Nikolay Aleksandrov says:
====================
bridge: improve cache utilization
This is the first set which begins to deal with the bad bridge cache
access patterns. The first patch rearranges the bridge and port structs
a little so the frequently (and closely) accessed members are in the same
cache line. The second patch then moves the garbage collection to a
workqueue trying to improve system responsiveness under load (many fdbs)
and more importantly removes the need to check if the matched entry is
expired in __br_fdb_get which was a major source of false-sharing.
The third patch is a preparation for the final one which
If properly configured, i.e. ports bound to CPUs (thus updating "updated"
locally) then the bridge's HitM goes from 100% to 0%, but even without
binding we get a win because previously every lookup that iterated over
the hash chain caused false-sharing due to the first cache line being
used for both mac/vid and used/updated fields.
Some results from tests I've run:
(note that these were run in good conditions for the baseline, everything
ran on a single NUMA node and there were only 3 fdbs)
1. baseline
100% Load HitM on the fdbs (between everyone who has done lookups and hit
one of the 3 hash chains of the communicating
src/dst fdbs)
Overall 5.06% Load HitM for the bridge, first place in the list
2. patched & ports bound to CPUs
0% Local load HitM, bridge is not even in the c2c report list
Also there's 3% consistent improvement in netperf tests.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Writing once per jiffy is enough to limit the bridge's false sharing.
After this change the bridge doesn't show up in the local load HitM stats.
Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fdb's used and updated fields are written to on every packet forward and
packet receive respectively. Thus if we are receiving packets from a
particular fdb, they'll cause false-sharing with everyone who has looked
it up (even if it didn't match, since mac/vid share cache line!). The
"used" field is even worse since it is updated on every packet forward
to that fdb, thus the standard config where X ports use a single gateway
results in 100% fdb false-sharing. Note that this patch does not prevent
the last scenario, but it makes it better for other bridge participants
which are not using that fdb (and are only doing lookups over it).
The point is with this move we make sure that only communicating parties
get the false-sharing, in a later patch we'll show how to avoid that too.
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the fdb garbage collector to a workqueue which fires at least 10
milliseconds apart and cleans chain by chain allowing for other tasks
to run in the meantime. When having thousands of fdbs the system is much
more responsive. Most importantly remove the need to check if the
matched entry has expired in __br_fdb_get that causes false-sharing and
is completely unnecessary if we cleanup entries, at worst we'll get 10ms
of traffic for that entry before it gets deleted.
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move around net_bridge so the vlan fields are in the beginning since
they're checked on every packet even if vlan filtering is disabled.
For the port move flags & vlan group to the beginning, so they're in the
same cache line with the port's state (both flags and state are checked
on each packet).
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The patch fixes the case when adding a zero value to the packet
pointer. The zero value could come from src_reg equals type
BPF_K or CONST_IMM. The patch fixes both, otherwise the verifer
reports the following error:
[...]
R0=imm0,min_value=0,max_value=0
R1=pkt(id=0,off=0,r=4)
R2=pkt_end R3=fp-12
R4=imm4,min_value=4,max_value=4
R5=pkt(id=0,off=4,r=4)
269: (bf) r2 = r0 // r2 becomes imm0
270: (77) r2 >>= 3
271: (bf) r4 = r1 // r4 becomes pkt ptr
272: (0f) r4 += r2 // r4 += 0
addition of negative constant to packet pointer is not allowed
Signed-off-by: William Tu <u9012063@gmail.com>
Signed-off-by: Mihai Budiu <mbudiu@vmware.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
These two tests are based on the work done for f23cc643f9. The first test is
just a basic one to make sure we don't allow AND'ing negative values, even if it
would result in a valid index for the array. The second is a cleaned up version
of the original testcase provided by Jann Horn that resulted in the commit.
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vivien Didelot says:
====================
net: dsa: add fabric notifier
When a switch fabric is composed of multiple switch chips, these chips
must be programmed accordingly when an event occurred on one of them.
Examples of such event include hardware bridging: when a Linux bridge
spans interconnected chips, they must be programmed to allow external
ports to ingress frames on their internal ports.
Another example is cross-chip hardware VLANs. Switch chips in-between
interconnected bridge ports must also configure a given VLAN to allow
packets to pass through them.
In order to support that, this patchset introduces a non-intrusive
notifier mechanism. It adds a notifier head in every DSA switch tree
(the said fabric), and a notifier block in every DSA switch chip.
When an even occurs, it is chained to all notifiers of the fabric.
Switch chips can react accordingly if they are cross-chip capable.
On a dynamic debug enabled system, bridging a port in a multi-chip
fabric will print something like this (ZII Rev B board):
# brctl addif br0 lan3
mv88e6085 0.1:00: crosschip DSA port 1.0 bridged to br0
mv88e6085 0.4:00: crosschip DSA port 1.0 bridged to br0
# brctl delif br0 lan3
mv88e6085 0.1:00: crosschip DSA port 1.0 unbridged from br0
mv88e6085 0.4:00: crosschip DSA port 1.0 unbridged from br0
Currently only bridging events are added. A patchset introducing support
for cross-chip hardware bridging configuration in mv88e6xxx will follow
right after. Then events for switchdev operations are next on the line.
We should note that non-switchdev events do not support rolling-back
switch-wide operations. We'll have to work on closer integration with
switchdev for that, like introducing new attributes or objects, to
benefit from the prepare and commit phases.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
A slave device will now notify the switch fabric once its port is
bridged or unbridged, instead of calling directly its switch operations.
This code allows propagating cross-chip bridging events in the fabric.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a notifier block per DSA switch, registered against a notifier head
in the switch fabric they belong to.
This infrastructure will allow to propagate fabric-wide events such as
port bridging, VLAN configuration, etc. If a DSA switch driver cares
about cross-chip configuration, such events can be caught.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The scope of the functions inside net/dsa/slave.c must be the slave
net_device pointer. Change to state setter helper accordingly to
simplify callers.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When an error is returned during the bridging of a port in a
NETDEV_CHANGEUPPER event, net/core/dev.c rolls back the operation.
Be consistent and unassign dp->bridge_dev when this happens.
In the meantime, add comments to document this behavior.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Simplify the code handling the slave netdevice notifier call by
providing a dsa_slave_changeupper helper for NETDEV_CHANGEUPPER, and so
on (only this event is supported at the moment.)
Return NOTIFY_DONE when we did not care about an event, and NOTIFY_OK
when we were concerned but no error occurred, as the API suggests.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the netdevice notifier block register code in slave.c and provide
helpers for dsa.c to register and unregister it.
At the same time, check for errors since (un)register_netdevice_notifier
may fail.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit abeffce ("net/mlx5e: Fix a -Wmaybe-uninitialized warning"), I fixed a
gcc warning for the ipv4 offload handling. Now we get the same warning for the
added ipv6 support:
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c:815:40: warning: 'out_dev' may be used uninitialized in this function [-Wmaybe-uninitialized]
We can apply the same workaround here as well.
Fixes: ce99f6b97f ("net/mlx5e: Support SRIOV TC encapsulation offloads for IPv6 tunnels")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes use of is_vlan_dev() function instead of flag
comparison which is exactly done by is_vlan_dev() helper function.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Jon Maxwell <jmaxwell37@gmail.com>
Acked-by: Johannes Thumshirn <jth@kernel.org>
Acked-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a "||" vs "|" typo here so we test 0x1 instead of 0x6.
Fixes: 1f8176f735 ("net/mlx4_en: Check the enabling pptx/pprx flags in SET_PORT wrapper flow")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit 364b605573 ("net: busy-poll: return busypolling status
to drivers"), napi_complete_done() returns a boolean that can be used
by drivers to conditionally rearm interrupts.
Testing with a 7142 shows a small latency improvement of ~100 ns.
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is to check if asoc->peer.prsctp_capable is set before
processing fwd tsn chunk, if not, it will return an ERROR to the
peer, just as rfc3758 section 3.3.1 demands.
Reported-by: Julian Cordes <julian.cordes@gmail.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jiri Pirko says:
====================
mlxsw: cleanup neigh handling
Ido says:
This series addresses long standing issues in the mlxsw driver
concerning neighbour reflection. It also prepares the code for follow-up
changes dealing with proper resource cleanup and nexthop reflection.
The first two patches convert the neighbour reflection code to use an
ordered workqueue, to prevent re-ordering of NEIGH_UPDATE events that
may happen following subsequent patches.
The third to fifth patches remove the ndo_neigh_{construct,destroy}
entry points from the driver, thereby relying only on NEIGH_UPDATE
events for neighbour reflection. This simplifies the code considerably.
Last patches are fallout and adjust nits in the code I noticed while
going over it.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We periodically ask the neighbouring system to try and resolve
neighbours that are used for nexthops, but aren't currently resolved.
However, 'nud_state' is protected by the neighbour lock, so we shouldn't
access it without taking it. Instead, we can simply check the
'connected' field of the neighbour entry, which we update upon
NEIGH_UPDATE events.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We only add neighbour entries that are also used for nexthops to
'nexthop_neighs_list', so when iterating over this list there's no need
to check that the entry is indeed used for nexthops.
Remove the redundant check.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 18bfb924f0 ("net: introduce default neigh_construct/destroy
ndo calls for L2 upper devices") we added these ndos to stacked devices
such as team and bond, so that calls will be propagated to mlxsw.
However, previous commit removed the reliance on these ndos and no new
users of these ndos have appeared since above mentioned commit. We can
therefore safely remove this dead code.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Up until now we had two interfaces for neighbour related configuration:
ndo_neigh_{construct,destroy} and NEIGH_UPDATE netevents. The ndos were
used to add and remove neighbours from the driver's cache, whereas the
netevent was used to reflect the neighbours into the device's tables.
However, if the NUD state of a neighbour isn't NUD_VALID or if the
neighbour is dead, then there's really no reason for us to keep it
inside our cache. The only exception to this rule are neighbours that
are also used for nexthops, which we periodically refresh to get them
resolved.
We can therefore eliminate the ndo entry point into the driver and
simplify the code, making it similar to the FIB reflection, which is
based solely on events. This also helps us avoid a locking issue, in
which the RIF cache was traversed without proper locking during
insertion into the neigh entry cache.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit 33b1341cd1 ("mlxsw: spectrum_router: Fix handling of
neighbour structure") we no longer use destination IP for neighbour
lookup, so remove it.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We currently associate each neighbour entry with a work item, so it's
not possible to have multiple events queued for the same neighbour
entry. However, this is about to be changed so that the neighbour entry
is only resolved when the work item is scheduled.
The above can result in a mismatch between the kernel's and the device's
neighbour table, unless the associated work items are processed in the
order in which they were submitted.
Do that by migrating the NEIGH_UPDATE work items to be processed in the
ordered workqueue which was recently introduced in mlxsw in commit
a3832b3189 ("mlxsw: core: Create an ordered workqueue for FIB
offload").
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We always use zero delay before queueing a work on the ordered workqueue
('mlxsw_owq'), so use work_struct directly instead of delayable work.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
-----BEGIN PGP SIGNATURE-----
iQFHBAABCgAxFiEE4bay/IylYqM/npjQHv7KIOw4HPYFAliYiW8THG1rbEBwZW5n
dXRyb25peC5kZQAKCRAe/sog7Dgc9jdWCACU/2SxD8WcIcP5HZgJMxv9K9LyRC++
qQb1SQKUiedAF0173IQAVawe105LEdZS08o8ovNCzcU0gLHmfAKysjXEnWVqpyLA
9zOBkl5XzjFsi9KI1TVsn/pUlk3yCq8w2azmt3mi/G5yLoAE0G46pzjbpOYyz67d
Gyoh4i9OhsQOmgpefkwtyPxlh5Xd28ohdPv8NCBb17No+1Dfs4qji1TTyVY9sqNr
8AwRl9BmTqDvZMUPRsdXk+XqZyITGLt+kMa80hd+XFmNkoWy6EKMbFiuVjkpZqtL
kX/Tfe5/mEiBegBvIUVkeOrA7MryNLOf0ZO+bLdNnSTJZJyIro1kAyIc
=kgvC
-----END PGP SIGNATURE-----
Merge tag 'linux-can-next-for-4.11-20170206' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next
Marc Kleine-Budde says:
====================
pull-request: can-next 2017-02-06
this is a pull request of 16 patches for net-next/master.
The first two patches by David Jander and me add the rx-offload
framework for CAN devices to the kernel. The remaining 14 patches
convert the flexcan driver to make use of it.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
HTGT register length is limited to 32 bytes and not 256 bytes.
Signed-off-by: Elad Raz <eladr@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mvneta itself does not support WOL, but the PHY might.
So pass the calls to the PHY
Signed-off-by: Jingju Hou <houjingj@marvell.com>
Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch switches the imx6 and vf610 based SoCs from the hardware FIFO
to the timestamp based rx offloading.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
The flexcan IP core has 64 mailboxes. For now they are configured for
RX as a hardware FIFO. This FIFO has a fixed depth of 6 CAN frames. In
some high load scenarios it turns out thas this buffer is too small.
In order to have a buffer larger than the 6 frames FIFO, this patch adds
support for timestamp based offloading via the generic rx-offload
infrastructure.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
In order to receive RTR frames in the non HW FIFO mode the RSS and EACEN bits
of the reg_ctrl2 have to be activated. As this has no side effect in the FIFO
mode, we do this unconditionally on cores with the reg_ctrl2.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Modern flexcan IP cores support two RX modes. One is using the 6 fames deep
hardware FIFO, the other is using up to 64 mailboxes (in non FIFO mode). For
now only the HW FIFO mode is activated.
In order to make use of the RX mailboxes the individual RX masking feature has
to be activated, otherwise matching mailboxes are overwritten during the
reception process. This however switches on the individual RX masking, which
uses reg_rximr registers for masking.
This patch activates the individual RX masking feature unconditionally and
initializes the mask registers (reg_rximr) with 0x0 == "don't care", which
switches off any filtering.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
This patch converts the flexcan driver to make use of the rx-offload
can_rx_offload_irq_offload_fifo() helper function. The idea is to read
the CAN frames already in the interrupt context, as the depth of the
flexcan HW FIFO is too shallow, resulting in too many missed frames.
During a normal NAPI poll the frames are the pushed into the upper
layers.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
This patch makes the TX mailbox selectable duing runtime. This is a preparation
patch to use of the hardware FIFO selectable via runtime. As the TX mailbox
number is different in HW FIFO and normal mode.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
This patch converts the define FLEXCAN_IFLAG_DEFAULT into the runtime
calculated value priv->reg_imask1_default. This is a preparation patch to make
the TX mailbox selectable during runtime, too.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
This patch changes the flexcan_irq() function to only return
IRQ_HANDLED, if the interrupt really has been handled, otherwise
IRQ_NONE is returned.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
This patch removed the not needed initialisation from the new_state,
rx_state, tx_state variabled.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
This patch converts the rx_errors and tx_errors from int into bool
values, to reflect their actual meaning.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Some CAN controllers don't implement a FIFO in hardware, but fill their
mailboxes in a particular order (from lowest to highest or highest to lowest).
This makes problems to read the frames in the correct order from the hardware,
as new frames might be filled into just read (low) mailboxes. This gets worse,
when following new frames are received into not read (higher) mailboxes.
On the bright side some these CAN controllers put a timestamp on each received
CAN frame. This patch adds support to offload CAN frames in interrupt context,
order them by timestamp and then transmitted in a NAPI context.
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Some CAN controllers have a usable FIFO already but can still benefit
from off-loading the CAN controller FIFO. The CAN frames of the FIFO are
read and put into a skb queue during interrupt and then transmitted in a
NAPI context.
Signed-off-by: David Jander <david@protonic.nl>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Eric Dumazet says:
====================
net: get rid of __napi_complete()
This patch series removes __napi_complete() calls, in an effort
to make NAPI API simpler and generalize GRO and napi_complete_done()
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
All __napi_complete() callers have been converted to
use the more standard napi_complete_done(),
we can now remove this NAPI method for good.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We plan to remove __napi_complete() soon,
this driver is the last user.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller