rds-stress experiments with request size 256 bytes, 8K acks,
using 16 threads show a 40% improvment when pskb_extract()
replaces the {skb_clone(..); pskb_pull(..); pskb_trim(..);}
pattern in the Rx path, so we leverage the perf gain with
this commit.
Signed-off-by: NSowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.

The existing code uses the sequence
	clone = skb_clone(..);
	pskb_pull(clone, off, ..);
	pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().

To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: NSowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This works on the same implementation principle as
codel*.h, i.e. there's a generic header with
structures and macros and a implementation header
carrying function definitions to include in given,
e.g. driver or module.

The fairness logic comes from
net/sched/sch_fq_codel.c but is generalized so it
is more flexible and easier to re-use.
Signed-off-by: NMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Michal Kazior says:

====================
codel: make it reuseable beyond qdiscs

There's an ongoing effort in fixing wireless
bufferbloat. As part of that fq_codel is being
ported into mac80211. To prevent code duplication
codel.h needs to be slightly modified before it
can be used in mac80211 (or other drivers FWIW).

For more background please see:

  https://www.spinics.net/lists/linux-wireless/msg149976.html
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It was impossible to include codel.h for the
purpose of having access to codel_params or
codel_vars structure definitions and using them
for embedding in other more complex structures.

This splits allows codel.h itself to be treated
like any other header file while codel_qdisc.h and
codel_impl.h contain function definitions with
logic that was previously in codel.h.

This copies over copyrights and doesn't involve
code changes other than adding a few additional
include directives to net/sched/sch*codel.c.
Signed-off-by: NMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This strips out qdisc specific bits from the code
and makes it slightly more reusable. Codel will be
used by wireless/mac80211 in the future.
Signed-off-by: NMichal Kazior <michal.kazior@tieto.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NPhil Sutter <phil@nwl.cc>
Acked-by: NSabrina Dubroca <sd@queasysnail.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit 751a587a ("route: fix breakage after moving lwtunnel state")
moved lwtstate to the end of dst_entry for 32bit archs. This makes it share
the cacheline with __refcnt which had an unkown effect on performance. For
this reason, the pointer was kept in place for 64bit archs.

However, later performance measurements showed this is of no concern. It
turns out that every performance sensitive path that accesses lwtstate
accesses also struct rtable or struct rt6_info which share the same cache
line.

Thus, to get rid of a few #ifdefs, move the field to the end of the struct
also for 64bit.
Signed-off-by: NJiri Benc <jbenc@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Yuval Mintz says:

====================
qed*: driver updates

[Was previous termed 'eeprom access et al.', but seemed a bit
inappropriate given we've dropped the eeprom patch for now.
Still waiting for some inputs on that one, BTW]

This patch series contains some ethtool-related enhancements.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The APIs for making this sort of configuration [e.g., via ethtool] are
already present in qede, but the current configuration flow in qed doesn't
respect it.
Signed-off-by: NSudarsana Reddy Kalluru <sudarsana.kalluru@qlogic.com>
Signed-off-by: NYuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There's some inconsistency in current logic determining whether the
link settings of a given interface can be changed; I.e., in all modes
other than the so-called `deault' mode the interfaces are forbidden from
changing the configuration - but even this rule is not applied to all
user APIs that may change the configuration.

Instead, let the core-module [qed] decide whether an interface can change
the configuration by supporting a new API function. We also revise the
current rule, allowing all interfaces to change their configurations while
laying the infrastructure for future modes where an interface would be
blocked from making such a configuration.
Signed-off-by: NYuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adds a getter for the interfaces private flags.
The only parameter currently supported is whether the interface is a
coupled function [required for supporting 100g].
Signed-off-by: NYuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There's a difference in statsitics' names starting at qed and
propagating to qede, where egress counters indicate ranges while ingress
counters indiciate high-end.
Align all statistcs to follow the same conventions - name indicates range.
Signed-off-by: NYuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We should call consume_skb(skb) when skb is properly consumed,
or kfree_skb(skb) when skb must be dropped in error case.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We now have proper per-listener but also per network namespace counters
for SYN packets that might be dropped.

We replace the kfree_skb() by consume_skb() to be drop monitor [1]
friendly, and remove an obsolete comment.
FastOpen SYN packets can carry payload in them just fine.

[1] perf record -a -g -e skb:kfree_skb sleep 1; perf report
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
10GbE Intel Wired LAN Driver Updates 2016-04-25

This series contains updates to ixgbe and ixgbevf.

Emil provides several patches, starting with the consolidation of the
logic behind configuring spoof checking.  Fixed an issue which was
causing link issues for backplane devices because x550em_a/x devices
did not have a default value for mac->ops.setup_link.  Refactored the
ethtool stats to bring the logic closer to how ixgbe handles stats and
sets up per-queue stats for ixgbevf.

Mark adds a new register to wait for previous register writes to complete
before issuing a register read, which is needed when slower links are
in use.  Fixed the flow control setup for x550em_a, the incorrect
fc_setup function was being used.

Don added a workaround for empty SFP+ cage crosstalk, since on some
systems the crosstalk could lead to link flap on empty SFP+ cages.

Jake converts ixgbe and ixgbevf to use the BIT() macro.

Alex Duyck adds support for partial GSO segmentation in the case of
tunnels for ixgbe and ixgbevf.  Then preps for HyperV by moving the API
negotiation into mac_ops.

Arnd Bergmann provides a fix for the ARM compile warnings in linux-next
by converting the use of a udelay() to msleep().
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Nicolas Dichtel says:

====================
netlink: align attributes when needed (patchset #2)

This is the continuation (series #2) of the work done to align netlink
attributes when these attributes contain some 64-bit fields.

In patch #3, I didn't modify the function ila_encap_nlsize(). I was waiting
feedback for this patch: http://patchwork.ozlabs.org/patch/613766/
If it's approved, there will be an update to switch nla_total_size() to
nla_total_size_64bit() after the merge of net in net-next.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

d894ba18 ("soreuseport: fix ordering for mixed v4/v6 sockets")
was merged as a bug fix to the net tree.  Two conflicting changes
were committed to net-next before the above fix was merged back to
net-next:
ca065d0c ("udp: no longer use SLAB_DESTROY_BY_RCU")
3b24d854 ("tcp/dccp: do not touch listener sk_refcnt under synflood")

These changes switched the datastructure used for TCP and UDP sockets
from hlist_nulls to hlist.  This patch applies the necessary parts
of the net tree fix to net-next which were not automatic as part of the
merge.

Fixes: 1602f49b ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net")
Signed-off-by: NCraig Gallek <kraig@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Valdis reported tons of stack dumps caused by WARN_ON() in
sock_owned_by_user()

This test needs to be relaxed if/when lockdep disables itself.

Note that other lockdep_sock_is_held() callers are all from
rcu_dereference_protected() sections which already are disabled
if/when lockdep has been disabled.

Fixes: fafc4e1e ("sock: tigthen lockdep checks for sock_owned_by_user")
Reported-by: NValdis Kletnieks <Valdis.Kletnieks@vt.edu>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The newly added x550em_a support causes a link failure on ARM because of
an overly long time passed into udelay():

ERROR: "__bad_udelay" [drivers/net/ethernet/intel/ixgbe/ixgbe.ko] undefined!

There are multiple variants of the ixgbe_acquire_swfw_sync_*() function,
and the other ones all use msleep(), so we can safely assume that all
callers are allowed to sleep, which makes msleep() a better replacement
than mdelay().
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Fixes: 49425dfc ("ixgbe: Add support for x550em_a 10G MAC type")
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch moves API negotiation into mac_ops.  The general idea here is
that with HyperV on the way we need to make certain that anything that will
have different versions between HyperV and a standard VF needs to be
abstracted enough so that we can have a separate function between the two
so we can avoid changes in one breaking something in the other.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds support for partial GSO segmentation in the case of
tunnels.  Specifically with this change the driver an perform segmentation
as long as the frame either has IPv6 inner headers, or we are allowed to
mangle the IP IDs on the inner header.  This is needed because we will not
be modifying any fields from the start of the start of the outer transport
header to the start of the inner transport header as we are treating them
like they are just a block of IP options.
Signed-off-by: NAlexander Duyck <aduyck@mirantis.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Also cleanup a case where we're bit shifting a value into place, and use
an unsigned constant. Make use of the unsigned postfix in places where
BIT() macro is not appropriate.
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Make use of GENMASK instead of open coding the equivalent operation
incorrectly.
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Several areas of ixgbe were written before widespread usage of the
BIT(n) macro. With the impending release of GCC 6 and its associated new
warnings, some usages such as (1 << 31) have been noted within the ixgbe
driver source. Fix these wholesale and prevent future issues by simply
using BIT macro instead of hand coded bit shifts.

Also fix a few shifts that are shifting values into place by using the
'u' prefix to indicate unsigned. It doesn't strictly matter in these
cases because we're not shifting by too large a value, but these are all
unsigned values and should be indicated as such.
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

It is possible on some systems that crosstalk could lead to link flap
on empty SFP+ cages.  A new NVM bit was defined to let SW know it
needs to implement the work around which consists of verifying that
there is a module in the cage before acting on the LSC.
Signed-off-by: NDon Skidmore <donald.c.skidmore@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Somehow the wrong fc_setup function was used for x550em_a, so
correct that. Also set setup_link to NULL as its value is
determined later, just like it is with X550EM_x.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Implement per-queue statistics for packets, bytes and busy poll
specific counters.
Signed-off-by: NEmil Tantilov <emil.s.tantilov@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This brings the logic closer to how we handle the stats in ixgbe and it
sets us up for introducing per-queue stats.

Use IXGBEVF_STAT and IXGBEVF_NETDEV_STAT for accessing the driver and
netdev stats respectively. This way we don't have to calculate the
stats based on register values which could lead to the counters not
being initialized properly when the interface is down.

IXGBEVF_QUEUE_STATS_LEN is set to include the number of queues.

Also some defines were renamed to use the IXGBEVF prefix.
Signed-off-by: NEmil Tantilov <emil.s.tantilov@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Use a new register to wait for previous register writes to complete
before issuing a register read. This is needed when slower links
are in use.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

RNDIS_STATUS_NETWORK_CHANGE event is handled as two "half events" --
media disconnect & connect. The second half should be added to the list
head, not to the tail. So all events are processed in normal order.
Signed-off-by: NHaiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: NK. Y. Srinivasan <kys@microsoft.com>
Reviewed-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>