openwrt/staging/blogic.git
10 years agotipc: change signature of tunnelling reception function
Jon Paul Maloy [Thu, 13 Feb 2014 22:29:12 +0000 (17:29 -0500)]
tipc: change signature of tunnelling reception function

After the earlier commits in this series related to the function
tipc_link_tunnel_rcv(), we can now go further and simplify its
signature.

The function now consumes all DUPLICATE packets, and only returns such
ORIGINAL packets that are ready for immediate delivery, i.e., no
more link level protocol processing needs to be done by the caller.
As a consequence, the the caller, tipc_rcv(), does not access the link
pointer after call return, and it becomes unnecessary to pass a link
pointer reference in the call. Instead, we now only pass it the tunnel
link's owner node, which is sufficient to find the destination link for
the tunnelled packet.

Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: change reception of tunnelled failover packets
Jon Paul Maloy [Thu, 13 Feb 2014 22:29:11 +0000 (17:29 -0500)]
tipc: change reception of tunnelled failover packets

When a link is reset, and there is a redundant link available, all
sender sockets will steer their subsequent traffic through the
remaining link. In order to guarantee preserved packet order and
cardinality during the transition, we tunnel the failing link's send
queue through the remaining link before we allow any sockets to use it.

In this commit, we change the algorithm for receiving failover
("ORIGINAL_MSG") packets in tipc_link_tunnel_rcv(), at the same time
delegating it to a new subfuncton, tipc_link_failover_rcv(). Instead
of directly returning an extracted inner packet to the packet reception
loop in tipc_rcv(), we first check if it is a message fragment, in which
case we append it to the reset link's fragment chain. If the fragment
chain is complete, we return the whole chain instead of the individual
buffer, eliminating any need for the tipc_rcv() loop to do reassembly of
tunneled packets.

This change makes it possible to further simplify tipc_link_tunnel_rcv(),
as well as the calling tipc_rcv() loop. We will do that in later
commits. It also makes it possible to identify a single spot in the code
where we can tell that a failover procedure is finished, something that
is useful when we are deleting links after a failover. This will also
be done in a later commit.

Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: change reception of tunnelled duplicate packets
Jon Paul Maloy [Thu, 13 Feb 2014 22:29:10 +0000 (17:29 -0500)]
tipc: change reception of tunnelled duplicate packets

When a second link to a destination comes up, some sender sockets will
steer their subsequent traffic through the new link. In order to
guarantee preserved packet order and cardinality for those sockets, we
tunnel a duplicate of the old link's send queue through the new link
before we open it for regular traffic. The last arriving packet copy,
on whichever link, will be dropped at the receiving end based on the
original sequence number, to ensure that only one copy is delivered to
the end receiver.

In this commit, we change the algorithm for receiving DUPLICATE_MSG
packets, at the same time delegating it to a new subfunction,
tipc_link_dup_rcv(). Instead of returning an extracted inner packet to
the packet reception loop in tipc_rcv(), we just add it to the receiving
(new) link's deferred packet queue. The packet will then be processed by
that link when it receives its first non-tunneled packet, i.e., at
latest when the changeover procedure is finished.

Because tipc_link_tunnel_rcv()/tipc_link_dup_rcv() now is consuming all
packets of type DUPLICATE_MSG, the calling tipc_rcv() function can omit
testing for this. This in turn means that the current conditional jump
to the label 'protocol_check' becomes redundant, and we can remove that
label.

Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: remove 'links' list from tipc_bearer struct
Ying Xue [Thu, 13 Feb 2014 22:29:09 +0000 (17:29 -0500)]
tipc: remove 'links' list from tipc_bearer struct

In our ongoing effort to simplify the TIPC locking structure,
we see a need to remove the linked list for tipc_links
in the bearer. This can be explained as follows.

Currently, we have three different ways to access a link,
via three different lists/tables:

1: Via a node hash table:
   Used by the time-critical outgoing/incoming data paths.
   (e.g. link_send_sections_fast() and tipc_recv_msg() ):

grab net_lock(read)
   find node from node hash table
   grab node_lock
       select link
       grab bearer_lock
          send_msg()
       release bearer_lock
   release node lock
release net_lock

2: Via a global linked list for nodes:
   Used by configuration commands (link_cmd_set_value())

grab net_lock(read)
   find node and link from global node list (using link name)
   grab node_lock
       update link
   release node lock
release net_lock

(Same locking order as above. No problem.)

3: Via the bearer's linked link list:
   Used by notifications from interface (e.g. tipc_disable_bearer() )

grab net_lock(write)
   grab bearer_lock
      get link ptr from bearer's link list
      get node from link
      grab node_lock
         delete link
      release node lock
   release bearer_lock
release net_lock

(Different order from above, but works because we grab the
outer net_lock in write mode first, excluding all other access.)

The first major goal in our simplification effort is to get rid
of the "big" net_lock, replacing it with rcu-locks when accessing
the node list and node hash array. This will come in a later patch
series.

But to get there we first need to rewrite access methods ##2 and 3,
since removal of net_lock would introduce three major problems:

a) In access method #2, we access the link before taking the
   protecting node_lock. This will not work once net_lock is gone,
   so we will have to change the access order. We will deal with
   this in a later commit in this series, "tipc: add node lock
   protection to link found by link_find_link()".

b) When the outer protection from net_lock is gone, taking
   bearer_lock and node_lock in opposite order of method 1) and 2)
   will become an obvious deadlock hazard. This is fixed in the
   commit ("tipc: remove bearer_lock from tipc_bearer struct")
   later in this series.

c) Similar to what is described in problem a), access method #3
   starts with using a link pointer that is unprotected by node_lock,
   in order to via that pointer find the correct node struct and
   lock it. Before we remove net_lock, this access order must be
   altered. This is what we do with this commit.

We can avoid introducing problem problem c) by even here using the
global node list to find the node, before accessing its links. When
we loop though the node list we use the own bearer identity as search
criteria, thus easily finding the links that are associated to the
resetting/disabling bearer. It should be noted that although this
method is somewhat slower than the current list traversal, it is in
no way time critical. This is only about resetting or deleting links,
something that must be considered relatively infrequent events.

As a bonus, we can get rid of the mutual pointers between links and
bearers. After this commit, pointer dependency go in one direction
only: from the link to the bearer.

This commit pre-empts introduction of problem c) as described above.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: redefine 'started' flag in struct link to bitmap
Ying Xue [Thu, 13 Feb 2014 22:29:08 +0000 (17:29 -0500)]
tipc: redefine 'started' flag in struct link to bitmap

Currently, the 'started' field in struct tipc_link represents only a
binary state, 'started' or 'not started'. We need it to represent
more link execution states in the coming commits in this series.
Hence, we rename the field to 'flags', and define the current
started/non-started state to be represented by the LSB bit of
that field.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: move code for deleting links from bearer.c to link.c
Ying Xue [Thu, 13 Feb 2014 22:29:07 +0000 (17:29 -0500)]
tipc: move code for deleting links from bearer.c to link.c

We break out the code for deleting attached links in the
function bearer_disable(), and define a new function named
tipc_link_delete_list() to do this job.

This commit incurs no functional changes, but makes the code of
function bearer_disable() cleaner. It is also a preparation
for a more important change to the bearer code, in a subsequent
commit in this series.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: move code for resetting links from bearer.c to link.c
Ying Xue [Thu, 13 Feb 2014 22:29:06 +0000 (17:29 -0500)]
tipc: move code for resetting links from bearer.c to link.c

We break out the code for resetting attached links in the
function tipc_reset_bearer(), and define a new function named
tipc_link_reset_list() to do this job.

This commit incurs no functional changes, but makes the code
of function tipc_reset_bearer() cleaner. It is also a preparation
for a more important change to the bearer code, in a subsequent
commit in this series.

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: stricter behavior of message reassembly function
Jon Paul Maloy [Thu, 13 Feb 2014 22:29:05 +0000 (17:29 -0500)]
tipc: stricter behavior of message reassembly function

The function tipc_link_recv_fragment(struct sk_buff **buf) currently
leaves the value of the input buffer pointer undefined when it returns,
except when the return code indicates that the reassembly is complete.
This despite the fact that it always consumes the input buffer.

Here, we enforce a stricter behavior by this function, ensuring that
the returned buffer pointer is non-NULL if and only if the reassembly
is complete. This makes it possible to test for the buffer pointer as
criteria for successful reassembly.

We also rename the function to tipc_link_frag_rcv(), which is both
shorter and more in line with common naming practice in the network
subsystem.

Apart from the new name, these changes have no impact on current
users of the function, but makes it more practical for use in some
planned future commits.

Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agotipc: explicitly include core.h in addr.h
Andreas Bofjäll [Wed, 12 Feb 2014 08:34:18 +0000 (09:34 +0100)]
tipc: explicitly include core.h in addr.h

The inline functions in addr.h uses tipc_own_addr which is exported by
core.h, but addr.h never actually includes it. It works because it is
explicitly included where this is used, but it looks a bit strange.

Include core.h in addr.h explicitly to make the dependency clearer.

Signed-off-by: Andreas Bofjäll <andreas.bofjall@ericsson.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: remove the redundant judgements for bond_option_queue_id_set()
dingtianhong [Wed, 12 Feb 2014 06:58:50 +0000 (14:58 +0800)]
bonding: remove the redundant judgements for bond_option_queue_id_set()

The dev_valid_name() will check the buffer length for input name, no need to
check it twice.

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Reviewed-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: remove the redundant judgements for bond_set_mac_address()
dingtianhong [Wed, 12 Feb 2014 06:58:49 +0000 (14:58 +0800)]
bonding: remove the redundant judgements for bond_set_mac_address()

The dev_set_mac_address() will check the dev->netdev_ops->ndo_set_mac_address,
so no need to check it in bond_set_mac_address().

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Reviewed-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: allow setting mac address of loopback device
WANG Cong [Wed, 12 Feb 2014 01:21:26 +0000 (17:21 -0800)]
net: allow setting mac address of loopback device

We are trying to mirror the local traffic from lo to eth0,
allowing setting mac address of lo to eth0 would make
the ether addresses in these packets correct, so that
we don't have to modify the ether header again.

Since usually no one cares about its mac address (all-zero),
it is safe to allow those who care to set its mac address.

Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Eric Dumazet <edumazet@google.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'i40e'
David S. Miller [Thu, 13 Feb 2014 22:27:33 +0000 (17:27 -0500)]
Merge branch 'i40e'

Aaron Brown says:

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

This series contains updates to i40e and i40evf, primarily reset
handling / refactoring along with a fair amount of minor cleanup.

Jesse fixes some spelling, bumps the version and other trivial fixes.
Akeem sets a bit that is needed before shutdown in the case of
tx_timeout recovery failure.  Mitch refactors reset handling along
with a whole bunch of clean up.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: refactor reset handling
Mitch Williams [Thu, 13 Feb 2014 11:48:53 +0000 (03:48 -0800)]
i40evf: refactor reset handling

Respond better to a VF reset event. When a reset is signaled by the
PF, or detected by the watchdog task, prevent the watchdog from
processing admin queue requests, and schedule the reset task.

In the reset task, wait first for the reset to start, then for it to
complete, then reinit the driver.

If the reset never appears to complete after a long, long time (>10
seconds is possible depending on what's going on with the PF driver),
then set a flag to indicate that PF communications have failed.

If this flag is set, check for the reset to complete in the watchdog,
and  attempt to do a full reinitialization of the driver from scratch.

With these changes the VF driver correctly handles a PF reset event
while running on bare metal, or in a VM.

Also update copyrights.

Change-ID: I93513efd0b50523a8345e7f6a33a5e4f8a2a5996
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: change type of flags variable
Mitch Williams [Thu, 13 Feb 2014 11:48:52 +0000 (03:48 -0800)]
i40evf: change type of flags variable

As pointed out by Dan Carpenter (from Oracle), the flags variable is
declared as a 64-bit long but all of the flags are defined as u32,
which may lead to unintended consequences. Fix this by declaring flags
as u32 (since we don't need any more than about a dozen flags right
now), and remove the volatile qualifier, since it's unnecessary and
just makes checkpatch cry.

Change-ID: I137d3bb1842bf7e9456b5929ca54e3b0ed45dcab
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
CC: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: don't store unnecessary array of strings
Mitch Williams [Thu, 13 Feb 2014 11:48:51 +0000 (03:48 -0800)]
i40evf: don't store unnecessary array of strings

Since we store the traffic vector names in the queue vector struct, we
don't need to maintain an array of strings for these names in the
adapter structure. Replace this array with a single string and use it
when allocating the misc irq vector.

Also update copyrights.

Change-ID: I664f096c3c008210d6a04a487163e8aa934fee5b
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: fix bogus comment
Mitch Williams [Thu, 13 Feb 2014 11:48:50 +0000 (03:48 -0800)]
i40evf: fix bogus comment

Locate the structure in the correct header file.

Change-ID: Ic7853131728812093a44a75d6b70953311a48dab
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: clean up adapter struct
Mitch Williams [Thu, 13 Feb 2014 11:48:49 +0000 (03:48 -0800)]
i40evf: clean up adapter struct

Remove a bunch of unused structure members that are just wasting
space. Remove a completely unused info structure definition as well.

Also update copyrights.

Change-ID: I028ab92d9b7bd13a832cf3363bd1dc6610d8a535
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: don't handle VF reset on unload
Mitch Williams [Thu, 13 Feb 2014 11:48:48 +0000 (03:48 -0800)]
i40e: don't handle VF reset on unload

Set the DOWN flag before attempting to disable VFs when unloading the
driver. Also, don't attempt to reset the VFs when the driver is
unloading, because the switch configuration will fail. This fixes a
panic on unload when VFs are enabled.

Change-ID: I25a6567e89c9687145f510ff4f630932412c5c5d
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: enable extant VFs
Mitch Williams [Thu, 13 Feb 2014 11:48:47 +0000 (03:48 -0800)]
i40e: enable extant VFs

If VFs are present when the driver loads, then set up some resources
so they can function.

Change-ID: I485916a811609a9990ce663d06dc645f625b07ff
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: reset VFs after PF reset
Mitch Williams [Thu, 13 Feb 2014 11:48:46 +0000 (03:48 -0800)]
i40e: reset VFs after PF reset

Reset all of the VFs after a PF reset, so that they are in a known
state, and the VF driver can detect the reset and reinit itself.

Change-ID: I93c5b3a0f8b1371d0da078f92de948b9d3a6413f
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: set VF state to active when reset is complete
Mitch Williams [Thu, 13 Feb 2014 11:48:45 +0000 (03:48 -0800)]
i40e: set VF state to active when reset is complete

Without this, the VF can never communicate with the PF after a VF
reset.

Change-ID: I8d10f1d0d0638d50d39f0aff263422e05d83ad83
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: remove dead code
Mitch Williams [Thu, 13 Feb 2014 11:48:44 +0000 (03:48 -0800)]
i40e: remove dead code

This looks like a cut and paste error. The code makes no sense where
it is, and accomplishes nothing. Since we've removed the goto, we can
also get rid of the extraneous brackets.

Change-ID: I9315e3eafeee0a5713c94b0dc57b58b60a849124
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: Setting i40e_down bit for tx_timeout
Akeem G Abodunrin [Thu, 13 Feb 2014 11:48:43 +0000 (03:48 -0800)]
i40e: Setting i40e_down bit for tx_timeout

If tx_timeout recovery failed, then it becomes necessary to set
i40e_down bit before actually shutdown the connection.

Change-ID: Iaac81df0e302116571827aa0cff450697fbb7fa3
Signed-off-by: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: clean up memsets
Mitch Williams [Thu, 13 Feb 2014 11:48:42 +0000 (03:48 -0800)]
i40evf: clean up memsets

As politely pointed out by Dave Miller, calls to memset do not need a
void pointer cast. Additionally, it is preferred to use sizeof(*the
actual object) instead of sizeof(type).

Change-ID: Id6a02429b7040111531f3865ea03fbe619167cb3
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Kevin Scott <kevin.c.scott@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40evf: trivial fixes
Jesse Brandeburg [Thu, 13 Feb 2014 11:48:41 +0000 (03:48 -0800)]
i40evf: trivial fixes

This change moves one operator up to the previous line and deletes
the duplicate declaration of ETH_ALEN.

Also update copyrights.

Change-ID: I88de73093b584e0f3b29d481ccd83fc4b1a1afa5
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: bump driver version
Jesse Brandeburg [Thu, 13 Feb 2014 11:48:40 +0000 (03:48 -0800)]
i40e: bump driver version

Update the driver version to 0.3.31-k.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoi40e: spelling error
Jesse Brandeburg [Thu, 13 Feb 2014 11:48:39 +0000 (03:48 -0800)]
i40e: spelling error

Fix a spelling error, s/extention/extension/.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoipx: implement shutdown()
Sabrina Dubroca [Sat, 8 Feb 2014 23:23:38 +0000 (00:23 +0100)]
ipx: implement shutdown()

IPX doesn't implement shutdown, which poses a problem to some users:

https://bugzilla.kernel.org/show_bug.cgi?id=67841

This patch is heavily based on the shutdown implementation for unix
sockets.

Reported-by: Bruno Jesus <00cpxxx@gmail.com>
Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'net_sched'
David S. Miller [Thu, 13 Feb 2014 00:23:53 +0000 (19:23 -0500)]
Merge branch 'net_sched'

Cong Wang says:

====================
net_sched: act: more cleanup and improvement

v2 -> v3:
* fix a mis-splitted patch
* keep hinfo as a pointer in ops

v1 -> v2:
* Fix a bug noticed by Jamal
* Drop patches already merged into net-next
* Add patch 5/5

Patches are cleanup's for the structures of tc actions, except patch 4
which is an improvement.

See each patch for details.
====================

Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: clean up tca_action_flush()
WANG Cong [Wed, 12 Feb 2014 01:07:35 +0000 (17:07 -0800)]
net_sched: act: clean up tca_action_flush()

We could allocate tc_action on stack in tca_action_flush(),
since it is not large.

Also, we could use create_a() in tcf_action_get_1().

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: refuse to remove bound action outside
WANG Cong [Wed, 12 Feb 2014 01:07:34 +0000 (17:07 -0800)]
net_sched: act: refuse to remove bound action outside

When an action is bonnd to a filter, there is no point to
remove it outside. Currently we just silently decrease the refcnt,
we should reject this explicitly with EPERM.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: move tcf_hashinfo_init() into tcf_register_action()
WANG Cong [Wed, 12 Feb 2014 01:07:33 +0000 (17:07 -0800)]
net_sched: act: move tcf_hashinfo_init() into tcf_register_action()

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: refactor cleanup ops
WANG Cong [Wed, 12 Feb 2014 01:07:32 +0000 (17:07 -0800)]
net_sched: act: refactor cleanup ops

For bindcnt and refcnt etc., they are common for all actions,
not need to repeat such operations for their own, they can be unified
now. Actions just need to do its specific cleanup if needed.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet_sched: act: hide struct tcf_common from API
WANG Cong [Wed, 12 Feb 2014 01:07:31 +0000 (17:07 -0800)]
net_sched: act: hide struct tcf_common from API

Now we can totally hide it from modules. tcf_hash_*() API's
will operate on struct tc_action, modules don't need to care about
the details.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'bnx2x'
David S. Miller [Thu, 13 Feb 2014 00:16:03 +0000 (19:16 -0500)]
Merge branch 'bnx2x'

Yuval Mintz says:

====================
bnx2x: Enhancements & semantic changes series

This patch series contains several semantic (or mostly semantic) patches,
as well as adding support for packet aggregations on the receive path
of windows VMs and updating bnx2x to the new FW recently accepted upstream.

Please consider applying these patches to `net-next'.

(This is a repost as net-next was still closed when this was previously
sent)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: utilize FW 7.8.19
Dmitry Kravkov [Wed, 12 Feb 2014 16:19:57 +0000 (18:19 +0200)]
bnx2x: utilize FW 7.8.19

    This new firmware fixes following bugs:
     1. HW attention appears and traffic stops when iSCSI firmware tries to
        retransmit iSCSI login command when the iSCSI login is carrying data
        not aligned to 4-bytes.
     2. FCoE traffic fails to run when running in switch-independent multi-function
        mode and there's more than one interface supporting FCoE on a given port.
     3. While two ports are running FCoE with at least one of them has a function
        number (>1) on the same engine in a 4-port device a zeroed CQE is given,
        causing FCoE traffic to stop.

Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: (semantic) revise scheduling of sp_rtnl
Yuval Mintz [Wed, 12 Feb 2014 16:19:56 +0000 (18:19 +0200)]
bnx2x: (semantic) revise scheduling of sp_rtnl

This removes the various points where the driver use bit operations in order
to schedule the sp_rtnl_task from the code, adding a single utility function
that does it instead.

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Fix bnx2x_panic_dump for VFs
Yuval Mintz [Wed, 12 Feb 2014 16:19:55 +0000 (18:19 +0200)]
bnx2x: Fix bnx2x_panic_dump for VFs

bnx2x_panic_dump() prints all kind of driver information, including slowpath
information. Since VFs don't initialize slowpath information, a VF reaching
this flow will likely cause a panic in the system as it will access NULL
pointers.

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Revise IOV vlan/mac validation
Yuval Mintz [Wed, 12 Feb 2014 16:19:54 +0000 (18:19 +0200)]
bnx2x: Revise IOV vlan/mac validation

There are several places in IOV related flows where PF needs to determine
whether a VF slowpath elements have already been configured (i.e., this
affect its ability to configure/remove classifications for the VF).

This patch changes the conditions for the validation and performs a cleaner
validation (e.g., by replacing several validations with a single one).

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Add support in PF driver for RSC
Michal Kalderon [Wed, 12 Feb 2014 16:19:53 +0000 (18:19 +0200)]
bnx2x: Add support in PF driver for RSC

This provides PF-side support for VFs assigned to a VM running windows
2012 with the RSC feature enabled.

Signed-off-by: Michal Kalderon <michals@broadcom.com>
Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Semantic Validate vlan/mac changes
Yuval Mintz [Wed, 12 Feb 2014 16:19:52 +0000 (18:19 +0200)]
bnx2x: Semantic Validate vlan/mac changes

This is purely semantic - break the flow in which PF validates the VF
classification filtering requirement is valid into several sub-functions
for better readable code.

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Remove unnecessary internal mem config
Ariel Elior [Wed, 12 Feb 2014 16:19:51 +0000 (18:19 +0200)]
bnx2x: Remove unnecessary internal mem config

Latest FW performs this autonomously, makes this code surplus.

Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: Remove unused iov code
Yuval Mintz [Wed, 12 Feb 2014 16:19:50 +0000 (18:19 +0200)]
bnx2x: Remove unused iov code

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobnx2x: [Debug] change verbosity of some prints
Yuval Mintz [Wed, 12 Feb 2014 16:19:49 +0000 (18:19 +0200)]
bnx2x: [Debug] change verbosity of some prints

There are some debug prints (mostly iov/statistics related) which clobber
system logs whenever their verbosity level is set for an interface.
This patch puts harsher verbosity requirements for such debug prints to be
printed.

Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: Ariel Elior <ariele@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: expose phydev->has_fixups to sysfs
Florian Fainelli [Wed, 12 Feb 2014 01:27:42 +0000 (17:27 -0800)]
net: phy: expose phydev->has_fixups to sysfs

Expose the PHY device has_fixups boolean as a sysfs property to help
troubleshooting PHY configurations.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: add "has_fixups" boolean property
Florian Fainelli [Wed, 12 Feb 2014 01:27:41 +0000 (17:27 -0800)]
net: phy: add "has_fixups" boolean property

Add a boolean property which indicates if the PHY has had any fixup
routine ran on it. We are later going to use that boolean to expose it
as a sysfs property to help troubleshooting.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: expose PHY device interface mode
Florian Fainelli [Wed, 12 Feb 2014 01:27:40 +0000 (17:27 -0800)]
net: phy: expose PHY device interface mode

Expose the PHY device interface mode through sysfs since this is an
useful piece of information for knowing how the attached networking
device will have configured its transmit/receive path.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: re-design phy_modes to be self-contained
Florian Fainelli [Wed, 12 Feb 2014 01:27:39 +0000 (17:27 -0800)]
net: phy: re-design phy_modes to be self-contained

of_get_phy_mode() uses a local array to map phy_interface_t values from
include/linux/net/phy.h to a string which is read from the 'phy-mode' or
'phy-connection-type' property. In preparation for exposing the PHY
interface mode through sysfs, perform the following:

- mode phy_modes from drivers/of/of_net.c to include/linux/phy.h such
  that it is right below the phy_interface_t enum
- make it a static inline function returning the string such that we can
  use it by just including include/linux/net/phy.h
- add a PHY_INTERFACE_MODE_MAX enum value to guard the iteration in
  of_get_phy_mode()

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: fix phy_{clear,config}_interrupt comment typos
Florian Fainelli [Wed, 12 Feb 2014 01:27:38 +0000 (17:27 -0800)]
net: phy: fix phy_{clear,config}_interrupt comment typos

The comments above phy_{clear,config}_interrupt used the word "on"
instead of "or", when talking about the return values of the functions,
fix these two typos.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: allow driver to implement their own aneg_done
Florian Fainelli [Wed, 12 Feb 2014 01:27:37 +0000 (17:27 -0800)]
net: phy: allow driver to implement their own aneg_done

Some PHYs out there can be very quirky with respect to how they would
report the auto-negotiation is completed. Allow drivers to override the
generic aneg_done() implementation by providing their own.

Since not all drivers have been updated yet to use genphy_aneg_done() as
aneg_done() callback, we explicitely check that this callback is valid
before calling into it.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: add genphy_aneg_done()
Florian Fainelli [Wed, 12 Feb 2014 01:27:36 +0000 (17:27 -0800)]
net: phy: add genphy_aneg_done()

In preparation for allowing PHY drivers to potentially override their
auto-negotiation done callback, move the contents of phy_aneg_done() to
genphy_aneg_done() since that function really is the generic
implementation based on the BMSR_ANEGCOMPLETE status.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: display human readable PHY speed settings
Florian Fainelli [Wed, 12 Feb 2014 01:27:35 +0000 (17:27 -0800)]
net: phy: display human readable PHY speed settings

Use a convenience function: phy_speed_to_str() which will display human
readable speeds.

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: update phy_print_status to show pause settings
Florian Fainelli [Wed, 12 Feb 2014 01:27:34 +0000 (17:27 -0800)]
net: phy: update phy_print_status to show pause settings

Update phy_print_status() to also display the PHY device pause settings
(rx/tx or off).

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agonet: phy: use network device in phy_print_status
Florian Fainelli [Wed, 12 Feb 2014 01:27:33 +0000 (17:27 -0800)]
net: phy: use network device in phy_print_status

phy_print_status() currently uses dev_name(&phydev->dev) which will
usually result in printing something along those lines for Device Tree
aware drivers:

libphy: f0b60000.etherne:0a - Link is Down
libphy: f0ba0000.etherne:00 - Link is Up - 1000/Full

This is not terribly useful for network administrators or users since we
expect a network interface name to be able to correlate link events with
interfaces. Update phy_print_status() to use netdev_info() with
phydev->attached_dev which is the backing network device for our PHY
device. The leading dash is removed since netdev_info() prefixes the
messages with "<interface>: " already.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'sfc'
David S. Miller [Wed, 12 Feb 2014 22:53:43 +0000 (17:53 -0500)]
Merge branch 'sfc'

Shradha Shah says:

====================
Cleanup patches for the SFC driver

This patch set consists of some cleanup and housekeeping
patches for the sfc driver.
These patches help to reduce the differences between the in-
tree and out-of-tree driver.

Ben Hutchings (12):
  sfc: Cache skb->data in local variable in efx_ptp_rx()
  sfc: Rewrite adjustment of PPS event in a clearer way
  sfc: Replace TSOH_OFFSET with the equivalent NET_IP_ALIGN
  sfc: Rename 'use_options' variable in tso_start() to clearer
    'use_opt_desc'
  sfc: Remove unused definitions of EF10 user-mode DMA descriptors
  sfc: Correct comment about number of TX queues used on EF10
  sfc: Preserve rx_frm_trunc counters when resizing DMA rings
  sfc: Use canonical pointer type for MAC address in
    efx_set_mac_address()
  sfc: Update product naming
  sfc: Cosmetic changes to self-test from the out-of-tree driver
  sfc: Fail self-test with -EBUSY, not -EIO, if the device is busy
  sfc: Add/remove blank lines to taste

Laurence Evans (1):
  sfc: Removed adhoc scheme to rate limit PTP event queue overflow
    message
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Add/remove blank lines to taste
Ben Hutchings [Wed, 12 Feb 2014 19:00:37 +0000 (19:00 +0000)]
sfc: Add/remove blank lines to taste

Remove trailing blank lines in several files.
Use only one blank line between functions.
Add a blank line as a separator in a few places.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Fail self-test with -EBUSY, not -EIO, if the device is busy
Ben Hutchings [Wed, 12 Feb 2014 19:00:28 +0000 (19:00 +0000)]
sfc: Fail self-test with -EBUSY, not -EIO, if the device is busy

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Cosmetic changes to self-test from the out-of-tree driver
Ben Hutchings [Wed, 12 Feb 2014 19:00:16 +0000 (19:00 +0000)]
sfc: Cosmetic changes to self-test from the out-of-tree driver

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Update product naming
Ben Hutchings [Wed, 12 Feb 2014 19:00:07 +0000 (19:00 +0000)]
sfc: Update product naming

We don't use 'Solarstorm' or 'Solarflare Communications' in full
any more.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Use canonical pointer type for MAC address in efx_set_mac_address()
Ben Hutchings [Wed, 12 Feb 2014 18:59:54 +0000 (18:59 +0000)]
sfc: Use canonical pointer type for MAC address in efx_set_mac_address()

Functions such as is_valid_ether_addr() expect u8 *, so use that
instead of char *.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Rename 'use_options' variable in tso_start() to clearer 'use_opt_desc'
Ben Hutchings [Wed, 12 Feb 2014 18:59:41 +0000 (18:59 +0000)]
sfc: Rename 'use_options' variable in tso_start() to clearer 'use_opt_desc'

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Preserve rx_frm_trunc counters when resizing DMA rings
Ben Hutchings [Wed, 12 Feb 2014 18:59:28 +0000 (18:59 +0000)]
sfc: Preserve rx_frm_trunc counters when resizing DMA rings

We allocate efx_channel structures with kzalloc() so we don't need to
zero-initialise individual fields in efx_probe_channel().  Further,
this function will be called again during DMA ring resizing and we
should not reset any statistics then.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Correct comment about number of TX queues used on EF10
Ben Hutchings [Wed, 12 Feb 2014 18:59:19 +0000 (18:59 +0000)]
sfc: Correct comment about number of TX queues used on EF10

EF10 implements option descriptors to switch TX checksum offload
on and off between packets.  We could therefore use a single
hardware TX queue per kernel TX queue, although we don't yet.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Remove unused definitions of EF10 user-mode DMA descriptors
Ben Hutchings [Wed, 12 Feb 2014 18:59:07 +0000 (18:59 +0000)]
sfc: Remove unused definitions of EF10 user-mode DMA descriptors

These DMA descriptor types will only be used by the userland
networking stack.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Replace TSOH_OFFSET with the equivalent NET_IP_ALIGN
Ben Hutchings [Wed, 12 Feb 2014 18:58:57 +0000 (18:58 +0000)]
sfc: Replace TSOH_OFFSET with the equivalent NET_IP_ALIGN

If CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS is defined then NET_IP_ALIGN
will be defined as 0, so this macro is redundant.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Rewrite adjustment of PPS event in a clearer way
Ben Hutchings [Wed, 12 Feb 2014 18:58:46 +0000 (18:58 +0000)]
sfc: Rewrite adjustment of PPS event in a clearer way

There is substantial latency in generation and handling of PPS events
from the NIC, which we have to correct for before passing a host
timestamp to the PPS subsystem.  We compare clocks with the MC,
giving us two offsets to subtract from the timestamp generated by
pps_get_ts():

(a) Time from the last good sync (where we got host and NIC timestamps
    for nearly the same instant) to the time we called pps_get_ts()
(b) Time from NIC top of second to the last good sync

We currently calculate (a) + (b) in a quite confusing way.
Instead, calculate (a) completely, then add (b) to it.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Cache skb->data in local variable in efx_ptp_rx()
Ben Hutchings [Wed, 12 Feb 2014 18:58:34 +0000 (18:58 +0000)]
sfc: Cache skb->data in local variable in efx_ptp_rx()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agosfc: Removed adhoc scheme to rate limit PTP event queue overflow message
Laurence Evans [Wed, 12 Feb 2014 18:58:24 +0000 (18:58 +0000)]
sfc: Removed adhoc scheme to rate limit PTP event queue overflow message

Use conventional net_ratelimit() instead.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'ethtool_docs'
David S. Miller [Wed, 12 Feb 2014 22:33:39 +0000 (17:33 -0500)]
Merge branch 'ethtool_docs'

Ben Hutchings says:

====================
ethtool: Improve documentation

This series converts most comments in <linux/ethtool.h> into kernel-doc
format, and expands them to describe the semantics as well as I
understand them.  It also fixes some kernel-doc formatting errors.

I started on this at Solarflare, but they graciously allowed me to take
it away and finish it off.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Fix unwanted section breaks in kernel-doc
Ben Hutchings [Wed, 12 Feb 2014 22:15:09 +0000 (22:15 +0000)]
ethtool: Fix unwanted section breaks in kernel-doc

A colon almost unavoidably starts a new section.  The script should be
changed to provide a way to avoid this, but for now reword the
comments to avoid using colons.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Move kernel-doc comment next to struct ethtool_dump definition
Ben Hutchings [Wed, 12 Feb 2014 22:14:59 +0000 (22:14 +0000)]
ethtool: Move kernel-doc comment next to struct ethtool_dump definition

The kernel-doc script does not tolerate the macro definition in between.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Document the general convention for VLAs in kernel space
Ben Hutchings [Wed, 12 Feb 2014 22:14:53 +0000 (22:14 +0000)]
ethtool: Document the general convention for VLAs in kernel space

Various ethtool command structures are declared with zero-length array
at the end which are intended to be variable-length in userland
(relying on lack of compiler bounds checking).  However, in the kernel
the structure and array are always allocated and passed to driver
operations separately.  Make that explicit.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_perm_addr
Ben Hutchings [Wed, 12 Feb 2014 22:14:48 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_perm_addr

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_stats
Ben Hutchings [Wed, 12 Feb 2014 22:14:43 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_stats

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_test
Ben Hutchings [Wed, 12 Feb 2014 22:14:38 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_test

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of string set types
Ben Hutchings [Wed, 12 Feb 2014 22:14:33 +0000 (22:14 +0000)]
ethtool: Expand documentation of string set types

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Update documentation of struct ethtool_pauseparam
Ben Hutchings [Wed, 12 Feb 2014 22:14:28 +0000 (22:14 +0000)]
ethtool: Update documentation of struct ethtool_pauseparam

Convert the inline comments to kernel-doc format.

Explicitly specify that non-zero autoneg is an error if link
autonegotiation is disabled.

Specify that pause capabilities should be advertised dependent on link
autonegotiation, not the autoneg flag here.  There is no way to
opt-out of pause frame autonegotiation, and this improves behaviour
when the link partner is configured to follow pause frame
autonegotiation and our interface is not.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_ringparam
Ben Hutchings [Wed, 12 Feb 2014 22:14:23 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_ringparam

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_eeprom
Ben Hutchings [Wed, 12 Feb 2014 22:14:17 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_eeprom

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_regs
Ben Hutchings [Wed, 12 Feb 2014 22:14:13 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_regs

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_wol
Ben Hutchings [Wed, 12 Feb 2014 22:14:07 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_wol

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_drvinfo
Ben Hutchings [Wed, 12 Feb 2014 22:14:03 +0000 (22:14 +0000)]
ethtool: Expand documentation of struct ethtool_drvinfo

Replace the inline comments (and some others below) with a full
explanation of the semantics, in kernel-doc format.  Specify which
strings may be empty.  Document the relationship with other commands.

Replace the 'deprecation' of some fields with a proper explanation of
the conversion to generalised string sets, as userland programs may
not be able to assume that ETHTOOL_GSSET_INFO is available.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoethtool: Expand documentation of struct ethtool_cmd
Ben Hutchings [Wed, 12 Feb 2014 22:13:57 +0000 (22:13 +0000)]
ethtool: Expand documentation of struct ethtool_cmd

struct ethtool_cmd has very limited documentation; it contains
several obscure or obsolete fields and several with non-obvious
interpretation.

Replace the inline comments (and some others below) with a full
explanation of the semantics as well as I understand them, in
kernel-doc format.  Formally deprecate some fields that seem to be of
historical use only.

Extend the comment about 32/64-bit compatibility to cover all
ethtool structures.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge tag 'microblaze-3.14-rc3' of git://git.monstr.eu/linux-2.6-microblaze
Linus Torvalds [Tue, 11 Feb 2014 20:24:35 +0000 (12:24 -0800)]
Merge tag 'microblaze-3.14-rc3' of git://git.monstr.eu/linux-2.6-microblaze

Pull microblaze fixes from Michal Simek:
 - Fix two compilation issues - HZ, readq/writeq
 - Fix stack protection support

* tag 'microblaze-3.14-rc3' of git://git.monstr.eu/linux-2.6-microblaze:
  microblaze: Fix a typo when disabling stack protection
  microblaze: Define readq and writeq IO helper function
  microblaze: Fix missing HZ macro

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Tue, 11 Feb 2014 20:23:50 +0000 (12:23 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 bugfixes from Martin Schwidefsky:
 "A collection a bug fixes.  Most of them are minor but two of them are
  more severe.  The linkage stack bug can be used by user space to force
  an oops, with panic_on_oops this is a denial-of-service.  And the dump
  memory detection issue can cause incomplete memory dumps"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/cio: improve cio_commit_config
  s390: fix kernel crash due to linkage stack instructions
  s390/dump: Fix dump memory detection
  s390/appldata: restore missing init_virt_timer()
  s390/qdio: correct program-controlled interruption checking
  s390/qdio: for_each macro correctness

10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 11 Feb 2014 20:05:55 +0000 (12:05 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking updates from David Miller:

 1) Fix flexcan build on big endian, from Arnd Bergmann

 2) Correctly attach cpsw to GPIO bitbang MDIO drive, from Stefan Roese

 3) udp_add_offload has to use GFP_ATOMIC since it can be invoked from
    non-sleepable contexts.  From Or Gerlitz

 4) vxlan_gro_receive() does not iterate over all possible flows
    properly, fix also from Or Gerlitz

 5) CAN core doesn't use a proper SKB destructor when it hooks up
    sockets to SKBs.  Fix from Oliver Hartkopp

 6) ip_tunnel_xmit() can use an uninitialized route pointer, fix from
    Eric Dumazet

 7) Fix address family assignment in IPVS, from Michal Kubecek

 8) Fix ath9k build on ARM, from Sujith Manoharan

 9) Make sure fail_over_mac only applies for the correct bonding modes,
    from Ding Tianhong

10) The udp offload code doesn't use RCU correctly, from Shlomo Pongratz

11) Handle gigabit features properly in generic PHY code, from Florian
    Fainelli

12) Don't blindly invoke link operations in
    rtnl_link_get_slave_info_data_size, they are optional.  Fix from
    Fernando Luis Vazquez Cao

13) Add USB IDs for Netgear Aircard 340U, from Bjørn Mork

14) Handle netlink packet padding properly in openvswitch, from Thomas
    Graf

15) Fix oops when deleting chains in nf_tables, from Patrick McHardy

16) Fix RX stalls in xen-netback driver, from Zoltan Kiss

17) Fix deadlock in mac80211 stack, from Emmanuel Grumbach

18) inet_nlmsg_size() forgets to consider ifa_cacheinfo, fix from Geert
    Uytterhoeven

19) tg3_change_mtu() can deadlock, fix from Nithin Sujir

20) Fix regression in setting SCTP local source addresses on accepted
    sockets, caused by some generic ipv6 socket changes.  Fix from
    Matija Glavinic Pecotic

21) IPPROTO_* must be pure defines, otherwise module aliases don't get
    constructed properly.  Fix from Jan Moskyto

22) IPV6 netconsole setup doesn't work properly unless an explicit
    source address is specified, fix from Sabrina Dubroca

23) Use __GFP_NORETRY for high order skb page allocations in
    sock_alloc_send_pskb and skb_page_frag_refill.  From Eric Dumazet

24) Fix a regression added in netconsole over bridging, from Cong Wang

25) TCP uses an artificial offset of 1ms for SRTT, but this doesn't jive
    well with TCP pacing which needs the SRTT to be accurate.  Fix from
    Eric Dumazet

26) Several cases of missing header file includes from Rashika Kheria

27) Add ZTE MF667 device ID to qmi_wwan driver, from Raymond Wanyoike

28) TCP Small Queues doesn't handle nonagle properly in some corner
    cases, fix from Eric Dumazet

29) Remove extraneous read_unlock in bond_enslave, whoops.  From Ding
    Tianhong

30) Fix 9p trans_virtio handling of vmalloc buffers, from Richard Yao

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (136 commits)
  6lowpan: fix lockdep splats
  alx: add missing stats_lock spinlock init
  9p/trans_virtio.c: Fix broken zero-copy on vmalloc() buffers
  bonding: remove unwanted bond lock for enslave processing
  USB2NET : SR9800 : One chip USB2.0 USB2NET SR9800 Device Driver Support
  tcp: tsq: fix nonagle handling
  bridge: Prevent possible race condition in br_fdb_change_mac_address
  bridge: Properly check if local fdb entry can be deleted when deleting vlan
  bridge: Properly check if local fdb entry can be deleted in br_fdb_delete_by_port
  bridge: Properly check if local fdb entry can be deleted in br_fdb_change_mac_address
  bridge: Fix the way to check if a local fdb entry can be deleted
  bridge: Change local fdb entries whenever mac address of bridge device changes
  bridge: Fix the way to find old local fdb entries in br_fdb_change_mac_address
  bridge: Fix the way to insert new local fdb entries in br_fdb_changeaddr
  bridge: Fix the way to find old local fdb entries in br_fdb_changeaddr
  tcp: correct code comment stating 3 min timeout for FIN_WAIT2, we only do 1 min
  net: vxge: Remove unused device pointer
  net: qmi_wwan: add ZTE MF667
  3c59x: Remove unused pointer in vortex_eisa_cleanup()
  net: fix 'ip rule' iif/oif device rename
  ...

10 years ago6lowpan: fix lockdep splats
Eric Dumazet [Mon, 10 Feb 2014 19:42:35 +0000 (11:42 -0800)]
6lowpan: fix lockdep splats

When a device ndo_start_xmit() calls again dev_queue_xmit(),
lockdep can complain because dev_queue_xmit() is re-entered and the
spinlocks protecting tx queues share a common lockdep class.

Same issue was fixed for bonding/l2tp/ppp in commits

0daa2303028a6 ("[PATCH] bonding: lockdep annotation")
49ee49202b4ac ("bonding: set qdisc_tx_busylock to avoid LOCKDEP splat")
23d3b8bfb8eb2 ("net: qdisc busylock needs lockdep annotations ")
303c07db487be ("ppp: set qdisc_tx_busylock to avoid LOCKDEP splat ")

Reported-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoalx: add missing stats_lock spinlock init
John Greene [Mon, 10 Feb 2014 19:34:04 +0000 (14:34 -0500)]
alx: add missing stats_lock spinlock init

Trivial fix for init time stack trace occuring in
alx_get_stats64 upon start up. Should have been part of
commit adding the spinlock:
f1b6b106 alx: add alx_get_stats64 operation

Signed-off-by: John Greene <jogreene@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years ago9p/trans_virtio.c: Fix broken zero-copy on vmalloc() buffers
Richard Yao [Sun, 9 Feb 2014 00:32:01 +0000 (19:32 -0500)]
9p/trans_virtio.c: Fix broken zero-copy on vmalloc() buffers

The 9p-virtio transport does zero copy on things larger than 1024 bytes
in size. It accomplishes this by returning the physical addresses of
pages to the virtio-pci device. At present, the translation is usually a
bit shift.

That approach produces an invalid page address when we read/write to
vmalloc buffers, such as those used for Linux kernel modules. Any
attempt to load a Linux kernel module from 9p-virtio produces the
following stack.

[<ffffffff814878ce>] p9_virtio_zc_request+0x45e/0x510
[<ffffffff814814ed>] p9_client_zc_rpc.constprop.16+0xfd/0x4f0
[<ffffffff814839dd>] p9_client_read+0x15d/0x240
[<ffffffff811c8440>] v9fs_fid_readn+0x50/0xa0
[<ffffffff811c84a0>] v9fs_file_readn+0x10/0x20
[<ffffffff811c84e7>] v9fs_file_read+0x37/0x70
[<ffffffff8114e3fb>] vfs_read+0x9b/0x160
[<ffffffff81153571>] kernel_read+0x41/0x60
[<ffffffff810c83ab>] copy_module_from_fd.isra.34+0xfb/0x180

Subsequently, QEMU will die printing:

qemu-system-x86_64: virtio: trying to map MMIO memory

This patch enables 9p-virtio to correctly handle this case. This not
only enables us to load Linux kernel modules off virtfs, but also
enables ZFS file-based vdevs on virtfs to be used without killing QEMU.

Special thanks to both Avi Kivity and Alexander Graf for their
interpretation of QEMU backtraces. Without their guidence, tracking down
this bug would have taken much longer. Also, special thanks to Linus
Torvalds for his insightful explanation of why this should use
is_vmalloc_addr() instead of is_vmalloc_or_module_addr():

https://lkml.org/lkml/2014/2/8/272

Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agobonding: remove unwanted bond lock for enslave processing
dingtianhong [Mon, 10 Feb 2014 08:33:59 +0000 (16:33 +0800)]
bonding: remove unwanted bond lock for enslave processing

The bond enslave processing don't hold bond->lock anymore,
so release an unlocked rw lock will cause warning message,
remove the unwanted read_unlock(&bond->lock).

Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Acked-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoUSB2NET : SR9800 : One chip USB2.0 USB2NET SR9800 Device Driver Support
Liu Junliang [Mon, 10 Feb 2014 06:31:42 +0000 (14:31 +0800)]
USB2NET : SR9800 : One chip USB2.0 USB2NET SR9800 Device Driver Support

Signed-off-by: Liu Junliang <liujunliang_ljl@163.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
10 years agoMerge branch 'akpm' (patches from Andrew Morton)
Linus Torvalds [Tue, 11 Feb 2014 00:03:16 +0000 (16:03 -0800)]
Merge branch 'akpm' (patches from Andrew Morton)

Merge misc fixes from Andrew Morton:
 "A bunch of fixes"

* emailed patches fron Andrew Morton <akpm@linux-foundation.org>:
  ocfs2: check existence of old dentry in ocfs2_link()
  ocfs2: update inode size after zeroing the hole
  ocfs2: fix issue that ocfs2_setattr() does not deal with new_i_size==i_size
  mm/memory-failure.c: move refcount only in !MF_COUNT_INCREASED
  smp.h: fix x86+cpu.c sparse warnings about arch nonboot CPU calls
  mm: fix page leak at nfs_symlink()
  slub: do not assert not having lock in removing freed partial
  gitignore: add all.config
  ocfs2: fix ocfs2_sync_file() if filesystem is readonly
  drivers/edac/edac_mc_sysfs.c: poll timeout cannot be zero
  fs/file.c:fdtable: avoid triggering OOMs from alloc_fdmem
  xen: properly account for _PAGE_NUMA during xen pte translations
  mm/slub.c: list_lock may not be held in some circumstances
  drivers/md/bcache/extents.c: use %zi to format size_t
  vmcore: prevent PT_NOTE p_memsz overflow during header update
  drivers/message/i2o/i2o_config.c: fix deadlock in compat_ioctl(I2OGETIOPS)
  Documentation/: update 00-INDEX files
  checkpatch: fix detection of git repository
  get_maintainer: fix detection of git repository
  drivers/misc/sgi-gru/grukdump.c: unlocking should be conditional in gru_dump_context()

10 years agoocfs2: check existence of old dentry in ocfs2_link()
Xue jiufei [Mon, 10 Feb 2014 22:25:54 +0000 (14:25 -0800)]
ocfs2: check existence of old dentry in ocfs2_link()

System call linkat first calls user_path_at(), check the existence of
old dentry, and then calls vfs_link()->ocfs2_link() to do the actual
work.  There may exist a race when Node A create a hard link for file
while node B rm it.

         Node A                          Node B
user_path_at()
  ->ocfs2_lookup(),
find old dentry exist
                                rm file, add inode say inodeA
                                to orphan_dir

call ocfs2_link(),create a
hard link for inodeA.

                                rm the link, add inodeA to orphan_dir
                                again

When orphan_scan work start, it calls ocfs2_queue_orphans() to do the
main work.  It first tranverses entrys in orphan_dir, linking all inodes
in this orphan_dir to a list look like this:

inodeA->inodeB->...->inodeA

When tranvering this list, it will fall into loop, calling iput() again
and again.  And finally trigger BUG_ON(inode->i_state & I_CLEAR).

Signed-off-by: joyce <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoocfs2: update inode size after zeroing the hole
Junxiao Bi [Mon, 10 Feb 2014 22:25:53 +0000 (14:25 -0800)]
ocfs2: update inode size after zeroing the hole

fs-writeback will release the dirty pages without page lock whose offset
are over inode size, the release happens at
block_write_full_page_endio().  If not update, dirty pages in file holes
may be released before flushed to the disk, then file holes will contain
some non-zero data, this will cause sparse file md5sum error.

To reproduce the bug, find a big sparse file with many holes, like vm
image file, its actual size should be bigger than available mem size to
make writeback work more frequently, tar it with -S option, then keep
untar it and check its md5sum again and again until you get a wrong
md5sum.

Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Younger Liu <younger.liu@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoocfs2: fix issue that ocfs2_setattr() does not deal with new_i_size==i_size
Younger Liu [Mon, 10 Feb 2014 22:25:51 +0000 (14:25 -0800)]
ocfs2: fix issue that ocfs2_setattr() does not deal with new_i_size==i_size

The issue scenario is as following:

- Create a small file and fallocate a large disk space for a file with
  FALLOC_FL_KEEP_SIZE option.

- ftruncate the file back to the original size again.  but the disk free
  space is not changed back.  This is a real bug that be fixed in this
  patch.

In order to solve the issue above, we modified ocfs2_setattr(), if
attr->ia_size != i_size_read(inode), It calls ocfs2_truncate_file(), and
truncate disk space to attr->ia_size.

Signed-off-by: Younger Liu <younger.liu@huawei.com>
Reviewed-by: Jie Liu <jeff.liu@oracle.com>
Tested-by: Jie Liu <jeff.liu@oracle.com>
Cc: Joel Becker <jlbec@evilplan.org>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Sunil Mushran <sunil.mushran@gmail.com>
Reviewed-by: Jensen <shencanquan@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm/memory-failure.c: move refcount only in !MF_COUNT_INCREASED
Naoya Horiguchi [Mon, 10 Feb 2014 22:25:50 +0000 (14:25 -0800)]
mm/memory-failure.c: move refcount only in !MF_COUNT_INCREASED

mce-test detected a test failure when injecting error to a thp tail
page.  This is because we take page refcount of the tail page in
madvise_hwpoison() while the fix in commit a3e0f9e47d5e
("mm/memory-failure.c: transfer page count from head page to tail page
after split thp") assumes that we always take refcount on the head page.

When a real memory error happens we take refcount on the head page where
memory_failure() is called without MF_COUNT_INCREASED set, so it seems
to me that testing memory error on thp tail page using madvise makes
little sense.

This patch cancels moving refcount in !MF_COUNT_INCREASED for valid
testing.

[akpm@linux-foundation.org: s/&&/&/]
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Chen Gong <gong.chen@linux.intel.com>
Cc: <stable@vger.kernel.org> [3.9+: a3e0f9e47d5e]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agosmp.h: fix x86+cpu.c sparse warnings about arch nonboot CPU calls
Paul Gortmaker [Mon, 10 Feb 2014 22:25:49 +0000 (14:25 -0800)]
smp.h: fix x86+cpu.c sparse warnings about arch nonboot CPU calls

Use what we already do for arch_disable_smp_support() to fix these:

  arch/x86/kernel/smpboot.c:1155:6: warning: symbol 'arch_enable_nonboot_cpus_begin' was not declared. Should it be static?
  arch/x86/kernel/smpboot.c:1160:6: warning: symbol 'arch_enable_nonboot_cpus_end' was not declared. Should it be static?
  kernel/cpu.c:512:13: warning: symbol 'arch_enable_nonboot_cpus_begin' was not declared. Should it be static?
  kernel/cpu.c:516:13: warning: symbol 'arch_enable_nonboot_cpus_end' was not declared. Should it be static?

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: fix page leak at nfs_symlink()
Rafael Aquini [Mon, 10 Feb 2014 22:25:48 +0000 (14:25 -0800)]
mm: fix page leak at nfs_symlink()

Changes in commit a0b8cab3b9b2 ("mm: remove lru parameter from
__pagevec_lru_add and remove parts of pagevec API") have introduced a
call to add_to_page_cache_lru() which causes a leak in nfs_symlink() as
now the page gets an extra refcount that is not dropped.

Jan Stancek observed and reported the leak effect while running test8
from Connectathon Testsuite.  After several iterations over the test
case, which creates several symlinks on a NFS mountpoint, the test
system was quickly getting into an out-of-memory scenario.

This patch fixes the page leak by dropping that extra refcount
add_to_page_cache_lru() is grabbing.

Signed-off-by: Jan Stancek <jstancek@redhat.com>
Signed-off-by: Rafael Aquini <aquini@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Trond Myklebust <trond.myklebust@primarydata.com>
Cc: <stable@vger.kernel.org> [3.11.x+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoslub: do not assert not having lock in removing freed partial
Steven Rostedt [Mon, 10 Feb 2014 22:25:46 +0000 (14:25 -0800)]
slub: do not assert not having lock in removing freed partial

Vladimir reported the following issue:

Commit c65c1877bd68 ("slub: use lockdep_assert_held") requires
remove_partial() to be called with n->list_lock held, but free_partial()
called from kmem_cache_close() on cache destruction does not follow this
rule, leading to a warning:

  WARNING: CPU: 0 PID: 2787 at mm/slub.c:1536 __kmem_cache_shutdown+0x1b2/0x1f0()
  Modules linked in:
  CPU: 0 PID: 2787 Comm: modprobe Tainted: G        W    3.14.0-rc1-mm1+ #1
  Hardware name:
   0000000000000600 ffff88003ae1dde8 ffffffff816d9583 0000000000000600
   0000000000000000 ffff88003ae1de28 ffffffff8107c107 0000000000000000
   ffff880037ab2b00 ffff88007c240d30 ffffea0001ee5280 ffffea0001ee52a0
  Call Trace:
    __kmem_cache_shutdown+0x1b2/0x1f0
    kmem_cache_destroy+0x43/0xf0
    xfs_destroy_zones+0x103/0x110 [xfs]
    exit_xfs_fs+0x38/0x4e4 [xfs]
    SyS_delete_module+0x19a/0x1f0
    system_call_fastpath+0x16/0x1b

His solution was to add a spinlock in order to quiet lockdep.  Although
there would be no contention to adding the lock, that lock also requires
disabling of interrupts which will have a larger impact on the system.

Instead of adding a spinlock to a location where it is not needed for
lockdep, make a __remove_partial() function that does not test if the
list_lock is held, as no one should have it due to it being freed.

Also added a __add_partial() function that does not do the lock
validation either, as it is not needed for the creation of the cache.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Reported-by: Vladimir Davydov <vdavydov@parallels.com>
Suggested-by: David Rientjes <rientjes@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Vladimir Davydov <vdavydov@parallels.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>