GitHub/moto-9609/android_kernel_motorola_exynos9610.git
8 years agoairo: prevent potential underflow in airo_set_freq()
Dan Carpenter [Tue, 10 May 2016 19:20:37 +0000 (22:20 +0300)]
airo: prevent potential underflow in airo_set_freq()

I'm not sure if this can underflow but Smatch complains.  It seems
harmless to add a check for negatives.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agomwifiex: fixup error messages
Wei-Ning Huang [Tue, 10 May 2016 06:21:23 +0000 (14:21 +0800)]
mwifiex: fixup error messages

Use dev_err instead of pr_err and add newline character at the end.

Signed-off-by: Wei-Ning Huang <wnhuang@chromium.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortlwifi: pci: use dev_kfree_skb_irq instead of kfree_skb in rtl_pci_reset_trx_ring
wang yanqing [Fri, 6 May 2016 16:33:53 +0000 (00:33 +0800)]
rtlwifi: pci: use dev_kfree_skb_irq instead of kfree_skb in rtl_pci_reset_trx_ring

We can't use kfree_skb in irq disable context, because spin_lock_irqsave
make sure we are always in irq disable context, use dev_kfree_skb_irq
instead of kfree_skb is better than dev_kfree_skb_any.

This patch fix below kernel warning:
[ 7612.095528] ------------[ cut here ]------------
[ 7612.095546] WARNING: CPU: 3 PID: 4460 at kernel/softirq.c:150 __local_bh_enable_ip+0x58/0x80()
[ 7612.095550] Modules linked in: rtl8723be x86_pkg_temp_thermal btcoexist rtl_pci rtlwifi rtl8723_common
[ 7612.095567] CPU: 3 PID: 4460 Comm: ifconfig Tainted: G        W       4.4.0+ #4
[ 7612.095570] Hardware name: LENOVO 20DFA04FCD/20DFA04FCD, BIOS J5ET48WW (1.19 ) 08/27/2015
[ 7612.095574]  00000000 00000000 da37fc70 c12ce7c5 00000000 da37fca0 c104cc59 c19d4454
[ 7612.095584]  00000003 0000116c c19d4784 00000096 c10508a8 c10508a8 00000200 c1b42400
[ 7612.095594]  f29be780 da37fcb0 c104ccad 00000009 00000000 da37fcbc c10508a8 f21f08b8
[ 7612.095604] Call Trace:
[ 7612.095614]  [<c12ce7c5>] dump_stack+0x41/0x5c
[ 7612.095620]  [<c104cc59>] warn_slowpath_common+0x89/0xc0
[ 7612.095628]  [<c10508a8>] ? __local_bh_enable_ip+0x58/0x80
[ 7612.095634]  [<c10508a8>] ? __local_bh_enable_ip+0x58/0x80
[ 7612.095640]  [<c104ccad>] warn_slowpath_null+0x1d/0x20
[ 7612.095646]  [<c10508a8>] __local_bh_enable_ip+0x58/0x80
[ 7612.095653]  [<c16b7d34>] destroy_conntrack+0x64/0xa0
[ 7612.095660]  [<c16b300f>] nf_conntrack_destroy+0xf/0x20
[ 7612.095665]  [<c1677565>] skb_release_head_state+0x55/0xa0
[ 7612.095670]  [<c16775bb>] skb_release_all+0xb/0x20
[ 7612.095674]  [<c167760b>] __kfree_skb+0xb/0x60
[ 7612.095679]  [<c16776f0>] kfree_skb+0x30/0x70
[ 7612.095686]  [<f81b869d>] ? rtl_pci_reset_trx_ring+0x22d/0x370 [rtl_pci]
[ 7612.095692]  [<f81b869d>] rtl_pci_reset_trx_ring+0x22d/0x370 [rtl_pci]
[ 7612.095698]  [<f81b87f9>] rtl_pci_start+0x19/0x190 [rtl_pci]
[ 7612.095705]  [<f81970e6>] rtl_op_start+0x56/0x90 [rtlwifi]
[ 7612.095712]  [<c17e3f16>] drv_start+0x36/0xc0
[ 7612.095717]  [<c17f5ab3>] ieee80211_do_open+0x2d3/0x890
[ 7612.095725]  [<c16820fe>] ? call_netdevice_notifiers_info+0x2e/0x60
[ 7612.095730]  [<c17f60bd>] ieee80211_open+0x4d/0x50
[ 7612.095736]  [<c16891b3>] __dev_open+0xa3/0x130
[ 7612.095742]  [<c183fa53>] ? _raw_spin_unlock_bh+0x13/0x20
[ 7612.095748]  [<c1689499>] __dev_change_flags+0x89/0x140
[ 7612.095753]  [<c127c70d>] ? selinux_capable+0xd/0x10
[ 7612.095759]  [<c1689589>] dev_change_flags+0x29/0x60
[ 7612.095765]  [<c1700b93>] devinet_ioctl+0x553/0x670
[ 7612.095772]  [<c12db758>] ? _copy_to_user+0x28/0x40
[ 7612.095777]  [<c17018b5>] inet_ioctl+0x85/0xb0
[ 7612.095783]  [<c166e647>] sock_ioctl+0x67/0x260
[ 7612.095788]  [<c166e5e0>] ? sock_fasync+0x80/0x80
[ 7612.095795]  [<c115c99b>] do_vfs_ioctl+0x6b/0x550
[ 7612.095800]  [<c127c812>] ? selinux_file_ioctl+0x102/0x1e0
[ 7612.095807]  [<c10a8914>] ? timekeeping_suspend+0x294/0x320
[ 7612.095813]  [<c10a256a>] ? __hrtimer_run_queues+0x14a/0x210
[ 7612.095820]  [<c1276e24>] ? security_file_ioctl+0x34/0x50
[ 7612.095827]  [<c115cef0>] SyS_ioctl+0x70/0x80
[ 7612.095832]  [<c1001804>] do_fast_syscall_32+0x84/0x120
[ 7612.095839]  [<c183ff91>] sysenter_past_esp+0x36/0x55
[ 7612.095844] ---[ end trace 97e9c637a20e8348 ]---

Signed-off-by: Wang YanQing <udknight@gmail.com>
Cc: Stable <stable@vger.kernel.org>
Acked-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortlwifi: Remove double check for cnt_after_linked
wang yanqing [Thu, 5 May 2016 15:08:22 +0000 (23:08 +0800)]
rtlwifi: Remove double check for cnt_after_linked

rtl_lps_enter does two successive check for cnt_after_linked
to make sure some time has elapsed after linked. The second
check isn't necessary, because if cnt_after_linked is bigger
than 5, it is bigger than 2 of course!

This patch remove the second check code.

Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agobrcmfmac: Add 4356 sdio support
Christian Daudt [Thu, 5 May 2016 00:55:20 +0000 (17:55 -0700)]
brcmfmac: Add 4356 sdio support

This adds support for the 4356-sdio wireless chip.

Signed-off-by: Christian Daudt <csd@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortlwifi: rtl818x: silence uninitialized variable warning
Dan Carpenter [Wed, 4 May 2016 06:16:39 +0000 (09:16 +0300)]
rtlwifi: rtl818x: silence uninitialized variable warning

What about if "rtlphy->pwrgroup_cnt" is 2?  In that case we would use an
uninitialized "chnlgroup" variable and probably crash.  Maybe that can't
happen for some reason which is not obvious but in that case this patch
is harmless.

Setting it to zero seems like a standard default in the surrounding code
so it's probably fine here as well.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agomwifiex: Drop unnecessary include pcieport_if.h
Lukas Wunner [Tue, 3 May 2016 19:44:40 +0000 (21:44 +0200)]
mwifiex: Drop unnecessary include pcieport_if.h

This header file is only needed for drivers binding to a PCI bridge
device allocated by drivers/pci/pcie/portdrv_core.c. The mwifiex driver
doesn't do that nor use any symbols defined in pcieport_if.h.

Cc: Amitkumar Karwar <akarwar@marvell.com>
Cc: Nishant Sarmukadam <nishants@marvell.com>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortlwifi: Fix logic error in enter/exit power-save mode
wang yanqing [Mon, 2 May 2016 16:38:36 +0000 (00:38 +0800)]
rtlwifi: Fix logic error in enter/exit power-save mode

In commit a269913c52ad ("rtlwifi: Rework rtl_lps_leave() and
rtl_lps_enter() to use work queue"), the tests for enter/exit
power-save mode were inverted. With this change applied, the
wifi connection becomes much more stable.

Fixes: a269913c52ad ("rtlwifi: Rework rtl_lps_leave() and rtl_lps_enter() to use work queue")
Signed-off-by: Wang YanQing <udknight@gmail.com>
CC: Stable <stable@vger.kernel.org> [3.10+]
Acked-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortlwifi: rtl818x: constify rtl_intf_ops structures
Julia Lawall [Sun, 1 May 2016 19:57:44 +0000 (21:57 +0200)]
rtlwifi: rtl818x: constify rtl_intf_ops structures

The rtl_intf_ops structures are never modified, so declare them as const.

Done with the help of Coccinelle.

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agowlcore/wl12xx: Fix fw logger over sdio
Guy Mishol [Thu, 14 Apr 2016 08:58:17 +0000 (11:58 +0300)]
wlcore/wl12xx: Fix fw logger over sdio

The commit fb724ed5c617 ("wlcore: Fix regression in
wlcore_set_partition()") fixed wl12xx functionality.
However, it reverted the support in fw logger
over sdio in wl18xx.

This patch reverts the changes made and also fixes
the original functionality issue introduced in wl12xx.

Fixes: fb724ed5c617 ("wlcore: Fix regression in wlcore_set_partition()")
Signed-off-by: Guy Mishol <guym@ti.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agoMerge tag 'iwlwifi-next-for-kalle-2016-05-10' of git://git.kernel.org/pub/scm/linux...
Kalle Valo [Wed, 11 May 2016 17:54:07 +0000 (20:54 +0300)]
Merge tag 'iwlwifi-next-for-kalle-2016-05-10' of git://git./linux/kernel/git/iwlwifi/iwlwifi-next

* work for RX multiqueue continues (Sara);
* dynamic queue allocation work continues (Liad);
* add Luca as maintainer;
* a bunch of fixes and improvements all over;

8 years agoMAINTAINERS: add myself as co-maintainer of the iwlwifi driver
Luca Coelho [Tue, 19 Apr 2016 09:03:45 +0000 (12:03 +0300)]
MAINTAINERS: add myself as co-maintainer of the iwlwifi driver

I'm starting to take a more active role in the iwlwifi driver
maintainership.

Cc: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Cc: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add default value to disable_11ac mod param description
Emmanuel Grumbach [Mon, 14 Mar 2016 13:21:06 +0000 (15:21 +0200)]
iwlwifi: add default value to disable_11ac mod param description

Small change to make it clear that the default value is false.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: grab NIC access only once on RX init
Sara Sharon [Tue, 12 Apr 2016 15:41:32 +0000 (18:41 +0300)]
iwlwifi: pcie: grab NIC access only once on RX init

When initializing RX we grab NIC access for every read and
write. This is redundant - we can just grab access once.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: use shadow registers for updating write pointer
Sara Sharon [Sun, 17 Apr 2016 12:08:59 +0000 (15:08 +0300)]
iwlwifi: pcie: use shadow registers for updating write pointer

The RX queues have a shadow register for the write pointer
that enables updates without grabbing NIC access. Use them
instead of the periphery registers because accessing those
is much more expensive.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support dqa-mode agg on non-shared queue
Liad Kaufman [Thu, 13 Aug 2015 16:16:08 +0000 (19:16 +0300)]
iwlwifi: mvm: support dqa-mode agg on non-shared queue

In non-shared queues, DQA requires re-configuring existing
queues to become aggregated rather than allocating a new
one. It also requires "un-aggregating" an existing queue
when aggregations are turned off.

Support this requirement for non-shared queues.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: avoid msleep() with short timeout
Johannes Berg [Wed, 13 Apr 2016 08:31:14 +0000 (10:31 +0200)]
iwlwifi: pcie: avoid msleep() with short timeout

Since msleep is based on jiffies, it can sleep for a long time.
Use usleep_range() instead to shorten the maximum time.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: extend device reset delay
Johannes Berg [Wed, 13 Apr 2016 08:24:59 +0000 (10:24 +0200)]
iwlwifi: pcie: extend device reset delay

Newer hardware generations will take longer to be accessible again
after reset, so we need to wait longer before continuing any flow
that did a reset.

Rather than make the wait time configurable, simply extend it for
all.

Since all of these code paths can sleep, use usleep_range() rather
than mdelay().

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: turn on SGI support for VHT 160MHz
Gregory Greenman [Tue, 12 Apr 2016 12:16:24 +0000 (15:16 +0300)]
iwlwifi: turn on SGI support for VHT 160MHz

Devices supporting VHT 160MHz width are supporting also Short GI.
Turn on this capability in vht cap.

Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: make configuration structs smaller
Johannes Berg [Tue, 12 Apr 2016 10:36:01 +0000 (12:36 +0200)]
iwlwifi: make configuration structs smaller

Since we have a lot of configuration structs (almost 70) saving
some memory in each one of them leads to an overall saving of
~2.6KiB of memory.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: don't access a nonexistent register upon assert
Emmanuel Grumbach [Tue, 12 Apr 2016 10:07:52 +0000 (13:07 +0300)]
iwlwifi: don't access a nonexistent register upon assert

The commit below added code to dump the content of FIFOs
that are present only on dual CPU products (8000 and up).
This broke 7265D whose firmware does advertise
IWL_UCODE_TLV_CAPA_EXTEND_SHARED_MEM_CFG but doesn't have 2
CPUs. The current code does check the length of the FIFO
before dumping them (and the nonexistent FIFO has a 0
length), but we still accessed a register to set the FIFO
number and that made the DMA unhappy.

The impact was a much longer recovery upon firmware assert.

Fixes: 5b086414293f ("iwlwifi: mvm: support dumping UMAC internal txfifos")
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: remove IWLWIFI_DEBUG_EXPERIMENTAL_UCODE
Emmanuel Grumbach [Sun, 10 Apr 2016 05:48:46 +0000 (08:48 +0300)]
iwlwifi: remove IWLWIFI_DEBUG_EXPERIMENTAL_UCODE

This Kconfig option allows to load a firmware for
debugging with a different name. This mechanism has not
been used for a few years now and replacing the firmware
file works as well.
Kill this Kconfig option and all the code that goes with it.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: use helpers to get iwl_mvm_sta
Sara Sharon [Sun, 10 Apr 2016 12:51:54 +0000 (15:51 +0300)]
iwlwifi: mvm: use helpers to get iwl_mvm_sta

Getting the mvm station out of station id requires dereferencing
the station id to get ieee80211_sta, then checking for pointer
validity and only then extract mvm station out.
Given that there are helpers to do it - use them instead of
duplicating the code whenever we need only mvm station.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: remove redundant alloc_ctx parameter
Sara Sharon [Sun, 10 Apr 2016 13:02:12 +0000 (16:02 +0300)]
iwlwifi: mvm: remove redundant alloc_ctx parameter

iwl_phy_db_set_section() is get called only from atomic
context, the alloc_ctx parameter is not needed. Remove it.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: make phy_db size dynamic
Sara Sharon [Sun, 10 Apr 2016 09:27:25 +0000 (12:27 +0300)]
iwlwifi: mvm: make phy_db size dynamic

Driver is agnostic to the number of the phy_db entries and
only serves the firmware as a pipe to move the data from init
image to RT image.
As the size of the arrays may change (as it does in 9000 device)
allocate it dynamically. Firmware sends the largest index first
so we can use this to know how much we should allocate.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: set correct vht capability
Sara Sharon [Thu, 7 Apr 2016 14:26:53 +0000 (17:26 +0300)]
iwlwifi: mvm: set correct vht capability

Our device supports only 160 GHz and not 80+80. Fix
VHT flag accordingly.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: loosen nssn comparison to reorder buffer head
Sara Sharon [Wed, 30 Mar 2016 17:04:48 +0000 (20:04 +0300)]
iwlwifi: mvm: loosen nssn comparison to reorder buffer head

Up till now, the reorder buffer uses standard spec based comparison
when comparing the buffer status to NSSN. This indeed works for the
regular case, since we shouldn't cross the 2048 boundary without
getting a frame release notification.
However, this is problematic due to packet filtering that may be
performed by the FW while we are in d0i3. Theoretically we may
filter over 2048 packets, and then the check of the NSSN will get
incorrect.
Change the comparison to always trust nssn unless it is 64 or less
frames behind the head - which might happen due to a timeout.
This new comparison is to be used only when comparing reorder buffer
head with nssn, and not when comparing the packet SN to nssn or
reorder buffer head.
Put this in a separate commit as the logic is a bit tricky and
stands for its own commit message.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: don't wake up the NIC when writing CSRs in MSIX mode
Haim Dreyfuss [Sun, 3 Apr 2016 17:15:26 +0000 (20:15 +0300)]
iwlwifi: pcie: don't wake up the NIC when writing CSRs in MSIX mode

CSR registers are always available even when the NIC is not awake, no
need to wake up the NIC before accessing them. This has a huge impact
when we re-enable an interrupt at the end of the ISR since waking up the
NIC can take some time.

Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: Fix firmware name maximum length definition
Haim Dreyfuss [Wed, 6 Apr 2016 07:45:05 +0000 (10:45 +0300)]
iwlwifi: Fix firmware name maximum length definition

Previous patch had changed firmware name convention for
new generation product. The firmware name is now longer
than the former convention. Adapt max firmware name length
to the new convention.

Fixes: e1ba684f762b ("iwlwifi: 8000: fix MODULE_FIRMWARE input")
Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: don't allow negative reference count
Sara Sharon [Mon, 4 Apr 2016 16:30:05 +0000 (19:30 +0300)]
iwlwifi: mvm: don't allow negative reference count

Currently code allows mvm reference to become negative and
only warns in case mvm reference is released while reference
counting is 0.
However, we better prevent this from happening at all since
iwl_mvm_unref() may race against iwl_mvm_unref_all_except()
which is called on restart.
As a result we might get the same reference unreferenced twice
ending with a negative value:
An example for an easily reproduced log:
    [ 2689.909166] iwl_mvm_ref Take mvm reference - type 8
    [ 2690.732716] iwl_mvm_unref_all_except Cleanup: remove mvm ref type 8 (1)
    [ 2690.849708] iwl_mvm_unref Leave mvm reference - type 8
    [ 2690.849721] WARNING: ... iwl_mvm_unref+0xb0/0xc0 [iwlmvm]()

If there will be yet another another restart iwl_mvm_unref_all_except
will run from 0 up to ref count, and since it is unsigned, we will throw
the transport ref count completely out of balance:
    iwl_mvm_unref_all_except[I] -- Cleanup: remove mvm ref type 8 (255)
    iwl_trans_slv_unref[I] -- rpm counter: 0
    iwl_trans_slv_unref[I] -- rpm counter: -1
    iwl_trans_slv_unref[I] -- rpm counter: -2
 ...
    iwl_trans_slv_unref[I] -- rpm counter: -253
    iwl_trans_slv_unref[I] -- rpm counter: -254

As there is no valid scenario where we can get to a negative
reference count - prevent it from happening.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add more registers to dump upon error
Golan Ben-Ami [Wed, 6 Apr 2016 08:59:50 +0000 (11:59 +0300)]
iwlwifi: mvm: add more registers to dump upon error

Add UREG, RXFC, RFH, WMAL and RL2P registers to the prph dump
upon error. These regesiters could help to debug MSI-X and other
issues.

These register should be dumped only when multi-queue rx is supported
so separate the prph ranges static array to two different arrays,
and enable dumping different prph ranges according to run-time
decision.

Signed-off-by: Golan Ben-Ami <golan.ben.ami@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add a new mvm reference type for RX data
Luca Coelho [Wed, 30 Mar 2016 12:05:56 +0000 (15:05 +0300)]
iwlwifi: mvm: add a new mvm reference type for RX data

When a data packet is received, we need to make sure that we stay
awake until it can be processed and wait a while before trying to
enter runtime_suspend os system_suspend again.  To do so, add a new
reference type for RX data and take the reference when sending the
packet to mac80211.  We only do this for data packets, all the other
RX packets sent by the firmware (e.g. notifications) are not a reason
to prevent suspend.

Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: pass station to mac80211 RX where known
Johannes Berg [Thu, 10 Mar 2016 10:55:44 +0000 (11:55 +0100)]
iwlwifi: mvm: pass station to mac80211 RX where known

When we've already looked up the transmitter station, we can just
pass it to mac80211 using the new ieee80211_rx_napi(). This saves
the overhead of looking it up in mac80211 again.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: consider VHT 160MHz while parsing NVM
Gregory Greenman [Sun, 3 Apr 2016 11:06:12 +0000 (14:06 +0300)]
iwlwifi: consider VHT 160MHz while parsing NVM

Devices belonging to 9000 family can support VHT 160MHz channel
width, so need to consider it when configuring VHT capabilities.
However, NVM file doesn't have a single bit specifying that 160MHz
is supported. This patch turns on 160MHz support in VHT capabilities
in case there's at least one channel supporting 160MHz.

Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: allow combining different phy images with mac images
Haim Dreyfuss [Sun, 3 Apr 2016 16:55:59 +0000 (19:55 +0300)]
iwlwifi: allow combining different phy images with mac images

Currently there is one to one function between device id to it's ucode.
The new generation devices allows to combine different phy and mac images.
Now we have two different ucode images with the same device id.
Read RF ID to identify phy image and overwrite it if needed.

Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: wake from runtime suspend before sending sync commands
Luca Coelho [Fri, 11 Mar 2016 10:12:16 +0000 (12:12 +0200)]
iwlwifi: wake from runtime suspend before sending sync commands

If a host command was queued while in runtime suspend, it would go out
before the D0I3_END_CMD was sent.  Sometimes it works, but sometimes
it fails, and it is obviously the wrong thing to do.

To fix this, have the opmode take a reference before sending a SYNC
command and make the pcie trans wait for the runtime state to become
active before actually queueing the command.

Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: allow a debug knob for Tx A-MSDU even if rate control forbids it
Emmanuel Grumbach [Sun, 3 Apr 2016 07:15:59 +0000 (10:15 +0300)]
iwlwifi: mvm: allow a debug knob for Tx A-MSDU even if rate control forbids it

There is a debugfs knob to configure the maximal length
of the A-MSDU. If this value is not 0 (which is the
default), allow Tx A-MSDU even if the rate control
disallows it.
While at it, add "unlikely" to the if that limits the
length of the A-MSDU based on the debugfs hook.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: Rename 9560 to 9260 and add new PCI IDs for it
Haim Dreyfuss [Sun, 3 Apr 2016 15:56:15 +0000 (18:56 +0300)]
iwlwifi: Rename 9560 to 9260 and add new PCI IDs for it

Rename 9560 to 9260.
Add new PCI ID for 9260 and change some entries from 5165 to 9260.
Also order the 9000 series.

Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add a flag to disable checksum
Sara Sharon [Sun, 3 Apr 2016 07:19:16 +0000 (10:19 +0300)]
iwlwifi: mvm: add a flag to disable checksum

Add a constant to allow disabling checksum. This will enable easier
debugging in early phases.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: advertise RSS queue usage
Johannes Berg [Wed, 30 Mar 2016 09:05:16 +0000 (11:05 +0200)]
iwlwifi: mvm: advertise RSS queue usage

In order for mac80211 to use per-CPU statistics for RSS RX, the
driver needs to advertise that it uses RSS. Do this when using
more than a single queue.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: trans: don't call the trans-specific ref/unref directly
Luca Coelho [Wed, 30 Mar 2016 17:59:27 +0000 (20:59 +0300)]
iwlwifi: trans: don't call the trans-specific ref/unref directly

It's cleaner to always call the iwl_trans_ref/unref() functions
instead of sometimes calling the trans-specific ops directly.  This
also prepares for moving some of the code from the trans-specific ops
to the common trans code.

Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support queue removal in ADD_STA hcmd
Liad Kaufman [Mon, 21 Sep 2015 12:14:23 +0000 (14:14 +0200)]
iwlwifi: mvm: support queue removal in ADD_STA hcmd

To indicate to the FW that a queue has been removed, an
existing flag in the ADD_STA HCMD (that hasn't been in use)
has been changed to indicate that a queue is being removed
from a STA.

Update this in the driver code.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: utilize the frame release infrastructure
Sara Sharon [Sun, 28 Feb 2016 13:41:47 +0000 (15:41 +0200)]
iwlwifi: mvm: utilize the frame release infrastructure

The firmware will send frame release notification in order
to release "stuck" frames on a queue where no more frames
arrive on.
Upon receiving the message the driver shall indicate the frames
up to the NSSN.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add reorder timeout per frame
Sara Sharon [Sun, 28 Feb 2016 18:28:17 +0000 (20:28 +0200)]
iwlwifi: mvm: add reorder timeout per frame

Add a timer in order to release expired frames from the
reorder buffer.
This is needed since some APs do not retransmit frames
to fill in the reorder holes and in TCP it results with
a complete stall of traffic.

This has a few side effects on the general design:

The nssn may not reflect the the head of the reorder buffer.
This situation is valid, and packets with SN lower than the
reorder buffer head will be dropped.

Another side effect is that since the reorder timer might expire
we need to lock the reorder buffer.
This however is fine since the locking is only inside a
single reorder buffer between RX path and reorder timeout and
there is no outside contention.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add reorder buffer per queue
Sara Sharon [Wed, 23 Mar 2016 14:32:02 +0000 (16:32 +0200)]
iwlwifi: mvm: add reorder buffer per queue

Next hardware will direct packets to core based on the TCP/UDP
streams.
This logic can create holes in reorder buffer since packets that
belong to other stream were directed to a different core.
However, those are valid holes and the packets can be indicated
in L3 order.

The hardware will utilize a mechanism of informing the driver of
the normalized ssn and the driver shall release all packets that
SN is lower than the nssn.
This enables managing the reorder across the queues without sharing
any data between them.

The reorder buffer is allocated and released directly in the RX path
in order to avoid various races between control path and rx path.
The code utilizes the internal messaging to notify rx queues of when
to delete the reorder buffer.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add firmware API name comment
Johannes Berg [Wed, 30 Mar 2016 09:57:01 +0000 (11:57 +0200)]
iwlwifi: mvm: add firmware API name comment

Add the firmware API name to the struct iwl_wowlan_gtk_status.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: add infrastructure for tracking BA session in driver
Sara Sharon [Sun, 20 Mar 2016 14:23:41 +0000 (16:23 +0200)]
iwlwifi: mvm: add infrastructure for tracking BA session in driver

According to the spec when a BA session is started there
is a timeout set for the session in the ADDBA request.
If there is not activity on the TA/TID then the session
expires and a DELBA is sent.
In order to check for the timeout, data must be shared
among the rx queues.
Add a timer that runs as long as BA session is active
for the station and stops aggregation session if needed.
This patch also lays the infrastructure for the reordering
buffer which will be enabled in the next patches.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: change RX sync notification to be an attribute and not a type
Sara Sharon [Wed, 23 Mar 2016 14:31:43 +0000 (16:31 +0200)]
iwlwifi: mvm: change RX sync notification to be an attribute and not a type

Currently the sync notification is a type of notification. However, it
is better fitted as an attribute of a notification, since there might
be another message in the payload (delba for instance) that should be
sent while control path is waiting for all queues to process.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: implement driver RX queues sync command
Sara Sharon [Thu, 18 Feb 2016 12:21:12 +0000 (14:21 +0200)]
iwlwifi: mvm: implement driver RX queues sync command

mac80211 will call the driver whenever there is a race between
RSS queues and control path that requires a processing of all
pending frames in RSS queues.
Implement that by utilizing the internal notification mechanism:
queue a message to all queues. When the message is received on
a queue it decrements the atomic counter. This guarantees that
all pending frames in the RX queue were processed since the message
is in order inside the queue.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agomwifiex: change sleep cookie poll count
Shengzhen Li [Tue, 26 Apr 2016 14:25:45 +0000 (07:25 -0700)]
mwifiex: change sleep cookie poll count

Sometimes current polling count is not sufficient.
This patch increases it to 100.

Signed-off-by: Shengzhen Li <szli@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agoiwlwifi: Edit the 8265 SDIO ID
Mordechai Goodstein [Tue, 29 Mar 2016 11:41:49 +0000 (14:41 +0300)]
iwlwifi: Edit the 8265 SDIO ID

Add new 8265 series SDIO ID.

Signed-off-by: Mordechai Goodstein <mordechay.goodstein@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support p2p device frames tx on dqa queue #2
Liad Kaufman [Sun, 9 Aug 2015 16:26:56 +0000 (19:26 +0300)]
iwlwifi: mvm: support p2p device frames tx on dqa queue #2

Support sending P2P device frames should be sent from
queue #2, as required in DQA mode.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: allocate queue for probe response in dqa mode
Liad Kaufman [Tue, 4 Aug 2015 12:19:18 +0000 (15:19 +0300)]
iwlwifi: mvm: allocate queue for probe response in dqa mode

In DQA mode, allocate a dedicated queue (#9) for P2P GO/soft
AP probe responses.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoMerge tag 'iwlwifi-for-kalle-2016-05-04' of git://git.kernel.org/pub/scm/linux/kernel...
Luca Coelho [Tue, 10 May 2016 08:56:24 +0000 (11:56 +0300)]
Merge tag 'iwlwifi-for-kalle-2016-05-04' of git://git./linux/kernel/git/iwlwifi/iwlwifi-fixes

* fix P2P rates (and possibly other issues)

Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoMerge tag 'mac80211-next-for-davem-2016-04-13' of git://git.kernel.org/pub/scm/linux...
Luca Coelho [Tue, 10 May 2016 07:30:56 +0000 (10:30 +0300)]
Merge tag 'mac80211-next-for-davem-2016-04-13' of git://git./linux/kernel/git/jberg/mac80211-next into master

To synchronize with Kalle, here's just a big change that affects
all drivers - removing the duplicated enum ieee80211_band and
replacing it by enum nl80211_band. On top of that, just a small
documentation update.

8 years agoiwlwifi: mvm: don't override the rate with the AMSDU len
Emmanuel Grumbach [Tue, 3 May 2016 09:08:43 +0000 (12:08 +0300)]
iwlwifi: mvm: don't override the rate with the AMSDU len

The TSO code creates A-MSDUs from a single large send. Each
A-MSDU is an skb and skb->len doesn't include the number of
bytes which need to be added for the headers being added
(subframe header, TCP header, IP header, SNAP, padding).

To be able to set the right value in the Tx command, we
put the number of bytes added by those headers in
driver_data in iwl_mvm_tx_tso and use this value in
iwl_mvm_set_tx_cmd.

The problem by setting this value in driver_data is that
it overrides the ieee80211_tx_info. The bug manifested
itself when we send P2P related frames in CCK since the
rate in ieee80211_tx_info is zero-ed. This of course is
a violation of the P2P specification.

To fix this, copy the original ieee80211_tx_info to the
stack and pass it to the functions which need it.
Assign the number of bytes added by the headers to the
driver_data inside the skb itself.

Fixes: a6d5e32f247c ("iwlwifi: mvm: send large SKBs to the transport")
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
8 years agortl8xxxu: Remove the now obsolete mbox_ext_reg info from rtl8xxxu_fileops
Jes Sorensen [Thu, 28 Apr 2016 19:19:15 +0000 (15:19 -0400)]
rtl8xxxu: Remove the now obsolete mbox_ext_reg info from rtl8xxxu_fileops

With two different h2c_cmd() functions, mbox_ext_reg and
mbox_ext_width are no longer needed.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: rtl8xxxu_prepare_calibrate() is never used on gen1
Jes Sorensen [Thu, 28 Apr 2016 19:19:14 +0000 (15:19 -0400)]
rtl8xxxu: rtl8xxxu_prepare_calibrate() is never used on gen1

Rename it to rtl8xxxu_gen2_prepare_calibrate() and remove the calls to
it from rtl8xxxu_gen1_phy_iq_calibrate()

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: Split rtl8723a_h2c_cmd() into a gen1 and a gen2 version
Jes Sorensen [Thu, 28 Apr 2016 19:19:13 +0000 (15:19 -0400)]
rtl8xxxu: Split rtl8723a_h2c_cmd() into a gen1 and a gen2 version

The H2C API is completely different between gen1 and gen2 parts, so
there is little point trying to treat this as a generic function. All
calls to *_h2c_cmd() will always come from a gen1 or a gen2 specific
function.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: Rename rtl8723a_disabled_to_emu() to rtl8xxxu_disabled_to_emu()
Jes Sorensen [Thu, 28 Apr 2016 19:19:12 +0000 (15:19 -0400)]
rtl8xxxu: Rename rtl8723a_disabled_to_emu() to rtl8xxxu_disabled_to_emu()

This function is generic to most of the chips, so change the name to
reflect this.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: rename rtl8723a_channel_group() to rtl8xxxu_gen1_channel_to_group()
Jes Sorensen [Thu, 28 Apr 2016 19:19:11 +0000 (15:19 -0400)]
rtl8xxxu: rename rtl8723a_channel_group() to rtl8xxxu_gen1_channel_to_group()

This function is generic for most (if not all) gen1 parts, so rename
it to reflect this.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: Rename rtl8723a_stop_tx_beacon() to rtl8xxxu_stop_tx_beacon()
Jes Sorensen [Thu, 28 Apr 2016 19:19:10 +0000 (15:19 -0400)]
rtl8xxxu: Rename rtl8723a_stop_tx_beacon() to rtl8xxxu_stop_tx_beacon()

There is nothing 8723au specific about this function, so rename it to
reflect this.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: move rtl8188[cr] and rtl8192c related code into rtl8xxxu_8192c.c
Jes Sorensen [Thu, 28 Apr 2016 19:19:09 +0000 (15:19 -0400)]
rtl8xxxu: move rtl8188[cr] and rtl8192c related code into rtl8xxxu_8192c.c

This moves the code for rtl8188c, rtl8188r, and rtl8192c into it's own
file. This is purely a code moving exercise, there is no change to the
code itself.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: move rtl8723a related code into rtl8xxxu_8723a.c
Jes Sorensen [Thu, 28 Apr 2016 19:19:08 +0000 (15:19 -0400)]
rtl8xxxu: move rtl8723a related code into rtl8xxxu_8723a.c

This moves the rtl8723a code into it's own file. This is purely a code
moving exercise, no code changes.

This device specific file is a lot smaller since the gen1 chips
(8723a, 8188c, 8188r, 8192c) share a lot more common code than the
gen2 chips.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: move rtl8723b related code into rtl8xxxu_8723b.c
Jes Sorensen [Thu, 28 Apr 2016 19:19:07 +0000 (15:19 -0400)]
rtl8xxxu: move rtl8723b related code into rtl8xxxu_8723b.c

This moves the rtl8723b code into it's own file. This is purely a code
moving exercise, no functional changes. This did expose
rtl723a_h2c_cmd() as a function that should be refactored into a gen1
and a gen2 version.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: move rtl8192e related code into rtl8xxxu_8192e.c
Jes Sorensen [Thu, 28 Apr 2016 19:19:06 +0000 (15:19 -0400)]
rtl8xxxu: move rtl8192e related code into rtl8xxxu_8192e.c

This moves the rtl8192e code into it's own file. This is purely a code
moving exercise, no code changes.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortl8xxxu: Rename rtl8xxxu.c to rtl8xxxu_core.c
Jes Sorensen [Thu, 28 Apr 2016 19:19:05 +0000 (15:19 -0400)]
rtl8xxxu: Rename rtl8xxxu.c to rtl8xxxu_core.c

This renames the core file to rtl8xxxu_core.c in order to allow us to
keep the module nake rtl8xxxu.ko when refactoring the code into
multiple files.

Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agortlwifi: rtl818x: Deinline indexed IO functions, save 21568 bytes
Denys Vlasenko [Mon, 25 Apr 2016 14:17:52 +0000 (16:17 +0200)]
rtlwifi: rtl818x: Deinline indexed IO functions, save 21568 bytes

rtl818x_ioread8_idx: 151 bytes, 29 calls
rtl818x_ioread16_idx: 151 bytes, 11 calls
rtl818x_ioread32_idx: 151 bytes, 5 calls
rtl818x_iowrite8_idx: 157 bytes, 117 calls
rtl818x_iowrite16_idx: 158 bytes, 74 calls
rtl818x_iowrite32_idx: 157 bytes, 22 calls

Each of these functions has a pair of mutex lock/unlock ops,
both of these ops perform atomic updates of memory (on x86, it boils down to
"lock cmpxchg %reg,mem" insn), which are 4-8 times more expensive than call+return.

    text     data      bss       dec     hex filename
95894242 20860288 35991552 152746082 91ab862 vmlinux_before
95872674 20860320 35991552 152724546 91a6442 vmlinux

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Larry Finger <Larry.Finger@lwfinger.net>
CC: Chaoming Li <chaoming_li@realsil.com.cn>
CC: linux-wireless@vger.kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
8 years agoMerge tag 'wireless-drivers-next-for-davem-2016-05-02' of git://git.kernel.org/pub...
David S. Miller [Tue, 3 May 2016 04:35:16 +0000 (00:35 -0400)]
Merge tag 'wireless-drivers-next-for-davem-2016-05-02' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers patches for 4.7

Major changes:

brcmfmac

* add support for nl80211 BSS_SELECT feature

mwifiex

* add platform specific wakeup interrupt support

ath10k

* implement set_tsf() for 10.2.4 branch
* remove rare MSI range support
* remove deprecated firmware API 1 support

ath9k

* add module parameter to invert LED polarity

wcn36xx

* fixes to get the driver properly working on Dragonboard 410c
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: relax expensive skb_unclone() in iptunnel_handle_offloads()
Eric Dumazet [Sat, 30 Apr 2016 17:19:29 +0000 (10:19 -0700)]
net: relax expensive skb_unclone() in iptunnel_handle_offloads()

Locally generated TCP GSO packets having to go through a GRE/SIT/IPIP
tunnel have to go through an expensive skb_unclone()

Reallocating skb->head is a lot of work.

Test should really check if a 'real clone' of the packet was done.

TCP does not care if the original gso_type is changed while the packet
travels in the stack.

This adds skb_header_unclone() which is a variant of skb_clone()
using skb_header_cloned() check instead of skb_cloned().

This variant can probably be used from other points.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonetdevice: shrink size of struct netdev_queue
Florian Westphal [Tue, 3 May 2016 01:29:09 +0000 (03:29 +0200)]
netdevice: shrink size of struct netdev_queue

- trans_timeout is incremented when tx queue timed out (tx watchdog).
- tx_maxrate is set via sysfs

Moving tx_maxrate to read-mostly part shrinks the struct by 64 bytes.
While at it, also move trans_timeout (it is out-of-place in the
'write-mostly' part).

Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'bridge-per-vlan-stats'
David S. Miller [Tue, 3 May 2016 02:27:06 +0000 (22:27 -0400)]
Merge branch 'bridge-per-vlan-stats'

Nikolay Aleksandrov says:

====================
bridge: per-vlan stats

This set adds support for bridge per-vlan statistics.
In order to be able to dump statistics for many vlans we need a way to
continue dumping after reaching maximum size, thus patches 01 and 02 extend
the new stats API with a per-device extended link stats attribute and
callback which can save its local state and continue where it left off
afterwards. I considered using the already existing "fill_xstats" callback
but it gets confusing since we need to separate the linkinfo dump from the
new stats api dump and adding a flag/argument to do that just looks messy.
I don't think the rtnl_link_ops size is an issue, so adding these seemed
like the cleaner approach.

Patches 03 and 04 add the stats support and netlink dump support
respectively. The stats accounting is controlled via a bridge option which
is default off, thus the performance impact is kept minimal.
I've tested this set with both old and modified iproute2, kmemleak on and
some traffic stress tests while adding/removing vlans and ports.

v3:
 - drop the RCU pvid patch and remove one pointer fetch as requested
 - make stats accounting optional with default to off, the option is in the
   same cache line as vlan_proto and vlan_enabled, so it is already fetched
   before the fast path check thus the performance impact is minimal, this
   also allows us to avoid one vlan lookup and return early when using pvid
 - rebased and retested

v2:
 - Improve the error checking, rename lidx to prividx and save the current
   idx user instead of restricting it to one in patch 01
 - squash patch 02 into 01 and remove the restriction
 - add callback descriptions, improve the size calculation and change the
   xstats message structure to have an embedding level per rtnl link type
   so we can avoid one call to get the link type (and thus filter on it)
   and also each link type can now have any number of private attributes
   inside
 - fix a problem where the vlan stats are not dumped if the bridge has 0
   vlans on it but has vlans on the ports, add bridge link type private
   attributes and also add paddings for future extensions to avoid at least
   a few netlink attributes and improve struct alignment
 - drop the is_skb_forwardable argument constifying patch as it's not
   needed anymore, but it's a nice cleanup which I'll send separately
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobridge: netlink: export per-vlan stats
Nikolay Aleksandrov [Sat, 30 Apr 2016 08:25:29 +0000 (10:25 +0200)]
bridge: netlink: export per-vlan stats

Add a new LINK_XSTATS_TYPE_BRIDGE attribute and implement the
RTM_GETSTATS callbacks for IFLA_STATS_LINK_XSTATS (fill_linkxstats and
get_linkxstats_size) in order to export the per-vlan stats.
The paddings were added because soon these fields will be needed for
per-port per-vlan stats (or something else if someone beats me to it) so
avoiding at least a few more netlink attributes.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobridge: vlan: learn to count
Nikolay Aleksandrov [Sat, 30 Apr 2016 08:25:28 +0000 (10:25 +0200)]
bridge: vlan: learn to count

Add support for per-VLAN Tx/Rx statistics. Every global vlan context gets
allocated a per-cpu stats which is then set in each per-port vlan context
for quick access. The br_allowed_ingress() common function is used to
account for Rx packets and the br_handle_vlan() common function is used
to account for Tx packets. Stats accounting is performed only if the
bridge-wide vlan_stats_enabled option is set either via sysfs or netlink.
A struct hole between vlan_enabled and vlan_proto is used for the new
option so it is in the same cache line. Currently it is binary (on/off)
but it is intentionally restricted to exactly 0 and 1 since other values
will be used in the future for different purposes (e.g. per-port stats).

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: rtnetlink: add linkxstats callbacks and attribute
Nikolay Aleksandrov [Sat, 30 Apr 2016 08:25:27 +0000 (10:25 +0200)]
net: rtnetlink: add linkxstats callbacks and attribute

Add callbacks to calculate the size and fill link extended statistics
which can be split into multiple messages and are dumped via the new
rtnl stats API (RTM_GETSTATS) with the IFLA_STATS_LINK_XSTATS attribute.
Also add that attribute to the idx mask check since it is expected to
be able to save state and resume dumping (e.g. future bridge per-vlan
stats will be dumped via this attribute and callbacks).
Each link type should nest its private attributes under the per-link type
attribute. This allows to have any number of separated private attributes
and to avoid one call to get the dev link type.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: rtnetlink: allow rtnl_fill_statsinfo to save private state counter
Nikolay Aleksandrov [Sat, 30 Apr 2016 08:25:26 +0000 (10:25 +0200)]
net: rtnetlink: allow rtnl_fill_statsinfo to save private state counter

The new prividx argument allows the current dumping device to save a
private state counter which would enable it to continue dumping from
where it left off. And the idxattr is used to save the current idx user
so multiple prividx using attributes can be requested at the same time
as suggested by Roopa Prabhu.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'ipv6-tunnel-cleanups'
David S. Miller [Mon, 2 May 2016 23:23:32 +0000 (19:23 -0400)]
Merge branch 'ipv6-tunnel-cleanups'

Tom Herbert says:

====================
net: Cleanup IPv6 ip tunnels

The IPv6 tunnel code is very different from IPv4 code. There is a lot
of redundancy with the IPv4 code, particularly in the GRE tunneling.

This patch set cleans up the tunnel code to make the IPv6 code look
more like the IPv4 code and use common functions between the two
stacks where possible.

This work should make it easier to maintain and extend the IPv6 ip
tunnels.

Items in this patch set:
  - Cleanup IPv6 tunnel receive path (ip6_tnl_rcv). Includes using
    gro_cells and exporting ip6_tnl_rcv so the ip6_gre can call it
  - Move GRE functions to common header file (tx functions) or
    gre_demux.c (rx functions like gre_parse_header)
  - Call common GRE functions from IPv6 GRE
  - Create ip6_tnl_xmit (to be like ip_tunnel_xmit)

Tested:
  Ran super_netperf tests for TCP_RR and TCP_STREAM for:
    - IPv4 over gre, gretap, gre6, gre6tap
    - IPv6 over gre, gretap, gre6, gre6tap
    - ipip
    - ip6ip6
    - ipip/gue
    - IPv6 over gre/gue
    - IPv4 over gre/gue
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agogre6: Cleanup GREv6 transmit path, call common GRE functions
Tom Herbert [Sat, 30 Apr 2016 00:12:21 +0000 (17:12 -0700)]
gre6: Cleanup GREv6 transmit path, call common GRE functions

Changes in GREv6 transmit path:
  - Call gre_checksum, remove gre6_checksum
  - Rename ip6gre_xmit2 to __gre6_xmit
  - Call gre_build_header utility function
  - Call ip6_tnl_xmit common function
  - Call ip6_tnl_change_mtu, eliminate ip6gre_tunnel_change_mtu

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: Generic tunnel cleanup
Tom Herbert [Sat, 30 Apr 2016 00:12:20 +0000 (17:12 -0700)]
ipv6: Generic tunnel cleanup

A few generic changes to generalize tunnels in IPv6:
  - Export ip6_tnl_change_mtu so that it can be called by ip6_gre
  - Add tun_hlen to ip6_tnl structure.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agogre: Create common functions for transmit
Tom Herbert [Sat, 30 Apr 2016 00:12:19 +0000 (17:12 -0700)]
gre: Create common functions for transmit

Create common functions for both IPv4 and IPv6 GRE in transmit. These
are put into gre.h.

Common functions are for:
  - GRE checksum calculation. Move gre_checksum to gre.h.
  - Building a GRE header. Move GRE build_header and rename
    gre_build_header.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: Create ip6_tnl_xmit
Tom Herbert [Sat, 30 Apr 2016 00:12:18 +0000 (17:12 -0700)]
ipv6: Create ip6_tnl_xmit

This patch renames ip6_tnl_xmit2 to ip6_tnl_xmit and exports it. Other
users like GRE will be able to call this. The original ip6_tnl_xmit
function is renamed to ip6_tnl_start_xmit (this is an ndo_start_xmit
function).

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agogre6: Cleanup GREv6 receive path, call common GRE functions
Tom Herbert [Sat, 30 Apr 2016 00:12:17 +0000 (17:12 -0700)]
gre6: Cleanup GREv6 receive path, call common GRE functions

- Create gre_rcv function. This calls gre_parse_header and ip6gre_rcv.
  - Call ip6_tnl_rcv. Doing this and using gre_parse_header eliminates
    most of the code in ip6gre_rcv.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agogre: Move utility functions to common headers
Tom Herbert [Sat, 30 Apr 2016 00:12:16 +0000 (17:12 -0700)]
gre: Move utility functions to common headers

Several of the GRE functions defined in net/ipv4/ip_gre.c are usable
for IPv6 GRE implementation (that is they are protocol agnostic).

These include:
  - GRE flag handling functions are move to gre.h
  - GRE build_header is moved to gre.h and renamed gre_build_header
  - parse_gre_header is moved to gre_demux.c and renamed gre_parse_header
  - iptunnel_pull_header is taken out of gre_parse_header. This is now
    done by caller. The header length is returned from gre_parse_header
    in an int* argument.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: Cleanup IPv6 tunnel receive path
Tom Herbert [Sat, 30 Apr 2016 00:12:15 +0000 (17:12 -0700)]
ipv6: Cleanup IPv6 tunnel receive path

Some basic changes to make IPv6 tunnel receive path look more like
IPv4 path:
  - Make ip6_tnl_rcv non-static so that GREv6 and others can call it
  - Make ip6_tnl_rcv look like ip_tunnel_rcv
  - Switch to gro_cells_receive
  - Make ip6_tnl_rcv non-static and export it

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'tcp-preempt'
David S. Miller [Mon, 2 May 2016 21:02:26 +0000 (17:02 -0400)]
Merge branch 'tcp-preempt'

Eric Dumazet says:

====================
net: make TCP preemptible

Most of TCP stack assumed it was running from BH handler.

This is great for most things, as TCP behavior is very sensitive
to scheduling artifacts.

However, the prequeue and backlog processing are problematic,
as they need to be flushed with BH being blocked.

To cope with modern needs, TCP sockets have big sk_rcvbuf values,
in the order of 16 MB, and soon 32 MB.
This means that backlog can hold thousands of packets, and things
like TCP coalescing or collapsing on this amount of packets can
lead to insane latency spikes, since BH are blocked for too long.

It is time to make UDP/TCP stacks preemptible.

Note that fast path still runs from BH handler.

v2: Added "tcp: make tcp_sendmsg() aware of socket backlog"
    to reduce latency problems of large sends.

v3: Fixed a typo in tcp_cdg.c
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: make tcp_sendmsg() aware of socket backlog
Eric Dumazet [Fri, 29 Apr 2016 21:16:53 +0000 (14:16 -0700)]
tcp: make tcp_sendmsg() aware of socket backlog

Large sendmsg()/write() hold socket lock for the duration of the call,
unless sk->sk_sndbuf limit is hit. This is bad because incoming packets
are parked into socket backlog for a long time.
Critical decisions like fast retransmit might be delayed.
Receivers have to maintain a big out of order queue with additional cpu
overhead, and also possible stalls in TX once windows are full.

Bidirectional flows are particularly hurt since the backlog can become
quite big if the copy from user space triggers IO (page faults)

Some applications learnt to use sendmsg() (or sendmmsg()) with small
chunks to avoid this issue.

Kernel should know better, right ?

Add a generic sk_flush_backlog() helper and use it right
before a new skb is allocated. Typically we put 64KB of payload
per skb (unless MSG_EOR is requested) and checking socket backlog
every 64KB gives good results.

As a matter of fact, tests with TSO/GSO disabled give very nice
results, as we manage to keep a small write queue and smaller
perceived rtt.

Note that sk_flush_backlog() maintains socket ownership,
so is not equivalent to a {release_sock(sk); lock_sock(sk);},
to ensure implicit atomicity rules that sendmsg() was
giving to (possibly buggy) applications.

In this simple implementation, I chose to not call tcp_release_cb(),
but we might consider this later.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: do not block BH while processing socket backlog
Eric Dumazet [Fri, 29 Apr 2016 21:16:52 +0000 (14:16 -0700)]
net: do not block BH while processing socket backlog

Socket backlog processing is a major latency source.

With current TCP socket sk_rcvbuf limits, I have sampled __release_sock()
holding cpu for more than 5 ms, and packets being dropped by the NIC
once ring buffer is filled.

All users are now ready to be called from process context,
we can unblock BH and let interrupts be serviced faster.

cond_resched_softirq() could be removed, as it has no more user.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosctp: prepare for socket backlog behavior change
Eric Dumazet [Fri, 29 Apr 2016 21:16:51 +0000 (14:16 -0700)]
sctp: prepare for socket backlog behavior change

sctp_inq_push() will soon be called without BH being blocked
when generic socket code flushes the socket backlog.

It is very possible SCTP can be converted to not rely on BH,
but this needs to be done by SCTP experts.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoudp: prepare for non BH masking at backlog processing
Eric Dumazet [Fri, 29 Apr 2016 21:16:50 +0000 (14:16 -0700)]
udp: prepare for non BH masking at backlog processing

UDP uses the generic socket backlog code, and this will soon
be changed to not disable BH when protocol is called back.

We need to use appropriate SNMP accessors.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodccp: do not assume DCCP code is non preemptible
Eric Dumazet [Fri, 29 Apr 2016 21:16:49 +0000 (14:16 -0700)]
dccp: do not assume DCCP code is non preemptible

DCCP uses the generic backlog code, and this will soon
be changed to not disable BH when protocol is called back.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: do not block bh during prequeue processing
Eric Dumazet [Fri, 29 Apr 2016 21:16:48 +0000 (14:16 -0700)]
tcp: do not block bh during prequeue processing

AFAIK, nothing in current TCP stack absolutely wants BH
being disabled once socket is owned by a thread running in
process context.

As mentioned in my prior patch ("tcp: give prequeue mode some care"),
processing a batch of packets might take time, better not block BH
at all.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: do not assume TCP code is non preemptible
Eric Dumazet [Fri, 29 Apr 2016 21:16:47 +0000 (14:16 -0700)]
tcp: do not assume TCP code is non preemptible

We want to to make TCP stack preemptible, as draining prequeue
and backlog queues can take lot of time.

Many SNMP updates were assuming that BH (and preemption) was disabled.

Need to convert some __NET_INC_STATS() calls to NET_INC_STATS()
and some __TCP_INC_STATS() to TCP_INC_STATS()

Before using this_cpu_ptr(net->ipv4.tcp_sk) in tcp_v4_send_reset()
and tcp_v4_send_ack(), we add an explicit preempt disabled section.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'xgene-channel-number'
David S. Miller [Mon, 2 May 2016 20:47:55 +0000 (16:47 -0400)]
Merge branch 'xgene-channel-number'

Iyappan Subramanian says:

====================
drivers: net: xgene: fix: Get channel number from device binding

This patch set adds 'channel' property to get ethernet to CPU channel number,
thus decoupling the Linux driver from static resource selection.

v2: Address review comments from v1
- removed irq reference from Linux driver
- added 'channel' property to get ethernet to CPU channel number

v1:
- Initial version
====================

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodtb: xgene: Add channel property
Iyappan Subramanian [Fri, 29 Apr 2016 18:10:15 +0000 (11:10 -0700)]
dtb: xgene: Add channel property

Added 'channel' property, describing ethernet to CPU channel number.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoDocumentation: dtb: xgene: Add channel property
Iyappan Subramanian [Fri, 29 Apr 2016 18:10:14 +0000 (11:10 -0700)]
Documentation: dtb: xgene: Add channel property

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrivers: net: xgene: Get channel number from device binding
Iyappan Subramanian [Fri, 29 Apr 2016 18:10:13 +0000 (11:10 -0700)]
drivers: net: xgene: Get channel number from device binding

This patch gets ethernet to CPU channel (prefetch buffer number) from
the newly added 'channel' property, thus decoupling Linux driver from
resource management.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'qed-selftests'
David S. Miller [Mon, 2 May 2016 04:16:45 +0000 (00:16 -0400)]
Merge branch 'qed-selftests'

Sudarsana Reddy Kalluru says:

====================
qed/qede: ethtool selftests support.

This series adds the driver support for following selftests:
1. Register test
2. Memory test
3. Clock test
4. Interrupt test
5. Internal loopback test
Patch (1) adds the qed driver infrastructure for selftests. Patches (2) and
(3) add qede driver support for ethtool selftests.

Please consider applying this series to "net-next".
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqede: add implementation for internal loopback test.
Sudarsana Reddy Kalluru [Fri, 29 Apr 2016 00:20:54 +0000 (20:20 -0400)]
qede: add implementation for internal loopback test.

This patch adds the qede implementation for internal loopback test.

Signed-off-by: Sudarsana Reddy Kalluru <sudarsana.kalluru@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Manish Chopra <manish.chopra@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqede: add support for selftests.
Sudarsana Reddy Kalluru [Fri, 29 Apr 2016 00:20:53 +0000 (20:20 -0400)]
qede: add support for selftests.

This patch adds the qede ethtool support for the following tests:
- interrupt test
- memory test
- register test
- clock test

Signed-off-by: Sudarsana Reddy Kalluru <sudarsana.kalluru@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: add infrastructure for device self tests.
Sudarsana Reddy Kalluru [Fri, 29 Apr 2016 00:20:52 +0000 (20:20 -0400)]
qed: add infrastructure for device self tests.

This patch adds the functionality and APIs needed for selftests.
It adds the ability to configure the link-mode which is required for the
implementation of loopback tests. It adds the APIs for clock test,
register test, interrupt test and memory test.

Signed-off-by: Sudarsana Reddy Kalluru <sudarsana.kalluru@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>