Saeed Mahameed says:
====================
mlx5-updates-2021-10-25
Misc updates for mlx5 driver:
1) Misc updates and cleanups:
- Don't write directly to netdev->dev_addr, From Jakub Kicinski
- Remove unnecessary checks for slow path flag in tc module
- Fix unused function warning of mlx5i_flow_type_mask
- Bridge, support replacing existing FDB entry
2) Sub Functions, Reduction in memory usage:
- Reduce flow counters bulk query buffer size
- Implement max_macs devlink parameter
- Add devlink vendor params to control Event Queue sizes
- Added SF life cycle trace points by Parav/
3) From Aya, Firmware health buffer reporting improvements
- Print health buffer by log level and more missing information
- Periodic update of host time to firmware
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
v1: Implement a more general statement as recommended by Eric Dumazet. The
sequence number will be advanced, so this check will fix the FIN case and
other cases.
A customer reported sockets stuck in the CLOSING state. A Vmcore revealed that
the write_queue was not empty as determined by tcp_write_queue_empty() but the
sk_buff containing the FIN flag had been freed and the socket was zombied in
that state. Corresponding pcaps show no FIN from the Linux kernel on the wire.
Some instrumentation was added to the kernel and it was found that there is a
timing window where tcp_sendmsg() can run after tcp_send_fin().
tcp_sendmsg() will hit an error, for example:
1269 ▹ if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))↩
1270 ▹ ▹ goto do_error;↩
tcp_remove_empty_skb() will then free the FIN sk_buff as "skb->len == 0". The
TCP socket is now wedged in the FIN-WAIT-1 state because the FIN is never sent.
If the other side sends a FIN packet the socket will transition to CLOSING and
remain that way until the system is rebooted.
Fix this by checking for the FIN flag in the sk_buff and don't free it if that
is the case. Testing confirmed that fixed the issue.
Fixes: fdfc5c8594 ("tcp: remove empty skb from write queue in error cases")
Signed-off-by: Jon Maxwell <jmaxwell37@gmail.com>
Reported-by: Monir Zouaoui <Monir.Zouaoui@mail.schwarz>
Reported-by: Simon Stier <simon.stier@mail.schwarz>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the check of !rc in (!rc && !resc_lock_params.b_granted) since it
is always true.
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Remove the check of !rc in (!rc && !params.b_granted) since it is always
true.
We should also use constant 0 for return.
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eric Dumazet says:
====================
tcp: receive path optimizations
This series aims to reduce cache line misses in RX path.
I am still working on better cache locality in tcp_sock but
this will wait few more weeks.
====================
Link: https://lore.kernel.org/r/20211025164825.259415-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Two kfree_skb() calls must be replaced by consume_skb()
for skbs that are not technically dropped.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
RFC 5082 IP_MINTTL option is rarely used on hosts.
Add a static key to remove from TCP fast path useless code,
and potential cache line miss to fetch inet_sk(sk)->min_ttl
Note that once ip4_min_ttl static key has been enabled,
it stays enabled until next boot.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
No report yet from KCSAN, yet worth documenting the races.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
RFC 5082 IPV6_MINHOPCOUNT is rarely used on hosts.
Add a static key to remove from TCP fast path useless code,
and potential cache line miss to fetch tcp_inet6_sk(sk)->min_hopcount
Note that once ip6_min_hopcount static key has been enabled,
it stays enabled until next boot.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
No report yet from KCSAN, yet worth documenting the races.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sk->sk_rx_queue_mapping can be modified locklessly,
add a couple of READ_ONCE()/WRITE_ONCE() to document this fact.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sk_rx_queue_mapping is located in a cache line that should be kept read mostly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sk_napi_id is located in a cache line that can be kept read mostly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Increase cache locality by moving rx_dst_coookie next to sk->sk_rx_dst
This removes one or two cache line misses in IPv6 early demux (TCP/UDP)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Increase cache locality by moving rx_dst_ifindex next to sk->sk_rx_dst
This is part of an effort to reduce cache line misses in TCP fast path.
This removes one cache line miss in early demux.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
rx_dropped, tx_dropped, rx_frame_errors and rx_crc_errors are being
wrongly fetched from the target container rather than source percpu
ones.
No idea if that goes from the vendor driver or was brainoed during
the refactoring, but fix it either way.
Fixes: a97c69ba4f ("net: ax88796c: ASIX AX88796C SPI Ethernet Adapter Driver")
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Acked-by: Łukasz Stelmach <l.stelmach@samsung.com>
Link: https://lore.kernel.org/r/20211023121148.113466-1-alobakin@pm.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently, max_macs is taking 70Kbytes of memory per function. This
size is not needed in all use cases, and is critical with large scale.
Hence, allow user to configure the number of max_macs.
For example, to reduce the number of max_macs to 1, execute::
$ devlink dev param set pci/0000:00:0b.0 name max_macs value 1 \
cmode driverinit
$ devlink dev reload pci/0000:00:0b.0
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Event EQ is an EQ which received the notification of almost all the
events generated by the NIC.
Currently, each event EQ is taking 512KB of memory. This size is not
needed in most use cases, and is critical with large scale. Hence,
allow user to configure the size of the event EQ.
For example to reduce event EQ size to 64, execute::
$ devlink resource set pci/0000:00:0b.0 path /event_eq_size/ size 64
$ devlink dev reload pci/0000:00:0b.0
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Currently, each I/O EQ is taking 128KB of memory. This size
is not needed in all use cases, and is critical with large scale.
Hence, allow user to configure the size of I/O EQs.
For example, to reduce I/O EQ size to 64, execute:
$ devlink resource set pci/0000:00:0b.0 path /io_eq_size/ size 64
$ devlink dev reload pci/0000:00:0b.0
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The SWITCHDEV_FDB_ADD_TO_DEVICE is used for both adding new and replacing
existing entry. Implement support for replacing existing FDB entries in
mlx5 offload code.
Signed-off-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Following two patterns in bridge code are used in multiple places where
similar code is duplicated:
- Lookup FDB entry from hashtable by address+vid pair.
- Notify software bridge and then delete existing FDB entry.
In order to improve code quality and prepare for following patch series
that also uses described patterns, extract the codes to dedicated helper
functions.
This commit doesn't change functionality.
Signed-off-by: Vlad Buslov <vladbu@nvidia.com>
Reviewed-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Firmware logs its asserts also to non-volatile memory. In order to
reduce drift between the NIC and the host, the driver sets the host
epoch-time to the firmware every hour.
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Add log macro which gets log level as a parameter. Use the severity
read from the health buffer and the new log macro to log the health buffer
with severity as log level. Prior to this patch, health buffer was
printed in error log level regardless of its severity. Now the user may
filter dmesg (--level) or change kernel log level to focus on different
severity levels of firmware errors.
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Currently, the flow counters bulk query buffer takes a little more than
512KB of memory, which is aligned to the next power of 2, to 1MB.
The buffer size determines the maximum number of flow counters that can
be queried at a time. Thus, having a bigger buffer can improve
performance for users that need to query many flow counters.
SFs don't use many flow counters and don't need a big buffer. Since this
size is critical with large scale, reduce the size of the bulk query
buffer for SFs.
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The cited commit is causing unused-function warning[1] when
CONFIG_MLX5_EN_RXNFC is not set.
Fix this by moving the function into the ifdef, where it's only used
[1]
warning: ‘mlx5i_flow_type_mask’ defined but not used [-Wunused-function]
Fixes: 9fbe1c25ec ("net/mlx5i: Enable Rx steering for IPoIB via ethtool")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
After previous changes, caller (mlx5e_tc_offload_fdb_rules()) already
checks for the slow path flag, and if set won't call offload/unoffload
sample.
Signed-off-by: Paul Blakey <paulb@nvidia.com>
Reviewed-by: Maor Dickman <maord@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Commit 406f42fa0d ("net-next: When a bond have a massive amount
of VLANs...") introduced a rbtree for faster Ethernet address look
up. To maintain netdev->dev_addr in this tree we need to make all
the writes to it go through appropriate helpers.
Reviewed-by: Marcel Holtmann <marcel@holtmann.org>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 406f42fa0d ("net-next: When a bond have a massive amount
of VLANs...") introduced a rbtree for faster Ethernet address look
up. To maintain netdev->dev_addr in this tree we need to make all
the writes to it go through appropriate helpers.
Convert bluetooth from memcpy(... ETH_ADDR) to eth_hw_addr_set():
@@
expression dev, np;
@@
- memcpy(dev->dev_addr, np, ETH_ALEN)
+ eth_hw_addr_set(dev, np)
Reviewed-by: Marcel Holtmann <marcel@holtmann.org>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
hw_addr is a uint AKA unsigned int. dev_addr_set() takes
a u8 *.
drivers/net/fddi/defza.c:1383:27: error: passing argument 2 of 'dev_addr_set' from incompatible pointer type [-Werror=incompatible-pointer-types]
Reported-by: kernel test robot <lkp@intel.com>
Fixes: 1e9258c389 ("fddi: defxx,defza: use dev_addr_set()")
Acked-by: Maciej W. Rozycki <macro@orcam.me.uk>
Link: https://lore.kernel.org/r/20211025160000.2803818-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
AES_CCM_128 and CHACHA20_POLY1305 are already supported by tls,
similar to setsockopt, getsockopt also needs to support these
two algorithms.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tls already supports the SM4 GCM/CCM algorithms. It is also necessary
to add support for these two algorithms in tls_crypto_context to avoid
potential issues caused by forced type conversion.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use 'bitmap_zalloc()' to simplify code, improve the semantic and avoid
some open-coded arithmetic in allocator arguments.
Also change the corresponding 'kfree()' into 'bitmap_free()' to keep
consistency.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 406f42fa0d ("net-next: When a bond have a massive amount
of VLANs...") introduced a rbtree for faster Ethernet address look
up. To maintain netdev->dev_addr in this tree we need to make all
the writes to it got through appropriate helpers.
In the case of catc we need a new temporary buffer to conform
to the rules for DMA coherency. That in turn necessitates
a reworking of error handling in probe().
Signed-off-by: Oliver Neukum <oneukum@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Kalle Valo says:
====================
wireless-drivers-next patches for v5.16
Third set of patches for v5.16. This time we have a small one to
quickly fix two mt76 build failures I had missed in my previous pull
request.
Major changes:
mt76
* fix linking when CONFIG_MMC is disabled
* fix dev_err() format warning
* mt7615: mt7622: fix ibss and meshpoint
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeroen de Borst says:
====================
gve: Add jumbo-frame support for GQ
This patchset introduces jumbo-frame support for the GQ queue format.
The device already supports jumbo-frames on TX. This introduces
multi-descriptor RX packets using a packet continuation bit.
A widely deployed driver has a bug with causes it to fail to load
when a MTU greater than 2048 bytes is configured. A jumbo-frame device
option is introduced to pass a jumbo-frame MTU only to drivers that
support it.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
A widely deployed driver has a bug that will cause the driver not
to load when a max_mtu > 2048 is present in the device descriptor.
To avoid this bug while still enabling jumbo frames, we present a lower
max_mtu in the device descriptor and pass the actual max_mtu in
a separate device option.
The driver supports 2 different queue formats. To enable features
on one queue format, but not the other, a supported_features mask
was added to the device options in the device descriptor.
Signed-off-by: Shailend Chand <shailend@google.com>
Signed-off-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This enables the driver to receive RX packets spread across multiple
buffers:
For a given multi-fragment packet the "packet continuation" bit is set
on all descriptors except the last one. These descriptors' payloads are
combined into a single SKB before the SKB is handed to the
networking stack.
This change adds a "packet buffer size" notion for RX queues. The
CreateRxQueue AdminQueue command sent to the device now includes the
packet_buffer_size.
We opt for a packet_buffer_size of PAGE_SIZE / 2 to give the
driver the opportunity to flip pages where we can instead of copying.
Signed-off-by: David Awogbemila <awogbemila@google.com>
Signed-off-by: Jeroen de Borst <jeroendb@google.com>
Reviewed-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This refactor moves the skb_head and skb_tail fields into a new
gve_rx_ctx struct. This new struct will contain information about the
current packet being processed. This is in preparation for
multi-descriptor RX packets.
Signed-off-by: David Awogbemila <awogbemila@google.com>
Signed-off-by: Jeroen de Borst <jeroendb@google.com>
Reviewed-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ido Schimmel says:
====================
selftests: mlxsw: Various updates
This patchset contains various updates to mlxsw selftests.
Patch #1 replaces open-coded compatibility checks with dedicated
helpers. These helpers are used to skip tests when run on incompatible
machines.
Patch #2 avoids spurious failures in some tests by using permanent
neighbours instead of reachable ones.
Patch #3 reduces the run time of a test by not iterating over all the
available trap policers.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of iterating over all the available trap policers, only perform
the tests with three policers: The first, the last and the one in the
middle of the range. On a Spectrum-3 system, this reduces the run time
from almost an hour to a few minutes.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The nexthop objects tests configure dummy reachable neighbours so that
the nexthops will have a MAC address and be programmed to the device.
Since these are dummy reachable neighbours, they can be transitioned by
the kernel to a failed state if they are around for too long. This can
happen, for example, if the "TIMEOUT" variable is configured with a too
high value.
Make the tests more robust by configuring the neighbours as permanent,
so that the tests do not depend on the configured timeout value.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A number of mlxsw-specific selftests currently detect whether they are run
on a compatible machine, and bail out silently when not. These tests are
however done in a somewhat impenetrable manner by directly comparing PCI
IDs against a blacklist or a whitelist, and bailing out silently if the
machine is not compatible.
Instead, add a helper, mlxsw_only_on_spectrum(), which allows specifying
the supported machines in a human-readable manner. If the current machine
is incompatible, the helper emits a SKIP message and returns an error code,
based on which the caller can gracefully bail out in a suitable way. This
allows a more readable conditions such as:
mlxsw_only_on_spectrum 2+ || return
Convert all existing open-coded guards to the new helper. Also add two new
guards to do_mark_test() and do_drop_test(), which are supported only on
Spectrum-2+, but the corresponding check was not there.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Luo Jie says:
====================
net: phy: Add qca8081 ethernet phy driver
This patch series add the qca8081 ethernet phy driver support, which
improve the wol feature, leverage at803x phy driver and add the fast
retrain, master/slave seed and CDT feature.
Changes in v7:
* update Reviewed-by tags.
Changes in v6:
* add Reviewed-by tags on the applicable patches.
Changes in v5:
* rebase the patches on net-next/master.
Changes in v4:
* handle other interrupts in set_wol.
* add genphy_c45_fast_retrain.
Changes in v3:
* correct a typo "excpet".
* remove the suffix "PHY" from phy name.
Changes in v2:
* add definitions of fast retrain related registers in mdio.h.
* break up the patch into small patches.
* improve the at803x legacy code.
Changes in v1:
* merge qca8081 phy driver into at803x.
* add cdt feature.
* leverage at803x phy driver helpers.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
To perform CDT of qca8081 phy:
1. disable hibernation.
2. force phy working in MDI mode.
3. force phy working in 1000BASE-T mode.
4. configure the related thresholds.
Signed-off-by: Luo Jie <luoj@codeaurora.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
