cfg80211_chan_def and ieee80211_channel recently gained a
frequency offset component. Handle this where it makes
sense (potentially required by S1G channels).
For IBSS, TDLS, CSA, and ROC we return -EOPNOTSUPP if a
channel with frequency offset is passed, since they may or
may not work. Once someone tests and verifies these
commands work on thos types of channels, we can remove
that error.
join_ocb and join_mesh look harmless because they use a
simple ieee80211_vif_use_channel(), which is using an
already verified channel, so we let those through.
Signed-off-by: Thomas Pedersen <thomas@adapt-ip.com>
Link: https://lore.kernel.org/r/20200402011810.22947-4-thomas@adapt-ip.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Some bands (S1G) define channels centered on a non-integer
MHz. Give ieee80211_channel and cfg80211_chan_def a
freq_offset component where the final frequency can be
expressed as:
MHZ_TO_KHZ(chan->center_freq) + chan->freq_offset;
Also provide some helper functions to do the frequency
conversion and test for equality.
Retain the existing interface to frequency and channel
conversion helpers, and expose new ones which handle
frequencies in units of KHz.
Some internal functions (net/wireless/chan.c) pass around
a frequency value. Convert these to units of KHz.
mesh, ibss, wext, etc. are currently ignored.
Signed-off-by: Thomas Pedersen <thomas@adapt-ip.com>
Link: https://lore.kernel.org/r/20200402011810.22947-3-thomas@adapt-ip.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
MHZ_TO_KHZ, and KHZ_TO_MHZ are useful to drivers and
elsewhere so export these in the common ieee80211 header.
Move the power helpers also because we might as well.
Signed-off-by: Thomas Pedersen <thomas@adapt-ip.com>
Link: https://lore.kernel.org/r/20200402011810.22947-2-thomas@adapt-ip.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This is actually true because there's no functional beacon distribution
and lets us get active scanning working - without it, mac80211 doesn't
respond to probe requests.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.554d1199b309.Id86fd36e3d88d2a75d6e0c6618fd93ce8fe84065@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The MLME logic had a workaround that allowed to continue an
association with an AP even if the AP did not provide any basic
rates in its supported rates in the association response, assuming
that the first (non basic) legacy rate could be used as a basic rate.
However, this did not consider the case where the AP (which is
obviously buggy) did not provide any legacy rate.
Fix this by failing the association, as this can result in
an unexpected failure in the low level driver and FW, e.g., in
rate scale logic etc.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.d70a1450d83f.I6e6ce5efda351a8544c0e7bfeee260fe3360d401@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Driver tells mac80211 to sends ADDBA with SSN (starting sequence number)
from the head of the queue, while the transmission of all the frames in the
queue may take a while, which causes the peer to time out. In order to
fix this scenario, add an option to defer ADDBA transmit until queue
is drained.
Signed-off-by: Mordechay Goodstein <mordechay.goodstein@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.0f27423fec75.If67daab123a27c1cbddef000d6a3f212aa6309ef@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
SAE AP may reject authentication with WLAN_STATUS_ANTI_CLOG_REQUIRED.
As the user space will immediately continue the authentication flow,
there is no need to destroy the authentication data in this case.
This saves unneeded station removal and releasing the channel.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.7483996157a8.I8040a842874aaf6d209df3fc8a2acb97a0bf508b@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Somehow we missed this for a long time, but similar to the extended
NSS support in VHT capabilities, we need to have this in Operating
Mode notification.
Implement it by
* parsing the 160/80+80 bit there and setting the bandwidth
appropriately
* having callers of ieee80211_get_vht_max_nss() pass in the current
max NSS value as received in the operating mode notification in
order to modify it appropriately depending on the extended NSS
bits.
This updates all drivers that use it, i.e. only iwlwifi/mvm.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>
Link: https://lore.kernel.org/r/iwlwifi.20200326150855.098483728cfa.I4e8c25d3288441759c2793247197229f0696a37d@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Convert a user space registration for processing multicast Action frames
(NL80211_CMD_REGISTER_FRAME with NL80211_ATTR_RECEIVE_MULTICAST) to a
new enum ieee80211_filter_flags bit FIF_MCAST_ACTION so that drivers can
update their RX filter parameters appropriately, if needed.
Signed-off-by: Jouni Malinen <jouni@codeaurora.org>
Link: https://lore.kernel.org/r/20200421144815.19175-1-jouni@codeaurora.org
[rename variables to rx_mcast_action_reg indicating action frames only]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
For DPP, there's a need to receive multicast action frames,
but many drivers need a special filter configuration for this.
Support announcing from userspace in the management registration
that multicast RX is required, with an extended feature flag if
the driver handles this.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Sergey Matyukevich <sergey.matyukevich.os@quantenna.com>
Link: https://lore.kernel.org/r/20200417124013.c46238801048.Ib041d437ce0bff28a0c6d5dc915f68f1d8591002@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Almost all drivers below cfg80211 get the API wrong (except for
cfg80211) and are unable to cope with multiple registrations for
the same frame type, which is valid due to the match filter.
This seems to indicate the API is wrong, and we should maintain
the full information in cfg80211 instead of the drivers.
Change the API to no longer inform the driver about individual
registrations and unregistrations, but rather every time about
the entire state of the entire wiphy and single wdev, whenever
it may have changed. This also simplifies the code in cfg80211
as it no longer has to track exactly what was unregistered and
can free things immediately.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-by: Arend van Spriel <arend.vanspriel@broadcom.com>
Reviewed-by: Sergey Matyukevich <sergey.matyukevich.os@quantenna.com>
Link: https://lore.kernel.org/r/20200417124300.f47f3828afc8.I7f81ef59c2c5a340d7075fb3c6d0e08e8aeffe07@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Report received Beacon frames that do not have a valid MME MIC when
beacon protection is enabled. This covers both the cases of no MME in
the received frame and invalid MIC in the MME.
Signed-off-by: Jouni Malinen <jouni@codeaurora.org>
Link: https://lore.kernel.org/r/20200401142548.6990-2-jouni@codeaurora.org
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Extend cfg80211_rx_unprot_mlme_mgmt() to cover indication of unprotected
Beacon frames in addition to the previously used Deauthentication and
Disassociation frames. The Beacon frame case is quite similar, but has
couple of exceptions: this is used both with fully unprotected and also
incorrectly protected frames and there is a rate limit on the events to
avoid unnecessary flooding netlink events in case something goes wrong.
Signed-off-by: Jouni Malinen <jouni@codeaurora.org>
Link: https://lore.kernel.org/r/20200401142548.6990-1-jouni@codeaurora.org
[add missing kernel-doc]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
There are two bugs with this, first, it shouldn't be called
on an interface that's down, and secondly, it should then be
called when the interface comes up.
Note that the currently only user (iwlwifi) doesn't seem to
care about either of these scenarios.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Link: https://lore.kernel.org/r/20200417111830.401d82c7a0bf.I5dc7d718816460c2d8d89c7af6c215f9e2b3078f@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Maintain the connection AID only in sdata->vif.bss_conf.aid, not
also in sdata->u.mgd.aid.
Keep setting that where we set ifmgd->aid before, which has the
side effect of exposing the AID to the driver before the station
entry (AP) is marked associated, in case it needs it then.
Requested-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Tested-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/20200417123802.085d4a322b0c.I2e7a2ceceea8c6880219f9e9ee4d4ac985fd295a@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Currently, wmediumd requires each used MAC address to be configured
as a station in the virtual air, but that doesn't make sense as any
station could have multiple MAC addresses, and even have randomized
ones in scanning, etc.
Add some code here to tell wmediumd of used MAC addresses, binding
them to the hardware address. Combined with a wmediumd patch that
makes it track the addresses this allows configuring just the radio
address (42:00:00:00:nn:00 unless the radio was manually created)
in wmediumd as a station, and all addresses that the station uses
are added/removed dynamically.
Tested with random scan, which without this and the corresponding
wmediumd change doesn't get anything through as the sender doesn't
exist as far as wmediumd is concerned (it's random).
Link: https://lore.kernel.org/r/20200323162358.b397b1a1acef.Ice0536e34e5d96c51f97c374ea8af9551347c7e8@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Tonghao Zhang says:
====================
openvswitch: expand meter tables and fix bug
The patch set expand or shrink the meter table when necessary.
and other patches fix bug or improve codes.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
When setting the meter rate to 4+Gbps, there is an
overflow, the meters don't work as expected.
Cc: Pravin B Shelar <pshelar@ovn.org>
Cc: Andy Zhou <azhou@ovn.org>
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>
Cc: Pravin B Shelar <pshelar@ovn.org>
Cc: Andy Zhou <azhou@ovn.org>
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>
Before invoking the ovs_meter_cmd_reply_stats, "meter"
was checked, so don't check it agin in that function.
Cc: Pravin B Shelar <pshelar@ovn.org>
Cc: Andy Zhou <azhou@ovn.org>
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>
Don't allow user to create meter unlimitedly, which may cause
to consume a large amount of kernel memory. The max number
supported is decided by physical memory and 20K meters as default.
Cc: Pravin B Shelar <pshelar@ovn.org>
Cc: Andy Zhou <azhou@ovn.org>
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>
In kernel datapath of Open vSwitch, there are only 1024
buckets of meter in one datapath. If installing more than
1024 (e.g. 8192) meters, it may lead to the performance drop.
But in some case, for example, Open vSwitch used as edge
gateway, there should be 20K at least, where meters used for
IP address bandwidth limitation.
[Open vSwitch userspace datapath has this issue too.]
For more scalable meter, this patch use meter array instead of
hash tables, and expand/shrink the array when necessary. So we
can install more meters than before in the datapath.
Introducing the struct *dp_meter_instance, it's easy to
expand meter though changing the *ti point in the struct
*dp_meter_table.
Cc: Pravin B Shelar <pshelar@ovn.org>
Cc: Andy Zhou <azhou@ovn.org>
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>
Currently the unsigned variable tmp is being checked for an negative
error return from the call to bcm_phy_read_rdb and this can never
be true since tmp is unsigned. Fix this by making tmp a plain int.
Addresses-Coverity: ("Unsigned compared against 0")
Fixes: 4406d36dfd ("net: phy: bcm54140: add hwmon support")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Michael Walle <michael@walle.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the following sparse warning:
drivers/net/ethernet/qlogic/qed/qed_ll2.c:2334:20: warning: symbol 'll2_cbs'
was not declared. Should it be static?
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Zou Wei <zou_wei@huawei.com>
Acked-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The CPSW misc IRQ need be enabled for CPTS event_pend IRQs processing. This
patch adds corresponding support to CPSW driver.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Hence CPTS IRQ support is in place the W_TS_PUSH events can be added.
PWM capable DmTimers can be used to generete input signals for CPTS on TI
AM335x/AM437x/DRA7 SoCs to be timestamped:
AM335x/AM437x: timer4 - timer7
DRA7/AM57xx: timer13 - timer16
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add CPTS IRQ support, but do not enable it. By default, the CPTS driver
will continue working using polling mode which is required for CPTS to
continue working on platforms other than CPSW, like Keystone 2.
The CPTS IRQ support is required to enable support for HW_TS_PUSH events.
The CPSW CPTS IRQ and HW_TS_PUSH events support will be enabled in follow
up patches.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now spinlock is used to synchronize everything which is not required. Add
mutex and use to sync access to PTP interface and PTP worker and use
spinlock only to sync FIFO/events processing.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now the tx timestamp processing happens from different contexts - softirq
and thread/PTP worker. Enabling IRQ will add one more hard_irq context.
This makes over all defered TX timestamp processing and locking
overcomplicated. Move tx timestamp processing to PTP worker always instead.
napi_rx->cpts_tx_timestamp
if ptp_packet then
push to txq
ptp_schedule_worker()
do_aux_work->cpts_overflow_check
cpts_process_events()
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now the CPTS driver performs packet (skb) parsing every time when it needs
to match packet to CPTS event (including ptp_classify_raw() calls).
This patch optimizes matching process by parsing packet only once upon
arrival and stores PTP specific data in skb->cb using the same fromat as in
CPTS HW event. As result, all future matching reduces to comparing two u32
values.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The CPTS HW latches and saves CPTS counter value in CPTS fifo immediately
after writing to CPSW_CPTS_PUSH.TS_PUSH (bit 0), so the total time that the
driver needs to read the CPTS timestamp is the time required CPSW_CPTS_PUSH
write to actually reach HW.
Hence switch CPTS driver to implement new .gettimex64() callback for more
precise measurement of the offset between a PHC and the system clock which
is measured as time between
write(CPSW_CPTS_PUSH)
read(CPSW_CPTS_PUSH)
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now CPTS driver .adjfreq() generates request to read CPTS current time
(CPTS_EV_PUSH) with intention to process all pending event using previous
frequency adjustment values before switching to the new ones. So
CPTS_EV_PUSH works as a marker to switch to the new frequency adjustment
values. Current code assumes that all job is done in .adjfreq(), but after
enabling IRQ this will not be true any more.
Hence save new frequency adjustment values (mult) and perform actual freq
adjustment in cpts_fifo_read() immediately after CPTS_EV_PUSH is received.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now CPTS HW time reading code is implemented in timecounter->cyclecounter
.read() callback and performs following operations:
timecounter_read() ->cc.read() -> cpts_systim_read()
- request current CPTS HW time CPTS_TS_PUSH.TS_PUSH = 1
- poll CPTS FIFO for CPTS_EV_PUSH event with current HW timestamp
This approach need to be changed for the future switch to PTP PHC
.gettimex64() callback, which require to separate requesting current CPTS
HW time and processing CPTS FIFO. And for the follow up patch, which
improves .adjfreq() implementation.
This patch moves code accessing CPTS HW out of timecounter code as
following:
- convert HW timestamp of every CPTS event to PTP time (us) and store it as
part struct cpts_event;
- add CPTS context field to store current CPTS HW time (counter) value and
update it on CPTS_EV_PUSH reception;
- move code accessing CPTS HW out of timecounter code and use current CPTS
HW time (counter) from CPTS context instead;
- ensure timecounter->cycle_last is updated on CPTS_EV_PUSH reception.
After this change CPTS timecounter will only perform timekeeper role
without actually accessing CPTS HW.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use dev_yy() API instead of pr_yy() for log outputs.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet says:
====================
net: napi: addition of napi_defer_hard_irqs
This patch series augments gro_glush_timeout feature with napi_defer_hard_irqs
As extensively described in first patch changelog, this can suppresss
the chit-chat traffic between NIC and host to signal interrupts and re-arming
them, since this can be an issue on high speed NIC with many queues.
The last patch in this series converts mlx4 TX completion to
napi_complete_done(), to enable this new mechanism.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to benefit from the new napi_defer_hard_irqs feature,
we need to use napi_complete_done() variant in this driver.
RX path is already using it, this patch implements TX completion side.
mlx4_en_process_tx_cq() now returns the amount of retired packets,
instead of a boolean, so that mlx4_en_poll_tx_cq() can pass
this value to napi_complete_done().
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
gro_flush_timeout and napi_defer_hard_irqs can be read
from napi_complete_done() while other cpus write the value,
whithout explicit synchronization.
Use READ_ONCE()/WRITE_ONCE() to annotate the races.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Back in commit 3b47d30396 ("net: gro: add a per device gro flush timer")
we added the ability to arm one high resolution timer, that we used
to keep not-complete packets in GRO engine a bit longer, hoping that further
frames might be added to them.
Since then, we added the napi_complete_done() interface, and commit
364b605573 ("net: busy-poll: return busypolling status to drivers")
allowed drivers to avoid re-arming NIC interrupts if we made a promise
that their NAPI poll() handler would be called in the near future.
This infrastructure can be leveraged, thanks to a new device parameter,
which allows to arm the napi hrtimer, instead of re-arming the device
hard IRQ.
We have noticed that on some servers with 32 RX queues or more, the chit-chat
between the NIC and the host caused by IRQ delivery and re-arming could hurt
throughput by ~20% on 100Gbit NIC.
In contrast, hrtimers are using local (percpu) resources and might have lower
cost.
The new tunable, named napi_defer_hard_irqs, is placed in the same hierarchy
than gro_flush_timeout (/sys/class/net/ethX/)
By default, both gro_flush_timeout and napi_defer_hard_irqs are zero.
This patch does not change the prior behavior of gro_flush_timeout
if used alone : NIC hard irqs should be rearmed as before.
One concrete usage can be :
echo 20000 >/sys/class/net/eth1/gro_flush_timeout
echo 10 >/sys/class/net/eth1/napi_defer_hard_irqs
If at least one packet is retired, then we will reset napi counter
to 10 (napi_defer_hard_irqs), ensuring at least 10 periodic scans
of the queue.
On busy queues, this should avoid NIC hard IRQ, while before this patch IRQ
avoidance was only possible if napi->poll() was exhausting its budget
and not call napi_complete_done().
This feature also can be used to work around some non-optimal NIC irq
coalescing strategies.
Having the ability to insert XX usec delays between each napi->poll()
can increase cache efficiency, since we increase batch sizes.
It also keeps serving cpus not idle too long, reducing tail latencies.
Co-developed-by: Luigi Rizzo <lrizzo@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sudarsana Reddy Kalluru says:
====================
qed*: Add support for pcie advanced error recovery.
The patch series adds qed/qede driver changes for PCIe Advanced Error
Recovery (AER) support.
Patch (1) adds qed changes to enable the device to send error messages
to root port when detected.
Patch (2) adds qede support for handling the detected errors (AERs).
Changes from previous version:
-------------------------------
v2: use pci_num_vf() instead of caching the value in edev.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The error recovery is handled by management firmware (MFW) with the help of
qed/qede drivers. Upon detecting the errors, driver informs MFW about this
event which in turn starts a recovery process. MFW sends ERROR_RECOVERY
notification to the driver which performs the required cleanup/recovery
from the driver side.
Signed-off-by: Sudarsana Reddy Kalluru <skalluru@marvell.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The patch enables the device to send error messages to root port when
an error is detected.
Signed-off-by: Sudarsana Reddy Kalluru <skalluru@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
gro_cells lib is used by different encapsulating netdevices, such as
geneve, macsec, vxlan etc. to speed up decapsulated traffic processing.
CPU tag is a sort of "encapsulation", and we can use the same mechs to
greatly improve overall DSA performance.
skbs are passed to the GRO layer after removing CPU tags, so we don't
need any new packet offload types as it was firstly proposed by me in
the first GRO-over-DSA variant [1].
The size of struct gro_cells is sizeof(void *), so hot struct
dsa_slave_priv becomes only 4/8 bytes bigger, and all critical fields
remain in one 32-byte cacheline.
The other positive side effect is that drivers for network devices
that can be shipped as CPU ports of DSA-driven switches can now use
napi_gro_frags() to pass skbs to kernel. Packets built that way are
completely non-linear and are likely being dropped without GRO.
This was tested on to-be-mainlined-soon Ethernet driver that uses
napi_gro_frags(), and the overall performance was on par with the
variant from [1], sometimes even better due to minimal overhead.
net.core.gro_normal_batch tuning may help to push it to the limit
on particular setups and platforms.
iperf3 IPoE VLAN NAT TCP forwarding (port1.218 -> port0) setup
on 1.2 GHz MIPS board:
5.7-rc2 baseline:
[ID] Interval Transfer Bitrate Retr
[ 5] 0.00-120.01 sec 9.00 GBytes 644 Mbits/sec 413 sender
[ 5] 0.00-120.00 sec 8.99 GBytes 644 Mbits/sec receiver
Iface RX packets TX packets
eth0 7097731 7097702
port0 426050 6671829
port1 6671681 425862
port1.218 6671677 425851
With this patch:
[ID] Interval Transfer Bitrate Retr
[ 5] 0.00-120.01 sec 12.2 GBytes 870 Mbits/sec 122 sender
[ 5] 0.00-120.00 sec 12.2 GBytes 870 Mbits/sec receiver
Iface RX packets TX packets
eth0 9474792 9474777
port0 455200 353288
port1 9019592 455035
port1.218 353144 455024
v2:
- Add some performance examples in the commit message;
- No functional changes.
[1] https://lore.kernel.org/netdev/20191230143028.27313-1-alobakin@dlink.ru/
Signed-off-by: Alexander Lobakin <bloodyreaper@yandex.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
