Report the current FW values for the VF attributes, but don't
save the FW values locally, only save the vf attributes that
are given to us from the user. This allows us to replay user
data, and doesn't end up confusing things like "who set the
mac address".
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The VF attributes that the user has set into the FW through
the PF can be lost over a FW crash recovery. Much like we
already replay the PF mac/vlan filters, we now add a replay
in the recovery path to be sure the FW has the up-to-date
VF configurations.
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/can/usb/kvaser_usb/kvaser_usb_leaf.c
2871edb32f ("can: kvaser_usb: Fix possible completions during init_completion")
abb8670938 ("can: kvaser_usb_leaf: Ignore stale bus-off after start")
8d21f5927a ("can: kvaser_usb_leaf: Fix improved state not being reported")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Under memory pressure, enetc_refill_rx_ring() may fail, and when called
during the enetc_open() -> enetc_setup_rxbdr() procedure, this is not
checked for.
An extreme case of memory pressure will result in exactly zero buffers
being allocated for the RX ring, and in such a case it is expected that
hardware drops all RX packets due to lack of buffers.
This does not happen, because the reset-default value of the consumer
and produces index is 0, and this makes the ENETC think that all buffers
have been initialized and that it owns them (when in reality none were).
The hardware guide explains this best:
| Configure the receive ring producer index register RBaPIR with a value
| of 0. The producer index is initially configured by software but owned
| by hardware after the ring has been enabled. Hardware increments the
| index when a frame is received which may consume one or more BDs.
| Hardware is not allowed to increment the producer index to match the
| consumer index since it is used to indicate an empty condition. The ring
| can hold at most RBLENR[LENGTH]-1 received BDs.
|
| Configure the receive ring consumer index register RBaCIR. The
| consumer index is owned by software and updated during operation of the
| of the BD ring by software, to indicate that any receive data occupied
| in the BD has been processed and it has been prepared for new data.
| - If consumer index and producer index are initialized to the same
| value, it indicates that all BDs in the ring have been prepared and
| hardware owns all of the entries.
| - If consumer index is initialized to producer index plus N, it would
| indicate N BDs have been prepared. Note that hardware cannot start if
| only a single buffer is prepared due to the restrictions described in
| (2).
| - Software may write consumer index to match producer index anytime
| while the ring is operational to indicate all received BDs prior have
| been processed and new BDs prepared for hardware.
Normally, the value of rx_ring->rcir (consumer index) is brought in sync
with the rx_ring->next_to_use software index, but this only happens if
page allocation ever succeeded.
When PI==CI==0, the hardware appears to receive frames and write them to
DMA address 0x0 (?!), then set the READY bit in the BD.
The enetc_clean_rx_ring() function (and its XDP derivative) is naturally
not prepared to handle such a condition. It will attempt to process
those frames using the rx_swbd structure associated with index i of the
RX ring, but that structure is not fully initialized (enetc_new_page()
does all of that). So what happens next is undefined behavior.
To operate using no buffer, we must initialize the CI to PI + 1, which
will block the hardware from advancing the CI any further, and drop
everything.
The issue was seen while adding support for zero-copy AF_XDP sockets,
where buffer memory comes from user space, which can even decide to
supply no buffers at all (example: "xdpsock --txonly"). However, the bug
is present also with the network stack code, even though it would take a
very determined person to trigger a page allocation failure at the
perfect time (a series of ifup/ifdown under memory pressure should
eventually reproduce it given enough retries).
Fixes: d4fd0404c1 ("enetc: Introduce basic PF and VF ENETC ethernet drivers")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Link: https://lore.kernel.org/r/20221027182925.3256653-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit at rx sa update operation passes the sci object
attribute, in the wrong endianness and not as expected by the HW
effectively create malformed hw sa context in case of update rx sa
consequently, HW produces unexpected MACsec packets which uses this
sa.
Fix by passing sci to create macsec object with the correct endianness,
while at it add __force u64 to prevent sparse check error of type
"sparse: error: incorrect type in assignment".
Fixes: aae3454e4d ("net/mlx5e: Add MACsec offload Rx command support")
Signed-off-by: Raed Salem <raeds@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-16-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit produces a sparse check error of type
"sparse: error: restricted __be64 degrades to integer". The
offending line wrongly did a bitwise operation between two different
storage types one of 64 bit when the other smaller side is 16 bit
which caused the above sparse error, furthermore bitwise operation
usage here is wrong in the first place as the constant MACSEC_PORT_ES
is not a bitwise field.
Fix by using the right mask to get the lower 16 bit if the sci number,
and use comparison operator '==' instead of bitwise '&' operator.
Fixes: 3b20949cb2 ("net/mlx5e: Add MACsec RX steering rules")
Signed-off-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-15-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit adds the support for update/delete MACsec Rx SA,
naturally, these operations need to check if the SA in question exists
to update/delete the SA and return error code otherwise, however they
do just the opposite i.e. return with error if the SA exists
Fix by change the check to return error in case the SA in question does
not exist, adjust error message and code accordingly.
Fixes: aae3454e4d ("net/mlx5e: Add MACsec offload Rx command support")
Signed-off-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-14-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit at update rx sa operation passes object attributes
to MACsec object create function without initializing/setting all
attributes fields leaving some of them with garbage values, therefore
violating the implicit assumption at create object function, which
assumes that all input object attributes fields are set.
Fix by initializing the object attributes struct to zero, thus leaving
unset fields with the legal zero value.
Fixes: aae3454e4d ("net/mlx5e: Add MACsec offload Rx command support")
Signed-off-by: Raed Salem <raeds@nvidia.com>
Reviewed-by: Lior Nahmanson <liorna@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-13-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When setting Bluefield to DPU NIC mode using mlxconfig tool + sync
firmware reset flow, we run into scenario where the host was not
eswitch manager at the time of mlx5 driver load but becomes eswitch manager
after the sync firmware reset flow. This results in null pointer
access of mpfs structure during mac filter add. This change prevents null
pointer access but mpfs table entries will not be added.
Fixes: 5ec697446f ("net/mlx5: Add support for devlink reload action fw activate")
Signed-off-by: Suresh Devarakonda <ramad@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Bodong Wang <bodong@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-12-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Update devlink health fw fatal reporter state to "healthy" is needed by
strictly calling devlink_health_reporter_state_update() after recovery
was done by PCI error handler. This is needed when fw_fatal reporter was
triggered due to PCI error. Poll health is called and set reporter state
to error. Health recovery failed (since EEH didn't re-enable the PCI).
PCI handlers keep on recover flow and succeed later without devlink
acknowledgment. Fix this by adding devlink state update at the end of
the PCI handler recovery process.
Fixes: 6181e5cb75 ("devlink: add support for reporter recovery completion")
Signed-off-by: Roy Novich <royno@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Aya Levin <ayal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-11-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reject TC rules that forward from internal port to internal port
as it is not supported.
This include rules that are explicitly have internal port as
the filter device as well as rules that apply on tunnel interfaces
as the route device for the tunnel interface can be an internal
port.
Fixes: 27484f7170 ("net/mlx5e: Offload tc rules that redirect to ovs internal port")
Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Maor Dickman <maord@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-9-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mlx5 SQs must select the timestamp format explicitly according to the
active clock mode, select the current active timestamp mode so ASO SQ create
will succeed.
This fixes the following error prints when trying to create ipsec ASO SQ
while the timestamp format is real time mode.
mlx5_cmd_out_err:778:(pid 34874): CREATE_SQ(0x904) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0xd61c0b), err(-22)
mlx5_aso_create_sq:285:(pid 34874): Failed to open aso wq sq, err=-22
mlx5e_ipsec_init:436:(pid 34874): IPSec initialization failed, -22
Fixes: cdd04f4d4d ("net/mlx5: Add support to create SQ and CQ for ASO")
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reported-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-7-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently encap slow path rules just forward to software without
setting the chain id miss register, so driver doesn't restore
the chain, and packets hitting this rule will restart from tc chain
0 instead of continuing to the chain the encap rule was on.
Fix this by setting the chain id miss register to the chain id mapping.
Fixes: 8f1e0b97cc ("net/mlx5: E-Switch, Mark miss packets with new chain id mapping")
Signed-off-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-6-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When tx_port_ts is set, the driver diverts all UPD traffic over PTP port
to a dedicated PTP-SQ. The SKBs are cached until the wire-CQE arrives.
When the packet size is greater then MTU, the firmware might drop it and
the packet won't be transmitted to the wire, hence the wire-CQE won't
reach the driver. In this case the SKBs are accumulated in the SKB fifo.
Add room check to consider the PTP-SQ SKB fifo, when the SKB fifo is
full, driver stops the queue resulting in a TX timeout. Devlink
TX-reporter can recover from it.
Fixes: 1880bc4e4a ("net/mlx5e: Add TX port timestamp support")
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-5-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When 2nd flow rules arrives, it will merge together with the
1st one if matcher criteria is the same.
If merge fails, driver will rollback the merge contents, and
reject the 2nd rule. At rollback stage, matcher can't be
disconnected unconditionally, otherise the 1st rule can't be
hit anymore.
Add logic to check if the matcher should be disconnected or not.
Fixes: cc2295cd54 ("net/mlx5: DR, Improve steering for empty or RX/TX-only matchers")
Signed-off-by: Rongwei Liu <rongweil@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-4-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
An offloaded SA stops receiving after about 2^32 + replay_window
packets. For example, when SA reaches <seq-hi 0x1, seq 0x2c>, all
subsequent packets get dropped with SA-icv-failure (integrity_failed).
To reproduce the bug:
- ConnectX-6 Dx with crypto enabled (FW 22.30.1004)
- ipsec.conf:
nic-offload = yes
replay-window = 32
esn = yes
salifetime=24h
- Run netperf for a long time to send more than 2^32 packets
netperf -H <device-under-test> -t TCP_STREAM -l 20000
When 2^32 + replay_window packets are received, the replay window
moves from the 2nd half of subspace (overlap=1) to the 1st half
(overlap=0). The driver then updates the 'esn' value in NIC
(i.e. seq_hi) as follows.
seq_hi = xfrm_replay_seqhi(seq_bottom)
new esn in NIC = seq_hi + 1
The +1 increment is wrong, as seq_hi already contains the correct
seq_hi. For example, when seq_hi=1, the driver actually tells NIC to
use seq_hi=2 (esn). This incorrect esn value causes all subsequent
packets to fail integrity checks (SA-icv-failure). So, do not
increment.
Fixes: cb01008390 ("net/mlx5: IPSec, Add support for ESN")
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
Acked-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20221026135153.154807-2-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Remove dir in nsim_dev_debugfs_init() when creating ports dir failed.
Otherwise, the netdevsim device will not be created next time. Kernel
reports an error: debugfs: Directory 'netdevsim1' with parent 'netdevsim'
already present!
Fixes: ab1d0cc004 ("netdevsim: change debugfs tree topology")
Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Marc Kleine-Budde says:
====================
pull-request: can 2022-10-27
Anssi Hannula fixes the use of the completions in the kvaser_usb
driver.
Biju Das contributes 2 patches for the rcar_canfd driver. A IRQ storm
that can be triggered by high CAN bus load and channel specific IRQ
handlers are fixed.
Yang Yingliang fixes the j1939 transport protocol by moving a
kfree_skb() out of a spin_lock_irqsave protected section.
* tag 'linux-can-fixes-for-6.1-20221027' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can:
can: j1939: transport: j1939_session_skb_drop_old(): spin_unlock_irqrestore() before kfree_skb()
can: rcar_canfd: fix channel specific IRQ handling for RZ/G2L
can: rcar_canfd: rcar_canfd_handle_global_receive(): fix IRQ storm on global FIFO receive
can: kvaser_usb: Fix possible completions during init_completion
====================
Link: https://lore.kernel.org/r/20221027114356.1939821-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Queueing packets doesn't guarantee their transmission. Update TX stats
after hardware confirms consuming submitted data.
This also fixes a possible race and NULL dereference.
bcm4908_enet_start_xmit() could try to access skb after freeing it in
the bcm4908_enet_poll_tx().
Reported-by: Florian Fainelli <f.fainelli@gmail.com>
Fixes: 4feffeadbc ("net: broadcom: bcm4908enet: add BCM4908 controller driver")
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20221027112430.8696-1-zajec5@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
No need to have the used_list - we don't need to keep track of the
used chunks, we only need to know the amount of used memory.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
When ICM chunk is freed, it might still be accessed by HW until we do
sync with HW. This sync is expensive operation, so we don't do it often.
Instead, when the chunk is freed, it is moved to the buddy's "hot memory"
list. Once sync is done, we traverse the hot list and finally free all
the chunks.
It appears that traversing a long list takes unusually long time due to cache
misses on many entries, which causes a big "hiccup" during rule insertion.
This patch deals with this issue the following way:
- Move hot chunks list from buddy to pool, so that the pool will
keep track of all its hot memory.
- Replace the list with pre-allocated array on the memory pool struct,
and store only the information that is needed to later free this
chunk in its buddy allocator.
This cost additional memory for the array that is dynamically
allocated, but it allows not to save long list of hot chunks,
so at peak times it actually saves memory due to the fact that
each array entry is much smaller than the chunk struct.
This way an overhead of traversing the long list is virtually removed:
the loop of freeing hot chunks takes ~27 msec instead of ~70 msec, where
most of it are the actual freeing activities.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Instead of hiding the math in the code, define a value that sets the
fraction of allowed hot memory of ICM pool.
Set the threshold for sync of ICM hot chunks to 1/4 of the pool
instead of 1/2 of the pool. Although we will have more syncs, each
sync will be shorter and will help with insertion rate stability.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
SW steering allocates/frees lots of htbl structs. Create a
separate kmem_cache and allocate htbls from this allocator.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
SW steering allocates/frees lots of icm_chunk structs. To make this more
efficiently, create a separate kmem_cache and allocate these chunks from
this allocator.
By doing this we observe that the alloc/free "hiccups" frequency has
become much lower, which allows for a more steady rule insersion rate.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Instead of allocating/freeing send info objects dynamically, manage them
in pool. The number of send info objects doesn't depend on rules, so after
pre-populating the pool with an initial batch of send info objects, the
pool is not expected to grow.
This way we save alloc/free during writing STEs to ICM, which can
sometimes take up to 40msec.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Don't wait for the whole table to be ready - write each row immediately.
This way we save allocations of the ste_send_info structure and improve
performance.
Signed-off-by: Erez Shitrit <erezsh@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Handle creation/destruction of all the domain's memory pools and other
memory-related fields in a separate init/uninit functions.
This simplifies error flow and allows cleaner addition of new pools.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Rather than cleaning the corresponding chunk's section of ste_arrays on
chunk deletion, initialize these areas upon chunk creation.
Chunk destruction tend to come in large batches (during pool syncing).
To reduce the "hiccup" in such cases, moving ste_arrays init from chunk
destruction to initialization.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
While creating rule, ste_arr is an array that is allocated at the start
of the function and freed at the end.
This memory allocation can sometimes lead to "hiccups" of up to 10ms.
However, the common use case is short chains of STEs. For such cases,
we can use a local buffer on stack instead.
Changes in v2:
Use small local array for short rules, allocate dynamically for long rules
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Remove an argument that can be extracted in the function.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Calling fast teardown as part of the normal unloading caused
a problem with SW steering - SW steering still needs to clear
its tables, write to ICM and poll for completions.
When teardown has been done, SW steering keeps polling the CQ
forever, because nobody flushes it.
This patch fixes the issue by checking the device state in
cases where no CQE was returned.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
If sync happens when the device is in fast teardown, just bail
and don't do anything, because the PCI device is not there any more.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Otherwise resources will never be freed and refcount will not be decreased.
Signed-off-by: Chris Mi <cmi@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Avoid the PHY library call unnecessarily into the suspend/resume
functions by setting phydev->mac_managed_pm to true. The SYSTEMPORT
driver essentially does exactly what mdio_bus_phy_resume() does by
calling phy_resume().
Fixes: fba863b816 ("net: phy: make PHY PM ops a no-op if MAC driver manages PHY PM")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20221025234201.2549360-1-f.fainelli@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently we assume that any filter table contains a fixed number
of entries. Like routing tables, the number of entries in a filter
table is limited only by the size of the IPA-local memory region
used to hold the table.
Stop assuming that a filter table has exactly 14 entries. Instead,
determine the number of entries in a routing table by dividing its
memory region size by the size of an entry. (Note that the first
"entry" in a filter table contains an endpoint bitmap.)
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently all platforms are assumed allot 8 routing table entries
for use by the modem. Instead, add a new configuration data entry
that defines the number of modem routing table entries, and record
that in the IPA structure.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently we assume that any routing table contains a fixed number
of entries. The number of entries in a routing table can actually
vary, depending only on the size of the IPA-local memory region used
to hold the table.
Stop assuming that a routing table has exactly 15 entries. Instead,
determine the number of entries in a routing table by dividing its
memory region size by the size of an entry.
The number of entries is computed early, when ipa_table_mem_valid()
is called by ipa_table_init().
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The non-hashed routing tables for IPv4 and IPv6 will be the same
size. And if supported, the hashed routing tables will be the same
size as the non-hashed tables.
Record the size (number of entries) of all routing tables in the IPA
structure. For now, initialize this field using IPA_ROUTE_TABLE_MAX,
and just do so when the first route table is validated.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
RX code can be more efficient with the build_skb(). Allocating actual
SKB around eth packet buffer - right before passing it up - results in
a better cache usage.
Without RPS (echo 0 > rps_cpus) BCM4908 NAT masq performance "jumps"
between two speeds: ~900 Mbps and 940 Mbps (it's a 4 CPUs SoC). This
change bumps the lower speed from 905 Mb/s to 918 Mb/s (tested using
single stream iperf 2.0.5 traffic).
There are more optimizations to consider. One obvious to try is GRO
however as BCM4908 doesn't do hw csum is may actually lower performance.
Sometimes. Some early testing:
┌─────────────────────────────────┬─────────────────────┬────────────────────┐
│ │ netif_receive_skb() │ napi_gro_receive() │
├─────────────────────────────────┼─────────────────────┼────────────────────┤
│ netdev_alloc_skb() │ 905 Mb/s │ 892 Mb/s │
│ napi_alloc_frag() + build_skb() │ 918 Mb/s │ 917 Mb/s │
└─────────────────────────────────┴─────────────────────┴────────────────────┘
Another ideas:
1. napi_build_skb()
2. skb_copy_from_linear_data() for small packets
Those need proper testing first though. That can be done later.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Link: https://lore.kernel.org/r/20221025132245.22871-1-zajec5@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
If of_device_register() returns error, the of node and the
name allocated in dev_set_name() is leaked, call put_device()
to give up the reference that was set in device_initialize(),
so that of node is put in logical_port_release() and the name
is freed in kobject_cleanup().
Fixes: 1acf2318dd ("ehea: dynamic add / remove port")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Link: https://lore.kernel.org/r/20221025130011.1071357-1-yangyingliang@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Bond agnostically interacts with TLS device-offload requests via the
.ndo_sk_get_lower_dev operation. Return value is true iff bond
guarantees fixed mapping between the TLS connection and a lower netdev.
Due to this nature, the bond TLS device offload features are not
explicitly controllable in the bond layer. As of today, these are
read-only values based on the evaluation of bond_sk_check(). However,
this indication might be incorrect and misleading, when the feature bits
are "fixed" by some dependency features. For example,
NETIF_F_HW_TLS_TX/RX are forcefully cleared in case the corresponding
checksum offload is disabled. But in fact the bond ability to still
offload TLS connections to the lower device is not hurt.
This means that these bits can not be trusted, and hence better become
unused.
This patch revives some old discussion [1] and proposes a much simpler
solution: Clear the bond's TLS features bits. Everyone should stop
reading them.
[1] https://lore.kernel.org/netdev/20210526095747.22446-1-tariqt@nvidia.com/
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Gal Pressman <gal@nvidia.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20221025105300.4718-1-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
It doesn't need extra macros for queue 0 & 4. Same macro could
be used for all 8 queues. Related queue/channel functions could
be combined together.
Original macro which has two same parameters is unsafe macro and
might have potential side effects. Each MTL RxQ DMA channel mask
is 4 bits, so using (0xf << chan) instead of GENMASK(x + 3, x) to
avoid unsafe macro.
Signed-off-by: Junxiao Chang <junxiao.chang@intel.com>
Link: https://lore.kernel.org/r/20221025081747.1884926-1-junxiao.chang@intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
