This reverts commit 11fd667dac.
dsa_slave_change_mtu() updates the MTU of the DSA master and of the
associated CPU port, but only if it detects a change to the master MTU.
The blamed commit in the Fixes: tag below addressed a regression where
dsa_slave_change_mtu() would return early and not do anything due to
ds->ops->port_change_mtu() not being implemented.
However, that commit also had the effect that the master MTU got set up
to the correct value by dsa_master_setup(), but the associated CPU port's
MTU did not get updated. This causes breakage for drivers that rely on
the ->port_change_mtu() DSA call to account for the tagging overhead on
the CPU port, and don't set up the initial MTU during the setup phase.
Things actually worked before because they were in a fragile equilibrium
where dsa_slave_change_mtu() was called before dsa_master_setup() was.
So dsa_slave_change_mtu() could actually detect a change and update the
CPU port MTU too.
Restore the code to the way things used to work by reverting the reorder
of dsa_tree_setup_master() and dsa_tree_setup_ports(). That change did
not have a concrete motivation going for it anyway, it just looked
better.
Fixes: 066dfc4290 ("Revert "net: dsa: stop updating master MTU from master.c"")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
DSA probing is atypical because a tree of devices must probe all at
once, so out of N switches which call dsa_tree_setup_routing_table()
during probe, for (N - 1) of them, "complete" will return false and they
will exit probing early. The Nth switch will set up the whole tree on
their behalf.
The implication is that for (N - 1) switches, the driver binds to the
device successfully, without doing anything. When the driver is bound,
the ->shutdown() method may run. But if the Nth switch has failed to
initialize the tree, there is nothing to do for the (N - 1) driver
instances, since the slave devices have not been created, etc. Moreover,
dsa_switch_shutdown() expects that the calling @ds has been in fact
initialized, so it jumps at dereferencing the various data structures,
which is incorrect.
Avoid the ensuing NULL pointer dereferences by simply checking whether
the Nth switch has previously set "ds->setup = true" for the switch
which is currently shutting down. The entire setup is serialized under
dsa2_mutex which we already hold.
Fixes: 0650bf52b3 ("net: dsa: be compatible with masters which unregister on shutdown")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20220318195443.275026-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The device_node pointer is returned by of_parse_phandle() with refcount
incremented. We should use of_node_put() on it when done.
Fixes: 6d4e5c570c ("net: dsa: get port type at parse time")
Signed-off-by: Miaoqian Lin <linmq006@gmail.com>
Link: https://lore.kernel.org/r/20220316082602.10785-1-linmq006@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
&cpu_db->fdbs and &cpu_db->mdbs may be uninitialized lists during some
call paths of felix_set_tag_protocol().
There was an attempt to avoid calling dsa_port_walk_fdbs() during setup
by using a "bool change" in the felix driver, but this doesn't work when
the tagging protocol is defined in the device tree, and a change is
triggered by DSA at pseudo-runtime:
dsa_tree_setup_switches
-> dsa_switch_setup
-> dsa_switch_setup_tag_protocol
-> ds->ops->change_tag_protocol
dsa_tree_setup_ports
-> dsa_port_setup
-> &dp->fdbs and &db->mdbs only get initialized here
So it seems like the only way to fix this is to move the initialization
of these lists earlier.
dsa_port_touch() is called from dsa_switch_touch_ports() which is called
from dsa_switch_parse_of(), and this runs completely before
dsa_tree_setup(). Similarly, dsa_switch_release_ports() runs after
dsa_tree_teardown().
Fixes: f9cef64fa2 ("net: dsa: felix: migrate host FDB and MDB entries when changing tag proto")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There has been recent work towards matching each switchdev object
addition with a corresponding deletion.
Therefore, having elements in the fdbs, mdbs, vlans lists at the time of
a shared (DSA, CPU) port's teardown is indicative of a bug somewhere
else, and not something that is to be expected.
We shouldn't try to silently paper over that. Instead, print a warning
and a stack trace.
This change is a prerequisite for moving the initialization/teardown of
these lists. Make it clear that clearing the lists isn't needed.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After the blamed commit, dsa_tree_setup_master() may exit without
calling rtnl_unlock(), fix that.
Fixes: c146f9bc19 ("net: dsa: hold rtnl_mutex when calling dsa_master_{setup,teardown}")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The blamed commit said one thing but did another. It explains that we
should restore the "return err" to the original "goto out_unwind_tagger",
but instead it replaced it with "goto out_unlock".
When DSA_NOTIFIER_TAG_PROTO fails after the first switch of a
multi-switch tree, the switches would end up not using the same tagging
protocol.
Fixes: 0b0e2ff103 ("net: dsa: restore error path of dsa_tree_change_tag_proto")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20220303154249.1854436-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the DSA_NOTIFIER_TAG_PROTO returns an error, the user space process
which initiated the protocol change exits the kernel processing while
still holding the rtnl_mutex. So any other process attempting to lock
the rtnl_mutex would deadlock after such event.
The error handling of DSA_NOTIFIER_TAG_PROTO was inadvertently changed
by the blamed commit, introducing this regression. We must still call
rtnl_unlock(), and we must still call DSA_NOTIFIER_TAG_PROTO for the old
protocol. The latter is due to the limiting design of notifier chains
for cross-chip operations, which don't have a built-in error recovery
mechanism - we should look into using notifier_call_chain_robust for that.
Fixes: dc452a471d ("net: dsa: introduce tagger-owned storage for private and shared data")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20220228141715.146485-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The main purpose of this change is to create a data structure for a LAG
as seen by DSA. This is similar to what we have for bridging - we pass a
copy of this structure by value to ->port_lag_join and ->port_lag_leave.
For now we keep the lag_dev, id and a reference count in it. Future
patches will add a list of FDB entries for the LAG (these also need to
be refcounted to work properly).
The LAG structure is created using dsa_port_lag_create() and destroyed
using dsa_port_lag_destroy(), just like we have for bridging.
Because now, the dsa_lag itself is refcounted, we can simplify
dsa_lag_map() and dsa_lag_unmap(). These functions need to keep a LAG in
the dst->lags array only as long as at least one port uses it. The
refcounting logic inside those functions can be removed now - they are
called only when we should perform the operation.
dsa_lag_dev() is renamed to dsa_lag_by_id() and now returns the dsa_lag
structure instead of the lag_dev net_device.
dsa_lag_foreach_port() now takes the dsa_lag structure as argument.
dst->lags holds an array of dsa_lag structures.
dsa_lag_map() now also saves the dsa_lag->id value, so that linear
walking of dst->lags in drivers using dsa_lag_id() is no longer
necessary. They can just look at lag.id.
dsa_port_lag_id_get() is a helper, similar to dsa_port_bridge_num_get(),
which can be used by drivers to get the LAG ID assigned by DSA to a
given port.
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 DSA LAG API will be changed to become more similar with the bridge
data structures, where struct dsa_bridge holds an unsigned int num,
which is generated by DSA and is one-based. We have a similar thing
going with the DSA LAG, except that isn't stored anywhere, it is
calculated dynamically by dsa_lag_id() by iterating through dst->lags.
The idea of encoding an invalid (or not requested) LAG ID as zero for
the purpose of simplifying checks in drivers means that the LAG IDs
passed by DSA to drivers need to be one-based too. So back-and-forth
conversion is needed when indexing the dst->lags array, as well as in
drivers which assume a zero-based index.
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 preparation of converting struct net_device *dp->lag_dev into a
struct dsa_lag *dp->lag, we need to rename, for consistency purposes,
all occurrences of the "lag" variable in the DSA core to "lag_dev".
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>
DSA now explicitly handles VLANs installed with the 'self' flag on the
bridge as host VLANs, instead of just replicating every bridge port VLAN
also on the CPU port and never deleting it, which is what it did before.
However, this leaves a corner case uncovered, as explained by
Tobias Waldekranz:
https://patchwork.kernel.org/project/netdevbpf/patch/20220209213044.2353153-6-vladimir.oltean@nxp.com/#24735260
Forwarding towards a bridge port VLAN installed on a bridge port foreign
to DSA (separate NIC, Wi-Fi AP) used to work by virtue of the fact that
DSA itself needed to have at least one port in that VLAN (therefore, it
also had the CPU port in said VLAN). However, now that the CPU port may
not be member of all VLANs that user ports are members of, we need to
ensure this isn't the case if software forwarding to a foreign interface
is required.
The solution is to treat bridge port VLANs on standalone interfaces in
the exact same way as host VLANs. From DSA's perspective, there is no
difference between local termination and software forwarding; packets in
that VLAN must reach the CPU in both cases.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, DSA programs VLANs on shared (DSA and CPU) ports each time it
does so on user ports. This is good for basic functionality but has
several limitations:
- the VLAN group which must reach the CPU may be radically different
from the VLAN group that must be autonomously forwarded by the switch.
In other words, the admin may want to isolate noisy stations and avoid
traffic from them going to the control processor of the switch, where
it would just waste useless cycles. The bridge already supports
independent control of VLAN groups on bridge ports and on the bridge
itself, and when VLAN-aware, it will drop packets in software anyway
if their VID isn't added as a 'self' entry towards the bridge device.
- Replaying host FDB entries may depend, for some drivers like mv88e6xxx,
on replaying the host VLANs as well. The 2 VLAN groups are
approximately the same in most regular cases, but there are corner
cases when timing matters, and DSA's approximation of replicating
VLANs on shared ports simply does not work.
- If a user makes the bridge (implicitly the CPU port) join a VLAN by
accident, there is no way for the CPU port to isolate itself from that
noisy VLAN except by rebooting the system. This is because for each
VLAN added on a user port, DSA will add it on shared ports too, but
for each VLAN deletion on a user port, it will remain installed on
shared ports, since DSA has no good indication of whether the VLAN is
still in use or not.
Now that the bridge driver emits well-balanced SWITCHDEV_OBJ_ID_PORT_VLAN
addition and removal events, DSA has a simple and straightforward task
of separating the bridge port VLANs (these have an orig_dev which is a
DSA slave interface, or a LAG interface) from the host VLANs (these have
an orig_dev which is a bridge interface), and to keep a simple reference
count of each VID on each shared port.
Forwarding VLANs must be installed on the bridge ports and on all DSA
ports interconnecting them. We don't have a good view of the exact
topology, so we simply install forwarding VLANs on all DSA ports, which
is what has been done until now.
Host VLANs must be installed primarily on the dedicated CPU port of each
bridge port. More subtly, they must also be installed on upstream-facing
and downstream-facing DSA ports that are connecting the bridge ports and
the CPU. This ensures that the mv88e6xxx's problem (VID of host FDB
entry may be absent from VTU) is still addressed even if that switch is
in a cross-chip setup, and it has no local CPU port.
Therefore:
- user ports contain only bridge port (forwarding) VLANs, and no
refcounting is necessary
- DSA ports contain both forwarding and host VLANs. Refcounting is
necessary among these 2 types.
- CPU ports contain only host VLANs. Refcounting is also necessary.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rafael reports that on a system with LX2160A and Marvell DSA switches,
if a reboot occurs while the DSA master (dpaa2-eth) is up, the following
panic can be seen:
systemd-shutdown[1]: Rebooting.
Unable to handle kernel paging request at virtual address 00a0000800000041
[00a0000800000041] address between user and kernel address ranges
Internal error: Oops: 96000004 [#1] PREEMPT SMP
CPU: 6 PID: 1 Comm: systemd-shutdow Not tainted 5.16.5-00042-g8f5585009b24 #32
pc : dsa_slave_netdevice_event+0x130/0x3e4
lr : raw_notifier_call_chain+0x50/0x6c
Call trace:
dsa_slave_netdevice_event+0x130/0x3e4
raw_notifier_call_chain+0x50/0x6c
call_netdevice_notifiers_info+0x54/0xa0
__dev_close_many+0x50/0x130
dev_close_many+0x84/0x120
unregister_netdevice_many+0x130/0x710
unregister_netdevice_queue+0x8c/0xd0
unregister_netdev+0x20/0x30
dpaa2_eth_remove+0x68/0x190
fsl_mc_driver_remove+0x20/0x5c
__device_release_driver+0x21c/0x220
device_release_driver_internal+0xac/0xb0
device_links_unbind_consumers+0xd4/0x100
__device_release_driver+0x94/0x220
device_release_driver+0x28/0x40
bus_remove_device+0x118/0x124
device_del+0x174/0x420
fsl_mc_device_remove+0x24/0x40
__fsl_mc_device_remove+0xc/0x20
device_for_each_child+0x58/0xa0
dprc_remove+0x90/0xb0
fsl_mc_driver_remove+0x20/0x5c
__device_release_driver+0x21c/0x220
device_release_driver+0x28/0x40
bus_remove_device+0x118/0x124
device_del+0x174/0x420
fsl_mc_bus_remove+0x80/0x100
fsl_mc_bus_shutdown+0xc/0x1c
platform_shutdown+0x20/0x30
device_shutdown+0x154/0x330
__do_sys_reboot+0x1cc/0x250
__arm64_sys_reboot+0x20/0x30
invoke_syscall.constprop.0+0x4c/0xe0
do_el0_svc+0x4c/0x150
el0_svc+0x24/0xb0
el0t_64_sync_handler+0xa8/0xb0
el0t_64_sync+0x178/0x17c
It can be seen from the stack trace that the problem is that the
deregistration of the master causes a dev_close(), which gets notified
as NETDEV_GOING_DOWN to dsa_slave_netdevice_event().
But dsa_switch_shutdown() has already run, and this has unregistered the
DSA slave interfaces, and yet, the NETDEV_GOING_DOWN handler attempts to
call dev_close_many() on those slave interfaces, leading to the problem.
The previous attempt to avoid the NETDEV_GOING_DOWN on the master after
dsa_switch_shutdown() was called seems improper. Unregistering the slave
interfaces is unnecessary and unhelpful. Instead, after the slaves have
stopped being uppers of the DSA master, we can now reset to NULL the
master->dsa_ptr pointer, which will make DSA start ignoring all future
notifier events on the master.
Fixes: 0650bf52b3 ("net: dsa: be compatible with masters which unregister on shutdown")
Reported-by: Rafael Richter <rafael.richter@gin.de>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order for switch driver to be able to make simple and reliable use of
the master tracking operations, they must also be notified of the
initial state of the DSA master, not just of the changes. This is
because they might enable certain features only during the time when
they know that the DSA master is up and running.
Therefore, this change explicitly checks the state of the DSA master
under the same rtnl_mutex as we were holding during the
dsa_master_setup() and dsa_master_teardown() call. The idea being that
if the DSA master became operational in between the moment in which it
became a DSA master (dsa_master_setup set dev->dsa_ptr) and the moment
when we checked for the master being up, there is a chance that we
would emit a ->master_state_change() call with no actual state change.
We need to avoid that by serializing the concurrent netdevice event with
us. If the netdevice event started before, we force it to finish before
we begin, because we take rtnl_lock before making netdev_uses_dsa()
return true. So we also handle that early event and do nothing on it.
Similarly, if the dev_open() attempt is concurrent with us, it will
attempt to take the rtnl_mutex, but we're holding it. We'll see that
the master flag IFF_UP isn't set, then when we release the rtnl_mutex
we'll process the NETDEV_UP notifier.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Certain drivers may need to send management traffic to the switch for
things like register access, FDB dump, etc, to accelerate what their
slow bus (SPI, I2C, MDIO) can already do.
Ethernet is faster (especially in bulk transactions) but is also more
unreliable, since the user may decide to bring the DSA master down (or
not bring it up), therefore severing the link between the host and the
attached switch.
Drivers needing Ethernet-based register access already should have
fallback logic to the slow bus if the Ethernet method fails, but that
fallback may be based on a timeout, and the I/O to the switch may slow
down to a halt if the master is down, because every Ethernet packet will
have to time out. The driver also doesn't have the option to turn off
Ethernet-based I/O momentarily, because it wouldn't know when to turn it
back on.
Which is where this change comes in. By tracking NETDEV_CHANGE,
NETDEV_UP and NETDEV_GOING_DOWN events on the DSA master, we should know
the exact interval of time during which this interface is reliably
available for traffic. Provide this information to switches so they can
use it as they wish.
An helper is added dsa_port_master_is_operational() to check if a master
port is operational.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is said that as soon as a network interface is registered, all its
resources should have already been prepared, so that it is available for
sending and receiving traffic. One of the resources needed by a DSA
slave interface is the master.
dsa_tree_setup
-> dsa_tree_setup_ports
-> dsa_port_setup
-> dsa_slave_create
-> register_netdevice
-> dsa_tree_setup_master
-> dsa_master_setup
-> sets up master->dsa_ptr, which enables reception
Therefore, there is a short period of time after register_netdevice()
during which the master isn't prepared to pass traffic to the DSA layer
(master->dsa_ptr is checked by eth_type_trans). Same thing during
unregistration, there is a time frame in which packets might be missed.
Note that this change opens us to another race: dsa_master_find_slave()
will get invoked potentially earlier than the slave creation, and later
than the slave deletion. Since dp->slave starts off as a NULL pointer,
the earlier calls aren't a problem, but the later calls are. To avoid
use-after-free, we should zeroize dp->slave before calling
dsa_slave_destroy().
In practice I cannot really test real life improvements brought by this
change, since in my systems, netdevice creation races with PHY autoneg
which takes a few seconds to complete, and that masks quite a few races.
Effects might be noticeable in a setup with fixed links all the way to
an external system.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After commit a57d8c217a ("net: dsa: flush switchdev workqueue before
tearing down CPU/DSA ports"), the port setup and teardown procedure
became asymmetric.
The fact of the matter is that user ports need the shared ports to be up
before they can be used for CPU-initiated termination. And since we
register net devices for the user ports, those won't be functional until
we also call the setup for the shared (CPU, DSA) ports. But we may do
that later, depending on the port numbering scheme of the hardware we
are dealing with.
It just makes sense that all shared ports are brought up before any user
port is. I can't pinpoint any issue due to the current behavior, but
let's change it nonetheless, for consistency's sake.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
DSA needs to simulate master tracking events when a binding is first
with a DSA master established and torn down, in order to give drivers
the simplifying guarantee that ->master_state_change calls are made
only when the master's readiness state to pass traffic changes.
master_state_change() provide a operational bool that DSA driver can use
to understand if DSA master is operational or not.
To avoid races, we need to block the reception of
NETDEV_UP/NETDEV_CHANGE/NETDEV_GOING_DOWN events in the netdev notifier
chain while we are changing the master's dev->dsa_ptr (this changes what
netdev_uses_dsa(dev) reports).
The dsa_master_setup() and dsa_master_teardown() functions optionally
require the rtnl_mutex to be held, if the tagger needs the master to be
promiscuous, these functions call dev_set_promiscuity(). Move the
rtnl_lock() from that function and make it top-level.
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>
On the NXP Bluebox 3 board which uses a multi-switch setup with sja1105,
the mechanism through which the tagger connects to the switch tree is
broken, due to improper DSA code design. At the time when tag_ops->connect()
is called in dsa_port_parse_cpu(), DSA hasn't finished "touching" all
the ports, so it doesn't know how large the tree is and how many ports
it has. It has just seen the first CPU port by this time. As a result,
this function will call the tagger's ->connect method too early, and the
tagger will connect only to the first switch from the tree.
This could be perhaps addressed a bit more simply by just moving the
tag_ops->connect(dst) call a bit later (for example in dsa_tree_setup),
but there is already a design inconsistency at present: on the switch
side, the notification is on a per-switch basis, but on the tagger side,
it is on a per-tree basis. Furthermore, the persistent storage itself is
per switch (ds->tagger_data). And the tagger connect and disconnect
procedures (at least the ones that exist currently) could see a fair bit
of simplification if they didn't have to iterate through the switches of
a tree.
To fix the issue, this change transforms tag_ops->connect(dst) into
tag_ops->connect(ds) and moves it somewhere where we already iterate
over all switches of a tree. That is in dsa_switch_setup_tag_protocol(),
which is a good placement because we already have there the connection
call to the switch side of things.
As for the dsa_tree_bind_tag_proto() method (called from the code path
that changes the tag protocol), things are a bit more complicated
because we receive the tree as argument, yet when we unwind on errors,
it would be nice to not call tag_ops->disconnect(ds) where we didn't
previously call tag_ops->connect(ds). We didn't have this problem before
because the tag_ops connection operations passed the entire dst before,
and this is more fine grained now. To solve the error rewind case using
the new API, we have to create yet one more cross-chip notifier for
disconnection, and stay connected with the old tag protocol to all the
switches in the tree until we've succeeded to connect with the new one
as well. So if something fails half way, the whole tree is still
connected to the old tagger. But there may still be leaks if the tagger
fails to connect to the 2nd out of 3 switches in a tree: somebody needs
to tell the tagger to disconnect from the first switch. Nothing comes
for free, and this was previously handled privately by the tagging
protocol driver before, but now we need to emit a disconnect cross-chip
notifier for that, because DSA has to take care of the unwind path. We
assume that the tagging protocol has connected to a switch if it has set
ds->tagger_data to something, otherwise we avoid calling its
disconnection method in the error rewind path.
The rest of the changes are in the tagging protocol drivers, and have to
do with the replacement of dst with ds. The iteration is removed and the
error unwind path is simplified, as mentioned above.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ansuel is working on register access over Ethernet for the qca8k switch
family. This requires the qca8k tagging protocol driver to receive
frames which aren't intended for the network stack, but instead for the
qca8k switch driver itself.
The dp->priv is currently the prevailing method for passing data back
and forth between the tagging protocol driver and the switch driver.
However, this method is riddled with caveats.
The DSA design allows in principle for any switch driver to return any
protocol it desires in ->get_tag_protocol(). The dsa_loop driver can be
modified to do just that. But in the current design, the memory behind
dp->priv has to be allocated by the switch driver, so if the tagging
protocol is paired to an unexpected switch driver, we may end up in NULL
pointer dereferences inside the kernel, or worse (a switch driver may
allocate dp->priv according to the expectations of a different tagger).
The latter possibility is even more plausible considering that DSA
switches can dynamically change tagging protocols in certain cases
(dsa <-> edsa, ocelot <-> ocelot-8021q), and the current design lends
itself to mistakes that are all too easy to make.
This patch proposes that the tagging protocol driver should manage its
own memory, instead of relying on the switch driver to do so.
After analyzing the different in-tree needs, it can be observed that the
required tagger storage is per switch, therefore a ds->tagger_data
pointer is introduced. In principle, per-port storage could also be
introduced, although there is no need for it at the moment. Future
changes will replace the current usage of dp->priv with ds->tagger_data.
We define a "binding" event between the DSA switch tree and the tagging
protocol. During this binding event, the tagging protocol's ->connect()
method is called first, and this may allocate some memory for each
switch of the tree. Then a cross-chip notifier is emitted for the
switches within that tree, and they are given the opportunity to fix up
the tagger's memory (for example, they might set up some function
pointers that represent virtual methods for consuming packets).
Because the memory is owned by the tagger, there exists a ->disconnect()
method for the tagger (which is the place to free the resources), but
there doesn't exist a ->disconnect() method for the switch driver.
This is part of the design. The switch driver should make minimal use of
the public part of the tagger data, and only after type-checking it
using the supplied "proto" argument.
In the code there are in fact two binding events, one is the initial
event in dsa_switch_setup_tag_protocol(). At this stage, the cross chip
notifier chains aren't initialized, so we call each switch's connect()
method by hand. Then there is dsa_tree_bind_tag_proto() during
dsa_tree_change_tag_proto(), and here we have an old protocol and a new
one. We first connect to the new one before disconnecting from the old
one, to simplify error handling a bit and to ensure we remain in a valid
state at all times.
Co-developed-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The main desire behind this is to provide coherent bridge information to
the fast path without locking.
For example, right now we set dp->bridge_dev and dp->bridge_num from
separate code paths, it is theoretically possible for a packet
transmission to read these two port properties consecutively and find a
bridge number which does not correspond with the bridge device.
Another desire is to start passing more complex bridge information to
dsa_switch_ops functions. For example, with FDB isolation, it is
expected that drivers will need to be passed the bridge which requested
an FDB/MDB entry to be offloaded, and along with that bridge_dev, the
associated bridge_num should be passed too, in case the driver might
want to implement an isolation scheme based on that number.
We already pass the {bridge_dev, bridge_num} pair to the TX forwarding
offload switch API, however we'd like to remove that and squash it into
the basic bridge join/leave API. So that means we need to pass this
pair to the bridge join/leave API.
During dsa_port_bridge_leave, first we unset dp->bridge_dev, then we
call the driver's .port_bridge_leave with what used to be our
dp->bridge_dev, but provided as an argument.
When bridge_dev and bridge_num get folded into a single structure, we
need to preserve this behavior in dsa_port_bridge_leave: we need a copy
of what used to be in dp->bridge.
Switch drivers check bridge membership by comparing dp->bridge_dev with
the provided bridge_dev, but now, if we provide the struct dsa_bridge as
a pointer, they cannot keep comparing dp->bridge to the provided
pointer, since this only points to an on-stack copy. To make this
obvious and prevent driver writers from forgetting and doing stupid
things, in this new API, the struct dsa_bridge is provided as a full
structure (not very large, contains an int and a pointer) instead of a
pointer. An explicit comparison function needs to be used to determine
bridge membership: dsa_port_offloads_bridge().
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The service where DSA assigns a unique bridge number for each forwarding
domain is useful even for drivers which do not implement the TX
forwarding offload feature.
For example, drivers might use the dp->bridge_num for FDB isolation.
So rename ds->num_fwd_offloading_bridges to ds->max_num_bridges, and
calculate a unique bridge_num for all drivers that set this value.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
I have seen too many bugs already due to the fact that we must encode an
invalid dp->bridge_num as a negative value, because the natural tendency
is to check that invalid value using (!dp->bridge_num). Latest example
can be seen in commit 1bec0f0506 ("net: dsa: fix bridge_num not
getting cleared after ports leaving the bridge").
Convert the existing users to assume that dp->bridge_num == 0 is the
encoding for invalid, and valid bridge numbers start from 1.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Now that the rtnl_mutex is going away for dsa_port_{host_,}fdb_{add,del},
no one is serializing access to the address lists that DSA keeps for the
purpose of reference counting on shared ports (CPU and cascade ports).
It can happen for one dsa_switch_do_fdb_del to do list_del on a dp->fdbs
element while another dsa_switch_do_fdb_{add,del} is traversing dp->fdbs.
We need to avoid that.
Currently dp->mdbs is not at risk, because dsa_switch_do_mdb_{add,del}
still runs under the rtnl_mutex. But it would be nice if it would not
depend on that being the case. So let's introduce a mutex per port (the
address lists are per port too) and share it between dp->mdbs and
dp->fdbs.
The place where we put the locking is interesting. It could be tempting
to put a DSA-level lock which still serializes calls to
.port_fdb_{add,del}, but it would still not avoid concurrency with other
driver code paths that are currently under rtnl_mutex (.port_fdb_dump,
.port_fast_age). So it would add a very false sense of security (and
adding a global switch-wide lock in DSA to resynchronize with the
rtnl_lock is also counterproductive and hard).
So the locking is intentionally done only where the dp->fdbs and dp->mdbs
lists are traversed. That means, from a driver perspective, that
.port_fdb_add will be called with the dp->addr_lists_lock mutex held on
the CPU port, but not held on user ports. This is done so that driver
writers are not encouraged to rely on any guarantee offered by
dp->addr_lists_lock.
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>
Now that the rtnl_mutex is going away for dsa_port_{host_,}fdb_{add,del},
no one is serializing access to the address lists that DSA keeps for the
purpose of reference counting on shared ports (CPU and cascade ports).
It can happen for one dsa_switch_do_fdb_del to do list_del on a dp->fdbs
element while another dsa_switch_do_fdb_{add,del} is traversing dp->fdbs.
We need to avoid that.
Currently dp->mdbs is not at risk, because dsa_switch_do_mdb_{add,del}
still runs under the rtnl_mutex. But it would be nice if it would not
depend on that being the case. So let's introduce a mutex per port (the
address lists are per port too) and share it between dp->mdbs and
dp->fdbs.
The place where we put the locking is interesting. It could be tempting
to put a DSA-level lock which still serializes calls to
.port_fdb_{add,del}, but it would still not avoid concurrency with other
driver code paths that are currently under rtnl_mutex (.port_fdb_dump,
.port_fast_age). So it would add a very false sense of security (and
adding a global switch-wide lock in DSA to resynchronize with the
rtnl_lock is also counterproductive and hard).
So the locking is intentionally done only where the dp->fdbs and dp->mdbs
lists are traversed. That means, from a driver perspective, that
.port_fdb_add will be called with the dp->addr_lists_lock mutex held on
the CPU port, but not held on user ports. This is done so that driver
writers are not encouraged to rely on any guarantee offered by
dp->addr_lists_lock.
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>
Find the remaining iterators over dst->ports that only filter for the
ports belonging to a certain switch, and replace those with the
dsa_switch_for_each_port helper that we have now.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ever since Vivien's conversion of the ds->ports array into a dst->ports
list, and the introduction of dsa_to_port, iterations through the ports
of a switch became quadratic whenever dsa_to_port was needed.
dsa_to_port can either be called directly, or indirectly through the
dsa_is_{user,cpu,dsa,unused}_port helpers.
Use the newly introduced dsa_switch_for_each_port() iteration macro
that works with the iterator variable being a struct dsa_port *dp
directly, and not an int i. It is an expensive variable to go from i to
dp, but cheap to go from dp to i.
This macro iterates through the entire ds->dst->ports list and filters
by the ports belonging just to the switch provided as argument.
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>
tools/testing/selftests/net/ioam6.sh
7b1700e009 ("selftests: net: modify IOAM tests for undef bits")
bf77b1400a ("selftests: net: Test for the IOAM encapsulation with IPv6")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
To be symmetric with the error unwind path of dsa_switch_setup(), call
dsa_switch_unregister_notifier() after ds->ops->teardown.
The implication is that ds->ops->teardown cannot emit cross-chip
notifiers. For example, currently the dsa_tag_8021q_unregister() call
from sja1105_teardown() does not propagate to the entire tree due to
this reason. However I cannot find an actual issue caused by this,
observed using code inspection.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20211012123735.2545742-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It was a documented fact that ds->ops->change_tag_protocol() offered
rtnetlink mutex protection to the switch driver, since there was an
ASSERT_RTNL right before the call in dsa_switch_change_tag_proto()
(initiated from sysfs).
The blamed commit introduced another call path for
ds->ops->change_tag_protocol() which does not hold the rtnl_mutex.
This is:
dsa_tree_setup
-> dsa_tree_setup_switches
-> dsa_switch_setup
-> dsa_switch_setup_tag_protocol
-> ds->ops->change_tag_protocol()
-> dsa_port_setup
-> dsa_slave_create
-> register_netdevice(slave_dev)
-> dsa_tree_setup_master
-> dsa_master_setup
-> dev->dsa_ptr = cpu_dp
The reason why the rtnl_mutex is held in the sysfs call path is to
ensure that, once the master and all the DSA interfaces are down (which
is required so that no packets flow), they remain down during the
tagging protocol change.
The above calling order illustrates the fact that it should not be risky
to change the initial tagging protocol to the one specified in the
device tree at the given time:
- packets cannot enter the dsa_switch_rcv() packet type handler since
netdev_uses_dsa() for the master will not yet return true, since
dev->dsa_ptr has not yet been populated
- packets cannot enter the dsa_slave_xmit() function because no DSA
interface has yet been registered
So from the DSA core's perspective, holding the rtnl_mutex is indeed not
necessary.
Yet, drivers may need to do things which need rtnl_mutex protection. For
example:
felix_set_tag_protocol
-> felix_setup_tag_8021q
-> dsa_tag_8021q_register
-> dsa_tag_8021q_setup
-> dsa_tag_8021q_port_setup
-> vlan_vid_add
-> ASSERT_RTNL
These drivers do not really have a choice to take the rtnl_mutex
themselves, since in the sysfs case, the rtnl_mutex is already held.
Fixes: deff710703 ("net: dsa: Allow default tag protocol to be overridden from DT")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The dp->bridge_num is zero-based, with -1 being the encoding for an
invalid value. But dsa_bridge_num_put used to check for an invalid value
by comparing bridge_num with 0, which is of course incorrect.
The result is that the bridge_num will never get cleared by
dsa_bridge_num_put, and further port joins to other bridges will get a
bridge_num larger than the previous one, and once all the available
bridges with TX forwarding offload supported by the hardware get
exhausted, the TX forwarding offload feature is simply disabled.
In the case of sja1105, 7 iterations of the loop below are enough to
exhaust the TX forwarding offload bits, and further bridge joins operate
without that feature.
ip link add br0 type bridge vlan_filtering 1
while :; do
ip link set sw0p2 master br0 && sleep 1
ip link set sw0p2 nomaster && sleep 1
done
This issue is enough of an indication that having the dp->bridge_num
invalid encoding be a negative number is prone to bugs, so this will be
changed to a one-based value, with the dp->bridge_num of zero being the
indication of no bridge. However, that is material for net-next.
Fixes: f5e165e72b ("net: dsa: track unique bridge numbers across all DSA switch trees")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This change prevents from users to access device before devlink
is fully configured.
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/mptcp/protocol.c
977d293e23 ("mptcp: ensure tx skbs always have the MPTCP ext")
efe686ffce ("mptcp: ensure tx skbs always have the MPTCP ext")
same patch merged in both trees, keep net-next.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
devlink_register() can't fail and always returns success, but all drivers
are obligated to check returned status anyway. This adds a lot of boilerplate
code to handle impossible flow.
Make devlink_register() void and simplify the drivers that use that
API call.
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Acked-by: Simon Horman <simon.horman@corigine.com>
Acked-by: Vladimir Oltean <olteanv@gmail.com> # dsa
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Linux device model permits both the ->shutdown and ->remove driver
methods to get called during a shutdown procedure. Example: a DSA switch
which sits on an SPI bus, and the SPI bus driver calls this on its
->shutdown method:
spi_unregister_controller
-> device_for_each_child(&ctlr->dev, NULL, __unregister);
-> spi_unregister_device(to_spi_device(dev));
-> device_del(&spi->dev);
So this is a simple pattern which can theoretically appear on any bus,
although the only other buses on which I've been able to find it are
I2C:
i2c_del_adapter
-> device_for_each_child(&adap->dev, NULL, __unregister_client);
-> i2c_unregister_device(client);
-> device_unregister(&client->dev);
The implication of this pattern is that devices on these buses can be
unregistered after having been shut down. The drivers for these devices
might choose to return early either from ->remove or ->shutdown if the
other callback has already run once, and they might choose that the
->shutdown method should only perform a subset of the teardown done by
->remove (to avoid unnecessary delays when rebooting).
So in other words, the device driver may choose on ->remove to not
do anything (therefore to not unregister an MDIO bus it has registered
on ->probe), because this ->remove is actually triggered by the
device_shutdown path, and its ->shutdown method has already run and done
the minimally required cleanup.
This used to be fine until the blamed commit, but now, the following
BUG_ON triggers:
void mdiobus_free(struct mii_bus *bus)
{
/* For compatibility with error handling in drivers. */
if (bus->state == MDIOBUS_ALLOCATED) {
kfree(bus);
return;
}
BUG_ON(bus->state != MDIOBUS_UNREGISTERED);
bus->state = MDIOBUS_RELEASED;
put_device(&bus->dev);
}
In other words, there is an attempt to free an MDIO bus which was not
unregistered. The attempt to free it comes from the devres release
callbacks of the SPI device, which are executed after the device is
unregistered.
I'm not saying that the fact that MDIO buses allocated using devres
would automatically get unregistered wasn't strange. I'm just saying
that the commit didn't care about auditing existing call paths in the
kernel, and now, the following code sequences are potentially buggy:
(a) devm_mdiobus_alloc followed by plain mdiobus_register, for a device
located on a bus that unregisters its children on shutdown. After
the blamed patch, either both the alloc and the register should use
devres, or none should.
(b) devm_mdiobus_alloc followed by plain mdiobus_register, and then no
mdiobus_unregister at all in the remove path. After the blamed
patch, nobody unregisters the MDIO bus anymore, so this is even more
buggy than the previous case which needs a specific bus
configuration to be seen, this one is an unconditional bug.
In this case, DSA falls into category (a), it tries to be helpful and
registers an MDIO bus on behalf of the switch, which might be on such a
bus. I've no idea why it does it under devres.
It does this on probe:
if (!ds->slave_mii_bus && ds->ops->phy_read)
alloc and register mdio bus
and this on remove:
if (ds->slave_mii_bus && ds->ops->phy_read)
unregister mdio bus
I _could_ imagine using devres because the condition used on remove is
different than the condition used on probe. So strictly speaking, DSA
cannot determine whether the ds->slave_mii_bus it sees on remove is the
ds->slave_mii_bus that _it_ has allocated on probe. Using devres would
have solved that problem. But nonetheless, the existing code already
proceeds to unregister the MDIO bus, even though it might be
unregistering an MDIO bus it has never registered. So I can only guess
that no driver that implements ds->ops->phy_read also allocates and
registers ds->slave_mii_bus itself.
So in that case, if unregistering is fine, freeing must be fine too.
Stop using devres and free the MDIO bus manually. This will make devres
stop attempting to free a still registered MDIO bus on ->shutdown.
Fixes: ac3a68d566 ("net: phy: don't abuse devres in devm_mdiobus_register()")
Reported-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the blamed commit, dsa_tree_teardown_switches() was split into two
smaller functions, dsa_tree_teardown_switches and dsa_tree_teardown_ports.
However, the error path of dsa_tree_setup stopped calling dsa_tree_teardown_ports.
Fixes: a57d8c217a ("net: dsa: flush switchdev workqueue before tearing down CPU/DSA ports")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 86f8b1c01a ("net: dsa: Do not make user port errors fatal")
decided it was fine to ignore errors on certain ports that fail to
probe, and go on with the ports that do probe fine.
Commit fb6ec87f72 ("net: dsa: Fix type was not set for devlink port")
noticed that devlink_port_type_eth_set(dlp, dp->slave); does not get
called, and devlink notices after a timeout of 3600 seconds and prints a
WARN_ON. So it went ahead to unregister the devlink port. And because
there exists an UNUSED port flavour, we actually re-register the devlink
port as UNUSED.
Commit 08156ba430 ("net: dsa: Add devlink port regions support to
DSA") added devlink port regions, which are set up by the driver and not
by DSA.
When we trigger the devlink port deregistration and reregistration as
unused, devlink now prints another WARN_ON, from here:
devlink_port_unregister:
WARN_ON(!list_empty(&devlink_port->region_list));
So the port still has regions, which makes sense, because they were set
up by the driver, and the driver doesn't know we're unregistering the
devlink port.
Somebody needs to tear them down, and optionally (actually it would be
nice, to be consistent) set them up again for the new devlink port.
But DSA's layering stays in our way quite badly here.
The options I've considered are:
1. Introduce a function in devlink to just change a port's type and
flavour. No dice, devlink keeps a lot of state, it really wants the
port to not be registered when you set its parameters, so changing
anything can only be done by destroying what we currently have and
recreating it.
2. Make DSA cache the parameters passed to dsa_devlink_port_region_create,
and the region returned, keep those in a list, then when the devlink
port unregister needs to take place, the existing devlink regions are
destroyed by DSA, and we replay the creation of new regions using the
cached parameters. Problem: mv88e6xxx keeps the region pointers in
chip->ports[port].region, and these will remain stale after DSA frees
them. There are many things DSA can do, but updating mv88e6xxx's
private pointers is not one of them.
3. Just let the driver do it (i.e. introduce a very specific method
called ds->ops->port_reinit_as_unused, which unregisters its devlink
port devlink regions, then the old devlink port, then registers the
new one, then the devlink port regions for it). While it does work,
as opposed to the others, it's pretty horrible from an API
perspective and we can do better.
4. Introduce a new pair of methods, ->port_setup and ->port_teardown,
which in the case of mv88e6xxx must register and unregister the
devlink port regions. Call these 2 methods when the port must be
reinitialized as unused.
Naturally, I went for the 4th approach.
Fixes: 08156ba430 ("net: dsa: Add devlink port regions support to DSA")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Lino reports that on his system with bcmgenet as DSA master and KSZ9897
as a switch, rebooting or shutting down never works properly.
What does the bcmgenet driver have special to trigger this, that other
DSA masters do not? It has an implementation of ->shutdown which simply
calls its ->remove implementation. Otherwise said, it unregisters its
network interface on shutdown.
This message can be seen in a loop, and it hangs the reboot process there:
unregister_netdevice: waiting for eth0 to become free. Usage count = 3
So why 3?
A usage count of 1 is normal for a registered network interface, and any
virtual interface which links itself as an upper of that will increment
it via dev_hold. In the case of DSA, this is the call path:
dsa_slave_create
-> netdev_upper_dev_link
-> __netdev_upper_dev_link
-> __netdev_adjacent_dev_insert
-> dev_hold
So a DSA switch with 3 interfaces will result in a usage count elevated
by two, and netdev_wait_allrefs will wait until they have gone away.
Other stacked interfaces, like VLAN, watch NETDEV_UNREGISTER events and
delete themselves, but DSA cannot just vanish and go poof, at most it
can unbind itself from the switch devices, but that must happen strictly
earlier compared to when the DSA master unregisters its net_device, so
reacting on the NETDEV_UNREGISTER event is way too late.
It seems that it is a pretty established pattern to have a driver's
->shutdown hook redirect to its ->remove hook, so the same code is
executed regardless of whether the driver is unbound from the device, or
the system is just shutting down. As Florian puts it, it is quite a big
hammer for bcmgenet to unregister its net_device during shutdown, but
having a common code path with the driver unbind helps ensure it is well
tested.
So DSA, for better or for worse, has to live with that and engage in an
arms race of implementing the ->shutdown hook too, from all individual
drivers, and do something sane when paired with masters that unregister
their net_device there. The only sane thing to do, of course, is to
unlink from the master.
However, complications arise really quickly.
The pattern of redirecting ->shutdown to ->remove is not unique to
bcmgenet or even to net_device drivers. In fact, SPI controllers do it
too (see dspi_shutdown -> dspi_remove), and presumably, I2C controllers
and MDIO controllers do it too (this is something I have not researched
too deeply, but even if this is not the case today, it is certainly
plausible to happen in the future, and must be taken into consideration).
Since DSA switches might be SPI devices, I2C devices, MDIO devices, the
insane implication is that for the exact same DSA switch device, we
might have both ->shutdown and ->remove getting called.
So we need to do something with that insane environment. The pattern
I've come up with is "if this, then not that", so if either ->shutdown
or ->remove gets called, we set the device's drvdata to NULL, and in the
other hook, we check whether the drvdata is NULL and just do nothing.
This is probably not necessary for platform devices, just for devices on
buses, but I would really insist for consistency among drivers, because
when code is copy-pasted, it is not always copy-pasted from the best
sources.
So depending on whether the DSA switch's ->remove or ->shutdown will get
called first, we cannot really guarantee even for the same driver if
rebooting will result in the same code path on all platforms. But
nonetheless, we need to do something minimally reasonable on ->shutdown
too to fix the bug. Of course, the ->remove will do more (a full
teardown of the tree, with all data structures freed, and this is why
the bug was not caught for so long). The new ->shutdown method is kept
separate from dsa_unregister_switch not because we couldn't have
unregistered the switch, but simply in the interest of doing something
quick and to the point.
The big question is: does the DSA switch's ->shutdown get called earlier
than the DSA master's ->shutdown? If not, there is still a risk that we
might still trigger the WARN_ON in unregister_netdevice that says we are
attempting to unregister a net_device which has uppers. That's no good.
Although the reference to the master net_device won't physically go away
even if DSA's ->shutdown comes afterwards, remember we have a dev_hold
on it.
The answer to that question lies in this comment above device_link_add:
* A side effect of the link creation is re-ordering of dpm_list and the
* devices_kset list by moving the consumer device and all devices depending
* on it to the ends of these lists (that does not happen to devices that have
* not been registered when this function is called).
so the fact that DSA uses device_link_add towards its master is not
exactly for nothing. device_shutdown() walks devices_kset from the back,
so this is our guarantee that DSA's shutdown happens before the master's
shutdown.
Fixes: 2f1e8ea726 ("net: dsa: link interfaces with the DSA master to get rid of lockdep warnings")
Link: https://lore.kernel.org/netdev/20210909095324.12978-1-LinoSanfilippo@gmx.de/
Reported-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sometimes when unbinding the mv88e6xxx driver on Turris MOX, these error
messages appear:
mv88e6085 d0032004.mdio-mii:12: port 1 failed to delete be:79:b4:9e:9e:96 vid 1 from fdb: -2
mv88e6085 d0032004.mdio-mii:12: port 1 failed to delete be:79:b4:9e:9e:96 vid 0 from fdb: -2
mv88e6085 d0032004.mdio-mii:12: port 1 failed to delete d8:58:d7:00:ca:6d vid 100 from fdb: -2
mv88e6085 d0032004.mdio-mii:12: port 1 failed to delete d8:58:d7:00:ca:6d vid 1 from fdb: -2
mv88e6085 d0032004.mdio-mii:12: port 1 failed to delete d8:58:d7:00:ca:6d vid 0 from fdb: -2
(and similarly for other ports)
What happens is that DSA has a policy "even if there are bugs, let's at
least not leak memory" and dsa_port_teardown() clears the dp->fdbs and
dp->mdbs lists, which are supposed to be empty.
But deleting that cleanup code, the warnings go away.
=> the FDB and MDB lists (used for refcounting on shared ports, aka CPU
and DSA ports) will eventually be empty, but are not empty by the time
we tear down those ports. Aka we are deleting them too soon.
The addresses that DSA complains about are host-trapped addresses: the
local addresses of the ports, and the MAC address of the bridge device.
The problem is that offloading those entries happens from a deferred
work item scheduled by the SWITCHDEV_FDB_DEL_TO_DEVICE handler, and this
races with the teardown of the CPU and DSA ports where the refcounting
is kept.
In fact, not only it races, but fundamentally speaking, if we iterate
through the port list linearly, we might end up tearing down the shared
ports even before we delete a DSA user port which has a bridge upper.
So as it turns out, we need to first tear down the user ports (and the
unused ones, for no better place of doing that), then the shared ports
(the CPU and DSA ports). In between, we need to ensure that all work
items scheduled by our switchdev handlers (which only run for user
ports, hence the reason why we tear them down first) have finished.
Fixes: 161ca59d39 ("net: dsa: reference count the MDB entries at the cross-chip notifier level")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20210914134726.2305133-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Right now, cross-tree bridging setups work somewhat by mistake.
In the case of cross-tree bridging with sja1105, all switch instances
need to agree upon a common VLAN ID for forwarding a packet that belongs
to a certain bridging domain.
With TX forwarding offload, the VLAN ID is the bridge VLAN for
VLAN-aware bridging, and the tag_8021q TX forwarding offload VID
(a VLAN which has non-zero VBID bits) for VLAN-unaware bridging.
The VBID for VLAN-unaware bridging is derived from the dp->bridge_num
value calculated by DSA independently for each switch tree.
If ports from one tree join one bridge, and ports from another tree join
another bridge, DSA will assign them the same bridge_num, even though
the bridges are different. If cross-tree bridging is supported, this
is an issue.
Modify DSA to calculate the bridge_num globally across all switch trees.
This has the implication for a driver that the dp->bridge_num value that
DSA will assign to its ports might not be contiguous, if there are
boards with multiple DSA drivers instantiated. Additionally, all
bridge_num values eat up towards each switch's
ds->num_fwd_offloading_bridges maximum, which is potentially unfortunate,
and can be seen as a limitation introduced by this patch. However, that
is the lesser evil for now.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vladimir Oltean