| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
During some fatal firmware error conditions, the PCI config space
register 0x2e which normally contains the subsystem ID will become
0xffff. This register will revert back to the normal value after
the chip has completed core reset. If we detect this condition,
we can poll this config register immediately for the value to revert.
Because we use config read cycles to poll this register, there is no
possibility of Master Abort if we happen to read it during core reset.
This speeds up recovery significantly as we don't have to wait for the
conservative min_time before polling MMIO to see if the firmware has
come out of reset. As soon as this register changes value we can
proceed to re-initialize the device.
Reviewed-by: Edwin Peer <[email protected]>
Reviewed-by: Vasundhara Volam <[email protected]>
Reviewed-by: Andy Gospodarek <[email protected]>
Signed-off-by: Michael Chan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Newer devices may have local context memory instead of relying on the
host for backing store. In these cases, HWRM_FUNC_BACKING_STORE_QCAPS
will return a zero entry size to indicate contexts for which the host
should not allocate backing store.
Selectively allocate context memory based on device capabilities and
only enable backing store for the appropriate contexts.
Signed-off-by: Edwin Peer <[email protected]>
Signed-off-by: Michael Chan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The main changes are the echo request/response from firmware for error
detection and the NO_FCS feature to transmit frames without FCS.
Signed-off-by: Michael Chan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Newer versions of the Xilinx AXI Ethernet core (specifically version 7.2 or
later) allow the core to be configured with a PHY interface mode of "Both",
allowing either 1000BaseX or SGMII modes to be selected at runtime. Add
support for this in the driver to allow better support for applications
which can use both fiber and copper SFP modules.
Signed-off-by: Robert Hancock <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Hook up the nway_reset ethtool operation to the corresponding phylink
function so that "ethtool -r" can be supported.
Signed-off-by: Robert Hancock <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This driver is set up to use a clock mapping in the device tree if it is
present, but still work without one for backward compatibility. However,
if getting the clock returns -EPROBE_DEFER, then we need to abort and
return that error from our driver initialization so that the probe can
be retried later after the clock is set up.
Move clock initialization to earlier in the process so we do not waste as
much effort if the clock is not yet available. Switch to use
devm_clk_get_optional and abort initialization on any error reported.
Also enable the clock regardless of whether the controller is using an MDIO
bus, as the clock is required in any case.
Fixes: 09a0354cadec267be7f ("net: axienet: Use clock framework to get device clock rate")
Signed-off-by: Robert Hancock <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue
Tony Nguyen says:
====================
40GbE Intel Wired LAN Driver Updates 2021-02-12
This series contains updates to i40e, ice, and ixgbe drivers.
Maciej does cleanups on the following drivers.
For i40e, removes redundant check for XDP prog, cleans up no longer
relevant information, and removes an unused function argument.
For ice, removes local variable use, instead returning values directly.
Moves skb pointer from buffer to ring and removes an unneeded check for
xdp_prog in zero copy path. Also removes a redundant MTU check when
changing it.
For i40e, ice, and ixgbe, stores the rx_offset in the Rx ring as
the value is constant so there's no need for continual calls.
Bjorn folds a decrement into a while statement.
====================
Signed-off-by: David S. Miller <[email protected]>
|
|
of_dev_get() and of_dev_put are just wrappers for get_device()/put_device()
on a platform_device. There's also already platform_device_{get,put}()
wrappers for this purpose. Let's update the few users and remove
of_dev_{get,put}().
Cc: Michael Ellerman <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: "David S. Miller" <[email protected]>
Cc: Jakub Kicinski <[email protected]>
Cc: Frank Rowand <[email protected]>
Cc: Patrice Chotard <[email protected]>
Cc: Felipe Balbi <[email protected]>
Cc: Julia Lawall <[email protected]>
Cc: Gilles Muller <[email protected]>
Cc: Nicolas Palix <[email protected]>
Cc: Michal Marek <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Rob Herring <[email protected]>
Reviewed-by: Greg Kroah-Hartman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
|
|
The supported indirect subcrq entries on Power8 is 16. Power9
supports 128. Redefined this value to 16 to minimize the driver from
having to reset when migrating between Power9 and Power8. In our rx/tx
performance testing, we found no performance difference between 16 and
128 at this time.
Fixes: f019fb6392e5 ("ibmvnic: Introduce indirect subordinate Command Response Queue buffer")
Signed-off-by: Dany Madden <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
We should not be unconditionally enabling address learning, since doing
that is actively detrimential when a port is standalone and not offloading
a bridge. Namely, if a port in the switch is standalone and others are
offloading the bridge, then we could enter a situation where we learn an
address towards the standalone port, but the bridged ports could not
forward the packet there, because the CPU is the only path between the
standalone and the bridged ports. The solution of course is to not
enable address learning unless the bridge asks for it.
We need to set up the initial port flags for no learning and flooding
everything, and also when the port joins and leaves the bridge.
The flood configuration was already configured ok for standalone mode
in ocelot_init, we just need to disable learning in ocelot_init_port.
Signed-off-by: Vladimir Oltean <[email protected]>
Reviewed-by: Alexandre Belloni <[email protected]>
Reviewed-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In preparation of offloading the bridge port flags which have
independent settings for unknown multicast and for broadcast, we should
also start reserving one destination Port Group ID for the flooding of
broadcast packets, to allow configuring it individually.
Signed-off-by: Vladimir Oltean <[email protected]>
Reviewed-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This switchdev attribute offers a counterproductive API for a driver
writer, because although br_switchdev_set_port_flag gets passed a
"flags" and a "mask", those are passed piecemeal to the driver, so while
the PRE_BRIDGE_FLAGS listener knows what changed because it has the
"mask", the BRIDGE_FLAGS listener doesn't, because it only has the final
value. But certain drivers can offload only certain combinations of
settings, like for example they cannot change unicast flooding
independently of multicast flooding - they must be both on or both off.
The way the information is passed to switchdev makes drivers not
expressive enough, and unable to reject this request ahead of time, in
the PRE_BRIDGE_FLAGS notifier, so they are forced to reject it during
the deferred BRIDGE_FLAGS attribute, where the rejection is currently
ignored.
This patch also changes drivers to make use of the "mask" field for edge
detection when possible.
Signed-off-by: Vladimir Oltean <[email protected]>
Reviewed-by: Grygorii Strashko <[email protected]>
Reviewed-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When a struct switchdev_attr is notified through switchdev, there is no
way to report informational messages, unlike for struct switchdev_obj.
Signed-off-by: Vladimir Oltean <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Reviewed-by: Florian Fainelli <[email protected]>
Reviewed-by: Nikolay Aleksandrov <[email protected]>
Reviewed-by: Grygorii Strashko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Fixes: 93efb0c656837 ("octeontx2-pf: Fix out-of-bounds read in otx2_get_fecparam()")
Signed-off-by: David S. Miller <[email protected]>
|
|
Code at line 967 implies that rsp->fwdata.supported_fec may be up to 4:
967: if (rsp->fwdata.supported_fec <= FEC_MAX_INDEX)
If rsp->fwdata.supported_fec evaluates to 4, then there is an
out-of-bounds read at line 971 because fec is an array with
a maximum of 4 elements:
954 const int fec[] = {
955 ETHTOOL_FEC_OFF,
956 ETHTOOL_FEC_BASER,
957 ETHTOOL_FEC_RS,
958 ETHTOOL_FEC_BASER | ETHTOOL_FEC_RS};
959 #define FEC_MAX_INDEX 4
971: fecparam->fec = fec[rsp->fwdata.supported_fec];
Fix this by properly indexing fec[] with rsp->fwdata.supported_fec - 1.
In this case the proper indexes 0 to 3 are used when
rsp->fwdata.supported_fec evaluates to a range of 1 to 4, correspondingly.
Fixes: d0cf9503e908 ("octeontx2-pf: ethtool fec mode support")
Addresses-Coverity-ID: 1501722 ("Out-of-bounds read")
Signed-off-by: Gustavo A. R. Silva <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
There is a spelling mistake in the text in array rpm_rx_stats_fields,
fix it.
Signed-off-by: Colin Ian King <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
hclge_rm_vport_all_mac_table() is bloated, so split it into
separate functions for readability and maintainability.
Signed-off-by: Hao Chen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To make it more readable and maintainable, split
hclgevf_set_rss_tuple() into two parts.
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To make it more readable and maintainable, split
hclge_set_rss_tuple() into two parts.
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
hclgevf_cmd_send() is bloated, so split it into separate
functions for readability and maintainability.
Signed-off-by: Yufeng Mo <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
hclge_cmd_send() is bloated, so split it into separate
functions for readability and maintainability.
Signed-off-by: Yufeng Mo <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
hclge_dbg_dump_qos_buf_cfg() is bloated, so split it into
separate functions for readability and maintainability.
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To improve code readability and maintainability, separate
the flow type parsing part and the converting part from
bloated hclgevf_get_rss_tuple().
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To improve code readability and maintainability, separate
the flow type parsing part and the converting part from
bloated hclge_get_rss_tuple().
Signed-off-by: Jian Shen <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To improve code readability and maintainability, separate
the command handling part and the status parsing part from
bloated hclge_set_vf_vlan_common().
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Use common ipv6_addr_any() to determine if an addr is ipv6 any addr.
Signed-off-by: Jiaran Zhang <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
As more commands are added, hns3_dbg_cmd_write() is going to
get more bloated, so move the part about command check into
a separate function.
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To improve code readability and maintainability, refactor
hclgevf_cmd_convert_err_code() with an array of imp_errcode
and common_errno mapping, instead of a bloated switch/case.
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To improve code readability and maintainability, refactor
hclge_cmd_convert_err_code() with an array of imp_errcode
and common_errno mapping, instead of a bloated switch/case.
Signed-off-by: Peng Li <[email protected]>
Signed-off-by: Huazhong Tan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Output of ixgbe_rx_offset() is based on ethtool's priv flag setting, which
when changed, causes PF reset (disables napi, frees irqs, loads
different Rx mem model, etc.). This means that within napi its result is
constant and there is no reason to call it per each processed frame.
Add new 'rx_offset' field to ixgbe_ring that is meant to hold the
ixgbe_rx_offset() result and use it within ixgbe_clean_rx_irq().
Furthermore, use it within ixgbe_alloc_mapped_page().
Last but not least, un-inline the function of interest as it lives in .c
file so let compiler do the decision about the inlining.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
Output of ice_rx_offset() is based on ethtool's priv flag setting, which
when changed, causes PF reset (disables napi, frees irqs, loads
different Rx mem model, etc.). This means that within napi its result is
constant and there is no reason to call it per each processed frame.
Add new 'rx_offset' field to ice_ring that is meant to hold the
ice_rx_offset() result and use it within ice_clean_rx_irq().
Furthermore, use it within ice_alloc_mapped_page().
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
Output of i40e_rx_offset() is based on ethtool's priv flag setting,
which when changed, causes PF reset (disables napi, frees irqs, loads
different Rx mem model, etc.). This means that within napi its result is
constant and there is no reason to call it per each processed frame.
Add new 'rx_offset' field to i40e_ring that is meant to hold the
i40e_rx_offset() result and use it within i40e_clean_rx_irq().
Furthermore, use it within i40e_alloc_mapped_page().
Last but not least, un-inline the function of interest so that compiler
makes the decision about inlining as it lives in .c file.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
Fold the count decrement into the while-statement.
Reviewed-by: Maciej Fijalkowski <[email protected]>
Signed-off-by: Björn Töpel <[email protected]>
Tested-by: Kiran Bhandare <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
Whole zero-copy variant of clean Rx IRQ is executed when xsk_pool is
attached to rx_ring and it can happen only when XDP program is present
on interface. Therefore it is safe to assume that program is always
!NULL and there is no need for checking it in ice_run_xdp_zc.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Kiran Bhandare <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
dev_validate_mtu checks that mtu value specified by user is not less
than min mtu and not greater than max allowed mtu. It is being done
before calling the ndo_change_mtu exposed by driver, so remove these
redundant checks in ice_change_mtu.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
Similar thing has been done in i40e, as there is no real need for having
the sk_buff pointer in each rx_buf. Non-eop frames can be simply handled
on that pointer moved upwards to rx_ring.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
There's no need for 'result' variable, we can directly return the
internal status based on action returned by xdp prog.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Kiran Bhandare <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
i40e_is_non_eop had a leftover comment and unused skb argument which was
used for placing the skb onto rx_buf in case when current buffer was
non-eop one. This is not relevant anymore as commit e72e56597ba1
("i40e/i40evf: Moves skb from i40e_rx_buffer to i40e_ring") pulled the
non-complete skb handling out of rx_bufs up to rx_ring. Therefore,
let's adjust the function arguments that i40e_is_non_eop takes.
Furthermore, since there is already a function responsible for bumping
the ntc, make use of that and drop that logic from i40e_is_non_eop so
that the scope of this function is limited to what the name actually
states.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
i40e_cleanup_headers has a statement about check against skb being
linear or not which is not relevant anymore, so let's remove it.
Same case for i40e_can_reuse_rx_page, it references things that are not
present there anymore.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Tony Brelinski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
Net core handles the case where netdev has no xdp prog attached and
current prog is NULL. Therefore, remove such check within
i40e_xdp_setup.
Reviewed-by: Björn Töpel <[email protected]>
Signed-off-by: Maciej Fijalkowski <[email protected]>
Tested-by: Kiran Bhandare <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>
|
|
No real dependency on XPS, but on RX queue mapping, which
is being selected by TLS_DEVICE.
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Maxim Mikityanskiy <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Check that tunnel offload is required before setting Software Parser
offsets to get Geneve HW offload. In case of Geneve packet we check HW
offload support of SWP in mlx5e_tunnel_features_check() and set features
accordingly, this should be reflected in skb offload requested by the
kernel and we should add the Software Parser offsets only if requested.
Otherwise, in case HW doesn't support SWP for Geneve, data path will
mistakenly try to offload Geneve SKBs with skb->encapsulation set,
regardless of whether offload was requested or not on this specific SKB.
Fixes: e3cfc7e6b7bd ("net/mlx5e: TX, Add geneve tunnel stateless offload support")
Signed-off-by: Moshe Shemesh <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
The ct entry object is accessed by the ct add, del, stats and restore
methods. In addition, it is referenced from several hash tables.
The lifetime of the ct entry object was not managed which triggered race
conditions as in the following kasan dump:
[ 3374.973945] ==================================================================
[ 3374.988552] BUG: KASAN: use-after-free in memcmp+0x4c/0x98
[ 3374.999590] Read of size 1 at addr ffff00036129ea55 by task ksoftirqd/1/15
[ 3375.016415] CPU: 1 PID: 15 Comm: ksoftirqd/1 Tainted: G O 5.4.31+ #1
[ 3375.055301] Call trace:
[ 3375.060214] dump_backtrace+0x0/0x238
[ 3375.067580] show_stack+0x24/0x30
[ 3375.074244] dump_stack+0xe0/0x118
[ 3375.081085] print_address_description.isra.9+0x74/0x3d0
[ 3375.091771] __kasan_report+0x198/0x1e8
[ 3375.099486] kasan_report+0xc/0x18
[ 3375.106324] __asan_load1+0x60/0x68
[ 3375.113338] memcmp+0x4c/0x98
[ 3375.119409] mlx5e_tc_ct_restore_flow+0x3a4/0x6f8 [mlx5_core]
[ 3375.131073] mlx5e_rep_tc_update_skb+0x1d4/0x2f0 [mlx5_core]
[ 3375.142553] mlx5e_handle_rx_cqe_rep+0x198/0x308 [mlx5_core]
[ 3375.154034] mlx5e_poll_rx_cq+0x2a0/0x1060 [mlx5_core]
[ 3375.164459] mlx5e_napi_poll+0x1d4/0xa78 [mlx5_core]
[ 3375.174453] net_rx_action+0x28c/0x7a8
[ 3375.182004] __do_softirq+0x1b4/0x5d0
Manage the lifetime of the ct entry object by using synchornization
mechanisms for concurrent access.
Fixes: ac991b48d43c ("net/mlx5e: CT: Offload established flows")
Signed-off-by: Roi Dayan <[email protected]>
Signed-off-by: Oz Shlomo <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Devlink reload can't be allowed on lag devices since reloading one lag
device will cause traffic on the bond to get stucked.
Users who wish to reload a lag device, need to remove the device from
the bond, and only then reload it.
Fixes: 4383cfcc65e7 ("net/mlx5: Add devlink reload")
Signed-off-by: Shay Drory <[email protected]>
Reviewed-by: Moshe Shemesh <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
In lag mode, setting roce enabled/disable of lag device have no effect.
e.g.: bond device (roce/vf_lag) roce status remain unchanged.
Therefore disable it and add an error message.
Fixes: cc9defcbb8fa ("net/mlx5: Handle "enable_roce" devlink param")
Signed-off-by: Shay Drory <[email protected]>
Reviewed-by: Moshe Shemesh <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
In dual port mode, setting roce enabled/disable for the slave device
have no effect. e.g.: the slave device roce status remain unchanged.
Therefore disable it and add an error message.
Enable or disable roce of the master device affect both master and slave
devices.
Fixes: cc9defcbb8fa ("net/mlx5: Handle "enable_roce" devlink param")
Signed-off-by: Shay Drory <[email protected]>
Reviewed-by: Moshe Shemesh <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Devlink reload can't be allowed on a multi port slave device, because
reload of slave device doesn't take effect.
The right flow is to disable devlink reload for multi port slave
device. Hence, disabling it in mlx5_core probing.
Fixes: 4383cfcc65e7 ("net/mlx5: Add devlink reload")
Signed-off-by: Shay Drory <[email protected]>
Reviewed-by: Moshe Shemesh <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
wait_for_resync is unreliable - if it timeouts, priv_rx will be freed
anyway. However, mlx5e_ktls_handle_get_psv_completion will be called
sooner or later, leading to use-after-free. For example, it can happen
if a CQ error happened, and ICOSQ stopped, but later on the queues are
destroyed, and ICOSQ is flushed with mlx5e_free_icosq_descs.
This patch converts the lifecycle of priv_rx to fully refcount-based, so
that the struct won't be freed before the refcount goes to zero.
Fixes: 0419d8c9d8f8 ("net/mlx5e: kTLS, Add kTLS RX resync support")
Signed-off-by: Maxim Mikityanskiy <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
The commit mentioned below has split the parameters of ICOSQ and async
ICOSQ, but it contained a typo: the CQ parameters were swapped for ICOSQ
and async ICOSQ. Async ICOSQ is longer than the normal ICOSQ, and the CQ
size must be the same as the size of the corresponding SQ, but due to
this bug, the CQ of async ICOSQ was much shorter than async ICOSQ
itself. It led to overflows of the CQ with such messages in dmesg, in
particular, when running multiple kTLS-offloaded streams:
mlx5_core 0000:08:00.0: cq_err_event_notifier:529:(pid 9422): CQ error
on CQN 0x406, syndrome 0x1
mlx5_core 0000:08:00.0 eth2: mlx5e_cq_error_event: cqn=0x000406
event=0x04
This commit fixes the issue by using the corresponding parameters for
ICOSQ and async ICOSQ.
Fixes: c293ac927fbb ("net/mlx5e: Refactor build channel params")
Signed-off-by: Maxim Mikityanskiy <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
The commit cited below switched from using napi_synchronize to
synchronize_rcu to have a guarantee that it will finish in finite time.
However, on average, synchronize_rcu takes more time than
napi_synchronize. Given that it's called multiple times per channel on
deactivation, it accumulates to a significant amount, which causes
timeouts in some applications (for example, when using bonding with
NetworkManager).
This commit replaces synchronize_rcu with synchronize_net, which is
faster when called under rtnl_lock, allowing to speed up the described
flow.
Fixes: 9c25a22dfb00 ("net/mlx5e: Use synchronize_rcu to sync with NAPI")
Signed-off-by: Maxim Mikityanskiy <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|