When using ifb+netem on ingress on SIT/IPIP/GRE traffic,
GRO packets are not properly processed.

Segmentation should not be forced, since ifb is already adding
quite a performance hit.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add multiqueue capabilities to ifb netdevice.

This removes last bottleneck for ingress when mq qdisc can be used
to shard load from multiple RX queues on physical device.

Tested:

# netem based setup, installed at receiver side
ETH=eth0
IFB=ifb10
EST="est 1sec 4sec" # Optional rate estimator
RTT_HALF=2ms
#REORDER=20us
#LOSS="loss 1"
TXQ=8

ip link add ifb10 numtxqueues $TXQ type ifb
ip link set dev $IFB up

tc qdisc add dev $ETH ingress 2>/dev/null

tc filter add dev $ETH parent ffff: \
   protocol ip u32 match u32 0 0 flowid 1:1 \
	action mirred egress redirect dev $IFB

tc qdisc del dev $IFB root 2>/dev/null

tc qdisc add dev $IFB root handle 1: mq
for i in `seq 1 $TXQ`
do
 slot=$( printf %x $(( i )) )
 tc qd add dev $IFB parent 1:$slot $EST netem \
	limit 100000 delay $RTT_HALF $REORDER $LOSS
done

lpaa24:~# tc -s -d qd sh dev ifb10
qdisc mq 1: root
 Sent 316544766 bytes 5265927 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 98880b 1648p requeues 0
qdisc netem 8002: parent 1:1 limit 100000 delay 2.0ms
 Sent 39601416 bytes 658721 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38235Kbit 79657pps backlog 12240b 204p requeues 0
qdisc netem 8003: parent 1:2 limit 100000 delay 2.0ms
 Sent 39472866 bytes 657227 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38234Kbit 79655pps backlog 10620b 176p requeues 0
qdisc netem 8004: parent 1:3 limit 100000 delay 2.0ms
 Sent 39703417 bytes 659699 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38320Kbit 79831pps backlog 12780b 213p requeues 0
qdisc netem 8005: parent 1:4 limit 100000 delay 2.0ms
 Sent 39565149 bytes 658011 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38174Kbit 79530pps backlog 11880b 198p requeues 0
qdisc netem 8006: parent 1:5 limit 100000 delay 2.0ms
 Sent 39506078 bytes 657354 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38195Kbit 79571pps backlog 12480b 208p requeues 0
qdisc netem 8007: parent 1:6 limit 100000 delay 2.0ms
 Sent 39675994 bytes 658849 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38323Kbit 79838pps backlog 12600b 210p requeues 0
qdisc netem 8008: parent 1:7 limit 100000 delay 2.0ms
 Sent 39532042 bytes 658367 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38177Kbit 79536pps backlog 13140b 219p requeues 0
qdisc netem 8009: parent 1:8 limit 100000 delay 2.0ms
 Sent 39488164 bytes 657705 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38192Kbit 79568pps backlog 13Kb 222p requeues 0
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When you redirect a VLAN device to any device, you end up with
crap in af_packet on the xmit path because hard_header_len is
not equal to skb->mac_len.  So the redirected packet contains
four extra bytes at the start which then gets interpreted as
part of the MAC address.

This patch fixes this by only pushing skb->mac_len.  We also
need to fix ifb because it tries to undo the pushing done by
act_mirred.
Signed-off-by: NHerbert Xu <herbert@gondor.apana.org.au>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Testing xmit_more support with netperf and connected UDP sockets,
I found strange dst refcount false sharing.

Current handling of IFF_XMIT_DST_RELEASE is not optimal.

Dropping dst in validate_xmit_skb() is certainly too late in case
packet was queued by cpu X but dequeued by cpu Y

The logical point to take care of drop/force is in __dev_queue_xmit()
before even taking qdisc lock.

As Julian Anastasov pointed out, need for skb_dst() might come from some
packet schedulers or classifiers.

This patch adds new helper to cleanly express needs of various drivers
or qdiscs/classifiers.

Drivers that need skb_dst() in their ndo_start_xmit() should call
following helper in their setup instead of the prior :

	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
->
	netif_keep_dst(dev);

Instead of using a single bit, we use two bits, one being
eventually rebuilt in bonding/team drivers.

The other one, is permanent and blocks IFF_XMIT_DST_RELEASE being
rebuilt in bonding/team. Eventually, we could add something
smarter later.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Extend alloc_netdev{,_mq{,s}}() to take name_assign_type as argument, and convert
all users to pass NET_NAME_UNKNOWN.

Coccinelle patch:

@@
expression sizeof_priv, name, setup, txqs, rxqs, count;
@@

(
-alloc_netdev_mqs(sizeof_priv, name, setup, txqs, rxqs)
+alloc_netdev_mqs(sizeof_priv, name, NET_NAME_UNKNOWN, setup, txqs, rxqs)
|
-alloc_netdev_mq(sizeof_priv, name, setup, count)
+alloc_netdev_mq(sizeof_priv, name, NET_NAME_UNKNOWN, setup, count)
|
-alloc_netdev(sizeof_priv, name, setup)
+alloc_netdev(sizeof_priv, name, NET_NAME_UNKNOWN, setup)
)

v9: move comments here from the wrong commit
Signed-off-by: NTom Gundersen <teg@jklm.no>
Reviewed-by: NDavid Herrmann <dh.herrmann@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Do not include vlan acceleration features in vlan_features as that
precludes correct Q-in-Q operation.
Signed-off-by: NVlad Yasevich <vyasevic@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Replace the bh safe variant with the hard irq safe variant.

We need a hard irq safe variant to deal with netpoll transmitting
packets from hard irq context, and we need it in most if not all of
the places using the bh safe variant.

Except on 32bit uni-processor the code is exactly the same so don't
bother with a bh variant, just have a hard irq safe variant that
everyone can use.
Signed-off-by: N"Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In order to enable lockdep on seqcount/seqlock structures, we
must explicitly initialize any locks.

The u64_stats_sync structure, uses a seqcount, and thus we need
to introduce a u64_stats_init() function and use it to initialize
the structure.

This unfortunately adds a lot of fairly trivial initialization code
to a number of drivers. But the benefit of ensuring correctness makes
this worth while.

Because these changes are required for lockdep to be enabled, and the
changes are quite trivial, I've not yet split this patch out into 30-some
separate patches, as I figured it would be better to get the various
maintainers thoughts on how to best merge this change along with
the seqcount lockdep enablement.

Feedback would be appreciated!
Signed-off-by: NJohn Stultz <john.stultz@linaro.org>
Acked-by: NJulian Anastasov <ja@ssi.bg>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: James Morris <jmorris@namei.org>
Cc: Jesse Gross <jesse@nicira.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Mirko Lindner <mlindner@marvell.com>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Roger Luethi <rl@hellgate.ch>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Simon Horman <horms@verge.net.au>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Wensong Zhang <wensong@linux-vs.org>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/1381186321-4906-2-git-send-email-john.stultz@linaro.orgSigned-off-by: NIngo Molnar <mingo@kernel.org>

If __rtnl_link_register() return faild when loading the ifb, it will
take the wrong path and get oops, so fix it just like dummy.
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

According to the commit 16b0dc29
(dummy: fix rcu_sched self-detected stalls)

Eric Dumazet fix the problem in dummy, but the ifb will occur the
same problem like the dummy modules.

Trying to "modprobe ifb numifbs=30000" triggers :

INFO: rcu_sched self-detected stall on CPU

After this splat, RTNL is locked and reboot is needed.

We must call cond_resched() to avoid this, even holding RTNL.
Signed-off-by: NDing Tianhong <dingtianhong@huawei.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

- macvlan: propagate STAG filtering capabilities from underlying device
- ifb: announce STAG tagging support in addition to CTAG tagging support
- veth: announce STAG tagging/stripping support in addition to CTAG support
Signed-off-by: NPatrick McHardy <kaber@trash.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Rename the hardware VLAN acceleration features to include "CTAG" to indicate
that they only support CTAGs. Follow up patches will introduce 802.1ad
server provider tagging (STAGs) and require the distinction for hardware not
supporting acclerating both.
Signed-off-by: NPatrick McHardy <kaber@trash.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ifb should lookup devices in the appropriate namespace.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Replace usage of random_ether_addr() with eth_hw_addr_random()
to set addr_assign_type correctly to NET_ADDR_RANDOM.

Change the trivial cases.

v2: adapt to renamed eth_hw_addr_random()
Signed-off-by: NDanny Kukawka <danny.kukawka@bisect.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Only distinct use is checking if NETIF_F_NOCACHE_COPY should be
enabled by default. The check heuristics is altered a bit here,
so it hits other people than before. The default shouldn't be
trusted for performance-critical cases anyway.

For all other uses NETIF_F_NO_CSUM is equivalent to NETIF_F_HW_CSUM.
Signed-off-by: NMichał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After the last patch, We are left in a state in which only drivers calling
ether_setup have IFF_TX_SKB_SHARING set (we assume that drivers touching real
hardware call ether_setup for their net_devices and don't hold any state in
their skbs.  There are a handful of drivers that violate this assumption of
course, and need to be fixed up.  This patch identifies those drivers, and marks
them as not being able to support the safe transmission of skbs by clearning the
IFF_TX_SKB_SHARING flag in priv_flags
Signed-off-by: NNeil Horman <nhorman@tuxdriver.com>
CC: Karsten Keil <isdn@linux-pingi.de>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: Patrick McHardy <kaber@trash.net>
CC: Krzysztof Halasa <khc@pm.waw.pl>
CC: "John W. Linville" <linville@tuxdriver.com>
CC: Greg Kroah-Hartman <gregkh@suse.de>
CC: Marcel Holtmann <marcel@holtmann.org>
CC: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Convert input functional block device to use 64 bit stats.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Acked-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

* remove interrupt.g inclusion from netdevice.h -- not needed
* fixup fallout, add interrupt.h and hardirq.h back where needed.
Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Force dev_alloc_name() to be called from register_netdevice() by
dev_get_valid_name(). That allows to remove multiple explicit
dev_alloc_name() calls.

The possibility to call dev_alloc_name in advance remains.

This also fixes veth creation regresion caused by
84c49d8cSigned-off-by: NJiri Pirko <jpirko@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Le lundi 03 janvier 2011 à 11:40 -0800, David Miller a écrit :
> From: Jarek Poplawski <jarkao2@gmail.com>
> Date: Mon, 3 Jan 2011 20:37:03 +0100
>
> > On Sun, Jan 02, 2011 at 09:24:36PM +0100, Eric Dumazet wrote:
> >> Le mercredi 29 décembre 2010 ?? 00:07 +0100, Jarek Poplawski a écrit :
> >>
> >> > Ingress is before vlans handler so these features and the
> >> > NETIF_F_HW_VLAN_TX flag seem useful for ifb considering
> >> > dev_hard_start_xmit() checks.
> >>
> >> OK, here is v2 of the patch then, thanks everybody.
> >>
> >>
> >> [PATCH v2 net-next-2.6] ifb: add performance flags
> >>
> >> IFB can use the full set of features flags (NETIF_F_SG |
> >> NETIF_F_FRAGLIST | NETIF_F_TSO | NETIF_F_NO_CSUM | NETIF_F_HIGHDMA) to
> >> avoid unnecessary split of some packets (GRO for example)
> >>
> >> Changli suggested to also set vlan_features,
> >
> > He also suggested more GSO flags of which especially NETIF_F_TSO6
> > seems interesting (wrt GRO)?
>
> I think at least TSO6 would very much be appropriate here.

Yes, why not, I am only wondering why loopback / dummy (and others ?)
only set NETIF_F_TSO :)

Since I want to play with ECN, I might also add NETIF_F_TSO_ECN ;)

For other flags, I really doubt it can matter on ifb ?

[PATCH v3 net-next-2.6] ifb: add performance flags

IFB can use the full set of features flags (NETIF_F_SG |
NETIF_F_FRAGLIST | NETIF_F_TSO | NETIF_F_NO_CSUM | NETIF_F_HIGHDMA) to
avoid unnecessary split of some packets (GRO for example)

Changli suggested to also set vlan_features, NETIF_F_TSO6,
NETIF_F_TSO_ECN.

Jarek suggested to add NETIF_F_HW_VLAN_TX as well.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Cc: Changli Gao <xiaosuo@gmail.com>
Cc: Jarek Poplawski <jarkao2@gmail.com>
Cc: Pawel Staszewski <pstaszewski@itcare.pl>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In ri_tasklet(), we run from softirq, so can directly handle packet
through netif_receive_skb() instead of netif_rx().
There is no risk of recursion.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Acked-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After recent ifb changes, we must use lockless __skb_dequeue() since
lock is not anymore initialized.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Cc: Jamal Hadi Salim <hadi@cyberus.ca>
Cc: Changli Gao <xiaosuo@gmail.com>
Acked-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

rq and tq are both protected by tx queue lock, so we can simply use
the lockless variants of skb_queue.

skb_queue_splice_tail_init() is used instead of the open coded and slow
one.
Signed-off-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These debug stats are not exported, and become useless.
Signed-off-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If we break the loop when there are still skbs in tq and no skb in
rq, the skbs will be left in txq until new skbs are enqueued into rq.
In rare cases, no new skb is queued, then these skbs will stay in rq
forever.

After this patch, if tq isn't empty when we break the loop, we goto
resched directly.
Signed-off-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that core network takes care of trans_start updates, dont do it
in drivers themselves, if possible. Drivers can avoid one cache miss
(on dev->trans_start) in their start_xmit() handler.

Exceptions are NETIF_F_LLTX drivers
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

To help grep games, rename iif to skb_iif
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Avoids touching dev refcount in hotpath
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

At this point (ri_tasklet()), RTNL or dev_base_lock are not held,
we must use dev_get_by_index() instead of __dev_get_by_index()
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These are all drivers that don't touch real hardware.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch converts the remaining occurences of raw return values to their
symbolic counterparts in ndo_start_xmit() functions that were missed by the
previous automatic conversion.

Additionally code that assumed the symbolic value of NETDEV_TX_OK to be zero
is changed to explicitly use NETDEV_TX_OK.
Signed-off-by: NPatrick McHardy <kaber@trash.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

One point of contention in high network loads is the dst_release() performed
when a transmited skb is freed. This is because NIC tx completion calls
dev_kree_skb() long after original call to dev_queue_xmit(skb).

CPU cache is cold and the atomic op in dst_release() stalls. On SMP, this is
quite visible if one CPU is 100% handling softirqs for a network device,
since dst_clone() is done by other cpus, involving cache line ping pongs.

It seems right place to release dst is in dev_hard_start_xmit(), for most
devices but ones that are virtual, and some exceptions.

David Miller suggested to define a new device flag, set in alloc_netdev_mq()
(so that most devices set it at init time), and carefuly unset in devices
which dont want a NULL skb->dst in their ndo_start_xmit().

List of devices that must clear this flag is :

- loopback device, because it calls netif_rx() and quoting Patrick :
    "ip_route_input() doesn't accept loopback addresses, so loopback packets
     already need to have a dst_entry attached."
- appletalk/ipddp.c : needs skb->dst in its xmit function

- And all devices that call again dev_queue_xmit() from their xmit function
(as some classifiers need skb->dst) : bonding, vlan, macvlan, eql, ifb, hdlc_fr
Signed-off-by: NEric Dumazet <dada1@cosmosbay.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch moves neigh_setup and hard_start_xmit into the network device ops
structure. For bisection, fix all the previously converted drivers as well.
Bonding driver took the biggest hit on this.

Added a prefetch of the hard_start_xmit in the fast path to try and reduce
any impact this would have.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Convert to new network device ops interface.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When support for multiple TX queues were added, the
netif_tx_lock() routines we converted to iterate over
all TX queues and grab each queue's spinlock.

This causes heartburn for lockdep and it's not a healthy
thing to do with lots of TX queues anyways.

So modify this to use a top-level lock and a "frozen"
state for the individual TX queues.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We can simply use the qdisc->q.lock for all of the
qdisc tree synchronization.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

alloc_netdev_mq() now allocates an array of netdev_queue
structures for TX, based upon the queue_count argument.

Furthermore, all accesses to the TX queues are now vectored
through the netdev_get_tx_queue() and netdev_for_each_tx_queue()
interfaces.  This makes it easy to grep the tree for all
things that want to get to a TX queue of a net device.

Problem spots which are not really multiqueue aware yet, and
only work with one queue, can easily be spotted by grepping
for all netdev_get_tx_queue() calls that pass in a zero index.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Every qdisc is assosciated with a queue, and in the case of ingress
qdiscs that will now be netdev->rx_queue so using that queue's lock is
the thing to do.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The lock is now an attribute of the device queue.

One thing to notice is that "suspicious" places
emerge which will need specific training about
multiple queue handling.  They are so marked with
explicit "netdev->rx_queue" and "netdev->tx_queue"
references.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>