The call to dpaa2_switch_port_link_state_update is a leftover from the
time when on DPAA2 platforms the PHYs were started at boot time so when
an ifconfig was issued on the associated interface, the link status
needed to be checked directly from the ndo_open() callback. This is not
needed anymore since we are now properly integrated with the PHY layer
thus a link interrupt will come directly from the PHY eventually without
the need to call the sync function.
Fix this up by removing the call to dpaa2_switch_port_link_state_update.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We should not enable the switch interfaces at probe time since this is
trigged by the open callback. Remove the call dpsw_enable() which does
exactly this.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The MC firmware supplies us the switch interface index for which an
interrupt was triggered. Use this to our advantage instead of looping
through all the switch ports and doing unnecessary work.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Request all interrupt sources to be read and then cleared on the DPSW
object. In the next patches we'll also add support for treating other
interrupts.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch intends to fix MAC internal delay doesn't work, due to use
of_property_read_u32() incorrectly, and improve this feature a bit:
1) check the delay value if valid, only program register when it's 2000ps.
2) only enable "enet_2x_txclk" clock when require MAC internal delay.
Fixes: fc539459e9 ("net: fec: add MAC internal delayed clock feature support")
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Link: https://lore.kernel.org/r/20210803052424.19008-1-qiangqing.zhang@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Don't populate the array faf_bits on the stack but instead it
static const. Makes the object code smaller by 175 bytes.
Before:
text data bss dec hex filename
9645 4552 0 14197 3775 ../freescale/dpaa2/dpaa2-eth-devlink.o
After:
text data bss dec hex filename
9406 4616 0 14022 36c6 ../freescale/dpaa2/dpaa2-eth-devlink.o
(gcc version 10.2.0)
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Link: https://lore.kernel.org/r/20210801152209.146359-1-colin.king@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When mirroring rules are added in shared filter blocks, the same
mirroring rule has to be configured on all the switch ports that are
part of the same block.
In case a switch port joins a shared block after mirroring filters have
been already added to it, then all the mirror rules should be offloaded
to the port. The reverse, removal of mirroring rules, has to be done at
block unbind.
For this purpose, the dpaa2_switch_block_offload_mirror() and
dpaa2_switch_block_unoffload_mirror() functions are added and called
upon binding and unbinding a switch port to/from a block.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Using the infrastructure added in the previous patch, extend tc-flower
support with FLOW_ACTION_MIRRED based on VLAN.
Tested with:
tc qdisc add dev eth8 ingress_block 1 clsact
tc filter add block 1 ingress protocol 802.1q flower skip_sw \
vlan_id 100 action mirred egress mirror dev eth6
tc filter del block 1 ingress pref 49152
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for per port mirroring for the DPAA2 switch. We support
only single mirror port, therefore we allow mirroring rules only as long
as the destination port is always the same.
Unlike all the actions (drop, redirect, trap) already supported by the
dpaa2-switch driver, adding mirroring filters in shared blocks is not
achieved by a singular ACL entry added in a table shared by the ports.
This is why, when a new mirror filter is added in a block we have to got
through all the switch ports sharing it and configure the filter
individually on all.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the necessary MC API for setting up and configuring the mirroring
feature on the DPSW DPAA2 object.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extract the necessary steps to offload a filter by using the ACL table
in a separate function - dpaa2_switch_cls_matchall_replace_acl().
This is intended to help with the code readability when the mirroring
support is added.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extract the necessary steps to offload a filter by using the ACL table
in a separate function - dpaa2_switch_cls_flower_replace_acl().
This is intended to help with the code readability when the mirroring
support is added.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Until now, shared filter blocks were implemented only by ACL tables
shared between ports. Going forward, when the mirroring support will be
added, this will not be true anymore.
Rename the dpaa2_switch_acl_tbl into dpaa2_switch_filter_block so that
we make it clear that the structure is used not only for filters that
use the ACL table but will be used for all the filters that are added in
a block.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Until now, the dpaa2_switch_tc_parse_action() function was used for all
the supported tc actions since all of them were implemented by adding
ACL table entries. In the next commits, the dpaa2-switch driver will
gain mirroring support which is not using the same HW feature.
Make sure that we specify the ACL in the function name so that we make
it clear that it's only used for specific actions.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
i.MX8QM ENET IP version support timing specification that MAC
integrate clock delay in RGMII mode, the delayed TXC/RXC as an
alternative option to work well with various PHYs.
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The i.MX8MQ ENET version support IEEE802.3az eee mode, add
eee mode tx lpi enable to support ethtool interface.
usage:
1. set sleep and wake timer to 5ms:
ethtool --set-eee eth0 eee on tx-lpi on tx-timer 5000
2. check the eee mode:
~# ethtool --show-eee eth0
EEE Settings for eth0:
EEE status: enabled - active
Tx LPI: 5000 (us)
Supported EEE link modes: 100baseT/Full
1000baseT/Full
Advertised EEE link modes: 100baseT/Full
1000baseT/Full
Link partner advertised EEE link modes: 100baseT/Full
Note: For realtime case and IEEE1588 ptp case, it should disable
EEE mode.
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ENET of imx8mq and imx8qm are basically the same as imx6sx,
but they have new features support based on imx6sx, like:
- imx8mq: supports IEEE 802.3az EEE standard.
- imx8qm: supports RGMII mode delayed clock.
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most users of ndo_do_ioctl are ethernet drivers that implement
the MII commands SIOCGMIIPHY/SIOCGMIIREG/SIOCSMIIREG, or hardware
timestamping with SIOCSHWTSTAMP/SIOCGHWTSTAMP.
Separate these from the few drivers that use ndo_do_ioctl to
implement SIOCBOND, SIOCBR and SIOCWANDEV commands.
This is a purely cosmetic change intended to help readers find
their way through the implementation.
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jay Vosburgh <j.vosburgh@gmail.com>
Cc: Veaceslav Falico <vfalico@gmail.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Vivien Didelot <vivien.didelot@gmail.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: Vladimir Oltean <olteanv@gmail.com>
Cc: Leon Romanovsky <leon@kernel.org>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow switchdevs to forward frames from the CPU in accordance with the
bridge configuration in the same way as is done between bridge
ports. This means that the bridge will only send a single skb towards
one of the ports under the switchdev's control, and expects the driver
to deliver the packet to all eligible ports in its domain.
Primarily this improves the performance of multicast flows with
multiple subscribers, as it allows the hardware to perform the frame
replication.
The basic flow between the driver and the bridge is as follows:
- When joining a bridge port, the switchdev driver calls
switchdev_bridge_port_offload() with tx_fwd_offload = true.
- The bridge sends offloadable skbs to one of the ports under the
switchdev's control using skb->offload_fwd_mark = true.
- The switchdev driver checks the skb->offload_fwd_mark field and lets
its FDB lookup select the destination port mask for this packet.
v1->v2:
- convert br_input_skb_cb::fwd_hwdoms to a plain unsigned long
- introduce a static key "br_switchdev_fwd_offload_used" to minimize the
impact of the newly introduced feature on all the setups which don't
have hardware that can make use of it
- introduce a check for nbp->flags & BR_FWD_OFFLOAD to optimize cache
line access
- reorder nbp_switchdev_frame_mark_accel() and br_handle_vlan() in
__br_forward()
- do not strip VLAN on egress if forwarding offload on VLAN-aware bridge
is being used
- propagate errors from .ndo_dfwd_add_station() if not EOPNOTSUPP
v2->v3:
- replace the solution based on .ndo_dfwd_add_station with a solution
based on switchdev_bridge_port_offload
- rename BR_FWD_OFFLOAD to BR_TX_FWD_OFFLOAD
v3->v4: rebase
v4->v5:
- make sure the static key is decremented on bridge port unoffload
- more function and variable renaming and comments for them:
br_switchdev_fwd_offload_used to br_switchdev_tx_fwd_offload
br_switchdev_accels_skb to br_switchdev_frame_uses_tx_fwd_offload
nbp_switchdev_frame_mark_tx_fwd to nbp_switchdev_frame_mark_tx_fwd_to_hwdom
nbp_switchdev_frame_mark_accel to nbp_switchdev_frame_mark_tx_fwd_offload
fwd_accel to tx_fwd_offload
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Any interraction with the buffer pool (seeding a buffer, acquire one) is
made through a software portal (SWP, a DPIO object).
There are circumstances where the dpaa2-switch driver probes on a DPSW
before any DPIO devices have been probed. In this case, seeding of the
buffer pool will lead to a panic since no SWPs are initialized.
To fix this, seed the buffer pool after making sure that the software
portals have been probed and are ready to be used.
Fixes: 0b1b713704 ("staging: dpaa2-switch: handle Rx path on control interface")
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Starting with commit 4f2673b3a2 ("net: bridge: add helper to replay
port and host-joined mdb entries"), DSA has introduced some bridge
helpers that replay switchdev events (FDB/MDB/VLAN additions and
deletions) that can be lost by the switchdev drivers in a variety of
circumstances:
- an IP multicast group was host-joined on the bridge itself before any
switchdev port joined the bridge, leading to the host MDB entries
missing in the hardware database.
- during the bridge creation process, the MAC address of the bridge was
added to the FDB as an entry pointing towards the bridge device
itself, but with no switchdev ports being part of the bridge yet, this
local FDB entry would remain unknown to the switchdev hardware
database.
- a VLAN/FDB/MDB was added to a bridge port that is a LAG interface,
before any switchdev port joined that LAG, leading to the hardware
database missing those entries.
- a switchdev port left a LAG that is a bridge port, while the LAG
remained part of the bridge, and all FDB/MDB/VLAN entries remained
installed in the hardware database of the switchdev port.
Also, since commit 0d2cfbd41c ("net: bridge: ignore switchdev events
for LAG ports which didn't request replay"), DSA introduced a method,
based on a const void *ctx, to ensure that two switchdev ports under the
same LAG that is a bridge port do not see the same MDB/VLAN entry being
replayed twice by the bridge, once for every bridge port that joins the
LAG.
With so many ordering corner cases being possible, it seems unreasonable
to expect a switchdev driver writer to get it right from the first try.
Therefore, now that DSA has experimented with the bridge replay helpers
for a little bit, we can move the code to the bridge driver where it is
more readily available to all switchdev drivers.
To convert the switchdev object replay helpers from "pull mode" (where
the driver asks for them) to a "push mode" (where the bridge offers them
automatically), the biggest problem is that the bridge needs to be aware
when a switchdev port joins and leaves, even when the switchdev is only
indirectly a bridge port (for example when the bridge port is a LAG
upper of the switchdev).
Luckily, we already have a hook for that, in the form of the newly
introduced switchdev_bridge_port_offload() and
switchdev_bridge_port_unoffload() calls. These offer a natural place for
hooking the object addition and deletion replays.
Extend the above 2 functions with:
- pointers to the switchdev atomic notifier (for FDB replays) and the
blocking notifier (for MDB and VLAN replays).
- the "const void *ctx" argument required for drivers to be able to
disambiguate between which port is targeted, when multiple ports are
lowers of the same LAG that is a bridge port. Most of the drivers pass
NULL to this argument, except the ones that support LAG offload and have
the proper context check already in place in the switchdev blocking
notifier handler.
Also unexport the replay helpers, since nobody except the bridge calls
them directly now.
Note that:
(a) we abuse the terminology slightly, because FDB entries are not
"switchdev objects", but we count them as objects nonetheless.
With no direct way to prove it, I think they are not modeled as
switchdev objects because those can only be installed by the bridge
to the hardware (as opposed to FDB entries which can be propagated
in the other direction too). This is merely an abuse of terms, FDB
entries are replayed too, despite not being objects.
(b) the bridge does not attempt to sync port attributes to newly joined
ports, just the countable stuff (the objects). The reason for this
is simple: no universal and symmetric way to sync and unsync them is
known. For example, VLAN filtering: what to do on unsync, disable or
leave it enabled? Similarly, STP state, ageing timer, etc etc. What
a switchdev port does when it becomes standalone again is not really
up to the bridge's competence, and the driver should deal with it.
On the other hand, replaying deletions of switchdev objects can be
seen a matter of cleanup and therefore be treated by the bridge,
hence this patch.
We make the replay helpers opt-in for drivers, because they might not
bring immediate benefits for them:
- nbp_vlan_init() is called _after_ netdev_master_upper_dev_link(),
so br_vlan_replay() should not do anything for the new drivers on
which we call it. The existing drivers where there was even a slight
possibility for there to exist a VLAN on a bridge port before they
join it are already guarded against this: mlxsw and prestera deny
joining LAG interfaces that are members of a bridge.
- br_fdb_replay() should now notify of local FDB entries, but I patched
all drivers except DSA to ignore these new entries in commit
2c4eca3ef7 ("net: bridge: switchdev: include local flag in FDB
notifications"). Driver authors can lift this restriction as they
wish, and when they do, they can also opt into the FDB replay
functionality.
- br_mdb_replay() should fix a real issue which is described in commit
4f2673b3a2 ("net: bridge: add helper to replay port and host-joined
mdb entries"). However most drivers do not offload the
SWITCHDEV_OBJ_ID_HOST_MDB to see this issue: only cpsw and am65_cpsw
offload this switchdev object, and I don't completely understand the
way in which they offload this switchdev object anyway. So I'll leave
it up to these drivers' respective maintainers to opt into
br_mdb_replay().
So most of the drivers pass NULL notifier blocks for the replay helpers,
except:
- dpaa2-switch which was already acked/regression-tested with the
helpers enabled (and there isn't much of a downside in having them)
- ocelot which already had replay logic in "pull" mode
- DSA which already had replay logic in "pull" mode
An important observation is that the drivers which don't currently
request bridge event replays don't even have the
switchdev_bridge_port_{offload,unoffload} calls placed in proper places
right now. This was done to avoid unnecessary rework for drivers which
might never even add support for this. For driver writers who wish to
add replay support, this can be used as a tentative placement guide:
https://patchwork.kernel.org/project/netdevbpf/patch/20210720134655.892334-11-vladimir.oltean@nxp.com/
Cc: Vadym Kochan <vkochan@marvell.com>
Cc: Taras Chornyi <tchornyi@marvell.com>
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>
Cc: Lars Povlsen <lars.povlsen@microchip.com>
Cc: Steen Hegelund <Steen.Hegelund@microchip.com>
Cc: UNGLinuxDriver@microchip.com
Cc: Claudiu Manoil <claudiu.manoil@nxp.com>
Cc: Alexandre Belloni <alexandre.belloni@bootlin.com>
Cc: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Ioana Ciornei <ioana.ciornei@nxp.com> # dpaa2-switch
Signed-off-by: David S. Miller <davem@davemloft.net>
On reception of an skb, the bridge checks if it was marked as 'already
forwarded in hardware' (checks if skb->offload_fwd_mark == 1), and if it
is, it assigns the source hardware domain of that skb based on the
hardware domain of the ingress port. Then during forwarding, it enforces
that the egress port must have a different hardware domain than the
ingress one (this is done in nbp_switchdev_allowed_egress).
Non-switchdev drivers don't report any physical switch id (neither
through devlink nor .ndo_get_port_parent_id), therefore the bridge
assigns them a hardware domain of 0, and packets coming from them will
always have skb->offload_fwd_mark = 0. So there aren't any restrictions.
Problems appear due to the fact that DSA would like to perform software
fallback for bonding and team interfaces that the physical switch cannot
offload.
+-- br0 ---+
/ / | \
/ / | \
/ | | bond0
/ | | / \
swp0 swp1 swp2 swp3 swp4
There, it is desirable that the presence of swp3 and swp4 under a
non-offloaded LAG does not preclude us from doing hardware bridging
beteen swp0, swp1 and swp2. The bandwidth of the CPU is often times high
enough that software bridging between {swp0,swp1,swp2} and bond0 is not
impractical.
But this creates an impossible paradox given the current way in which
port hardware domains are assigned. When the driver receives a packet
from swp0 (say, due to flooding), it must set skb->offload_fwd_mark to
something.
- If we set it to 0, then the bridge will forward it towards swp1, swp2
and bond0. But the switch has already forwarded it towards swp1 and
swp2 (not to bond0, remember, that isn't offloaded, so as far as the
switch is concerned, ports swp3 and swp4 are not looking up the FDB,
and the entire bond0 is a destination that is strictly behind the
CPU). But we don't want duplicated traffic towards swp1 and swp2, so
it's not ok to set skb->offload_fwd_mark = 0.
- If we set it to 1, then the bridge will not forward the skb towards
the ports with the same switchdev mark, i.e. not to swp1, swp2 and
bond0. Towards swp1 and swp2 that's ok, but towards bond0? It should
have forwarded the skb there.
So the real issue is that bond0 will be assigned the same hardware
domain as {swp0,swp1,swp2}, because the function that assigns hardware
domains to bridge ports, nbp_switchdev_add(), recurses through bond0's
lower interfaces until it finds something that implements devlink (calls
dev_get_port_parent_id with bool recurse = true). This is a problem
because the fact that bond0 can be offloaded by swp3 and swp4 in our
example is merely an assumption.
A solution is to give the bridge explicit hints as to what hardware
domain it should use for each port.
Currently, the bridging offload is very 'silent': a driver registers a
netdevice notifier, which is put on the netns's notifier chain, and
which sniffs around for NETDEV_CHANGEUPPER events where the upper is a
bridge, and the lower is an interface it knows about (one registered by
this driver, normally). Then, from within that notifier, it does a bunch
of stuff behind the bridge's back, without the bridge necessarily
knowing that there's somebody offloading that port. It looks like this:
ip link set swp0 master br0
|
v
br_add_if() calls netdev_master_upper_dev_link()
|
v
call_netdevice_notifiers
|
v
dsa_slave_netdevice_event
|
v
oh, hey! it's for me!
|
v
.port_bridge_join
What we do to solve the conundrum is to be less silent, and change the
switchdev drivers to present themselves to the bridge. Something like this:
ip link set swp0 master br0
|
v
br_add_if() calls netdev_master_upper_dev_link()
|
v bridge: Aye! I'll use this
call_netdevice_notifiers ^ ppid as the
| | hardware domain for
v | this port, and zero
dsa_slave_netdevice_event | if I got nothing.
| |
v |
oh, hey! it's for me! |
| |
v |
.port_bridge_join |
| |
+------------------------+
switchdev_bridge_port_offload(swp0, swp0)
Then stacked interfaces (like bond0 on top of swp3/swp4) would be
treated differently in DSA, depending on whether we can or cannot
offload them.
The offload case:
ip link set bond0 master br0
|
v
br_add_if() calls netdev_master_upper_dev_link()
|
v bridge: Aye! I'll use this
call_netdevice_notifiers ^ ppid as the
| | switchdev mark for
v | bond0.
dsa_slave_netdevice_event | Coincidentally (or not),
| | bond0 and swp0, swp1, swp2
v | all have the same switchdev
hmm, it's not quite for me, | mark now, since the ASIC
but my driver has already | is able to forward towards
called .port_lag_join | all these ports in hw.
for it, because I have |
a port with dp->lag_dev == bond0. |
| |
v |
.port_bridge_join |
for swp3 and swp4 |
| |
+------------------------+
switchdev_bridge_port_offload(bond0, swp3)
switchdev_bridge_port_offload(bond0, swp4)
And the non-offload case:
ip link set bond0 master br0
|
v
br_add_if() calls netdev_master_upper_dev_link()
|
v bridge waiting:
call_netdevice_notifiers ^ huh, switchdev_bridge_port_offload
| | wasn't called, okay, I'll use a
v | hwdom of zero for this one.
dsa_slave_netdevice_event : Then packets received on swp0 will
| : not be software-forwarded towards
v : swp1, but they will towards bond0.
it's not for me, but
bond0 is an upper of swp3
and swp4, but their dp->lag_dev
is NULL because they couldn't
offload it.
Basically we can draw the conclusion that the lowers of a bridge port
can come and go, so depending on the configuration of lowers for a
bridge port, it can dynamically toggle between offloaded and unoffloaded.
Therefore, we need an equivalent switchdev_bridge_port_unoffload too.
This patch changes the way any switchdev driver interacts with the
bridge. From now on, everybody needs to call switchdev_bridge_port_offload
and switchdev_bridge_port_unoffload, otherwise the bridge will treat the
port as non-offloaded and allow software flooding to other ports from
the same ASIC.
Note that these functions lay the ground for a more complex handshake
between switchdev drivers and the bridge in the future.
For drivers that will request a replay of the switchdev objects when
they offload and unoffload a bridge port (DSA, dpaa2-switch, ocelot), we
place the call to switchdev_bridge_port_unoffload() strategically inside
the NETDEV_PRECHANGEUPPER notifier's code path, and not inside
NETDEV_CHANGEUPPER. This is because the switchdev object replay helpers
need the netdev adjacency lists to be valid, and that is only true in
NETDEV_PRECHANGEUPPER.
Cc: Vadym Kochan <vkochan@marvell.com>
Cc: Taras Chornyi <tchornyi@marvell.com>
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>
Cc: Lars Povlsen <lars.povlsen@microchip.com>
Cc: Steen Hegelund <Steen.Hegelund@microchip.com>
Cc: UNGLinuxDriver@microchip.com
Cc: Claudiu Manoil <claudiu.manoil@nxp.com>
Cc: Alexandre Belloni <alexandre.belloni@bootlin.com>
Cc: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Ioana Ciornei <ioana.ciornei@nxp.com> # dpaa2-switch: regression
Acked-by: Ioana Ciornei <ioana.ciornei@nxp.com> # dpaa2-switch
Tested-by: Horatiu Vultur <horatiu.vultur@microchip.com> # ocelot-switch
Signed-off-by: David S. Miller <davem@davemloft.net>
Make more room for some extra code in the NETDEV_PRECHANGEUPPER handler
by moving what already exists into a dedicated function.
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to propagate the extack argument for
dpaa2_switch_port_bridge_join to use it in a future patch, and it looks
like there is already an error message there which is currently printed
to the console. Move it over netlink so it is properly transmitted to
user space.
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set SUPPORTED_FIBRE to mac_dev->if_support. It allows proper usage of
PHYs with optical/fiber support.
Signed-off-by: Maxim Kochetkov <fido_max@inbox.ru>
Acked-by: Madalin Bucur <madalin.bucur@oss.nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Daniel Borkmann says:
====================
pull-request: bpf-next 2021-06-28
The following pull-request contains BPF updates for your *net-next* tree.
We've added 37 non-merge commits during the last 12 day(s) which contain
a total of 56 files changed, 394 insertions(+), 380 deletions(-).
The main changes are:
1) XDP driver RCU cleanups, from Toke Høiland-Jørgensen and Paul E. McKenney.
2) Fix bpf_skb_change_proto() IPv4/v6 GSO handling, from Maciej Żenczykowski.
3) Fix false positive kmemleak report for BPF ringbuf alloc, from Rustam Kovhaev.
4) Fix x86 JIT's extable offset calculation for PROBE_LDX NULL, from Ravi Bangoria.
5) Enable libbpf fallback probing with tracing under RHEL7, from Jonathan Edwards.
6) Clean up x86 JIT to remove unused cnt tracking from EMIT macro, from Jiri Olsa.
7) Netlink cleanups for libbpf to please Coverity, from Kumar Kartikeya Dwivedi.
8) Allow to retrieve ancestor cgroup id in tracing programs, from Namhyung Kim.
9) Fix lirc BPF program query to use user-provided prog_cnt, from Sean Young.
10) Add initial libbpf doc including generated kdoc for its API, from Grant Seltzer.
11) Make xdp_rxq_info_unreg_mem_model() more robust, from Jakub Kicinski.
12) Fix up bpfilter startup log-level to info level, from Gary Lin.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In the case where the driver asks for a replay of a certain type of
event (port object or attribute) for a bridge port that is a LAG, it may
do so because this port has just joined the LAG.
But there might already be other switchdev ports in that LAG, and it is
preferable that those preexisting switchdev ports do not act upon the
replayed event.
The solution is to add a context to switchdev events, which is NULL most
of the time (when the bridge layer initiates the call) but which can be
set to a value controlled by the switchdev driver when a replay is
requested. The driver can then check the context to figure out if all
ports within the LAG should act upon the switchdev event, or just the
ones that match the context.
We have to modify all switchdev_handle_* helper functions as well as the
prototypes in the drivers that use these helpers too, because these
helpers hide the underlying struct switchdev_notifier_info from us and
there is no way to retrieve the context otherwise.
The context structure will be populated and used in later patches.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The newly implemented fwnode_mdbiobus_register turned out to be
problematic - in case the fwnode_/of_/acpi_mdio are built as
modules, a dependency cycle can be observed during the depmod phase of
modules_install, eg.:
depmod: ERROR: Cycle detected: fwnode_mdio -> of_mdio -> fwnode_mdio
depmod: ERROR: Found 2 modules in dependency cycles!
OR:
depmod: ERROR: Cycle detected: acpi_mdio -> fwnode_mdio -> acpi_mdio
depmod: ERROR: Found 2 modules in dependency cycles!
A possible solution could be to rework fwnode_mdiobus_register,
so that to merge the contents of acpi_mdiobus_register and
of_mdiobus_register. However feasible, such change would
be very intrusive and affect huge amount of the of_mdiobus_register
users.
Since there are currently 2 users of ACPI and MDIO
(xgmac_mdio and mvmdio), withdraw the fwnode_mdbiobus_register
and roll back to a simple 'if' condition in affected drivers.
Fixes: 62a6ef6a99 ("net: mdiobus: Introduce fwnode_mdbiobus_register()")
Signed-off-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The dpaa and dpaa2 drivers have rcu_read_lock()/rcu_read_unlock() pairs
around XDP program invocations. However, the actual lifetime of the objects
referred by the XDP program invocation is longer, all the way through to
the call to xdp_do_flush(), making the scope of the rcu_read_lock() too
small. This turns out to be harmless because it all happens in a single
NAPI poll cycle (and thus under local_bh_disable()), but it makes the
rcu_read_lock() misleading.
Rather than extend the scope of the rcu_read_lock(), just get rid of it
entirely. With the addition of RCU annotations to the XDP_REDIRECT map
types that take bh execution into account, lockdep even understands this to
be safe, so there's really no reason to keep it around.
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Camelia Groza <camelia.groza@nxp.com>
Cc: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Cc: Madalin Bucur <madalin.bucur@nxp.com>
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://lore.kernel.org/bpf/20210624160609.292325-11-toke@redhat.com
Utilize the newly added helper routine
for registering the MDIO bus via fwnode_
interface.
Signed-off-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As we know that AVB is enabled by default, and the ENET IP design is
queue 0 for best effort, queue 1&2 for AVB Class A&B. Bandwidth of each
queue 1&2 set in driver is 50%, TX bandwidth fluctuated when selecting
tx queues randomly with FEC_QUIRK_HAS_AVB quirk available.
This patch adds ndo_select_queue callback to select queues for
transmitting to fix this issue. It will always return queue 0 if this is
not a vlan packet, and return queue 1 or 2 based on priority of vlan
packet.
You may complain that in fact we only use single queue for trasmitting
if we are not targeted to VLAN. Yes, but seems we have no choice, since
AVB is enabled when the driver probed, we can't switch this feature
dynamicly. After compare multiple queues to single queue, TX throughput
almost no improvement.
One way we can implemet is to configure the driver to multiple queues
with Round-robin scheme by default. Then add ndo_setup_tc callback to
enable/disable AVB feature for users. Unfortunately, ENET AVB IP seems
not follow the standard 802.1Qav spec. We only can program
DMAnCFG[IDLE_SLOPE] field to calculate bandwidth fraction. And idle
slope is restricted to certain valus (a total of 19). It's far away from
CBS QDisc implemented in Linux TC framework. If you strongly suggest to do
this, I think we only can support limited numbers of bandwidth and reject
others, but it's really urgly and wried.
With this patch, VLAN tagged packets route to queue 0/1/2 based on vlan
priority; VLAN untagged packets route to queue 0.
Tested-by: Frieder Schrempf <frieder.schrempf@kontron.de>
Reported-by: Frieder Schrempf <frieder.schrempf@kontron.de>
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Frieder Schrempf reported a TX throuthput issue [1], it happens quite often
that the measured bandwidth in TX direction drops from its expected/nominal
value to something like ~50% (for 100M) or ~67% (for 1G) connections.
[1] https://lore.kernel.org/linux-arm-kernel/421cc86c-b66f-b372-32f7-21e59f9a98bc@kontron.de/
The issue becomes clear after digging into it, Net core would select
queues when transmitting packets. Since FEC have not impletemented
ndo_select_queue callback yet, so it will call netdev_pick_tx to select
queues randomly.
For i.MX6SX ENET IP with AVB support, driver default enables this
feature. According to the setting of QOS/RCMRn/DMAnCFG registers, AVB
configured to Credit-based scheme, 50% bandwidth of each queue 1&2.
With below tests let me think more:
1) With FEC_QUIRK_HAS_AVB quirk, can reproduce TX bandwidth fluctuations issue.
2) Without FEC_QUIRK_HAS_AVB quirk, can't reproduce TX bandwidth fluctuations issue.
The related difference with or w/o FEC_QUIRK_HAS_AVB quirk is that, whether we
program FTYPE field of TxBD or not. As I describe above, AVB feature is
enabled by default. With FEC_QUIRK_HAS_AVB quirk, frames in queue 0
marked as non-AVB, and frames in queue 1&2 marked as AVB Class A&B. It's
unreasonable if frames in queue 1&2 are not required to be time-sensitive.
So when Net core select tx queues ramdomly, Credit-based scheme would work
and lead to TX bandwidth fluctuated. On the other hand, w/o
FEC_QUIRK_HAS_AVB quirk, frames in queue 1&2 are all marked as non-AVB, so
Credit-based scheme would not work.
Till now, how can we fix this TX throughput issue? Yes, please remove
FEC_QUIRK_HAS_AVB quirk if you suffer it from time-nonsensitive networking.
However, this quirk is used to indicate i.MX6SX, other setting depends
on it. So this patch adds a new quirk FEC_QUIRK_HAS_MULTI_QUEUES to
represent i.MX6SX, it is safe for us remove FEC_QUIRK_HAS_AVB quirk
now.
FEC_QUIRK_HAS_AVB quirk is set by default in the driver, and users may
not know much about driver details, they would waste effort to find the
root cause, that is not we want. The following patch is a implementation
to fix it and users don't need to modify the driver.
Tested-by: Frieder Schrempf <frieder.schrempf@kontron.de>
Reported-by: Frieder Schrempf <frieder.schrempf@kontron.de>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Trivial conflicts in net/can/isotp.c and
tools/testing/selftests/net/mptcp/mptcp_connect.sh
scaled_ppm_to_ppb() was moved from drivers/ptp/ptp_clock.c
to include/linux/ptp_clock_kernel.h in -next so re-apply
the fix there.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Devices with RMON support has a 16-bit RDRP counter. It provides: "Receive
dropped packets counter. Increments for frames received which are streamed
to system but are later dropped due to lack of system resources."
To handle more than 2^16 dropped packets, a carry bit in CAR1 register is
set on overflow, so we enable irq when this is set, extending the counter
to 2^64 for handling situations where lots of packets are missed (e.g.
during heavy network storms).
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
These are for carry status and interrupt mask bits of statistics registers.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The memset on CAMx is wrong, as it actually unmasks all carry irq's,
which we clearly are not interested in.
The memset on CARx registers is just pointless, as they are W1C.
So let's just stop the memset before CAR1.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The CAR1 and CAR2 registers are W1C style registers, to the memset does not
actually clear them.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
No reason to wrap counter values at 2^32. Especially the bytes counters
can wrap pretty fast on Gbit networks.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
No reason to produce the legacy net_device_stats struct, only to have it
converted to rtnl_link_stats64. And as a bonus, this allows for improving
counter size to 64 bit.
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit da722186f6 ("net: fec: set GPR bit on suspend by DT configuration.")
refactor the fec_devtype, need adjust ptp driver accordingly.
Fixes: da722186f6 ("net: fec: set GPR bit on suspend by DT configuration.")
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add clock rate zero check to fix coverity issue of "divide by 0".
Fixes: commit 85bd1798b2 ("net: fec: fix spin_lock dead lock")
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Modify dpaa2_mac_get_node() to get the dpmac fwnode from either
DT or ACPI.
Modify dpaa2_mac_get_if_mode() to get interface mode from dpmac_node
which is a fwnode.
Modify dpaa2_pcs_create() to create pcs from dpmac_node fwnode.
Modify dpaa2_mac_connect() to support ACPI along with DT.
Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Rafael J. Wysocki <rafael@kernel.org> # from the ACPI side
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Depending on the device node type, call the specific OF or ACPI
mdiobus_register function.
Note: For both ACPI and DT cases, endianness of MDIO controllers
need to be specified using the "little-endian" property.
Signed-off-by: Calvin Johnson <calvin.johnson@oss.nxp.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Acked-by: Grant Likely <grant.likely@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use devm_platform_get_and_ioremap_resource() to simplify
code.
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The supplied buffer for the MAC address might not be aligned. Thus
doing a 32bit (or 16bit) access could be on an unaligned address. For
now, enetc is only used on aarch64 which can do unaligned accesses, thus
there is no error. In any case, be correct and use the get/put_unaligned
helpers.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After the refactoring introduced in commit 87614b931c ("net: enetc:
create a common enetc_pf_to_port helper"), enetc_pf_to_port was coded up
to return -1 in case the passed PCIe device does not have a recognized
BDF.
Make sure the -1 value is checked by the callers, to appease static
checkers.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Issue identified with Coccinelle.
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
When booting a board with DPAA2 interfaces defined statically via DPL
(as opposed to creating them dynamically using restool), the driver will
print an unspecific error message.
This change adds the error code to the message, and avoids printing
altogether if the error code is EPROBE_DEFER, because that is not a
cause of alarm.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Name the debugfs directory after the DPNI object instead of the netdev
name since this can be changed after probe by udev rules.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the DPNI object is connected to a DPMAC, setup the of_node to point
to the DTS device node of that specific MAC. This enables other drivers,
for example the DSA subsystem, to find the net_device by its device
node.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If MAC address read from nvmem efuse by calling .of_get_mac_address(),
but nvmem efuse is registered later than the driver, then it
return -EPROBE_DEFER value. So modify the driver to support
defer probe when read MAC address from nvmem efuse.
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the memory allocated for cbd_base is failed, it should
free the memory allocated for the queues, otherwise it causes
memory leak.
And if the memory allocated for the queues is failed, it can
return error directly.
Fixes: 59d0f74656 ("net: fec: init multi queue date structure")
Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case ethernet driver is enabled and INET is disabled, selftest will
fail to build.
Reported-by: Randy Dunlap <rdunlap@infradead.org>
Fixes: 3e1e58d64c ("net: add generic selftest support")
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20210428130947.29649-1-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The previous patch to support PTP Sync packet one-step timestamping
described one-step timestamping packet handling logic as below in
commit message:
- Trasmit packet immediately if no other one in transfer, or queue to
skb queue if there is already one in transfer.
The test_and_set_bit_lock() is used here to lock and check state.
- Start a work when complete transfer on hardware, to release the bit
lock and to send one skb in skb queue if has.
There was not problem of the description, but there was a mistake in
implementation. The locking/test_and_set_bit_lock() should be put in
enetc_start_xmit() which may be called by worker, rather than in
enetc_xmit(). Otherwise, the worker calling enetc_start_xmit() after
bit lock released is not able to lock again for transfer.
Fixes: 7294380c52 ("enetc: support PTP Sync packet one-step timestamping")
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A link time bug that I had fixed before has come back now that
another sub-module was added to the enetc driver:
ERROR: modpost: "enetc_ierb_register_pf" [drivers/net/ethernet/freescale/enetc/fsl-enetc.ko] undefined!
The problem is that the enetc Makefile is not actually used for
the ierb module if that is the only built-in driver in there
and everything else is a loadable module.
Fix it by always entering the directory this time, regardless
of which symbols are configured. This should reliably fix the
problem and prevent it from coming back another time.
Fixes: 112463ddbe ("net: dsa: felix: fix link error")
Fixes: e7d48e5fbf ("net: enetc: add a mini driver for the Integrated Endpoint Register Block")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that enetc supports flow control we have to make sure the settings in
the IERB are correct. Therefore, we actually depend on the enetc-ierb
module. Previously it was possible that this module was disabled while the
enetc was enabled. Fix it by automatically select the enetc-ierb module.
Fixes: e7d48e5fbf ("net: enetc: add a mini driver for the Integrated Endpoint Register Block")
Signed-off-by: Michael Walle <michael@walle.cc>
Acked-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With this patch FEC on iMX will able to run generic net selftests
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the ENETC receive path, a frame received by the MAC is first stored
in a 256KB 'FIFO' memory, then transferred to DRAM when enqueuing it to
the RX ring. The FIFO is a shared resource for all ENETC ports, but
every port keeps track of its own memory utilization, on RX and on TX.
There is a setting for RX rings through which they can either operate in
'lossy' mode (where the lack of a free buffer causes an immediate
discard of the frame) or in 'lossless' mode (where the lack of a free
buffer in the ring makes the frame stay longer in the FIFO).
In turn, when the memory utilization of the FIFO exceeds a certain
margin, the MAC can be configured to emit PAUSE frames.
There is enough FIFO memory to buffer up to 3 MTU-sized frames per RX
port while not jeopardizing the other use cases (jumbo frames), and
also not consume bytes from the port TX allocations. Also, 3 MTU-sized
frames worth of memory is enough to ensure zero loss for 64 byte packets
at 1G line rate.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The NXP ENETC is a 4-port Ethernet controller which 'smells' to
operating systems like 4 distinct PCIe PFs with SR-IOV, each PF having
its own driver instance, but in fact there are some hardware resources
which are shared between all ports, like for example the 256 KB SRAM
FIFO between the MACs and the Host Transfer Agent which DMAs frames to
DRAM.
To hide the stuff that cannot be neatly exposed per port, the hardware
designers came up with this idea of having a dedicated register block
which is supposed to be populated by the bootloader, and contains
everything configuration-related: MAC addresses, FIFO partitioning, etc.
When a port is reset using PCIe Function Level Reset, its defaults are
transferred from the IERB configuration. Most of the time, the settings
made through the IERB are read-only in the port's memory space (if they
are even visible), so they cannot be modified at runtime.
Linux doesn't have any advanced FIFO partitioning requirements at all,
but when reading through the hardware manual, it became clear that, even
though there are many good 'recommendations' for default values, many of
them were not actually put in practice on LS1028A. So we end up with a
default configuration that:
(a) does not have enough TX and RX byte credits to support the max MTU
of 9600 (which the Linux driver claims already) properly (at full speed)
(b) allows the FIFO to be overrun with RX traffic, potentially
overwriting internal data structures.
The last part sounds a bit catastrophic, but it isn't. Frames are
supposed to transit the FIFO for a very short time, but they can
actually accumulate there under 2 conditions:
(a) there is very severe congestion on DRAM memory, or
(b) the RX rings visible to the operating system were configured for
lossless operation, and they just ran out of free buffers to copy
the frame to. This is what is used to put backpressure onto the MAC
with flow control.
So since ENETC has not supported flow control thus far, RX FIFO overruns
were never seen with Linux. But with the addition of flow control, we
should configure some registers to prevent this from happening. What we
are trying to protect against are bad actors which continue to send us
traffic despite the fact that we have signaled a PAUSE condition. Of
course we can't be lossless in that case, but it is best to configure
the FIFO to do tail dropping rather than letting it overrun.
So in a nutshell, this driver is a fixup for all the IERB default values
that should have been but aren't.
The IERB configuration needs to be done _before_ the PFs are enabled.
So every PF searches for the presence of the "fsl,ls1028a-enetc-ierb"
node in the device tree, and if it finds it, it "registers" with the
IERB, which means that it requests the IERB to fix up its default
values. This is done through -EPROBE_DEFER. The IERB driver is part of
the fsl_enetc module, but is technically a platform driver, since the
IERB is a good old fashioned MMIO region, as opposed to ENETC ports
which pretend to be PCIe devices.
The driver was already configuring ENETC_PTXMBAR (FIFO allocation for
TX) because due to an omission, TXMBAR is a read/write register in the
PF memory space. But the manual is quite clear that the formula for this
should depend upon the TX byte credits (TXBCR). In turn, the TX byte
credits are only readable/writable through the IERB. So if we want to
ensure that the TXBCR register also has a value that is correct and in
line with TXMBAR, there is simply no way this can be done from the PF
driver, access to the IERB is needed.
I could have modified U-Boot to fix up the IERB values, but that is
quite undesirable, as old U-Boot versions are likely to be floating
around for quite some time from now.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Even though ENETC interfaces are exposed as individual PCIe PFs with
their own driver instances, the ENETC is still fundamentally a
multi-port Ethernet controller, and some parts of the IP take a port
number (as can be seen in the PSFP implementation).
Create a common helper that can be used outside of the TSN code for
retrieving the ENETC port number based on the PF number. This is only
correct for LS1028A, the only Linux-capable instantiation of ENETC thus
far.
Note that ENETC port 3 is PF 6. The TSN code did not care about this
because ENETC port 3 does not support TSN, so the wrong mapping done by
enetc_get_port for PF 6 could have never been hit.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Described in fd5736bf9f ("enetc: Workaround for MDIO register access
issue") is a workaround for a hardware bug that requires a register
access of the MDIO controller to never happen concurrently with a
register access of a port PF. To avoid that, a mutual exclusion scheme
with rwlocks was implemented - the port PF accessors are the 'read'
side, and the MDIO accessors are the 'write' side.
When we do XDP_REDIRECT between two ENETC interfaces, all is fine
because the MDIO lock is already taken from the NAPI poll loop.
But when the ingress interface is not ENETC, just the egress is, the
MDIO lock is not taken, so we might access the port PF registers
concurrently with MDIO, which will make the link flap due to wrong
values returned from the PHY.
To avoid this, let's just slap an enetc_lock_mdio/enetc_unlock_mdio at
the beginning and ending of enetc_xdp_xmit. The fact that the MDIO lock
is designed as a rwlock is important here, because the read side is
reentrant (that is one of the main reasons why we chose it). Usually,
the way we benefit of its reentrancy is by running the data path
concurrently on both CPUs, but in this case, we benefit from the
reentrancy by taking the lock even when the lock is already taken
(and that's the situation where ENETC is both the ingress and the egress
interface for XDP_REDIRECT, which was fine before and still is fine now).
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the TX ring is congested, enetc_xdp_tx() returns false for the
current XDP frame (represented as an array of software BDs).
This array of software TX BDs is constructed in enetc_rx_swbd_to_xdp_tx_swbd
from software BDs freshly cleaned from the RX ring. The issue is that we
scrub the RX software BDs too soon, more precisely before we know that
we can enqueue the TX BDs successfully into the TX ring.
If we can't enqueue them (and enetc_xdp_tx returns false), we call
enetc_xdp_drop which attempts to recycle the buffers held by the RX
software BDs. But because we scrubbed those RX BDs already, two things
happen:
(a) we leak their memory
(b) we populate the RX software BD ring with an all-zero rx_swbd
structure, which makes the buffer refill path allocate more memory.
enetc_refill_rx_ring
-> if (unlikely(!rx_swbd->page))
-> enetc_new_page
That is a recipe for fast OOM.
Fixes: 7ed2bc8007 ("net: enetc: add support for XDP_TX")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the XDP program returns an invalid action, we should free the RX
buffer.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is possible for one CPU to perform TX hashing (see netdev_pick_tx)
between the 8 ENETC TX rings, and the TX hashing to select TX queue 1.
At the same time, it is possible for the other CPU to already use TX
ring 1 for XDP (either XDP_TX or XDP_REDIRECT). Since there is no mutual
exclusion between XDP and the network stack, we run into an issue
because the ENETC TX procedure is not reentrant.
The obvious approach would be to just make XDP take the lock of the
network stack's TX queue corresponding to the ring it's about to enqueue
in.
For XDP_REDIRECT, this is quite straightforward, a lock at the beginning
and end of enetc_xdp_xmit() should do the trick.
But for XDP_TX, it's a bit more complicated. For one, we do TX batching
all by ourselves for frames with the XDP_TX verdict. This is something
we would like to keep the way it is, for performance reasons. But
batching means that the network stack's lock should be kept from the
first enqueued XDP_TX frame and until we ring the doorbell. That is
mostly fine, except for cases when in the same NAPI loop we have mixed
XDP_TX and XDP_REDIRECT frames. So if enetc_xdp_xmit() gets called while
we are holding the lock from the RX NAPI, then bam, deadlock. The naive
answer could be 'just flush the XDP_TX frames first, then release the
network stack's TX queue lock, then call xdp_do_flush_map()'. But even
xdp_do_redirect() is capable of flushing the batched XDP_REDIRECT
frames, so unless we unlock/relock the TX queue around xdp_do_redirect(),
there simply isn't any clean way to protect XDP_TX from concurrent
network stack .ndo_start_xmit() on another CPU.
So we need to take a different approach, and that is to reserve two
rings for the sole use of XDP. We leave TX rings
0..ndev->real_num_tx_queues-1 to be handled by the network stack, and we
pick them from the end of the priv->tx_ring array.
We make an effort to keep the mapping done by enetc_alloc_msix() which
decides which CPU handles the TX completions of which TX ring in its
NAPI poll. So the XDP TX ring of CPU 0 is handled by TX ring 6, and the
XDP TX ring of CPU 1 is handled by TX ring 7.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that commit d6a2829e82 ("net: enetc: increase RX ring default
size") has increased the RX ring size, it is quite easy to congest the
TX rings when the traffic is predominantly XDP_TX, as the RX ring is
quite a bit larger than the TX one.
Since we bit the bullet and did the expensive thing already (larger RX
rings consume more memory pages), it seems quite foolish to keep the TX
rings small. So make them equally sized with TX.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
xdp_do_redirect already contains:
-> dev_map_enqueue
-> __xdp_enqueue
-> bq_enqueue
-> bq_xmit_all // if we have more than 16 frames
So the logic from enetc will never be hit, because ENETC_DEFAULT_TX_WORK
is 128.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the code path below fails:
enetc_clean_rx_ring_xdp // XDP_PASS
-> enetc_build_skb
-> enetc_map_rx_buff_to_skb
-> build_skb
enetc_clean_rx_ring_xdp will 'break', but that 'break' instruction isn't
strong enough to actually break the NAPI poll loop, just the switch/case
statement for XDP actions. So we increment rx_frm_cnt and go to the next
frames minding our own business.
Instead let's do what the skb NAPI poll function does, and break the
loop now, waiting for the memory pressure to go away. Otherwise the next
calls to build_skb() are likely to fail too.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When receiving a frame with errors, currently we do nothing with it (we
don't construct an skb or an xdp_buff), we just exit the NAPI poll loop.
Let's put the buffer back into the RX ring (similar to XDP_DROP).
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
enetc_put_xdp_buff has nothing to do with XDP, frankly, it is just a
helper to populate the recycle end of the shadow RX BD ring
(next_to_alloc) with a given buffer.
On the other hand, enetc_put_rx_buff plays more tricks than its name
would suggest.
So let's rename enetc_put_rx_buff into enetc_flip_rx_buff to reflect the
half-page buffer reuse tricks that it employs, and enetc_put_xdp_buff
into enetc_put_rx_buff which suggests a more garden-variety operation.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Later in enetc_clean_tx_ring we have:
/* Scrub the swbd here so we don't have to do that
* when we reuse it during xmit
*/
memset(tx_swbd, 0, sizeof(*tx_swbd));
So these assignments are unnecessary.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Gianfar used to enable all 8 Rx queues (DMA rings) per
ethernet device, even though the controller can only
support 2 interrupt lines at most. This meant that
multiple Rx queues would have to be grouped per NAPI poll
routine, and the CPU would have to split the budget and
service them in a round robin manner. The overhead of
this scheme proved to outweight the potential benefits.
The alternative was to introduce the "Single Queue" polling
mode, supporting one Rx queue per NAPI, which became the
default packet processing option and helped improve the
performance of the driver.
MQ_POLLING also relies on undocumeted device tree properties
to specify how to map the 8 Rx and Tx queues to a given
interrupt line (aka "interrupt group"). Using module parameters
to enable this mode wasn't an option either. Long story short,
MQ_POLLING became obsolete, now it is just dead code, and no
one asked for it so far.
For the Tx queues, multi-queue support (more than 1 Tx queue
per CPU) could be revisited by adding tc MQPRIO support, but
again, one has to consider that there are only 2 interrupt lines.
So the NAPI poll routine would have to service multiple Tx rings.
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As explained in bugfix commit 6ab4c3117a ("net: bridge: don't notify
switchdev for local FDB addresses") as well as in this discussion:
https://lore.kernel.org/netdev/20210117193009.io3nungdwuzmo5f7@skbuf/
the switchdev notifiers for FDB entries managed to have a zero-day bug,
which was that drivers would not know what to do with local FDB entries,
because they were not told that they are local. The bug fix was to
simply not notify them of those addresses.
Let us now add the 'is_local' bit to bridge FDB entries, and make all
drivers ignore these entries by their own choice.
Co-developed-by: Tobias Waldekranz <tobias@waldekranz.com>
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert system_wq queue_work() to schedule_work() which is
a wrapper around it, since the former is a rare construct.
Fixes: 7294380c52 ("enetc: support PTP Sync packet one-step timestamping")
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Normally, the bootloader will already initialize the MAC address
registers of the ENETC and the driver will just use them or generate a
random one, if it is not initialized.
Add a new way to provide the MAC address: via device tree. Besides the
usual 'mac-address' property, there is also the possibility to fetch it
via a NVMEM provider. The sl28 board stores the MAC address in the SPI
NOR flash OTP region. Having this will allow linux to fetch the MAC
address from there without being dependent on the bootloader.
No in-tree boards have the device tree properties set, thus for these,
this is a no-op.
Signed-off-by: Michael Walle <michael@walle.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since we added the dpaa2_switch_acl_entry_add() function in the previous
patches to hide all the details of actually adding the ACL entry by
issuing a firmware command, let's use it also for adding a CPU trap for
the STP frames.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support TC_SETUP_CLSMATCHALL by using the same ACL table entries
framework as for tc flower. Adding a matchall rule is done by installing
an entry which has a mask of all zeroes, thus matching on any packet.
This can be used as a catch-all type of rule if used correctly, ie the
priority of the matchall filter should be kept as the lowest one in the
entire filter block.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for tc flower hardware offload on the ingress
path. Shared filter blocks are supported by sharing a single ACL table
between multiple ports.
The following flow keys are supported:
- Ethernet: dst_mac/src_mac
- IPv4: dst_ip/src_ip/ip_proto/tos
- VLAN: vlan_id/vlan_prio/vlan_tpid/vlan_dei
- L4: dst_port/src_port
As per flow actions, the following are supported:
- drop
- mirred egress redirect
- trap
Each ACL entry (filter) can be setup with only one of the listed
actions.
A sorted single linked list is used to keep the ACL entries by their
order of priority. When adding a new filter, this enables us to quickly
ascertain if the new entry has the highest priority of the entire block
or if we should make some space in the ACL table by increasing the
priority of the filters already in the table.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change the default ACL trap rule for STP frames to have the highest
priority.
In the same ACL table will reside both default rules added by the driver
for its internal use as well as rules added with tc flower. In this
case, the default rules such as the STP one that we already have should
have the highest priority.
Also, remove the check for a full ACL table since we already know that
it's sized so that we don't hit this case. The last thing changes is
that default trap filters will not be counted in the acl_tbl's num_rules
variable since their number doesn't change.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce a new structure - dpaa2_switch_acl_tbl - to hold all data
related to an ACL table: number of rules added, ACL table id, etc.
This will be used more in the next patches when adding support for
sharing an ACL table between ports.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
of_get_mac_address() returns a "const void*" pointer to a MAC address.
Lately, support to fetch the MAC address by an NVMEM provider was added.
But this will only work with platform devices. It will not work with
PCI devices (e.g. of an integrated root complex) and esp. not with DSA
ports.
There is an of_* variant of the nvmem binding which works without
devices. The returned data of a nvmem_cell_read() has to be freed after
use. On the other hand the return of_get_mac_address() points to some
static data without a lifetime. The trick for now, was to allocate a
device resource managed buffer which is then returned. This will only
work if we have an actual device.
Change it, so that the caller of of_get_mac_address() has to supply a
buffer where the MAC address is written to. Unfortunately, this will
touch all drivers which use the of_get_mac_address().
Usually the code looks like:
const char *addr;
addr = of_get_mac_address(np);
if (!IS_ERR(addr))
ether_addr_copy(ndev->dev_addr, addr);
This can then be simply rewritten as:
of_get_mac_address(np, ndev->dev_addr);
Sometimes is_valid_ether_addr() is used to test the MAC address.
of_get_mac_address() already makes sure, it just returns a valid MAC
address. Thus we can just test its return code. But we have to be
careful if there are still other sources for the MAC address before the
of_get_mac_address(). In this case we have to keep the
is_valid_ether_addr() call.
The following coccinelle patch was used to convert common cases to the
new style. Afterwards, I've manually gone over the drivers and fixed the
return code variable: either used a new one or if one was already
available use that. Mansour Moufid, thanks for that coccinelle patch!
<spml>
@a@
identifier x;
expression y, z;
@@
- x = of_get_mac_address(y);
+ x = of_get_mac_address(y, z);
<...
- ether_addr_copy(z, x);
...>
@@
identifier a.x;
@@
- if (<+... x ...+>) {}
@@
identifier a.x;
@@
if (<+... x ...+>) {
...
}
- else {}
@@
identifier a.x;
expression e;
@@
- if (<+... x ...+>@e)
- {}
- else
+ if (!(e))
{...}
@@
expression x, y, z;
@@
- x = of_get_mac_address(y, z);
+ of_get_mac_address(y, z);
... when != x
</spml>
All drivers, except drivers/net/ethernet/aeroflex/greth.c, were
compile-time tested.
Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is to add support for PTP Sync packet one-step timestamping.
Since ENETC single-step register has to be configured dynamically per
packet for correctionField offeset and UDP checksum update, current
one-step timestamping packet has to be sent only when the last one
completes transmitting on hardware. So, on the TX, this patch handles
one-step timestamping packet as below:
- Trasmit packet immediately if no other one in transfer, or queue to
skb queue if there is already one in transfer.
The test_and_set_bit_lock() is used here to lock and check state.
- Start a work when complete transfer on hardware, to release the bit
lock and to send one skb in skb queue if has.
And the configuration for one-step timestamping on ENETC before
transmitting is,
- Set one-step timestamping flag in extension BD.
- Write 30 bits current timestamp in tstamp field of extension BD.
- Update PTP Sync packet originTimestamp field with current timestamp.
- Configure single-step register for correctionField offeset and UDP
checksum update.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Mark TX timestamp type per skb on skb->cb[0], instead of
global variable for all skbs. This is a preparation for
one step timestamp support.
For one-step timestamping enablement, there will be both
one-step and two-step PTP messages to transfer. And a skb
queue is needed for one-step PTP messages making sure
start to send current message only after the last one
completed on hardware. (ENETC single-step register has to
be dynamically configured per message.) So, marking TX
timestamp type per skb is required.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
MAINTAINERS
- keep Chandrasekar
drivers/net/ethernet/mellanox/mlx5/core/en_main.c
- simple fix + trust the code re-added to param.c in -next is fine
include/linux/bpf.h
- trivial
include/linux/ethtool.h
- trivial, fix kdoc while at it
include/linux/skmsg.h
- move to relevant place in tcp.c, comment re-wrapped
net/core/skmsg.c
- add the sk = sk // sk = NULL around calls
net/tipc/crypto.c
- trivial
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Even if the current mapping is correct for the 1 CPU and 2 CPU cases
(currently enetc is included in SoCs with up to 2 CPUs only), better
use a generic rule for the mapping to cover all possible cases.
The number of CPUs is the same as the number of interrupt vectors:
Per device Tx rings -
device_tx_ring[idx], where idx = 0..n_rings_total-1
Per interrupt vector Tx rings -
int_vector[i].ring[j], where i = 0..n_int_vects-1
j = 0..n_rings_per_v-1
Mapping rule -
n_rings_per_v = n_rings_total / n_int_vects
for i = 0..n_int_vects - 1:
for j = 0..n_rings_per_v - 1:
idx = n_int_vects * j + i
int_vector[i].ring[j] <- device_tx_ring[idx]
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20210409071613.28912-1-claudiu.manoil@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The blamed commit introduced a bit in the TX software buffer descriptor
structure for determining whether a BD is final or not; we rearm the TX
interrupt vector for every frame (hence final BD) transmitted.
But there is a problem with the patch: it replaced a condition whose
expression is a bool which was evaluated at the beginning of the "while"
loop with a bool expression that is evaluated on the spot: tx_swbd->is_eof.
The problem with the latter expression is that the tx_swbd has already
been incremented at that stage, so the tx_swbd->is_eof check is in fact
with the _next_ software BD. Which is _not_ final.
The effect is that the CPU is in 100% load with ksoftirqd because it
does not acknowledge the TX interrupt, so the handler keeps getting
called again and again.
The fix is to restore the code structure, and keep the local bool is_eof
variable, just to assign it the tx_swbd->is_eof value instead of
!!tx_swbd->skb.
Fixes: d504498d2e ("net: enetc: add a dedicated is_eof bit in the TX software BD")
Reported-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Link: https://lore.kernel.org/r/20210409192759.3895104-1-olteanv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This loop will try to unmap enetc_unmap_tx_buff[-1] and crash.
Fixes: 9d2b68cc10 ("net: enetc: add support for XDP_REDIRECT")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/YHBHfCY/yv3EnM9z@mwanda
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the new mac_managed_pm flag to work around an issue with KSZ8081 PHY
that becomes unstable when a soft reset is triggered during aneg.
Reported-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Tested-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It's useful, especially for debugging purposes, to have the Rx copybreak
value changeable at runtime. Export it as an ethtool tunable.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
DMA unmapping, allocating a new buffer and DMA mapping it back on the
refill path is really not that efficient. Proper buffer recycling (page
pool, flipping the page and using the other half) cannot be done for
DPAA2 since it's not a ring based controller but it rather deals with
multiple queues which all get their buffers from the same buffer pool on
Rx.
To circumvent these limitations, add support for Rx copybreak. For small
sized packets instead of creating a skb around the buffer in which the
frame was received, allocate a new sk buffer altogether, copy the
contents of the frame and release the initial page back into the buffer
pool.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rename the dpaa2_eth_xdp_release_buf function into dpaa2_eth_recycle_buf
since in the next patches we'll be using the same recycle mechanism for
the normal stack path beside for XDP_DROP.
Also, rename the array which holds the buffers to be recycled so that it
does not have any reference to XDP.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver implementation of the XDP_REDIRECT action reuses parts from
XDP_TX, most notably the enetc_xdp_tx function which transmits an array
of TX software BDs. Only this time, the buffers don't have DMA mappings,
we need to create them.
When a BPF program reaches the XDP_REDIRECT verdict for a frame, we can
employ the same buffer reuse strategy as for the normal processing path
and for XDP_PASS: we can flip to the other page half and seed that to
the RX ring.
Note that scatter/gather support is there, but disabled due to lack of
multi-buffer support in XDP (which is added by this series):
https://patchwork.kernel.org/project/netdevbpf/cover/cover.1616179034.git.lorenzo@kernel.org/
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As explained in the XDP_TX patch, when receiving a burst of frames with
the XDP_TX verdict, there is a momentary dip in the number of available
RX buffers. The system will eventually recover as TX completions will
start kicking in and refilling our RX BD ring again. But until that
happens, we need to survive with as few out-of-buffer discards as
possible.
This increases the memory footprint of the driver in order to avoid
discards at 2.5Gbps line rate 64B packet sizes, the maximum speed
available for testing on 1 port on NXP LS1028A.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For reflecting packets back into the interface they came from, we create
an array of TX software BDs derived from the RX software BDs. Therefore,
we need to extend the TX software BD structure to contain most of the
stuff that's already present in the RX software BD structure, for
reasons that will become evident in a moment.
For a frame with the XDP_TX verdict, we don't reuse any buffer right
away as we do for XDP_DROP (the same page half) or XDP_PASS (the other
page half, same as the skb code path).
Because the buffer transfers ownership from the RX ring to the TX ring,
reusing any page half right away is very dangerous. So what we can do is
we can recycle the same page half as soon as TX is complete.
The code path is:
enetc_poll
-> enetc_clean_rx_ring_xdp
-> enetc_xdp_tx
-> enetc_refill_rx_ring
(time passes, another MSI interrupt is raised)
enetc_poll
-> enetc_clean_tx_ring
-> enetc_recycle_xdp_tx_buff
But that creates a problem, because there is a potentially large time
window between enetc_xdp_tx and enetc_recycle_xdp_tx_buff, period in
which we'll have less and less RX buffers.
Basically, when the ship starts sinking, the knee-jerk reaction is to
let enetc_refill_rx_ring do what it does for the standard skb code path
(refill every 16 consumed buffers), but that turns out to be very
inefficient. The problem is that we have no rx_swbd->page at our
disposal from the enetc_reuse_page path, so enetc_refill_rx_ring would
have to call enetc_new_page for every buffer that we refill (if we
choose to refill at this early stage). Very inefficient, it only makes
the problem worse, because page allocation is an expensive process, and
CPU time is exactly what we're lacking.
Additionally, there is an even bigger problem: if we let
enetc_refill_rx_ring top up the ring's buffers again from the RX path,
remember that the buffers sent to transmission haven't disappeared
anywhere. They will be eventually sent, and processed in
enetc_clean_tx_ring, and an attempt will be made to recycle them.
But surprise, the RX ring is already full of new buffers, because we
were premature in deciding that we should refill. So not only we took
the expensive decision of allocating new pages, but now we must throw
away perfectly good and reusable buffers.
So what we do is we implement an elastic refill mechanism, which keeps
track of the number of in-flight XDP_TX buffer descriptors. We top up
the RX ring only up to the total ring capacity minus the number of BDs
that are in flight (because we know that those BDs will return to us
eventually).
The enetc driver manages 1 RX ring per CPU, and the default TX ring
management is the same. So we do XDP_TX towards the TX ring of the same
index, because it is affined to the same CPU. This will probably not
produce great results when we have a tc-taprio/tc-mqprio qdisc on the
interface, because in that case, the number of TX rings might be
greater, but I didn't add any checks for that yet (mostly because I
didn't know what checks to add).
It should also be noted that we need to change the DMA mapping direction
for RX buffers, since they may now be reflected into the TX ring of the
same device. We choose to use DMA_BIDIRECTIONAL instead of unmapping and
remapping as DMA_TO_DEVICE, because performance is better this way.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the RX ring, enetc uses an allocation scheme based on pages split
into two buffers, which is already very efficient in terms of preventing
reallocations / maximizing reuse, so I see no reason why I would change
that.
+--------+--------+--------+--------+--------+--------+--------+
| | | | | | | |
| half B | half B | half B | half B | half B | half B | half B |
| | | | | | | |
+--------+--------+--------+--------+--------+--------+--------+
| | | | | | | |
| half A | half A | half A | half A | half A | half A | half A | RX ring
| | | | | | | |
+--------+--------+--------+--------+--------+--------+--------+
^ ^
| |
next_to_clean next_to_alloc
next_to_use
+--------+--------+--------+--------+--------+
| | | | | |
| half B | half B | half B | half B | half B |
| | | | | |
+--------+--------+--------+--------+--------+--------+--------+
| | | | | | | |
| half B | half B | half A | half A | half A | half A | half A | RX ring
| | | | | | | |
+--------+--------+--------+--------+--------+--------+--------+
| | | ^ ^
| half A | half A | | |
| | | next_to_clean next_to_use
+--------+--------+
^
|
next_to_alloc
then when enetc_refill_rx_ring is called, whose purpose is to advance
next_to_use, it sees that it can take buffers up to next_to_alloc, and
it says "oh, hey, rx_swbd->page isn't NULL, I don't need to allocate
one!".
The only problem is that for default PAGE_SIZE values of 4096, buffer
sizes are 2048 bytes. While this is enough for normal skb allocations at
an MTU of 1500 bytes, for XDP it isn't, because the XDP headroom is 256
bytes, and including skb_shared_info and alignment, we end up being able
to make use of only 1472 bytes, which is insufficient for the default
MTU.
To solve that problem, we implement scatter/gather processing in the
driver, because we would really like to keep the existing allocation
scheme. A packet of 1500 bytes is received in a buffer of 1472 bytes and
another one of 28 bytes.
Because the headroom required by XDP is different (and much larger) than
the one required by the network stack, whenever a BPF program is added
or deleted on the port, we drain the existing RX buffers and seed new
ones with the required headroom. We also keep the required headroom in
rx_ring->buffer_offset.
The simplest way to implement XDP_PASS, where an skb must be created, is
to create an xdp_buff based on the next_to_clean RX BDs, but not clear
those BDs from the RX ring yet, just keep the original index at which
the BDs for this frame started. Then, if the verdict is XDP_PASS,
instead of converting the xdb_buff to an skb, we replay a call to
enetc_build_skb (just as in the normal enetc_clean_rx_ring case),
starting from the original BD index.
We would also like to be minimally invasive to the regular RX data path,
and not check whether there is a BPF program attached to the ring on
every packet. So we create a separate RX ring processing function for
XDP.
Because we only install/remove the BPF program while the interface is
down, we forgo the rcu_read_lock() in enetc_clean_rx_ring, since there
shouldn't be any circumstance in which we are processing packets and
there is a potentially freed BPF program attached to the RX ring.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For XDP_TX, we need to call enetc_reuse_page from enetc_clean_tx_ring,
so we need to avoid a forward declaration.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With the future introduction of some new fields into enetc_tx_swbd such
as is_xdp_tx, is_xdp_redirect etc, we need not only to set these bits
to true from the XDP_TX/XDP_REDIRECT code path, but also to false from
the old code paths.
This is because TX software buffer descriptors are kept in a ring that
is shadow of the hardware TX ring, so these structures keep getting
reused, and there is always the possibility that when a software BD is
reused (after we ran a full circle through the TX ring), the old user of
the tx_swbd had set is_xdp_tx = true, and now we are sending a regular
skb, which would need to set is_xdp_tx = false.
To be minimally invasive to the old code paths, let's just scrub the
software TX BD in the TX confirmation path (enetc_clean_tx_ring), once
we know that nobody uses this software TX BD (tx_ring->next_to_clean
hasn't yet been updated, and the TX paths check enetc_bd_unused which
tells them if there's any more space in the TX ring for a new enqueue).
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the transmit path, if we have a scatter/gather frame, it is put into
multiple software buffer descriptors, the last of which has the skb
pointer populated (which is necessary for rearming the TX MSI vector and
for collecting the two-step TX timestamp from the TX confirmation path).
At the moment, this is sufficient, but with XDP_TX, we'll need to
service TX software buffer descriptors that don't have an skb pointer,
however they might be final nonetheless. So add a dedicated bit for
final software BDs that we populate and check explicitly. Also, we keep
looking just for an skb when doing TX timestamping, because we don't
want/need that for XDP.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to build an skb from two code paths now: from the plain RX data
path and from the XDP data path when the verdict is XDP_PASS.
Create a new enetc_build_skb function which contains the essential steps
for building an skb based on the first and last positions of buffer
descriptors within the RX ring.
We also squash the enetc_process_skb function into enetc_build_skb,
because what that function did wasn't very meaningful on its own.
The "rx_frm_cnt++" instruction has been moved around napi_gro_receive
for cosmetic reasons, to be in the same spot as rx_byte_cnt++, which
itself must be before napi_gro_receive, because that's when we lose
ownership of the skb.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
