set and get tx-power functions.

Implementation of tx-power (get and set) related functions are
still assuming mW interface. This is wrong as functions use dbm
(or mbm) nowadays. As a result a tx power configuration could
result in wrong power configuration.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Add support for the BCM4350 PCIE device.

This patch adds support fo the BRCM4350 2x2 11ac PCIE device.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Fix exception handling.

In some exception situations the ifp->vif was not properly assigned
which could result in crash.

Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: complete usb tx data with multi endpoints

This patch do the work to TX data with specific USB
endpoint. At the same time, update data_sent flag
according to multi port status. And is_port_ready
API is added for BSS interface to check if current
used usb data endpoint is available or not.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: add usb multi endpoints resync support

This patch add support for usb multi endpoints resync.
Once multi channel event is received from firmware,
update usb_mc_setp flag to block TX data until setup is over.
And available data endpoint will be attached to BSS interface.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: add usb tx data multi endpoints support

This patch add support for USB interface to TX data
with different endpoint. And previous TX information
are saved in new designed structure.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: add usb multi channel event process support

This patch add multi channel event process for USB multi tx data endpoints.
Driver receives firmware multi channel event only in case that new
connection is setup or interface is disconnect. Different BSS interface
need update used usb endpoint.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: move usb specific data_sent update to usb.c

This patch move data_sent flag update to usb.c file.
>From now all data_sent update cases only happened in specific file:
sdio.c, usb.c, pcie.c. Outside ot these files, it is only allowed
to check the value of data_ent.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: fix tx data_sent issue for usb interface

This patch fix missing tx data_sent flag update for usb interface.
Except USB interface, data_sent flag has been updated in specific
file such as sdio.c and pcie.c. So only USB interface type need check
when TX data completed.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: Suppress -ENOSR error for data traffic on USB

We have preallocated buffer pool for Tx data. During high
data traffic, all buffers are submitted to USB and driver needs
to wait until one of the buffers get available for next Tx
packet.

"data: -ENOSR is returned" errors is expected in this case. Let's
lower the priority of this message.

Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: correct paused tx data packet counter

This patch fixes observed issues while updating counter for
number of paused data packets in wmm queue when trying to
delete packet or delete station entry.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: avoid memsetting PCIe event buffer

Preallocated PCIe buffer is being reused for all PCIe interface
events. Physical address of the buffer is shared with firmware
so that it can perform DMA on it. As event length is specified
in the header, there should not be a problem if the buffer gets
overwritten.
We will save some cycles by avoiding memset everytime while
submitting the buffer to firmware.

Fixes: 2728cecdc7d6bf3d21(mwifiex: corrections in PCIe event skb)
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: fix NULL pointer dereference during hidden SSID scan

This NULL pointer dereference is observed during suspend resume
stress test. All pending commands are cancelled when system goes
into suspend state. There a corner case in which host may receive
response for last scan command after this and try to trigger extra
active scan for hidden SSIDs.

The issue is fixed by adding a NULL check to skip that extra scan.

Fixes: 2375fa2b36feaf34 (mwifiex: fix unable to connect hidden SSID..)
Cc: <stable@vger.kernel.org> [v4.2+]
Signed-off-by: Aniket Nagarnaik <aniketn@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: don't always include ht/vht info in tdls confirm frame

Current TDLS implementation always includes ht/vht information
in tdls setup confirm frame which causes teardown by legacy
peer station after TDLS handshake. We will inclue ht/vht
capabilities in tdls setup confirm frame only if peer station
supports it to fix this problem.

Signed-off-by: Aniket Nagarnaik <aniketn@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: NULL check for cfg80211_inform_bss()

cfg80211_inform_bss would return null in some cases, such as
memory allocation failure.

This patch adds sanity check for this case, to avoid possible
issues when above corner case is hit.

Signed-off-by: Xinming Hu <huxm@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: fix driver init failure under memory pressure

64k Tx and Rx buffers are allocated during driver initialization
for SDIO level data aggregations. When host is under memory
pressure situation, kzalloc() request for 64k may fail.

We will try allocating 32k buffers and disable our rx single port
aggreagation feature in this situation.

If the allocation still fails, we will disable our sdio multport
aggregation feature as well. In this way, we will transmit and
receive packets one by one, thus reduce the demand for big
memory.

Signed-off-by: Xinming Hu <huxm@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: minor corrections in multiport aggregation code

This patch includes below changes
1) Check if multiport aggregation feature flag while doing rx
length validation check.
2) Reset buffer size variables when buffers are freed.

Signed-off-by: Xinming Hu <huxm@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: update amsdu tx packet time stamp

This patch fixes the issue of delay time in A-MSDU tx packet.
In generated new A-MSDU packet, the time stamp is initialized to zero.
Choose the time of first MSDU packet as aggregated packet's time.

Signed-off-by: Zhaoyang Liu <liuzy@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: fix typo in del_virtual_intf() function

Interface counters should be incremented in add_virtual_intf()
and decremented in del_virtual_intf() function. This patch
corrects it.

Signed-off-by: Xinming Hu <huxm@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

wlcore: align reg_ch_conf_last[] to 64bit

Align fields reg_ch_conf_last and reg_ch_conf_pending of
struct wl1271{} to 64bit.

Without this, on 64bit ARM, wlcore_set_pending_regdomain_ch() fails at
the point it calls set_bit(ch_bit_idx, (long*)wl->reg_ch_conf_pending);
Here is the error message while doing iw wlan0 scan or connect:

[   10.666857] wlcore: IRQ work
[   10.670046] wlcore: intr: 0x40 (fw_rx_counter = 1, drv_rx_counter = 0, tx_results_counter = 0)
[   10.678697] wlcore: WL1271_ACX_INTR_DATA
[   10.682810] Unhandled fault: alignment fault (0x96000021) at 0xffffffc037a817f4
[   10.690139] Internal error: : 96000021 [#1] PREEMPT SMP
[   10.695366] Modules linked in:
[   10.698437] CPU: 3 PID: 894 Comm: irq/60-wl18xx Tainted: G        W       4.2.0-rc6-linaro-hikey #2
[   10.707501] Hardware name: HiKey Development Board (DT)
[   10.712733] task: ffffffc03a9d1680 ti: ffffffc039e18000 task.ti: ffffffc039e18000
[   10.720239] PC is at set_bit+0x14/0x30
[   10.724002] LR is at wlcore_set_pending_regdomain_ch+0x40/0x4c

Signed-off-by: Guodong Xu <guodong.xu@linaro.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

ath9k: Add support for OCB mode

The patch adds support for "outside the context of a BSS"(OCB) mode
to ath9k driver and extends debugfs files by OCB ralated information.

This patch was tested on AR9380-AL1A cards.

Signed-off-by: Jan Kaisrlik <kaisrja1@fel.cvut.cz>
Cc: Michal Sojka <sojkam1@fel.cvut.cz>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

carl9170: fix bad rssi reading

Fix rssi calculation error which was introduced in otus to ar9170
porting.

Signed-off-by: Hiroaki KAWAI <hiroaki.kawai@gmail.com>
Acked-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

net/wireless/wl18xx: Add missing MODULE_FIRMWARE

Fixes the output of 'modinfo --field firmware'.

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

ath9k: enable hw manual peak calibration for QCA9561

This patch fix https://lists.openwrt.org/pipermail/openwrt-devel/
2015-August/034979.html. As the peak detect calibration is set
incorrectly.

Signed-off-by: Miaoqing Pan <miaoqing@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

mwifiex: Make mwifiex_dbg a function, reduce object size

The mwifiex_dbg macro has two tests that could be consolidated
into a function reducing overall object size ~10KB (~4%).

So convert the macro into a function.

$ size drivers/net/wireless/mwifiex/built-in.o* (x86-64 defconfig)
   text    data     bss     dec     hex filename
 233102    8628    4809  246539   3c30b drivers/net/wireless/mwifiex/built-in.o.new
 243949    8628    4809  257386   3ed6a drivers/net/wireless/mwifiex/built-in.o.old

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: Reset PCIE devices after recognition.

When PCIE type devices are being FW reloaded without being properly
reset then the device ends up in a locked state, requiring the
device to be completely powered down. This patch adds a reset
through watchdog at the moment the device (cores) has been
recognized. This will solve warm reboot issues.

Cc: Rafal Milecki <zajec5@gmail.com>
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: introduce brcmf_net_detach() function

In case of error during brcmf_bus_start() the network interfaces were
freed using free_netdev(). However, the interfaces may have additional
memory allocated which is not freed. The netdev has destructor set to
brcmf_cfg80211_free_netdev() which frees the additional memory if
allocated and call free_netdev(). The brcmf_net_detach() either calls
brcmf_cfg80211_free_netdev() directly or uses unregister_netdev() when
struct net_device::reg_state indicates the netdev was registered.

Reported-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: change prototype for brcmf_fws_hdrpull()

Instead of passing ifidx and drvr just pass struct brcmf_if pointer
which holds both parameters.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: remove ifidx parameter from brcmf_fws_txstatus_suppressed()

The brcmf_fws_txstatus_suppressed() function prototype specifies an
ifidx parameter which is not used within the function implementation.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: add dedicated debug level for firmware console logging

Both PCIe and SDIO devices have the possibility to log the firmware
console output in kernel log. For PCIe it is logged when PCIE debug
level is enabled. For SDIO it is logged when user specifies a non-zero
console interval through debugfs. This patch tries to make it a
bit more consistent. The firmware console output is only logged when
FWCON debug level is enabled.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Reviewed-by: Pontus Fuchs <pontusf@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: add mapping for interface index to bsscfg index

Because the P2P Device interface in firmware uses the same interface
index as the primary interface we use the bsscfg index as index in the
struct brcmf_pub::iflist. However, in the data path we get the interface
index and not the bsscfg index. So we need a mapping of interface index
to bsscfg index, which can be determined upon handle adding the interface.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: pass struct brcmf_if instance in brcmf_txfinalize()

Most call sites of brcmf_txfinalize already have struct brcmf_if
instance so pass that to brcmf_txfinalize() as the function
needs it anyway.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: use brcmf_get_ifp() to map ifidx to struct brcmf_if instance

The knowledge on how to map the interface index to a struct brcmf_if
instance is in brcmf_get_ifp() so use that function when only the
interface index is known instead of accessing brcmf_pub::iflist
directly.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: correct detection of p2pdev interface event

The p2pdev interface is setup in firmware resulting in a interface
event. This event has role and no-if flag. When role is p2p client
and no-if flag is set it indicates that this is the p2pdev interface.
This info is used in handling the event and adding interface in the
driver.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: only call brcmf_cfg80211_detach() when attach was successful

In brcmf_bus_start() the function brcmf_cfg80211_attach() is called which
may fail. If this happens we should not call brcmf_cfg80211_detach() in
the failure path as it will result in NULL pointer dereference:

  brcmf_fweh_activate_events: Set event_msgs error (-5)
  brcmf_bus_start: failed: -5
  brcmf_sdio_firmware_callback: dongle is not responding
  BUG: unable to handle kernel NULL pointer dereference at 0000000000000068
  IP: [<ffffffff811e8f08>] kernfs_find_ns+0x18/0xd0
  PGD 0
  Oops: 0000 [#1] SMP
  Modules linked in: brcmfmac(O) brcmutil(O) cfg80211 auth_rpcgss
  CPU: 1 PID: 45 Comm: kworker/1:1 Tainted: G           O
  Hardware name: Dell Inc. Latitude E6410/07XJP9, BIOS A07 02/15/2011
  Workqueue: events request_firmware_work_func
  task: ffff880036c09ac0 ti: ffff880036dd4000 task.ti: ffff880036dd4000
  RIP: 0010:[<ffffffff811e8f08>]  [<ffffffff811e8f08>] kernfs_find_ns+0x18/0xd0
  RSP: 0018:ffff880036dd7a28  EFLAGS: 00010246
  RAX: ffff880036c09ac0 RBX: 0000000000000000 RCX: 000000007fffffff
  RDX: 0000000000000000 RSI: ffffffff816578b9 RDI: 0000000000000000
  RBP: ffff880036dd7a48 R08: 0000000000000000 R09: ffff880036c0b340
  R10: 00000000000002ec R11: ffff880036dd7b08 R12: ffffffff816578b9
  R13: 0000000000000000 R14: ffffffff816578b9 R15: ffff8800c6c87000
  FS:  0000000000000000(0000) GS:ffff88012bc40000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
  CR2: 0000000000000068 CR3: 0000000001a0b000 CR4: 00000000000006e0
  Stack:
   0000000000000000 ffffffff816578b9 0000000000000000 ffff8800c0d003c8
   ffff880036dd7a78 ffffffff811e8ff5 0000000ffffffff1 ffffffff81a9b060
   ffff8800c789f880 ffff8800c0d00000 ffff880036dd7a98 ffffffff811ebe0d
  Call Trace:
   [<ffffffff811e8ff5>] kernfs_find_and_get_ns+0x35/0x60
   [<ffffffff811ebe0d>] sysfs_unmerge_group+0x1d/0x60
   [<ffffffff81404ef2>] dpm_sysfs_remove+0x22/0x60
   [<ffffffff813f9db9>] device_del+0x49/0x240
   [<ffffffff815da768>] rfkill_unregister+0x58/0xc0
   [<ffffffffa06bd91b>] wiphy_unregister+0xab/0x2f0 [cfg80211]
   [<ffffffffa0742fe3>] brcmf_cfg80211_detach+0x23/0x50 [brcmfmac]
   [<ffffffffa074d986>] brcmf_detach+0x86/0xe0 [brcmfmac]
   [<ffffffffa0757de8>] brcmf_sdio_remove+0x48/0x120 [brcmfmac]
   [<ffffffffa0758ed9>] brcmf_sdiod_remove+0x29/0xd0 [brcmfmac]
   [<ffffffffa0759031>] brcmf_ops_sdio_remove+0xb1/0x110 [brcmfmac]
   [<ffffffffa001c267>] sdio_bus_remove+0x37/0x100 [mmc_core]
   [<ffffffff813fe026>] __device_release_driver+0x96/0x130
   [<ffffffff813fe0e3>] device_release_driver+0x23/0x30
   [<ffffffffa0754bc8>] brcmf_sdio_firmware_callback+0x2a8/0x5d0 [brcmfmac]
   [<ffffffffa074deaf>] brcmf_fw_request_nvram_done+0x15f/0x5e0 [brcmfmac]
   [<ffffffff8140142f>] ? devres_add+0x3f/0x50
   [<ffffffff810642b5>] ? usermodehelper_read_unlock+0x15/0x20
   [<ffffffff81400000>] ? platform_match+0x70/0xa0
   [<ffffffff8140f400>] request_firmware_work_func+0x30/0x60
   [<ffffffff8106828c>] process_one_work+0x14c/0x3d0
   [<ffffffff8106862a>] worker_thread+0x11a/0x450
   [<ffffffff81068510>] ? process_one_work+0x3d0/0x3d0
   [<ffffffff8106d692>] kthread+0xd2/0xf0
   [<ffffffff8106d5c0>] ? kthread_create_on_node+0x180/0x180
   [<ffffffff815ed35f>] ret_from_fork+0x3f/0x70
   [<ffffffff8106d5c0>] ? kthread_create_on_node+0x180/0x180
  Code: e9 40 fe ff ff 48 89 d8 eb 87 66 0f 1f 84 00 00 00 00 00 66 66 66 66
90 55 48 89 e5 41 56 49 89 f6 41 55 49 89 d5 31 d2 41 54 53 <0f> b7
47 68 48 8b 5f 48 66 c1 e8 05 83 e0 01 4d 85 ed 0f b6 c8
  RIP  [<ffffffff811e8f08>] kernfs_find_ns+0x18/0xd0
   RSP <ffff880036dd7a28>
  CR2: 0000000000000068
  ---[ end trace 87d6ec0d3fe46740 ]---

Reported-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: change parameters for brcmf_remove_interface()

Just pass the interface to be removed, ie. the struct brcmf_if instance.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: make brcmf_proto_hdrpull() return struct brcmf_if instance

Avoid spreading the ifidx in the driver, but have it return the
struct brcmf_if instance.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

brcmfmac: consolidate ifp lookup in driver core

In rx path the firmware provide an interface index which is used to
map to a struct brcmf_if instance. However, this involves some trick
that is done in two places. This is changed by having driver core
providing brcmf_get_ifp() function.

Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

Merge tag 'iwlwifi-next-for-kalle-2015-09-21' of git://git./linux/kernel/git/iwlwifi/iwlwifi-next

* some debugfs improvements;
* fix signedness in beacon statistics;
* deinline some functions to reduce size when device tracing is enabled;
* filter beacons out in AP mode when no stations are associated;
* deprecate firmwares version -12;
* fix a runtime PM vs. legacy suspend race;
* one-liner fix for a ToF bug;
* clean-ups in the rx code;
* small debugging improvement;
* fix WoWLAN with new firmware versions;

iwlwifi: mvm: add debug print for d0i3 exit indication

In order to verify d0i3 flow, add debug print to indicate
d0i3 exit was completed (right after tx was re-enabled),
along with the wakeup reasons.

Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: configure wowlan configuration only if connected

Recent fw version added assert to make sure wowlan configuration
is configured only when a station is connected.

Change the driver behavior to pass this configuration only
if we indeed have ap station id (i.e. connected).

Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: move RX API into its own file

The RX API is currently mixed up into the general fw-api.h
file, but we're going to need to extend it significantly in
the future, so move it to its own file.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove some unused defines from RX API

Remove some unused values from the RX API; these were used
with older firmware API that didn't have the RX energy API,
support for which was removed a long time ago.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove pointless cfg_phy_cnt length check

Since the driver can never configure the data here, this field
will always be reported as 0 by the firmware. Even if this was
not the case, however, it wouldn't matter since the extra data
would be added beyond the end of the phy_info structure we use
in the driver, so wouldn't harm anything in this code either.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove useless debug message from RX

This message is useless - it's in the good case that always
happens so enabling it doesn't really help. Just remove it.
There are other ways to debug this (e.g. tracing) so there's
no need to add a message in the bad case.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: make sure AP is operating for ToF

It's possible for an AP interface to be UP but not actually
operating (i.e. not beaconing etc.) - in this case it can't
actually do ToF, so check for it.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove IWL_UCODE_TLV_API_STATS_V10 TLV flag

This flag is set in all supported firmwares.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luciano Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove IWL_UCODE_TLV_API_ASYNC_DTM TLV flag

This flag is set in all supported firmwares.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luciano Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove IWL_UCODE_TLV_API_SINGLE_SCAN_EBS TLV flag

All the supported firmwares have this flag set.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luciano Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove IWL_UCODE_TLV_API_TX_POWER_DEV TLV flag

All the supported firmwares use the new API.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luciano Coelho <luciano.coelho@intel.com>

iwlwifi: mvm: remove IWL_UCODE_TLV_API_HDC_PHASE_0 TLV flag

All the supported firwmares support the new API.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luciano Coelho <luciano.coelho@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

ath9k_htc: introduce support for different fw versions

Current kernel support only one fw name with theoretically only one
fw version located in “firmware/htc_[9271|7010].fw”. Which is ok so far we
have only one fw version (1.3). After we realised new fw 1.4, we faced
compatibility problem which was decided to solve by firmware name and
location:
- new firmware is located now in
firmware/ath9k_htc/htc_[9271|7010]-1.4.0.fw
- old version 1.3 should be on old place, so old kernel have no issues
with it.
- new kernels including this patch should be able to try different
supported (min..max) fw version.
- new kernel should be able to support old fw location too. At least for
now.

At same time this patch will add new module option which should allow user
to play with development  fw version without replacing stable one. If user
will set “ath9k_htc use_dev_fw=1” module will try to find
firmware/ath9k_htc/htc_[9271|7010]-1.dev.0.fw first and if it fails, use
stable version: for example...1.4.0.fw.

Signed-off-by: Oleksij Rempel <linux@rempel-privat.de>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>

sch_dsmark: improve memory locality

Memory placement in sch_dsmark is silly : Better place mask/value
in the same cache line.

Also, we can embed small arrays in the first cache line and
remove a potential cache miss.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'bcmgenet-irq-coalesce'

Florian Fainelli says:

====================
net: bcmgenet: Interrupt coalescing

This patch series adds support for interrupt coalescing for GENET
adapters.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: Implement RX coalescing control knobs

Add support for the ethtool rx-frames coalescing parameter which allows
defining the number of RX interrupts per frames received. The RDMA
engine supports a configurable timeout with a resolution of
approximately 8.192 us.

We can no longer enable the BDONE/PDONE interrupts as those would
fire for each packet/buffer received, which would defeat the MBDONE
interrupt purpose. The MBDONE interrupt is guaranteed to correspond to a
PDONE/BDONE interrupt when the threshold is set to 1.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: bcmgenet: Implement TX coalescing control knobs

Configuring the ethtool tx-frames property, which translates into N
packets before a TX interrupt is the simplest configuration scheme
because it requires no locking neither at the softare nor hardware
level, and is completely indepedent from the link speed. Since ethtool
does not allow per-tx queue coalescing parameters, we apply the same
setting to any transmit queue.

We can no longer enable the BDONE/PDONE interrupts as those would fire
for each packet/buffer received, which would defeat the MBDONE interrupt
purpose. The MBDONE interrupt is guaranteed to correspond to a
PDONE/BDONE interrupt when the threshold is set to 1, but offers
interrupt coalescing when the value is > 1.

Since the HW is configured to generate an interrupt when the ring
becomes emtpy, we have to deny any timeout/timer settings coming from
user-space to indicate we can only generate an interrupt very <N>
packets.

While we are at it, fix the DMA_INTR_THRESHOLD_MASK value which was off
by one bit (0xff vs. 0x1ff).

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

lan78xx: Remove not defined MAC_CR_GMII_EN_ bit from MAC_CR.

Remove not defined MAC_CR_GMII_EN_ bit from MAC_CR.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

lan78xx: Create lan78xx_get_mdix_status() and lan78xx_set_mdix_status() for MDIX control.

Create lan78xx_get_mdix_status() and lan78xx_set_mdix_status() for MDIX control.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

lan78xx: Remove phy defines in lan78xx.h and use defines in include/linux/microchipphy.h

Remove phy defines in lan78xx.h and use defines in include/linux/microchipphy.h.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

lan78xx: Update to use phylib instead of mii_if_info.

Update to use phylib instead of mii_if_info.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

lan78xx: Add PHYLIB and MICROCHIP_PHY as default config.

Add PHYLIB and MICROCHIP_PHY as default configuration for lan78xx.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

lan78xx: Check device ready bit (PMT_CTL_READY_) after reset the PHY

Check device ready bit (PMT_CTL_READY_) after reset the PHY.
Device may not be ready even if PHY_RST_ is cleared depends on configuration.

Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Initialize table in fib result

Sergey, Richard and Fabio reported an oops in ip_route_input_noref. e.g., from Richard:

[    0.877040] BUG: unable to handle kernel NULL pointer dereference at 0000000000000056
[    0.877597] IP: [<ffffffff8155b5e2>] ip_route_input_noref+0x1a2/0xb00
[    0.877597] PGD 3fa14067 PUD 3fa6e067 PMD 0
[    0.877597] Oops: 0000 [#1] SMP
[    0.877597] Modules linked in: virtio_net virtio_pci virtio_ring virtio
[    0.877597] CPU: 1 PID: 119 Comm: ifconfig Not tainted 4.2.0+ #1
[    0.877597] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[    0.877597] task: ffff88003fab0bc0 ti: ffff88003faa8000 task.ti: ffff88003faa8000
[    0.877597] RIP: 0010:[<ffffffff8155b5e2>]  [<ffffffff8155b5e2>] ip_route_input_noref+0x1a2/0xb00
[    0.877597] RSP: 0018:ffff88003ed03ba0  EFLAGS: 00010202
[    0.877597] RAX: 0000000000000046 RBX: 00000000ffffff8f RCX: 0000000000000020
[    0.877597] RDX: ffff88003fab50b8 RSI: 0000000000000200 RDI: ffffffff8152b4b8
[    0.877597] RBP: ffff88003ed03c50 R08: 0000000000000000 R09: 0000000000000000
[    0.877597] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88003fab6f00
[    0.877597] R13: ffff88003fab5000 R14: 0000000000000000 R15: ffffffff81cb5600
[    0.877597] FS:  00007f6de5751700(0000) GS:ffff88003ed00000(0000) knlGS:0000000000000000
[    0.877597] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    0.877597] CR2: 0000000000000056 CR3: 000000003fa6d000 CR4: 00000000000006e0
[    0.877597] Stack:
[    0.877597]  0000000000000000 0000000000000046 ffff88003fffa600 ffff88003ed03be0
[    0.877597]  ffff88003f9e2c00 697da8c0017da8c0 ffff880000000000 000000000007fd00
[    0.877597]  0000000000000000 0000000000000046 0000000000000000 0000000400000000
[    0.877597] Call Trace:
[    0.877597]  <IRQ>
[    0.877597]  [<ffffffff812bfa1f>] ? cpumask_next_and+0x2f/0x40
[    0.877597]  [<ffffffff8158e13c>] arp_process+0x39c/0x690
[    0.877597]  [<ffffffff8158e57e>] arp_rcv+0x13e/0x170
[    0.877597]  [<ffffffff8151feec>] __netif_receive_skb_core+0x60c/0xa00
[    0.877597]  [<ffffffff81515795>] ? __build_skb+0x25/0x100
[    0.877597]  [<ffffffff81515795>] ? __build_skb+0x25/0x100
[    0.877597]  [<ffffffff81521ff6>] __netif_receive_skb+0x16/0x70
[    0.877597]  [<ffffffff81522078>] netif_receive_skb_internal+0x28/0x90
[    0.877597]  [<ffffffff8152288f>] napi_gro_receive+0x7f/0xd0
[    0.877597]  [<ffffffffa0017906>] virtnet_receive+0x256/0x910 [virtio_net]
[    0.877597]  [<ffffffffa0017fd8>] virtnet_poll+0x18/0x80 [virtio_net]
[    0.877597]  [<ffffffff815234cd>] net_rx_action+0x1dd/0x2f0
[    0.877597]  [<ffffffff81053228>] __do_softirq+0x98/0x260
[    0.877597]  [<ffffffff8164969c>] do_softirq_own_stack+0x1c/0x30

The root cause is use of res.table uninitialized.

Thanks to Nikolay for noticing the uninitialized use amongst the maze of
gotos.

As Nikolay pointed out the second initialization is not required to fix
the oops, but rather to fix a related problem where a valid lookup should
be invalidated before creating the rth entry.

Fixes: b7503e0cdb5d ("net: Add FIB table id to rtable")
Reported-by: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Reported-by: Richard Alpe <richard.alpe@ericsson.com>
Reported-by: Fabio Estevam <festevam@gmail.com>
Tested-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Tested-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'bpf_avoid_clone'

Alexei Starovoitov says:

====================
bpf: performance improvements

v1->v2: dropped redundant iff_up check in patch 2

At plumbers we discussed different options on how to get rid of skb_clone
from bpf_clone_redirect(), the patch 2 implements the best option.
Patch 1 adds 'integrated exts' to cls_bpf to improve performance by
combining simple actions into bpf classifier.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

bpf: add bpf_redirect() helper

Existing bpf_clone_redirect() helper clones skb before redirecting
it to RX or TX of destination netdev.
Introduce bpf_redirect() helper that does that without cloning.

Benchmarked with two hosts using 10G ixgbe NICs.
One host is doing line rate pktgen.
Another host is configured as:
$ tc qdisc add dev $dev ingress
$ tc filter add dev $dev root pref 10 u32 match u32 0 0 flowid 1:2 \
   action bpf run object-file tcbpf1_kern.o section clone_redirect_xmit drop
so it receives the packet on $dev and immediately xmits it on $dev + 1
The section 'clone_redirect_xmit' in tcbpf1_kern.o file has the program
that does bpf_clone_redirect() and performance is 2.0 Mpps

$ tc filter add dev $dev root pref 10 u32 match u32 0 0 flowid 1:2 \
   action bpf run object-file tcbpf1_kern.o section redirect_xmit drop
which is using bpf_redirect() - 2.4 Mpps

and using cls_bpf with integrated actions as:
$ tc filter add dev $dev root pref 10 \
  bpf run object-file tcbpf1_kern.o section redirect_xmit integ_act classid 1
performance is 2.5 Mpps

To summarize:
u32+act_bpf using clone_redirect - 2.0 Mpps
u32+act_bpf using redirect - 2.4 Mpps
cls_bpf using redirect - 2.5 Mpps

For comparison linux bridge in this setup is doing 2.1 Mpps
and ixgbe rx + drop in ip_rcv - 7.8 Mpps

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cls_bpf: introduce integrated actions

Often cls_bpf classifier is used with single action drop attached.
Optimize this use case and let cls_bpf return both classid and action.
For backwards compatibility reasons enable this feature under
TCA_BPF_FLAG_ACT_DIRECT flag.

Then more interesting programs like the following are easier to write:
int cls_bpf_prog(struct __sk_buff *skb)
{
  /* classify arp, ip, ipv6 into different traffic classes
   * and drop all other packets
   */
  switch (skb->protocol) {
  case htons(ETH_P_ARP):
    skb->tc_classid = 1;
    break;
  case htons(ETH_P_IP):
    skb->tc_classid = 2;
    break;
  case htons(ETH_P_IPV6):
    skb->tc_classid = 3;
    break;
  default:
    return TC_ACT_SHOT;
  }

  return TC_ACT_OK;
}

Joint work with Daniel Borkmann.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: only check perm protocol when register proto

The permanent protocol nodes are at the head of the list,
So only need check all these nodes.

No matter the new node is permanent or not,
insert the new node after the last permanent protocol node,

If the new node conflicts with existing permanent node,
return error.

Signed-off-by: Martin Zhang <martinbj2008@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bonding: use l4 hash if available

If skb carries a l4 hash, no need to perform a flow dissection.

Performance is slightly better :

lpaa5:~# ./super_netperf 200 -H lpaa6 -t TCP_RR -l 100
2.39012e+06
lpaa5:~# ./super_netperf 200 -H lpaa6 -t TCP_RR -l 100
2.39393e+06
lpaa5:~# ./super_netperf 200 -H lpaa6 -t TCP_RR -l 100
2.39988e+06

After patch :

lpaa5:~# ./super_netperf 200 -H lpaa6 -t TCP_RR -l 100
2.43579e+06
lpaa5:~# ./super_netperf 200 -H lpaa6 -t TCP_RR -l 100
2.44304e+06
lpaa5:~# ./super_netperf 200 -H lpaa6 -t TCP_RR -l 100
2.44312e+06

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <tom@herbertland.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: provide skb->hash to synack packets

In commit b73c3d0e4f0e ("net: Save TX flow hash in sock and set in skbuf
on xmit"), Tom provided a l4 hash to most outgoing TCP packets.

We'd like to provide one as well for SYNACK packets, so that all packets
of a given flow share same txhash, to later enable bonding driver to
also use skb->hash to perform slave selection.

Note that a SYNACK retransmit shuffles the tx hash, as Tom did
in commit 265f94ff54d62 ("net: Recompute sk_txhash on negative routing
advice") for established sockets.

This has nice effect making TCP flows resilient to some kind of black
holes, even at connection establish phase.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <tom@herbertland.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Acked-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'nf_hook_netns'

Eric W. Biederman says:

====================
Passing net through the netfilter hooks

My primary goal with this patchset and it's follow ups is to cleanup the
network routing paths so that we do not look at the output device to
derive the network namespace.  My plan is to pass the network namespace
of the transmitting socket through the output path, to replace code that
looks at the output network device today.  Once that is done we can have
routes with output devices outside of the current network namespace.
Which should allow reception and transmission of packets in network
namespaces to be as fast as normal packet reception and transmission
with early demux disabled, because it will same code path.

Once skb_dst(skb)->dev is a little better under control I think it will
also be possible to use rcu to cleanup the ancient hack that sets
dst->dev to loopback_dev when a network device is removed.

The work to get there is a series of code cleanups.  I am starting with
passing net into the netfilter hooks and into the functions that are
called after the netfilter hooks.  This removes from netfilter the
need to guess which network namespace it is working on.

To get there I perform a series of minor prep patches so the big changes
at the end are possible to audit without getting lost in the noise.  In
particular I have a lot of patches computing net into a local variable
and then using it through out the function.

So this patchset encompases removing dead code, sorting out the _sk
functions that were added last time someone pushed a prototype change
through the post netfilter functions.  Cleaning up individual functions
use of the network namespace.  Passing net into the netfilter hooks.
Passing net into the post netfilter functions.  Using state->net in
the netfilter code where it is available and trivially usable.

Pablo, Dave I don't know whose tree this makes more sense to go
through.  I am assuming at least initially Pablos as netfilter is
involved.  From what I have seen there will be a lot of back and forth
between the netfilter code paths and the routing code paths.

The patches are also available (against 4.3-rc1) at:
git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/net-next.git master
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: Add blank lines in callers of netfilter hooks

In code review it was noticed that I had failed to add some blank lines
in places where they are customarily used.  Taking a second look at the
code I have to agree blank lines would be nice so I have added them
here.

Reported-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: Pass net into okfn

This is immediately motivated by the bridge code that chains functions that
call into netfilter.  Without passing net into the okfns the bridge code would
need to guess about the best expression for the network namespace to process
packets in.

As net is frequently one of the first things computed in continuation functions
after netfilter has done it's job passing in the desired network namespace is in
many cases a code simplification.

To support this change the function dst_output_okfn is introduced to
simplify passing dst_output as an okfn.  For the moment dst_output_okfn
just silently drops the struct net.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: Use nf_hook_state.net

Instead of saying "net = dev_net(state->in?state->in:state->out)"
just say "state->net".  As that information is now availabe,
much less confusing and much less error prone.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netfilter: Pass struct net into the netfilter hooks

Pass a network namespace parameter into the netfilter hooks.  At the
call site of the netfilter hooks the path a packet is taking through
the network stack is well known which allows the network namespace to
be easily and reliabily.

This allows the replacement of magic code like
"dev_net(state->in?:state->out)" that appears at the start of most
netfilter hooks with "state->net".

In almost all cases the network namespace passed in is derived
from the first network device passed in, guaranteeing those
paths will not see any changes in practice.

The exceptions are:
xfrm/xfrm_output.c:xfrm_output_resume()         xs_net(skb_dst(skb)->xfrm)
ipvs/ip_vs_xmit.c:ip_vs_nat_send_or_cont()      ip_vs_conn_net(cp)
ipvs/ip_vs_xmit.c:ip_vs_send_or_cont()          ip_vs_conn_net(cp)
ipv4/raw.c:raw_send_hdrinc()                    sock_net(sk)
ipv6/ip6_output.c:ip6_xmit() sock_net(sk)
ipv6/ndisc.c:ndisc_send_skb()                   dev_net(skb->dev) not dev_net(dst->dev)
ipv6/raw.c:raw6_send_hdrinc()                   sock_net(sk)
br_netfilter_hooks.c:br_nf_pre_routing_finish() dev_net(skb->dev) before skb->dev is set to nf_bridge->physindev

In all cases these exceptions seem to be a better expression for the
network namespace the packet is being processed in then the historic
"dev_net(in?in:out)".  I am documenting them in case something odd
pops up and someone starts trying to track down what happened.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: Add br_netif_receive_skb remove netif_receive_skb_sk

netif_receive_skb_sk is only called once in the bridge code, replace
it with a bridge specific function that calls netif_receive_skb.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: Cache net in br_nf_pre_routing_finish

This is prep work for passing net to the netfilter hooks.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: Pass net into br_nf_push_frag_xmit

When struct net starts being passed through the ipv4 and ipv6 fragment
routines br_nf_push_frag_xmit will need to take a net parameter.
Prepare br_nf_push_frag_xmit before that is needed and introduce
br_nf_push_frag_xmit_sk for the call sites that still need the old
calling conventions.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: Pass net into br_nf_ip_fragment

This is a prep work for passing struct net through ip_do_fragment and
later the netfilter okfn.   Doing this independently makes the later
code changes clearer.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Compute net once in raw6_send_hdrinc

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Cache net in ip6_output

Keep net in a local variable so I can use it in NF_HOOK_COND
when I pass struct net to all of the netfilter hooks.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Only compute net once in ip6_finish_output2

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Don't recompute net in ip6_rcv

Avoid silly redundant code

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Remove dev_queue_xmit_sk

A function with weird arguments that it will never use to accomdate a
netfilter callback prototype is absolutely in the core of the
networking stack.  Frankly it does not make sense and it causes a lot
of confusion as to why arguments that are never used are being passed
to the function.

As I am preparing to make a second change to arguments to the okfn even
the names stops making sense.

As I have removed the two callers of this function remove this confusion
from the networking stack.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bridge: Introduce br_send_bpdu_finish

The function dev_queue_xmit_skb_sk is unncessary and very confusing.
Introduce br_send_bpdu_finish to remove the need for dev_queue_xmit_skb_sk,
and have br_send_bpdu_finish call dev_queue_xmit.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

arp: Introduce arp_xmit_finish

The function dev_queue_xmit_skb_sk is unncessary and very confusing.
Introduce arp_xmit_finish to remove the need for dev_queue_xmit_skb_sk,
and have arp_xmit_finish call dev_queue_xmit.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Only compute net once in ip6mr_forward2_finish

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Only compute net once in ipmr_forward_finish

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Only compute net once in ip_rcv_finish

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Only compute net once in ip_finish_output2

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Explicitly compute net in ip_fragment

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Only compute net once in ip_do_fragment

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Don't recompute net in ipmr_queue_xmit

Calling dev_net(dev) for is just silly.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Remember the net in ip_output and ip_mc_output

This is a prepatory patch to passing net int the netfilter hooks,
where net will be used again.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>