Prior to the blamed commit, the bridge_num range was from
0 to ds->max_num_bridges - 1. After the commit, it is from
1 to ds->max_num_bridges.
So this check:
if (bridge_num >= max)
return 0;
must be updated to:
if (bridge_num > max)
return 0;
in order to allow the last bridge_num value (==max) to be used.
This is easiest visible when a driver sets ds->max_num_bridges=1.
The observed behaviour is that even the first created bridge triggers
the netlink extack "Range of offloadable bridges exceeded" warning, and
is handled in software rather than being offloaded.
Fixes: 3f9bb0301d ("net: dsa: make dp->bridge_num one-based")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20260120211039.3228999-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
of_find_net_device_by_node() searches net devices by their /sys/class/net/,
entry. It is documented in its kernel-doc that:
* If successful, returns a pointer to the net_device with the embedded
* struct device refcount incremented by one, or NULL on failure. The
* refcount must be dropped when done with the net_device.
We are missing a put_device(&conduit->dev) which we could place at the
end of dsa_tree_find_first_conduit(). But to explain why calling
put_device() right away is safe is the same as to explain why the chosen
solution is different.
The code is very poorly split: dsa_tree_find_first_conduit() was first
introduced in commit 95f510d0b7 ("net: dsa: allow the DSA master to be
seen and changed through rtnetlink") but was first used several commits
later, in commit acc43b7bf5 ("net: dsa: allow masters to join a LAG").
Assume there is a switch with 2 CPU ports and 2 conduits, eno2 and eno3.
When we create a LAG (bonding or team device) and place eno2 and eno3
beneath it, we create a 3rd conduit (the LAG device itself), but this is
slightly different than the first two.
Namely, the cpu_dp->conduit pointer of the CPU ports does not change,
and remains pointing towards the physical Ethernet controllers which are
now LAG ports. Only 2 things change:
- the LAG device has a dev->dsa_ptr which marks it as a DSA conduit
- dsa_port_to_conduit(user port) finds the LAG and not the physical
conduit, because of the dp->cpu_port_in_lag bit being set.
When the LAG device is destroyed, dsa_tree_migrate_ports_from_lag_conduit()
is called and this is where dsa_tree_find_first_conduit() kicks in.
This is the logical mistake and the reason why introducing code in one
patch and using it from another is bad practice. I didn't realize that I
don't have to call of_find_net_device_by_node() again; the cpu_dp->conduit
association was never undone, and is still available for direct (re)use.
There's only one concern - maybe the conduit disappeared in the
meantime, but the netdev_hold() call we made during dsa_port_parse_cpu()
(see previous change) ensures that this was not the case.
Therefore, fixing the code means reimplementing it in the simplest way.
I am blaming the time of use, since this is what "git blame" would show
if we were to monitor for the conduit's kobject's refcount remaining
elevated instead of being freed.
Tested on the NXP LS1028A, using the steps from
Documentation/networking/dsa/configuration.rst section "Affinity of user
ports to CPU ports", followed by (extra prints added by me):
$ ip link del bond0
mscc_felix 0000:00:00.5 swp3: Link is Down
bond0 (unregistering): (slave eno2): Releasing backup interface
fsl_enetc 0000:00:00.2 eno2: Link is Down
mscc_felix 0000:00:00.5 swp0: bond0 disappeared, migrating to eno2
mscc_felix 0000:00:00.5 swp1: bond0 disappeared, migrating to eno2
mscc_felix 0000:00:00.5 swp2: bond0 disappeared, migrating to eno2
mscc_felix 0000:00:00.5 swp3: bond0 disappeared, migrating to eno2
Fixes: acc43b7bf5 ("net: dsa: allow masters to join a LAG")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251215150236.3931670-2-vladimir.oltean@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Problem description
-------------------
DSA has a mumbo-jumbo of reference handling of the conduit net device
and its kobject which, sadly, is just wrong and doesn't make sense.
There are two distinct problems.
1. The OF path, which uses of_find_net_device_by_node(), never releases
the elevated refcount on the conduit's kobject. Nominally, the OF and
non-OF paths should result in objects having identical reference
counts taken, and it is already suspicious that
dsa_dev_to_net_device() has a put_device() call which is missing in
dsa_port_parse_of(), but we can actually even verify that an issue
exists. With CONFIG_DEBUG_KOBJECT_RELEASE=y, if we run this command
"before" and "after" applying this patch:
(unbind the conduit driver for net device eno2)
echo 0000:00:00.2 > /sys/bus/pci/drivers/fsl_enetc/unbind
we see these lines in the output diff which appear only with the patch
applied:
kobject: 'eno2' (ffff002009a3a6b8): kobject_release, parent 0000000000000000 (delayed 1000)
kobject: '109' (ffff0020099d59a0): kobject_release, parent 0000000000000000 (delayed 1000)
2. After we find the conduit interface one way (OF) or another (non-OF),
it can get unregistered at any time, and DSA remains with a long-lived,
but in this case stale, cpu_dp->conduit pointer. Holding the net
device's underlying kobject isn't actually of much help, it just
prevents it from being freed (but we never need that kobject
directly). What helps us to prevent the net device from being
unregistered is the parallel netdev reference mechanism (dev_hold()
and dev_put()).
Actually we actually use that netdev tracker mechanism implicitly on
user ports since commit 2f1e8ea726 ("net: dsa: link interfaces with
the DSA master to get rid of lockdep warnings"), via netdev_upper_dev_link().
But time still passes at DSA switch probe time between the initial
of_find_net_device_by_node() code and the user port creation time, time
during which the conduit could unregister itself and DSA wouldn't know
about it.
So we have to run of_find_net_device_by_node() under rtnl_lock() to
prevent that from happening, and release the lock only with the netdev
tracker having acquired the reference.
Do we need to keep the reference until dsa_unregister_switch() /
dsa_switch_shutdown()?
1: Maybe yes. A switch device will still be registered even if all user
ports failed to probe, see commit 86f8b1c01a ("net: dsa: Do not
make user port errors fatal"), and the cpu_dp->conduit pointers
remain valid. I haven't audited all call paths to see whether they
will actually use the conduit in lack of any user port, but if they
do, it seems safer to not rely on user ports for that reference.
2. Definitely yes. We support changing the conduit which a user port is
associated to, and we can get into a situation where we've moved all
user ports away from a conduit, thus no longer hold any reference to
it via the net device tracker. But we shouldn't let it go nonetheless
- see the next change in relation to dsa_tree_find_first_conduit()
and LAG conduits which disappear.
We have to be prepared to return to the physical conduit, so the CPU
port must explicitly keep another reference to it. This is also to
say: the user ports and their CPU ports may not always keep a
reference to the same conduit net device, and both are needed.
As for the conduit's kobject for the /sys/class/net/ entry, we don't
care about it, we can release it as soon as we hold the net device
object itself.
History and blame attribution
-----------------------------
The code has been refactored so many times, it is very difficult to
follow and properly attribute a blame, but I'll try to make a short
history which I hope to be correct.
We have two distinct probing paths:
- one for OF, introduced in 2016 in commit 83c0afaec7 ("net: dsa: Add
new binding implementation")
- one for non-OF, introduced in 2017 in commit 71e0bbde0d ("net: dsa:
Add support for platform data")
These are both complete rewrites of the original probing paths (which
used struct dsa_switch_driver and other weird stuff, instead of regular
devices on their respective buses for register access, like MDIO, SPI,
I2C etc):
- one for OF, introduced in 2013 in commit 5e95329b70 ("dsa: add
device tree bindings to register DSA switches")
- one for non-OF, introduced in 2008 in commit 91da11f870 ("net:
Distributed Switch Architecture protocol support")
except for tiny bits and pieces like dsa_dev_to_net_device() which were
seemingly carried over since the original commit, and used to this day.
The point is that the original probing paths received a fix in 2015 in
the form of commit 679fb46c57 ("net: dsa: Add missing master netdev
dev_put() calls"), but the fix never made it into the "new" (dsa2)
probing paths that can still be traced to today, and the fixed probing
path was later deleted in 2019 in commit 93e86b3bc8 ("net: dsa: Remove
legacy probing support").
That is to say, the new probing paths were never quite correct in this
area.
The existence of the legacy probing support which was deleted in 2019
explains why dsa_dev_to_net_device() returns a conduit with elevated
refcount (because it was supposed to be released during
dsa_remove_dst()). After the removal of the legacy code, the only user
of dsa_dev_to_net_device() calls dev_put(conduit) immediately after this
function returns. This pattern makes no sense today, and can only be
interpreted historically to understand why dev_hold() was there in the
first place.
Change details
--------------
Today we have a better netdev tracking infrastructure which we should
use. Logically netdev_hold() belongs in common code
(dsa_port_parse_cpu(), where dp->conduit is assigned), but there is a
tradeoff to be made with the rtnl_lock() section which would become a
bit too long if we did that - dsa_port_parse_cpu() also calls
request_module(). So we duplicate a bit of logic in order for the
callers of dsa_port_parse_cpu() to be the ones responsible of holding
the conduit reference and releasing it on error. This shortens the
rtnl_lock() section significantly.
In the dsa_switch_probe() error path, dsa_switch_release_ports() will be
called in a number of situations, one being where dsa_port_parse_cpu()
maybe didn't get the chance to run at all (a different port failed
earlier, etc). So we have to test for the conduit being NULL prior to
calling netdev_put().
There have still been so many transformations to the code since the
blamed commits (rename master -> conduit, commit 0650bf52b3 ("net:
dsa: be compatible with masters which unregister on shutdown")), that it
only makes sense to fix the code using the best methods available today
and see how it can be backported to stable later. I suspect the fix
cannot even be backported to kernels which lack dsa_switch_shutdown(),
and I suspect this is also maybe why the long-lived conduit reference
didn't make it into the new DSA probing paths at the time (problems
during shutdown).
Because dsa_dev_to_net_device() has a single call site and has to be
changed anyway, the logic was just absorbed into the non-OF
dsa_port_parse().
Tested on the ocelot/felix switch and on dsa_loop, both on the NXP
LS1028A with CONFIG_DEBUG_KOBJECT_RELEASE=y.
Reported-by: Ma Ke <make24@iscas.ac.cn>
Closes: https://lore.kernel.org/netdev/20251214131204.4684-1-make24@iscas.ac.cn/
Fixes: 83c0afaec7 ("net: dsa: Add new binding implementation")
Fixes: 71e0bbde0d ("net: dsa: Add support for platform data")
Reviewed-by: Jonas Gorski <jonas.gorski@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251215150236.3931670-1-vladimir.oltean@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
It turns out that HSR offloads are so fine-grained that many DSA
switches can do a small part even though they weren't specifically
designed for the protocols supported by that driver (HSR and PRP).
Specifically NETIF_F_HW_HSR_DUP - it is simple packet duplication on
transmit, towards all (aka 2) ports members of the HSR device.
For many DSA switches, we know how to duplicate a packet, even though we
never typically use that feature. The transmit port mask from the
tagging protocol can have multiple bits set, and the switch should send
the packet once to every port with a bit set from that mask.
Nonetheless, not all tagging protocols are like this, and sometimes the
port is a single numeric value rather than a bit mask. For that reason,
and also because switches can sometimes change tagging protocols for
different ones, we need to make HSR offload helpers opt-in.
For devices that can do nothing else HSR-specific, we introduce
dsa_port_simple_hsr_join() and dsa_port_simple_hsr_leave(). These
functions monitor when two user ports of the same switch are part of the
same HSR device, and when that condition is true, they toggle the
NETIF_F_HW_HSR_DUP feature flag of both net devices.
Normally only dsa_port_simple_hsr_join() and dsa_port_simple_hsr_leave()
are needed. The dsa_port_simple_hsr_validate() helper is just to see
what kind of configuration could be offloadable using the generic
helpers. This is used by switch drivers which are not currently using
the right tagging protocol to offload this HSR ring, but could in
principle offload it after changing the tagger.
Suggested-by: David Yang <mmyangfl@gmail.com>
Cc: "Alvin Šipraga" <alsi@bang-olufsen.dk>
Cc: Chester A. Unal" <chester.a.unal@arinc9.com>
Cc: "Clément Léger" <clement.leger@bootlin.com>
Cc: Daniel Golle <daniel@makrotopia.org>
Cc: DENG Qingfang <dqfext@gmail.com>
Cc: Florian Fainelli <florian.fainelli@broadcom.com>
Cc: George McCollister <george.mccollister@gmail.com>
Cc: Hauke Mehrtens <hauke@hauke-m.de>
Cc: Jonas Gorski <jonas.gorski@gmail.com>
Cc: Kurt Kanzenbach <kurt@linutronix.de>
Cc: Linus Walleij <linus.walleij@linaro.org>
Cc: Sean Wang <sean.wang@mediatek.com>
Cc: UNGLinuxDriver@microchip.com
Cc: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251130131657.65080-6-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This mirrors what we do in dsa_port_lag_leave() and
dsa_port_bridge_leave(): when ds->ops->port_hsr_join() returns
-EOPNOTSUPP, we fall back to a software implementation where dp->hsr_dev
is NULL, and the unoffloaded port is no longer bothered with calls from
the HSR layer.
This helps, for example, with interlink ports which current DSA drivers
don't know how to offload. We have to check only in port_hsr_join() for
the port type, then in port_hsr_leave() we are sure we're dealing only
with known port types.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251130131657.65080-4-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The "xrs700x" is the original DSA tagging protocol with HSR TX
replication support, we now essentially move that logic to the
dsa_xmit_port_mask() helper. The end result is something akin to
hellcreek_xmit() (but reminds me I should also take care of
skb_checksum_help() for tail taggers in the core).
The implementation differences to dsa_xmit_port_mask() are immaterial.
Cc: George McCollister <george.mccollister@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251127120902.292555-15-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The "ocelot" and "seville" tagging protocols populate a bit mask for the
TX ports, so we can use dsa_xmit_port_mask() to centralize the decision
of how to set that field.
This protocol used BIT_ULL() rather than simple BIT() to silence Smatch,
as explained in commit 1f778d500d ("net: mscc: ocelot: avoid type
promotion when calling ocelot_ifh_set_dest"). I would expect that this
tool no longer complains now, when the BIT(dp->index) is hidden inside
the dsa_xmit_port_mask() function, the return value of which is promoted
to u64.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251127120902.292555-9-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Many tagging protocols deal with the transmit port mask being a bit
mask, and set it to BIT(dp->index). Not a big deal.
Also, some tagging protocols are written for switches which support HSR
offload (including packet duplication offload), there we see a walk
using dsa_hsr_foreach_port() to find the other port in the same switch
that's member of the HSR, and set that bit in the port mask too.
That isn't sufficiently interesting either, until you come to realize
that there isn't anything special in the second case that switches just
in the first one can't do too.
It just becomes a matter of "is it wise to do it? are sufficient people
using HSR/PRP with generic off-the-shelf switches to justify add an
extra test in the data path?" - the answer to which is probably "it
depends". It isn't _much_ worse to not have HSR offload at all, so as to
make it impractical, esp. with a rich OS like Linux. But the HSR users
are rather specialized in industrial networking.
Anyway, the change acts on the premise that we're going to have support
for this, it should be uniformly implemented for everyone, and that if
we find some sort of balance, we can keep everyone relatively happy.
So I've disabled that logic if CONFIG_HSR isn't enabled, and I've tilted
the branch predictor to say it's unlikely we're transmitting through a
port with this capability currently active. On branch miss, we're still
going to save the transmission of one packet, so there's some remaining
benefit there too. I don't _think_ we need to jump to static keys yet.
The helper returns a 32-bit zero-based unsigned number, that callers
have to transpose using FIELD_PREP(). It is not the first time we assume
DSA switches won't be larger than 32 ports - dsa_user_ports() has that
assumption baked into it too.
One last development note about why pass the "skb" argument when this
isn't used. Looking at the compiled code on arm64, which is identical
both with and without it, the answer is "why not?" - who knows what
other features dependent on the skb may be handled in the future.
Link: https://lore.kernel.org/netdev/20251126093240.2853294-4-mmyangfl@gmail.com/
Cc: "Alvin Šipraga" <alsi@bang-olufsen.dk>
Cc: Chester A. Unal" <chester.a.unal@arinc9.com>
Cc: "Clément Léger" <clement.leger@bootlin.com>
Cc: Daniel Golle <daniel@makrotopia.org>
Cc: David Yang <mmyangfl@gmail.com>
Cc: DENG Qingfang <dqfext@gmail.com>
Cc: Florian Fainelli <florian.fainelli@broadcom.com>
Cc: George McCollister <george.mccollister@gmail.com>
Cc: Hauke Mehrtens <hauke@hauke-m.de>
Cc: Jonas Gorski <jonas.gorski@gmail.com>
Cc: Kurt Kanzenbach <kurt@linutronix.de>
Cc: Linus Walleij <linus.walleij@linaro.org>
Cc: Sean Wang <sean.wang@mediatek.com>
Cc: UNGLinuxDriver@microchip.com
Cc: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251127120902.292555-2-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently there is no way to see packet counters on cascade ports, and
no clarity on how the API for that would look like.
Because it's something that is currently needed, just extend the hack
where ethtool -S on the conduit interface dumps CPU port counters, and
also use it to dump counters of cascade ports.
Note that the "pXX_" naming convention changes to "sXX_pYY", to
distinguish between ports having the same index but belonging to
different switches. This has a slight chance of causing regressions to
existing tooling:
- grepping for "p04_counter_name" still works, but might return more
than one string now
- grepping for " p04_counter_name" no longer works
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251122112311.138784-4-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In theory this would have been seen by now, but it seems that all
drivers used as DSA conduit interfaces thus far have had ethtool_ops
set, and it's hard to even find modern Ethernet drivers (and not VF
ones) which don't use ethtool.
Here is the unfiltered list of drivers which register any sort of
net_device but don't set its ethtool_ops pointer. I don't think any of
them 'risks' being used as a DSA conduit, maybe except for moxart,
rnpbge and icssm, I'm not sure.
- drivers/net/can/dev/dev.c
- drivers/net/wwan/qcom_bam_dmux.c
- drivers/net/wwan/t7xx/t7xx_netdev.c
- drivers/net/arcnet/arcnet.c
- drivers/net/hamradio/
- drivers/net/slip/slip.c
- drivers/net/ethernet/ezchip/nps_enet.c
- drivers/net/ethernet/moxa/moxart_ether.c
- drivers/net/ethernet/wangxun/txgbevf/txgbevf_main.c
- drivers/net/ethernet/wangxun/ngbevf/ngbevf_main.c
- drivers/net/ethernet/huawei/hinic3/hinic3_main.c
- drivers/net/ethernet/i825xx/
- drivers/net/ethernet/ti/icssm/icssm_prueth.c
- drivers/net/ethernet/seeq/
- drivers/net/ethernet/litex/litex_liteeth.c
- drivers/net/ethernet/sunplus/spl2sw_driver.c
- drivers/net/ethernet/mucse/rnpgbe/rnpgbe_main.c
- drivers/net/ipa/
- drivers/net/wireless/microchip/wilc1000/
- drivers/net/wireless/mediatek/mt76/dma.c
- drivers/net/wireless/ath/ath12k/
- drivers/net/wireless/ath/ath11k/
- drivers/net/wireless/ath/ath6kl/
- drivers/net/wireless/ath/ath10k/
- drivers/net/wireless/intel/iwlwifi/pcie/gen1_2/trans.c
- drivers/net/wireless/virtual/mac80211_hwsim.c
- drivers/net/wireless/quantenna/qtnfmac/pcie/pcie.c
- drivers/net/wireless/realtek/rtw89/core.c
- drivers/net/wireless/realtek/rtw88/pci.c
- drivers/net/caif/
- drivers/net/plip/
- drivers/net/wan/
- drivers/net/mctp/
- drivers/net/ppp/
- drivers/net/thunderbolt/
Nonetheless, it's good for the framework not to make such assumptions,
and not panic when coming across such kind of host device in the future.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251122112311.138784-2-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allow devlink_param::get() handlers to report error messages via
extack. This function is called in a few different contexts, but not
all of them will have an valid extack to use.
When devlink_param::get() is called from param_get_doit or
param_get_dumpit contexts, pass the extack through so that drivers can
report errors when retrieving param values. devlink_param::get() is
called from the context of devlink_param_notify(), pass NULL in for
the extack.
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Daniel Zahka <daniel.zahka@gmail.com>
Link: https://patch.msgid.link/20251119025038.651131-2-daniel.zahka@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cross-merge networking fixes after downstream PR (net-6.18-rc6).
No conflicts, adjacent changes in:
drivers/net/phy/micrel.c
96a9178a29 ("net: phy: micrel: lan8814 fix reset of the QSGMII interface")
61b7ade9ba ("net: phy: micrel: Add support for non PTP SKUs for lan8814")
and a trivial one in tools/testing/selftests/drivers/net/Makefile.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Broadcom switches locally terminate link local traffic and do not
forward it, so we should not mark it as offloaded.
In some situations we still want/need to flood this traffic, e.g. if STP
is disabled, or it is explicitly enabled via the group_fwd_mask. But if
the skb is marked as offloaded, the kernel will assume this was already
done in hardware, and the packets never reach other bridge ports.
So ensure that link local traffic is never marked as offloaded, so that
the kernel can forward/flood these packets in software if needed.
Since the local termination in not configurable, check the destination
MAC, and never mark packets as offloaded if it is a link local ether
address.
While modern switches set the tag reason code to BRCM_EG_RC_PROT_TERM
for trapped link local traffic, they also set it for link local traffic
that is flooded (01:80:c2:00:00:10 to 01:80:c2:00:00:2f), so we cannot
use it and need to look at the destination address for them as well.
Fixes: 964dbf186e ("net: dsa: tag_brcm: add support for legacy tags")
Fixes: 0e62f543be ("net: dsa: Fix duplicate frames flooded by learning")
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://patch.msgid.link/20251109134635.243951-1-jonas.gorski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cross-merge networking fixes after downstream PR (net-6.18-rc5).
Conflicts:
drivers/net/wireless/ath/ath12k/mac.c
9222582ec5 ("Revert "wifi: ath12k: Fix missing station power save configuration"")
6917e268c4 ("wifi: ath12k: Defer vdev bring-up until CSA finalize to avoid stale beacon")
https://lore.kernel.org/11cece9f7e36c12efd732baa5718239b1bf8c950.camel@sipsolutions.net
Adjacent changes:
drivers/net/ethernet/intel/Kconfig
b1d16f7c00 ("libie: depend on DEBUG_FS when building LIBIE_FWLOG")
93f53db9f9 ("ice: switch to Page Pool")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The internal switch on BCM63XX SoCs will unconditionally add 802.1Q VLAN
tags on egress to CPU when 802.1Q mode is enabled. We do this
unconditionally since commit ed409f3bba ("net: dsa: b53: Configure
VLANs while not filtering").
This is fine for VLAN aware bridges, but for standalone ports and vlan
unaware bridges this means all packets are tagged with the default VID,
which is 0.
While the kernel will treat that like untagged, this can break userspace
applications processing raw packets, expecting untagged traffic, like
STP daemons.
This also breaks several bridge tests, where the tcpdump output then
does not match the expected output anymore.
Since 0 isn't a valid VID, just strip out the VLAN tag if we encounter
it, unless the priority field is set, since that would be a valid tag
again.
Fixes: 964dbf186e ("net: dsa: tag_brcm: add support for legacy tags")
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/20251027194621.133301-1-jonas.gorski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit cc34acd577 ("docs: net: document new locking reality")
introduced netif_ vs dev_ function semantics: the former expects locked
netdev, the latter takes care of the locking. We don't strictly
follow this semantics on either side, but there are more dev_xxx handlers
now that don't fit. Rename them to netif_xxx where appropriate.
netif_close_many is used only by vlan/dsa and one mtk driver, so move it into
NETDEV_INTERNAL namespace.
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250717172333.1288349-8-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for legacy Broadcom FCS tags, which are similar to
DSA_TAG_PROTO_BRCM_LEGACY.
BCM5325 and BCM5365 switches require including the original FCS value and
length, as opposed to BCM63xx switches.
Adding the original FCS value and length to DSA_TAG_PROTO_BRCM_LEGACY would
impact performance of BCM63xx switches, so it's better to create a new tag.
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://patch.msgid.link/20250614080000.1884236-3-noltari@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
New timestamping API was introduced in commit 66f7223039 ("net: add
NDOs for configuring hardware timestamping") from kernel v6.6. It is
time to convert DSA to the new API, so that the ndo_eth_ioctl() path can
be removed completely.
Move the ds->ops->port_hwtstamp_get() and ds->ops->port_hwtstamp_set()
calls from dsa_user_ioctl() to dsa_user_hwtstamp_get() and
dsa_user_hwtstamp_set().
Due to the fact that the underlying ifreq type changes to
kernel_hwtstamp_config, the drivers and the Ocelot switchdev front-end,
all hooked up directly or indirectly, must also be converted all at once.
The conversion also updates the comment from dsa_port_supports_hwtstamp(),
which is no longer true because kernel_hwtstamp_config is kernel memory
and does not need copy_to_user(). I've deliberated whether it is
necessary to also update "err != -EOPNOTSUPP" to a more general "!err",
but all drivers now either return 0 or -EOPNOTSUPP.
The existing logic from the ocelot_ioctl() function, to avoid
configuring timestamping if the PHY supports the operation, is obsoleted
by more advanced core logic in dev_set_hwtstamp_phylib().
This is only a partial preparation for proper PHY timestamping support.
None of these switch driver currently sets up PTP traps for PHY
timestamping, so setting dev->see_all_hwtstamp_requests is not yet
necessary and the conversion is relatively trivial.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # felix, sja1105, mv88e6xxx
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20250508095236.887789-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If complete = true in dsa_tree_setup(), it means that we are the last
switch of the tree which is successfully probing, and we should be
setting up all switches from our probe path.
After "complete" becomes true, dsa_tree_setup_cpu_ports() or any
subsequent function may fail. If that happens, the entire tree setup is
in limbo: the first N-1 switches have successfully finished probing
(doing nothing but having allocated persistent memory in the tree's
dst->ports, and maybe dst->rtable), and switch N failed to probe, ending
the tree setup process before anything is tangible from the user's PoV.
If switch N fails to probe, its memory (ports) will be freed and removed
from dst->ports. However, the dst->rtable elements pointing to its ports,
as created by dsa_link_touch(), will remain there, and will lead to
use-after-free if dereferenced.
If dsa_tree_setup_switches() returns -EPROBE_DEFER, which is entirely
possible because that is where ds->ops->setup() is, we get a kasan
report like this:
==================================================================
BUG: KASAN: slab-use-after-free in mv88e6xxx_setup_upstream_port+0x240/0x568
Read of size 8 at addr ffff000004f56020 by task kworker/u8:3/42
Call trace:
__asan_report_load8_noabort+0x20/0x30
mv88e6xxx_setup_upstream_port+0x240/0x568
mv88e6xxx_setup+0xebc/0x1eb0
dsa_register_switch+0x1af4/0x2ae0
mv88e6xxx_register_switch+0x1b8/0x2a8
mv88e6xxx_probe+0xc4c/0xf60
mdio_probe+0x78/0xb8
really_probe+0x2b8/0x5a8
__driver_probe_device+0x164/0x298
driver_probe_device+0x78/0x258
__device_attach_driver+0x274/0x350
Allocated by task 42:
__kasan_kmalloc+0x84/0xa0
__kmalloc_cache_noprof+0x298/0x490
dsa_switch_touch_ports+0x174/0x3d8
dsa_register_switch+0x800/0x2ae0
mv88e6xxx_register_switch+0x1b8/0x2a8
mv88e6xxx_probe+0xc4c/0xf60
mdio_probe+0x78/0xb8
really_probe+0x2b8/0x5a8
__driver_probe_device+0x164/0x298
driver_probe_device+0x78/0x258
__device_attach_driver+0x274/0x350
Freed by task 42:
__kasan_slab_free+0x48/0x68
kfree+0x138/0x418
dsa_register_switch+0x2694/0x2ae0
mv88e6xxx_register_switch+0x1b8/0x2a8
mv88e6xxx_probe+0xc4c/0xf60
mdio_probe+0x78/0xb8
really_probe+0x2b8/0x5a8
__driver_probe_device+0x164/0x298
driver_probe_device+0x78/0x258
__device_attach_driver+0x274/0x350
The simplest way to fix the bug is to delete the routing table in its
entirety. dsa_tree_setup_routing_table() has no problem in regenerating
it even if we deleted links between ports other than those of switch N,
because dsa_link_touch() first checks whether the port pair already
exists in dst->rtable, allocating if not.
The deletion of the routing table in its entirety already exists in
dsa_tree_teardown(), so refactor that into a function that can also be
called from the tree setup error path.
In my analysis of the commit to blame, it is the one which added
dsa_link elements to dst->rtable. Prior to that, each switch had its own
ds->rtable which is freed when the switch fails to probe. But the tree
is potentially persistent memory.
Fixes: c5f51765a1 ("net: dsa: list DSA links in the fabric")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20250414213001.2957964-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As explained in many places such as commit b117e1e8a8 ("net: dsa:
delete dsa_legacy_fdb_add and dsa_legacy_fdb_del"), DSA is written given
the assumption that higher layers have balanced additions/deletions.
As such, it only makes sense to be extremely vocal when those
assumptions are violated and the driver unbinds with entries still
present.
But Ido Schimmel points out a very simple situation where that is wrong:
https://lore.kernel.org/netdev/ZDazSM5UsPPjQuKr@shredder/
(also briefly discussed by me in the aforementioned commit).
Basically, while the bridge bypass operations are not something that DSA
explicitly documents, and for the majority of DSA drivers this API
simply causes them to go to promiscuous mode, that isn't the case for
all drivers. Some have the necessary requirements for bridge bypass
operations to do something useful - see dsa_switch_supports_uc_filtering().
Although in tools/testing/selftests/net/forwarding/local_termination.sh,
we made an effort to popularize better mechanisms to manage address
filters on DSA interfaces from user space - namely macvlan for unicast,
and setsockopt(IP_ADD_MEMBERSHIP) - through mtools - for multicast, the
fact is that 'bridge fdb add ... self static local' also exists as
kernel UAPI, and might be useful to someone, even if only for a quick
hack.
It seems counter-productive to block that path by implementing shim
.ndo_fdb_add and .ndo_fdb_del operations which just return -EOPNOTSUPP
in order to prevent the ndo_dflt_fdb_add() and ndo_dflt_fdb_del() from
running, although we could do that.
Accepting that cleanup is necessary seems to be the only option.
Especially since we appear to be coming back at this from a different
angle as well. Russell King is noticing that the WARN_ON() triggers even
for VLANs:
https://lore.kernel.org/netdev/Z_li8Bj8bD4-BYKQ@shell.armlinux.org.uk/
What happens in the bug report above is that dsa_port_do_vlan_del() fails,
then the VLAN entry lingers on, and then we warn on unbind and leak it.
This is not a straight revert of the blamed commit, but we now add an
informational print to the kernel log (to still have a way to see
that bugs exist), and some extra comments gathered from past years'
experience, to justify the logic.
Fixes: 0832cd9f1f ("net: dsa: warn if port lists aren't empty in dsa_port_teardown")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20250414212930.2956310-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move the more esoteric helpers for netdev instance lock to
a dedicated header. This avoids growing netdevice.h to infinity
and makes rebuilding the kernel much faster (after touching
the header with the helpers).
The main netdev_lock() / netdev_unlock() functions are used
in static inlines in netdevice.h and will probably be used
most commonly, so keep them in netdevice.h.
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250307183006.2312761-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Protect all ethtool callbacks and PHY related state with the netdev
instance lock, for drivers which want / need to have their ops
instance-locked. Basically take the lock everywhere we take rtnl_lock.
It was tempting to take the lock in ethnl_ops_begin(), but turns
out we actually nest those calls (when generating notifications).
Tested-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Cc: Saeed Mahameed <saeed@kernel.org>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250305163732.2766420-11-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Support hw SCM_TSTAMP_SND case for bpf timestamping.
Add a new sock_ops callback, BPF_SOCK_OPS_TSTAMP_SND_HW_CB. This
callback will occur at the same timestamping point as the user
space's hardware SCM_TSTAMP_SND. The BPF program can use it to
get the same SCM_TSTAMP_SND timestamp without modifying the
user-space application.
To avoid increasing the code complexity, replace SKBTX_HW_TSTAMP
with SKBTX_HW_TSTAMP_NOBPF instead of changing numerous callers
from driver side using SKBTX_HW_TSTAMP. The new definition of
SKBTX_HW_TSTAMP means the combination tests of socket timestamping
and bpf timestamping. After this patch, drivers can work under the
bpf timestamping.
Considering some drivers don't assign the skb with hardware
timestamp, this patch does the assignment and then BPF program
can acquire the hwstamp from skb directly.
Signed-off-by: Jason Xing <kerneljasonxing@gmail.com>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250220072940.99994-9-kerneljasonxing@gmail.com
In order to allow DSA drivers to use phylink managed EEE, we need to
change the behaviour of the DSA's .set_eee() ethtool method.
Implementation of the DSA .set_mac_eee() method becomes optional with
phylink managed EEE as it is only used to validate the EEE parameters
supplied from userspace. The rest of the EEE state management should
be left to phylink.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/E1thR9l-003vXC-9F@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull driver core and debugfs updates from Greg KH:
"Here is the big set of driver core and debugfs updates for 6.14-rc1.
Included in here is a bunch of driver core, PCI, OF, and platform rust
bindings (all acked by the different subsystem maintainers), hence the
merge conflict with the rust tree, and some driver core api updates to
mark things as const, which will also require some fixups due to new
stuff coming in through other trees in this merge window.
There are also a bunch of debugfs updates from Al, and there is at
least one user that does have a regression with these, but Al is
working on tracking down the fix for it. In my use (and everyone
else's linux-next use), it does not seem like a big issue at the
moment.
Here's a short list of the things in here:
- driver core rust bindings for PCI, platform, OF, and some i/o
functions.
We are almost at the "write a real driver in rust" stage now,
depending on what you want to do.
- misc device rust bindings and a sample driver to show how to use
them
- debugfs cleanups in the fs as well as the users of the fs api for
places where drivers got it wrong or were unnecessarily doing
things in complex ways.
- driver core const work, making more of the api take const * for
different parameters to make the rust bindings easier overall.
- other small fixes and updates
All of these have been in linux-next with all of the aforementioned
merge conflicts, and the one debugfs issue, which looks to be resolved
"soon""
* tag 'driver-core-6.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core: (95 commits)
rust: device: Use as_char_ptr() to avoid explicit cast
rust: device: Replace CString with CStr in property_present()
devcoredump: Constify 'struct bin_attribute'
devcoredump: Define 'struct bin_attribute' through macro
rust: device: Add property_present()
saner replacement for debugfs_rename()
orangefs-debugfs: don't mess with ->d_name
octeontx2: don't mess with ->d_parent or ->d_parent->d_name
arm_scmi: don't mess with ->d_parent->d_name
slub: don't mess with ->d_name
sof-client-ipc-flood-test: don't mess with ->d_name
qat: don't mess with ->d_name
xhci: don't mess with ->d_iname
mtu3: don't mess wiht ->d_iname
greybus/camera - stop messing with ->d_iname
mediatek: stop messing with ->d_iname
netdevsim: don't embed file_operations into your structs
b43legacy: make use of debugfs_get_aux()
b43: stop embedding struct file_operations into their objects
carl9170: stop embedding file_operations into their objects
...
Pull networking updates from Paolo Abeni:
"This is slightly smaller than usual, with the most interesting work
being still around RTNL scope reduction.
Core:
- More core refactoring to reduce the RTNL lock contention, including
preparatory work for the per-network namespace RTNL lock, replacing
RTNL lock with a per device-one to protect NAPI-related net device
data and moving synchronize_net() calls outside such lock.
- Extend drop reasons usage, adding net scheduler, AF_UNIX, bridge
and more specific TCP coverage.
- Reduce network namespace tear-down time by removing per-subsystems
synchronize_net() in tipc and sched.
- Add flow label selector support for fib rules, allowing traffic
redirection based on such header field.
Netfilter:
- Do not remove netdev basechain when last device is gone, allowing
netdev basechains without devices.
- Revisit the flowtable teardown strategy, dealing better with fin,
reset and re-open events.
- Scale-up IP-vs connection dumping by avoiding linear search on each
restart.
Protocols:
- A significant XDP socket refactor, consolidating and optimizing
several helpers into the core
- Better scaling of ICMP rate-limiting, by removing false-sharing in
inet peers handling.
- Introduces netlink notifications for multicast IPv4 and IPv6
address changes.
- Add ipsec support for IP-TFS/AggFrag encapsulation, allowing
aggregation and fragmentation of the inner IP.
- Add sysctl to configure TIME-WAIT reuse delay for TCP sockets, to
avoid local port exhaustion issues when the average connection
lifetime is very short.
- Support updating keys (re-keying) for connections using kernel TLS
(for TLS 1.3 only).
- Support ipv4-mapped ipv6 address clients in smc-r v2.
- Add support for jumbo data packet transmission in RxRPC sockets,
gluing multiple data packets in a single UDP packet.
- Support RxRPC RACK-TLP to manage packet loss and retransmission in
conjunction with the congestion control algorithm.
Driver API:
- Introduce a unified and structured interface for reporting PHY
statistics, exposing consistent data across different H/W via
ethtool.
- Make timestamping selectable, allow the user to select the desired
hwtstamp provider (PHY or MAC) administratively.
- Add support for configuring a header-data-split threshold (HDS)
value via ethtool, to deal with partial or buggy H/W
implementation.
- Consolidate DSA drivers Energy Efficiency Ethernet support.
- Add EEE management to phylink, making use of the phylib
implementation.
- Add phylib support for in-band capabilities negotiation.
- Simplify how phylib-enabled mac drivers expose the supported
interfaces.
Tests and tooling:
- Make the YNL tool package-friendly to make it easier to deploy it
separately from the kernel.
- Increase TCP selftest coverage importing several packetdrill
test-cases.
- Regenerate the ethtool uapi header from the YNL spec, to ease
maintenance and future development.
- Add YNL support for decoding the link types used in net self-tests,
allowing a single build to run both net and drivers/net.
Drivers:
- Ethernet high-speed NICs:
- nVidia/Mellanox (mlx5):
- add cross E-Switch QoS support
- add SW Steering support for ConnectX-8
- implement support for HW-Managed Flow Steering, improving the
rule deletion/insertion rate
- support for multi-host LAG
- Intel (ixgbe, ice, igb):
- ice: add support for devlink health events
- ixgbe: add initial support for E610 chipset variant
- igb: add support for AF_XDP zero-copy
- Meta:
- add support for basic RSS config
- allow changing the number of channels
- add hardware monitoring support
- Broadcom (bnxt):
- implement TCP data split and HDS threshold ethtool support,
enabling Device Memory TCP.
- Marvell Octeon:
- implement egress ipsec offload support for the cn10k family
- Hisilicon (HIBMC):
- implement unicast MAC filtering
- Ethernet NICs embedded and virtual:
- Convert UDP tunnel drivers to NETDEV_PCPU_STAT_DSTATS, avoiding
contented atomic operations for drop counters
- Freescale:
- quicc: phylink conversion
- enetc: support Tx and Rx checksum offload and improve TSO
performances
- MediaTek:
- airoha: introduce support for ETS and HTB Qdisc offload
- Microchip:
- lan78XX USB: preparation work for phylink conversion
- Synopsys (stmmac):
- support DWMAC IP on NXP Automotive SoCs S32G2xx/S32G3xx/S32R45
- refactor EEE support to leverage the new driver API
- optimize DMA and cache access to increase raw RX performances
by 40%
- TI:
- icssg-prueth: add multicast filtering support for VLAN
interface
- netkit:
- add ability to configure head/tailroom
- VXLAN:
- accepts packets with user-defined reserved bit
- Ethernet switches:
- Microchip:
- lan969x: add RGMII support
- lan969x: improve TX and RX performance using the FDMA engine
- nVidia/Mellanox:
- move Tx header handling to PCI driver, to ease XDP support
- Ethernet PHYs:
- Texas Instruments DP83822:
- add support for GPIO2 clock output
- Realtek:
- 8169: add support for RTL8125D rev.b
- rtl822x: add hwmon support for the temperature sensor
- Microchip:
- add support for RDS PTP hardware
- consolidate periodic output signal generation
- CAN:
- several DT-bindings to DT schema conversions
- tcan4x5x:
- add HW standby support
- support nWKRQ voltage selection
- kvaser:
- allowing Bus Error Reporting runtime configuration
- WiFi:
- the on-going Multi-Link Operation (MLO) effort continues,
affecting both the stack and in drivers
- mac80211/cfg80211:
- Emergency Preparedness Communication Services (EPCS) station
mode support
- support for adding and removing station links for MLO
- add support for WiFi 7/EHT mesh over 320 MHz channels
- report Tx power info for each link
- RealTek (rtw88):
- enable USB Rx aggregation and USB 3 to improve performance
- LED support
- RealTek (rtw89):
- refactor power save to support Multi-Link Operations
- add support for RTL8922AE-VS variant
- MediaTek (mt76):
- single wiphy multiband support (preparation for MLO)
- p2p device support
- add TP-Link TXE50UH USB adapter support
- Qualcomm (ath10k):
- support for the QCA6698AQ IP core
- Qualcomm (ath12k):
- enable MLO for QCN9274
- Bluetooth:
- Allow sysfs to trigger hdev reset, to allow recovering devices
not responsive from user-space
- MediaTek: add support for MT7922, MT7925, MT7921e devices
- Realtek: add support for RTL8851BE devices
- Qualcomm: add support for WCN785x devices
- ISO: allow BIG re-sync"
* tag 'net-next-6.14' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next: (1386 commits)
net/rose: prevent integer overflows in rose_setsockopt()
net: phylink: fix regression when binding a PHY
net: ethernet: ti: am65-cpsw: streamline TX queue creation and cleanup
net: ethernet: ti: am65-cpsw: streamline RX queue creation and cleanup
net: ethernet: ti: am65-cpsw: ensure proper channel cleanup in error path
ipv6: Convert inet6_rtm_deladdr() to per-netns RTNL.
ipv6: Convert inet6_rtm_newaddr() to per-netns RTNL.
ipv6: Move lifetime validation to inet6_rtm_newaddr().
ipv6: Set cfg.ifa_flags before device lookup in inet6_rtm_newaddr().
ipv6: Pass dev to inet6_addr_add().
ipv6: Convert inet6_ioctl() to per-netns RTNL.
ipv6: Hold rtnl_net_lock() in addrconf_init() and addrconf_cleanup().
ipv6: Hold rtnl_net_lock() in addrconf_dad_work().
ipv6: Hold rtnl_net_lock() in addrconf_verify_work().
ipv6: Convert net.ipv6.conf.${DEV}.XXX sysctl to per-netns RTNL.
ipv6: Add __in6_dev_get_rtnl_net().
net: stmmac: Drop redundant skb_mark_for_recycle() for SKB frags
net: mii: Fix the Speed display when the network cable is not connected
sysctl net: Remove macro checks for CONFIG_SYSCTL
eth: bnxt: update header sizing defaults
...
Pull kthread updates from Frederic Weisbecker:
"Kthreads affinity follow either of 4 existing different patterns:
1) Per-CPU kthreads must stay affine to a single CPU and never
execute relevant code on any other CPU. This is currently handled
by smpboot code which takes care of CPU-hotplug operations.
Affinity here is a correctness constraint.
2) Some kthreads _have_ to be affine to a specific set of CPUs and
can't run anywhere else. The affinity is set through
kthread_bind_mask() and the subsystem takes care by itself to
handle CPU-hotplug operations. Affinity here is assumed to be a
correctness constraint.
3) Per-node kthreads _prefer_ to be affine to a specific NUMA node.
This is not a correctness constraint but merely a preference in
terms of memory locality. kswapd and kcompactd both fall into this
category. The affinity is set manually like for any other task and
CPU-hotplug is supposed to be handled by the relevant subsystem so
that the task is properly reaffined whenever a given CPU from the
node comes up. Also care should be taken so that the node affinity
doesn't cross isolated (nohz_full) cpumask boundaries.
4) Similar to the previous point except kthreads have a _preferred_
affinity different than a node. Both RCU boost kthreads and RCU
exp kworkers fall into this category as they refer to "RCU nodes"
from a distinctly distributed tree.
Currently the preferred affinity patterns (3 and 4) have at least 4
identified users, with more or less success when it comes to handle
CPU-hotplug operations and CPU isolation. Each of which do it in its
own ad-hoc way.
This is an infrastructure proposal to handle this with the following
API changes:
- kthread_create_on_node() automatically affines the created kthread
to its target node unless it has been set as per-cpu or bound with
kthread_bind[_mask]() before the first wake-up.
- kthread_affine_preferred() is a new function that can be called
right after kthread_create_on_node() to specify a preferred
affinity different than the specified node.
When the preferred affinity can't be applied because the possible
targets are offline or isolated (nohz_full), the kthread is affine to
the housekeeping CPUs (which means to all online CPUs most of the time
or only the non-nohz_full CPUs when nohz_full= is set).
kswapd, kcompactd, RCU boost kthreads and RCU exp kworkers have been
converted, along with a few old drivers.
Summary of the changes:
- Consolidate a bunch of ad-hoc implementations of
kthread_run_on_cpu()
- Introduce task_cpu_fallback_mask() that defines the default last
resort affinity of a task to become nohz_full aware
- Add some correctness check to ensure kthread_bind() is always
called before the first kthread wake up.
- Default affine kthread to its preferred node.
- Convert kswapd / kcompactd and remove their halfway working ad-hoc
affinity implementation
- Implement kthreads preferred affinity
- Unify kthread worker and kthread API's style
- Convert RCU kthreads to the new API and remove the ad-hoc affinity
implementation"
* tag 'kthread-for-6.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks:
kthread: modify kernel-doc function name to match code
rcu: Use kthread preferred affinity for RCU exp kworkers
treewide: Introduce kthread_run_worker[_on_cpu]()
kthread: Unify kthread_create_on_cpu() and kthread_create_worker_on_cpu() automatic format
rcu: Use kthread preferred affinity for RCU boost
kthread: Implement preferred affinity
mm: Create/affine kswapd to its preferred node
mm: Create/affine kcompactd to its preferred node
kthread: Default affine kthread to its preferred NUMA node
kthread: Make sure kthread hasn't started while binding it
sched,arm64: Handle CPU isolation on last resort fallback rq selection
arm64: Exclude nohz_full CPUs from 32bits el0 support
lib: test_objpool: Use kthread_run_on_cpu()
kallsyms: Use kthread_run_on_cpu()
soc/qman: test: Use kthread_run_on_cpu()
arm/bL_switcher: Use kthread_run_on_cpu()
We need the debugfs / driver-core fixes in here as well for testing and
to build on top of.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kthread_create() creates a kthread without running it yet. kthread_run()
creates a kthread and runs it.
On the other hand, kthread_create_worker() creates a kthread worker and
runs it.
This difference in behaviours is confusing. Also there is no way to
create a kthread worker and affine it using kthread_bind_mask() or
kthread_affine_preferred() before starting it.
Consolidate the behaviours and introduce kthread_run_worker[_on_cpu]()
that behaves just like kthread_run(). kthread_create_worker[_on_cpu]()
will now only create a kthread worker without starting it.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
