drivers/net/ethernet/freescale/dpaa/dpaa_eth.c:2734:34: warning:
tentative array definition assumed to have one element
static const struct of_device_id dpaa_match[];
^
1 warning generated.
Turns out that since this driver was introduced in commit 9ad1a3749333
("dpaa_eth: add support for DPAA Ethernet"), this declaration has been
unused. Remove it to silence the warning.
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com> Acked-by: Madalin Bucur <madalin.bucur@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Patrick Ruddy [Mon, 1 Oct 2018 08:41:27 +0000 (09:41 +0100)]
ipv6: add vrf table handling code for ipv6 mcast
The code to obtain the correct table for the incoming interface was
missing for IPv6. This has been added along with the table creation
notification to fib rules for the RTNL_FAMILY_IP6MR address family.
Signed-off-by: Patrick Ruddy <pruddy@vyatta.att-mail.com> Signed-off-by: Mike Manning <mmanning@vyatta.att-mail.com> Reviewed-by: David Ahern <dsahern@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Robert Shearman [Mon, 1 Oct 2018 08:40:23 +0000 (09:40 +0100)]
ipv4: Allow sending multicast packets on specific i/f using VRF socket
It is useful to be able to use the same socket for listening in a
specific VRF, as for sending multicast packets out of a specific
interface. However, the bound device on the socket currently takes
precedence and results in the packets not being sent.
Relax the condition on overriding the output interface to use for
sending packets out of UDP, raw and ping sockets to allow multicast
packets to be sent using the specified multicast interface.
Signed-off-by: Robert Shearman <rshearma@vyatta.att-mail.com> Signed-off-by: Mike Manning <mmanning@vyatta.att-mail.com> Reviewed-by: David Ahern <dsahern@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ido Schimmel [Mon, 1 Oct 2018 08:57:01 +0000 (11:57 +0300)]
bridge: mcast: Default back to multicast enabled state
Commit 13cefad2f2c1 ("net: bridge: convert and rename mcast disabled")
converted the 'multicast_disabled' field to an option bit named
'BROPT_MULTICAST_ENABLED'.
While the old field was implicitly initialized to 0, the new field is
not initialized, resulting in the bridge defaulting to multicast
disabled state and breaking existing applications.
Fix this by explicitly initializing the option.
Fixes: 13cefad2f2c1 ("net: bridge: convert and rename mcast disabled") Signed-off-by: Ido Schimmel <idosch@mellanox.com> Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
====================
dpaa2-eth: Add support for Rx flow classification
The Management Complex (MC) firmware initially allowed the
configuration of a single key to be used both for Rx flow hashing
and flow classification. This prevented us from supporting
Rx flow classification independently of the hash key configuration.
Newer firmware versions expose separate commands for
configuring the two types of keys, so we can use them to
introduce Rx classification support. For frames that don't match
any classification rule, we fall back to statistical distribution
based on the current hash key.
The first patch in this set updates the Rx hashing code to use
the new firmware API for key config. Subsequent patches introduce
the firmware API for configuring the classification and actual
support for adding and deleting rules via ethtool.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Ioana Radulescu [Mon, 1 Oct 2018 10:44:58 +0000 (13:44 +0300)]
dpaa2-eth: Add ethtool support for flow classification
Add support for inserting and deleting Rx flow classification
rules through ethtool.
We support classification based on some header fields for
flow-types ether, ip4, tcp4, udp4 and sctp4.
Rx queues are core affine, so the action argument effectively
selects on which cpu the matching frame will be processed.
Discarding the frame is also supported.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ioana Radulescu [Mon, 1 Oct 2018 10:44:57 +0000 (13:44 +0300)]
dpaa2-eth: Configure Rx flow classification key
For firmware versions that support it, configure an Rx flow
classification key at probe time.
Hardware expects all rules in the classification table to share
the same key. So we setup a key containing all supported fields
at driver init and when a user adds classification rules through
ethtool, we will just mask out the unused header fields.
Since the key composition process is the same for flow
classification and hashing, reuse existing code where possible.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ioana Radulescu [Mon, 1 Oct 2018 10:44:56 +0000 (13:44 +0300)]
dpaa2-eth: Rename structure
Since the array of supported header fields will be used for
Rx flow classification as well, rename it from "hash_fields" to
the more inclusive "dist_fields".
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ioana Radulescu [Mon, 1 Oct 2018 10:44:55 +0000 (13:44 +0300)]
dpaa2-eth: Use new API for Rx flow hashing
The Management Complex (MC) firmware initially allowed the
configuration of a single key to be used both for Rx flow hashing
and flow classification. This prevented us from supporting
Rx flow classification through ethtool.
Starting with version 10.7.0, the Management Complex(MC) offers
a new set of APIs for separate configuration of Rx hashing and
classification keys.
Update the Rx flow hashing support to use the new API, if available.
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Ben Dooks [Mon, 1 Oct 2018 16:02:43 +0000 (17:02 +0100)]
net: usbnet: make driver_info const
The driver_info field that is used for describing each of the usb-net
drivers using the usbnet.c core all declare their information as const
and the usbnet.c itself does not try and modify the struct.
It is therefore a good idea to make this const in the usbnet.c structure
in case anyone tries to modify it.
Signed-off-by: Ben Dooks <ben-linux@fluff.org> Signed-off-by: Ben Dooks <ben.dooks@codethink.co.uk> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Tue, 2 Oct 2018 06:24:26 +0000 (23:24 -0700)]
tcp: do not release socket ownership in tcp_close()
syzkaller was able to hit the WARN_ON(sock_owned_by_user(sk));
in tcp_close()
While a socket is being closed, it is very possible other
threads find it in rtnetlink dump.
tcp_get_info() will acquire the socket lock for a short amount
of time (slow = lock_sock_fast(sk)/unlock_sock_fast(sk, slow);),
enough to trigger the warning.
Fixes: 67db3e4bfbc9 ("tcp: no longer hold ehash lock while calling tcp_get_info()") Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: syzbot <syzkaller@googlegroups.com> Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller [Tue, 2 Oct 2018 18:46:29 +0000 (11:46 -0700)]
Merge tag 'wireless-drivers-next-for-davem-2018-10-02' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next
Kalle Valo says:
====================
wireless-drivers-next patches for 4.20
First set of new features for 4.20. mt76 driver is going through major
refactoring and that's why there are so many mt76 patches. iwlwifi is
also under heavy development and smaller changes to other drivers.
Also wireless-drivers was merged to fix a conflict between the two trees.
Major changes:
ath10k
* limit available channels via DT ieee80211-freq-limit
wil6210
* add 802.11r Fast Roaming support for AP and station modes
* add support for channel 4
iwlwifi
* new FW API handling
* some improvements in the PCI recovery mechanism
* enable a new scanning feature;
* continued work on HE (mostly radiotap)
* TKIP implementation in new devices
* work continues for new 22560 hardware
mt76
* add support for Alfa AWUS036ACM
* lots of refactoring to make it easier to add new hardware support
* prepare for adding mt76x0e (pci-e variant) support
* add CONFIG_MT76x0E kconfig symbol
brcmfmac
* add support CYW89342 mini-PCIe device
* add 4-way handshake offload detection for FT-802.1X
* enable NL80211_EXT_FEATURE_CQM_RSSI_LIST
* fix for proper support of 160MHz bandwidth
rtl8xxxu
* add rtl8188ctv support
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller [Tue, 2 Oct 2018 18:39:09 +0000 (11:39 -0700)]
Merge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue
Jeff Kirsher says:
====================
100GbE Intel Wired LAN Driver Updates 2018-10-02
This series contains updates to ice driver only.
Anirudh expands the use of VSI handles across the rest of the driver,
which includes refactoring the code to correctly use VSI handles. After
a reset, ensure that all configurations for a VSI get re-applied before
moving on to rebuilding the next VSI.
Dave fixed the driver to check the current link state after reset to
ensure that the correct link state of a port is reported. Fixed an
issue where if the driver is unloaded when traffic is in progress,
errors are generated.
Preethi breaks up the IRQ tracker into a software and hardware IRQ
tracker, where the software IRQ tracker tracks only the PF's IRQ
requests and does not play any role in the VF initialization. The
hardware IRQ tracker represents the device's interrupt space and will be
looked up to see if the device has run our of interrupts when a
interrupt has to be allocated in the device for either PF or VF.
Md Fahad adds support for enabling/disabling RSS via ethtool.
Brett aligns the ice_reset_req enum values to the values that the
hardware understands. Also added initial support for dynamic interrupt
moderation in the ice driver.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add driver support for reading/configuring the 20G link speed via ethtool.
Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com> Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Add driver support for configuring/reading the 20G link speed.
Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com> Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Currently there is no support for dynamic interrupt moderation. This
patch adds some initial code to support this. The following changes
were made:
1. Currently we are using multiple members to store the interrupt
granularity (itr_gran_25/50/100/200). This is not necessary because
we can query the device to determine what the interrupt granularity
should be set to, done by a new function ice_get_itr_intrl_gran.
2. Added intrl to ice_q_vector structure to support interrupt rate
limiting.
3. Added the function ice_intrl_usecs_to_reg for converting to a value
in usecs that the device understands.
4. Added call to write to the GLINT_RATE register. Disable intrl by
default for now.
5. Changed rx/tx_itr_setting to itr_setting because having both seems
redundant because a ring is either Tx or Rx.
6. Initialize itr_setting for both Tx/Rx rings in ice_vsi_alloc_rings()
Signed-off-by: Brett Creeley <brett.creeley@intel.com> Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Align ice_reset_req enum values to hardware reset values
Currently the ice_reset_req enum values have to be translated into
a different set of values that the hardware understands for the same
reset types. Avoid this translation by aligning ice_reset_req enum
values to the ones that the hardware understands.
Also add and else if block to check for ICE_RESET_EMPR and put a dev_dbg
message in the else case.
Signed-off-by: Brett Creeley <brett.creeley@intel.com> Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch implements ethtool hook for enabling/disabling
RSS. While disabling RSS, the LUT should be cleared. And
the LUT should be reconfigured while enabling RSS.
Signed-off-by: Md Fahad Iqbal Polash <md.fahad.iqbal.polash@intel.com> Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Split irq_tracker into sw_irq_tracker and hw_irq_tracker
For the PF driver, when mapping interrupts to queues, we need to request
IRQs from the kernel and we also have to allocate interrupts from
the device.
Similarly, when the VF driver (iavf.ko) initializes, it requests the kernel
IRQs that it needs but it can't directly allocate interrupts in the device.
Instead, it sends a mailbox message to the ice driver, which then allocates
interrupts in the device on the VF driver's behalf.
Currently both these cases end up having to reserve entries in
pf->irq_tracker but irq_tracker itself is sized based on how many vectors
the PF driver needs. Under the right circumstances, the VF driver can fail
to get entries in irq_tracker, which will result in the VF driver failing
probe.
To fix this, sw_irq_tracker and hw_irq_tracker are introduced. The
sw_irq_tracker tracks only the PF's IRQ request and doesn't play any
role in VF init. hw_irq_tracker represents the device's interrupt space.
When interrupts have to be allocated in the device for either PF or VF,
hw_irq_tracker will be looked up to see if the device has run out of
interrupts.
Signed-off-by: Preethi Banala <preethi.banala@intel.com> Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Dave Ertman [Thu, 20 Sep 2018 00:23:15 +0000 (17:23 -0700)]
ice: Check for actual link state of port after reset
We are currently replaying the link state of a port after a reset, but
it is possible that the link state of a port can change during the reset
process. So check for the current link state of a port during the rebuild
process of a reset.
Signed-off-by: Dave Ertman <david.m.ertman@intel.com> Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Currently, switch filters get replayed after reset. In addition to
filters, other VSI attributes (like RSS configuration, Tx scheduler
configuration, etc.) also need to be replayed after reset.
Thus, instead of replaying based on functional blocks (i.e. replay
all filters for all VSIs, followed by RSS configuration replay for
all VSIs, and so on), it makes more sense to have the replay centered
around a VSI. In other words, replay all configurations for a VSI before
moving on to rebuilding the next VSI.
To that effect, this patch introduces a VSI replay framework in a new
function ice_vsi_replay_all. Currently it only replays switch filters,
but it will be expanded in the future to replay additional VSI attributes.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
A VSI handle is just a number the driver maintains to uniquely identify
a VSI. A VSI handle is backed by a VSI number in the hardware. When
interacting when the hardware, VSI handles are converted into VSI numbers.
In commit 0f9d5027a749 ("ice: Refactor VSI allocation, deletion and
rebuild flow"), VSI handles were introduced but it was used only
when creating and deleting VSIs. This patch is part one of two patches
that expands the use of VSI handles across the rest of the driver. Also
in this patch, certain parts of the code had to be refactored to correctly
use VSI handles.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In this patch series, update the ibmvnic driver to use driver-defined
queue limits instead of limits imposed by the Virtual I/O server
management partition. For some deviced, initial max queue size and
amount limits, despite their definition, can actually be exceeded if
the client driver requests it. With this in mind, define a private
ethtool flag that toggles the use of driver-defined limits. These
limits are currently more than what supported hardware will likely
allow, so the driver will attempt to get as close as possible to
the user request but may not fully succeed.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Thomas Falcon [Fri, 28 Sep 2018 23:38:26 +0000 (18:38 -0500)]
ibmvnic: Add ethtool private flag for driver-defined queue limits
When choosing channel amounts and ring sizes, the maximums in the
ibmvnic driver are defined by the virtual i/o server management
partition. Even though they are defined as maximums, the client
driver may in fact successfully request resources that exceed
these limits, which are mostly dependent on a user's hardware
With this in mind, provide an ethtool flag that when enabled will
allow the user to request resources limited by driver-defined
maximums instead of limits defined by the management partition.
The driver will try to honor the user's request but may not allowed
by the management partition. In this case, the driver requests
as close as it can get to the desired amount until it succeeds.
Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Thomas Falcon [Fri, 28 Sep 2018 23:38:25 +0000 (18:38 -0500)]
ibmvnic: Introduce driver limits for ring sizes
Introduce driver-defined maximums for queue ring sizes. Devices
available for IBM vNIC today will likely not allow this amount,
but this should give us some leeway for future devices that may
support larger ring sizes.
Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Thomas Falcon [Fri, 28 Sep 2018 23:38:24 +0000 (18:38 -0500)]
ibmvnic: Increase maximum queue size limit
Increase queue size limit to 16. Devices available for IBM vNIC today
will not allow this amount, but this should give us some leeway for
future devices that may support more RX or TX queues.
Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Rob Herring [Fri, 28 Sep 2018 22:04:59 +0000 (17:04 -0500)]
net: phy: Convert to using %pOFn instead of device_node.name
In preparation to remove the node name pointer from struct device_node,
convert printf users to use the %pOFn format specifier.
Cc: Andrew Lunn <andrew@lunn.ch> Cc: "David S. Miller" <davem@davemloft.net> Cc: netdev@vger.kernel.org Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: Rob Herring <robh@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Fri, 28 Sep 2018 17:28:44 +0000 (10:28 -0700)]
tcp/fq: move back to CLOCK_MONOTONIC
In the recent TCP/EDT patch series, I switched TCP and sch_fq
clocks from MONOTONIC to TAI, in order to meet the choice done
earlier for sch_etf packet scheduler.
But sure enough, this broke some setups were the TAI clock
jumps forward (by almost 50 year...), as reported
by Leonard Crestez.
If we want to converge later, we'll probably need to add
an skb field to differentiate the clock bases, or a socket option.
In the meantime, an UDP application will need to use CLOCK_MONOTONIC
base for its SCM_TXTIME timestamps if using fq packet scheduler.
Fixes: 72b0094f9182 ("tcp: switch tcp_clock_ns() to CLOCK_TAI base") Fixes: 142537e41923 ("net_sched: sch_fq: switch to CLOCK_TAI") Fixes: fd2bca2aa789 ("tcp: switch internal pacing timer to CLOCK_TAI") Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: Leonard Crestez <leonard.crestez@nxp.com> Tested-by: Leonard Crestez <leonard.crestez@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
selftests/tls: Fix recv(MSG_PEEK) & splice() test cases
TLS test cases splice_from_pipe, send_and_splice &
recv_peek_multiple_records expect to receive a given nummber of bytes
and then compare them against the number of bytes which were sent.
Therefore, system call recv() must not return before receiving the
requested number of bytes, otherwise the subsequent memcmp() fails.
This patch passes MSG_WAITALL flag to recv() so that it does not return
prematurely before requested number of bytes are copied to receive
buffer.
Signed-off-by: Vakul Garg <vakul.garg@nxp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
hv_netvsc: Fix rndis_per_packet_info internal field initialization
The RSC feature -- a bit field "internal" was added here with total
size unchanged:
struct rndis_per_packet_info {
u32 size;
u32 type:31;
u32 internal:1;
u32 ppi_offset;
};
On TX path, we put rndis msg into skb head room, which is not zeroed
before passing to us. We do not use the "internal" field in TX path,
but it may impact older hosts which use the entire 32 bits as "type".
To fix the bug, this patch sets the field "internal" to zero.
Fixes: c8e4eff4675f ("hv_netvsc: Add support for LRO/RSC in the vSwitch") Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Using mod_delayed_work() allows to simplify handling delayed work and
removes the need for the sync parameter in phy_trigger_machine().
Also introduce a helper phy_queue_state_machine() to encapsulate the
low-level delayed work calls. No functional change intended.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
====================
net: systemport: Turn on offloads by default
Up until now, we had added all the code necessary to turn on RX/TX
checksum offloads at runtime, but there is no reason why they have to be
disabled by default given that this gives a slight performance
improvement.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
net: systemport: Add software counters to track reallocations
When inserting the TSB, keep track of how many times we had to do it and
if there was a failure in doing so, this helps profile the driver for
possibly incorrect headroom settings.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
We can turn on the RX/TX checksum offloads by default and make sure that
those are properly reflected back to e.g: stacked devices such as VLAN
or DSA.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
net: systemport: Utilize bcm_sysport_set_features() during resume/open
During driver resume and open, the HW may have lost its context/state,
utilize bcm_sysport_set_features() to make sure we do restore the
correct set of features that were previously configured.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for unconditionally enabling TX and RX checksum offloads,
refactor bcm_sysport_set_features() a bit such that
__netdev_update_features() during register_netdev() can make sure that
features are correctly programmed during network device registration.
Since we can now be called during register_netdev() with clocks gated,
we need to temporarily turn them on/off in order to have a successful
register programming.
We also move the CRC forward setting read into
bcm_sysport_set_features() since priv->crc_fwd matters while turning on
RX checksum offload, that way we are guaranteed they are in sync in case
we ever add support for NETIF_F_RXFCS at some point in the future.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Cong Wang [Thu, 27 Sep 2018 20:42:19 +0000 (13:42 -0700)]
net_sched: fix a crash in tc_new_tfilter()
When tcf_block_find() fails, it already rollbacks the qdisc refcnt,
so its caller doesn't need to clean up this again. Avoid calling
qdisc_put() again by resetting qdisc to NULL for callers.
Reported-by: syzbot+37b8770e6d5a8220a039@syzkaller.appspotmail.com Fixes: e368fdb61d8e ("net: sched: use Qdisc rcu API instead of relying on rtnl lock") Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
This adds further netlink attribute validation:
* min/max/range validation
* validation through a custom function pointer
This is useful to
* reduce boilerplate code in command handling code, if attributes
are used commonly across different commands
* get more extended ACK error messages/attribute pointers
* ensure attributes are valid even when ignored
(though this might be a problem when converting existing code)
Changes since v1:
* split off validate_type from type and use that for min/max/range
and function; this is better because the range is limited to the
range of s16 and so things like "u16 with minimum value 1" couldn't
be expressed earlier
* add macros for this, e.g. NLA_POLICY_MIN(NLA_U16, 1) for the case
mentioned in the previous bullet
Using this pretty much in all places where applicable in nl80211
reduces the code size there by about 1.8KiB, with just a minimal
code increase in lib/nlattr.o.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Johannes Berg [Thu, 27 Sep 2018 09:28:36 +0000 (11:28 +0200)]
netlink: add validation function to policy
Add the ability to have an arbitrary validation function attached
to a netlink policy that doesn't already use the validation_data
pointer in another way.
This can be useful to validate for example the content of a binary
attribute, like in nl80211 the "(information) elements", which must
be valid streams of "u8 type, u8 length, u8 value[length]".
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Johannes Berg [Thu, 27 Sep 2018 09:28:35 +0000 (11:28 +0200)]
netlink: add attribute range validation to policy
Without further bloating the policy structs, we can overload
the `validation_data' pointer with a struct of s16 min, max
and use those to validate ranges in NLA_{U,S}{8,16,32,64}
attributes.
It may sound strange to validate NLA_U32 with a s16 max, but
in many cases NLA_U32 is used for enums etc. since there's no
size benefit in using a smaller attribute width anyway, due
to netlink attribute alignment; in cases like that it's still
useful, particularly when the attribute really transports an
enum value.
Doing so lets us remove quite a bit of validation code, if we
can be sure that these attributes aren't used by userspace in
places where they're ignored today.
To achieve all this, split the 'type' field and introduce a
new 'validation_type' field which indicates what further
validation (beyond the validation prescribed by the type of
the attribute) is done. This currently allows for no further
validation (the default), as well as min, max and range checks.
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
====================
Support of Flow Director in HNS3 Ethernet Driver for HiP08 Rev2 SoC
This patch-set adds the support of FD(Flow Director) in the HNS3 PF driver
for HiP08 Rev2(0x21) SoC of Hisilicon. FD can be used in filtering the flows
and deciding to drop the flow or forward it to paricular queue.
Configuration consists of rules with input keys and actions. The rules are
stored in TCAM.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:47 +0000 (12:46 +0100)]
net: hns3: Add support for enable/disable flow director
This patch adds switch for flow director with ethtool command
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:46 +0000 (12:46 +0100)]
net: hns3: Remove all flow director rules when unload hns3 driver
This patch removes all flow director rules when unload hns3 driver.
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:45 +0000 (12:46 +0100)]
net: hns3: Add reset handle for flow director
When doing reset, remove all entries in TCAM block, and keep flow
director rules list. After finishing reset, restore all entries.
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:44 +0000 (12:46 +0100)]
net: hns3: Add support for rule query of flow director
This patch adds support for querying rule number and rule details
by ethtool commands.
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:43 +0000 (12:46 +0100)]
net: hns3: Add support for rule add/delete for flow director
This patch adds support for add and delete rule by ethtool commands.
HNS3 driver supports several flow types, include ETHER_FLOW,
IP_USER_FLOW, TCP_V4_FLOW, UDP_V4_FLOW, SCTP_V4_FLOW, IPV6_USER_FLOW,
TCP_V6_FLOW, UDP_V6_FLOW and SCTP_V6_FLOW.
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:42 +0000 (12:46 +0100)]
net: hns3: Add input key and action config support for flow director
Each flow director rule consists of input key and action. The input key
is the condition for matching, includes tuples of L2/L3/L4 header.
Action is the behaviour when a packet matches with the input key, such
as drop the packet, or forward to a specified queue.
The input key is stored in the tcam blocks, Each bit of input key can
be masked.
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jian Shen [Mon, 1 Oct 2018 11:46:41 +0000 (12:46 +0100)]
net: hns3: Add flow director initialization
Flow director is a new feature supported by hardware with revision 0x21.
This patch adds flow direcor initialization for each PF. It queries flow
director mode and tcam resource from firmware, selects tuples used for
input key.
Signed-off-by: Jian Shen <shenjian15@huawei.com> Signed-off-by: Peng Li <lipeng321@huawei.com> Signed-off-by: Salil Mehta <salil.mehta@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:16 +0000 (23:04 +0200)]
net: phy: Replace phy driver features u32 with link_mode bitmap
This is one step in allowing phylib to make use of link_mode bitmaps,
instead of u32 for supported and advertised features. Convert the phy
drivers to use bitmaps to indicates the features they support.
Build bitmap equivalents of the u32 values at runtime, and have the
drivers point to the appropriate bitmap. These bitmaps are shared, and
we don't want a driver to modify them. So mark them __ro_after_init.
Within phylib, the features bitmap is currently turned back into a
u32. This will be removed once the whole of phylib, and the drivers
are converted to use bitmaps.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:15 +0000 (23:04 +0200)]
net: ethernet: xgbe: expand PHY_GBIT_FEAUTRES
The macro PHY_GBIT_FEAUTRES needs to change into a bitmap in order to
support link_modes. Remove its use from xgde by replacing it with its
definition.
Probably, the current behavior is wrong. It probably should be
ANDing not assigning.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:14 +0000 (23:04 +0200)]
net: phy: Add limkmode equivalents to some of the MII ethtool helpers
Add helpers which take a linkmode rather than a u32 ethtool for
advertising settings.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:13 +0000 (23:04 +0200)]
net: phy: Add helper for advertise to lcl value
Add a helper to convert the local advertising to an LCL capabilities,
which is then used to resolve pause flow control settings.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:12 +0000 (23:04 +0200)]
net: phy: Add helper to convert MII ADV register to a linkmode
The phy_mii_ioctl can be used to write a value into the MII_ADVERTISE
register in the PHY. Since this changes the state of the PHY, we need
to make the same change to phydev->advertising. Add a helper which can
convert the register value to a linkmode.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:10 +0000 (23:04 +0200)]
net: phy: Add phydev_warn()
Not all new style LINK_MODE bits can be converted into old style
SUPPORTED bits. We need to warn when such a conversion is attempted.
Add a helper for this.
Convert all pr_warn() calls to phydev_warn() where possible.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Andrew Lunn [Sat, 29 Sep 2018 21:04:09 +0000 (23:04 +0200)]
net: phy: Move linkmode helpers to somewhere public
phylink has some useful helpers to working with linkmode bitmaps.
Move them to there own header so other code can use them.
Signed-off-by: Andrew Lunn <andrew@lunn.ch> Acked-by: Florian Fainelli <f.fainelli@gmail.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski [Wed, 26 Sep 2018 22:35:31 +0000 (15:35 -0700)]
nfp: warn on experimental TLV types
Reserve two TLV types for feature development, and warn in the driver
if they ever leak into production.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by: Simon Horman <simon.horman@netronome.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Moritz Fischer [Thu, 27 Sep 2018 19:48:51 +0000 (12:48 -0700)]
net: nixge: Address compiler warnings when building for i386
Address compiler warning reported by kbuild autobuilders
when building for i386 as a result of dma_addr_t size on
different architectures.
warning: cast to pointer from integer of different size
[-Wint-to-pointer-cast]
Fixes: 7e8d5755be0e ("net: nixge: Add support for 64-bit platforms") Signed-off-by: Moritz Fischer <mdf@kernel.org> Cc: Arnd Bergmann <arnd@arndb.de> Signed-off-by: David S. Miller <davem@davemloft.net>
tcp: adjust rcv zerocopy hints based on frag sizes
When SKBs are coalesced, we can have SKBs with different
frag sizes. Some with PAGE_SIZE and some not with PAGE_SIZE.
Since recv_skip_hint is always set to the full SKB size,
it can overestimate the amount that should be read using
normal read for coalesced packets.
Change the recv_skip_hint so that it only includes the first
frags that are not of PAGE_SIZE.
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
tcp: set recv_skip_hint when tcp_inq is less than PAGE_SIZE
When we have less than PAGE_SIZE of data on receive queue,
we set recv_skip_hint to 0. Instead, set it to the actual
number of bytes available.
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
1) Make xfrmi_get_link_net() static to silence a sparse warning.
From Wei Yongjun.
2) Remove a unused esph pointer definition in esp_input().
From Haishuang Yan.
3) Allow the NIC driver to quietly refuse xfrm offload
in case it does not support it, the SA is created
without offload in this case.
From Shannon Nelson.
Please pull or let me know if there are problems.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller [Mon, 1 Oct 2018 22:49:17 +0000 (15:49 -0700)]
Merge tag 'mlx5e-updates-2018-10-01' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
mlx5e-updates-2018-10-01
This series includes updates to mlx5e ethernet netdevice driver:
From Or Gerlitz:
1) Support masks for l3/l4 filters in ethtool flow steering
2) Report checksum unnecessary also when the L3 checksum flag on the
cqe is set and there's no L4 header
3) Allow reporting of checksum unnecessary, using an ethtool private flag.
From Gavi Teitz and Or, VF representors netdevs performance improvements
4) Allow striding RQ in VF representor and bigger RQ size, ~3X performance improvement
5) Enable stateless offloads for VF representor, csum and TSO, 1.5X performance improvement
6) RSS Support for VF representors
6.1) Allow flow table destination fir VF representor steering rule.
6.2) Create RSS flow table per representor netdev
6.3) Expose mlx5e RSS ethtool to be used by representor netdevs
6.4) Enable multi-queue and RSS for VF representors, using mlx5e existing infrastructure
for managing a multi-queue RX RSS tables.
From Alaa Hleihel:
7) Cache the system image guid, The system image guid is a read-only field
Read this once and save it on the core device.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Yuchung Cheng [Mon, 1 Oct 2018 22:42:32 +0000 (15:42 -0700)]
tcp: start receiver buffer autotuning sooner
Previously receiver buffer auto-tuning starts after receiving
one advertised window amount of data. After the initial receiver
buffer was raised by patch a337531b942b ("tcp: up initial rmem to
128KB and SYN rwin to around 64KB"), the reciver buffer may take
too long to start raising. To address this issue, this patch lowers
the initial bytes expected to receive roughly the expected sender's
initial window.
Fixes: a337531b942b ("tcp: up initial rmem to 128KB and SYN rwin to around 64KB") Signed-off-by: Yuchung Cheng <ycheng@google.com> Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Neal Cardwell <ncardwell@google.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller [Mon, 1 Oct 2018 22:43:58 +0000 (15:43 -0700)]
Merge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next-queue
Jeff Kirsher says:
====================
100GbE Intel Wired LAN Driver Updates 2018-10-01
This series contains updates to ice driver only.
Anirudh provides several changes to "prep" the driver for upcoming
features. Specifically, the functions that are used for PF VSI/netdev
setup will also be used in SR-IOV support and to allow the reuse of
these functions, code needs to move.
Dave provides the only other change in the series, updates the driver to
protect the reset patch in its entirety. This is done by adding the
various bit checks to determine if a reset is scheduled/initiated and
whether it came from the software or firmware.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Dave Ertman [Thu, 20 Sep 2018 00:23:11 +0000 (17:23 -0700)]
ice: Change pf state behavior to protect reset path
Currently, there is no bit, or set of bits, that protect the entirety
of the reset path.
If the reset is originated by the driver, then the relevant
one of the following bits will be set when the reset is scheduled:
__ICE_PFR_REQ
__ICE_CORER_REQ
__ICE_GLOBR_REQ
This bit will not be cleared until after the rebuild has completed.
If the reset is originated by the FW, then the first the driver knows of
it will be the reception of the OICR interrupt. The __ICE_RESET_OICR_RECV
bit will be set in the interrupt handler. This will also be the indicator
in a SW originated reset that we have completed the pre-OICR tasks and
have informed the FW that a reset was requested.
To utilize these bits, change the function:
ice_is_reset_recovery_pending()
to be:
ice_is_reset_in_progress()
The new function will check all of the above bits in the pf->state and
will return a true if one or more of these bits are set.
Signed-off-by: Dave Ertman <david.m.ertman@intel.com> Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 7/7
This patch completes the code move out of ice_main.c
The following top level functions and related dependency functions) were
moved to ice_lib.c:
ice_vsi_setup
ice_vsi_cfg_tc
The following functions were made static again:
ice_vsi_setup_vector_base
ice_vsi_alloc_q_vectors
ice_vsi_get_qs
void ice_vsi_map_rings_to_vectors
ice_vsi_alloc_rings
ice_vsi_set_rss_params
ice_vsi_set_num_qs
ice_get_free_slot
ice_vsi_init
ice_vsi_alloc_arrays
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 6/7
This patch continues the code move out of ice_main.c
The following top level functions (and related dependency functions) were
moved to ice_lib.c:
ice_vsi_setup_vector_base
ice_vsi_alloc_q_vectors
ice_vsi_get_qs
The following functions were made static again:
ice_vsi_free_arrays
ice_vsi_clear_rings
Also, in this patch, the netdev and NAPI registration logic was de-coupled
from the VSI creation logic (ice_vsi_setup) as for SR-IOV, while we want to
create VF VSIs using ice_vsi_setup, we don't want to create netdevs.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 5/7
This patch continues the code move out of ice_main.c
The following top level functions (and related dependency functions) were
moved to ice_lib.c:
ice_vsi_clear
ice_vsi_close
ice_vsi_free_arrays
ice_vsi_map_rings_to_vectors
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 4/7
This patch continues the code move out of ice_main.c
The following top level functions (and related dependency functions) were
moved to ice_lib.c:
ice_vsi_alloc_rings
ice_vsi_set_rss_params
ice_vsi_set_num_qs
ice_get_free_slot
ice_vsi_init
ice_vsi_clear_rings
ice_vsi_alloc_arrays
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 3/7
This patch continues the code move out of ice_main.c
The following top level functions (and related dependency functions) were
moved to ice_lib.c:
ice_vsi_delete
ice_free_res
ice_get_res
ice_is_reset_recovery_pending
ice_vsi_put_qs
ice_vsi_dis_irq
ice_vsi_free_irq
ice_vsi_free_rx_rings
ice_vsi_free_tx_rings
ice_msix_clean_rings
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 2/7
This patch continues the code move out of ice_main.c
The following top level functions (and related dependency functions) were
moved to ice_lib.c:
ice_vsi_start_rx_rings
ice_vsi_stop_rx_rings
ice_vsi_stop_tx_rings
ice_vsi_cfg_rxqs
ice_vsi_cfg_txqs
ice_vsi_cfg_msix
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice: Move common functions out of ice_main.c part 1/7
The functions that are used for PF VSI/netdev setup will also be used
for SR-IOV support. To allow reuse of these functions, move these
functions out of ice_main.c to ice_common.c/ice_lib.c
This move is done across multiple patches. Each patch moves a few
functions and may have minor adjustments. For example, a function that was
previously static in ice_main.c will be made non-static temporarily in
its new location to allow the driver to build cleanly. These adjustments
will be removed in subsequent patches where more code is moved out of
ice_main.c
In this particular patch, the following functions were moved out of
ice_main.c:
int ice_add_mac_to_list
ice_free_fltr_list
ice_stat_update40
ice_stat_update32
ice_update_eth_stats
ice_vsi_add_vlan
ice_vsi_kill_vlan
ice_vsi_manage_vlan_insertion
ice_vsi_manage_vlan_stripping
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com> Tested-by: Andrew Bowers <andrewx.bowers@intel.com> Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The system image guid is a read-only field which is used by the TC
offloads code to determine if two mlx5 devices belong to the same
ASIC while adding flows.
Read this once and save it on the core device rather than querying each
time an offloaded flow is added.
Or Gerlitz [Sun, 1 Jul 2018 08:58:38 +0000 (08:58 +0000)]
net/mlx5e: Allow reporting of checksum unnecessary
Currently we practically never report checksum unnecessary, because
for all IP packets we take the checksum complete path.
Enable non-default runs with reprorting checksum unnecessary, using
an ethtool private flag. This can be useful for performance evals
and other explorations.
net/mlx5e: Enable multi-queue and RSS for VF representors
Increased the amount of channels the representors can open to be the
amount of CPUs. The default amount opened remains one.
Used the standard NIC netdev functions to:
* Set RSS params when building the representors' params.
* Setup an indirect TIR and RQT for the representors upon
initialization.
* Create a TTC flow table for the representors' indirect TIR (when
creating the TTC table, mlx5e_set_ttc_basic_params() is not called,
in order to avoid setting the inner_ttc param, which is not needed).
Added ethtool control to the representors for setting and querying
the amount of open channels. Additionally, included logic in the
representors' ethtool set channels handler which controls a
representor's vport rx rule, so that if there is one open channel
the rx rule steers traffic to the representor's direct TIR, whereas
if there is more than one channel, the rx rule steers traffic to the
new TTC flow table.
Gavi Teitz [Thu, 16 Aug 2018 21:28:53 +0000 (00:28 +0300)]
net/mlx5: E-Switch, Provide flow dest when creating vport rx rule
Currently the destination for the representor e-switch rx rule is
a TIR number. Towards changing that to potentially be a flow table,
as part of enabling RSS for representors, modify the signature of
the related e-switch API to get a flow destination.
Gavi Teitz [Thu, 16 Aug 2018 12:12:27 +0000 (15:12 +0300)]
net/mlx5e: Enable stateless offloads for VF representor netdevs
Enabled checksum and TSO offloads for the representors, in
order to increase their performance, which is required to
increase the performance of flows that cannot be offloaded.
Checksum offloads contribute to a general acceleration of all
traffic (to around 150%), whereas the TSO offload contributes
to a prominent acceleration of the representor's TX for traffic
flows with larger than MTU sized packets (to around 200%). This
is the usual case for TCP streams, as the PF, which serves as
the uplink representor, and the VF representors employ GRO before
forwarding the packets to the representor.
GRO was enabled implicitly for the representors beforehand, and
is explicitly enabled here to ensure that the representors preserve
the performance boost it provides (of around 200%) when working in
tandem with the TSO offload by the forwardee, which is the standard
case as both the PF and the VF representors employ HW TSO.
The impact of these changes can be seen in the following
measurements taken on a setup of a VM over a VF, connected
to OVS via the VF representor, to an external host:
Before current changes:
TCP Throughput [Gb/s]
External host to VM ~ 10.5
VM to external host ~ 23.5
With just checksum offloads enabled:
TCP Throughput [Gb/s]
External host to VM ~ 14.9
VM to external host ~ 28.5
With the TSO offload also enabled:
TCP Throughput [Gb/s]
External host to VM ~ 30.5
Gavi Teitz [Thu, 16 Aug 2018 11:25:24 +0000 (14:25 +0300)]
net/mlx5e: Change VF representors' RQ type
The representors' RQ size was not large enough for them to achieve
high enough performance, and therefore needed to be enlarged, while
suffering a minimum hit to its memory usage. To achieve this the
representors RQ size was increased, and its type was changed to be a
striding RQ if it is supported.
Towards that goal the following changes were made:
* Extracted the sequence for setting the standard netdev's RQ parmas
into a function
* Replaced the sequence for setting the representor's RQ params with
the standard sequence
The impact of this change can be seen in the following measurements
taken on a setup of a VM over a VF, connected to OVS via the VF
representor, to an external host:
Before current change:
TCP Throughput [Gb/s]
VM to external host ~ 7.2
With the current change (measured with a striding RQ):
TCP Throughput [Gb/s]
VM to external host ~ 23.5
Each representor now consumes 2 [MB] of memory for its packet
buffers.