Increasing the indirection RQ table size from 128 to 256 improves the
packet distribution over the NIC HW queues for various cases.
Let's take a look at the following scenario:
Assuming RSS result distributed uniformly and indirection table is filled
with queues in a cyclic manner.
Let N be the number of queues on a given setup.
If 256%N = 128%N = 0, then all queues have the same probability to be
chosen for a given RSS result.
This case doesn't improves nor degrade by this change.
If 256%N != 0 and 128%N != 0, there is a remainder which will favor some
queues. Increasing the indirection RQ table size to 256 reduce the ratio
between the favored queues probability to be selected to the rest of the
queues and improves the distribution.
For example, let's assume the number of queues is 56.
For a table size of 128, we have 128%56=16 queues which will have a 3/128
probability to be chosen and 2/128 for the rest 40.
16 queues have 1.5 times the probability to be chosen over the other 40.
For a table size of 256, we have 256%56=32 queues which will have a 5/256
probability to be chosen and 4/256 probability for the rest 24 queues.
Here 32 queues have 1.25 more probability to be chosen over the other 24.
This shows that the larger indirection table size would more likely cause
an even distribution.
This change also aligns our mlx5 driver's indirection table size with
other vendors.
Signed-off-by: Noam Stolero <noams@nvidia.com>
Reviewed-by: Tal Gilboa <talgi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The channel's napi is first needed upon activation, not creation.
Minimize its enabled scope by moving it from the channel's open/close
stage into the activate/deactivate stage.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Also cleanup neigh in profile disable.
This is for logical separation.
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
To avoid false lock dependency warning set the tc_ht lock
class different than the lock class of the ht being used when deleting
last flow from a group and then deleting a group, we get into del_sw_flow_group()
which call rhashtable_destroy on fg->ftes_hash which will take ht->mutex but
it's different than the ht->mutex here.
======================================================
WARNING: possible circular locking dependency detected
5.11.0-rc4_net_next_mlx5_949fdcc #1 Not tainted
------------------------------------------------------
modprobe/12950 is trying to acquire lock:
ffff88816510f910 (&node->lock){++++}-{3:3}, at: mlx5_del_flow_rules+0x2a/0x210 [mlx5_core]
but task is already holding lock:
ffff88815834e3e8 (&ht->mutex){+.+.}-{3:3}, at: rhashtable_free_and_destroy+0x37/0x340
which lock already depends on the new lock.
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Since its profile dependent let's init the vxlan info
as part of profile initialization.
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
We actually initialize priv and not netdev. The only call to
set netdev carrier will be moved in the following commit.
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Port nic netdevice will be used as uplink representor in downstream
patches. Add change profile method to allow changing a mlx5e netdevice
profile dynamically.
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
1) Initialize netdevice features and structures on netdevice allocation
and outside of the mlx5e profile.
2) As now mlx5e netdevice private params will be setup on profile init only
after netdevice features are already set, we add a call to
netde_update_features() to resolve any conflict.
This is nice since we reuse the fix_features ndo code if a profile
wants different default features, instead of duplicating features
conflict resolution code on profile initialization.
3) With this we achieve total separation between mlx5e profiles and
netdevices, and will allow replacing mlx5e profiles on the fly to reuse
the same netdevice for multiple profiles.
e.g. for uplink representor profile as shown in the following patch
4) Profile callbacks are not allowed to touch netdev->features directly
anymore, since in downstream patch we will detach/attach netdev
dynamically to profile, hence we move the code dealing with
netdev->features from profile->init() to fix_features ndo, and we
will call netdev_update_features() on
mlx5e_attach_netdev(profile, netdev);
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Lijun Pan says:
====================
rework the memory barrier for SCRQ entry
This series rework the memory barrier for SCRQ (Sub-Command-Response
Queue) entry.
====================
Link: https://lore.kernel.org/r/20210130011905.1485-1-ljp@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
rmb() can be removed since:
1. pending_scrq() has dma_rmb() at the function end;
2. dma_rmb(), though weaker, is enough here.
Signed-off-by: Lijun Pan <ljp@linux.ibm.com>
Acked-by: Dwip Banerjee <dnbanerg@us.ibm.com>
Acked-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move the dma_rmb() between pending_scrq() and ibmvnic_next_scrq()
into the end of pending_scrq() to save the duplicated code since
this dma_rmb will be used 3 times.
Signed-off-by: Lijun Pan <ljp@linux.ibm.com>
Acked-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for Connect-X6DX Software steering
This series adds SW Steering support for Connect-X6DX.
Since the STE and actions formats are different on this new HW,
we implemented the HW specific STEv1 layer on the infrastructure
implemented in previous mlx5 DR patchset to support all the
functionalities as previous devices.
Most of the code in this series very is low level HW specific, we
implement the function pointers for the generic SW steering layer.
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEGhZs6bAKwk/OTgTpSD+KveBX+j4FAmAUwK8ACgkQSD+KveBX
+j61CAgAnNwFuk8PQjdU6TR+nlPl0oQ0mVUIyWvP8giuyXPxTFLxo8wJVKnh9tfB
JFHgaEzOXhdE6n3+/vknlN/NsFUpt6Kbg2cBXc65btEKKCdcm/D3Db45TUwu0o3d
HE5cEWnJm/Qqvy7JvoVpzbNDcNh91AIdpWt95AxRYBFgWbcKvyz/Bq+DSb22grYz
bSU2HMKZKpXtHbxOV0BsZ9b2si6hpIMKRXIofT3F5yVmx6t8M174NmD4u2h6VVaa
v7dvZp7ItbnD61iJnKRLa3zftBptifDB/2wsei3W4wmfdAA1Uw9B2tPNJKboxMSa
8hWiBWE6U72rG6uz4fWd9V0mgkELXA==
=CTbF
-----END PGP SIGNATURE-----
Merge tag 'mlx5-dr-2021-01-29' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
mlx5-dr-2021-01-29
Add support for Connect-X6DX Software steering
This series adds SW Steering support for Connect-X6DX.
Since the STE and actions formats are different on this new HW,
we implemented the HW specific STEv1 layer on the infrastructure
implemented in previous mlx5 DR patchset to support all the
functionalities as previous devices.
Most of the code in this series very is low level HW specific, we
implement the function pointers for the generic SW steering layer.
* tag 'mlx5-dr-2021-01-29' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux:
net/mlx5: DR, Allow SW steering for sw_owner_v2 devices
net/mlx5: DR, Copy all 64B whenever replacing STE in the head of miss-list
net/mlx5: DR, Use HW specific logic API when writing STE
net/mlx5: DR, Use the right size when writing partial STE into HW
net/mlx5: DR, Add STEv1 modify header logic
net/mlx5: DR, Add STEv1 action apply logic
net/mlx5: DR, Add STEv1 setters and getters
net/mlx5: DR, Allow native protocol support for HW STEv1
net/mlx5: DR, Add HW STEv1 match logic
net/mlx5: DR, Add match STEv1 structs to ifc
net/mlx5: DR, Fix potential shift wrapping of 32-bit value
====================
Link: https://lore.kernel.org/r/20210130022618.317351-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kurt Kanzenbach says:
====================
net: dsa: hellcreek: Report tables sizes
Florian, Andrew and Vladimir suggested at some point to use devlink for
reporting tables, features and debugging counters instead of using debugfs and
printk.
So, start by reporting the VLAN and FDB table sizes.
====================
Link: https://lore.kernel.org/r/20210130135934.22870-1-kurt@kmk-computers.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Report the FDB table size and occupancy via devlink. The actual size depends on
the used Hellcreek version:
|root@tsn:~# devlink resource show platform/ff240000.switch
|platform/ff240000.switch:
| name VLAN size 4096 occ 2 unit entry dpipe_tables none
| name FDB size 256 occ 6 unit entry dpipe_tables none
Suggested-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Kurt Kanzenbach <kurt@kmk-computers.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The VLAN membership configuration is cached in software already. So, it can be
reported via devlink. Add support for it:
|root@tsn:~# devlink resource show platform/ff240000.switch
|platform/ff240000.switch:
| name VLAN size 4096 occ 4 unit entry dpipe_tables none
Signed-off-by: Kurt Kanzenbach <kurt@kmk-computers.de>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This commit shrinks inet_connection_sock by 4 bytes, by shrinking
icsk_mtup.enabled from 32 bits to 1 bit, and shrinking
icsk_mtup.probe_size from s32 to an unsuigned 31 bit field.
This is to save space to compensate for the recent introduction of a
new u32 in inet_connection_sock, icsk_probes_tstamp, in the recent bug
fix commit 9d9b1ee0b2 ("tcp: fix TCP_USER_TIMEOUT with zero window").
This should not change functionality, since icsk_mtup.enabled is only
ever set to 0 or 1, and icsk_mtup.probe_size can only be either 0
or a positive MTU value returned by tcp_mss_to_mtu()
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20210129185438.1813237-1-ncardwell.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We're moving to netlink-only options, so add comments in the bridge's
sysfs files to warn against adding any new sysfs entries.
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We decided to stop adding new sysfs bridge options and continue with
netlink only, so remove hosts limit sysfs support.
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Vladimir Oltean says:
====================
tag_8021q for Ocelot switches
The Felix switch inside LS1028A has an issue. It has a 2.5G CPU port,
and the external ports, in the majority of use cases, run at 1G. This
means that, when the CPU injects traffic into the switch, it is very
easy to run into congestion. This is not to say that it is impossible to
enter congestion even with all ports running at the same speed, just
that the default configuration is already very prone to that by design.
Normally, the way to deal with that is using Ethernet flow control
(PAUSE frames).
However, this functionality is not working today with the ENETC - Felix
switch pair. The hardware issue is undergoing documentation right now as
an erratum within NXP, but several customers have been requesting a
reasonable workaround for it.
In truth, the LS1028A has 2 internal port pairs. The lack of flow control
is an issue only when NPI mode (Node Processor Interface, aka the mode
where the "CPU port module", which carries DSA-style tagged packets, is
connected to a regular Ethernet port) is used, and NPI mode is supported
by Felix on a single port.
In past BSPs, we have had setups where both internal port pairs were
enabled. We were advertising the following setup:
"data port" "control port"
(2.5G) (1G)
eno2 eno3
^ ^
| |
| regular | DSA-tagged
| frames | frames
| |
v v
swp4 swp5
This works but is highly unpractical, due to NXP shifting the task of
designing a functional system (choosing which port to use, depending on
type of traffic required) up to the end user. The swpN interfaces would
have to be bridged with swp4, in order for the eno2 "data port" to have
access to the outside network. And the swpN interfaces would still be
capable of IP networking. So running a DHCP client would give us two IP
interfaces from the same subnet, one assigned to eno2, and the other to
swpN (0, 1, 2, 3).
Also, the dual port design doesn't scale. When attaching another DSA
switch to a Felix port, the end result is that the "data port" cannot
carry any meaningful data to the external world, since it lacks the DSA
tags required to traverse the sja1105 switches below. All that traffic
needs to go through the "control port".
So in newer BSPs there was a desire to simplify that setup, and only
have one internal port pair:
eno2 eno3
^
|
| DSA-tagged x disabled
| frames
|
v
swp4 swp5
However, this setup only exacerbates the issue of not having flow
control on the NPI port, since that is the only port now. Also, there
are use cases that still require the "data port", such as IEEE 802.1CB
(TSN stream identification doesn't work over an NPI port), source
MAC address learning over NPI, etc.
Again, there is a desire to keep the simplicity of the single internal
port setup, while regaining the benefits of having a dedicated data port
as well. And this series attempts to deliver just that.
So the NPI functionality is disabled conditionally. Its purpose was:
- To ensure individually addressable ports on TX. This can be replaced
by using some designated VLAN tags which are pushed by the DSA tagger
code, then removed by the switch (so they are invisible to the outside
world and to the user).
- To ensure source port identification on RX. Again, this can be
replaced by using some designated VLAN tags to encapsulate all RX
traffic (each VLAN uniquely identifies a source port). The DSA tagger
determines which port it was based on the VLAN number, then removes
that header.
- To deliver PTP timestamps. This cannot be obtained through VLAN
headers, so we need to take a step back and see how else we can do
that. The Microchip Ocelot-1 (VSC7514 MIPS) driver performs manual
injection/extraction from the CPU port module using register-based
MMIO, and not over Ethernet. We will need to do the same from DSA,
which makes this tagger a sort of hybrid between DSA and pure
switchdev.
====================
Link: https://lore.kernel.org/r/20210129010009.3959398-1-olteanv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Unlike sja1105, the only other user of the software-defined tag_8021q.c
tagger format, the implementation we choose for the Felix DSA switch
driver preserves full functionality under a vlan_filtering bridge
(i.e. IP termination works through the DSA user ports under all
circumstances).
The tag_8021q protocol just wants:
- Identifying the ingress switch port based on the RX VLAN ID, as seen
by the CPU. We achieve this by using the TCAM engines (which are also
used for tc-flower offload) to push the RX VLAN as a second, outer
tag, on egress towards the CPU port.
- Steering traffic injected into the switch from the network stack
towards the correct front port based on the TX VLAN, and consuming
(popping) that header on the switch's egress.
A tc-flower pseudocode of the static configuration done by the driver
would look like this:
$ tc qdisc add dev <cpu-port> clsact
$ for eth in swp0 swp1 swp2 swp3; do \
tc filter add dev <cpu-port> egress flower indev ${eth} \
action vlan push id <rxvlan> protocol 802.1ad; \
tc filter add dev <cpu-port> ingress protocol 802.1Q flower
vlan_id <txvlan> action vlan pop \
action mirred egress redirect dev ${eth}; \
done
but of course since DSA does not register network interfaces for the CPU
port, this configuration would be impossible for the user to do. Also,
due to the same reason, it is impossible for the user to inadvertently
delete these rules using tc. These rules do not collide in any way with
tc-flower, they just consume some TCAM space, which is something we can
live with.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
There are use cases for which the existing tagger, based on the NPI
(Node Processor Interface) functionality, is insufficient.
Namely:
- Frames injected through the NPI port bypass the frame analyzer, so no
source address learning is performed, no TSN stream classification,
etc.
- Flow control is not functional over an NPI port (PAUSE frames are
encapsulated in the same Extraction Frame Header as all other frames)
- There can be at most one NPI port configured for an Ocelot switch. But
in NXP LS1028A and T1040 there are two Ethernet CPU ports. The non-NPI
port is currently either disabled, or operated as a plain user port
(albeit an internally-facing one). Having the ability to configure the
two CPU ports symmetrically could pave the way for e.g. creating a LAG
between them, to increase bandwidth seamlessly for the system.
So there is a desire to have an alternative to the NPI mode. This change
keeps the default tagger for the Seville and Felix switches as "ocelot",
but it can be changed via the following device attribute:
echo ocelot-8021q > /sys/class/<dsa-master>/dsa/tagging
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In expectation of the new tag_ocelot_8021q tagger implementation, we
need to be able to do runtime switchover between one tagger and another.
So we must structure the existing code for the current NPI-based tagger
in a certain way.
We move the felix_npi_port_init function in expectation of the future
driver configuration necessary for tag_ocelot_8021q: we would like to
not have the NPI-related bits interspersed with the tag_8021q bits.
The conversion from this:
ocelot_write_rix(ocelot,
ANA_PGID_PGID_PGID(GENMASK(ocelot->num_phys_ports, 0)),
ANA_PGID_PGID, PGID_UC);
to this:
cpu_flood = ANA_PGID_PGID_PGID(BIT(ocelot->num_phys_ports));
ocelot_rmw_rix(ocelot, cpu_flood, cpu_flood, ANA_PGID_PGID, PGID_UC);
is perhaps non-trivial, but is nonetheless non-functional. The PGID_UC
(replicator for unknown unicast) is already configured out of hardware
reset to flood to all ports except ocelot->num_phys_ports (the CPU port
module). All we change is that we use a read-modify-write to only add
the CPU port module to the unknown unicast replicator, as opposed to
doing a full write to the register.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently DSA exposes the following sysfs:
$ cat /sys/class/net/eno2/dsa/tagging
ocelot
which is a read-only device attribute, introduced in the kernel as
commit 98cdb48071 ("net: dsa: Expose tagging protocol to user-space"),
and used by libpcap since its commit 993db3800d7d ("Add support for DSA
link-layer types").
It would be nice if we could extend this device attribute by making it
writable:
$ echo ocelot-8021q > /sys/class/net/eno2/dsa/tagging
This is useful with DSA switches that can make use of more than one
tagging protocol. It may be useful in dsa_loop in the future too, to
perform offline testing of various taggers, or for changing between dsa
and edsa on Marvell switches, if that is desirable.
In terms of implementation, drivers can support this feature by
implementing .change_tag_protocol, which should always leave the switch
in a consistent state: either with the new protocol if things went well,
or with the old one if something failed. Teardown of the old protocol,
if necessary, must be handled by the driver.
Some things remain as before:
- The .get_tag_protocol is currently only called at probe time, to load
the initial tagging protocol driver. Nonetheless, new drivers should
report the tagging protocol in current use now.
- The driver should manage by itself the initial setup of tagging
protocol, no later than the .setup() method, as well as destroying
resources used by the last tagger in use, no earlier than the
.teardown() method.
For multi-switch DSA trees, error handling is a bit more complicated,
since e.g. the 5th out of 7 switches may fail to change the tag
protocol. When that happens, a revert to the original tag protocol is
attempted, but that may fail too, leaving the tree in an inconsistent
state despite each individual switch implementing .change_tag_protocol
transactionally. Since the intersection between drivers that implement
.change_tag_protocol and drivers that support D in DSA is currently the
empty set, the possibility for this error to happen is ignored for now.
Testing:
$ insmod mscc_felix.ko
[ 79.549784] mscc_felix 0000:00:00.5: Adding to iommu group 14
[ 79.565712] mscc_felix 0000:00:00.5: Failed to register DSA switch: -517
$ insmod tag_ocelot.ko
$ rmmod mscc_felix.ko
$ insmod mscc_felix.ko
[ 97.261724] libphy: VSC9959 internal MDIO bus: probed
[ 97.267363] mscc_felix 0000:00:00.5: Found PCS at internal MDIO address 0
[ 97.274998] mscc_felix 0000:00:00.5: Found PCS at internal MDIO address 1
[ 97.282561] mscc_felix 0000:00:00.5: Found PCS at internal MDIO address 2
[ 97.289700] mscc_felix 0000:00:00.5: Found PCS at internal MDIO address 3
[ 97.599163] mscc_felix 0000:00:00.5 swp0 (uninitialized): PHY [0000:00:00.3:10] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[ 97.862034] mscc_felix 0000:00:00.5 swp1 (uninitialized): PHY [0000:00:00.3:11] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[ 97.950731] mscc_felix 0000:00:00.5 swp0: configuring for inband/qsgmii link mode
[ 97.964278] 8021q: adding VLAN 0 to HW filter on device swp0
[ 98.146161] mscc_felix 0000:00:00.5 swp2 (uninitialized): PHY [0000:00:00.3:12] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[ 98.238649] mscc_felix 0000:00:00.5 swp1: configuring for inband/qsgmii link mode
[ 98.251845] 8021q: adding VLAN 0 to HW filter on device swp1
[ 98.433916] mscc_felix 0000:00:00.5 swp3 (uninitialized): PHY [0000:00:00.3:13] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[ 98.485542] mscc_felix 0000:00:00.5: configuring for fixed/internal link mode
[ 98.503584] mscc_felix 0000:00:00.5: Link is Up - 2.5Gbps/Full - flow control rx/tx
[ 98.527948] device eno2 entered promiscuous mode
[ 98.544755] DSA: tree 0 setup
$ ping 10.0.0.1
PING 10.0.0.1 (10.0.0.1): 56 data bytes
64 bytes from 10.0.0.1: seq=0 ttl=64 time=2.337 ms
64 bytes from 10.0.0.1: seq=1 ttl=64 time=0.754 ms
^C
- 10.0.0.1 ping statistics -
2 packets transmitted, 2 packets received, 0% packet loss
round-trip min/avg/max = 0.754/1.545/2.337 ms
$ cat /sys/class/net/eno2/dsa/tagging
ocelot
$ cat ./test_ocelot_8021q.sh
#!/bin/bash
ip link set swp0 down
ip link set swp1 down
ip link set swp2 down
ip link set swp3 down
ip link set swp5 down
ip link set eno2 down
echo ocelot-8021q > /sys/class/net/eno2/dsa/tagging
ip link set eno2 up
ip link set swp0 up
ip link set swp1 up
ip link set swp2 up
ip link set swp3 up
ip link set swp5 up
$ ./test_ocelot_8021q.sh
./test_ocelot_8021q.sh: line 9: echo: write error: Protocol not available
$ rmmod tag_ocelot.ko
rmmod: can't unload module 'tag_ocelot': Resource temporarily unavailable
$ insmod tag_ocelot_8021q.ko
$ ./test_ocelot_8021q.sh
$ cat /sys/class/net/eno2/dsa/tagging
ocelot-8021q
$ rmmod tag_ocelot.ko
$ rmmod tag_ocelot_8021q.ko
rmmod: can't unload module 'tag_ocelot_8021q': Resource temporarily unavailable
$ ping 10.0.0.1
PING 10.0.0.1 (10.0.0.1): 56 data bytes
64 bytes from 10.0.0.1: seq=0 ttl=64 time=0.953 ms
64 bytes from 10.0.0.1: seq=1 ttl=64 time=0.787 ms
64 bytes from 10.0.0.1: seq=2 ttl=64 time=0.771 ms
$ rmmod mscc_felix.ko
[ 645.544426] mscc_felix 0000:00:00.5: Link is Down
[ 645.838608] DSA: tree 0 torn down
$ rmmod tag_ocelot_8021q.ko
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cascading DSA switches can be done multiple ways. There is the brute
force approach / tag stacking, where one upstream switch, located
between leaf switches and the host Ethernet controller, will just
happily transport the DSA header of those leaf switches as payload.
For this kind of setups, DSA works without any special kind of treatment
compared to a single switch - they just aren't aware of each other.
Then there's the approach where the upstream switch understands the tags
it transports from its leaves below, as it doesn't push a tag of its own,
but it routes based on the source port & switch id information present
in that tag (as opposed to DMAC & VID) and it strips the tag when
egressing a front-facing port. Currently only Marvell implements the
latter, and Marvell DSA trees contain only Marvell switches.
So it is safe to say that DSA trees already have a single tag protocol
shared by all switches, and in fact this is what makes the switches able
to understand each other. This fact is also implied by the fact that
currently, the tagging protocol is reported as part of a sysfs installed
on the DSA master and not per port, so it must be the same for all the
ports connected to that DSA master regardless of the switch that they
belong to.
It's time to make this official and enforce it (yes, this also means we
won't have any "switch understands tag to some extent but is not able to
speak it" hardware oddities that we'll support in the future).
This is needed due to the imminent introduction of the dsa_switch_ops::
change_tag_protocol driver API. When that is introduced, we'll have
to notify switches of the tagging protocol that they're configured to
use. Currently the tag_ops structure pointer is held only for CPU ports.
But there are switches which don't have CPU ports and nonetheless still
need to be configured. These would be Marvell leaf switches whose
upstream port is just a DSA link. How do we inform these of their
tagging protocol setup/deletion?
One answer to the above would be: iterate through the DSA switch tree's
ports once, list the CPU ports, get their tag_ops, then iterate again
now that we have it, and notify everybody of that tag_ops. But what to
do if conflicts appear between one cpu_dp->tag_ops and another? There's
no escaping the fact that conflict resolution needs to be done, so we
can be upfront about it.
Ease our work and just keep the master copy of the tag_ops inside the
struct dsa_switch_tree. Reference counting is now moved to be per-tree
too, instead of per-CPU port.
There are many places in the data path that access master->dsa_ptr->tag_ops
and we would introduce unnecessary performance penalty going through yet
another indirection, so keep those right where they are.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The existence of dsa_broadcast has generated some confusion in the past:
https://www.mail-archive.com/netdev@vger.kernel.org/msg365042.html
So let's document the existing dsa_port_notify and dsa_broadcast
functions and explain when each of them should be used.
Also, in fact, the in-between function has always been there but was
lacking a name, and is the main reason for this patch: dsa_tree_notify.
Refactor dsa_broadcast to use it.
This patch also moves dsa_broadcast (a top-level function) to dsa2.c,
where it really belonged in the first place, but had no companion so it
stood with dsa_port_notify.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Context: Ocelot switches put the injection/extraction frame header in
front of the Ethernet header. When used in NPI mode, a DSA master would
see junk instead of the destination MAC address, and it would most
likely drop the packets. So the Ocelot frame header can have an optional
prefix, which is just "ff:ff:ff:ff:ff:fe > ff:ff:ff:ff:ff:ff" padding
put before the actual tag (still before the real Ethernet header) such
that the DSA master thinks it's looking at a broadcast frame with a
strange EtherType.
Unfortunately, a lesson learned in commit 69df578c5f ("net: mscc:
ocelot: eliminate confusion between CPU and NPI port") seems to have
been forgotten in the meanwhile.
The CPU port module and the NPI port have independent settings for the
length of the tag prefix. However, the driver is using the same variable
to program both of them.
There is no reason really to use any tag prefix with the CPU port
module, since that is not connected to any Ethernet port. So this patch
makes the inj_prefix and xtr_prefix variables apply only to the NPI
port (which the switchdev ocelot_vsc7514 driver does not use).
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Applying the bridge forwarding mask currently is done only on the STP
state changes for any port. But it depends on both STP state changes,
and bonding interface state changes. Export the bit that recalculates
the forwarding mask so that it could be reused, and call it when a port
starts and stops offloading a bonding interface.
Now that the logic is split into a separate function, we can rename "p"
into "port", since the "port" variable was already taken in
ocelot_bridge_stp_state_set. Also, we can rename "i" into "lag", to make
it more clear what is it that we're iterating through.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We will be adding some private VCAP filters that should not interfere in
any way with the filters added using tc-flower. So we need to allocate
some IDs which will not be used by tc.
Currently ocelot uses an u32 id derived from the flow cookie, which in
itself is an unsigned long. This is a problem in itself, since on 64 bit
systems, sizeof(unsigned long)=8, so the driver is already truncating
these.
Create a struct ocelot_vcap_id which contains the full unsigned long
cookie from tc, as well as a boolean that is supposed to namespace the
filters added by tc with the ones that aren't.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The Felix driver will need to preinstall some VCAP filters for its
tag_8021q implementation (outside of the tc-flower offload logic), so
these need to be exported to the common includes.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The sja1105 implementation can be blind about this, but the felix driver
doesn't do exactly what it's being told, so it needs to know whether it
is a TX or an RX VLAN, so it can install the appropriate type of TCAM
rule.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
buf_info structures in RX & TX queues are private driver data that
do not need to be visible to the device. Although there is physical
address and length in the queue descriptor that points to these
structures, their layout is not standardized, and device never looks
at them.
So lets allocate these structures in non-DMA-able memory, and fill
physical address as all-ones and length as zero in the queue
descriptor.
That should alleviate worries brought by Martin Radev in
https://lists.osuosl.org/pipermail/intel-wired-lan/Week-of-Mon-20210104/022829.html
that malicious vmxnet3 device could subvert SVM/TDX guarantees.
Signed-off-by: Petr Vandrovec <petr@vmware.com>
Signed-off-by: Ronak Doshi <doshir@vmware.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use cache friendly helpers to better use cpu caches
while reading /proc/net/netstat
Tested on a platform with 256 threads (AMD Rome)
Before: 305 usec spent in netstat_seq_show()
After: 130 usec spent in netstat_seq_show()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20210128162145.1703601-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The commit 41b14fb872 ("net: Do not clear the sock TX queue in
sk_set_socket()") removes sk_tx_queue_clear() from sk_set_socket() and adds
it instead in sk_alloc() and sk_clone_lock() to fix an issue introduced in
the commit e022f0b4a0 ("net: Introduce sk_tx_queue_mapping"). On the
other hand, the original commit had already put sk_tx_queue_clear() in
sk_prot_alloc(): the callee of sk_alloc() and sk_clone_lock(). Thus
sk_tx_queue_clear() is called twice in each path.
If we remove sk_tx_queue_clear() in sk_alloc() and sk_clone_lock(), it
currently works well because (i) sk_tx_queue_mapping is defined between
sk_dontcopy_begin and sk_dontcopy_end, and (ii) sock_copy() called after
sk_prot_alloc() in sk_clone_lock() does not overwrite sk_tx_queue_mapping.
However, if we move sk_tx_queue_mapping out of the no copy area, it
introduces a bug unintentionally.
Therefore, this patch adds a compile-time check to take care of the order
of sock_copy() and sk_tx_queue_clear() and removes sk_tx_queue_clear() from
sk_prot_alloc() so that it does the only allocation and its callers
initialize fields.
CC: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20210128150217.6060-1-kuniyu@amazon.co.jp
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In order to query tm info of nodes, priority and qset
for debugging, adds three debugfs files tm_nodes,
tm_priority and tm_qset in newly created tm directory.
Unlike previous debugfs commands, these three files
just support read ops, so they only support to use cat
command to dump their info.
The new tm file style is acccording to suggestion from
Jakub Kicinski's opinion as link https://lkml.org/lkml/2020/9/29/2101.
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add some interfaces to get information of tm priority and qset,
then they can be used by debugfs.
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Xin Long says:
====================
net: add support for ip generic checksum offload for gre
This patchset it to add ip generic csum processing first in
skb_csum_hwoffload_help() in Patch 1/2 and then add csum
offload support for GRE header in Patch 2/2.
====================
Link: https://lore.kernel.org/r/cover.1611825446.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch is to add csum offload support for gre header:
On the TX path in gre_build_header(), when CHECKSUM_PARTIAL's set
for inner proto, it will calculate the csum for outer proto, and
inner csum will be offloaded later. Otherwise, CHECKSUM_PARTIAL
and csum_start/offset will be set for outer proto, and the outer
csum will be offloaded later.
On the GSO path in gre_gso_segment(), when CHECKSUM_PARTIAL is
not set for inner proto and the hardware supports csum offload,
CHECKSUM_PARTIAL and csum_start/offset will be set for outer
proto, and outer csum will be offloaded later. Otherwise, it
will do csum for outer proto by calling gso_make_checksum().
Note that SCTP has to do the csum by itself for non GSO path in
sctp_packet_pack(), as gre_build_header() can't handle the csum
with CHECKSUM_PARTIAL set for SCTP CRC csum offload.
v1->v2:
- remove the SCTP part, as GRE dev doesn't support SCTP CRC CSUM
and it will always do checksum for SCTP in sctp_packet_pack()
when it's not a GSO packet.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
NETIF_F_IP|IPV6_CSUM feature flag indicates UDP and TCP csum offload
while NETIF_F_HW_CSUM feature flag indicates ip generic csum offload
for HW, which includes not only for TCP/UDP csum, but also for other
protocols' csum like GRE's.
However, in skb_csum_hwoffload_help() it only checks features against
NETIF_F_CSUM_MASK(NETIF_F_HW|IP|IPV6_CSUM). So if it's a non TCP/UDP
packet and the features doesn't support NETIF_F_HW_CSUM, but supports
NETIF_F_IP|IPV6_CSUM only, it would still return 0 and leave the HW
to do csum.
This patch is to support ip generic csum processing by checking
NETIF_F_HW_CSUM for all protocols, and check (NETIF_F_IP_CSUM |
NETIF_F_IPV6_CSUM) only for TCP and UDP.
Note that we're using skb->csum_offset to check if it's a TCP/UDP
proctol, this might be fragile. However, as Alex said, for now we
only have a few L4 protocols that are requesting Tx csum offload,
we'd better fix this until a new protocol comes with a same csum
offset.
v1->v2:
- not extend skb->csum_not_inet, but use skb->csum_offset to tell
if it's an UDP/TCP csum packet.
v2->v3:
- add a note in the changelog, as Willem suggested.
Suggested-by: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
-----BEGIN PGP SIGNATURE-----
iQFHBAABCgAxFiEEK3kIWJt9yTYMP3ehqclaivrt76kFAmATyC0THG1rbEBwZW5n
dXRyb25peC5kZQAKCRCpyVqK+u3vqUMNB/4s7WkirDS+fMDpdB+vRjqguvJ/R4Ef
loplrX5NN8437xZUrA0x4d/jQ59jXOOT7F3MjEN2kU389FPikVmpw6owUSbxQhOD
cZ8BK1WhBNJBezOITfO4TWGowKsBN3guEBolcqBrCtMt97mxQ5u9rB0Ke0uFyGZ7
2ws1BpGK/wQBeORkyHRS7hY8dlyZbFJkUl4J9oKYuhvIwHc0x26+LZFFWvs8GFFu
NSnDqwhg4ImI27/wamyHGSWukk2IdUgKVwtNzUpSzEWmeEtp2EHFTgY56ErizCce
8EihbIB15brChkMO2J8mrHoJxzWR99kMqWuJ/EKdiycxtMx4+fj3wz91
=QfkK
-----END PGP SIGNATURE-----
Merge tag 'linux-can-next-for-5.12-20210129' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can-next
Marc Kleine-Budde says:
====================
linux-can-next-for-5.12-20210129
All patches are by me and target the mcp251xfd driver. The first 4
patches update the information regarding the "85% of (FSYSCLK/2)"
errata. The other 4 are misc cleanups, unitfy error messages, add
missing postfix to a macro, simplify the return of a function, and
make use of dev_err_probe() in the mcp251xfd_probe() function.
====================
Link: https://lore.kernel.org/r/20210129084302.3040284-1-mkl@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the new mhi_get_free_desc_count helper to track queue usage
instead of relying on the locally maintained rx_queued count.
Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The RX queue size can be determined at runtime by retrieving the
number of available transfer descriptors.
Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This section was missed during the conversion to ReST, so convert it in the
same style as the surrounding section titles.
Signed-off-by: Jan Luebbe <jlu@pengutronix.de>
Link: https://lore.kernel.org/r/20210128111930.29473-1-jlu@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This converts the driver to use the new tasklet API introduced in
commit 12cc923f1c ("tasklet: Introduce new initialization API")
Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
Link: https://lore.kernel.org/r/20210127173256.13954-2-kernel@esmil.dk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Previously a temporary tasklet structure was initialized on the stack
using DECLARE_TASKLET_OLD() and then copied over and modified. Nothing
else in the kernel seems to use this pattern, so let's just call
tasklet_init() like everyone else.
Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
Link: https://lore.kernel.org/r/20210127173256.13954-1-kernel@esmil.dk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
