openwrt/staging/blogic.git
8 years agoiwlwifi: pcie: replace possible_cpus() with online_cpus() in MSIX mode
Haim Dreyfuss [Tue, 26 Jul 2016 15:03:07 +0000 (18:03 +0300)]
iwlwifi: pcie: replace possible_cpus() with online_cpus() in MSIX mode

In MSIX mode the number of irq depends on the number of
possible cpus existing on the host.
This cause to bug in case there are offline cores.
Take into account only the online CPUs instead.
Also save it in temporary variable.

Fixes: commit 2e5d4a8f61dc ("iwlwifi: pcie: Add new configuration to enable MSIX")
Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: bump max API to 26
Emmanuel Grumbach [Tue, 26 Jul 2016 20:22:01 +0000 (23:22 +0300)]
iwlwifi: mvm: bump max API to 26

The driver now support version 26 of the firmware APIs.

Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: change indentation of iwl_pcie_set_interrupt_capa()
Sara Sharon [Thu, 21 Jul 2016 12:39:29 +0000 (15:39 +0300)]
iwlwifi: pcie: change indentation of iwl_pcie_set_interrupt_capa()

Function is very indented. Go to msi section if needed to avoid
it and by that make the code more readable.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: use setup_timer instead of init_timer and data fields
Wei Yongjun [Tue, 12 Jul 2016 11:40:57 +0000 (11:40 +0000)]
iwlwifi: mvm: use setup_timer instead of init_timer and data fields

Use setup_timer function instead of initializing timer with the function
and data fields

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add the new 8275 series
Oren Givon [Thu, 7 Jul 2016 06:11:56 +0000 (09:11 +0300)]
iwlwifi: add the new 8275 series

Add a new config struct for the new 8275 series and add
the first PCI ID for it.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add the new 9560 series
Oren Givon [Thu, 7 Jul 2016 07:40:27 +0000 (10:40 +0300)]
iwlwifi: add the new 9560 series

Add a new config struct for the new 9560 series and add
the 4 new PCI IDs for it.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: Set affinity mask for rx interrupt vectors per cpu
Haim Dreyfuss [Sun, 13 Mar 2016 15:51:59 +0000 (17:51 +0200)]
iwlwifi: pcie: Set affinity mask for rx interrupt vectors per cpu

In order to utilize the host's CPUs in the most efficient way
we bind each rx interrupt vector to each CPU on the host.
Each rx interrupt is prioritized to execute only on the designated CPU
rather than any CPU.
Processor affinity takes advantage of the fact that some remnants of
a process that was run on a given processor may remain in that
processor's memory state for example, data in the CPU cache after
another process is run on that CPU. Scheduling that process to execute
on the same processor could result in an efficient use of process by
reducing performance-degrading situations such as cache misses
and parallel processing.

Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: free reserved queue on STA removal
Liad Kaufman [Thu, 7 Jul 2016 10:25:59 +0000 (13:25 +0300)]
iwlwifi: mvm: free reserved queue on STA removal

When a STA is removed in DQA mode, if no traffic went through
its reserved queue, the txq continues to be marked as
reserved and no STA can use it.

Make sure that in such a case the reserved queue is marked
as free when the STA is removed.

Fixes: commit 24afba7690e4 ("iwlwifi: mvm: support bss dynamic alloc/dealloc of queues")
Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: fix pending frames tracking on tx resp
Liad Kaufman [Thu, 7 Jul 2016 08:00:26 +0000 (11:00 +0300)]
iwlwifi: mvm: fix pending frames tracking on tx resp

In iwl_mvm_rx_tx_cmd_single(), when checking if a given TID is
aggregated, the driver doesn't check whether or not the queue
itself can be aggregated. For example, a management queue might
be marked as aggregated if TID 0 is aggregated on a (different)
data queue.

Make sure that mgmt frames are sent with TID IWL_TID_NON_QOS,
and in this way make sure no mixups of this sort happen.

Fixes: commit 24afba7690e4 ("iwlwifi: mvm: support bss dynamic alloc/dealloc of queues")
Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: Configure shared interrupt vector in MSIX mode
Haim Dreyfuss [Sun, 20 Mar 2016 15:57:22 +0000 (17:57 +0200)]
iwlwifi: pcie: Configure shared interrupt vector in MSIX mode

In case the OS provides fewer interrupts than requested, different
causes will share the same interrupt vector as follow:
1.One interrupt less: non rx causes shared with FBQ.
2.Two interrupts less: non rx causes shared with FBQ and RSS.
3.More than two interrupts: we will use fewer RSS queues.

Also make the request depend on the number of online CPUs
instead of possible CPUs.

Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support new BA notification response
Sara Sharon [Sun, 17 Jul 2016 11:24:55 +0000 (14:24 +0300)]
iwlwifi: mvm: support new BA notification response

Support new format. TX response will not be sent anymore,
so all needed data is in the BA response.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: merge iwl_queue and iwl_txq
Sara Sharon [Thu, 7 Jul 2016 15:17:45 +0000 (18:17 +0300)]
iwlwifi: pcie: merge iwl_queue and iwl_txq

The original intent was to have the general iwl_queue shared
between RX and TX queues, but it is not the actual status.
Since it is not shared with any struct but iwl_txq, it adds
unnecessary complexity. Merge those structs.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: change byte count table for a000 devices
Sara Sharon [Mon, 4 Jul 2016 11:34:26 +0000 (14:34 +0300)]
iwlwifi: change byte count table for a000 devices

Since TFD was enlarged to 256 bytes, the fetch of the TFD
itself is very expensive.
To make DRAM to SRAM more efficient, bits 12-13 will indicate
the number of 64 byte chunks that should be transferred to
SRAM.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: assign and access a000 TFD & TBs
Sara Sharon [Sun, 26 Jun 2016 10:17:56 +0000 (13:17 +0300)]
iwlwifi: pcie: assign and access a000 TFD & TBs

Previous patch introduced the new formats. This patch
allocates the new structures and adjusts code accordingly.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: introduce trans API to get byte count table
Sara Sharon [Mon, 4 Jul 2016 12:40:11 +0000 (15:40 +0300)]
iwlwifi: introduce trans API to get byte count table

In future HW the byte count table address will be configured
by ucode per queue. Add API to expose the byte count table to
the opmode

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support new shared memory config API
Sara Sharon [Tue, 5 Jul 2016 14:37:58 +0000 (17:37 +0300)]
iwlwifi: mvm: support new shared memory config API

In a000 devices we have 15 fifos, so in the shared memory
config the number of tx fifos in the array was changed
accordingly.
As it is in the middle of the struct, the parsing code needs
to be duplicated.
To minimize the duplication, do not save variables we never
actually use.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: remove dump of locked registers
Sara Sharon [Wed, 13 Jul 2016 13:53:36 +0000 (16:53 +0300)]
iwlwifi: mvm: remove dump of locked registers

Firmware may lock those registers for access. This results
in 9000 devices with a bus stall and an endless loop of 0x5a5a5a.
Don't dump those registers.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: introduce new tfd and tb formats
Sara Sharon [Thu, 23 Jun 2016 13:31:40 +0000 (16:31 +0300)]
iwlwifi: pcie: introduce new tfd and tb formats

New hardware supports bigger TFDs and TBs.
Introduce the new formats and adjust defines and code
relying on old format.
Changing the actual TFD allocation is trickier and
deferred to the next patch.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: fix semicolon.cocci warnings
kbuild test robot [Sun, 10 Jul 2016 21:01:12 +0000 (05:01 +0800)]
iwlwifi: fix semicolon.cocci warnings

drivers/net/wireless/intel/iwlwifi/iwl-io.c:243:2-3: Unneeded semicolon

 Remove unneeded semicolon.

Generated by: scripts/coccinelle/misc/semicolon.cocci

CC: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: don't free queue after delba in dqa
Liad Kaufman [Sun, 26 Jun 2016 11:45:12 +0000 (14:45 +0300)]
iwlwifi: mvm: don't free queue after delba in dqa

In DQA mode, a delBA might free the queue although it
shouldn't. Fix that.

Fixes: cf941e174ee2 ("iwlwifi: mvm: support dqa-mode agg on non-shared queue")
Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: call a different txq_enable function
Sara Sharon [Thu, 30 Jun 2016 13:14:02 +0000 (16:14 +0300)]
iwlwifi: mvm: call a different txq_enable function

Since the SCD_QUEUE_CFG command was introduced the driver
calls iwl_trans_txq_enable_cfg() with a NULL for scd_cfg
parameter.
This makes the transport avoid writing to the SCD pointers,
since it can cause races with firmware, which is also accessing
the registers.
The transport only updates the write pointer in that case.
Fix a wrong call to iwl_trans_txq_enable() which caused a
scd_cfg parameter to be sent to transport, resulting with an
access to SCD registers.

Fixes: 58f2cc57dc6a ("iwlwifi: mvm: support dqa-mode scd queue redirection")
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: remove dead code
Sara Sharon [Wed, 29 Jun 2016 09:23:06 +0000 (12:23 +0300)]
iwlwifi: pcie: remove dead code

If device family is 8000 then iwl_pcie_load_cpu_sections()
won't be called at all (iwl_pcie_load_cpu_sections_8000() is
called in that case) so this piece of code never gets called.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: fix ucode load flow for a000 devices
Sara Sharon [Wed, 29 Jun 2016 09:08:48 +0000 (12:08 +0300)]
iwlwifi: pcie: fix ucode load flow for a000 devices

Turns out we should access TFH relative addresses.
Also, the FH_UCODE_LOAD_STATUS was replaced by
UREG_UCODE_LOAD_STATUS.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: pcie: refrain from SCD accesses
Sara Sharon [Thu, 30 Jun 2016 13:36:24 +0000 (16:36 +0300)]
iwlwifi: pcie: refrain from SCD accesses

Up till now we accessed SCD configuration only for initial
configuration and for enabling command queue.
For a000 generation the command queue is open by default
and firmware configures the rest. No driver SCD accesses
are expected. Make sure this is the case.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add the new 9170 series
Oren Givon [Thu, 7 Jul 2016 06:40:12 +0000 (09:40 +0300)]
iwlwifi: add the new 9170 series

Add a new config struct for the new 9170 series and add
the first PCI ID for it.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add the new 9270 series
Oren Givon [Thu, 7 Jul 2016 06:23:56 +0000 (09:23 +0300)]
iwlwifi: add the new 9270 series

Add a new config struct for the new 9270 series and add
the first PCI ID for it.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add new 9460 series PCI IDs
Oren Givon [Thu, 7 Jul 2016 07:31:17 +0000 (10:31 +0300)]
iwlwifi: add new 9460 series PCI IDs

Add 4 more new 9460 series PCI IDs.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: add a new series 9460 with new PCI ID
Oren Givon [Sun, 26 Jun 2016 11:15:27 +0000 (14:15 +0300)]
iwlwifi: add a new series 9460 with new PCI ID

Add a new series to the 9000 series called 9460.
In addition, add a new PCI ID that is the 9460 new series.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: rename and reorder 9000 series configuration structs
Oren Givon [Thu, 23 Jun 2016 11:51:41 +0000 (14:51 +0300)]
iwlwifi: rename and reorder 9000 series configuration structs

Rename and reorder the 9000 series configuration structs:
- struct containing configuration of 5165 was renamed to 9000.

Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support txq tid owner change
Liad Kaufman [Tue, 8 Mar 2016 12:29:51 +0000 (14:29 +0200)]
iwlwifi: mvm: support txq tid owner change

Every active TXQ is assigned to a TID given through the
SCD_CONFIG_CMD, and acts as an identifier in the FW. However,
there may be cases this ownership needs to be changed.

For example, in the following scenario:
 1. TID x is owner of a queue
 2. Due to a shortage of queues, TID y and z share with x
 3. TID x becomes inactive and needs to be removed from the
    shared queue.
In this scenario, if another queue is freed and traffic on x
continues, we can't allocate it a new queue as long as it is
the owner of the first queue.

Support moving ownership of a TXQ to a different TID (same
STA) without stopping the queue.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: use defines for SCD_CONFIG_CMD enablement
Liad Kaufman [Tue, 8 Mar 2016 08:41:32 +0000 (10:41 +0200)]
iwlwifi: mvm: use defines for SCD_CONFIG_CMD enablement

Due to the addition of another option in the SCD_CONFIG_CMD's
%enable field, change the assignment of this field to use
defines rather than hard-code the value itself.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: re-assign old queues after hw restart in dqa mode
Liad Kaufman [Tue, 2 Feb 2016 14:02:46 +0000 (16:02 +0200)]
iwlwifi: mvm: re-assign old queues after hw restart in dqa mode

When working in DQA mode, if a queue is shared and a HW restart
occurs, there might be a possible race condition between
stations on the queues, and an existing queue might be left
with no queues.

To solve this, make sure in DQA mode to re-assign the same
queues as before the HW restart.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: keep track of tid associated with each queue
Liad Kaufman [Tue, 2 Feb 2016 13:43:32 +0000 (15:43 +0200)]
iwlwifi: mvm: keep track of tid associated with each queue

When sending the SCD_QUEUE_CONFIG command, the queue is
associated to a specific TID. If later there is a need to
use this TID on a different queue instead, it first needs to
be unassociated from the first queue.

Keep track for every queue what TID is associated with it.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: re-aggregate shared queue after unsharing
Liad Kaufman [Wed, 23 Dec 2015 14:03:46 +0000 (16:03 +0200)]
iwlwifi: mvm: re-aggregate shared queue after unsharing

When a shared queue becomes unshared, aggregations should be
re-enabled if they've existed before. Make sure that they do
this, if required.

Signed-off-by: Liad Kaufman <liad.kaufman@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support new paging command format
Sara Sharon [Mon, 4 Jul 2016 08:52:07 +0000 (11:52 +0300)]
iwlwifi: mvm: support new paging command format

For a000 devices there is a support of 64 bit DMA addressing.
The paging command was changed accordingly - support it.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: support GMAC protocol
Ayala Beker [Mon, 11 Apr 2016 08:37:38 +0000 (11:37 +0300)]
iwlwifi: mvm: support GMAC protocol

Add support for installing and removing GMAC key
for newer FW versions that support GCM and MFP.
GMAC provides authentication and integrity for multicast management
frames.

Firmware API was changed, update the driver accordingly.

Signed-off-by: Ayala Beker <ayala.beker@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoiwlwifi: mvm: allow same PN for de-aggregated AMSDU
Sara Sharon [Sun, 6 Mar 2016 13:08:55 +0000 (15:08 +0200)]
iwlwifi: mvm: allow same PN for de-aggregated AMSDU

The 9000 hardware will de-aggregate AMSDUs. In the process
it will copy the mac header "as is" to the new MPDUs.
This means driver should allow the same PN for MPDUs originated
from the same AMSDU.
Do that by incrementing the PN only for the last MPDU in the
sequence.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Thu, 18 Aug 2016 05:17:32 +0000 (01:17 -0400)]
Merge git://git./linux/kernel/git/davem/net

Minor overlapping changes for both merge conflicts.

Resolution work done by Stephen Rothwell was used
as a reference.

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 18 Aug 2016 00:26:58 +0000 (17:26 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Buffers powersave frame test is reversed in cfg80211, fix from Felix
    Fietkau.

 2) Remove bogus WARN_ON in openvswitch, from Jarno Rajahalme.

 3) Fix some tg3 ethtool logic bugs, and one that would cause no
    interrupts to be generated when rx-coalescing is set to 0.  From
    Satish Baddipadige and Siva Reddy Kallam.

 4) QLCNIC mailbox corruption and napi budget handling fix from Manish
    Chopra.

 5) Fix fib_trie logic when walking the trie during /proc/net/route
    output than can access a stale node pointer.  From David Forster.

 6) Several sctp_diag fixes from Phil Sutter.

 7) PAUSE frame handling fixes in mlxsw driver from Ido Schimmel.

 8) Checksum fixup fixes in bpf from Daniel Borkmann.

 9) Memork leaks in nfnetlink, from Liping Zhang.

10) Use after free in rxrpc, from David Howells.

11) Use after free in new skb_array code of macvtap driver, from Jason
    Wang.

12) Calipso resource leak, from Colin Ian King.

13) mediatek bug fixes (missing stats sync init, etc.) from Sean Wang.

14) Fix bpf non-linear packet write helpers, from Daniel Borkmann.

15) Fix lockdep splats in macsec, from Sabrina Dubroca.

16) hv_netvsc bug fixes from Vitaly Kuznetsov, mostly to do with VF
    handling.

17) Various tc-action bug fixes, from CONG Wang.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (116 commits)
  net_sched: allow flushing tc police actions
  net_sched: unify the init logic for act_police
  net_sched: convert tcf_exts from list to pointer array
  net_sched: move tc offload macros to pkt_cls.h
  net_sched: fix a typo in tc_for_each_action()
  net_sched: remove an unnecessary list_del()
  net_sched: remove the leftover cleanup_a()
  mlxsw: spectrum: Allow packets to be trapped from any PG
  mlxsw: spectrum: Unmap 802.1Q FID before destroying it
  mlxsw: spectrum: Add missing rollbacks in error path
  mlxsw: reg: Fix missing op field fill-up
  mlxsw: spectrum: Trap loop-backed packets
  mlxsw: spectrum: Add missing packet traps
  mlxsw: spectrum: Mark port as active before registering it
  mlxsw: spectrum: Create PVID vPort before registering netdevice
  mlxsw: spectrum: Remove redundant errors from the code
  mlxsw: spectrum: Don't return upon error in removal path
  i40e: check for and deal with non-contiguous TCs
  ixgbe: Re-enable ability to toggle VLAN filtering
  ixgbe: Force VLNCTRL.VFE to be set in all VMDq paths
  ...

8 years agoMerge branch 'strparser'
David S. Miller [Wed, 17 Aug 2016 23:37:04 +0000 (19:37 -0400)]
Merge branch 'strparser'

Tom Herbert says:

====================
strp: Stream parser for messages

This patch set introduces a utility for parsing application layer
protocol messages in a TCP stream. This is a generalization of the
mechanism implemented of Kernel Connection Multiplexor.

This patch set adapts KCM to use the strparser. We expect that kTLS
can use this mechanism also. RDS would probably be another candidate
to use a common stream parsing mechanism.

The API includes a context structure, a set of callbacks, utility
functions, and a data ready function. The callbacks include
a parse_msg function that is called to perform parsing (e.g.
BPF parsing in case of KCM), and a rcv_msg function that is called
when a full message has been completed.

For strparser we specify the return codes from the parser to allow
the backend to indicate that control of the socket should be
transferred back to userspace to handle some exceptions in the
stream: The return values are:

      >0 : indicates length of successfully parsed message
       0  : indicates more data must be received to parse the message
       -ESTRPIPE : current message should not be processed by the
          kernel, return control of the socket to userspace which
          can proceed to read the messages itself
       other < 0 : Error is parsing, give control back to userspace
          assuming that synchronization is lost and the stream
          is unrecoverable (application expected to close TCP socket)

There is one issue I haven't been able to fully resolve. If parse_msg
returns ESTRPIPE (wants control back to userspace) the parser may
already have consumed some bytes of the message. There is no way to
put bytes back into the TCP receive queue and tcp_read_sock does not
allow an easy way to peek messages. In lieu of a better solution, we
return ENODATA on the socket to indicate that the data stream is
unrecoverable (application needs to close socket). This condition
should only happen if an application layer message header is split
across two skbuffs and parsing just the first skbuff wasn't sufficient
to determine the that transfer to userspace is needed.

This patch set contains:

  - strparser implementation
  - changes to kcm to use strparser
  - strparser.txt documentation

v2:
  - Add copyright notice to C files
  - Remove GPL module license from strparser.c
  - Add report of rxpause

v3:
  - Restore GPL module license
  - Use EXPORT_SYMBOL_GPL

v4:
  - Removed unused function, changed another to be static as suggested
    by davem
  - Rewoked data_ready to be called from upper layer, no longer requires
    taking over socket data_ready callback as suggested by Lance Chao

Tested:
  - Ran a KCM thrash test for 24 hours. No behavioral or performance
    differences observed.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agostrparser: Documentation
Tom Herbert [Mon, 15 Aug 2016 21:51:03 +0000 (14:51 -0700)]
strparser: Documentation

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agokcm: Use stream parser
Tom Herbert [Mon, 15 Aug 2016 21:51:02 +0000 (14:51 -0700)]
kcm: Use stream parser

Adapt KCM to use the stream parser. This mostly involves removing
the RX handling and setting up the strparser using the interface.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agostrparser: Stream parser for messages
Tom Herbert [Mon, 15 Aug 2016 21:51:01 +0000 (14:51 -0700)]
strparser: Stream parser for messages

This patch introduces a utility for parsing application layer protocol
messages in a TCP stream. This is a generalization of the mechanism
implemented of Kernel Connection Multiplexor.

The API includes a context structure, a set of callbacks, utility
functions, and a data ready function.

A stream parser instance is defined by a strparse structure that
is bound to a TCP socket. The function to initialize the structure
is:

int strp_init(struct strparser *strp, struct sock *csk,
              struct strp_callbacks *cb);

csk is the TCP socket being bound to and cb are the parser callbacks.

The upper layer calls strp_tcp_data_ready when data is ready on the lower
socket for strparser to process. This should be called from a data_ready
callback that is set on the socket:

void strp_tcp_data_ready(struct strparser *strp);

A parser is bound to a TCP socket by setting data_ready function to
strp_tcp_data_ready so that all receive indications on the socket
go through the parser. This is assumes that sk_user_data is set to
the strparser structure.

There are four callbacks.
 - parse_msg is called to parse the message (returns length or error).
 - rcv_msg is called when a complete message has been received
 - read_sock_done is called when data_ready function exits
 - abort_parser is called to abort the parser

The input to parse_msg is an skbuff which contains next message under
construction. The backend processing of parse_msg will parse the
application layer protocol headers to determine the length of
the message in the stream. The possible return values are:

   >0 : indicates length of successfully parsed message
   0  : indicates more data must be received to parse the message
   -ESTRPIPE : current message should not be processed by the
      kernel, return control of the socket to userspace which
      can proceed to read the messages itself
   other < 0 : Error is parsing, give control back to userspace
      assuming that synchronzation is lost and the stream
      is unrecoverable (application expected to close TCP socket)

In the case of error return (< 0) strparse will stop the parser
and report and error to userspace. The application must deal
with the error. To handle the error the strparser is unbound
from the TCP socket. If the error indicates that the stream
TCP socket is at recoverable point (ESTRPIPE) then the application
can read the TCP socket to process the stream. Once the application
has dealt with the exceptions in the stream, it may again bind the
socket to a strparser to continue data operations.

Note that ENODATA may be returned to the application. In this case
parse_msg returned -ESTRPIPE, however strparser was unable to maintain
synchronization of the stream (i.e. some of the message in question
was already read by the parser).

strp_pause and strp_unpause are used to provide flow control. For
instance, if rcv_msg is called but the upper layer can't immediately
consume the message it can hold the message and pause strparser.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ipconfig: Fix more use after free
Thierry Reding [Tue, 16 Aug 2016 14:45:38 +0000 (16:45 +0200)]
net: ipconfig: Fix more use after free

While commit 9c706a49d660 ("net: ipconfig: fix use after free") avoids
the use after free, the resulting code still ends up calling both the
ic_setup_if() and ic_setup_routes() after calling ic_close_devs(), and
access to the device is still required.

Move the call to ic_close_devs() to the very end of the function.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'tc_action-fixes'
David S. Miller [Wed, 17 Aug 2016 23:27:58 +0000 (19:27 -0400)]
Merge branch 'tc_action-fixes'

Cong Wang says:

====================
net_sched: tc action fixes and updates

This patchset fixes a few regressions caused by the previous
code refactor and more. Thanks to Jamal for catching them!

Note, patch 3/7 and 4/7 are not strictly necessary for this patchset,
I just want to carry them together.

---
v4: adjust an indention for Jamal
    add two more patches

v3: avoid list for fast path, suggested by Jamal

v2: replace flex_array with regular dynamic array
    keep tcf_action_stats_update() in act_api.h
    fix macro typos found by Amir
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: allow flushing tc police actions
Roman Mashak [Sun, 14 Aug 2016 05:35:02 +0000 (22:35 -0700)]
net_sched: allow flushing tc police actions

The act_police uses its own code to walk the
action hashtable, which leads to that we could
not flush standalone tc police actions, so just
switch to tcf_generic_walker() like other actions.

(Joint work from Roman and Cong.)

Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: unify the init logic for act_police
WANG Cong [Sun, 14 Aug 2016 05:35:01 +0000 (22:35 -0700)]
net_sched: unify the init logic for act_police

Jamal reported a crash when we create a police action
with a specific index, this is because the init logic
is not correct, we should always create one for this
case. Just unify the logic with other tc actions.

Fixes: a03e6fe56971 ("act_police: fix a crash during removal")
Reported-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: convert tcf_exts from list to pointer array
WANG Cong [Sun, 14 Aug 2016 05:35:00 +0000 (22:35 -0700)]
net_sched: convert tcf_exts from list to pointer array

As pointed out by Jamal, an action could be shared by
multiple filters, so we can't use list to chain them
any more after we get rid of the original tc_action.
Instead, we could just save pointers to these actions
in tcf_exts, since they are refcount'ed, so convert
the list to an array of pointers.

The "ugly" part is the action API still accepts list
as a parameter, I just introduce a helper function to
convert the array of pointers to a list, instead of
relying on the C99 feature to iterate the array.

Fixes: a85a970af265 ("net_sched: move tc_action into tcf_common")
Reported-by: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: move tc offload macros to pkt_cls.h
WANG Cong [Sun, 14 Aug 2016 05:34:59 +0000 (22:34 -0700)]
net_sched: move tc offload macros to pkt_cls.h

struct tcf_exts belongs to filters, should not be visible
to plain tc actions.

Cc: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: fix a typo in tc_for_each_action()
WANG Cong [Sun, 14 Aug 2016 05:34:58 +0000 (22:34 -0700)]
net_sched: fix a typo in tc_for_each_action()

It is harmless because all users pass 'a' to this macro.

Fixes: 00175aec941e ("net/sched: Macro instead of CONFIG_NET_CLS_ACT ifdef")
Cc: Amir Vadai <amir@vadai.me>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: remove an unnecessary list_del()
WANG Cong [Sun, 14 Aug 2016 05:34:57 +0000 (22:34 -0700)]
net_sched: remove an unnecessary list_del()

This list_del() for tc action is not needed actually,
because we only use this list to chain bulk operations,
therefore should not be carried for latter operations.

Fixes: ec0595cc4495 ("net_sched: get rid of struct tcf_common")
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: remove the leftover cleanup_a()
WANG Cong [Sun, 14 Aug 2016 05:34:56 +0000 (22:34 -0700)]
net_sched: remove the leftover cleanup_a()

After refactoring tc_action into tcf_common, we no
longer need to cleanup temporary "actions" in list,
they are permanently stored in the hashtable.

Fixes: a85a970af265 ("net_sched: move tc_action into tcf_common")
Reported-by: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'batadv-next-for-davem-20160816' of git://git.open-mesh.org/linux-merge
David S. Miller [Wed, 17 Aug 2016 23:22:13 +0000 (19:22 -0400)]
Merge tag 'batadv-next-for-davem-20160816' of git://git.open-mesh.org/linux-merge

Simon Wunderlich says:

====================
pull request for net-next: batman-adv 2016-08-16

This feature patchset is all about adding netlink support, which should
supersede our debugfs configuration interface in the long run. It is
especially necessary when batman-adv should be used in different
namespaces, since debugfs can not differentiate between those.

More specifically, the following changes are included:

 - Two fixes for namespace handling by Andrew Lunn, checking also the
   namespaces for parent interfaces, and supress debugfs entries
   for non-default netns

 - Implement various netlink commands for the new interface, by
   Matthias Schiffer, Andrew Lunn, Sven Eckelmann and Simon Wunderlich
   (13 patches):
    * routing algorithm list
    * hardif list
    * translation tables (local and global)
    * TTVN for the translation tables
    * originator and neighbor tables for B.A.T.M.A.N. IV
      and B.A.T.M.A.N. V
    * gateway dump functionality for B.A.T.M.A.N. IV
      and B.A.T.M.A.N. V
    * Bridge Loop Avoidance claims, and corresponding BLA group
    * Bridge Loop Avoidance backbone tables

 - Finally, mark batman-adv as netns compatible, by Andrew Lunn (1 patch)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net...
David S. Miller [Wed, 17 Aug 2016 23:20:24 +0000 (19:20 -0400)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/net-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2016-08-16

This series contains fixes to e1000e, igb, ixgbe and i40e.

Kshitiz Gupta provides a fix for igb to resolve the PHY delay compensation
math in several functions.

Jarod Wilson provides a fix for e1000e which had to broken up into 2
patches, first is prepares the driver for expanding the list of NICs
that have occasional ~10 hour clock jumps when being used for PTP.
Second patch actually fixes i218 silicon which has been experiencing
the clock jumps while using PTP.

Alex provides 2 patches for ixgbe now that he is back at Intel.  First
fixes setting VLNCTRL.VFE bit, which was left unchanged in earlier patches
which resulted in disabling VLAN filtering for all the VFs.  Second
corrects the support for disabling the VLAN tag filtering via the
feature bit.

Lastly, David fixes i40e which was causing a kernel panic when
non-contiguous traffic classes or traffic classes not starting with TC0,
were configured on a link partner switch.  To fix this, changed the
logic when determining the total number of TCs enabled.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'mlxsw-fixes'
David S. Miller [Wed, 17 Aug 2016 23:18:34 +0000 (19:18 -0400)]
Merge branch 'mlxsw-fixes'

Jiri Pirko says:

====================
mlxsw: IPv4 UC router fixes

Ido says:
Patches 1-3 fix a long standing problem in the driver's init sequence,
which manifests itself quite often when routing daemons try to configure
an IP address on registered netdevs that don't yet have an associated
vPort.

Patches 4-9 add missing packet traps for the router to work properly and
also fix ordering issue following the recent changes to the driver's init
sequence.

The last patch isn't related to the router, but fixes a general problem
in which under certain conditions packets aren't trapped to CPU.

v1->v2:
- Change order of patch 7
- Add patch 6 following Ilan's comment
- Add patchset name and cover letter
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Allow packets to be trapped from any PG
Ido Schimmel [Wed, 17 Aug 2016 14:39:37 +0000 (16:39 +0200)]
mlxsw: spectrum: Allow packets to be trapped from any PG

When packets enter the device they are classified to a priority group
(PG) buffer based on their PCP value. After their egress port and
traffic class are determined they are moved to the switch's shared
buffer and await transmission, if:

(Ingress{Port}.Usage < Thres && Ingress{Port,PG}.Usage < Thres &&
 Egress{Port}.Usage < Thres && Egress{Port,TC}.Usage < Thres)
||
(Ingress{Port}.Usage < Min || Ingress{Port,PG} < Min ||
 Egress{Port}.Usage < Min || Egress{Port,TC}.Usage < Min)

Packets scheduled to transmission through CPU port (trapped to CPU) use
traffic class 7, which has a zero maximum and minimum quotas. However,
when such packets arrive from PG 0 they are admitted to the shared
buffer as PG 0 has a non-zero minimum quota.

Allow all packets to be trapped to the CPU - regardless of the PG they
were classified to - by assigning a 10KB minimum quota for CPU port and
TC7.

Fixes: 8e8dfe9fdf06 ("mlxsw: spectrum: Add IEEE 802.1Qaz ETS support")
Reported-by: Tamir Winetroub <tamirw@mellanox.com>
Tested-by: Tamir Winetroub <tamirw@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Unmap 802.1Q FID before destroying it
Ido Schimmel [Wed, 17 Aug 2016 14:39:36 +0000 (16:39 +0200)]
mlxsw: spectrum: Unmap 802.1Q FID before destroying it

Before destroying the 802.1Q FID we should first remove the VID-to-FID
mapping. This makes mlxsw_sp_fid_destroy() symmetric with regards to
mlxsw_sp_fid_create().

Fixes: 14d39461b3f4 ("mlxsw: spectrum: Use per-FID struct for the VLAN-aware bridge")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Add missing rollbacks in error path
Ido Schimmel [Wed, 17 Aug 2016 14:39:35 +0000 (16:39 +0200)]
mlxsw: spectrum: Add missing rollbacks in error path

While going over the code I noticed we are missing two rollbacks in the
port's creation error path. Add them and adjust the place of one of them
in the port's removal sequence so that both are symmetric.

Fixes: 56ade8fe3fe1 ("mlxsw: spectrum: Add initial support for Spectrum ASIC")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: reg: Fix missing op field fill-up
Jiri Pirko [Wed, 17 Aug 2016 14:39:34 +0000 (16:39 +0200)]
mlxsw: reg: Fix missing op field fill-up

Ralue pack function needs to set op, otherwise it is 0 for add always.

Fixes: d5a1c749d22 ("mlxsw: reg: Add Router Algorithmic LPM Unicast Entry Register definition")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Trap loop-backed packets
Ido Schimmel [Wed, 17 Aug 2016 14:39:33 +0000 (16:39 +0200)]
mlxsw: spectrum: Trap loop-backed packets

One of the conditions to generate an ICMP Redirect Message is that "the
packet is being forwarded out the same physical interface that it was
received from" (RFC 1812).

Therefore, we need to be able to trap such packets and let the kernel
decide what to do with them.

For each RIF, enable the loop-back filter, which will raise the LBERROR
trap whenever the ingress RIF equals the egress RIF.

Fixes: 99724c18fc66 ("mlxsw: spectrum: Introduce support for router interfaces")
Reported-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Add missing packet traps
Elad Raz [Wed, 17 Aug 2016 14:39:32 +0000 (16:39 +0200)]
mlxsw: spectrum: Add missing packet traps

Add the following traps:

1) MTU Error: Trap packets whose size is bigger than the egress RIF's
MTU. If DF bit isn't set, traffic will continue to be routed in slow
path.

2) TTL Error: Trap packets whose TTL expired. This allows traceroute to
work properly.

3) OSPF packets.

Fixes: 7b27ce7bb9cd ("mlxsw: spectrum: Add traps needed for router implementation")
Signed-off-by: Elad Raz <eladr@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Mark port as active before registering it
Ido Schimmel [Wed, 17 Aug 2016 14:39:31 +0000 (16:39 +0200)]
mlxsw: spectrum: Mark port as active before registering it

Commit bbf2a4757b30 ("mlxsw: spectrum: Initialize ports at the end of
init sequence") moved ports initialization to the end of the init
sequence, which means ports are the first to be removed during fini.

Since the FDB delayed work is still active when ports are removed it's
possible for it to process FDB notifications of inactive ports,
resulting in a warning message.

Fix that by marking ports as inactive only after unregistering them. The
NETDEV_UNREGISTER event will invoke bridge's driver port removal
sequence that will cause the FDB (and FDB notifications) to be flushed.

Fixes: bbf2a4757b30 ("mlxsw: spectrum: Initialize ports at the end of init sequence")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Create PVID vPort before registering netdevice
Ido Schimmel [Wed, 17 Aug 2016 14:39:30 +0000 (16:39 +0200)]
mlxsw: spectrum: Create PVID vPort before registering netdevice

After registering a netdevice it's possible for user space applications
to configure an IP address on it. From the driver's perspective, this
means a router interface (RIF) should be created for the PVID vPort.

Therefore, we must create the PVID vPort before registering the
netdevice.

Fixes: 99724c18fc66 ("mlxsw: spectrum: Introduce support for router interfaces")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Remove redundant errors from the code
Ido Schimmel [Wed, 17 Aug 2016 14:39:29 +0000 (16:39 +0200)]
mlxsw: spectrum: Remove redundant errors from the code

Currently, when device configuration fails we emit errors to the kernel
log despite the fact we already get these from the EMAD transaction
layer, so remove them.

In addition to being unnecessary, removing these error messages will
allow us to reuse mlxsw_sp_port_add_vid() to create the PVID vPort
before registering the netdevice.

Fixes: 99724c18fc66 ("mlxsw: spectrum: Introduce support for router interfaces")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Don't return upon error in removal path
Ido Schimmel [Wed, 17 Aug 2016 14:39:28 +0000 (16:39 +0200)]
mlxsw: spectrum: Don't return upon error in removal path

When removing a VLAN filter from the device we shouldn't return upon the
first error we encounter, as otherwise we'll have resources that will
never be freed nor used.

Instead, we should keep trying to free as much resources as possible in
a best effort mode.

Remove the error message as well, since we already get these from the
EMAD transaction code.

Fixes: 99724c18fc66 ("mlxsw: spectrum: Introduce support for router interfaces")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'for-v4.8-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/sre/linux...
Linus Torvalds [Wed, 17 Aug 2016 19:10:22 +0000 (12:10 -0700)]
Merge tag 'for-v4.8-rc' of git://git./linux/kernel/git/sre/linux-power-supply

Pull power supply fixes from Sebastian Reichel.

* tag 'for-v4.8-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-power-supply:
  power_supply: tps65217-charger: fix missing platform_set_drvdata()
  power: reset: hisi-reboot: Unmap region obtained by of_iomap
  power: reset: reboot-mode: fix build error of missing ioremap/iounmap on UM
  power: supply: max17042_battery: fix model download bug.

8 years agoMerge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
Linus Torvalds [Tue, 16 Aug 2016 22:51:57 +0000 (15:51 -0700)]
Merge tag 'for_linus' of git://git./linux/kernel/git/mst/vhost

Pull virtio/vhost fixes from Michael Tsirkin:
 - test fixes
 - a vsock fix

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
  tools/virtio: add dma stubs
  vhost/test: fix after swiotlb changes
  vhost/vsock: drop space available check for TX vq
  ringtest: test build fix

8 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Tue, 16 Aug 2016 22:50:22 +0000 (15:50 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Martin Schwidefsky:
 "A couple of bug fixes, minor cleanup and a change to the default
  config"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/dasd: fix failing CUIR assignment under LPAR
  s390/pageattr: handle numpages parameter correctly
  s390/dasd: fix hanging device after clear subchannel
  s390/qdio: avoid reschedule of outbound tasklet once killed
  s390/qdio: remove checks for ccw device internal state
  s390/qdio: fix double return code evaluation
  s390/qdio: get rid of spin_lock_irqsave usage
  s390/cio: remove subchannel_id from ccw_device_private
  s390/qdio: obtain subchannel_id via ccw_device_get_schid()
  s390/cio: stop using subchannel_id from ccw_device_private
  s390/config: make the vector optimized crc function builtin
  s390/lib: fix memcmp and strstr
  s390/crc32-vx: Fix checksum calculation for small sizes
  s390: clarify compressed image code path

8 years agoMerge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Tue, 16 Aug 2016 20:37:26 +0000 (13:37 -0700)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fixes from Herbert Xu:
 "This fixes the following issues:

   - Missing ULL suffixes for 64-bit constants in sha3.
   - Two caam AEAD regressions.
   - Bogus setkey hooks in non-hmac caam hashes.
   - Missing kbuild dependency for powerpc crc32c"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: caam - fix non-hmac hashes
  crypto: powerpc - CRYPT_CRC32C_VPMSUM should depend on ALTIVEC
  crypto: caam - defer aead_set_sh_desc in case of zero authsize
  crypto: caam - fix echainiv(authenc) encrypt shared descriptor
  crypto: sha3 - Add missing ULL suffixes for 64-bit constants

8 years agoi40e: check for and deal with non-contiguous TCs
Dave Ertman [Fri, 12 Aug 2016 16:56:32 +0000 (09:56 -0700)]
i40e: check for and deal with non-contiguous TCs

The i40e driver was causing a kernel panic when
non-contiguous Traffic Classes, or Traffic Classes not
starting with TC0, were configured on a link partner switch.
i40e does not support non-contiguous TCs.

To fix this, the patch changes the logic when determining
the total number of TCs enabled.  Before, this would use the
highest TC number enabled and assume that all TCs below it were
also enabled.  Now, we create a bitmask of enabled TCs and scan
it to determine not only the number of TCs, but also if the set
of enabled TCs starts at zero and is contiguous.  If not, then
DCB is disabled by only returning one TC.

Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoixgbe: Re-enable ability to toggle VLAN filtering
Alexander Duyck [Fri, 12 Aug 2016 16:53:39 +0000 (09:53 -0700)]
ixgbe: Re-enable ability to toggle VLAN filtering

Back when I submitted the GSO code I messed up and dropped the support for
disabling the VLAN tag filtering via the feature bit.  This patch
re-enables the use of the NETIF_F_HW_VLAN_CTAG_FILTER to enable/disable the
VLAN filtering independent of toggling promiscuous mode.

Fixes: b83e30104b ("ixgbe/ixgbevf: Add support for GSO partial")
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoixgbe: Force VLNCTRL.VFE to be set in all VMDq paths
Alexander Duyck [Thu, 11 Aug 2016 21:51:56 +0000 (14:51 -0700)]
ixgbe: Force VLNCTRL.VFE to be set in all VMDq paths

When I was adding the code for enabling VLAN promiscuous mode with SR-IOV
enabled I had inadvertently left the VLNCTRL.VFE bit unchanged as I has
assumed there was code in another path that was setting it when we enabled
SR-IOV.  This wasn't the case and as a result we were just disabling VLAN
filtering for all the VFs apparently.

Also the previous patches were always clearing CFIEN which was always set
to 0 by the hardware anyway so I am dropping the redundant bit clearing.

Fixes: 16369564915a ("ixgbe: Add support for VLAN promiscuous with SR-IOV")
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoMerge tag 'pinctrl-v4.8-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Tue, 16 Aug 2016 18:49:02 +0000 (11:49 -0700)]
Merge tag 'pinctrl-v4.8-2' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fixes from Linus Walleij:
 "Here are a few pin control fixes for the v4.8 series, nothing special
  about them:

   - Add the missing <linux/io.h> header to the Intel Merrifield driver
     to get rid of build mess.

   - Drop two instances of pinctrl_unregister() called for drivers using
     devm_* resource management.

   - Remove the default debounce time for the AMD driver"

* tag 'pinctrl-v4.8-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: intel: merrifield: Add missed header
  pinctrl/amd: Remove the default de-bounce time
  pinctrl: pistachio: Drop pinctrl_unregister for devm_ registered device
  pinctrl: meson: Drop pinctrl_unregister for devm_ registered device

8 years agoe1000e: fix PTP on e1000_pch_lpt variants
Jarod Wilson [Tue, 26 Jul 2016 18:25:35 +0000 (14:25 -0400)]
e1000e: fix PTP on e1000_pch_lpt variants

I've got reports that the Intel I-218V NIC in Intel NUC5i5RYH systems used
as a PTP slave experiences random ~10 hour clock jumps, which are resolved
if the same workaround for the 82574 and 82583 is employed, so set the
appropriate flag2 in e1000_pch_lpt_info too.

Reported-by: Rupesh Patel <rupatel@redhat.com>
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: factor out systim sanitization
Jarod Wilson [Tue, 26 Jul 2016 18:25:34 +0000 (14:25 -0400)]
e1000e: factor out systim sanitization

This is prepatory work for an expanding list of adapter families that have
occasional ~10 hour clock jumps when being used for PTP. Factor out the
sanitization function and convert to using a feature (bug) flag, per
suggestion from Jesse Brandeburg.

Littering functional code with device-specific checks is much messier than
simply checking a flag, and having device-specific init set flags as needed.
There are probably a number of other cases in the e1000e code that
could/should be converted similarly.

Suggested-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigb: fix adjusting PTP timestamps for Tx/Rx latency
Kshitiz Gupta [Sat, 16 Jul 2016 07:23:45 +0000 (02:23 -0500)]
igb: fix adjusting PTP timestamps for Tx/Rx latency

Fix PHY delay compensation math in igb_ptp_tx_hwtstamp() and
igb_ptp_rx_rgtstamp. Add PHY delay compensation in
igb_ptp_rx_pktstamp().

In the IGB driver, there are two functions that retrieve timestamps
received by the PHY - igb_ptp_rx_rgtstamp() and igb_ptp_rx_pktstamp().
The previous commit only changed igb_ptp_rx_rgtstamp(), and the change
was incorrect.

There are two instances in which PHY delay compensations should be
made:

- Before the packet transmission over the PHY, the latency between
  when the packet is timestamped and transmission of the packets,
  should be an add operation, but it is currently a subtract.

- After the packets are received from the PHY, the latency between
  the receiving and timestamping of the packets should be a subtract
  operation, but it is currently an add.

Signed-off-by: Kshitiz Gupta <kshitiz.gupta@ni.com>
Fixes: 3f544d2 (igb: adjust ptp timestamps for tx/rx latency)
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoMerge branch 'mediatek-fixes'
David S. Miller [Tue, 16 Aug 2016 06:02:45 +0000 (23:02 -0700)]
Merge branch 'mediatek-fixes'

Sean Wang says:

====================
mediatek: Fix warning and issue

This patch set fixes the following warning and issues

v1 -> v2: Fix message typos and add coverletter

v2 -> v3: Split from the previous series for submitting bug fixes
as a series targeting 'net'
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: mediatek: fix runtime warning raised by inconsistent struct device...
sean.wang@mediatek.com [Tue, 16 Aug 2016 05:55:15 +0000 (13:55 +0800)]
net: ethernet: mediatek: fix runtime warning raised by inconsistent struct device pointers passed to DMA API

Runtime warning occurs if DMA-API debug feature is enabled that would be
raised by pointers passed to DMA API as arguments to inconsistent struct
device objects, so that the patch makes them usage aligned between DMA
operations such as dma_map_*() and dma_unmap_*() to eliminate the warning.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: mediatek: fix flow control settings on GMAC0 is not being enabled...
sean.wang@mediatek.com [Tue, 16 Aug 2016 05:55:14 +0000 (13:55 +0800)]
net: ethernet: mediatek: fix flow control settings on GMAC0 is not being enabled properly

Commit 08ef55c6f257acf3bdc6940813f80e8f0f5d90ec
("net-next: mediatek: fix gigabit and flow control advertisement")
had supported proper flow control settings for GMAC1. But for GMAC0,

1.GMAC0 shares the common logic with GMAC1 inside mtk_phy_link_adjust()
to adapt various settings for the target phy.

2.GMAC0 uses fixed-phy to connect to a builtin gigabit switch with
fixed link speed as commit 0c72c50f6f93b0c3daa9ea35d89ab3a933c7b5a0
("net-next: mediatek: add fixed-phy support") describes.

3.However, fixed-phy doesn't enable SUPPORTED_Pause & SUPPORTED_Asym_Pause
supported flag on default that would cause mtk_phy_link_adjust() not to
enable flow control setting on GMAC0 properly and cause packet dropped
when high traffic.

Due to these reasons, the patch adds SUPPORTED_Pause & SUPPORTED_Asym_Pause
supported flags on fixed-phy used by the driver to have proper handling on
the both GMAC with the shared common logic.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: mediatek: fix RMII mode and add REVMII supported by GMAC
sean.wang@mediatek.com [Tue, 16 Aug 2016 05:55:13 +0000 (13:55 +0800)]
net: ethernet: mediatek: fix RMII mode and add REVMII supported by GMAC

The patch fixes up the incorrect setup of reduced MII (RMII) on GMAC
and adds the supplement for the setup of reverse MII (REVMII) on GMAC
, and rearranges the error handling for invalid PHY argument.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'dsa-abstract-PHY-access'
David S. Miller [Mon, 15 Aug 2016 23:43:56 +0000 (16:43 -0700)]
Merge branch 'dsa-abstract-PHY-access'

Vivien Didelot says:

====================
net: dsa: abstract PHY accesses

The Marvell 88E6xxx switch chips have different way to access the PHY
devices registers.

Old chips use a direct access to the PHY registers. Next chips have a
PHY Polling Unit (PPU) which needs to be disabled before accessing PHY
registers. Newer chips have an indirect access to the PHY devices so
that disabling the PPU is not necessary.

This patchset abstracts these accesses behind a new mv88e6xxx_phy_* API.

It also has the side effect to fix the temperature access code for
88E61xx chips which were using the wrong PHY access functions.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: use the new PHY API
Vivien Didelot [Mon, 15 Aug 2016 21:19:02 +0000 (17:19 -0400)]
net: dsa: mv88e6xxx: use the new PHY API

This commit replaces every MDIO direct or indirect access with the new
generic mv88e6xxx_phy_* routines.

This allows us to get rid of the mv88e6xxx_mdio_{read,write}_{,in}direct
and {_,}mv88e6xxx_mdio_page_{read,write} functions.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: describe PHY page and SerDes
Vivien Didelot [Mon, 15 Aug 2016 21:19:01 +0000 (17:19 -0400)]
net: dsa: mv88e6xxx: describe PHY page and SerDes

Add mv88e6xxx_phy_page_{read,write} routines and use them to access the
SerDes PHY device registers.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: abstract PHY ops
Vivien Didelot [Mon, 15 Aug 2016 21:19:00 +0000 (17:19 -0400)]
net: dsa: mv88e6xxx: abstract PHY ops

Old chips use a direct access to the PHY devices registers. Next chips
have a PHY Polling Unit (PPU) which needs to be disabled before
accessing PHY registers. Newer chips have an indirect access to the PHY
devices so that disabling the PPU is not necessary.

Introduce a new phy_ops structure in the chip to describe the required
PHY access routines.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: rework Global2 SMI PHY access
Vivien Didelot [Mon, 15 Aug 2016 21:18:59 +0000 (17:18 -0400)]
net: dsa: mv88e6xxx: rework Global2 SMI PHY access

Describe the presence of the Global2 SMI PHY registers, used to
indirectly access the internal SMI devices registers on some chips.

Also temporarily forward declare mv88e6xxx_g2_smi_phy_{read,write} to
use them in mv88e6xxx_mdio_{read,write}_indirect, before getting rid of
the later.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: describe Multi-chip registers
Vivien Didelot [Mon, 15 Aug 2016 21:18:58 +0000 (17:18 -0400)]
net: dsa: mv88e6xxx: describe Multi-chip registers

Add flags to describe the presence of SMI Command and Data registers
used to indirectly access internal SMI devices registers when the switch
SMI address is not zero.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: rename _mv88e6xxx_wait
Vivien Didelot [Mon, 15 Aug 2016 21:18:57 +0000 (17:18 -0400)]
net: dsa: mv88e6xxx: rename _mv88e6xxx_wait

Now that there is no locked version of the wait routine anymore, rename
the _ prefixed version and make it use the new read API.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: mediatek: remove unnecessary platform_set_drvdata()
Wei Yongjun [Mon, 15 Aug 2016 22:51:48 +0000 (22:51 +0000)]
net: mediatek: remove unnecessary platform_set_drvdata()

The driver core clears the driver data to NULL after device_release
or on probe failure. Thus, it is not needed to manually clear the
device driver data to NULL.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: thunderx: Remove unnecessary pci_set_drvdata()
Wei Yongjun [Mon, 15 Aug 2016 22:51:29 +0000 (22:51 +0000)]
net: thunderx: Remove unnecessary pci_set_drvdata()

The driver core clears the driver data to NULL after device_release
or on probe failure. Thus, it is not needed to manually clear the
device driver data to NULL.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ena: Fix error return code in ena_device_init()
Wei Yongjun [Mon, 15 Aug 2016 22:51:04 +0000 (22:51 +0000)]
net: ena: Fix error return code in ena_device_init()

Fix to return a negative error code from the invalid dma width
error handling case instead of 0.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ena: Remove unnecessary pci_set_drvdata()
Wei Yongjun [Mon, 15 Aug 2016 22:50:34 +0000 (22:50 +0000)]
net: ena: Remove unnecessary pci_set_drvdata()

The driver core clears the driver data to NULL after device_release
or on probe failure. Thus, it is not needed to manually clear the
device driver data to NULL.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: phy: Fix return value check in xgmiitorgmii_probe()
Wei Yongjun [Mon, 15 Aug 2016 22:34:57 +0000 (22:34 +0000)]
net: phy: Fix return value check in xgmiitorgmii_probe()

In case of error, the function of_parse_phandle() returns NULL
pointer not ERR_PTR(). The IS_ERR() test in the return value check
should be replaced with NULL test.

Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agopower_supply: tps65217-charger: fix missing platform_set_drvdata()
Wei Yongjun [Tue, 26 Jul 2016 14:49:04 +0000 (14:49 +0000)]
power_supply: tps65217-charger: fix missing platform_set_drvdata()

Add missing platform_set_drvdata() in tps65217_charger_probe(), otherwise
calling platform_get_drvdata() in remove returns NULL.

This is detected by Coccinelle semantic patch.

Fixes: 3636859b280c ("power_supply: Add support for tps65217-charger")
Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Signed-off-by: Sebastian Reichel <sre@kernel.org>
8 years agotipc: fix NULL pointer dereference in shutdown()
Vegard Nossum [Sat, 23 Jul 2016 06:15:04 +0000 (08:15 +0200)]
tipc: fix NULL pointer dereference in shutdown()

tipc_msg_create() can return a NULL skb and if so, we shouldn't try to
call tipc_node_xmit_skb() on it.

    general protection fault: 0000 [#1] PREEMPT SMP KASAN
    CPU: 3 PID: 30298 Comm: trinity-c0 Not tainted 4.7.0-rc7+ #19
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014
    task: ffff8800baf09980 ti: ffff8800595b8000 task.ti: ffff8800595b8000
    RIP: 0010:[<ffffffff830bb46b>]  [<ffffffff830bb46b>] tipc_node_xmit_skb+0x6b/0x140
    RSP: 0018:ffff8800595bfce8  EFLAGS: 00010246
    RAX: 0000000000000000 RBX: 0000000000000000 RCX: 000000003023b0e0
    RDX: 0000000000000000 RSI: dffffc0000000000 RDI: ffffffff83d12580
    RBP: ffff8800595bfd78 R08: ffffed000b2b7f32 R09: 0000000000000000
    R10: fffffbfff0759725 R11: 0000000000000000 R12: 1ffff1000b2b7f9f
    R13: ffff8800595bfd58 R14: ffffffff83d12580 R15: dffffc0000000000
    FS:  00007fcdde242700(0000) GS:ffff88011af80000(0000) knlGS:0000000000000000
    CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
    CR2: 00007fcddde1db10 CR3: 000000006874b000 CR4: 00000000000006e0
    DR0: 00007fcdde248000 DR1: 00007fcddd73d000 DR2: 00007fcdde248000
    DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000090602
    Stack:
     0000000000000018 0000000000000018 0000000041b58ab3 ffffffff83954208
     ffffffff830bb400 ffff8800595bfd30 ffffffff8309d767 0000000000000018
     0000000000000018 ffff8800595bfd78 ffffffff8309da1a 00000000810ee611
    Call Trace:
     [<ffffffff830c84a3>] tipc_shutdown+0x553/0x880
     [<ffffffff825b4a3b>] SyS_shutdown+0x14b/0x170
     [<ffffffff8100334c>] do_syscall_64+0x19c/0x410
     [<ffffffff83295ca5>] entry_SYSCALL64_slow_path+0x25/0x25
    Code: 90 00 b4 0b 83 c7 00 f1 f1 f1 f1 4c 8d 6d e0 c7 40 04 00 00 00 f4 c7 40 08 f3 f3 f3 f3 48 89 d8 48 c1 e8 03 c7 45 b4 00 00 00 00 <80> 3c 30 00 75 78 48 8d 7b 08 49 8d 75 c0 48 b8 00 00 00 00 00
    RIP  [<ffffffff830bb46b>] tipc_node_xmit_skb+0x6b/0x140
     RSP <ffff8800595bfce8>
    ---[ end trace 57b0484e351e71f1 ]---

I feel like we should maybe return -ENOMEM or -ENOBUFS, but I'm not sure
userspace is equipped to handle that. Anyway, this is better than a GPF
and looks somewhat consistent with other tipc_msg_create() callers.

Signed-off-by: Vegard Nossum <vegard.nossum@oracle.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'hv_netvsc-VF-removal-fixes'
David S. Miller [Mon, 15 Aug 2016 20:48:08 +0000 (13:48 -0700)]
Merge branch 'hv_netvsc-VF-removal-fixes'

Vitaly Kuznetsov says:

====================
hv_netvsc: fixes for VF removal path

Kernel crash is reported after VF is removed and detached from netvsc
device. Turns out we have multiple different (but related) issues on the
VF removal path which I'm trying to address with PATCHes 2-5 of this
series. PATCH1 is required to support the change.

Changes since v1:
- Re-arrange patches in the series to not introduce new issues [David Miller]
- Add PATCH5 which fixes a new issue I discovered while testing.
- Add Haiyang' A-b tags to PATCH1-4

With regards to Stephen's suggestion: I believe that switching to using RCU
and eliminating vf_use_cnt/vf_inject is the right thing to do long-term, we
can either put this on top of this series or do it later in net-next.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: fix bonding devices check in netvsc_netdev_event()
Vitaly Kuznetsov [Mon, 15 Aug 2016 15:48:43 +0000 (17:48 +0200)]
hv_netvsc: fix bonding devices check in netvsc_netdev_event()

Bonding driver sets IFF_BONDING on both master (the bonding device) and
slave (the real NIC) devices and in netvsc_netdev_event() we want to skip
master devices only. Currently, there is an uncertainty when a slave
interface is removed: if bonding module comes first in netdev_chain it
clears IFF_BONDING flag on the netdev and netvsc_netdev_event() correctly
handles NETDEV_UNREGISTER event, but in case netvsc comes first on the
chain it sees the device with IFF_BONDING still attached and skips it. As
we still hold vf_netdev pointer to the device we crash on the next inject.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Acked-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: protect module refcount by checking net_device_ctx->vf_netdev
Vitaly Kuznetsov [Mon, 15 Aug 2016 15:48:42 +0000 (17:48 +0200)]
hv_netvsc: protect module refcount by checking net_device_ctx->vf_netdev

We're not guaranteed to see NETDEV_REGISTER/NETDEV_UNREGISTER notifications
only once per VF but we increase/decrease module refcount unconditionally.
Check vf_netdev to make sure we don't take/release it twice. We presume
that only one VF per netvsc device may exist.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Acked-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: reset vf_inject on VF removal
Vitaly Kuznetsov [Mon, 15 Aug 2016 15:48:41 +0000 (17:48 +0200)]
hv_netvsc: reset vf_inject on VF removal

We reset vf_inject on VF going down (netvsc_vf_down()) but we don't on
VF removal (netvsc_unregister_vf()) so vf_inject stays 'true' while
vf_netdev is already NULL and we're trying to inject packets into NULL
net device in netvsc_recv_callback() causing kernel to crash.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Acked-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: avoid deadlocks between rtnl lock and vf_use_cnt wait
Vitaly Kuznetsov [Mon, 15 Aug 2016 15:48:40 +0000 (17:48 +0200)]
hv_netvsc: avoid deadlocks between rtnl lock and vf_use_cnt wait

Here is a deadlock scenario:
- netvsc_vf_up() schedules netvsc_notify_peers() work and quits.
- netvsc_vf_down() runs before netvsc_notify_peers() gets executed. As it
  is being executed from netdev notifier chain we hold rtnl lock when we
  get here.
- we enter while (atomic_read(&net_device_ctx->vf_use_cnt) != 0) loop and
  wait till netvsc_notify_peers() drops vf_use_cnt.
- netvsc_notify_peers() starts on some other CPU but netdev_notify_peers()
  will hang on rtnl_lock().
- deadlock!

Instead of introducing additional synchronization I suggest we drop
gwrk.dwrk completely and call NETDEV_NOTIFY_PEERS directly. As we're
acting under rtnl lock this is legitimate.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Acked-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: don't lose VF information
Vitaly Kuznetsov [Mon, 15 Aug 2016 15:48:39 +0000 (17:48 +0200)]
hv_netvsc: don't lose VF information

struct netvsc_device is not suitable for storing VF information as this
structure is being destroyed on MTU change / set channel operation (see
rndis_filter_device_remove()). Move all VF related stuff to struct
net_device_context which is persistent.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Acked-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>