When calling the flow_free() to free the flow, we call many times
(cpu_possible_mask, eg. 128 as default) cpumask_next(). That will
take up our CPU usage if we call the flow_free() frequently.
When we put all packets to userspace via upcall, and OvS will send
them back via netlink to ovs_packet_cmd_execute(will call flow_free).
The test topo is shown as below. VM01 sends TCP packets to VM02,
and OvS forward packtets. When testing, we use perf to report the
system performance.
VM01 --- OvS-VM --- VM02
Without this patch, perf-top show as below: The flow_free() is
3.02% CPU usage.
4.23% [kernel] [k] _raw_spin_unlock_irqrestore
3.62% [kernel] [k] __do_softirq
3.16% [kernel] [k] __memcpy
3.02% [kernel] [k] flow_free
2.42% libc-2.17.so [.] __memcpy_ssse3_back
2.18% [kernel] [k] copy_user_generic_unrolled
2.17% [kernel] [k] find_next_bit
When applied this patch, perf-top show as below: Not shown on
the list anymore.
4.11% [kernel] [k] _raw_spin_unlock_irqrestore
3.79% [kernel] [k] __do_softirq
3.46% [kernel] [k] __memcpy
2.73% libc-2.17.so [.] __memcpy_ssse3_back
2.25% [kernel] [k] copy_user_generic_unrolled
1.89% libc-2.17.so [.] _int_malloc
1.53% ovs-vswitchd [.] xlate_actions
With this patch, the TCP throughput(we dont use Megaflow Cache
+ Microflow Cache) between VMs is 1.18Gbs/sec up to 1.30Gbs/sec
(maybe ~10% performance imporve).
This patch adds cpumask struct, the cpu_used_mask stores the cpu_id
that the flow used. And we only check the flow_stats on the cpu we
used, and it is unncessary to check all possible cpu when getting,
cleaning, and updating the flow_stats. Adding the cpu_used_mask to
sw_flow struct does’t increase the cacheline number.
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using flow stats per NUMA node, use it per CPU. When using
megaflows, the stats lock can be a bottleneck in scalability.
On a E5-2690 12-core system, usual throughput went from ~4Mpps to
~15Mpps when forwarding between two 40GbE ports with a single flow
configured on the datapath.
This has been tested on a system with possible CPUs 0-7,16-23. After
module removal, there were no corruption on the slab cache.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@redhat.com>
Cc: pravin shelar <pshelar@ovn.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
On a system with only node 1 as possible, all statistics is going to be
accounted on node 0 as it will have a single writer.
However, when getting and clearing the statistics, node 0 is not going
to be considered, as it's not a possible node.
Tested that statistics are not zero on a system with only node 1
possible. Also compile-tested with CONFIG_NUMA off.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/usb/asix_common.c
net/ipv4/inet_connection_sock.c
net/switchdev/switchdev.c
In the inet_connection_sock.c case the request socket hashing scheme
is completely different in net-next.
The other two conflicts were overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Store tunnel protocol (AF_INET or AF_INET6) in sw_flow_key. This field now
also acts as an indicator whether the flow contains tunnel data (this was
previously indicated by tun_key.u.ipv4.dst being set but with IPv6 addresses
in an union with IPv4 ones this won't work anymore).
The new field was added to a hole in sw_flow_key.
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
When openvswitch tries allocate memory from offline numa node 0:
stats = kmem_cache_alloc_node(flow_stats_cache, GFP_KERNEL | __GFP_ZERO, 0)
It catches VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid))
[ replaced with VM_WARN_ON(!node_online(nid)) recently ] in linux/gfp.h
This patch disables numa affinity in this case.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When support for megaflows was introduced, OVS needed to start
installing flows with a mask applied to them. Since masking is an
expensive operation, OVS also had an optimization that would only
take the parts of the flow keys that were covered by a non-zero
mask. The values stored in the remaining pieces should not matter
because they are masked out.
While this works fine for the purposes of matching (which must always
look at the mask), serialization to netlink can be problematic. Since
the flow and the mask are serialized separately, the uninitialized
portions of the flow can be encoded with whatever values happen to be
present.
In terms of functionality, this has little effect since these fields
will be masked out by definition. However, it leaks kernel memory to
userspace, which is a potential security vulnerability. It is also
possible that other code paths could look at the masked key and get
uninitialized data, although this does not currently appear to be an
issue in practice.
This removes the mask optimization for flows that are being installed.
This was always intended to be the case as the mask optimizations were
really targetting per-packet flow operations.
Fixes: 03f0d916 ("openvswitch: Mega flow implementation")
Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the IPv6 addresses as an union with IPv4 ones. When using IPv4, the
newly introduced padding after the IPv4 addresses needs to be zeroed out.
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
net/bridge/br_mdb.c
br_mdb.c conflict was a function call being removed to fix a bug in
'net' but whose signature was changed in 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
Some architectures like POWER can have a NUMA node_possible_map that
contains sparse entries. This causes memory corruption with openvswitch
since it allocates flow_cache with a multiple of num_possible_nodes() and
assumes the node variable returned by for_each_node will index into
flow->stats[node].
Use nr_node_ids to allocate a maximal sparse array instead of
num_possible_nodes().
The crash was noticed after 3af229f2 was applied as it changed the
node_possible_map to match node_online_map on boot.
Fixes: 3af229f207
Signed-off-by: Chris J Arges <chris.j.arges@canonical.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Utilize the new metadata dst to attach encapsulation instructions to
the skb. The existing egress_tun_info via the OVS_CB() is left in
place until all tunnel vports have been converted to the new method.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously, flows were manipulated by userspace specifying a full,
unmasked flow key. This adds significant burden onto flow
serialization/deserialization, particularly when dumping flows.
This patch adds an alternative way to refer to flows using a
variable-length "unique flow identifier" (UFID). At flow setup time,
userspace may specify a UFID for a flow, which is stored with the flow
and inserted into a separate table for lookup, in addition to the
standard flow table. Flows created using a UFID must be fetched or
deleted using the UFID.
All flow dump operations may now be made more terse with OVS_UFID_F_*
flags. For example, the OVS_UFID_F_OMIT_KEY flag allows responses to
omit the flow key from a datapath operation if the flow has a
corresponding UFID. This significantly reduces the time spent assembling
and transacting netlink messages. With all OVS_UFID_F_OMIT_* flags
enabled, the datapath only returns the UFID and statistics for each flow
during flow dump, increasing ovs-vswitchd revalidator performance by 40%
or more.
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rework so that ovs_flow_tbl_insert() calls flow_{key,mask}_insert().
This tidies up a future patch.
Signed-off-by: Joe Stringer <joestringer@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch effectively reverts commit 500f808726 ("net: ovs: use CRC32
accelerated flow hash if available"), and other remaining arch_fast_hash()
users such as from nfsd via commit 6282cd5655 ("NFSD: Don't hand out
delegations for 30 seconds after recalling them.") where it has been used
as a hash function for bloom filtering.
While we think that these users are actually not much of concern, it has
been requested to remove the arch_fast_hash() library bits that arose
from [1] entirely as per recent discussion [2]. The main argument is that
using it as a hash may introduce bias due to its linearity (see avalanche
criterion) and thus makes it less clear (though we tried to document that)
when this security/performance trade-off is actually acceptable for a
general purpose library function.
Lets therefore avoid any further confusion on this matter and remove it to
prevent any future accidental misuse of it. For the time being, this is
going to make hashing of flow keys a bit more expensive in the ovs case,
but future work could reevaluate a different hashing discipline.
[1] https://patchwork.ozlabs.org/patch/299369/
[2] https://patchwork.ozlabs.org/patch/418756/
Cc: Neil Brown <neilb@suse.de>
Cc: Francesco Fusco <fusco@ntop.org>
Cc: Jesse Gross <jesse@nicira.com>
Cc: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ths simplifies flow-table-destroy API. No need to pass explicit
parameter about context.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Acked-by: Thomas Graf <tgraf@redhat.com>
Due to the race condition in userspace, there is chance that two
overlapping megaflows could be installed in datapath. And this
causes userspace unable to delete the less inclusive megaflow flow
even after it timeout, since the flow_del logic will stop at the
first match of masked flow.
This commit fixes the bug by making the kernel flow_del and flow_get
logic check all masks in that case.
Introduced by 03f0d916a (openvswitch: Mega flow implementation).
Signed-off-by: Alex Wang <alexw@nicira.com>
Acked-by: Andy Zhou <azhou@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Incorrect struct name was confusing, even though otherwise
inconsequental.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Masks are inserted when flows are inserted to the table, so it is
logical to correspondingly remove masks when flows are removed from
the table, in ovs_flow_table_remove().
This allows ovs_flow_free() to be called without locking, which will
be used by later patches.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Keep kernel flow stats for each NUMA node rather than each (logical)
CPU. This avoids using the per-CPU allocator and removes most of the
kernel-side OVS locking overhead otherwise on the top of perf reports
and allows OVS to scale better with higher number of threads.
With 9 handlers and 4 revalidators netperf TCP_CRR test flow setup
rate doubles on a server with two hyper-threaded physical CPUs (16
logical cores each) compared to the current OVS master. Tested with
non-trivial flow table with a TCP port match rule forcing all new
connections with unique port numbers to OVS userspace. The IP
addresses are still wildcarded, so the kernel flows are not considered
as exact match 5-tuple flows. This type of flows can be expected to
appear in large numbers as the result of more effective wildcarding
made possible by improvements in OVS userspace flow classifier.
Perf results for this test (master):
Events: 305K cycles
+ 8.43% ovs-vswitchd [kernel.kallsyms] [k] mutex_spin_on_owner
+ 5.64% ovs-vswitchd [kernel.kallsyms] [k] __ticket_spin_lock
+ 4.75% ovs-vswitchd ovs-vswitchd [.] find_match_wc
+ 3.32% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_lock
+ 2.61% ovs-vswitchd [kernel.kallsyms] [k] pcpu_alloc_area
+ 2.19% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask_range
+ 2.03% swapper [kernel.kallsyms] [k] intel_idle
+ 1.84% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_unlock
+ 1.64% ovs-vswitchd ovs-vswitchd [.] classifier_lookup
+ 1.58% ovs-vswitchd libc-2.15.so [.] 0x7f4e6
+ 1.07% ovs-vswitchd [kernel.kallsyms] [k] memset
+ 1.03% netperf [kernel.kallsyms] [k] __ticket_spin_lock
+ 0.92% swapper [kernel.kallsyms] [k] __ticket_spin_lock
...
And after this patch:
Events: 356K cycles
+ 6.85% ovs-vswitchd ovs-vswitchd [.] find_match_wc
+ 4.63% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_lock
+ 3.06% ovs-vswitchd [kernel.kallsyms] [k] __ticket_spin_lock
+ 2.81% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask_range
+ 2.51% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_unlock
+ 2.27% ovs-vswitchd ovs-vswitchd [.] classifier_lookup
+ 1.84% ovs-vswitchd libc-2.15.so [.] 0x15d30f
+ 1.74% ovs-vswitchd [kernel.kallsyms] [k] mutex_spin_on_owner
+ 1.47% swapper [kernel.kallsyms] [k] intel_idle
+ 1.34% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask
+ 1.33% ovs-vswitchd ovs-vswitchd [.] rule_actions_unref
+ 1.16% ovs-vswitchd ovs-vswitchd [.] hindex_node_with_hash
+ 1.16% ovs-vswitchd ovs-vswitchd [.] do_xlate_actions
+ 1.09% ovs-vswitchd ovs-vswitchd [.] ofproto_rule_ref
+ 1.01% netperf [kernel.kallsyms] [k] __ticket_spin_lock
...
There is a small increase in kernel spinlock overhead due to the same
spinlock being shared between multiple cores of the same physical CPU,
but that is barely visible in the netperf TCP_CRR test performance
(maybe ~1% performance drop, hard to tell exactly due to variance in
the test results), when testing for kernel module throughput (with no
userspace activity, handful of kernel flows).
On flow setup, a single stats instance is allocated (for the NUMA node
0). As CPUs from multiple NUMA nodes start updating stats, new
NUMA-node specific stats instances are allocated. This allocation on
the packet processing code path is made to never block or look for
emergency memory pools, minimizing the allocation latency. If the
allocation fails, the existing preallocated stats instance is used.
Also, if only CPUs from one NUMA-node are updating the preallocated
stats instance, no additional stats instances are allocated. This
eliminates the need to pre-allocate stats instances that will not be
used, also relieving the stats reader from the burden of reading stats
that are never used.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
The 5-tuple optimization becomes unnecessary with a later per-NUMA
node stats patch. Remove it first to make the changes easier to
grasp.
Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
In few functions, const formal parameters are assigned or cast to
non-const.
These changes suppress warnings if compiled with -Wcast-qual.
Signed-off-by: Daniele Di Proietto <daniele.di.proietto@gmail.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
ovs_flow_free() is not called under ovs-lock during packet
execute path (ovs_packet_cmd_execute()). Since packet execute
does not touch flow->mask, there is no need to take that
lock either. So move assert in case where flow->mask is checked.
Found by code inspection.
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>
