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>

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>

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>

Kenneth Klette Jonassen says:

====================
tcp: SACK RTTM changes for congestion control

This patch series improves SACK RTT measurements for congestion control:
  o Picks the latest sequence SACKed for RTT, i.e. most accurate delay
    signal.
  o Calls the congestion control's pkts_acked hook with SACK RTTMs
    even when not sequentially ACKing new data.

V2: amend misleading comment
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Invoking pkts_acked is currently conditioned on FLAG_ACKED:
receiving a cumulative ACK of new data, or ACK with SYN flag set.

Remove this condition so that CC may get RTT measurements from all SACKs.

Cc: Yuchung Cheng <ycheng@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Kenneth Klette Jonassen <kennetkl@ifi.uio.no>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp_sacktag_one() always picks the earliest sequence SACKed for RTT.
This might not make sense for congestion control in cases where:

  1. ACKs are lost, i.e. a SACK following a lost SACK covers both
     new and old segments at the receiver.
  2. The receiver disregards the RFC 5681 recommendation to immediately
     ACK out-of-order segments.

Give congestion control a RTT for the latest segment SACKed, which is the
most accurate RTT estimate, but preserve the conservative RTT for RTO.

Removes the call to skb_mstamp_get() in tcp_sacktag_one().

Cc: Yuchung Cheng <ycheng@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Kenneth Klette Jonassen <kennetkl@ifi.uio.no>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Later patch passes two values set in tcp_sacktag_one() to
tcp_clean_rtx_queue(). Prepare passing them via struct tcp_sacktag_state.
Acked-by: Yuchung Cheng <ycheng@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Kenneth Klette Jonassen <kennetkl@ifi.uio.no>
Signed-off-by: David S. Miller <davem@davemloft.net>

Thomas Graf says:

====================
rhashtable self-test improvements

This series improves the rhashtable self-test to:
  * Avoid allocation of test objects
  * Measure the time of test runs
  * Use the iterator to walk the table for consistency
  * Account for failed insertions due to memory pressure or
    utilization pressure
  * Ignore failed insertions when checking for consistency
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Account for failed inserts due to memory pressure or EBUSY and
ignore failed entries during the consistency check.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

As resizes may continue to run in the background, use walker to
ensure we see all entries. Also print the encountered number
of rehashes queued up while traversing.

This may lead to warnings due to entries being seen multiple
times. We consider them non-fatal.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

By far the most expensive part of the selftest was the allocation
of entries. Using a static array allows to measure the rhashtable
operations.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

This only blows up the size of the test structure for no gain
in test coverage. Reduces size of test_obj from 24 to 16 bytes.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make test configurable by allowing to specify all relevant knobs
through module parameters.

Do several test runs and measure the average time it takes to
insert & remove all entries. Note, a deferred resize might still
continue to run in the background.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Alexander Duyck says:

====================
A few minor clean-ups to eth_type_trans

This series addresses a few minor issues I found in eth_type_trans that
that allow us to gain back something like 3 or more cycles per packet.

The first change is to drop the byte swap since it isn't necessary.  On x86
we could just check the first byte and compare that against the upper 8
bits of the Ethertype to determine if we are dealing with a size value or
not.

The second makes it so that the value we read in to test for multicast can
be used for the address comparison.  This allows us to avoid a second read
of the destination address.

The final change is to avoid some unneeded instructions in computing the
Ethernet header pointer.  When we start the call the Ethernet header is at
skb->data, so we just use that rather than computing mac_header, and then
adding that back to skb->head.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Avoid recomputing the Ethernet header location and instead just use the
pointer provided by skb->data.  The problem with using eth_hdr is that the
compiler wasn't smart enough to realize that skb->head + skb->mac_header
was the same thing as skb->data before it added ETH_HLEN.  By just caching
it off before calling skb_pull_inline we can avoid a few unnecessary
instructions.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This change makes it so that we process the address in
is_multicast_ether_addr at the same size as the other calls.  This allows
us to avoid duplicate reads when used with other calls such as
is_zero_ether_addr or eth_addr_copy.  In addition I have added a 64 bit
version of the function so in eth_type_trans we can process the destination
address as a 64 bit value throughout.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This change takes advantage of the fact that ETH_P_802_3_MIN is aligned to
512 so as a result we can actually ignore the lower 8b when comparing the
Ethertype to ETH_P_802_3_MIN.  This allows us to avoid a byte swap by simply
masking the value and comparing it to the byte swapped value for
ETH_P_802_3_MIN.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch divides the IPv6 flow label space into two ranges:
0-7ffff is reserved for flow label manager, 80000-fffff will be
used for creating auto flow labels (per RFC6438). This only affects how
labels are set on transmit, it does not affect receive. This range split
can be disbaled by systcl.

Background:

IPv6 flow labels have been an unmitigated disappointment thus far
in the lifetime of IPv6. Support in HW devices to use them for ECMP
is lacking, and OSes don't turn them on by default. If we had these
we could get much better hashing in IPv6 networks without resorting
to DPI, possibly eliminating some of the motivations to to define new
encaps in UDP just for getting ECMP.

Unfortunately, the initial specfications of IPv6 did not clarify
how they are to be used. There has always been a vague concept that
these can be used for ECMP, flow hashing, etc. and we do now have a
good standard how to this in RFC6438. The problem is that flow labels
can be either stateful or stateless (as in RFC6438), and we are
presented with the possibility that a stateless label may collide
with a stateful one.  Attempts to split the flow label space were
rejected in IETF. When we added support in Linux for RFC6438, we
could not turn on flow labels by default due to this conflict.

This patch splits the flow label space and should give us
a path to enabling auto flow labels by default for all IPv6 packets.
This is an API change so we need to consider compatibility with
existing deployment. The stateful range is chosen to be the lower
values in hopes that most uses would have chosen small numbers.

Once we resolve the stateless/stateful issue, we can proceed to
look at enabling RFC6438 flow labels by default (starting with
scaled testing).
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In my earlier commit:
653437d0 ("ipv6: Stop /128 route from disappearing after pmtu update"),
there was a horrible typo.  Instead of checking RTF_LOCAL on
rt->rt6i_flags, it was checked on rt->dst.flags.  This patch fixes
it.
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hajime Tazaki <tazaki@sfc.wide.ad.jp>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Not used.

pedit sets TC_MUNGED when packet content was altered, but all the core
does is unset MUNGED again and then set OK2MUNGE.

And the latter isn't tested anywhere. So lets remove both
TC_MUNGED and TC_OK2MUNGE.
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The whole hlist will be moved, so not need to call hlist_del before
add the hlist_node to other hlist_head.
Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge net into net-next.
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull networking fixes from David Miller:

 1) Receive packet length needs to be adjust by 2 on RX to accomodate
    the two padding bytes in altera_tse driver.  From Vlastimil Setka.

 2) If rx frame is dropped due to out of memory in macb driver, we leave
    the receive ring descriptors in an undefined state.  From Punnaiah
    Choudary Kalluri

 3) Some netlink subsystems erroneously signal NLM_F_MULTI.  That is
    only for dumps.  Fix from Nicolas Dichtel.

 4) Fix mis-use of raw rt->rt_pmtu value in ipv4, one must always go via
    the ipv4_mtu() helper.  From Herbert Xu.

 5) Fix null deref in bridge netfilter, and miscalculated lengths in
    jump/goto nf_tables verdicts.  From Florian Westphal.

 6) Unhash ping sockets properly.

 7) Software implementation of BPF divide did 64/32 rather than 64/64
    bit divide.  The JITs got it right.  Fix from Alexei Starovoitov.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (30 commits)
  ipv4: Missing sk_nulls_node_init() in ping_unhash().
  net: fec: Fix RGMII-ID mode
  net/mlx4_en: Schedule napi when RX buffers allocation fails
  netxen_nic: use spin_[un]lock_bh around tx_clean_lock
  net/mlx4_core: Fix unaligned accesses
  mlx4_en: Use correct loop cursor in error path.
  cxgb4: Fix MC1 memory offset calculation
  bnx2x: Delay during kdump load
  net: Fix Kernel Panic in bonding driver debugfs file: rlb_hash_table
  net: dsa: Fix scope of eeprom-length property
  net: macb: Fix race condition in driver when Rx frame is dropped
  hv_netvsc: Fix a bug in netvsc_start_xmit()
  altera_tse: Correct rx packet length
  mlx4: Fix tx ring affinity_mask creation
  tipc: fix problem with parallel link synchronization mechanism
  tipc: remove wrong use of NLM_F_MULTI
  bridge/nl: remove wrong use of NLM_F_MULTI
  bridge/mdb: remove wrong use of NLM_F_MULTI
  net: sched: act_connmark: don't zap skb->nfct
  trivial: net: systemport: bcmsysport.h: fix 0x0x prefix
  ...

Here the "other side" refers to the guest or host.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

With my job change kernel work will be "own time"; I'm keeping lguest
and modules (and the virtio standards work), but virtio kernel has to
go.

This makes it clear that Michael is in charge.  He's good, but having
me watch over his shoulder won't help.

Good luck Michael!
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>