Some CSR registers have to be configured also
in case of suspend/resume with unified image
(which doesn't includes reconfiguration flow).
Reuse the existing d3_suspend/d3_resume trans ops,
while making sure some configurations are a bit
different, according to the wowlan type.
After this change, we no longer need the special
wowlan_d0i3 configurations done in iwl_pci_resume,
as they are already being done in the d3_resume op.
Signed-off-by: Eliad Peller <eliad@wizery.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
CQM overwrites a few thresholds in the bf command. On the other hand,
when entering D0i3 the thresholds are set to higher values on purpose,
so ignore CQM in this case.
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
The slow filtering threshold should be higher in D0i3 case.
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
TCP software implementation on the host requires extensive computing
power. Offloading even some of the TCP/IP stack to the NIC might save
a significant overhead. In order to enable this feature on our hw,
we need to configure it first. Once done, we mark this capability,
to be advertised later to the OS via ieee80211_register_hw.
The driver Rx indications for TCP Checksum is integrated within the
standard Rx status. The driver responds to those indications as follows:
If the frame was tested by hw and checksum ok report CHECKSUM_UNNECESSARY.
Otherwise, report CHECKSUM_NONE.
Signed-off-by: Avri Altman <avri.altman@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
The firmware debug infrastructure allows the user to
provide a firmware that will toggle a few registers to
configure the debugging capabilities.
On certain devices, certain operations are forbidden.
Executing a forbidden operation will cause the hardware to
die in a way that only driver unload / load will bring it
back to life.
Fortunately, there is a way to know in advance if those
operations will be accepted by the device. This is where
the new PRPH_BLOCKBIT operation plays its role. If the bit
X from PRPH register Y is set, then we should prevent any
further register configuration. When that happens, drop a
line in the kernel log since this is really an error state:
the user won't have his device configured as he expected.
Add operations that will be used in the future:
INDIRECT_ASSIGN, INDIRECT_SETBIT, and INDIRECT_CLEARBIT.
Other debugging configurations (such as destination
configuration for the monitor) will take place in any case.
Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
In iwl_mvm_tx_skb_non_sta(), in case of managed interface,
use the AP station for multicast frames instead of the auxiliary
station as otherwise the frames can be sent to an absent P2P GO as
the FW does not block transmissions for the auxiliary station
since it is not associated with the station MAC context.
Note that this is not possible for unicast frames, as a TDLS
discovery response is sent without a station entry, and in this
case the P2P GO NoA should not block transmission to the peer.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Allow the transport layer to return an error upon suspend.
Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
Reviewed-by: Luciano Coelho <luciano.coelho@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
This reverts commit 088070a2f6.
When working in d0i3_on_idle mode, we explicitly go out
of d0i3 on resume (so other potential commands could
be sent).
However, D0I3_DEFER_WAKEUP is currently cleared on
resume complete (which happens only later on), causing
d0i3 exit to timeout.
Since mac80211 was modified to accept incoming frames
once drv_resume was called, we can safely revert this
patch, and handle the pending work on iwl_mvm_resume().
Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Existing UMAC commands already use the long header, but are sent
with group 0 and the long header inserted manually. Move them to
the group 1 to take advantage of the header building in the low-
level transport.
Existing firmware ignores the group_id field (it's reserved) and
the first firmware that really supports long command headers can
parse all commands in both group 0 (with short header) and group
1 (with long header.)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
As the firmware is slowly running out of command IDs and grouping of
commands is desirable anyway, the firmware is extending the command
header from 4 bytes to 8 bytes to introduce a group (in place of the
former flags field, since that's always 0 on commands and thus can
be easily used to distinguish between the two.
In order to support this most easily in the driver widen the command
command ID used in the command sending functions and encode the new
values (group and version) in the ID. That way existing code doesn't
have to be changed (since the higher bits are 0 automatically) and
newer code can easily use the new ID generation function to create a
value to use in place of just the command ID.
Signed-off-by: Aviya Erenfeld <aviya.erenfeld@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
ToF is a time based method for measurement of the WiFi device
location within a WiFi environment. The driver functionality provided
by this patch is the interface for communication with FW and receiving
location related updates from the FW. The interface provided by this
patch is via debugfs.
Signed-off-by: Gregory Greenman <gregory.greenman@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
All the supported firmwares support this API.
This includes removing dwell per band, as band is no longer a factor
in calculating the dwell. Only basic dwell is used and FW will calculate
the actual dwell time.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
The 'flags' field really has been reserved in the firmware API for a
very long time, probably since 4965. As a consequence, the field is
always 0 and checking for a IWL_CMD_FAILED_MSK flag makes no sense.
Rename the field to 'reserved', get rid of IWL_CMD_FAILED_MSK and
all the code for it.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
In iwlmvm firmwares, the Byte count written in the scheduler
byte count table is in DWORDs and not in bytes.
We should check that this value fits in the 12 bits and
the value can be either in bits of in DWORD or bytes
depending on the firmware. Check the value after the
translation to DWORDs is done (if needed).
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
In a few places, we were disabling interrupts but didn't
make sure that the interrupt handler has finished running.
Add calls to synchronize_irq() to ensure we finish handling
the interrupts before we free resources or other things that
could lead to a crash if the interrupt were to be handled
later.
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
When the firmware crashes, we can't expect the Tx queues to
progress. Cancel their timer.
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Since the time-event is sent with the immediate flag set, there is
no need to sample the device time.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
With the previous patch series, no opmode continues using the
command or handler_status (i.e. the return value from the RX)
so it can be removed now.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
In the mvm driver, neither the old command nor the return value
are used, so remove them.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
After the previous patches, the command that's passed in nor the
return value are used any more, so can be removed.
While at it, make some functions static.
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
This makes the logging a little less useful, but as they're mostly
synchronous commands it won't matter much. It gets rid of the
dependency on the input command, which this is the only user of.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
This driver currently has some very confusing ADD_STA response handling
that runs asynchronously in the background for all of the commands, but
is only really necessary for synchronous ones (the really asynchronous
ones can only be done for already existing stations), and for the sync
ones it actually waits for the RX handler to return a status code.
Rework this to keep the debug printing in the handler, but do the code
that's supposed to have an effect only for sync commands in the command
sending function.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
The current key offset assignment algorithm always uses the lowest
unused key offset, which will potentially lead to issues when the
firmware will change to take the key material for TX from the key
table rather than from the TX command.
In order to avoid those issues (and avoid forgetting about them)
change the key offset allocation algorithm now to avoid reusing key
offsets quickly.
The new algorithm always picks as the next offset the least recently
freed offset, i.e. the offset that has been unused for the longest
amount of time. This is implemented by having a generation counter
for each key offset that is incremented every time a key is deleted,
except for the one that's deleted, which is reset to zero. Thus the
highest counter is the key that's been unused longest.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
During NIC initialization shared HW is reset and this disables the
scheduler. Some HW platforms do not activate the scheduler after it.
Consequently all HCMD sent by the driver stay at the queues which cause
to queue stuck.
Set the scheduler to work on auto active mode so it would be activated upon
change over one of the queues' write pointer.
Signed-off-by: Haim Dreyfuss <haim.dreyfuss@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
In multiple locations there are checks for whether the label in hand
is a reserved label or not using the arbritray value of 16. Factor
this out into a #define for better maintainability and for
documentation.
Signed-off-by: Robert Shearman <rshearma@brocade.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Robert Shearman says:
====================
lwtunnel: encap locally-generated ipv4 packets
Locally-generated IPv4 packets, such as from applications running on
the host or traceroute/ping currently don't have lwtunnel output
redirected encap applied. However, they should do in the same way as
for forwarded packets and this patch series addresses that.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
lwtunnel encap is applied for forwarded packets, but not for
locally-generated packets. This is because the output function is not
overridden in __mkroute_output, unlike it is in __mkroute_input.
The lwtunnel state is correctly set on the rth through the call to
rt_set_nexthop, so all that needs to be done is to override the dst
output function to be lwtunnel_output if there is lwtunnel state
present and it requires output redirection.
Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the locally-generated packet path skb->protocol may not be set and
this is required for the lwtunnel encap in order to get the lwtstate.
This would otherwise have been set by ip_output or ip6_output so set
skb->protocol prior to calling the lwtunnel encap
function. Additionally set skb->dev in case it is needed further down
the transmit path.
Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of trying to access br->vlan_enabled directly use the provided
helper br_vlan_enabled().
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the introduction of the BPF action in d23b8ad8ab ("tc: add BPF
based action"), late binding was not working as expected. I.e. setting
the action part for a classifier only via 'bpf index <num>', where <num>
is the index of an existing action, is being rejected by the kernel due
to other missing parameters.
It doesn't make sense to require these parameters such as BPF opcodes
etc, as they are not going to be used anyway: in this case, they're just
allocated/parsed and then freed again w/o doing anything meaningful.
Instead, parse and verify the remaining parameters *after* the test on
tcf_hash_check(), when we really know that we're dealing with creation
of a new action or replacement of an existing one and where late binding
is thus irrelevant.
After patch, test case is now working:
FOO="1,6 0 0 4294967295,"
tc actions add action bpf bytecode "$FOO"
tc filter add dev foo parent 1: bpf bytecode "$FOO" flowid 1:1 action bpf index 1
tc actions show action bpf
action order 0: bpf bytecode '1,6 0 0 4294967295' default-action pipe
index 1 ref 2 bind 1
tc filter show dev foo
filter protocol all pref 49152 bpf
filter protocol all pref 49152 bpf handle 0x1 flowid 1:1 bytecode '1,6 0 0 4294967295'
action order 1: bpf bytecode '1,6 0 0 4294967295' default-action pipe
index 1 ref 2 bind 1
Late binding of a BPF action can be useful for preloading maps (e.g. before
they hit traffic) in case of eBPF programs, or to share a single eBPF action
with multiple classifiers.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
At switch setup, _mv88e6xxx_stats_wait was called without holding the
SMI mutex. Fix this by requesting the lock for this call.
Also, return the _mv88e6xxx_stats_wait code, since it may fail.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Before this patch when a vid was not specified, the entry was added with
vid 0 which is useless when vlan_filtering is enabled. This patch makes
the entry to be added on all configured vlans when vlan filtering is
enabled and respectively deleted from all, if the entry vid is 0.
This is also closer to the way fdb works with regard to vid 0 and vlan
filtering.
Example:
Setup:
$ bridge vlan add vid 256 dev eth4
$ bridge vlan add vid 1024 dev eth4
$ bridge vlan add vid 64 dev eth3
$ bridge vlan add vid 128 dev eth3
$ bridge vlan
port vlan ids
eth3 1 PVID Egress Untagged
64
128
eth4 1 PVID Egress Untagged
256
1024
$ echo 1 > /sys/class/net/br0/bridge/vlan_filtering
Before:
$ bridge mdb add dev br0 port eth3 grp 239.0.0.1
$ bridge mdb
dev br0 port eth3 grp 239.0.0.1 temp
After:
$ bridge mdb add dev br0 port eth3 grp 239.0.0.1
$ bridge mdb
dev br0 port eth3 grp 239.0.0.1 temp vid 1
dev br0 port eth3 grp 239.0.0.1 temp vid 128
dev br0 port eth3 grp 239.0.0.1 temp vid 64
Signed-off-by: Satish Ashok <sashok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Toshiaki Makita says:
====================
Stacked vlan TSO for virtual devices
Basically virtual devices do not need to segment double tagged packets.
This patch set adds TSO feature for double tagged packets to several
virtual devices, which can be realized by simply setting
.ndo_features_check to passthru_features_check.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Tap devices don't need to segment multiple tagged packets.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bridge devices don't need to segment multiple tagged packets since thier
ports can segment them.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Veth devices don't need to segment multiple tagged packets.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Macvlan/macvtap devices don't need to segment multiple tagged packets
since the lower devices can segment them.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Straightforward patch to add GRO processing to virtio_net.
napi_complete_done() usage allows more aggressive aggregation,
opted-in by setting /sys/class/net/xxx/gro_flush_timeout
Tested:
Setting /sys/class/net/xxx/gro_flush_timeout to 1000 nsec,
Rick Jones reported following results.
One VM of each on a pair of OpenStack compute nodes with E5-2650Lv3 CPUs
and Intel 82599ES-based NICs. So, two "before" and two "after" VMs.
The OpenStack compute nodes were running OpenStack Kilo, with VxLAN
encapsulation being used through OVS so no GRO coming-up the host
stack. The compute nodes themselves were running a 3.14-based kernel.
Single-stream netperf, CPU utilizations and thus service demands are
based on intra-guest reported CPU.
Throughput Mbit/s, bigger is better
Min Median Average Max
4.2.0-rc3+ 1364 1686 1678 1938
4.2.0-rc3+flush1k 1824 2269 2275 2647
Send Service Demand, smaller is better
Min Median Average Max
4.2.0-rc3+ 0.236 0.558 0.524 0.802
4.2.0-rc3+flush1k 0.176 0.503 0.471 0.738
Receive Service Demand, smaller is better.
Min Median Average Max
4.2.0-rc3+ 1.906 2.188 2.191 2.531
4.2.0-rc3+flush1k 0.448 0.529 0.533 0.692
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Rick Jones <rick.jones2@hp.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Suggested-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rocker supports the transmission of scattered packets, so let the kernel
know about it by setting the NETIF_F_SG bit in the device's features.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add skb->hash to the __sk_buff offset map, so it can be accessed from
an eBPF program. We currently already do this for classic BPF filters,
but not yet on eBPF, it might be useful as a demuxer in combination with
helpers like bpf_clone_redirect(), toy example:
__section("cls-lb") int ingress_main(struct __sk_buff *skb)
{
unsigned int which = 3 + (skb->hash & 7);
/* bpf_skb_store_bytes(skb, ...); */
/* bpf_l{3,4}_csum_replace(skb, ...); */
bpf_clone_redirect(skb, which, 0);
return -1;
}
I was thinking whether to add skb_get_hash(), but then concluded the
raw skb->hash seems fine in this case: we can directly access the hash
w/o extra eBPF helper function call, it's filled out by many NICs on
ingress, and in case the entropy level would not be sufficient, people
can still implement their own specific sw fallback hash mix anyway.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 05cc5a39dd ("bnx2x: add vlan filtering offload") has introduced
an incorrect logic for checking whether pvid should be configured for
a vf, causing the hypervisor driver to send unneeded ramrods for all of
the vfs each time a pvid has changed.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: Ariel Elior <Ariel.Elior@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
arch/s390/net/bpf_jit_comp.c
drivers/net/ethernet/ti/netcp_ethss.c
net/bridge/br_multicast.c
net/ipv4/ip_fragment.c
All four conflicts were cases of simple overlapping
changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking fixes from David Miller:
1) Must teardown SR-IOV before unregistering netdev in igb driver, from
Alex Williamson.
2) Fix ipv6 route unreachable crash in IPVS, from Alex Gartrell.
3) Default route selection in ipv4 should take the prefix length, table
ID, and TOS into account, from Julian Anastasov.
4) sch_plug must have a reset method in order to purge all buffered
packets when the qdisc is reset, likewise for sch_choke, from WANG
Cong.
5) Fix deadlock and races in slave_changelink/br_setport in bridging.
From Nikolay Aleksandrov.
6) mlx4 bug fixes (wrong index in port even propagation to VFs,
overzealous BUG_ON assertion, etc.) from Ido Shamay, Jack
Morgenstein, and Or Gerlitz.
7) Turn off klog message about SCTP userspace interface compat that
makes no sense at all, from Daniel Borkmann.
8) Fix unbounded restarts of inet frag eviction process, causing NMI
watchdog soft lockup messages, from Florian Westphal.
9) Suspend/resume fixes for r8152 from Hayes Wang.
10) Fix busy loop when MSG_WAITALL|MSG_PEEK is used in TCP recv, from
Sabrina Dubroca.
11) Fix performance regression when removing a lot of routes from the
ipv4 routing tables, from Alexander Duyck.
12) Fix device leak in AF_PACKET, from Lars Westerhoff.
13) AF_PACKET also has a header length comparison bug due to signedness,
from Alexander Drozdov.
14) Fix bug in EBPF tail call generation on x86, from Daniel Borkmann.
15) Memory leaks, TSO stats, watchdog timeout and other fixes to
thunderx driver from Sunil Goutham and Thanneeru Srinivasulu.
16) act_bpf can leak memory when replacing programs, from Daniel
Borkmann.
17) WOL packet fixes in gianfar driver, from Claudiu Manoil.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (79 commits)
stmmac: fix missing MODULE_LICENSE in stmmac_platform
gianfar: Enable device wakeup when appropriate
gianfar: Fix suspend/resume for wol magic packet
gianfar: Fix warning when CONFIG_PM off
act_pedit: check binding before calling tcf_hash_release()
net: sk_clone_lock() should only do get_net() if the parent is not a kernel socket
net: sched: fix refcount imbalance in actions
r8152: reset device when tx timeout
r8152: add pre_reset and post_reset
qlcnic: Fix corruption while copying
act_bpf: fix memory leaks when replacing bpf programs
net: thunderx: Fix for crash while BGX teardown
net: thunderx: Add PCI driver shutdown routine
net: thunderx: Fix crash when changing rss with mutliple traffic flows
net: thunderx: Set watchdog timeout value
net: thunderx: Wakeup TXQ only if CQE_TX are processed
net: thunderx: Suppress alloc_pages() failure warnings
net: thunderx: Fix TSO packet statistic
net: thunderx: Fix memory leak when changing queue count
net: thunderx: Fix RQ_DROP miscalculation
...
Tom Herbert says:
====================
ipv6: Turn on auto IPv6 flow labels by default
BSD (MacOS) has already turned on flow labels by default and this does
not seem to be causing any problems in the Internet. Let's go ahead
and turn them on by default. We'll continue to monitor for any devices
start choking on them.
Flow labels are important since they are the desired solution for
network devices to perform ECMP and RSS (RFC6437 and RFC6438).
Traditionally, devices perform a 5-tuple hash on packets that
includes port numbers. For the most part, these devices can only
compute 5-tuple hashes for TCP and UDP. This severely limits our ability
to get good network load balancing for other protocols (IPIP, GRE,ESP,
etc.), and hence we are limited in using other protocols. Unfortunately,
this method is accepted as the de facto standard to the extent that
there are several proposals to encapsulate protocols in UDP _just_ for
the purposes for getting ECMP to work. With hosts generating flow labels
and devices taking them as input into ECMP (several already do), we can
start to fix this fundamental problem.
This patch set:
- Changes IPV6_FLOWINFO sockopt to be opt-out of flow labels for
connections rather than opt-in
- Disable flow label state ranges sysctl by default
- Enable auto flow labels sysctl by default
v2:
- Added functions to create an skb->hash based on flowi4 and flowi6.
These are called in output path when creating a packet
- Call skb_get_hash_flowi6 in ip6_make_flowlabel
- Implement the auto_flowlabels sysctl as a mode for auto flowlabels.
There are four modes which correspond to flow labels being enabled
and whether socket option can be used to opt in or opt out of
using them
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Initialize auto_flowlabels to one. This enables automatic flow labels,
individual socket may disable them using the IPV6_AUTOFLOWLABEL socket
option.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Per RFC6437 stateful flow labels (e.g. labels set by flow label manager)
cannot "disturb" nodes taking part in stateless flow labels. While the
ranges only reduce the flow label entropy by one bit, it is conceivable
that this might bias the algorithm on some routers causing a load
imbalance. For best results on the Internet we really need the full
20 bits.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change the meaning of net.ipv6.auto_flowlabels to provide a mode for
automatic flow labels generation. There are four modes:
0: flow labels are disabled
1: flow labels are enabled, sockets can opt-out
2: flow labels are allowed, sockets can opt-in
3: flow labels are enabled and enforced, no opt-out for sockets
np->autoflowlabel is initialized according to the sysctl value.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can't call skb_get_hash here since the packet is not complete to do
flow_dissector. Create hash based on flowi6 instead.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
