All implementations of these two methods are dummies that always
return -EINVAL.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Delete the doubled word "be" in a comment.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Cc: netdev@vger.kernel.org
Cc: linux-rdma@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Delete the doubled word "the" in a comment.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch add MDIX configuration ability for AR9331 and AR8035. Theoretically
it should work on other Atheros PHYs, but I was able to test only this
two.
Since I have no certified reference HW able to detect or configure MDIX, this
functionality was confirmed by oscilloscope.
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Michael Walle says:
====================
net: enetc: remove bootloader dependency
These patches were picked from the following series:
https://lore.kernel.org/netdev/1567779344-30965-1-git-send-email-claudiu.manoil@nxp.com/
They have never been resent. I've picked them up, addressed Andrews
comments, fixed some more bugs and asked Claudiu if I can keep their SOB
tags; he agreed. I've tested this on our board which happens to have a
bootloader which doesn't do the enetc setup in all cases. Though, only
SGMII mode was tested.
changes since v6:
- dropped _LPA_ infix for USXGMII constants
changes since v5:
- fixed pcs->autoneg_complete and pcs->link assignment. Thanks Vladimir.
changes since v4:
- moved (and renamed) the USXGMII constants to include/uapi/linux/mdio.h.
Suggested by Russell King.
changes since v3:
- rebased to latest net-next where devm_mdiobus_free() was removed.
replace it by mdiobus_free(). The internal MDIO bus is optional, if
there is any error, we try to run with the bootloader default PCS
settings, thus in the error case, we need to free the mdiobus.
changes since v2:
- removed SOBs from "net: enetc: Initialize SerDes for SGMII and USXGMII
protocols" because almost everything has changed.
- get a phy_device for the internal PCS PHY so we can use the phy_
functions instead of raw mdiobus writes
- reuse macros already defined in fsl_mdio.h, move missing bits from
felix to fsl_mdio.h, because they share the same PCS PHY building
block
- added 2500BaseX mode (based on felix init routine)
- changed xgmii mode to usxgmii mode, because it is actually USXGMII and
felix does the same.
- fixed devad, which is 0x1f (MMD_VEND2)
changes since v1:
- mdiobus id is '"imdio-%s", dev_name(dev)' because the plain dev_name()
is used by the emdio.
- use mdiobus_write() instead of imdio->write(imdio, ..), since this is
already a full featured mdiobus
- set phy_mask to ~0 to avoid scanning the bus
- use phy_interface_mode_is_rgmii(phy_mode) to also include the RGMII
modes with pad delays.
- move enetc_imdio_init() to enetc_pf.c, there shouldn't be any other
users, should it?
- renamed serdes to SerDes
- printing the error code of mdiobus_register() in the error path
- call mdiobus_unregister() on _remove()
- call devm_mdiobus_free() if mdiobus_register() fails, since an
error is not fatal
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Use DT information rather than in-band information from bootloader to
set up MAC for XGMII. For RGMII use the DT indication in addition to
RGMII defaults in hardware.
However, this implies that PHY connection information needs to be
extracted before netdevice creation, when the ENETC Port MAC is
being configured.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: Michael Walle <michael@walle.cc>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ENETC has ethernet MACs capable of SGMII, 2500BaseX and USXGMII. But in
order to use these protocols some SerDes configurations need to be
performed. The SerDes is configurable via an internal PCS PHY which is
connected to an internal MDIO bus at address 0.
This patch basically removes the dependency on bootloader regarding
SerDes initialization.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that there are USXGMII constants available, drop the old definitions
and reuse the generic ones.
Signed-off-by: Michael Walle <michael@walle.cc>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The constants are taken from the USXGMII Singleport Copper Interface
specification. The naming are based on the SGMII ones, but with an MDIO_
prefix.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Mark Starovoytov says:
====================
net: atlantic: add support for FW 4.x
This patch set adds support for FW 4.x, which is about to get into the
production for some products.
4.x is mostly compatible with 3.x, save for soft reset, which requires
the acquisition of 2 additional semaphores.
Other differences (e.g. absence of PTP support) are handled via
capabilities.
Note: 4.x targets specific products only. 3.x is still the main firmware
branch, which should be used by most users (at least for now).
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for FW 4.x, which is about to get into the
production for some products.
4.x is mostly compatible with 3.x, save for soft reset, which requires
the acquisition of 2 additional semaphores.
Other differences (e.g. absence of PTP support) are handled via
capabilities.
Note: 4.x targets specific products only. 3.x is still the main firmware
branch, which should be used by most users (at least for now).
Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com>
Signed-off-by: Mark Starovoytov <mstarovoitov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch aligns the return value of hw_atl_utils_ver_match function with
its name.
Change the return type to bool, because it's better aligned with the actual
usage. Return true when the version matches, false otherwise.
Signed-off-by: Mark Starovoytov <mstarovoitov@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following the removal of an unused variable assignment (remove
unused variable 'pm_csr') the associated register read can also go,
as the read also occurs in the subsequent et1310_in_phy_coma()
call.
Signed-off-by: Mark Einon <mark.einon@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Gcc report warning as follows:
drivers/net/ethernet/agere/et131x.c:953:6: warning:
variable 'pm_csr' set but not used [-Wunused-but-set-variable]
953 | u32 pm_csr;
| ^~~~~~
drivers/net/ethernet/agere/et131x.c:1002:6⚠️
variable 'pm_csr' set but not used [-Wunused-but-set-variable]
1002 | u32 pm_csr;
| ^~~~~~
drivers/net/ethernet/agere/et131x.c:3446:8: warning:
variable 'pm_csr' set but not used [-Wunused-but-set-variable]
3446 | u32 pm_csr;
| ^~~~~~
After commit 38df6492eb ("et131x: Add PCIe gigabit ethernet driver
et131x to drivers/net"), 'pm_csr' is never used in these functions,
so removing it to avoid build warning.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com>
Acked-by: Mark Einon <mark.einon@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Gcc report warning as follows:
drivers/net/ethernet/brocade/bna/bfa_ioc.c:1538:6: warning:
variable 't' set but not used [-Wunused-but-set-variable]
1538 | u32 t;
| ^
After commit c107ba171f ("bna: Firmware Patch Simplification"),
't' is never used, so removing it to avoid build warning.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The fact that NETIF_F_HW_TC is not set should be a sufficient
indication to the user that TC offloads are not supported.
No need to bother users of older firmware versions with
pointless warnings on every boot.
Also, since the support is optional, bnxt_init_tc() should not
return an error in case FW is old, similarly to how error
is not returned when CONFIG_BNXT_FLOWER_OFFLOAD is not set.
With that we can add an error message to the caller, to warn
about actual unexpected failures.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes:
1) Fix build break when CONFIG_XPS is not set
2) Fix missing switch_id for representors
Updates:
1) IPsec XFRM RX offloads from Raed and Huy.
- Added IPSec RX steering flow tables to NIC RX
- Refactoring of the existing FPGA IPSec, to add support
for ConnectX IPsec.
- RX data path handling for IPSec traffic
- Synchronize offloading device ESN with xfrm received SN
2) Parav allows E-Switch to siwtch to switchdev mode directly without
the need to go through legacy mode first.
3) From Tariq, Misc updates including:
3.1) indirect calls for RX and XDP handlers
3.2) Make MLX5_EN_TLS non-prompt as it should always be enabled when
TLS and MLX5_EN are selected.
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEGhZs6bAKwk/OTgTpSD+KveBX+j4FAl8Q5J0ACgkQSD+KveBX
+j5zhggAm/8ILhtG04BBKeQGay+m4CCg9qK7BrIavU3ta2t+DQdAxE+XmmHl+W2F
DfL5sR0AiV8z8v6OF6Yjrh49Ys6k7LFh6msFP2vyVkUC6t02zRv7WYMlZn44Igqb
Jg8n4Q806y5g2RJRmV/QFz9nOq8jxL/CXxA7eLCMiRSQKHl3LQ3TXbvvLJRY6ab2
aZT9fhi6lJWhe7Rii932oUM+USikmilFgB0tBoSgVQ9fxa+cNTuMb2y/IKHQo5pi
O9OUUKbPgYy3+xah+FCPLMx4izyv8F36XA7z6fGhtsM74pmFvC5e2eWOoqriWeBO
8SL2m2+FSUnuoI6S2wKsBl5dePdezQ==
=p788
-----END PGP SIGNATURE-----
Merge tag 'mlx5-updates-2020-07-16' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
mlx5-updates-2020-07-16
Fixes:
1) Fix build break when CONFIG_XPS is not set
2) Fix missing switch_id for representors
Updates:
1) IPsec XFRM RX offloads from Raed and Huy.
- Added IPSec RX steering flow tables to NIC RX
- Refactoring of the existing FPGA IPSec, to add support
for ConnectX IPsec.
- RX data path handling for IPSec traffic
- Synchronize offloading device ESN with xfrm received SN
2) Parav allows E-Switch to siwtch to switchdev mode directly without
the need to go through legacy mode first.
3) From Tariq, Misc updates including:
3.1) indirect calls for RX and XDP handlers
3.2) Make MLX5_EN_TLS non-prompt as it should always be enabled when
TLS and MLX5_EN are selected.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Avoid a memset after a call to 'dma_alloc_coherent()'.
This is useless since
commit 518a2f1925 ("dma-mapping: zero memory returned from dma_alloc_*")
Replace a kmalloc+memset with a corresponding kzalloc.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
Priyaranjan Jha says:
====================
tcp: improve handling of DSACK covering multiple segments
Currently, while processing DSACK, we assume DSACK covers only one
segment. This leads to significant underestimation of no. of duplicate
segments with LRO/GRO. Also, the existing SNMP counters, TCPDSACKRecv
and TCPDSACKOfoRecv, make similar assumption for DSACK, which makes them
unusable for estimating spurious retransmit rates.
This patch series fixes the segment accounting with DSACK, by estimating
number of duplicate segments based on: (DSACKed sequence range) / MSS.
It also introduces a new SNMP counter, TCPDSACKRecvSegs, which tracks
the estimated number of duplicate segments.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
There are two existing SNMP counters, TCPDSACKRecv and TCPDSACKOfoRecv,
which are incremented depending on whether the DSACKed range is below
the cumulative ACK sequence number or not. Unfortunately, these both
implicitly assume each DSACK covers only one segment. This makes these
counters unusable for estimating spurious retransmit rates,
or real/non-spurious loss rate.
This patch introduces a new SNMP counter, TCPDSACKRecvSegs, which tracks
the estimated number of duplicate segments based on:
(DSACKed sequence range) / MSS. This counter is usable for estimating
spurious retransmit rates, or real/non-spurious loss rate.
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, while processing DSACK, we assume DSACK covers only one
segment. This leads to significant underestimation of DSACKs with
LRO/GRO. This patch fixes segment accounting with DSACK by estimating
segment count from DSACK sequence range / MSS.
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yousuk Seung <ysseung@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
since commit d47a721520 ("mptcp: fix race in subflow_data_ready()"), it
is possible to observe a regression in MP_JOIN kselftests. For sockets in
TCP_CLOSE state, it's not sufficient to just wake up the main socket: we
also need to ensure that received data are made available to the reader.
Silence the WARN_ON_ONCE() in these cases: it preserves the syzkaller fix
and restores kselftests when they are ran as follows:
# while true; do
> make KBUILD_OUTPUT=/tmp/kselftest TARGETS=net/mptcp kselftest
> done
Reported-by: Florian Westphal <fw@strlen.de>
Fixes: d47a721520 ("mptcp: fix race in subflow_data_ready()")
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/47
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Reviewed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bjørn Mork says:
====================
usbnet: multicast filter support for cdc ncm devices
This revives a 2 year old patch set from Miguel Rodríguez
Pérez, which appears to have been lost somewhere along the
way. I've based it on the last version I found (v4), and
added one patch which I believe must have been missing in
the original.
I kept Oliver's ack on one of the patches, since both the patch and
the motivation still is the same. Hope this is OK..
Thanks to the anonymous user <wxcafe@wxcafe.net> for bringing up this
problem in https://bugs.debian.org/965074
This is only build and load tested by me. I don't have any device
where I can test the actual functionality.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We set set_rx_mode to usbnet_cdc_update_filter provided
by cdc_ether that simply admits all multicast traffic
if there is more than one multicast filter configured.
Signed-off-by: Miguel Rodríguez Pérez <miguel@det.uvigo.gal>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
The cdc_ncm driver overrides the net_device_ops structure used by usbnet
to be able to hook into .ndo_change_mtu. However, the structure was
missing the .ndo_set_rx_mode field, preventing the driver from
hooking into usbnet's set_rx_mode. This patch adds the missing callback to
usbnet_set_rx_mode in net_device_ops.
Signed-off-by: Miguel Rodríguez Pérez <miguel@det.uvigo.gal>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
This function can be reused by other usbnet minidrivers.
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
This makes the function available to other drivers, like cdc_ncm.
Signed-off-by: Miguel Rodríguez Pérez <miguel@det.uvigo.gal>
Acked-by: Oliver Neukum <oneukum@suse.com>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
usbnet_cdc_update_filter was getting the interface number from the
usb_interface struct in cdc_state->control. However, cdc_ncm does
not initialize that structure in its bind function, but uses
cdc_ncm_ctx instead. Getting intf directly from struct usbnet solves
the problem.
Signed-off-by: Miguel Rodríguez Pérez <miguel@det.uvigo.gal>
Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch reorders the masks array every 4 seconds based on their
usage count. This greatly reduces the masks per packet hit, and
hence the overall performance. Especially in the OVS/OVN case for
OpenShift.
Here are some results from the OVS/OVN OpenShift test, which use
8 pods, each pod having 512 uperf connections, each connection
sends a 64-byte request and gets a 1024-byte response (TCP).
All uperf clients are on 1 worker node while all uperf servers are
on the other worker node.
Kernel without this patch : 7.71 Gbps
Kernel with this patch applied: 14.52 Gbps
We also run some tests to verify the rebalance activity does not
lower the flow insertion rate, which does not.
Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
Tested-by: Andrew Theurer <atheurer@redhat.com>
Reviewed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some Cotsworks SFF have invalid data in the first few bytes of the
module EEPROM. This results in these modules not being detected as
valid modules.
Address this by poking the correct EEPROM values into the module
EEPROM when the model/PN match and the existing module EEPROM contents
are not correct.
Signed-off-by: Chris Healy <cphealy@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
At the time of introduction, in commit bdeced75b1 ("net: dsa: felix:
Add PCS operations for PHYLINK"), support for the Lynx PCS inside Felix
was relying, for USXGMII support, on the fact that get_phy_device() is
able to parse the Lynx PCS "device-in-package" registers for this C45
MDIO device and identify it correctly.
However, this was actually working somewhat by mistake (in the sense
that, even though it was detected, it was detected for the wrong
reasons).
The get_phy_c45_ids() function works by iterating through all MMDs
starting from 1 (MDIO_MMD_PMAPMD) and stops at the first one which
returns a non-zero value in the "device-in-package" register pair,
proceeding to see what that non-zero value is.
For the Felix PCS, the first MMD (1, for the PMA/PMD) returns a non-zero
value of 0xffffffff in the "device-in-package" registers. There is a
code branch which is supposed to treat this case and flag it as wrong,
and normally, this would have caught my attention when adding initial
support for this PCS:
if ((devs_in_pkg & 0x1fffffff) == 0x1fffffff) {
/* If mostly Fs, there is no device there, then let's probe
* MMD 0, as some 10G PHYs have zero Devices In package,
* e.g. Cortina CS4315/CS4340 PHY.
*/
However, this code never actually kicked in, it seems, because this
snippet from get_phy_c45_devs_in_pkg() was basically sabotaging itself,
by returning 0xfffffffe instead of 0xffffffff:
/* Bit 0 doesn't represent a device, it indicates c22 regs presence */
*devices_in_package &= ~BIT(0);
Then the rest of the code just carried on thinking "ok, MMD 1 (PMA/PMD)
says that there are 31 devices in that package, each having a device id
of ffff:ffff, that's perfectly fine, let's go ahead and probe this PHY
device".
But after cleanup commit 320ed3bf90 ("net: phy: split
devices_in_package"), this got "fixed", and now devs_in_pkg is no longer
0xfffffffe, but 0xffffffff. So now, get_phy_device is returning -ENODEV
for the Lynx PCS, because the semantics have remained mostly unchanged:
the loop stops at the first MMD that returns a non-zero value, and that
is MMD 1.
But the Lynx PCS is simply a clause 37 PCS which implements the required
MAC-side functionality for USXGMII (when operated in C45 mode, which is
where C45 devices-in-package detection is relevant to). Of course it
will fail the PMD/PMA test (MMD 1), since it is not a PHY. But it does
implement detection for MDIO_MMD_PCS (3):
- MDIO_DEVS1=0x008a, MDIO_DEVS2=0x0000,
- MDIO_DEVID1=0x0083, MDIO_DEVID2=0xe400
Let get_phy_c45_ids() continue searching for valid MMDs, and don't
assume that every phy_device has a PMA/PMD MMD implemented.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Petr Machata says:
====================
net: sched: Do not drop root lock in tcf_qevent_handle()
Mirred currently does not mix well with blocks executed after the qdisc
root lock is taken. This includes classification blocks (such as in PRIO,
ETS, DRR qdiscs) and qevents. The locking caused by the packet mirrored by
mirred can cause deadlocks: either when the thread of execution attempts to
take the lock a second time, or when two threads end up waiting on each
other's locks.
The qevent patchset attempted to not introduce further badness of this
sort, and dropped the lock before executing the qevent block. However this
lead to too little locking and races between qdisc configuration and packet
enqueue in the RED qdisc.
Before the deadlock issues are solved in a way that can be applied across
many qdiscs reasonably easily, do for qevents what is done for the
classification blocks and just keep holding the root lock.
That is done in patch #1. Patch #2 then drops the now unnecessary root_lock
argument from Qdisc_ops.enqueue.
====================
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Mirred currently does not mix well with blocks executed after the qdisc
root lock is taken. This includes classification blocks (such as in PRIO,
ETS, DRR qdiscs) and qevents. The locking caused by the packet mirrored by
mirred can cause deadlocks: either when the thread of execution attempts to
take the lock a second time, or when two threads end up waiting on each
other's locks.
The qevent patchset attempted to not introduce further badness of this
sort, and dropped the lock before executing the qevent block. However this
lead to too little locking and races between qdisc configuration and packet
enqueue in the RED qdisc.
Before the deadlock issues are solved in a way that can be applied across
many qdiscs reasonably easily, do for qevents what is done for the
classification blocks and just keep holding the root lock.
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The 128 bits ct_label field is matched using a 32 bit hardware register.
As such, only the lower 32 bits of ct_label field are offloaded. Change
this logic to support setting and matching higher bits too.
Map the 128 bits data to a unique 32 bits ID. Matching is done as exact
match of the mapping ID of key & mask.
Signed-off-by: Eli Britstein <elibr@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Maor Dickman <maord@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
UMR WQEs are posted in bulks, and HW is notified once per a bulk.
Reduce the number of completions by requesting such only for
the last WQE of the bulk.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Use INDIRECT_CALL_2() helper to avoid the cost of the indirect call
when/if CONFIG_RETPOLINE=y.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Use INDIRECT_CALL_2() helper to avoid the cost of the indirect call
when/if CONFIG_RETPOLINE=y.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Synchronize offloading device ESN with xfrm received SN
by updating an existing IPsec HW context with the new SN.
Signed-off-by: Raed Salem <raeds@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
On receive flow inspect received packets for IPsec offload indication
using the cqe, for IPsec offloaded packets propagate offload status
and stack handle to stack for further processing.
Supported statuses:
- Offload ok.
- Authentication failure.
- Bad trailer indication.
Connect-X IPsec does not use mlx5e_ipsec_handle_rx_cqe.
For RX only offload, we see the BW gain. Below is the iperf3
performance report on two server of 24 cores Intel(R) Xeon(R)
CPU E5-2620 v3 @ 2.40GHz with ConnectX6-DX.
We use one thread per IPsec tunnel.
---------------------------------------------------------------------
Mode | Num tunnel | BW | Send CPU util | Recv CPU util
| | (Gbps) | (Average %) | (Average %)
---------------------------------------------------------------------
Cryto offload | 1 | 4.6 | 4.2 | 14.5
---------------------------------------------------------------------
Cryto offload | 24 | 38 | 73 | 63
---------------------------------------------------------------------
Non-offload | 1 | 4 | 4 | 13
---------------------------------------------------------------------
Non-offload | 24 | 23 | 52 | 67
Signed-off-by: Raed Salem <raeds@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Introduce decrypt FT, the RX error FT and the default rules.
The IPsec RX decrypt flow table is pointed by the TTC
(Traffic Type Classifier) ESP steering rules.
The decrypt flow table has two flow groups. The first flow group
keeps the decrypt steering rule programmed via the "ip xfrm s" interface.
The second flow group has a default rule to forward all non-offloaded
ESP packet to the TTC ESP default RSS TIR.
The RX error flow table is the destination of the decrypt steering rules
in the IPsec RX decrypt flow table. It has a fixed rule with single
copy action that copies ipsec_syndrome to metadata_regB[0:6]. The IPsec
syndrome is used to filter out non-ipsec packet and to return the IPsec
crypto offload status in Rx flow. The destination of RX error flow table
is the TTC ESP default RSS TIR.
All the FTs (decrypt FT and error FT) are created only when IPsec SAs
are added. If there is no IPsec SAs, the FTs are removed.
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
This patch adds support for Connect-X IPsec crypto offload
by implementing the IPsec acceleration layer needed routines,
which delegates IPsec offloads to Connect-X routines.
In Connect-X IPsec, a Security Association (SA) is added or deleted
via allocating a HW context of an encryption/decryption key and
a HW context of a matching SA (IPsec object).
The Security Policy (SP) is added or deleted by creating matching Tx/Rx
steering rules whith an action of encryption/decryption respectively,
executed using the previously allocated SA HW context.
When new xfrm state (SA) is added:
- Use a separate crypto key HW context.
- Create a separate IPsec context in HW to inlcude the SA properties:
- aes-gcm salt.
- ICV properties (ICV length, implicit IV).
- on supported devices also update ESN.
- associate the allocated crypto key with this IPsec context.
Introduce a new compilation flag MLX5_IPSEC for it.
Downstream patches will implement the Rx,Tx steering
and will add the update esn.
Signed-off-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
This to set the base for downstream patches to support
the new IPsec implementation of the Connect-X family.
Following modifications made:
- Remove accel layer dependency from MLX5_FPGA_IPSEC.
- Introduce accel_ipsec_ops, each IPsec device will
have to support these ops.
Signed-off-by: Raed Salem <raeds@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Currently only ECPF allows enabling eswitch when SR-IOV is disabled.
Enable PF also to enable eswitch when SR-IOV is disabled.
Load VF vports when eswitch is already enabled.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Christoph Hellwig