Mahesh Bandewar says:
====================
Multicast processing in IPvlan
Dan Willems pointed out that autoconf in IPvlan is broken because of the
way broadcast bit gets set. Since broadcast processing is a real performance
drain, the broadcast bit in multicast filter was only set when the interface
was configured with IPv4 address. In autoconf scenario, when there are
no addresses configured; this logic did not work and it wouldn't allow
DHCPv4 to work. The only way was to add protocol specific hacks to avoid
processing unnecessary broadcast burdon.
This jugglery could be avoided if these multicast / broadcast packets are taken
out of fast-path and are processed in a work-queue. This will enable us to add
broadcast bit in all multicast filters without any impact on performance of
the virtual device. This patch series just does that.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Earlier tricks of setting broadcast bit only when IPv4 address is added
onto interface are not good enough especially when autoconf comes in play.
Setting them on always is performance drag but now that multicast /
broadcast is not processed in fast-path; enabling broadcast will let
autoconf work correctly without affecting performance characteristics of
the device.
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Processing multicast / broadcast in fast path is performance draining
and having more links means more cloning and bringing performance
down further.
Broadcast; in particular, need to be given to all the virtual links.
Earlier tricks of enabling broadcast bit for IPv4 only interfaces are not
really working since it fails autoconf. Which means enabling broadcast
for all the links if protocol specific hacks do not have to be added into
the driver.
This patch defers all (incoming as well as outgoing) multicast traffic to
a work-queue leaving only the unicast traffic in the fast-path. Now if we
need to apply any additional tricks to further reduce the impact of this
(multicast / broadcast) type of traffic, it can be implemented while
processing this work without affecting the fast-path.
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Alexander Duyck says:
====================
Add eth_proto_is_802_3 to provide improved means of checking Ethertype
This patch series implements and makes use of eth_proto_is_802_3(). The
idea behind the function is to provide an optimized means of testing to
determine if a given Ethertype value is a length or 802.3 protocol number.
The standard path for this was to use ntohs(proto) and then perform a
comparison. This adds a slight cost as it usually requires either a 16b
rotate or byte swap which can cost 1 cycle or more depending on the
processor.
I had previously addressed this for eth_type_trans, however in doing so I had
overlooked checking with sparse and had introduced a couple sparse warnings.
The first patch in this series fixes those sparse warnings as well as does
some additional optimization for big endian systems. In addition it pushes
the code out into a separate function which can then be used in the other
patches to reduce the instruction count/processing time in those functions
as well.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace "ntohs(proto) >= ETH_P_802_3_MIN" w/ eth_proto_is_802_3(proto).
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace "ntohs(proto) >= ETH_P_802_3_MIN" w/ eth_proto_is_802_3(proto).
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace "ntohs(proto) >= ETH_P_802_3_MIN" w/ eth_proto_is_802_3(proto).
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace "ntohs(proto) >= ETH_P_802_3_MIN" w/ eth_proto_is_802_3(proto).
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This change does two things. First it fixes a sparse error for the fact
that the __be16 degrades to an integer. Since that is actually what I am
kind of doing I am simply working around that by forcing both sides of the
comparison to u16.
Also I realized on some compilers I was generating another instruction for
big endian systems such as PowerPC since it was masking the value before
doing the comparison. So to resolve that I have simply pulled the mask out
and wrapped it in an #ifndef __BIG_ENDIAN.
Lastly I pulled this all out into its own function. I notices there are
similar checks in a number of other places so this function can be reused
there to help reduce overhead in these paths as well.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
BR_GROUPFWD_RESTRICTED bitmask restricts users from setting values to
/sys/class/net/brX/bridge/group_fwd_mask that allow forwarding of
some IEEE 802.1D Table 7-10 Reserved addresses:
(MAC Control) 802.3 01-80-C2-00-00-01
(Link Aggregation) 802.3 01-80-C2-00-00-02
802.1AB LLDP 01-80-C2-00-00-0E
Change BR_GROUPFWD_RESTRICTED to allow to forward LLDP frames and document
group_fwd_mask.
e.g.
echo 16384 > /sys/class/net/brX/bridge/group_fwd_mask
allows to forward LLDP frames.
This may be needed for bridge setups used for network troubleshooting or
any other scenario where forwarding of LLDP frames is desired (e.g. bridge
connecting a virtual machine to real switch transmitting LLDP frames that
virtual machine needs to receive).
Tested on a simple bridge setup with two interfaces and host transmitting
LLDP frames on one side of this bridge (used lldpd). Setting group_fwd_mask
as described above lets LLDP frames traverse bridge.
Signed-off-by: Bernhard Thaler <bernhard.thaler@wvnet.at>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch allows a server application to get the TCP SYN headers for
its passive connections. This is useful if the server is doing
fingerprinting of clients based on SYN packet contents.
Two socket options are added: TCP_SAVE_SYN and TCP_SAVED_SYN.
The first is used on a socket to enable saving the SYN headers
for child connections. This can be set before or after the listen()
call.
The latter is used to retrieve the SYN headers for passive connections,
if the parent listener has enabled TCP_SAVE_SYN.
TCP_SAVED_SYN is read once, it frees the saved SYN headers.
The data returned in TCP_SAVED_SYN are network (IPv4/IPv6) and TCP
headers.
Original patch was written by Tom Herbert, I changed it to not hold
a full skb (and associated dst and conntracking reference).
We have used such patch for about 3 years at Google.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
> net/core/skbuff.c:4108:13: sparse: incorrect type in assignment (different base types)
> net/ipv6/mcast_snoop.c:63 ipv6_mc_check_exthdrs() warn: unsigned 'offset' is never less than zero.
Introduced by 9afd85c9e4
("net: Export IGMP/MLD message validation code")
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
Intel Wired LAN Driver Updates 2015-05-04
This series contains updates to igb, e100, e1000e and ixgbe.
Todd cleans up igb_enable_mas() since it should only be called for the
82575 silicon and has no clear return, so modify the function to void.
Jean Sacren found upon inspection that 'err' did not need to be
initialized, since it is immediately overwritten.
Alex Duyck provides two patches for e1000e, the first cleans up the
handling VLAN_HLEN as a part of max frame size. Fixes the issue:
c751a3d58c ("e1000e: Correctly include VLAN_HLEN when changing
interface MTU"). The second fixes an issue where the driver was not
allowing jumbo frames to be enabled when CRC stripping was disabled,
however it was allowing CRC stripping to be disabled while jumbo frames
were enabled.
Jeff (me) fixes a warning found on PPC where the use of do_div() needed
to use u64 arg and not s64.
Mark provides three ixgbe patches, first to fix the Intel On-chip System
Fabric (IOSF) Sideband message interfaces, to serialize access using both
PHY bits in the SWFW_SEMAPHORE register. Then fixes how semaphore bits
were released, since they should be released in reverse of the order that
they were taken. Lastly updates ixgbe to use a signed type to hold
error codes, since error codes are negative, so consistently use signed
types when handling them.
v2: dropped the previous #6-#8 patches by Hiroshi Shimanoto based on
feedback from Or Gerlitz (and David Miller) that it appears there
needs to be further discussion on how this gets implemented.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Ying Xue says:
====================
tipc: cleanup topology server
Not only function names declared in subscr.c are very confused, but
also topology server's locking policy is not designed very well, for
instance, usually leading to panic in some special corner cases.
In this series, we attempt to eliminate the confusion of function names
and simplify topology server's locking policy to solve above mentioned
issues. More importantly, the change will make relevant code easily
understandable and maintainable.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Once tipc_conn_new() returns NULL, the connection should be shut
down immediately, otherwise, oops may happen due to the NULL pointer.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently subscriber's lock protects not only subscriber's subscription
list but also all subscriptions linked into the list. However, as all
members of subscription are never changed after they are initialized,
it's unnecessary for subscription to be protected under subscriber's
lock. If the lock is used to only protect subscriber's subscription
list, the adjustment not only makes the locking policy simpler, but
also helps to avoid a deadlock which may happen once creating a
subscription is failed.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
At present subscriber's lock is used to protect the subscription list
of subscriber as well as subscriptions linked into the list. While one
or all subscriptions are deleted through iterating the list, the
subscriber's lock must be held. Meanwhile, as deletion of subscription
may happen in subscription timer's handler, the lock must be grabbed
in the function as well. When subscription's timer is terminated with
del_timer_sync() during above iteration, subscriber's lock has to be
temporarily released, otherwise, deadlock may occur. However, the
temporary release may cause the double free of a subscription as the
subscription is not disconnected from the subscription list.
Now if a reference counter is introduced to subscriber, subscription's
timer can be asynchronously stopped with del_timer(). As a result, the
issue is not only able to be fixed, but also relevant code is pretty
readable and understandable.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introducing a new function makes the purpose of tipc_subscrb_connect_cb
callback routine more clear.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a topology server accepts a connection request from its client,
it allocates a connection instance and a tipc_subscriber structure
object. The former is used to communicate with client, and the latter
is often treated as a subscriber which manages all subscription events
requested from a same client. When a topology server receives a request
of subscribing name services from a client through the connection, it
creates a tipc_subscription structure instance which is seen as a
subscription recording what name services are subscribed. In order to
manage all subscriptions from a same client, topology server links
them into the subscrp_list of the subscriber. So subscriber and
subscription completely represents different meanings respectively,
but function names associated with them make us so confused that we
are unable to easily tell which function is against subscriber and
which is to subscription. So we want to eliminate the confusion by
renaming them.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Linus Lüssing says:
====================
Exporting IGMP/MLD checking from bridge code
The multicast optimizations in batman-adv are yet only usable and
enabled in non-bridged scenarios. To be able to support bridged setups
batman-adv needs to be able to detect IGMP/MLD queriers and reports on
mesh nodes without bridges, too. See the following link for details:
http://www.open-mesh.org/projects/batman-adv/wiki/Multicast-optimizations-listener-reports
To avoid duplicate code between the bridge and batman-adv, the IGMP/MLD
message validation code is moved from the bridge to the IPv4/IPv6 stack.
On the way, some refactoring to increase readability and to iron out
some subtle differences between the IGMP and MLD parsing code is done.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
With this patch, the IGMP and MLD message validation functions are moved
from the bridge code to IPv4/IPv6 multicast files. Some small
refactoring was done to enhance readibility and to iron out some
differences in behaviour between the IGMP and MLD parsing code (e.g. the
skb-cloning of MLD messages is now only done if necessary, just like the
IGMP part always did).
Finally, these IGMP and MLD message validation functions are exported so
that not only the bridge can use it but batman-adv later, too.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
Signed-off-by: David S. Miller <davem@davemloft.net>
Let's use these new, neat helpers.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Because error codes are negative, it only makes sense to
consistently use signed types when handling them. Also remove
some explicit comparisons with 0 on these variables.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The global semaphore bits should be released in the reverse of the
order that they were taken, so correct that.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
IOSF is the Intel On-chip System Fabric used in SOCs. IOSF SB is
the IOSF SideBand message interface. This patch serializes IOSF SB
access using both phy bits in the SWFW_SEMAPHORE register. It also
adds a helper function to wait for IOSF SB accesses to complete.
Use the new function to perform this wait before each access, as
specified in the datasheet, in addition to using it to wait for
IOSF SB read/write completion.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
We were using s64 for lat_ns (latency nano-second value) since in
our calculations a negative value could be a resultant. For negative
values, we then assign lat_ns to be zero, so the value passed to
do_div() was never negative, but do_div() expects the argument type
to be u64, so do a cast to resolve a compile warning seen on
PowerPC.
CC: Yanjiang Jin <yanjiang.jin@windriver.com>
CC: Yanir Lubetkin <yanirx.lubetkin@intel.com>
Reported-by: Yanjiang Jin <yanjiang.jin@windriver.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
The driver wasn't allowing jumbo frames to be
enabled when CRC stripping was disabled, however it was allowing CRC
stripping to be disabled while jumbo frames were enabled. This fixes that by
making it so that the NETIF_F_RXFCS flag cannot be set when jumbo frames are
enabled on 82579 and newer parts.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When the VLAN_HLEN was added to the calculation for the maximum frame size
there seems to have been a number of issues added to the driver.
The first issue is that in some cases the maximum frame size for a device
never really reached the actual maximum frame size as the VLAN header
length was not included the calculation for that value. As a result some
parts only supported a maximum frame size of either 1496 in the case of
parts that didn't support jumbo frames, and 8996 in the case of the parts
that do.
The second issue is the fact that there were several checks that weren't
updated so as a result setting an MTU of 1500 was treated as enabling jumbo
frames as the calculated value was 1522 instead of 1518. I have addressed
those by replacing ETH_FRAME_LEN with VLAN_ETH_FRAME_LEN where appropriate.
The final issue was the fact that lowering the MTU below 1500 would cause
the driver to allocate 2K buffers for the rings. This is an old issue that
was fixed several years ago in igb/ixgbe and I am addressing now by just
replacing == with a <= so that we always just round up to 1522 for anything
that isn't a jumbo frame.
Fixes: c751a3d58c ("e1000e: Correctly include VLAN_HLEN when changing interface MTU")
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
'err' will be overwritten so no need to initialize it to zero.
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
igb_enable_mas() should only be called for the 82575 and has no clear
return so changing it to void. Also simplify the odd conditional
expression.
Signed-off-by: Todd Fujinaka <todd.fujinaka@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Francois Romieu says:
====================
via-rhine rework
The series applies against davem-next as of
9dd3c79749 ("drivers: net: xgene: fix kbuild
warnings").
Patches #1..#4 avoid holes in the receive ring.
Patch #5 is a small leftover cleanup for #1..#4.
Patches #6 and #7 are fairly simple barrier stuff.
Patch #8 closes some SMP transmit races - not that anyone really
complained about these but it's a bit hard to handwave that they
can be safely ignored. Some testing, especially SMP testing of
course, would be welcome.
. Changes since #2:
- added dma_rmb barrier in vlan related patch 6.
- s/wmb/dma_wmb/ in (*new*) patch 7 of 8.
- added explicit SMP barriers in (*new*) patch 8 of 8.
. Changes since #1:
- turned wmb() into dma_wmb() as suggested by davem and Alexander Duyck
in patch 1 of 6.
- forgot to reset rx_head_desc in rhine_reset_rbufs in patch 4 of 6.
- removed rx_head_desc altogether in (*new*) patch 5 of 6
- remoed some vlan receive uglyness in (*new*) patch 6 of 6.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
7ab87ff4c7 ("via-rhine: move work from
irq handler to softirq and beyond") forgot to explicitely control the
lifespan of the tx_dirty and tx_cur pointers.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Follow the now usual transmit descriptor update path:
1. content change
2. dma_wmb
3. ownership change
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The NAPI receive path depends on desc->rx_status but it does not
enforce any explicit receive barrier.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver no longer produces holes in its receive ring so rx_head_desc
only duplicates cur_rx.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rationales:
- throttle work under memory pressure
- lower receive descriptor recycling latency for the network adapter
- lower the maintenance burden of uncommon paths
The patch is twofold:
- it fails early if the receive ring can't be completely initialized
at dev->open() time
- it drops packets on the floor in the napi receive handler so as to
keep the received ring full
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's used to initialize the receive ring but it will actually shine when
the receive poll code is reworked.
Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tom Herbert says:
====================
net: Eliminate calls to flow_dissector and introduce flow_keys_digest
In this patch set we add skb_get_hash_perturb which gets the skbuff
hash for a packet and perturbs it using a provided key and jhash1.
This function is used in serveral qdiscs and eliminates many calls
to flow_dissector and jhash3 to get a perturbed hash for a packet.
To handle the sch_choke issue (passes flow_keys in skbuff cb) we
add flow_keys_digest which is a digest of a flow constructed
from a flow_keys structure.
This is the second version of these patches I posted a while ago,
and is prerequisite work to increasing the size of the flow_keys
structure and hashing over it (full IPv6 address, flow label, VLAN ID,
etc.).
Version 2:
- Add keyval parameter to __flow_hash_from_keys which allows caller to
set the initval for jhash
- Perturb always does flow dissection and creates hash based on
input perturb value which acts as the keyval to __flow_hash_from_keys
- Added a _flow_keys_digest_data which is used in make_flow_keys_digest.
This fills out the digest by populating individual fields instead
of copying the whole structure.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Call make_flow_keys_digest to get a digest from flow keys and
use that to pass skbuff cb and for comparing flows.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some users of flow keys (well just sch_choke now) need to pass
flow_keys in skbuff cb, and use them for exact comparisons of flows
so that skb->hash is not sufficient. In order to increase size of
the flow_keys structure, we introduce another structure for
the purpose of passing flow keys in skbuff cb. We limit this structure
to sixteen bytes, and we will technically treat this as a digest of
flow_keys struct hence its name flow_keys_digest. In the first
incaranation we just copy the flow_keys structure up to 16 bytes--
this is the same information previously passed in the cb. In the
future, we'll adapt this for larger flow_keys and could use something
like SHA-1 over the whole flow_keys to improve the quality of the
digest.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call skb_get_hash_perturb instead of doing skb_flow_dissect and then
jhash by hand.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call skb_get_hash_perturb instead of doing skb_flow_dissect and then
jhash by hand.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call skb_get_hash_perturb instead of doing skb_flow_dissect and then
jhash by hand.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Call skb_get_hash_perturb instead of doing skb_flow_dissect and then
jhash by hand.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This calls flow_disect and __skb_get_hash to procure a hash for a
packet. Input includes a key to initialize jhash. This function
does not set skb->hash.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In setups with a global scope address on an interface, and a lesser
scope address on an interface sending IGMP reports, the reports can be
sent using the other interfaces global scope address rather than the
local interface address. RFC 2236 suggests:
Ignore the Report if you cannot identify the source address of
the packet as belonging to a subnet assigned to the interface on
which the packet was received.
since such reports could be forged.
Look at the protocol when deciding if a RT_SCOPE_LINK address should
be used for the packet.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
TC classifiers/actions were converted to RCU by John in the series:
http://thread.gmane.org/gmane.linux.network/329739/focus=329739
and many follow on patches.
This is the last patch from that series that finally drops
ingress spin_lock.
Single cpu ingress+u32 performance goes from 22.9 Mpps to 24.5 Mpps.
In two cpu case when both cores are receiving traffic on the same
device and go into the same ingress+u32 the performance jumps
from 4.5 + 4.5 Mpps to 23.5 + 23.5 Mpps
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
