netdev_wait_allrefs() waits that all references to a device vanishes.

It currently uses a _very_ pessimistic 250 ms delay between each probe.
Some users reported that no more than 4 devices can be dismantled per
second, this is a pretty serious problem for some setups.

Most of the time, a refcount is about to be released by an RCU callback,
that is still in flight because rollback_registered_many() uses a
synchronize_rcu() call instead of rcu_barrier(). Problem is visible if
number of online cpus is one, because synchronize_rcu() is then a no op.

time to remove 50 ipip tunnels on a UP machine :

before patch : real 11.910s
after patch : real 1.250s
Reported-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Reported-by: NOctavian Purdila <opurdila@ixiacom.com>
Reported-by: NBenjamin LaHaise <bcrl@kvack.org>
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When a net device is implementing the select_queue callback and is part of
a bridge, frames coming from the bridge already have a tx queue associated
to the socket (introduced in commit a4ee3ce3,
"net: Use sk_tx_queue_mapping for connected sockets"). The call to
sk_tx_queue_get will then return the tx queue used by the bridge instead
of calling the select_queue callback.

In case of mac80211 this broke QoS which is implemented by using the
select_queue callback. Furthermore it introduced problems with rt2x00
because frames with the same TID and RA sometimes appeared on different
tx queues which the hw cannot handle correctly.

Fix this by always calling select_queue first if it is available and only
afterwards use the socket tx queue mapping.
Signed-off-by: NHelmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

>Xin Xiaohui wrote:
> I looked into the code dev_gro_receive(), found the code here:
> if the frags[0] is pulled to 0, then the page will be released,
> and memmove() frags left.
> Is that right? I'm not sure if memmove do right or not, but
> frags[0].size is never set after memove at least. what I think
> a simple way is not to do anything if we found frags[0].size == 0.
> The patch is as followed.
...

This version of the patch fixes the bug directly in memmove.
Reported-by: N"Xin, Xiaohui" <xiaohui.xin@intel.com>
Signed-off-by: NJarek Poplawski <jarkao2@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Although netif_rx() isn't expected to be called in process context with
preemption enabled, it'd better handle this case. And this is why get_cpu()
is used in the non-RPS #ifdef branch. If tree RCU is selected,
rcu_read_lock() won't disable preemption, so preempt_disable() should be
called explictly.
Signed-off-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The netpoll_rx_on() check in __napi_gro_receive() skips part of the
"common" GRO_NORMAL path, especially "pull:" in dev_gro_receive(),
where at least eth header should be copied for entirely paged skbs.
Signed-off-by: NJarek Poplawski <jarkao2@gmail.com>
Acked-by: NHerbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This reverts commit 15e83ed7.

As explained by Johannes Berg, the optimization made here is
invalid.  Or, at best, incomplete.

Not only destructor invocation, but conntract entry releasing
must be executed outside of hw IRQ context.

So just checking "skb->destructor" is insufficient.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit ab95bfe0 replaces bridge and macvlan
hooks in __netif_receive_skb(), so dev.c doesn't need to include their headers.
Signed-off-by: NChangli Gao <xiaosuo@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If user misconfigures ingress and causes a redirection loop, don't
overwhelm the log.  This is also a error case so make it unlikely.
Found by inspection, luckily not in real system.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With conn-track zones and probably with different network
namespaces, the netfilter logic needs to be re-calculated
on packet receive.  If the netfilter logic is not reset,
it will not be recalculated properly.  This patch adds
the nf_reset logic to dev_forward_skb.
Signed-off-by: NBen Greear <greearb@candelatech.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since struct netdev_queue tx_bytes/tx_packets/tx_dropped are already
protected by _xmit_lock, its easy to convert these fields to u64 instead
of unsigned long.
This completes 64bit stats for devices using them (vlan, macvlan, ...)

Strictly, we could avoid the locking in dev_txq_stats_fold() on 64bit
arches, but its slow path and we prefer keep it simple.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds a new networking option to allow hardware time stamps
from PHY devices. When enabled, likely candidates among incoming and
outgoing network packets are offered to the PHY driver for possible
time stamping. When accepted by the PHY driver, incoming packets are
deferred for later delivery by the driver.

The patch also adds phylib driver methods for the SIOCSHWTSTAMP ioctl
and callbacks for transmit and receive time stamping. Drivers may
optionally implement these functions.
Signed-off-by: NRichard Cochran <richard.cochran@omicron.at>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix problem in reading the tx_queue recorded in a socket.  In
dev_pick_tx, the TX queue is read by doing a check with
sk_tx_queue_recorded on the socket, followed by a sk_tx_queue_get.
The problem is that there is not mutual exclusion across these
calls in the socket so it it is possible that the queue in the
sock can be invalidated after sk_tx_queue_recorded is called so
that sk_tx_queue get returns -1, which sets 65535 in queue_index
and thus dev_pick_tx returns 65536 which is a bogus queue and
can cause crash in dev_queue_xmit.

We fix this by only calling sk_tx_queue_get which does the proper
checks.  The interface is that sk_tx_queue_get returns the TX queue
if the sock argument is non-NULL and TX queue is recorded, else it
returns -1.  sk_tx_queue_recorded is no longer used so it can be
completely removed.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

commit fc6055a5 (net: Introduce skb_orphan_try()) added early
orphaning of skbs.

This unfortunately added a performance regression in skb_tx_hash() in
case of stacked devices (bonding, vlans, ...)

Since skb->sk is now NULL, we cannot access sk->sk_hash anymore to
spread tx packets to multiple NIC queues on multiqueue devices.

skb_tx_hash() in this case only uses skb->protocol, same value for all
flows.

skb_orphan_try() can copy sk->sk_hash into skb->rxhash and skb_tx_hash()
can use this saved sk_hash value to compute its internal hash value.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Document that dev_get_stats() returns the same stats pointer it was
given.  Remove const qualification from the returned pointer since the
caller may do what it likes with that structure.
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In commit be1f3c2c "net: Enable 64-bit
net device statistics on 32-bit architectures" I redefined struct
net_device_stats so that it could be used in a union with struct
rtnl_link_stats64, avoiding the need for explicit copying or
conversion between the two.  However, this is unsafe because there is
no locking required and no lock consistently held around calls to
dev_get_stats() and use of the statistics structure it returns.

In commit 28172739 "net: fix 64 bit
counters on 32 bit arches" Eric Dumazet dealt with that problem by
requiring callers of dev_get_stats() to provide storage for the
result.  This means that the net_device::stats64 field and the padding
in struct net_device_stats are now redundant, so remove them.

Update the comment on net_device_ops::ndo_get_stats64 to reflect its
new usage.

Change dev_txq_stats_fold() to use struct rtnl_link_stats64, since
that is what all its callers are really using and it is no longer
going to be compatible with struct net_device_stats.

Eric Dumazet suggested the separate function for the structure
conversion.
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Acked-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There is a small possibility that a reader gets incorrect values on 32
bit arches. SNMP applications could catch incorrect counters when a
32bit high part is changed by another stats consumer/provider.

One way to solve this is to add a rtnl_link_stats64 param to all
ndo_get_stats64() methods, and also add such a parameter to
dev_get_stats().

Rule is that we are not allowed to use dev->stats64 as a temporary
storage for 64bit stats, but a caller provided area (usually on stack)

Old drivers (only providing get_stats() method) need no changes.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Reduces an x86 defconfig text and data ~2k.
text is smaller, data is larger.

$ size vmlinux*
   text	   data	    bss	    dec	    hex	filename
7198862	 720112	1366288	9285262	 8dae8e	vmlinux
7205273	 716016	1366288	9287577	 8db799	vmlinux.device_h

Uses %pV and struct va_format
Format arguments are verified before printk
Signed-off-by: NJoe Perches <joe@perches.com>
Acked-by: NGreg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Reducing real_num_queues needs to flush the qdisc otherwise
skbs with queue_mappings greater then real_num_tx_queues can
be sent to the underlying driver.

The flow for this is,

dev_queue_xmit()
	dev_pick_tx()
		skb_tx_hash()  => hash using real_num_tx_queues
		skb_set_queue_mapping()
	...
	qdisc_enqueue_root() => enqueue skb on txq from hash
...
dev->real_num_tx_queues -= n
...
sch_direct_xmit()
	dev_hard_start_xmit()
		ndo_start_xmit(skb,dev) => skb queue set with old hash

skbs are enqueued on the qdisc with skb->queue_mapping set
0 < queue_mappings < real_num_tx_queues.  When the driver
decreases real_num_tx_queues skb's may be dequeued from the
qdisc with a queue_mapping greater then real_num_tx_queues.

This fixes a case in ixgbe where this was occurring with DCB
and FCoE. Because the driver is using queue_mapping to map
skbs to tx descriptor rings we can potentially map skbs to
rings that no longer exist.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is only noticed by people that are not doing everything correct in
the first place.
Signed-off-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

netif_needs_gso() is checked twice in the TX path once,
before submitting the skb to the qdisc and once after
it is dequeued from the qdisc just before calling
ndo_hard_start().  This opens a window for a user to
change the gso/tso or tx checksum settings that can
cause netif_needs_gso to be true in one check and false
in the other.

Specifically, changing TX checksum setting may cause
the warning in skb_gso_segment() to be triggered if
the checksum is calculated earlier.

This consolidates the netif_needs_gso() calls so that
the stack only checks if gso is needed in
dev_hard_start_xmit().
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: NHerbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Register net_bridge_port pointer as rx_handler data pointer. As br_port is
removed from struct net_device, another netdev priv_flag is added to indicate
the device serves as a bridge port. Also rcuized pointers are now correctly
dereferenced in br_fdb.c and in netfilter parts.
Signed-off-by: NJiri Pirko <jpirko@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add possibility to register rx_handler data pointer along with a rx_handler.
Signed-off-by: NJiri Pirko <jpirko@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use struct rtnl_link_stats64 as the statistics structure.

On 32-bit architectures, insert 32 bits of padding after/before each
field of struct net_device_stats to make its layout compatible with
struct rtnl_link_stats64.  Add an anonymous union in net_device; move
stats into the union and add struct rtnl_link_stats64 stats64.

Add net_device_ops::ndo_get_stats64, implementations of which will
return a pointer to struct rtnl_link_stats64.  Drivers that implement
this operation must not update the structure asynchronously.

Change dev_get_stats() to call ndo_get_stats64 if available, and to
return a pointer to struct rtnl_link_stats64.  Change callers of
dev_get_stats() accordingly.
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently, the accelerated receive path for VLAN's will
drop packets if the real device is an inactive slave and
is not one of the special pkts tested for in
skb_bond_should_drop().  This behavior is different then
the non-accelerated path and for pkts over a bonded vlan.

For example,

vlanx -> bond0 -> ethx

will be dropped in the vlan path and not delivered to any
packet handlers at all.  However,

bond0 -> vlanx -> ethx

and

bond0 -> ethx

will be delivered to handlers that match the exact dev,
because the VLAN path checks the real_dev which is not a
slave and netif_recv_skb() doesn't drop frames but only
delivers them to exact matches.

This patch adds a sk_buff flag which is used for tagging
skbs that would previously been dropped and allows the
skb to continue to skb_netif_recv().  Here we add
logic to check for the deliver_no_wcard flag and if it
is set only deliver to handlers that match exactly.  This
makes both paths above consistent and gives pkt handlers
a way to identify skbs that come from inactive slaves.
Without this patch in some configurations skbs will be
delivered to handlers with exact matches and in others
be dropped out right in the vlan path.

I have tested the following 4 configurations in failover modes
and load balancing modes.

# bond0 -> ethx

# vlanx -> bond0 -> ethx

# bond0 -> vlanx -> ethx

# bond0 -> ethx
            |
  vlanx -> --
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

BugLink: http://bugs.launchpad.net/bugs/591416

There are a number of network drivers (bridge, bonding, etc) that are not yet
receive multi-queue enabled and use alloc_netdev(), so don't print a
num_rx_queues imbalance warning in that case.

Also, only print the warning once for those drivers that _are_ multi-queue
enabled.
Signed-off-by: NTim Gardner <tim.gardner@canonical.com>
Acked-by: NEric Dumazet <eric.dumazet@gmail.com>

- dev_get_by_flags() changed to dev_get_by_flags_rcu()

- ipv6_sock_ac_join() dont touch dev & idev refcounts
- ipv6_sock_ac_drop() dont touch dev & idev refcounts
- ipv6_sock_ac_close() dont touch dev & idev refcounts
- ipv6_dev_ac_dec() dount touch idev refcount
- ipv6_chk_acast_addr() dont touch idev refcount
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
CC: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The extra assertion to allow packet munging only when there are
no other ptypes listening which may have worked around an old bug
is unnecessary. It is sufficient to check if the skb is cloned before
trampling on it. Thanks to Herbert Xu for being persistent and patient
in getting this across.
[Note that cloning checks and assertions are the general rule used
by tc actions (documentation/networking/tc-actions-env-rules.txt)].
Signed-off-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We can avoid an unecessary cache miss by checking if the skb is non-linear
before accessing gso_size/gso_type in skb_warn_if_lro, the same can also be
done to avoid a cache miss on nr_frags if data_len is 0.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

What this patch does is it removes two receive frame hooks (for bridge and for
macvlan) from __netif_receive_skb. These are replaced them with a single
hook for both. It only supports one hook per device because it makes no
sense to do bridging and macvlan on the same device.

Then a network driver (of virtual netdev like macvlan or bridge) can register
an rx_handler for needed net device.
Signed-off-by: NJiri Pirko <jpirko@redhat.com>
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When many cpus compete for sending frames on a given qdisc, the qdisc
spinlock suffers from very high contention.

The cpu owning __QDISC_STATE_RUNNING bit has same priority to acquire
the lock, and cannot dequeue packets fast enough, since it must wait for
this lock for each dequeued packet.

One solution to this problem is to force all cpus spinning on a second
lock before trying to get the main lock, when/if they see
__QDISC_STATE_RUNNING already set.

The owning cpu then compete with at most one other cpu for the main
lock, allowing for higher dequeueing rate.

Based on a previous patch from Alexander Duyck. I added the heuristic to
avoid the atomic in fast path, and put the new lock far away from the
cache line used by the dequeue worker. Also try to release the busylock
lock as late as possible.

Tests with following script gave a boost from ~50.000 pps to ~600.000
pps on a dual quad core machine (E5450 @3.00GHz), tg3 driver.
(A single netperf flow can reach ~800.000 pps on this platform)

for j in `seq 0 3`; do
  for i in `seq 0 7`; do
    netperf -H 192.168.0.1 -t UDP_STREAM -l 60 -N -T $i -- -m 6 &
  done
done
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Acked-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If a skb is received on an inactive bond that does not meet
the special cases checked for by skb_bond_should_drop it should
only be delivered to exact matches as the comment in
netif_receive_skb() says.

However because null_or_bond could also be null this is not
always true.  This patch renames null_or_bond to orig_or_bond
and initializes it to orig_dev.  This keeps the intent of
null_or_bond to pass frames received on VLAN interfaces stacked
on bonding interfaces without invalidating the statement for
null_or_orig.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Define three helpers to manipulate QDISC_STATE_RUNNIG flag, that a
second patch will move on another location.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

netpoll does an interesting work in zap_completion_queue(), but this was
before we did skb orphaning before delivering packets to device.

It now makes sense to add a test in dev_kfree_skb_irq() to not queue a
skb if already orphaned, and to remove netpoll zap_completion_queue() as
a bonus.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch saves 224 bytes of text on my machine.

__this_cpu_inc() generates a single instruction, using no scratch
registers :

  65 ff 04 25 a8 30 01 00      incl   %gs:0x130a8

instead of :

  48 c7 c2 80 30 01 00         mov    $0x13080,%rdx
  65 48 8b 04 25 88 ea 00 00   mov    %gs:0xea88,%rax
  83 44 10 28 01               addl   $0x1,0x28(%rax,%rdx,1)
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

the commit:

commit d9031024
Author: Octavian Purdila <opurdila@ixiacom.com>
Date:   Wed Nov 18 02:36:59 2009 +0000

    net: device name allocation cleanups

introduced a bug when there is a hash collision making impossible
to rename a device with eth%d. This bug is very hard to reproduce
and appears rarely.

The problem is coming from we don't pass a temporary buffer to
__dev_alloc_name but 'dev->name' which is modified by the function.

A detailed explanation is here:

http://marc.info/?l=linux-netdev&m=127417784011987&w=2

Changelog:
 V2 : replaced strings comparison by pointers comparison
Signed-off-by: NDaniel Lezcano <daniel.lezcano@free.fr>
Reviewed-by: NOctavian Purdila <opurdila@ixiacom.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This reverts commit aaf8cdc3.

Drivers like the ipw2100 call device_create_group when they
are initialized and device_remove_group when they are shutdown.
Moving them between namespaces deletes their sysfs groups early.

In particular the following call chain results.
netdev_unregister_kobject -> device_del -> kobject_del -> sysfs_remove_dir
With sysfs_remove_dir recursively deleting all of it's subdirectories,
and nothing adding them back.

Ouch!

Therefore we need to call something that ultimate calls sysfs_mv_dir
as that sysfs function can move sysfs directories between namespaces
without deleting their subdirectories or their contents.   Allowing
us to avoid placing extra boiler plate into every driver that does
something interesting with sysfs.

Currently the function that provides that capability is device_rename.
That is the code works without nasty side effects as originally written.

So remove the misguided fix for moving devices between namespaces.  The
bug in the kobject layer that inspired it has now been recognized and
fixed.
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NGreg Kroah-Hartman <gregkh@suse.de>

Fix some issues introduced in batch skb dequeuing for input_pkt_queue.
The primary issue it that the queue head must be incremented only
after a packet has been processed, that is only after
__netif_receive_skb has been called.  This is needed for the mechanism
to prevent OOO packet in RFS.  Also when flushing the input_pkt_queue
and process_queue, the process queue should be done first to prevent
OOO packets.

Because the input_pkt_queue has been effectively split into two queues,
the calculation of the tail ptr is no longer correct.  The correct value
would be head+input_pkt_queue->len+process_queue->len.  To avoid
this calculation we added an explict input_queue_tail in softnet_data.
The tail value is simply incremented when queuing to input_pkt_queue.
Signed-off-by: NTom Herbert <therbert@google.com>
Acked-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use low order bit of skb->_skb_dst to tell dst is not refcounted.

Change _skb_dst to _skb_refdst to make sure all uses are catched.

skb_dst() returns the dst, regardless of noref bit set or not, but
with a lockdep check to make sure a noref dst is not given if current
user is not rcu protected.

New skb_dst_set_noref() helper to set an notrefcounted dst on a skb.
(with lockdep check)

skb_dst_drop() drops a reference only if skb dst was refcounted.

skb_dst_force() helper is used to force a refcount on dst, when skb
is queued and not anymore RCU protected.

Use skb_dst_force() in __sk_add_backlog(), __dev_xmit_skb() if
!IFF_XMIT_DST_RELEASE or skb enqueued on qdisc queue, in
sock_queue_rcv_skb(), in __nf_queue().

Use skb_dst_force() in dev_requeue_skb().

Note: dst_use_noref() still dirties dst, we might transform it
later to do one dirtying per jiffies.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If CONFIG_SMP=y, then we own a queue spinlock, we can avoid the atomic
test_and_set_bit() from napi_schedule_prep().

We now have same number of atomic ops per netif_rx() calls than with
pre-RPS kernel.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With RPS inclusion, skb timestamping is not consistent in RX path.

If netif_receive_skb() is used, its deferred after RPS dispatch.

If netif_rx() is used, its done before RPS dispatch.

This can give strange tcpdump timestamps results.

I think timestamping should be done as soon as possible in the receive
path, to get meaningful values (ie timestamps taken at the time packet
was delivered by NIC driver to our stack), even if NAPI already can
defer timestamping a bit (RPS can help to reduce the gap)

Tom Herbert prefer to sample timestamps after RPS dispatch. In case
sampling is expensive (HPET/acpi_pm on x86), this makes sense.

Let admins switch from one mode to another, using a new
sysctl, /proc/sys/net/core/netdev_tstamp_prequeue

Its default value (1), means timestamps are taken as soon as possible,
before backlog queueing, giving accurate timestamps.

Setting a 0 value permits to sample timestamps when processing backlog,
after RPS dispatch, to lower the load of the pre-RPS cpu.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>