Monitor LAG events via the NETDEV_CHANGEUPPER/NETDEV_CHANGELOWERSTATE
notifiers to maintain a list of offloadable groups. Sync these groups with
HW via a delayed workqueue to prevent excessive re-configuration. When the
workqueue is triggered it may generate multiple control messages for
different groups. These messages are linked via a batch ID and flags to
indicate a new batch and the end of a batch.
Update private data in each repr to track their LAG lower state flags. The
state of a repr is used to determine the active netdevs that can be
offloaded. For example, in active-backup mode, we only offload the netdev
currently active.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
LAG upper event notifiers contain the tx type used by the LAG device.
Extend this to also include the hash policy used for tx types that
utilize hashing.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a bitmap to each flower repr to track its state if it is enslaved by a
bond. This LAG state may be different to the port state - for example, the
port may be up but LAG state may be down due to the selection in an
active/backup bond.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Check if the fw contains the _abi_flower_balance_sync_enable symbol. If it
does then write a 1 to this indicating that the driver is willing to
receive NIC to kernel LAG related control messages.
If the write is successful, update the list of extra features supported by
the fw and add a stub to accept LAG cmsgs.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the guest network adapter is not configured with DeviceNaming
enabled on the host, then the query for friendly name will return
success but with a zero length name. Which then leads to a garbage value
(stack contents) for ifalias.
Fix is simple, just don't set name if host doesn't return it.
Fixes: 0fe554a46a ("hv_netvsc: propogate Hyper-V friendly name into interface alias")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 624dbf55a3 ("driver/net: enic: Try DMA 64 first, then
failover to DMA") DMA mask was changed from 40 bits to 64 bits.
Hardware actually supports only 47 bits.
Fixes: 624dbf55a3 ("driver/net: enic: Try DMA 64 first, then failover to DMA")
Signed-off-by: Govindarajulu Varadarajan <gvaradar@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The PPPIOCDETACH ioctl effectively tries to "close" the given ppp file
before f_count has reached 0, which is fundamentally a bad idea. It
does check 'f_count < 2', which excludes concurrent operations on the
file since they would only be possible with a shared fd table, in which
case each fdget() would take a file reference. However, it fails to
account for the fact that even with 'f_count == 1' the file can still be
linked into epoll instances. As reported by syzbot, this can trivially
be used to cause a use-after-free.
Yet, the only known user of PPPIOCDETACH is pppd versions older than
ppp-2.4.2, which was released almost 15 years ago (November 2003).
Also, PPPIOCDETACH apparently stopped working reliably at around the
same time, when the f_count check was added to the kernel, e.g. see
https://lkml.org/lkml/2002/12/31/83. Also, the current 'f_count < 2'
check makes PPPIOCDETACH only work in single-threaded applications; it
always fails if called from a multithreaded application.
All pppd versions released in the last 15 years just close() the file
descriptor instead.
Therefore, instead of hacking around this bug by exporting epoll
internals to modules, and probably missing other related bugs, just
remove the PPPIOCDETACH ioctl and see if anyone actually notices. Leave
a stub in place that prints a one-time warning and returns EINVAL.
Reported-by: syzbot+16363c99d4134717c05b@syzkaller.appspotmail.com
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
Reviewed-by: Guillaume Nault <g.nault@alphalink.fr>
Tested-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
Alexei Starovoitov says:
====================
pull-request: bpf-next 2018-05-24
The following pull-request contains BPF updates for your *net-next* tree.
The main changes are:
1) Björn Töpel cleans up AF_XDP (removes rebind, explicit cache alignment from uapi, etc).
2) David Ahern adds mtu checks to bpf_ipv{4,6}_fib_lookup() helpers.
3) Jesper Dangaard Brouer adds bulking support to ndo_xdp_xmit.
4) Jiong Wang adds support for indirect and arithmetic shifts to NFP
5) Martin KaFai Lau cleans up BTF uapi and makes the btf_header extensible.
6) Mathieu Xhonneux adds an End.BPF action to seg6local with BPF helpers allowing
to edit/grow/shrink a SRH and apply on a packet generic SRv6 actions.
7) Sandipan Das adds support for bpf2bpf function calls in ppc64 JIT.
8) Yonghong Song adds BPF_TASK_FD_QUERY command for introspection of tracing events.
9) other misc fixes from Gustavo A. R. Silva, Sirio Balmelli, John Fastabend, and Magnus Karlsson
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce a recovery hard reset to handle reset failure as a result of
change of device context following a transport event, such as a
backing device failover or partition migration. These operations reset
the device context to its initial state. If this occurs during a reset,
any initialization commands are likely to fail with an invalid state
error as backing device firmware requests reinitialization.
When this happens, make one more attempt by performing a hard reset,
which frees any resources currently allocated and performs device
initialization. If a transport event occurs during a device reset, a
flag is set which will trigger a new hard reset following the
completionof the current reset event.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set device resetting state at the earliest possible point: as soon as a
reset is successfully scheduled. The reset state is toggled off when
all resets have been processed to completion.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of having one initialization routine for all cases, create
a separate, simpler function for standard initialization, such as during
device probe. Use the original initialization function to handle
device reset scenarios. The goal of this patch is to avoid having
a single, cluttered init function to handle all possible
scenarios.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If setting the link state is not successful, print a warning
with the resulting return code and return it to be handled
by the caller.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If device init is interrupted by a failover, set the init return
code so that it can be checked and handled appropriately by the
init routine.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Check whether CRQ command is successful before awaiting a response
from the management partition. If the command was not successful, the
driver may hang waiting for a response that will never come.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce an "active" state for a IBM vNIC Command-Response Queue. A CRQ
is considered active once it has initialized or linked with its partner by
sending an initialization request and getting a successful response back
from the management partition. Until this has happened, do not allow CRQ
commands to be sent other than the initialization request.
This change will avoid a protocol error in case of a device transport
event occurring during a initialization. When the driver receives a
transport event notification indicating that the backing hardware
has changed and needs reinitialization, any further commands other
than the initialization handshake with the VIOS management partition
will result in an invalid state error. Instead of sending a command
that will be returned with an error, print a warning and return an
error that will be handled by the caller.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set adapter NAPI state as disabled if they are removed. This will allow
them to be enabled again if reallocated in case of a hard reset.
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
t4_prep_fw doesn't check for card_fw pointer before store the read data,
which could lead to a NULL pointer dereference if kvzalloc failed.
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch change the API for ndo_xdp_xmit to support bulking
xdp_frames.
When kernel is compiled with CONFIG_RETPOLINE, XDP sees a huge slowdown.
Most of the slowdown is caused by DMA API indirect function calls, but
also the net_device->ndo_xdp_xmit() call.
Benchmarked patch with CONFIG_RETPOLINE, using xdp_redirect_map with
single flow/core test (CPU E5-1650 v4 @ 3.60GHz), showed
performance improved:
for driver ixgbe: 6,042,682 pps -> 6,853,768 pps = +811,086 pps
for driver i40e : 6,187,169 pps -> 6,724,519 pps = +537,350 pps
With frames avail as a bulk inside the driver ndo_xdp_xmit call,
further optimizations are possible, like bulk DMA-mapping for TX.
Testing without CONFIG_RETPOLINE show the same performance for
physical NIC drivers.
The virtual NIC driver tun sees a huge performance boost, as it can
avoid doing per frame producer locking, but instead amortize the
locking cost over the bulk.
V2: Fix compile errors reported by kbuild test robot <lkp@intel.com>
V4: Isolated ndo, driver changes and callers.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Sandbox QP Commands are retired in the order they are sent. Outstanding
commands are stored in a linked-list in the order they appear. Once a
response is received and the callback gets called, we pull the first
element off the pending list, assuming they correspond.
Sending a message and adding it to the pending list is not done atomically,
hence there is an opportunity for a race between concurrent requests.
Bind both send and add under a critical section.
Fixes: bebb23e6cb ("net/mlx5: Accel, Add IPSec acceleration interface")
Signed-off-by: Yossi Kuperman <yossiku@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
When RXFCS feature is enabled, the HW do not strip the FCS data,
however it is not present in the checksum calculated by the HW.
Fix that by manually calculating the FCS checksum and adding it to the SKB
checksum field.
Add helper function to find the FCS data for all SKB forms (linear,
one fragment or more).
Fixes: 102722fc68 ("net/mlx5e: Add support for RXFCS feature flag")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Add dcbnl's set/get buffer configuration callback that allows user to
set/get buffer size configuration and priority to buffer mapping.
By default, firmware controls receive buffer configuration and priority
of buffer mapping based on the changes in pfc settings. When set buffer
call back is triggered, the buffer configuration changes to manual mode.
The manual mode means mlx5 driver will adjust the buffer configuration
accordingly based on the changes in pfc settings.
ConnectX buffer stride is 128 Bytes. If the buffer size is not multiple
of 128, the buffer size will be rounded down to the nearest multiple of
128.
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Add APIs for buffer configuration based on the changes in
pfc configuration, cable len, buffer size configuration,
and priority to buffer mapping.
Note that the xoff fomula is as below
xoff = ((301+2.16 * len [m]) * speed [Gbps] + 2.72 MTU [B]
xoff_threshold = buffer_size - xoff
xon_threshold = xoff_threshold - MTU
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Move four below functions from en_ethtool.c to en/port.c. These
functions are used by both en_ethtool.c and en_main.c. Future code
can use these functions without ethtool link mode dependency.
u32 mlx5e_port_ptys2speed(u32 eth_proto_oper);
int mlx5e_port_linkspeed(struct mlx5_core_dev *mdev, u32 *speed);
int mlx5e_port_max_linkspeed(struct mlx5_core_dev *mdev, u32 *speed);
u32 mlx5e_port_speed2linkmodes(u32 speed);
Delete the speed field from table mlx5e_build_ptys2ethtool_map. This
table only keeps the mapping between the mlx5e link mode and
ethtool link mode. Add new table mlx5e_link_speed for translation
from mlx5e link mode to actual speed.
Signed-off-by: Huy Nguyen <huyn@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
mlx5-updates-2018-05-17
mlx5 core dirver updates for both net-next and rdma-next branches.
From Christophe JAILLET, first three patche to use kvfree where needed.
From: Or Gerlitz <ogerlitz@mellanox.com>
Next six patches from Roi and Co adds support for merged
sriov e-switch which comes to serve cases where both PFs, VFs set
on them and both uplinks are to be used in single v-switch SW model.
When merged e-switch is supported, the per-port e-switch is logically
merged into one e-switch that spans both physical ports and all the VFs.
This model allows to offload TC eswitch rules between VFs belonging
to different PFs (and hence have different eswitch affinity), it also
sets the some of the foundations needed for uplink LAG support.
* tag 'mlx5-updates-2018-05-17' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
net/mlx5e: Explicitly set source e-switch in offloaded TC rules
net/mlx5: Add source e-switch owner
net/mlx5e: Explicitly set destination e-switch in FDB rules
net/mlx5: Add destination e-switch owner
net/mlx5: Properly handle a vport destination when setting FTE
net/mlx5: Add merged e-switch cap
IB/mlx5: Use 'kvfree()' for memory allocated by 'kvzalloc()'
net/mlx5: Eswitch, Use 'kvfree()' for memory allocated by 'kvzalloc()'
net/mlx5: Vport, Use 'kvfree()' for memory allocated by 'kvzalloc()'
After changing speed to 100Mbps as a result of auto-negotiation (AN),
some 10/100/1000Mbps SFPs indicate a successful link (no faults or loss
of signal), but cannot successfully transmit or receive data. These
SFPs required an extra auto-negotiation (AN) after the speed change in
order to operate properly. Add a quirk for these SFPs so that if the
outcome of the AN actually results in changing to a new speed, re-initiate
AN at that new speed.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using a quirk to make the BelFuse 1GBT-SFP06 part look like
a 1000baseX part, program the SFP PHY to support SGMII and 10/100/1000
baseT.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a KR re-driver is present, indicate the FEC support is available
during auto-negotiation.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Link training is always attempted when in KR mode, but the code is
structured to check if link training has been enabled before attempting
to perform it. Since that check will always be true, simplify the code
to always enable and start link training during KR auto-negotiation.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add ethtool support to show and set the device channel configuration.
Changing the channel configuration will result in a device restart.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to support being able to dynamically set/change the number of
Rx and Tx channels, update the code to:
- Move alloc and free of device memory into callable functions
- Move setting of the real number of Rx and Tx channels to device startup
- Move mapping of the RSS channels to device startup
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add ethtool support to show and set the number of the Rx and Tx ring
descriptors. Changing the ring configuration will result in a device
restart.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver currently sets an indication of whether the SFP supports, and
that the driver can obtain, diagnostics data. This isn't currently used
by the driver and the logic to set this indicator is flawed because the
field is cleared each time the SFP is checked and only set when a new SFP
is detected. Remove this field and the logic supporting it.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The comm_owned field can hide logic where double locking is attempted
and prevent multiple threads for the same device from accessing the
mutex properly. Remove the comm_owned field and use the mutex API
exclusively for gaining ownership. The current driver has been audited
and is obtaining communications ownership properly.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Read and save the port property registers once during the device probe
and then use the saved values as they are needed.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A debug output print statement uses the wrong variable to output the
maximum Rx channel count (cut and paste error, basically). Fix the
statement to use the proper variable.
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
trigger an L1 configure operation when a transceiver module
is inserted in order to cause current "sticky" options like
Requested Forward Error Correction to be reapplied.
Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
MDI Port Capabilities bit definitions were inconsistent with
regard to the MDI enum values. 2 bits used to define MDI in
the port capabilities are not really separable, it's a 2-bit
field with 4 different values. Change the port capability bit
definitions to be "AUTO" and "STRAIGHT" in order to get them
to line up with the enum's.
Signed-off-by: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Johannes Berg says:
For this round, we have various things all over the place, notably
* a fix for a race in aggregation, which I want to let
bake for a bit longer before sending to stable
* some new statistics (ACK RSSI, TXQ)
* TXQ configuration
* preparations for HE, particularly radiotap
* replace confusing "country IE" by "country element" since it's
not referring to Ireland
Note that I merged net-next to get a fix from mac80211 that got
there via net, to apply one patch that would otherwise conflict.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
spin_lock/unlock was used instead of spin_un/lock_irq
in a procedure used in process space, on a spinlock
which can be grabbed in an interrupt.
This caused the stack trace below to be displayed (on kernel
4.17.0-rc1 compiled with Lock Debugging enabled):
[ 154.661474] WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
[ 154.668909] 4.17.0-rc1-rdma_rc_mlx+ #3 Tainted: G I
[ 154.675856] -----------------------------------------------------
[ 154.682706] modprobe/10159 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
[ 154.690254] 00000000f3b0e495 (&(&qp_table->lock)->rlock){+.+.}, at: mlx4_qp_remove+0x20/0x50 [mlx4_core]
[ 154.700927]
and this task is already holding:
[ 154.707461] 0000000094373b5d (&(&cq->lock)->rlock/1){....}, at: destroy_qp_common+0x111/0x560 [mlx4_ib]
[ 154.718028] which would create a new lock dependency:
[ 154.723705] (&(&cq->lock)->rlock/1){....} -> (&(&qp_table->lock)->rlock){+.+.}
[ 154.731922]
but this new dependency connects a SOFTIRQ-irq-safe lock:
[ 154.740798] (&(&cq->lock)->rlock){..-.}
[ 154.740800]
... which became SOFTIRQ-irq-safe at:
[ 154.752163] _raw_spin_lock_irqsave+0x3e/0x50
[ 154.757163] mlx4_ib_poll_cq+0x36/0x900 [mlx4_ib]
[ 154.762554] ipoib_tx_poll+0x4a/0xf0 [ib_ipoib]
...
to a SOFTIRQ-irq-unsafe lock:
[ 154.815603] (&(&qp_table->lock)->rlock){+.+.}
[ 154.815604]
... which became SOFTIRQ-irq-unsafe at:
[ 154.827718] ...
[ 154.827720] _raw_spin_lock+0x35/0x50
[ 154.833912] mlx4_qp_lookup+0x1e/0x50 [mlx4_core]
[ 154.839302] mlx4_flow_attach+0x3f/0x3d0 [mlx4_core]
Since mlx4_qp_lookup() is called only in process space, we can
simply replace the spin_un/lock calls with spin_un/lock_irq calls.
Fixes: 6dc06c08be ("net/mlx4: Fix the check in attaching steering rules")
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement adjust_link function that allows to overwrite default CPU port
setting using fixed-link device tree subnode.
Signed-off-by: Michal Vokáč <michal.vokac@ysoft.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
By default autonegotiation is enabled to configure MAC on all ports.
For the CPU port autonegotiation can not be used so we need to set
some sensible defaults manually.
This patch forces the default setting of the CPU port to 1000Mbps/full
duplex which is the chip maximum capability.
Also correct size of the bit field used to configure link speed.
Fixes: 6b93fb4648 ("net-next: dsa: add new driver for qca8xxx family")
Signed-off-by: Michal Vokáč <michal.vokac@ysoft.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a port is brought up/down do not enable/disable only the TXMAC
but the RXMAC as well. This is essential for the CPU port to work.
Fixes: 6b93fb4648 ("net-next: dsa: add new driver for qca8xxx family")
Signed-off-by: Michal Vokáč <michal.vokac@ysoft.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
