| Age | Commit message (Collapse) | Author |
|---|
|
Drivers like mlx5 expose NIC's vlan_features such as
NETIF_F_GSO_UDP_TUNNEL & NETIF_F_GSO_UDP_TUNNEL_CSUM which are
later not propagated when the underlying devices are bonded and
a vlan device created on top of the bond.
Right now, the more cumbersome workaround for this is to create
the vlan on top of the mlx5 and then enslave the vlan devices
to a bond.
To fix this, add NETIF_F_GSO_ENCAP_ALL to BOND_VLAN_FEATURES
such that bond_compute_features() can probe and propagate the
vlan_features from the slave devices up to the vlan device.
Given the following bond:
# ethtool -i enp2s0f{0,1}np{0,1}
driver: mlx5_core
[...]
# ethtool -k enp2s0f0np0 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: on
rx-udp-gro-forwarding: off
# ethtool -k enp2s0f1np1 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: on
rx-udp-gro-forwarding: off
# ethtool -k bond0 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: off [fixed]
rx-udp-gro-forwarding: off
Before:
# ethtool -k bond0.100 | grep udp
tx-udp_tnl-segmentation: off [requested on]
tx-udp_tnl-csum-segmentation: off [requested on]
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: off [fixed]
rx-udp-gro-forwarding: off
After:
# ethtool -k bond0.100 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: off [fixed]
rx-udp-gro-forwarding: off
Various users have run into this reporting performance issues when
configuring Cilium in vxlan tunneling mode and having the combination
of bond & vlan for the core devices connecting the Kubernetes cluster
to the outside world.
Fixes: a9b3ace44c7d ("bonding: fix vlan_features computing")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: Ido Schimmel <idosch@idosch.org>
Cc: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://patch.msgid.link/20241210141245.327886-3-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
If a bonding device has slave devices, then the current logic to derive
the feature set for the master bond device is limited in that flags which
are fully supported by the underlying slave devices cannot be propagated
up to vlan devices which sit on top of bond devices. Instead, these get
blindly masked out via current NETIF_F_ALL_FOR_ALL logic.
vlan_features and mpls_features should reuse netdev_base_features() in
order derive the set in the same way as ndo_fix_features before iterating
through the slave devices to refine the feature set.
Fixes: a9b3ace44c7d ("bonding: fix vlan_features computing")
Fixes: 2e770b507ccd ("net: bonding: Inherit MPLS features from slave devices")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: Ido Schimmel <idosch@idosch.org>
Cc: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://patch.msgid.link/20241210141245.327886-2-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Both bonding and team driver have logic to derive the base feature
flags before iterating over their slave devices to refine the set
via netdev_increment_features().
Add a small helper netdev_base_features() so this can be reused
instead of having it open-coded multiple times.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: Ido Schimmel <idosch@idosch.org>
Cc: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20241210141245.327886-1-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
|
Russell King says:
====================
net: dsa: cleanup EEE (part 1)
First part of DSA EEE cleanups.
Patch 1 removes a useless test that is always false. dp->pl will always
be set for user ports, so !dp->pl in the EEE methods will always be
false.
Patch 2 adds support for a new DSA support_eee() method, which tells
DSA whether the DSA driver supports EEE, and thus whether the ethtool
set_eee() and get_eee() methods should return -EOPNOTSUPP.
Patch 3 adds a trivial implementation for this new method which
indicates that EEE is supported.
Patches 4..8 adds implementations for .supports_eee() to all drivers
that support EEE in some form.
Patch 9 switches the core DSA code to require a .supports_eee()
implementation if DSA is supported. Any DSA driver that doesn't
implement this method after this patch will not support the ethtool
EEE methods.
Part 2 will remove the (now) useless .get_mac_eee() DSA operation.
====================
Link: https://patch.msgid.link/Z1hNkEb13FMuDQiY@shell.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Now that we have updated all drivers, switch DSA to require an
implementation of the .support_eee() method for EEE to be usable,
rather than defaulting to being permissive when not implemented.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL14e-006cZy-AT@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Implement the .support_eee() method by reusing the ksz_validate_eee()
method as a template, renaming the function, changing the return type
and values, and removing it from the ksz_set_mac_eee() and
ksz_get_mac_eee() methods.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL14Z-006cZs-6o@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Implement the .support_eee() method by using the generic helper as all
user ports support EEE.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL14U-006cZm-2K@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Implement the .support_eee() method by using the generic helper as all
user ports support EEE.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL14O-006cZg-VM@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Implement the .support_eee() method by using the generic helper as all
user ports support EEE.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL14J-006cZa-Rh@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Implement the .support_eee() method to indicate that EEE is not
supported by two switch variants, rather than making these checks in
the .set_mac_eee() and .get_mac_eee() methods.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL14E-006cZU-Nc@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Provide a trivial implementation for the .support_eee() method which
switch drivers can use to simply indicate that they support EEE on
all their user ports.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL149-006cZJ-JJ@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add a hook to determine whether the switch supports EEE. This will
return false if the switch does not, or true if it does. If the
method is not implemented, we assume (currently) that the switch
supports EEE.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL144-006cZD-El@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
When user ports are initialised, a phylink instance is always created,
and so dp->pl will always be non-NULL. The EEE methods are only used
for user ports, so checking for dp->pl to be NULL makes no sense. No
other phylink-calling method implements similar checks in DSA. Remove
this unnecessary check.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/E1tL13z-006cZ7-BZ@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
In general, 'qlen' of any classful qdisc should keep track of the
number of packets that the qdisc itself and all of its children holds.
In case of netem, 'qlen' only accounts for the packets in its internal
tfifo. When netem is used with a child qdisc, the child qdisc can use
'qdisc_tree_reduce_backlog' to inform its parent, netem, about created
or dropped SKBs. This function updates 'qlen' and the backlog statistics
of netem, but netem does not account for changes made by a child qdisc.
'qlen' then indicates the wrong number of packets in the tfifo.
If a child qdisc creates new SKBs during enqueue and informs its parent
about this, netem's 'qlen' value is increased. When netem dequeues the
newly created SKBs from the child, the 'qlen' in netem is not updated.
If 'qlen' reaches the configured sch->limit, the enqueue function stops
working, even though the tfifo is not full.
Reproduce the bug:
Ensure that the sender machine has GSO enabled. Configure netem as root
qdisc and tbf as its child on the outgoing interface of the machine
as follows:
$ tc qdisc add dev <oif> root handle 1: netem delay 100ms limit 100
$ tc qdisc add dev <oif> parent 1:0 tbf rate 50Mbit burst 1542 latency 50ms
Send bulk TCP traffic out via this interface, e.g., by running an iPerf3
client on the machine. Check the qdisc statistics:
$ tc -s qdisc show dev <oif>
Statistics after 10s of iPerf3 TCP test before the fix (note that
netem's backlog > limit, netem stopped accepting packets):
qdisc netem 1: root refcnt 2 limit 1000 delay 100ms
Sent 2767766 bytes 1848 pkt (dropped 652, overlimits 0 requeues 0)
backlog 4294528236b 1155p requeues 0
qdisc tbf 10: parent 1:1 rate 50Mbit burst 1537b lat 50ms
Sent 2767766 bytes 1848 pkt (dropped 327, overlimits 7601 requeues 0)
backlog 0b 0p requeues 0
Statistics after the fix:
qdisc netem 1: root refcnt 2 limit 1000 delay 100ms
Sent 37766372 bytes 24974 pkt (dropped 9, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
qdisc tbf 10: parent 1:1 rate 50Mbit burst 1537b lat 50ms
Sent 37766372 bytes 24974 pkt (dropped 327, overlimits 96017 requeues 0)
backlog 0b 0p requeues 0
tbf segments the GSO SKBs (tbf_segment) and updates the netem's 'qlen'.
The interface fully stops transferring packets and "locks". In this case,
the child qdisc and tfifo are empty, but 'qlen' indicates the tfifo is at
its limit and no more packets are accepted.
This patch adds a counter for the entries in the tfifo. Netem's 'qlen' is
only decreased when a packet is returned by its dequeue function, and not
during enqueuing into the child qdisc. External updates to 'qlen' are thus
accounted for and only the behavior of the backlog statistics changes. As
in other qdiscs, 'qlen' then keeps track of how many packets are held in
netem and all of its children. As before, sch->limit remains as the
maximum number of packets in the tfifo. The same applies to netem's
backlog statistics.
Fixes: 50612537e9ab ("netem: fix classful handling")
Signed-off-by: Martin Ottens <martin.ottens@fau.de>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20241210131412.1837202-1-martin.ottens@fau.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
git://git.open-mesh.org/linux-merge
Simon Wunderlich says:
====================
Here are some batman-adv bugfixes:
- fix TT unitialized data and size limit issues, by Remi Pommarel
(3 patches)
* tag 'batadv-net-pullrequest-20241210' of git://git.open-mesh.org/linux-merge:
batman-adv: Do not let TT changes list grows indefinitely
batman-adv: Remove uninitialized data in full table TT response
batman-adv: Do not send uninitialized TT changes
====================
Link: https://patch.msgid.link/20241210135024.39068-1-sw@simonwunderlich.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
With this port schedule:
tc qdisc replace dev $send_if parent root handle 100 taprio \
num_tc 8 queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
map 0 1 2 3 4 5 6 7 \
base-time 0 cycle-time 10000 \
sched-entry S 01 1250 \
sched-entry S 02 1250 \
sched-entry S 04 1250 \
sched-entry S 08 1250 \
sched-entry S 10 1250 \
sched-entry S 20 1250 \
sched-entry S 40 1250 \
sched-entry S 80 1250 \
flags 2
ptp4l would fail to take TX timestamps of Pdelay_Resp messages like this:
increasing tx_timestamp_timeout may correct this issue, but it is likely caused by a driver bug
ptp4l[4134.168]: port 2: send peer delay response failed
It turns out that the driver can't take their TX timestamps because it
can't transmit them in the first place. And there's nothing special
about the Pdelay_Resp packets - they're just regular 68 byte packets.
But with this taprio configuration, the switch would refuse to send even
the ETH_ZLEN minimum packet size.
This should have definitely not been the case. When applying the taprio
config, the driver prints:
mscc_felix 0000:00:00.5: port 0 tc 0 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 1 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 2 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 3 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 4 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 5 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 6 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 7 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
and thus, everything under 132 bytes - ETH_FCS_LEN should have been sent
without problems. Yet it's not.
For the forwarding path, the configuration is fine, yet packets injected
from Linux get stuck with this schedule no matter what.
The first hint that the static guard bands are the cause of the problem
is that reverting Michael Walle's commit 297c4de6f780 ("net: dsa: felix:
re-enable TAS guard band mode") made things work. It must be that the
guard bands are calculated incorrectly.
I remembered that there is a magic constant in the driver, set to 33 ns
for no logical reason other than experimentation, which says "never let
the static guard bands get so large as to leave less than this amount of
remaining space in the time slot, because the queue system will refuse
to schedule packets otherwise, and they will get stuck". I had a hunch
that my previous experimentally-determined value was only good for
packets coming from the forwarding path, and that the CPU injection path
needed more.
I came to the new value of 35 ns through binary search, after seeing
that with 544 ns (the bit time required to send the Pdelay_Resp packet
at gigabit) it works. Again, this is purely experimental, there's no
logic and the manual doesn't say anything.
The new driver prints for this schedule look like this:
mscc_felix 0000:00:00.5: port 0 tc 0 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 1 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 2 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 3 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 4 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 5 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 6 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 7 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
So yes, the maximum MTU is now even smaller by 1 byte than before.
This is maybe counter-intuitive, but makes more sense with a diagram of
one time slot.
Before:
Gate open Gate close
| |
v 1250 ns total time slot duration v
<---------------------------------------------------->
<----><---------------------------------------------->
33 ns 1217 ns static guard band
useful
Gate open Gate close
| |
v 1250 ns total time slot duration v
<---------------------------------------------------->
<-----><--------------------------------------------->
35 ns 1215 ns static guard band
useful
The static guard band implemented by this switch hardware directly
determines the maximum allowable MTU for that traffic class. The larger
it is, the earlier the switch will stop scheduling frames for
transmission, because otherwise they might overrun the gate close time
(and avoiding that is the entire purpose of Michael's patch).
So, we now have guard bands smaller by 2 ns, thus, in this particular
case, we lose a byte of the maximum MTU.
Fixes: 11afdc6526de ("net: dsa: felix: tc-taprio intervals smaller than MTU should send at least one packet")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Michael Walle <mwalle@kernel.org>
Link: https://patch.msgid.link/20241210132640.3426788-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Rather than maintaining a private copy of the LPI timer, make use of
the LPI timer maintained by phylib. In any case, phylib overwrites the
value of tx_lpi_timer set by the driver in phy_ethtool_get_eee().
Note that feb->eee.tx_lpi_timer is initialised to zero, which is just
the same with phylib's copy, so there should be no functional change.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Tested-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/E1tKzVS-006c67-IJ@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The device does not support multiple mirror actions per rule and the
driver rejects such configuration:
# tc filter add dev swp1 ingress pref 1 proto ip flower skip_sw action mirred egress mirror dev swp2 action mirred egress mirror dev swp3
Error: mlxsw_spectrum: Multiple mirror actions per rule are not supported.
We have an error talking to the kernel
Internally, the sample action is implemented by the device by mirroring
to the CPU port. Therefore, mixing sample and mirror actions in a single
rule does not work correctly and results in the last action effect.
Solve by rejecting such misconfiguration:
# tc filter add dev swp1 ingress pref 1 proto ip flower skip_sw action mirred egress mirror dev swp2 action sample rate 100 group 1
Error: mlxsw_spectrum: Sample action after mirror action is not supported.
We have an error talking to the kernel
# tc filter add dev swp1 ingress pref 1 proto ip flower skip_sw action sample rate 100 group 1 action mirred egress mirror dev swp2
Error: mlxsw_spectrum: Mirror action after sample action is not supported.
We have an error talking to the kernel
Reported-by: Vladyslav Mykhaliuk <vmykhaliuk@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Link: https://patch.msgid.link/d6c979914e8706dbe1dedbaf29ffffb0b8d71166.1733822570.git.petrm@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
When `skb_splice_from_iter` was introduced, it inadvertently added
checksumming for AF_UNIX sockets. This resulted in significant
slowdowns, for example when using sendfile over unix sockets.
Using the test code in [1] in my test setup (2G single core qemu),
the client receives a 1000M file in:
- without the patch: 1482ms (+/- 36ms)
- with the patch: 652.5ms (+/- 22.9ms)
This commit addresses the issue by marking checksumming as unnecessary in
`unix_stream_sendmsg`
Cc: stable@vger.kernel.org
Signed-off-by: Frederik Deweerdt <deweerdt.lkml@gmail.com>
Fixes: 2e910b95329c ("net: Add a function to splice pages into an skbuff for MSG_SPLICE_PAGES")
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Joe Damato <jdamato@fastly.com>
Link: https://patch.msgid.link/Z1fMaHkRf8cfubuE@xiberoa
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add the following Telit FE910C04 compositions:
0x10c0: rmnet + tty (AT/NMEA) + tty (AT) + tty (diag)
T: Bus=02 Lev=01 Prnt=03 Port=06 Cnt=01 Dev#= 13 Spd=480 MxCh= 0
D: Ver= 2.00 Cls=00(>ifc ) Sub=00 Prot=00 MxPS=64 #Cfgs= 1
P: Vendor=1bc7 ProdID=10c0 Rev=05.15
S: Manufacturer=Telit Cinterion
S: Product=FE910
S: SerialNumber=f71b8b32
C: #Ifs= 4 Cfg#= 1 Atr=e0 MxPwr=500mA
I: If#= 0 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=50 Driver=qmi_wwan
E: Ad=01(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=81(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=82(I) Atr=03(Int.) MxPS= 8 Ivl=32ms
I: If#= 1 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=60 Driver=option
E: Ad=02(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=83(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=84(I) Atr=03(Int.) MxPS= 10 Ivl=32ms
I: If#= 2 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=40 Driver=option
E: Ad=03(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=85(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=86(I) Atr=03(Int.) MxPS= 10 Ivl=32ms
I: If#= 3 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=30 Driver=option
E: Ad=04(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=87(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
0x10c4: rmnet + tty (AT) + tty (AT) + tty (diag)
T: Bus=02 Lev=01 Prnt=03 Port=06 Cnt=01 Dev#= 14 Spd=480 MxCh= 0
D: Ver= 2.00 Cls=00(>ifc ) Sub=00 Prot=00 MxPS=64 #Cfgs= 1
P: Vendor=1bc7 ProdID=10c4 Rev=05.15
S: Manufacturer=Telit Cinterion
S: Product=FE910
S: SerialNumber=f71b8b32
C: #Ifs= 4 Cfg#= 1 Atr=e0 MxPwr=500mA
I: If#= 0 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=50 Driver=qmi_wwan
E: Ad=01(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=81(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=82(I) Atr=03(Int.) MxPS= 8 Ivl=32ms
I: If#= 1 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=40 Driver=option
E: Ad=02(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=83(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=84(I) Atr=03(Int.) MxPS= 10 Ivl=32ms
I: If#= 2 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=40 Driver=option
E: Ad=03(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=85(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=86(I) Atr=03(Int.) MxPS= 10 Ivl=32ms
I: If#= 3 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=30 Driver=option
E: Ad=04(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=87(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
0x10c8: rmnet + tty (AT) + tty (diag) + DPL (data packet logging) + adb
T: Bus=02 Lev=01 Prnt=03 Port=06 Cnt=01 Dev#= 17 Spd=480 MxCh= 0
D: Ver= 2.00 Cls=00(>ifc ) Sub=00 Prot=00 MxPS=64 #Cfgs= 1
P: Vendor=1bc7 ProdID=10c8 Rev=05.15
S: Manufacturer=Telit Cinterion
S: Product=FE910
S: SerialNumber=f71b8b32
C: #Ifs= 5 Cfg#= 1 Atr=e0 MxPwr=500mA
I: If#= 0 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=50 Driver=qmi_wwan
E: Ad=01(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=81(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=82(I) Atr=03(Int.) MxPS= 8 Ivl=32ms
I: If#= 1 Alt= 0 #EPs= 3 Cls=ff(vend.) Sub=ff Prot=40 Driver=option
E: Ad=02(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=83(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=84(I) Atr=03(Int.) MxPS= 10 Ivl=32ms
I: If#= 2 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=ff Prot=30 Driver=option
E: Ad=03(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=85(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
I: If#= 3 Alt= 0 #EPs= 1 Cls=ff(vend.) Sub=ff Prot=80 Driver=(none)
E: Ad=86(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
I: If#= 4 Alt= 0 #EPs= 2 Cls=ff(vend.) Sub=42 Prot=01 Driver=(none)
E: Ad=04(O) Atr=02(Bulk) MxPS= 512 Ivl=0ms
E: Ad=87(I) Atr=02(Bulk) MxPS= 512 Ivl=0ms
Signed-off-by: Daniele Palmas <dnlplm@gmail.com>
Link: https://patch.msgid.link/20241209151821.3688829-1-dnlplm@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Maxim Levitsky says:
====================
MANA: Fix few memory leaks in mana_gd_setup_irqs
Fix 2 minor memory leaks in the mana driver,
introduced by commit
====================
Link: https://patch.msgid.link/20241209175751.287738-1-mlevitsk@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
gc->irq_contexts is not freeded if one of the later operations
fail.
Suggested-by: Michael Kelley <mhklinux@outlook.com>
Fixes: 8afefc361209 ("net: mana: Assigning IRQ affinity on HT cores")
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Saurabh Sengar <ssengar@linux.microsoft.com>
Reviewed-by: Yury Norov <yury.norov@gmail.com>
Link: https://patch.msgid.link/20241209175751.287738-3-mlevitsk@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Commit 8afefc361209 ("net: mana: Assigning IRQ affinity on HT cores")
added memory allocation in mana_gd_setup_irqs of 'irqs' but the code
doesn't free this temporary array in the success path.
This was caught by kmemleak.
Fixes: 8afefc361209 ("net: mana: Assigning IRQ affinity on HT cores")
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Saurabh Sengar <ssengar@linux.microsoft.com>
Reviewed-by: Yury Norov <yury.norov@gmail.com>
Link: https://patch.msgid.link/20241209175751.287738-2-mlevitsk@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Jakub Sitnicki says:
====================
Make TIME-WAIT reuse delay deterministic and configurable
This patch set is an effort to enable faster reuse of TIME-WAIT sockets.
We have recently talked about the motivation and the idea at Plumbers [1].
Experiment in production
------------------------
We are restarting our experiment on a small set of production nodes as the
code has slightly changed since v1 [2], and there are still a few weeks of
development window to soak the changes. We will report back if we observe
any regressions.
Packetdrill tests
-----------------
The packetdrill tests for TIME-WAIT reuse [3] did not change since v1.
Although we are not touching PAWS code any more, I would still like to add
tests to cover PAWS reject after TW reuse. This, however, requires patching
packetdrill as I mentioned in the last cover letter [2].
[1] https://lpc.events/event/18/contributions/1962/
[2] https://lore.kernel.org/r/20241113-jakub-krn-909-poc-msec-tw-tstamp-v2-0-b0a335247304@cloudflare.com
[3] https://github.com/google/packetdrill/pull/90
v1: https://lore.kernel.org/20241204-jakub-krn-909-poc-msec-tw-tstamp-v1-0-8b54467a0f34@cloudflare.com
RFCv2: https://lore.kernel.org/20241113-jakub-krn-909-poc-msec-tw-tstamp-v2-0-b0a335247304@cloudflare.com
RFCv1: https://lore.kernel.org/20240819-jakub-krn-909-poc-msec-tw-tstamp-v1-1-6567b5006fbe@cloudflare.com
====================
Link: https://patch.msgid.link/20241209-jakub-krn-909-poc-msec-tw-tstamp-v2-0-66aca0eed03e@cloudflare.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Today we have a hardcoded delay of 1 sec before a TIME-WAIT socket can be
reused by reopening a connection. This is a safe choice based on an
assumption that the other TCP timestamp clock frequency, which is unknown
to us, may be as low as 1 Hz (RFC 7323, section 5.4).
However, this means that in the presence of short lived connections with an
RTT of couple of milliseconds, the time during which a 4-tuple is blocked
from reuse can be orders of magnitude longer that the connection lifetime.
Combined with a reduced pool of ephemeral ports, when using
IP_LOCAL_PORT_RANGE to share an egress IP address between hosts [1], the
long TIME-WAIT reuse delay can lead to port exhaustion, where all available
4-tuples are tied up in TIME-WAIT state.
Turn the reuse delay into a per-netns setting so that sysadmins can make
more aggressive assumptions about remote TCP timestamp clock frequency and
shorten the delay in order to allow connections to reincarnate faster.
Note that applications can completely bypass the TIME-WAIT delay protection
already today by locking the local port with bind() before connecting. Such
immediate connection reuse may result in PAWS failing to detect old
duplicate segments, leaving us with just the sequence number check as a
safety net.
This new configurable offers a trade off where the sysadmin can balance
between the risk of PAWS detection failing to act versus exhausting ports
by having sockets tied up in TIME-WAIT state for too long.
[1] https://lpc.events/event/16/contributions/1349/
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Link: https://patch.msgid.link/20241209-jakub-krn-909-poc-msec-tw-tstamp-v2-2-66aca0eed03e@cloudflare.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Prepare ground for TIME-WAIT socket reuse with subsecond delay.
Today the last TS.Recent update timestamp, recorded in seconds and stored
tp->ts_recent_stamp and tw->tw_ts_recent_stamp fields, has two purposes.
Firstly, it is used to track the age of the last recorded TS.Recent value
to detect when that value becomes outdated due to potential wrap-around of
the other TCP timestamp clock (RFC 7323, section 5.5).
For this purpose a second-based timestamp is completely sufficient as even
in the worst case scenario of a peer using a high resolution microsecond
timestamp, the wrap-around interval is ~36 minutes long.
Secondly, it serves as a threshold value for allowing TIME-WAIT socket
reuse. A TIME-WAIT socket can be reused only once the virtual 1 Hz clock,
ktime_get_seconds, is past the TS.Recent update timestamp.
The purpose behind delaying the TIME-WAIT socket reuse is to wait for the
other TCP timestamp clock to tick at least once before reusing the
connection. It is only then that the PAWS mechanism for the reopened
connection can detect old duplicate segments from the previous connection
incarnation (RFC 7323, appendix B.2).
In this case using a timestamp with second resolution not only blocks the
way toward allowing faster TIME-WAIT reuse after shorter subsecond delay,
but also makes it impossible to reliably delay TW reuse by one second.
As Eric Dumazet has pointed out [1], due to timestamp rounding, the TW
reuse delay will actually be between (0, 1] seconds, and 0.5 seconds on
average. We delay TW reuse for one full second only when last TS.Recent
update coincides with our virtual 1 Hz clock tick.
Considering the above, introduce a dedicated field to store a millisecond
timestamp of transition into the TIME-WAIT state. Place it in an existing
4-byte hole inside inet_timewait_sock structure to avoid an additional
memory cost.
Use the new timestamp to (i) reliably delay TIME-WAIT reuse by one second,
and (ii) prepare for configurable subsecond reuse delay in the subsequent
change.
We assume here that a full one second delay was the original intention in
[2] because it accounts for the worst case scenario of the other TCP using
the slowest recommended 1 Hz timestamp clock.
A more involved alternative would be to change the resolution of the last
TS.Recent update timestamp, tw->tw_ts_recent_stamp, to milliseconds.
[1] https://lore.kernel.org/netdev/CANn89iKB4GFd8sVzCbRttqw_96o3i2wDhX-3DraQtsceNGYwug@mail.gmail.com/
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b8439924316d5bcb266d165b93d632a4b4b859af
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Link: https://patch.msgid.link/20241209-jakub-krn-909-poc-msec-tw-tstamp-v2-1-66aca0eed03e@cloudflare.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Eric Dumazet says:
====================
ipv6: mcast: add data-race annotations
ipv6_chk_mcast_addr() and igmp6_mcf_seq_show() are reading
fields under RCU. Add missing annotations.
====================
Link: https://patch.msgid.link/20241210183352.86530-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
psf->sf_count[MCAST_XXX] fields are read locklessly from
ipv6_chk_mcast_addr() and igmp6_mcf_seq_show().
Add READ_ONCE() and WRITE_ONCE() annotations accordingly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20241210183352.86530-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
mc->mca_sfcount[MCAST_EXCLUDE] is read locklessly from
ipv6_chk_mcast_addr().
Add READ_ONCE() and WRITE_ONCE() annotations accordingly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20241210183352.86530-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add a label and two gotos to shorten lines by two tabulations,
to ease code review of following patches.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20241210183352.86530-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Jacob Keller says:
====================
lib: packing: introduce and use (un)pack_fields
This series improves the packing library with a new API for packing or
unpacking a large number of fields at once with minimal code footprint. The
API is then used to replace bespoke packing logic in the ice driver,
preparing it to handle unpacking in the future. Finally, the ice driver has
a few other cleanups related to the packing logic.
The pack_fields and unpack_fields functions have the following improvements
over the existing pack() and unpack() API:
1. Packing or unpacking a large number of fields takes significantly less
code. This significantly reduces the .text size for an increase in the
.data size which is much smaller.
2. The unpacked data can be stored in sizes smaller than u64 variables.
This reduces the storage requirement both for runtime data structures,
and for the rodata defining the fields. This scales with the number of
fields used.
3. Most of the error checking is done at compile time, rather than
runtime, via CHECK_PACKED_FIELD macros.
The actual packing and unpacking code still uses the u64 size
variables. However, these are converted to the appropriate field sizes when
storing or reading the data from the buffer.
This version now uses significantly improved macro checks, thanks to the
work of Vladimir. We now only need 300 lines of macro for the generated
checks. In addition, each new check only requires 4 lines of code for its
macro implementation and 1 extra line in the CHECK_PACKED_FIELDS macro.
This is significantly better than previous versions which required ~2700
lines.
The CHECK_PACKED_FIELDS macro uses __builtin_choose_expr to select the
appropriately sized CHECK_PACKED_FIELDS_N macro. This enables directly
adding CHECK_PACKED_FIELDS calls into the pack_fields and unpack_fields
macros. Drivers no longer need to call the CHECK_PACKED_FIELDS_N macros
directly, and we do not need to modify Kbuild or introduce multiple CONFIG
options.
The code for the CHECK_PACKED_FIELDS_(0..50) and CHECK_PACKED_FIELDS itself
can be generated from the C program in scripts/gen_packed_field_checks.c.
This little C program may be used in the future to update the checks to
more sizes if a driver with more than 50 fields appears in the future.
The total amount of required code is now much smaller, and we don't
anticipate needing to increase the size very often. Thus, it makes sense to
simply commit the result directly instead of attempting to modify Kbuild to
automatically generate it.
This version uses the 5-argument format of pack_fields and unpack_fields,
with the size of the packed buffer passed as one of the arguments. We do
enforce that the compiler can tell its a constant using
__builtin_constant_p(), ensuring that the size checks are handled at
compile time. We could reduce these to 4 arguments and require that the
passed in pbuf be of a type which has the appropriate size. I opted against
that because it makes the API less flexible and a bit less natural to use
in existing code.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
v9: https://lore.kernel.org/20241204-packing-pack-fields-and-ice-implementation-v9-0-81c8f2bd7323@intel.com
v8: https://lore.kernel.org/20241203-packing-pack-fields-and-ice-implementation-v8-0-2ed68edfe583@intel.com
v7: https://lore.kernel.org/20241202-packing-pack-fields-and-ice-implementation-v7-0-ed22e38e6c65@intel.com
v6: https://lore.kernel.org/20241118-packing-pack-fields-and-ice-implementation-v6-0-6af8b658a6c3@intel.com
v5: https://lore.kernel.org/20241111-packing-pack-fields-and-ice-implementation-v5-0-80c07349e6b7@intel.com
v4: https://lore.kernel.org/20241108-packing-pack-fields-and-ice-implementation-v4-0-81a9f42c30e5@intel.com
v3: https://lore.kernel.org/20241107-packing-pack-fields-and-ice-implementation-v3-0-27c566ac2436@intel.com
v2: https://lore.kernel.org/20241025-packing-pack-fields-and-ice-implementation-v2-0-734776c88e40@intel.com
v1: https://lore.kernel.org/20241011-packing-pack-fields-and-ice-implementation-v1-0-d9b1f7500740@intel.com
====================
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-0-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The ice_copy_rxq_ctx_to_hw() and ice_write_rxq_ctx() functions perform some
defensive checks which are typically frowned upon by kernel style
guidelines.
In particular, NULL checks on buffers which point to the stack are
discouraged, especially when the functions are static and only called once.
Checks of this sort only serve to hide potential programming error, as we
will not produce the normal crash dump on a NULL access.
In addition, ice_copy_rxq_ctx_to_hw() cannot fail in another way, so could
be made void.
Future support for VF Live Migration will need to introduce an inverse
function for reading Rx queue context from HW registers to unpack it, as
well as functions to pack and unpack Tx queue context from HW.
Rather than copying these style issues into the new functions, lets first
cleanup the existing code.
For the ice_copy_rxq_ctx_to_hw() function:
* Move the Rx queue index check out of this function.
* Convert the function to a void return.
* Use a simple int variable instead of a u8 for the for loop index, and
initialize it inside the for loop.
* Update the function description to better align with kernel doc style.
For the ice_write_rxq_ctx() function:
* Move the Rx queue index check into this function.
* Update the function description with a Returns: to align with kernel doc
style.
These changes align the existing write functions to current kernel
style, and will align with the style of the new functions added when we
implement live migration in a future series.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-10-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The ice_write_rxq_ctx() function is responsible for programming the Rx
Queue context into hardware. It receives the configuration in unpacked form
via the ice_rlan_ctx structure.
This function unconditionally modifies the context to set the prefetch
enable bit. This was done by commit c31a5c25bb19 ("ice: Always set prefena
when configuring an Rx queue"). Setting this bit makes sense, since
prefetching descriptors is almost always the preferred behavior.
However, the ice_write_rxq_ctx() function is not the place that actually
defines the queue context. We initialize the Rx Queue context in
ice_setup_rx_ctx(). It is surprising to have the Rx queue context changed
by a function who's responsibility is to program the given context to
hardware.
Following the principle of least surprise, move the setting of the prefetch
enable bit out of ice_write_rxq_ctx() and into the ice_setup_rx_ctx().
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-9-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The ice_rlan_ctx and ice_tlan_ctx structures have some fields which are
intentionally sized larger than necessary relative to the packed sizes the
data must fit into. This was done because the original ice_set_ctx()
function and its helpers did not correctly handle packing when the packed
bits straddled a byte. This is no longer the case with the use of the
<linux/packing.h> implementation.
Save some bytes in these structures by sizing the variables to the number
of bytes the actual bitpacked fields fit into.
There are a couple of gaps left in the structure, which is a result of the
fields being in the order they appear in the packed bit layout, but where
alignment forces some extra gaps. We could fix this, saving ~8 bytes from
each structure. However, these structures are not used heavily, and the
resulting savings is minimal:
$ bloat-o-meter ice-before-reorder.ko ice-after-reorder.ko
add/remove: 0/0 grow/shrink: 1/1 up/down: 26/-70 (-44)
Function old new delta
ice_vsi_cfg_txq 1873 1899 +26
ice_setup_rx_ctx.constprop 1529 1459 -70
Total: Before=1459555, After=1459511, chg -0.00%
Thus, the fields are left in the same order as the packed bit layout,
despite the gaps this causes.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-8-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The ice driver needs to write the Tx and Rx queue context when programming
Tx and Rx queues. This is currently done using some bespoke custom logic
via the ice_set_ctx() and its helper functions, along with bit position
definitions in the ice_tlan_ctx_info and ice_rlan_ctx_info structures.
This logic does work, but is problematic for several reasons:
1) ice_set_ctx requires a helper function for each byte size being packed,
as it uses a separate function to pack u8, u16, u32, and u64 fields.
This requires 4 functions which contain near-duplicate logic with the
types changed out.
2) The logic in the ice_pack_ctx_word, ice_pack_ctx_dword, and
ice_pack_ctx_qword does not handle values which straddle alignment
boundaries very well. This requires that several fields in the
ice_tlan_ctx_info and ice_rlan_ctx_info be a size larger than their bit
size should require.
3) Future support for live migration will require adding unpacking
functions to take the packed hardware context and unpack it into the
ice_rlan_ctx and ice_tlan_ctx structures. Implementing this would
require implementing ice_get_ctx, and its associated helper functions,
which essentially doubles the amount of code required.
The Linux kernel has had a packing library that can handle this logic since
commit 554aae35007e ("lib: Add support for generic packing operations").
The library was recently extended with support for packing or unpacking an
array of fields, with a similar structure as the ice_ctx_ele structure.
Replace the ice-specific ice_set_ctx() logic with the recently added
pack_fields and packed_field_s infrastructure from <linux/packing.h>
For API simplicity, the Tx and Rx queue context are programmed using
separate ice_pack_txq_ctx() and ice_pack_rxq_ctx(). This avoids needing to
export the packed_field_s arrays. The functions can pointers to the
appropriate ice_txq_ctx_buf_t and ice_rxq_ctx_buf_t types, ensuring that
only buffers of the appropriate size are passed.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-7-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The ice Tx and Rx queue context are currently stored as arrays of bytes
with defined size (ICE_RXQ_CTX_SZ and ICE_TXQ_CTX_SZ). The packed queue
context is often passed to other functions as a simple u8 * pointer, which
does not allow tracking the size. This makes the queue context API easy to
misuse, as you can pass an arbitrary u8 array or pointer.
Introduce wrapper typedefs which use a __packed structure that has the
proper fixed size for the Tx and Rx context buffers. This enables the
compiler to track the size of the value and ensures that passing the wrong
buffer size will be detected by the compiler.
The existing APIs do not benefit much from this change, however the
wrapping structures will be used to simplify the arguments of new packing
functions based on the recently introduced pack_fields API.
Co-developed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-6-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The int_q_state field of the ice_tlan_ctx structure represents the internal
queue state. However, we never actually need to assign this or read this
during normal operation. In fact, trying to unpack it would not be possible
as it is larger than a u64. Remove this field from the ice_tlan_ctx
structure, and remove its packing field from the ice_tlan_ctx_info array.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-5-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Extend the documentation for the packing library, covering the intended use
for the recently added APIs. This includes the pack() and unpack() macros,
as well as the pack_fields() and unpack_fields() macros.
Add a note that the packing() API is now deprecated in favor of pack() and
unpack().
For the pack_fields() and unpack_fields() APIs, explain the rationale for
when a driver may want to select this API. Provide an example which shows
how to define the fields and call the pack_fields() and unpack_fields()
macros.
Co-developed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-4-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This is new API which caters to the following requirements:
- Pack or unpack a large number of fields to/from a buffer with a small
code footprint. The current alternative is to open-code a large number
of calls to pack() and unpack(), or to use packing() to reduce that
number to half. But packing() is not const-correct.
- Use unpacked numbers stored in variables smaller than u64. This
reduces the rodata footprint of the stored field arrays.
- Perform error checking at compile time, rather than runtime, and return
void from the API functions. Because the C preprocessor can't generate
variable length code (loops), this is a bit tricky to do with macros.
To handle this, implement macros which sanity check the packed field
definitions based on their size. Finally, a single macro with a chain of
__builtin_choose_expr() is used to select the appropriate macros. We
enforce the use of ascending or descending order to avoid O(N^2) scaling
when checking for overlap. Note that the macros are written with care to
ensure that the compilers can correctly evaluate the resulting code at
compile time. In particular, care was taken with avoiding too many nested
statement expressions. Nested statement expressions trip up some
compilers, especially when passing down variables created in previous
statement expressions.
There are two key design choices intended to keep the overall macro code
size small. First, the definition of each CHECK_PACKED_FIELDS_N macro is
implemented recursively, by calling the N-1 macro. This avoids needing
the code to repeat multiple times.
Second, the CHECK_PACKED_FIELD macro enforces that the fields in the
array are sorted in order. This allows checking for overlap only with
neighboring fields, rather than the general overlap case where each field
would need to be checked against other fields.
The overlap checks use the first two fields to determine the order of the
remaining fields, thus allowing either ascending or descending order.
This enables drivers the flexibility to keep the fields ordered in which
ever order most naturally fits their hardware design and its associated
documentation.
The CHECK_PACKED_FIELDS macro is directly called from within pack_fields
and unpack_fields, ensuring that all drivers using the API receive the
benefits of the compile-time checks. Users do not need to directly call
any of the macros directly.
The CHECK_PACKED_FIELDS and its helper macros CHECK_PACKED_FIELDS_(0..50)
are generated using a simple C program in scripts/gen_packed_field_checks.c
This program can be compiled on demand and executed to generate the
macro code in include/linux/packing.h. This will aid in the event that a
driver needs more than 50 fields. The generator can be updated with a new
size, and used to update the packing.h header file. In practice, the ice
driver will need to support 27 fields, and the sja1105 driver will need
to support 0 fields. This on-demand generation avoids the need to modify
Kbuild. We do not anticipate the maximum number of fields to grow very
often.
- Reduced rodata footprint for the storage of the packed field arrays.
To that end, we have struct packed_field_u8 and packed_field_u16, which
define the fields with the associated type. More can be added as
needed (unlikely for now). On these types, the same generic pack_fields()
and unpack_fields() API can be used, thanks to the new C11 _Generic()
selection feature, which can call pack_fields_u8() or pack_fields_16(),
depending on the type of the "fields" array - a simplistic form of
polymorphism. It is evaluated at compile time which function will actually
be called.
Over time, packing() is expected to be completely replaced either with
pack() or with pack_fields().
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Co-developed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-3-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Most of the sanity checks in pack() and unpack() can be covered at
compile time. There is only one exception, and that is truncation of the
uval during a pack() operation.
We'd like the error-less __pack() to catch that condition as well. But
at the same time, it is currently the responsibility of consumer drivers
(currently just sja1105) to print anything at all when this error
occurs, and then discard the return code.
We can just print a loud warning in the library code and continue with
the truncated __pack() operation. In practice, having the warning is
very important, see commit 24deec6b9e4a ("net: dsa: sja1105: disallow
C45 transactions on the BASE-TX MDIO bus") where the bug was caught
exactly by noticing this print.
Add the first print to the packing library, and at the same time remove
the print for the same condition from the sja1105 driver, to avoid
double printing.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-2-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
A future variant of the API, which works on arrays of packed_field
structures, will make most of these checks redundant. The idea will be
that we want to perform sanity checks at compile time, not once
for every function call.
Introduce new variants of pack() and unpack(), which elide the sanity
checks, assuming that the input was pre-sanitized.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241210-packing-pack-fields-and-ice-implementation-v10-1-ee56a47479ac@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
get_Bprotocol4id() was added in 2008 in
commit 1b2b03f8e514 ("Add mISDN core files")
but hasn't been used.
Remove it.
Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Link: https://patch.msgid.link/20241211005802.258279-1-linux@treblig.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Define static branch variable "cn10k_ipsec_sa_enabled"
in "otx2_txrx.c". This fixes below compilation error
when CONFIG_XFRM_OFFLOAD is disabled.
drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.o:(__jump_table+0x8): undefined reference to `cn10k_ipsec_sa_enabled'
drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.o:(__jump_table+0x18): undefined reference to `cn10k_ipsec_sa_enabled'
drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.o:(__jump_table+0x28): undefined reference to `cn10k_ipsec_sa_enabled'
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202412110505.ZKDzGRMv-lkp@intel.com/
Fixes: 6a77a158848a ("cn10k-ipsec: Process outbound ipsec crypto offload")
Signed-off-by: Bharat Bhushan <bbhushan2@marvell.com>
Link: https://patch.msgid.link/20241211062419.2587111-1-bbhushan2@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The last use of gve_adminq_set_mtu() was removed by
commit 37149e9374bf ("gve: Implement packet continuation for RX.")
Remove it.
Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Praveen Kaligineedi <pkaligineedi@google.com>
Link: https://patch.msgid.link/20241211001927.253161-1-linux@treblig.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Commit b35108a51cf7 ("jiffies: Define secs_to_jiffies()") introduced
secs_to_jiffies(). As the value here is a multiple of 1000, use
secs_to_jiffies() instead of msecs_to_jiffies to avoid the multiplication.
This is converted using scripts/coccinelle/misc/secs_to_jiffies.cocci with
the following Coccinelle rules:
@@ constant C; @@
- msecs_to_jiffies(C * 1000)
+ secs_to_jiffies(C)
@@ constant C; @@
- msecs_to_jiffies(C * MSEC_PER_SEC)
+ secs_to_jiffies(C)
Signed-off-by: Easwar Hariharan <eahariha@linux.microsoft.com>
Reviewed-by: Louis Peens <louis.peens@corigine.com>
Link: https://patch.msgid.link/20241210-converge-secs-to-jiffies-v3-20-59479891e658@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
This is part of an effort to assign a section in MAINTAINERS to header
files related to Networking. In this case the files named ethool.h.
Signed-off-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20241210-mnt-ethtool-h-v1-1-2a40b567939d@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
nf_tables_chain_destroy can sleep, it can't be used from call_rcu
callbacks.
Moreover, nf_tables_rule_release() is only safe for error unwinding,
while transaction mutex is held and the to-be-desroyed rule was not
exposed to either dataplane or dumps, as it deactives+frees without
the required synchronize_rcu() in-between.
nft_rule_expr_deactivate() callbacks will change ->use counters
of other chains/sets, see e.g. nft_lookup .deactivate callback, these
must be serialized via transaction mutex.
Also add a few lockdep asserts to make this more explicit.
Calling synchronize_rcu() isn't ideal, but fixing this without is hard
and way more intrusive. As-is, we can get:
WARNING: .. net/netfilter/nf_tables_api.c:5515 nft_set_destroy+0x..
Workqueue: events nf_tables_trans_destroy_work
RIP: 0010:nft_set_destroy+0x3fe/0x5c0
Call Trace:
<TASK>
nf_tables_trans_destroy_work+0x6b7/0xad0
process_one_work+0x64a/0xce0
worker_thread+0x613/0x10d0
In case the synchronize_rcu becomes an issue, we can explore alternatives.
One way would be to allocate nft_trans_rule objects + one nft_trans_chain
object, deactivate the rules + the chain and then defer the freeing to the
nft destroy workqueue. We'd still need to keep the synchronize_rcu path as
a fallback to handle -ENOMEM corner cases though.
Reported-by: syzbot+b26935466701e56cfdc2@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/67478d92.050a0220.253251.0062.GAE@google.com/T/
Fixes: c03d278fdf35 ("netfilter: nf_tables: wait for rcu grace period on net_device removal")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
Deletion of the last rule referencing a given idletimer may happen at
the same time as a read of its file in sysfs:
| ======================================================
| WARNING: possible circular locking dependency detected
| 6.12.0-rc7-01692-g5e9a28f41134-dirty #594 Not tainted
| ------------------------------------------------------
| iptables/3303 is trying to acquire lock:
| ffff8881057e04b8 (kn->active#48){++++}-{0:0}, at: __kernfs_remove+0x20
|
| but task is already holding lock:
| ffffffffa0249068 (list_mutex){+.+.}-{3:3}, at: idletimer_tg_destroy_v]
|
| which lock already depends on the new lock.
A simple reproducer is:
| #!/bin/bash
|
| while true; do
| iptables -A INPUT -i foo -j IDLETIMER --timeout 10 --label "testme"
| iptables -D INPUT -i foo -j IDLETIMER --timeout 10 --label "testme"
| done &
| while true; do
| cat /sys/class/xt_idletimer/timers/testme >/dev/null
| done
Avoid this by freeing list_mutex right after deleting the element from
the list, then continuing with the teardown.
Fixes: 0902b469bd25 ("netfilter: xtables: idletimer target implementation")
Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
On some systems, neighbor discoveries from ns1 for fec0:42::1 (i.e., the
martian trap address) would happen at the wrong time and cause
false-negative test result.
Problem analysis also discovered that IPv6 martian ping test was broken
in that sent neighbor discoveries, not echo requests were inadvertently
trapped
Avoid the race condition by introducing the neighbors to each other
upfront. Also pin down the firewall rules to matching on echo requests
only.
Fixes: efb056e5f1f0 ("netfilter: ip6t_rpfilter: Fix regression with VRF interfaces")
Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This reverts commit 5c26d2f1d3f5e4be3e196526bead29ecb139cf91.
It turns out that we can't do this, because while the old behavior of
ignoring ignorable code points was most definitely wrong, we have
case-folding filesystems with on-disk hash values with that wrong
behavior.
So now you can't look up those names, because they hash to something
different.
Of course, it's also entirely possible that in the meantime people have
created *new* files with the new ("more correct") case folding logic,
and reverting will just make other things break.
The correct solution is to not do case folding in filesystems, but
sadly, people seem to never really understand that. People still see it
as a feature, not a bug.
Reported-by: Qi Han <hanqi@vivo.com>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=219586
Cc: Gabriel Krisman Bertazi <krisman@suse.de>
Requested-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
