skb_segment by default transfers allocated wmem from the gso skb
to the tail of the segment list. This underreports real truesize
of the list, especially if the tail might be dropped.
Similar to tcp_gso_segment, update wmem_alloc with the aggregate
list truesize and make each segment responsible for its own
share by setting skb->destructor.
Clear gso_skb->destructor prior to calling skb_segment to skip
the default assignment to tail.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Support generic segmentation offload for udp datagrams. Callers can
concatenate and send at once the payload of multiple datagrams with
the same destination.
To set segment size, the caller sets socket option UDP_SEGMENT to the
length of each discrete payload. This value must be smaller than or
equal to the relevant MTU.
A follow-up patch adds cmsg UDP_SEGMENT to specify segment size on a
per send call basis.
Total byte length may then exceed MTU. If not an exact multiple of
segment size, the last segment will be shorter.
The implementation adds a gso_size field to the udp socket, ip(v6)
cmsg cookie and inet_cork structure to be able to set the value at
setsockopt or cmsg time and to work with both lockless and corked
paths.
Initial benchmark numbers show UDP GSO about as expensive as TCP GSO.
tcp tso
3197 MB/s 54232 msg/s 54232 calls/s
6,457,754,262 cycles
tcp gso
1765 MB/s 29939 msg/s 29939 calls/s
11,203,021,806 cycles
tcp without tso/gso *
739 MB/s 12548 msg/s 12548 calls/s
11,205,483,630 cycles
udp
876 MB/s 14873 msg/s 624666 calls/s
11,205,777,429 cycles
udp gso
2139 MB/s 36282 msg/s 36282 calls/s
11,204,374,561 cycles
[*] after reverting commit 0a6b2a1dc2
("tcp: switch to GSO being always on")
Measured total system cycles ('-a') for one core while pinning both
the network receive path and benchmark process to that core:
perf stat -a -C 12 -e cycles \
./udpgso_bench_tx -C 12 -4 -D "$DST" -l 4
Note the reduction in calls/s with GSO. Bytes per syscall drops
increases from 1470 to 61818.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement generic segmentation offload support for udp datagrams. A
follow-up patch adds support to the protocol stack to generate such
packets.
UDP GSO is not UFO. UFO fragments a single large datagram. GSO splits
a large payload into a number of discrete UDP datagrams.
The implementation adds a GSO type SKB_UDP_GSO_L4 to differentiate it
from UFO (SKB_UDP_GSO).
IPPROTO_UDPLITE is excluded, as that protocol has no gso handler
registered.
[ Export __udp_gso_segment for ipv6. -DaveM ]
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
UDP segmentation offload needs access to inet_cork in the udp layer.
Pass the struct to ip(6)_make_skb instead of allocating it on the
stack in that function itself.
This patch is a noop otherwise.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
1GbE Intel Wired LAN Driver Updates 2018-04-25
This series enables some ethtool and tc-flower filters to be offloaded
to igb-based network controllers. This is useful when the system
configuration wants to steer kinds of traffic to a specific hardware
queue for i210 devices only.
The first two patch in the series are bug fixes.
The basis of this series is to export the internal API used to
configure address filters, so they can be used by ethtool, and
extending the functionality so an source address can be handled.
Then, we enable the tc-flower offloading implementation to re-use the
same infrastructure as ethtool, and storing them in the per-adapter
"nfc" (Network Filter Config?) list. But for consistency, for
destructive access they are separated, i.e. an filter added by
tc-flower can only be removed by tc-flower, but ethtool can read them
all.
Only support for VLAN Prio, Source and Destination MAC Address, and
Ethertype is enabled for now.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Daniel Borkmann says:
====================
pull-request: bpf 2018-04-25
The following pull-request contains BPF updates for your *net* tree.
The main changes are:
1) Fix to clear the percpu metadata_dst that could otherwise carry
stale ip_tunnel_info, from William.
2) Fix that reduces the number of passes in x64 JIT with regards to
dead code sanitation to avoid risk of prog rejection, from Gianluca.
3) Several fixes of sockmap programs, besides others, fixing a double
page_put() in error path, missing refcount hold for pinned sockmap,
adding required -target bpf for clang in sample Makefile, from John.
4) Fix to disable preemption in __BPF_PROG_RUN_ARRAY() paths, from Roman.
5) Fix tools/bpf/ Makefile with regards to a lex/yacc build error
seen on older gcc-5, from John.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix build error found with Ubuntu shipped gcc-5
~/git/bpf/tools/bpf$ make all
Auto-detecting system features:
... libbfd: [ OFF ]
... disassembler-four-args: [ OFF ]
CC bpf_jit_disasm.o
LINK bpf_jit_disasm
CC bpf_dbg.o
/home/john/git/bpf/tools/bpf/bpf_dbg.c: In function ‘cmd_load’:
/home/john/git/bpf/tools/bpf/bpf_dbg.c:1077:13: warning: ‘cont’ may be used uninitialized in this function [-Wmaybe-uninitialized]
} else if (matches(subcmd, "pcap") == 0) {
^
LINK bpf_dbg
CC bpf_asm.o
make: *** No rule to make target `bpf_exp.yacc.o', needed by `bpf_asm'. Stop.
Fixes: 5a8997f207 ("tools: bpf: respect output directory during build")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Jeff Kirsher says:
====================
10GbE Intel Wired LAN Driver Updates 2018-04-25
This series represents yet another phase of the macvlan cleanup Alex has
been working on.
The main goal of these changes is to make it so that we only support
offloading what we can actually offload and we don't break any existing
functionality. So for example we were claiming to advertise source mode
macvlan and we were doing nothing of the sort, so support for that has been
dropped.
The biggest change with this set is that broadcast/multicast replication is
no longer being supported in software. Alex dropped it as it leads to
scaling issues when a broadcast frame has to be replicated up to 64 times.
Beyond that this set goes through and optimized the time needed to bring up
and tear down the macvlan interfaces on ixgbe and provides a clean way for
us to disable the macvlan offload when needed.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The function rds_ib_setup_qp is calling rds_ib_get_client_data and
should correspondingly call rds_ib_dev_put. This call was lost in
the non-error path with the introduction of error handling done in
commit 3b12f73a5c ("rds: ib: add error handle")
Signed-off-by: Dag Moxnes <dag.moxnes@oracle.com>
Reviewed-by: Håkon Bugge <haakon.bugge@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The internal CLC socket should exist till the SMC-socket is released.
Function tcp_listen_worker() releases the internal CLC socket of a
listen socket, if an smc_close_active() is called. This function
is called for the final release(), but it is called for shutdown
SHUT_RDWR as well. This opens a door for protection faults, if
socket calls using the internal CLC socket are called for a
shutdown listen socket.
With the changes of
commit 3d50206759 ("net/smc: simplify wait when closing listen socket")
there is no need anymore to release the internal CLC socket in
function tcp_listen_worker((). It is sufficient to release it in
smc_release().
Fixes: 127f497058 ("net/smc: release clcsock from tcp_listen_worker")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Reported-by: syzbot+9045fc589fcd196ef522@syzkaller.appspotmail.com
Reported-by: syzbot+28a2c86cf19c81d871fa@syzkaller.appspotmail.com
Reported-by: syzbot+9605e6cace1b5efd4a0a@syzkaller.appspotmail.com
Reported-by: syzbot+cf9012c597c8379d535c@syzkaller.appspotmail.com
Signed-off-by: David S. Miller <davem@davemloft.net>
The check for len > 0 is always true and hence is redundant as
this check is already being made to execute the code inside the
while-loop. Hence it is redundant and can be removed.
Cleans up cppcheck warning:
drivers/net/hamradio/mkiss.c:220: (warning) Identical inner 'if'
condition is always true.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are two identical nested if statements, the second is redundant
and can be removed. Also clean up white space formatting.
Cleans up cppcheck warning:
drivers/net/ethernet/amd/amd8111e.c:1080: (warning) Identical inner 'if'
condition is always true.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After Commit 4f00878126 ("sctp: apply rhashtable api to send/recv
path"), there's no place using sctp_assoc_is_match, so remove it.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This allows filters added by tc-flower and specifying MAC addresses,
Ethernet types, and the VLAN priority field, to be offloaded to the
controller.
This reuses most of the infrastructure used by ethtool, but clsflower
filters are kept in a separated list, so they are invisible to
ethtool.
To setup clsflower offloading:
$ tc qdisc replace dev eth0 handle 100: parent root mqprio \
num_tc 3 map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 \
queues 1@0 1@1 2@2 hw 0
(clsflower offloading depends on the netword driver to be configured
with multiple traffic classes, we use mqprio's 'num_tc' parameter to
set it to 3)
$ tc qdisc add dev eth0 ingress
Examples of filters:
$ tc filter add dev eth0 parent ffff: flower \
dst_mac aa:aa:aa:aa:aa:aa \
hw_tc 2 skip_sw
(just a simple filter filtering for the destination MAC address and
steering that traffic to queue 2)
$ tc filter add dev enp2s0 parent ffff: proto 0x22f0 flower \
src_mac cc:cc:cc:cc:cc:cc \
hw_tc 1 skip_sw
(as the i210 doesn't support steering traffic based on the source
address alone, we need to use another steering traffic, in this case
we are using the ethernet type (0x22f0) to steer traffic to queue 1)
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Jakub Kicinski says:
====================
nfp: flower tc block support and nfp PCI updates
This series improves the nfp PCIe code by making use of the new
pcie_print_link_status() helper and resetting NFP locks when
driver loads. This can help us avoid lock ups after host crashes
and is rebooted with PCIe reset or when kdump kernel is loaded.
The flower changes come from John, he says:
This patchset fixes offload issues when multiple repr netdevs are bound to
a tc block and filter rules added. Previously the rule would be passed to
the reprs and would be rejected in all but the first as the cookie value
will indicate a duplicate. The first patch extends the flow lookup
function to consider both host context and ingress netdev along with the
cookie value. This means that a rule with a given cookie can exist
multiple times assuming the ingress netdev is different. The host context
ensures that stats from fw are associated with the correct instance of the
rule.
The second patch protects against rejecting add/del/stat messages when a
rule has a repr as both an ingress port and an egress dev. In such cases a
callback can be triggered twice (once for ingress and once for egress)
and can lead to duplicate rule detection or incorrect double calls.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
If a flower rule has a repr both as ingress and egress port then 2
callbacks may be generated for the same rule request.
Add an indicator to each flow as to whether or not it was added from an
ingress registered cb. If so then ignore add/del/stat requests to it from
an egress cb.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When multiple netdevs are attached to a tc offload block and register for
callbacks, a rule added to the block will be propogated to all netdevs.
Previously these were detected as duplicates (based on cookie) and
rejected. Modify the rule nfp lookup function to optionally include an
ingress netdev and a host context along with the cookie value when
searching for a rule. When a new rule is passed to the driver, the netdev
the rule is to be attached to is considered when searching for dublicates.
When a stats update is received from HW, the host context is used
alongside the cookie to map to the correct host rule.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
To aid debugging of performance issues caused by limited PCIe
bandwidth print the PCIe link information on probe.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
NFP locks record the owner when held, for PCIe devices the owner
ID will be the PCIe link number. When driver loads it should scan
known locks and if they indicate that they are held by local
endpoint but the driver doesn't hold them - release them.
Locks can be left taken for instance when kernel gets kexec-ed or
after a crash. Management FW tries to clean up stale locks too,
but it currently depends on PCIe link going down which doesn't
always happen.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds basic functions needed to implement offloading for filters
created by tc-flower.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This adds the capability of configuring the queue steering of arriving
packets based on their source and destination MAC addresses.
Source address steering (i.e. driving traffic to a specific queue),
for the i210, does not work, but filtering does (i.e. accepting
traffic based on the source address). So, trying to add a filter
specifying only a source address will be an error.
In practical terms this adds support for the following use cases,
characterized by these examples:
$ ethtool -N eth0 flow-type ether dst aa:aa:aa:aa:aa:aa action 0
(this will direct packets with destination address "aa:aa:aa:aa:aa:aa"
to the RX queue 0)
$ ethtool -N eth0 flow-type ether src 44:44:44:44:44:44 \
proto 0x22f0 action 3
(this will direct packets with source address "44:44:44:44:44:44" and
ethertype 0x22f0 to the RX queue 3)
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This allows igb_add_filter()/igb_erase_filter() to work on filters
that include MAC addresses (both source and destination).
For now, this only exposes the functionality, the next commit glues
ethtool into this. Later in this series, these APIs are used to allow
offloading of cls_flower filters.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Users expect that when adding a steering filter for the local MAC
address, that all the traffic directed to that address will go to some
queue.
Currently, it's not possible to configure entries in the "in use"
state, which is the normal state of the local MAC address entry (it is
the default), this patch allows to override the steering configuration
of "in use" entries, if the filter to be added match the address and
address type (source or destination) of an existing entry.
There is a bit of a special handling for entries referring to the
local MAC address, when they are removed, only the steering
configuration is reset.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
On some igb models (82575 and i210) the MAC address filters can
control to which queue the packet will be assigned.
This extends the 'state' with one more state to signify that queue
selection should be enabled for that filter.
As 82575 parts are no longer easily obtained (and this was developed
against i210), only support for the i210 model is enabled.
These functions are exported and will be used in the next patch.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Makes it possible to direct packets to queues based on their source
address. Documents the expected usage of the 'flags' parameter.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This will allow functionality depending on the hardware being traffic
class aware to work. In particular the tc-flower offloading checks
verifies that this bit is set.
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
On the RAH registers there are semantic differences on the meaning of
the "queue" parameter for traffic steering depending on the controller
model: there is the 82575 meaning, which "queue" means a RX Hardware
Queue, and the i350 meaning, where it is a reception pool.
The previous behaviour was having no effect for i210 based controllers
because the QSEL bit of the RAH register wasn't being set.
This patch separates the condition in discrete cases, so the different
handling is clearer.
Fixes: 83c21335c8 ("igb: improve MAC filter handling")
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Move the check on FRA_L3MDEV attribute to helper to improve the
readability of fib_nl2rule. Update the extack messages to be
clear when the configuration option is disabled versus an invalid
value has been passed.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's currently written as:
if (!tchunk->tsn_gap_acked) { [1]
tchunk->tsn_gap_acked = 1;
...
}
if (TSN_lte(tsn, sack_ctsn)) {
if (!tchunk->tsn_gap_acked) {
/* SFR-CACC processing */
...
}
}
Which causes the SFR-CACC processing on ack reception to never process,
as tchunk->tsn_gap_acked is always true by then. Block [1] was
moved to that position by the commit marked below.
This patch fixes it by doing SFR-CACC processing earlier, before
tsn_gap_acked is set to true.
Fixes: 31b02e1549 ("sctp: Failover transmitted list on transport delete")
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sctp_make_sack() make changes to the asoc and this cast is just
bypassing the const attribute. As there is no need to have the const
there, just remove it and fix the violation.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch extends NTF_EXT_LEARNED support to the neighbour system.
Example use-case: An Ethernet VPN implementation (eg in FRR routing suite)
can use this flag to add dynamic reachable external neigh entires
learned via control plane. The use of neigh NTF_EXT_LEARNED in this
patch is consistent with its use with bridge and vxlan fdb entries.
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The addrconf_ifdown() evaluates keep_addr_on_down state twice. There
is no need to do it.
Cc: David Ahern <dsahern@gmail.com>
Signed-off-by: Ivan Vecera <cera@cera.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
ECMP (equal-cost multipath) hashes are typically computed on the packets'
5-tuple(src IP, dst IP, src port, dst port, L4 proto).
For encapsulated packets, the L4 data is not readily available and ECMP
hashing will often revert to (src IP, dst IP). This will lead to traffic
polarization on a single ECMP path, causing congestion and waste of network
capacity.
In IPv6, the 20-bit flow label field is also used as part of the ECMP hash.
In the lack of L4 data, the hashing will be on (src IP, dst IP, flow
label). Having a non-zero flow label is thus important for proper traffic
load balancing when L4 data is unavailable (i.e., when packets are
encapsulated).
Currently, the seg6_do_srh_encap() function extracts the original packet's
flow label and set it as the outer IPv6 flow label. There are two issues
with this behaviour:
a) There is no guarantee that the inner flow label is set by the source.
b) If the original packet is not IPv6, the flow label will be set to
zero (e.g., IPv4 or L2 encap).
This patch adds a function, named seg6_make_flowlabel(), that computes a
flow label from a given skb. It supports IPv6, IPv4 and L2 payloads, and
leverages the per namespace 'seg6_flowlabel" sysctl value.
The currently support behaviours are as follows:
-1 set flowlabel to zero.
0 copy flowlabel from Inner paceket in case of Inner IPv6
(Set flowlabel to 0 in case IPv4/L2)
1 Compute the flowlabel using seg6_make_flowlabel()
This patch has been tested for IPv6, IPv4, and L2 traffic.
Signed-off-by: Ahmed Abdelsalam <amsalam20@gmail.com>
Acked-by: David Lebrun <dlebrun@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some MDIO busses will error out when trying to read a phy address with no
phy present at that address. In that case, probing the bus will fail
because __mdiobus_register() is scanning the bus for all possible phys
addresses.
In case MII_PHYSID1 returns -EIO or -ENODEV, consider there is no phy at
this address and set the phy ID to 0xffffffff which is then properly
handled in get_phy_device().
Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Because the order of the parameters passes to 'hlist_add_behind()' was
inverted, the 'parent' node was added "behind" the 'input', as input
is not in the list, this causes the 'input' node to be lost.
Fixes: 0e71def252 ("igb: add support of RX network flow classification")
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The original implementation for macvlan offload has us performing a full
port reset every time we added a new macvlan. This shouldn't be necessary
and can be avoided with a few behavior changes.
This patches updates the logic for the queues so that we have essentially 3
possible configurations for macvlan offload. They consist of 15 macvlans
with 4 queues per macvlan, 31 macvlans with 2 queues per macvlan, and 63
macvlans with 1 queue per macvlan. As macvlans are added you will encounter
up to 3 total resets if you add all the way up to 63, and after that the
device will stay in the mode supporting up to 63 macvlans until the L2FW
flag is cleared.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch drops the real_adapter member from the fwd_adapter structure.
The general idea behind the change is that the real_adapter is carrying
unnecessary data since we could always just grab the adapter structure
from netdev_priv(macvlan->lowerdev) if we really needed to get at it.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Both the ixgbe and fm10k drivers support destination filtering.
Instead of adding a ton of complexity to support either source or passthru
mode we can instead just avoid offloading them for now. Doing this we avoid
leaking packets into interfaces that aren't meant to receive them.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch provides a basic function to allow a lower device to disable
macvlan offload if it was previously enabled on a given macvlan. The idea
here is to allow for recovery from failure should the lowerdev run out of
resources.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds a function indicating if a given macvlan can fully supports
destination filtering, especially as it relates to unicast traffic. For
those macvlan interfaces that do not support destination filtering such
passthru or source mode filtering we should not be enabling offload
support.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
It doesn't make sense to define macvlan_count_rx as a static inline and
then add a forward declaration after that as an extern. I am dropping the
extern declaration since it seems like it is something that likely got
missed when the function was made an inline.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Drop dead code now that we shouldn't be receiving broadcast or multicast
frames on the queues associated to the macvlan netdev.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This change makes it so that we use a software path for packets that are
going to be locally switched between two macvlan interfaces on the same
device. In addition we resort to software replication of broadcast and
multicast packets instead of offloading that to hardware.
The general idea is that using the device for east/west traffic local to
the system is extremely inefficient. We can only support up to whatever the
PCIe limit is for any given device so this caps us at somewhere around 20G
for devices supported by ixgbe. This is compounded even further when you
take broadcast and multicast into account as a single 10G port can come to
a crawl as a packet is replicated up to 60+ times in some cases. In order
to get away from that I am implementing changes so that we handle
broadcast/multicast replication and east/west local traffic all in
software.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This change renames the fwd_priv member to accel_priv as this more
accurately reflects the actual purpose of this value. In addition I am
adding an accessor which will allow us to further abstract this in the
future if needed.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Drop the code for handling macvlan specific unicast lists. It isn't needed
since we don't take any efforts to maintain it when we bring the interface
up and it takes the slow path anyway.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Commit 2a5418a13f ("bpf: improve dead code sanitizing") replaced dead
code with a series of ja-1 instructions, for safety. That made JIT
compilation much more complex for some BPF programs. One instance of such
programs is, for example:
bool flag = false
...
/* A bunch of other code */
...
if (flag)
do_something()
In some cases llvm is not able to remove at compile time the code for
do_something(), so the generated BPF program ends up with a large amount
of dead instructions. In one specific real life example, there are two
series of ~500 and ~1000 dead instructions in the program. When the
verifier replaces them with a series of ja-1 instructions, it causes an
interesting behavior at JIT time.
During the first pass, since all the instructions are estimated at 64
bytes, the ja-1 instructions end up being translated as 5 bytes JMP
instructions (0xE9), since the jump offsets become increasingly large (>
127) as each instruction gets discovered to be 5 bytes instead of the
estimated 64.
Starting from the second pass, the first N instructions of the ja-1
sequence get translated into 2 bytes JMPs (0xEB) because the jump offsets
become <= 127 this time. In particular, N is defined as roughly 127 / (5
- 2) ~= 42. So, each further pass will make the subsequent N JMP
instructions shrink from 5 to 2 bytes, making the image shrink every time.
This means that in order to have the entire program converge, there need
to be, in the real example above, at least ~1000 / 42 ~= 24 passes just
for translating the dead code. If we add this number to the passes needed
to translate the other non dead code, it brings such program to 40+
passes, and JIT doesn't complete. Ultimately the userspace loader fails
because such BPF program was supposed to be part of a prog array owner
being JITed.
While it is certainly possible to try to refactor such programs to help
the compiler remove dead code, the behavior is not really intuitive and it
puts further burden on the BPF developer who is not expecting such
behavior. To make things worse, such programs are working just fine in all
the kernel releases prior to the ja-1 fix.
A possible approach to mitigate this behavior consists into noticing that
for ja-1 instructions we don't really need to rely on the estimated size
of the previous and current instructions, we know that a -1 BPF jump
offset can be safely translated into a 0xEB instruction with a jump offset
of -2.
Such fix brings the BPF program in the previous example to complete again
in ~9 passes.
Fixes: 2a5418a13f ("bpf: improve dead code sanitizing")
Signed-off-by: Gianluca Borello <g.borello@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Jakub Kicinski says:
====================
This set adds an optimization run to the NFP jit to turn ADD and SUB
instructions with negative immediate into the opposite operation with
a positive immediate. NFP can fit small immediates into the instructions
but it can't ever fit negative immediates. Addition of small negative
immediates is quite common in BPF programs for stack address calculations,
therefore this optimization gives us non-negligible savings in instruction
count (up to 4%).
====================
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Comparison instruction requires a subtraction. If the constant
is negative we are more likely to fit it into a NFP instruction
directly if we change the sign and use addition.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
There are quite a few compare instructions now, use a table
to translate BPF instruction code to NFP instruction parameters
instead of parameterizing helpers. This saves LOC and makes
future extensions easier.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
