| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
(rcv) timestamp
skb->tstamp was first used as the (rcv) timestamp.
The major usage is to report it to the user (e.g. SO_TIMESTAMP).
Later, skb->tstamp is also set as the (future) delivery_time (e.g. EDT in TCP)
during egress and used by the qdisc (e.g. sch_fq) to make decision on when
the skb can be passed to the dev.
Currently, there is no way to tell skb->tstamp having the (rcv) timestamp
or the delivery_time, so it is always reset to 0 whenever forwarded
between egress and ingress.
While it makes sense to always clear the (rcv) timestamp in skb->tstamp
to avoid confusing sch_fq that expects the delivery_time, it is a
performance issue [0] to clear the delivery_time if the skb finally
egress to a fq@phy-dev. For example, when forwarding from egress to
ingress and then finally back to egress:
tcp-sender => veth@netns => veth@hostns => fq@eth0@hostns
^ ^
reset rest
This patch adds one bit skb->mono_delivery_time to flag the skb->tstamp
is storing the mono delivery_time (EDT) instead of the (rcv) timestamp.
The current use case is to keep the TCP mono delivery_time (EDT) and
to be used with sch_fq. A latter patch will also allow tc-bpf@ingress
to read and change the mono delivery_time.
In the future, another bit (e.g. skb->user_delivery_time) can be added
for the SCM_TXTIME where the clock base is tracked by sk->sk_clockid.
[ This patch is a prep work. The following patches will
get the other parts of the stack ready first. Then another patch
after that will finally set the skb->mono_delivery_time. ]
skb_set_delivery_time() function is added. It is used by the tcp_output.c
and during ip[6] fragmentation to assign the delivery_time to
the skb->tstamp and also set the skb->mono_delivery_time.
A note on the change in ip_send_unicast_reply() in ip_output.c.
It is only used by TCP to send reset/ack out of a ctl_sk.
Like the new skb_set_delivery_time(), this patch sets
the skb->mono_delivery_time to 0 for now as a place
holder. It will be enabled in a latter patch.
A similar case in tcp_ipv6 can be done with
skb_set_delivery_time() in tcp_v6_send_response().
[0] (slide 22): https://linuxplumbersconf.org/event/11/contributions/953/attachments/867/1658/LPC_2021_BPF_Datapath_Extensions.pdf
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The "ocelot" and "ocelot-8021q" tagging protocols make use of different
hardware resources, and host FDB entries have different destination
ports in the switch analyzer module, practically speaking.
So when the user requests a tagging protocol change, the driver must
migrate all host FDB and MDB entries from the NPI port (in fact CPU port
module) towards the same physical port, but this time used as a regular
port.
It is pointless for the felix driver to keep a copy of the host
addresses, when we can create and export DSA helpers for walking through
the addresses that it already needs to keep on the CPU port, for
refcounting purposes.
felix_classify_db() is moved up to avoid a forward declaration.
We pass "bool change" because dp->fdbs and dp->mdbs are uninitialized
lists when felix_setup() first calls felix_set_tag_protocol(), so we
need to avoid calling dsa_port_walk_fdbs() during probe time.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
DSA can treat IFF_PROMISC and IFF_ALLMULTI on standalone user ports as
signifying whether packets with an unknown MAC DA will be received or
not. Since known MAC DAs are handled by FDB/MDB entries, this means that
promiscuity is analogous to including/excluding the CPU port from the
flood domain of those packets.
There are two ways to signal CPU flooding to drivers.
The first (chosen here) is to synthesize a call to
ds->ops->port_bridge_flags() for the CPU port, with a mask of
BR_FLOOD | BR_MCAST_FLOOD. This has the effect of turning on egress
flooding on the CPU port regardless of source.
The alternative would be to create a new ds->ops->port_host_flood()
which is called per user port. Some switches (sja1105) have a flood
domain that is managed per {ingress port, egress port} pair, so it would
make more sense for this kind of switch to not flood the CPU from port A
if just port B requires it. Nonetheless, the sja1105 has other quirks
that prevent it from making use of unicast filtering, and without a
concrete user making use of this feature, I chose not to implement it.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
To be able to safely turn off CPU flooding for standalone ports, we need
to ensure that the dev_addr of each DSA slave interface is installed as
a standalone host FDB entry for compatible switches.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In preparation of disabling flooding towards the CPU in standalone ports
mode, identify the addresses requested by upper interfaces and use the
new API for DSA FDB isolation to request the hardware driver to offload
these as FDB or MDB objects. The objects belong to the user port's
database, and are installed pointing towards the CPU port.
Because dev_uc_add()/dev_mc_add() is VLAN-unaware, we offload to the
port standalone database addresses with VID 0 (also VLAN-unaware).
So this excludes switches with global VLAN filtering from supporting
unicast filtering, because there, it is possible for a port of a switch
to join a VLAN-aware bridge, and this changes the VLAN awareness of
standalone ports, requiring VLAN-aware standalone host FDB entries.
For the same reason, hellcreek, which requires VLAN awareness in
standalone mode, is also exempted from unicast filtering.
We create "standalone" variants of dsa_port_host_fdb_add() and
dsa_port_host_mdb_add() (and the _del coresponding functions).
We also create a separate work item type for handling deferred
standalone host FDB/MDB entries compared to the switchdev one.
This is done for the purpose of clarity - the procedure for offloading a
bridge FDB entry is different than offloading a standalone one, and
the switchdev event work handles only FDBs anyway, not MDBs.
Deferral is needed for standalone entries because ndo_set_rx_mode runs
in atomic context. We could probably optimize things a little by first
queuing up all entries that need to be offloaded, and scheduling the
work item just once, but the data structures that we can pass through
__dev_uc_sync() and __dev_mc_sync() are limiting (there is nothing like
a void *priv), so we'd have to keep the list of queued events somewhere
in struct dsa_switch, and possibly a lock for it. Too complicated for
now.
Adding the address to the master is handled by dev_uc_sync(), adding it
to the hardware is handled by __dev_uc_sync(). So this is the reason why
dsa_port_standalone_host_fdb_add() does not call dev_uc_add(). Not that
it had the rtnl_mutex anyway - ndo_set_rx_mode has it, but is atomic.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We are preparing to add API in port.c that adds FDB and MDB entries that
correspond to the port's standalone database. Rename the existing
methods to make it clear that the FDB and MDB entries offloaded come
from the bridge database.
Since the function names lengthen in dsa_slave_switchdev_event_work(),
we place "addr" and "vid" in temporary variables, to shorten those.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Lennert Buytenhek explains in commit df02c6ff2e39 ("dsa: fix master
interface allmulti/promisc handling"), dated Nov 2008, that changing the
promiscuity of interfaces that are down (here the master) is broken.
This fact regarding promisc/allmulti has changed since commit
b6c40d68ff64 ("net: only invoke dev->change_rx_flags when device is UP")
by Vlad Yasevich, dated Nov 2013.
Therefore, DSA now has unnecessary complexity to handle master state
transitions from down to up. In fact, syncing the unicast and multicast
addresses can happen completely asynchronously to the administrative
state changes.
This change reduces that complexity by effectively fully reverting
commit df02c6ff2e39 ("dsa: fix master interface allmulti/promisc
handling").
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This reverts commit 17f7ae16aef1f58bc4af4c7a16b8778a91a30255.
The commit brought a new socket state LLCP_DISCONNECTING, which was
never set, only read, so socket could never set to such state.
Remove the dead code.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
nfc_llcp_sock_link() is called in all paths (bind/connect) as a last
action, still protected with lock_sock(). When cleaning up in
llcp_sock_release(), call nfc_llcp_sock_unlink() in a mirrored way:
earlier and still under the lock_sock().
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Use test_bit() instead of open-coding it, just like in other places
touching the bitmap.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Coding style encourages centralized exiting of functions, so rewrite
llcp_sock_bind() error paths to use such pattern. This reduces the
duplicated cleanup code, make success path visually shorter and also
cleans up the errors in proper order (in reversed way from
initialization).
No functional impact expected.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The llcp_sock_connect() error paths were using a mixed way of central
exit (goto) and cleanup
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Nullify the llcp_sock->dev on llcp_sock_connect() error paths,
symmetrically to the code llcp_sock_bind(). The non-NULL value of
llcp_sock->dev is used in a few places to check whether the socket is
still valid.
There was no particular issue observed with missing NULL assignment in
connect() error path, however a similar case - in the bind() error path
- was triggereable. That one was fixed in commit 4ac06a1e013c ("nfc:
fix NULL ptr dereference in llcp_sock_getname() after failed connect"),
so the change here seems logical as well.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The offloaded HW stats are designed to allow per-netdevice enablement and
disablement. Add an attribute, IFLA_STATS_SET_OFFLOAD_XSTATS_L3_STATS,
which should be carried by the RTM_SETSTATS message, and expresses a desire
to toggle L3 offload xstats on or off.
As part of the above, add an exported function rtnl_offload_xstats_notify()
that drivers can use when they have installed or deinstalled the counters
backing the HW stats.
At this point, it is possible to enable, disable and query L3 offload
xstats on netdevices. (However there is no driver actually implementing
these.)
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The offloaded HW stats are designed to allow per-netdevice enablement and
disablement. These stats are only accessible through RTM_GETSTATS, and
therefore should be toggled by a RTM_SETSTATS message. Add it, and the
necessary skeleton handler.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a new IFLA_STATS_LINK_OFFLOAD_XSTATS child attribute,
IFLA_OFFLOAD_XSTATS_L3_STATS, to carry statistics for traffic that takes
place in a HW router.
The offloaded HW stats are designed to allow per-netdevice enablement and
disablement. Additionally, as a netdevice is configured, it may become or
cease being suitable for binding of a HW counter. Both of these aspects
need to be communicated to the userspace. To that end, add another child
attribute, IFLA_OFFLOAD_XSTATS_HW_S_INFO:
- attr nest IFLA_OFFLOAD_XSTATS_HW_S_INFO
- attr nest IFLA_OFFLOAD_XSTATS_L3_STATS
- attr IFLA_OFFLOAD_XSTATS_HW_S_INFO_REQUEST
- {0,1} as u8
- attr IFLA_OFFLOAD_XSTATS_HW_S_INFO_USED
- {0,1} as u8
Thus this one attribute is a nest that can be used to carry information
about various types of HW statistics, and indexing is very simply done by
wrapping the information for a given statistics suite into the attribute
that carries the suite is the RTM_GETSTATS query. At the same time, because
_HW_S_INFO is nested directly below IFLA_STATS_LINK_OFFLOAD_XSTATS, it is
possible through filtering to request only the metadata about individual
statistics suites, without having to hit the HW to get the actual counters.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Offloading switch device drivers may be able to collect statistics of the
traffic taking place in the HW datapath that pertains to a certain soft
netdevice, such as VLAN. Add the necessary infrastructure to allow exposing
these statistics to the offloaded netdevice in question. The API was shaped
by the following considerations:
- Collection of HW statistics is not free: there may be a finite number of
counters, and the act of counting may have a performance impact. It is
therefore necessary to allow toggling whether HW counting should be done
for any particular SW netdevice.
- As the drivers are loaded and removed, a particular device may get
offloaded and unoffloaded again. At the same time, the statistics values
need to stay monotonic (modulo the eventual 64-bit wraparound),
increasing only to reflect traffic measured in the device.
To that end, the netdevice keeps around a lazily-allocated copy of struct
rtnl_link_stats64. Device drivers then contribute to the values kept
therein at various points. Even as the driver goes away, the struct stays
around to maintain the statistics values.
- Different HW devices may be able to count different things. The
motivation behind this patch in particular is exposure of HW counters on
Nvidia Spectrum switches, where the only practical approach to counting
traffic on offloaded soft netdevices currently is to use router interface
counters, and count L3 traffic. Correspondingly that is the statistics
suite added in this patch.
Other devices may be able to measure different kinds of traffic, and for
that reason, the APIs are built to allow uniform access to different
statistics suites.
- Because soft netdevices and offloading drivers are only loosely bound, a
netdevice uses a notifier chain to communicate with the drivers. Several
new notifiers, NETDEV_OFFLOAD_XSTATS_*, have been added to carry messages
to the offloading drivers.
- Devices can have various conditions for when a particular counter is
available. As the device is configured and reconfigured, the device
offload may become or cease being suitable for counter binding. A
netdevice can use a notifier type NETDEV_OFFLOAD_XSTATS_REPORT_USED to
ping offloading drivers and determine whether anyone currently implements
a given statistics suite. This information can then be propagated to user
space.
When the driver decides to unoffload a netdevice, it can use a
newly-added function, netdev_offload_xstats_report_delta(), to record
outstanding collected statistics, before destroying the HW counter.
This patch adds a helper, call_netdevice_notifiers_info_robust(), for
dispatching a notifier with the possibility of unwind when one of the
consumers bails. Given the wish to eventually get rid of the global
notifier block altogether, this helper only invokes the per-netns notifier
block.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Obtaining stats for the IFLA_STATS_LINK_OFFLOAD_XSTATS nest involves a HW
access, and can fail for more reasons than just netlink message size
exhaustion. Therefore do not always return -EMSGSIZE on the failure path,
but respect the error code provided by the callee. Set the error explicitly
where it is reasonable to assume -EMSGSIZE as the failure reason.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Later patches add handlers for more HW-backed statistics. An extack will be
useful when communicating HW / driver errors to the client. Add the
arguments as appropriate.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The filter_mask field of RTM_GETSTATS header determines which top-level
attributes should be included in the netlink response. This saves
processing time by only including the bits that the user cares about
instead of always dumping everything. This is doubly important for
HW-backed statistics that would typically require a trip to the device to
fetch the stats.
So far there was only one HW-backed stat suite per attribute. However,
IFLA_STATS_LINK_OFFLOAD_XSTATS is a nest, and will gain a new stat suite in
the following patches. It would therefore be advantageous to be able to
filter within that nest, and select just one or the other HW-backed
statistics suite.
Extend rtnetlink so that RTM_GETSTATS permits attributes in the payload.
The scheme is as follows:
- RTM_GETSTATS
- struct if_stats_msg
- attr nest IFLA_STATS_GET_FILTERS
- attr IFLA_STATS_LINK_OFFLOAD_XSTATS
- u32 filter_mask
This scheme reuses the existing enumerators by nesting them in a dedicated
context attribute. This is covered by policies as usual, therefore a
gradual opt-in is possible. Currently only IFLA_STATS_LINK_OFFLOAD_XSTATS
nest has filtering enabled, because for the SW counters the issue does not
seem to be that important.
rtnl_offload_xstats_get_size() and _fill() are extended to observe the
requested filters.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The IFLA_STATS_LINK_OFFLOAD_XSTATS attribute is a nest whose child
attributes carry various special hardware statistics. The code that handles
this nest was written with the idea that all these statistics would be
exposed by the device driver of a physical netdevice.
In the following patches, a new attribute is added to the abovementioned
nest, which however can be defined for some soft netdevices. The NDO-based
approach to querying these does not work, because it is not the soft
netdevice driver that exposes these statistics, but an offloading NIC
driver that does so.
The current code does not scale well to this usage. Simply rewrite it back
to the pattern seen in other fill-like and get_size-like functions
elsewhere.
Extract to helpers the code that is concerned with handling specifically
NDO-backed statistics so that it can be easily reused should more such
statistics be added.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The currently used names rtnl_get_offload_stats() and
rtnl_get_offload_stats_size() do not clearly show the namespace. The former
function additionally seems to have been named this way in accordance with
the NDO name, as opposed to the naming used in the rtnetlink.c file (and
indeed elsewhere in the netlink handling code). As more and
differently-flavored attributes are introduced, a common clear prefix is
needed for all related functions.
Rename the functions to follow the rtnl_offload_xstats_* naming scheme.
Signed-off-by: Petr Machata <petrm@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Adds a function page_pool_get_stats which can be used by drivers to obtain
stats for a specified page_pool.
Signed-off-by: Joe Damato <jdamato@fastly.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add per-cpu stats tracking page pool recycling events:
- cached: recycling placed page in the page pool cache
- cache_full: page pool cache was full
- ring: page placed into the ptr ring
- ring_full: page released from page pool because the ptr ring was full
- released_refcnt: page released (and not recycled) because refcnt > 1
Signed-off-by: Joe Damato <jdamato@fastly.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add per-pool statistics counters for the allocation path of a page pool.
These stats are incremented in softirq context, so no locking or per-cpu
variables are needed.
This code is disabled by default and a kernel config option is provided for
users who wish to enable them.
The statistics added are:
- fast: successful fast path allocations
- slow: slow path order-0 allocations
- slow_high_order: slow path high order allocations
- empty: ptr ring is empty, so a slow path allocation was forced.
- refill: an allocation which triggered a refill of the cache
- waive: pages obtained from the ptr ring that cannot be added to
the cache due to a NUMA mismatch.
Signed-off-by: Joe Damato <jdamato@fastly.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Network drivers such as igb or igc call eth_get_headlen() to determine the
header length for their to be constructed skbs in receive path.
When running HSR on top of these drivers, it results in triggering BUG_ON() in
skb_pull(). The reason is the skb headlen is not sufficient for HSR to work
correctly. skb_pull() notices that.
For instance, eth_get_headlen() returns 14 bytes for TCP traffic over HSR which
is not correct. The problem is, the flow dissection code does not take HSR into
account. Therefore, add support for it.
Reported-by: Anthony Harivel <anthony.harivel@linutronix.de>
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Link: https://lore.kernel.org/r/20220228195856.88187-1-kurt@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Eliminate the following coccicheck warning:
./net/openvswitch/flow.c:379:2-3: Unneeded semicolon
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220227132208.24658-1-yang.lee@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Add extack message to return exact message to user when adding invalid
filter with conflict flags for TC action.
In previous implement we just return EINVAL which is confusing for user.
Signed-off-by: Baowen Zheng <baowen.zheng@corigine.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Link: https://lore.kernel.org/r/1646191769-17761-1-git-send-email-baowen.zheng@corigine.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
git://git.open-mesh.org/linux-merge
Simon Wunderlich says:
====================
This cleanup patchset includes the following patches:
- bump version strings, by Simon Wunderlich
- Remove redundant 'flush_workqueue()' calls, by Christophe JAILLET
- Migrate to linux/container_of.h, by Sven Eckelmann
- Demote batadv-on-batadv skip error message, by Sven Eckelmann
* tag 'batadv-next-pullrequest-20220302' of git://git.open-mesh.org/linux-merge:
batman-adv: Demote batadv-on-batadv skip error message
batman-adv: Migrate to linux/container_of.h
batman-adv: Remove redundant 'flush_workqueue()' calls
batman-adv: Start new development cycle
====================
Link: https://lore.kernel.org/r/20220302163522.102842-1-sw@simonwunderlich.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
The error message "Cannot find parent device" was shown for users of
macvtap (on batadv devices) whenever the macvtap was moved to a different
netns. This happens because macvtap doesn't provide an implementation for
rtnl_link_ops->get_link_net.
The situation for which this message is printed is actually not an error
but just a warning that the optional sanity check was skipped. So demote
the message from error to warning and adjust the text to better explain
what happened.
Reported-by: Leonardo Mörlein <freifunk@irrelefant.net>
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
|
|
The commit d2a8ebbf8192 ("kernel.h: split out container_of() and
typeof_member() macros") introduced a new header for the container_of
related macros from (previously) linux/kernel.h.
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Simon Wunderlich <sw@simonwunderlich.de>
|
|
Fix build:
include/linux/minmax.h:45:25: note: in expansion of macro ‘__careful_cmp’
45 | #define min(x, y) __careful_cmp(x, y, <)
| ^~~~~~~~~~~~~
net/smc/smc_tx.c:150:24: note: in expansion of macro ‘min’
150 | corking_size = min(sock_net(&smc->sk)->smc.sysctl_autocorking_size,
| ^~~
Fixes: 12bbb0d163a9 ("net/smc: add sysctl for autocorking")
Link: https://lore.kernel.org/r/20220301222446.1271127-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
|
Send data all the way down to the RDMA device is a time
consuming operation(get a new slot, maybe do RDMA Write
and send a CDC, etc). Moving those operations from BH
to user context is good for performance.
If the sock_lock is hold by user, we don't try to send
data out in the BH context, but just mark we should
send. Since the user will release the sock_lock soon, we
can do the sending there.
Add smc_release_cb() which will be called in release_sock()
and try send in the callback if needed.
This patch moves the sending part out from BH if sock lock
is hold by user. In my testing environment, this saves about
20% softirq in the qperf 4K tcp_bw test in the sender side
with no noticeable throughput drop.
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When we are handling softirq workload, enable hardirq may
again interrupt the current routine of softirq, and then
try to raise softirq again. This only wastes CPU cycles
and won't have any real gain.
Since IB_CQ_REPORT_MISSED_EVENTS already make sure if
ib_req_notify_cq() returns 0, it is safe to wait for the
next event, with no need to poll the CQ again in this case.
This patch disables hardirq during the processing of softirq,
and re-arm the CQ after softirq is done. Somehow like NAPI.
Co-developed-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
rmbe_update_limit is used to limit announcing receive
window updating too frequently. RFC7609 request a minimal
increase in the window size of 10% of the receive buffer
space. But current implementation used:
min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2)
and SOCK_MIN_SNDBUF / 2 == 2304 Bytes, which is almost
always less then 10% of the receive buffer space.
This causes the receiver always sending CDC message to
update its consumer cursor when it consumes more then 2K
of data. And as a result, we may encounter something like
"TCP silly window syndrome" when sending 2.5~8K message.
This patch fixes this using max(rmbe_size / 10, SOCK_MIN_SNDBUF / 2).
With this patch and SMC autocorking enabled, qperf 2K/4K/8K
tcp_bw test shows 45%/75%/40% increase in throughput respectively.
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In commit ea785a1a573b("net/smc: Send directly when
TCP_CORK is cleared"), we don't use delayed work
to implement cork.
This patch use the same algorithm, removes the
delayed work when setting TCP_NODELAY and send
directly in setsockopt(). This also makes the
TCP_NODELAY the same as TCP.
Cc: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This add a new sysctl: net.smc.autocorking_size
We can dynamically change the behaviour of autocorking
by change the value of autocorking_size.
Setting to 0 disables autocorking in SMC
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch adds autocorking support for SMC which could improve
throughput for small message by x3+.
The main idea is borrowed from TCP autocorking with some RDMA
specific modification:
1. The first message should never cork to make sure we won't
bring extra latency
2. If we have posted any Tx WRs to the NIC that have not
completed, cork the new messages until:
a) Receive CQE for the last Tx WR
b) We have corked enough message on the connection
3. Try to push the corked data out when we receive CQE of
the last Tx WR to prevent the corked messages hang in
the send queue.
Both SMC autocorking and TCP autocorking check the TX completion
to decide whether we should cork or not. The difference is
when we got a SMC Tx WR completion, the data have been confirmed
by the RNIC while TCP TX completion just tells us the data
have been sent out by the local NIC.
Add an atomic variable tx_pushing in smc_connection to make
sure only one can send to let it cork more and save CDC slot.
SMC autocorking should not bring extra latency since the first
message will always been sent out immediately.
The qperf tcp_bw test shows more than x4 increase under small
message size with Mellanox connectX4-Lx, same result with other
throughput benchmarks like sockperf/netperf.
The qperf tcp_lat test shows SMC autocorking has not increase any
ping-pong latency.
Test command:
client: smc_run taskset -c 1 qperf smc-server -oo msg_size:1:64K:*2 \
-t 30 -vu tcp_{bw|lat}
server: smc_run taskset -c 1 qperf
=== Bandwidth ====
MsgSize(Bytes) SMC-NoCork TCP SMC-AutoCorking
1 0.578 MB/s 2.392 MB/s(313.57%) 2.647 MB/s(357.72%)
2 1.159 MB/s 4.780 MB/s(312.53%) 5.153 MB/s(344.71%)
4 2.283 MB/s 10.266 MB/s(349.77%) 10.363 MB/s(354.02%)
8 4.668 MB/s 19.040 MB/s(307.86%) 21.215 MB/s(354.45%)
16 9.147 MB/s 38.904 MB/s(325.31%) 41.740 MB/s(356.32%)
32 18.369 MB/s 79.587 MB/s(333.25%) 82.392 MB/s(348.52%)
64 36.562 MB/s 148.668 MB/s(306.61%) 161.564 MB/s(341.89%)
128 72.961 MB/s 274.913 MB/s(276.80%) 325.363 MB/s(345.94%)
256 144.705 MB/s 512.059 MB/s(253.86%) 633.743 MB/s(337.96%)
512 288.873 MB/s 884.977 MB/s(206.35%) 1250.681 MB/s(332.95%)
1024 574.180 MB/s 1337.736 MB/s(132.98%) 2246.121 MB/s(291.19%)
2048 1095.192 MB/s 1865.952 MB/s( 70.38%) 2057.767 MB/s( 87.89%)
4096 2066.157 MB/s 2380.337 MB/s( 15.21%) 2173.983 MB/s( 5.22%)
8192 3717.198 MB/s 2733.073 MB/s(-26.47%) 3491.223 MB/s( -6.08%)
16384 4742.221 MB/s 2958.693 MB/s(-37.61%) 4637.692 MB/s( -2.20%)
32768 5349.550 MB/s 3061.285 MB/s(-42.77%) 5385.796 MB/s( 0.68%)
65536 5162.919 MB/s 3731.408 MB/s(-27.73%) 5223.890 MB/s( 1.18%)
==== Latency ====
MsgSize(Bytes) SMC-NoCork TCP SMC-AutoCorking
1 10.540 us 11.938 us( 13.26%) 10.573 us( 0.31%)
2 10.996 us 11.992 us( 9.06%) 10.269 us( -6.61%)
4 10.229 us 11.687 us( 14.25%) 10.240 us( 0.11%)
8 10.203 us 11.653 us( 14.21%) 10.402 us( 1.95%)
16 10.530 us 11.313 us( 7.44%) 10.599 us( 0.66%)
32 10.241 us 11.586 us( 13.13%) 10.223 us( -0.18%)
64 10.693 us 11.652 us( 8.97%) 10.251 us( -4.13%)
128 10.597 us 11.579 us( 9.27%) 10.494 us( -0.97%)
256 10.409 us 11.957 us( 14.87%) 10.710 us( 2.89%)
512 11.088 us 12.505 us( 12.78%) 10.547 us( -4.88%)
1024 11.240 us 12.255 us( 9.03%) 10.787 us( -4.03%)
2048 11.485 us 16.970 us( 47.76%) 11.256 us( -1.99%)
4096 12.077 us 13.948 us( 15.49%) 12.230 us( 1.27%)
8192 13.683 us 16.693 us( 22.00%) 13.786 us( 0.75%)
16384 16.470 us 23.615 us( 43.38%) 16.459 us( -0.07%)
32768 22.540 us 40.966 us( 81.75%) 23.284 us( 3.30%)
65536 34.192 us 73.003 us(113.51%) 34.233 us( 0.12%)
With SMC autocorking support, we can archive better throughput
than TCP in most message sizes without any latency trade-off.
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch add sysctl interface to support container environment
for SMC as we talk in the mail list.
Link: https://lore.kernel.org/netdev/20220224020253.GF5443@linux.alibaba.com
Co-developed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Use the helper function time_is_{before,after}_jiffies() to improve
code readability.
Signed-off-by: Wang Qing <wangqing@vivo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Both rps_sock_flow_sysctl() and flow_limit_cpu_sysctl()
are using synchronize_rcu() right before freeing memory
either by vfree() or kfree()
They can switch to kvfree_rcu(ptr) and kfree_rcu(ptr) to benefit
from asynchronous mode, instead of blocking the current thread.
Note that kvfree_rcu(ptr) and kfree_rcu(ptr) eventually can
have to use synchronize_rcu() in some memory pressure cases.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This also calls trace_smc_tx_sendmsg() even if data is corked. For ease
of understanding, if statements are not expanded here.
Link: https://lore.kernel.org/all/f4166712-9a1e-51a0-409d-b7df25a66c52@linux.ibm.com/
Fixes: 139653bc6635 ("net/smc: Remove corked dealyed work")
Suggested-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The current police offload action entry is missing exceed/notexceed
actions and parameters that can be configured by tc police action.
Add the missing parameters as a pre-step for offloading police actions
to hardware.
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
As FDB isolation cannot be enforced between VLAN-aware bridges in lack
of hardware assistance like extra FID bits, it seems plausible that many
DSA switches cannot do it. Therefore, they need to reject configurations
with multiple VLAN-aware bridges from the two code paths that can
transition towards that state:
- joining a VLAN-aware bridge
- toggling VLAN awareness on an existing bridge
The .port_vlan_filtering method already propagates the netlink extack to
the driver, let's propagate it from .port_bridge_join too, to make sure
that the driver can use the same function for both.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For DSA, to encourage drivers to perform FDB isolation simply means to
track which bridge does each FDB and MDB entry belong to. It then
becomes the driver responsibility to use something that makes the FDB
entry from one bridge not match the FDB lookup of ports from other
bridges.
The top-level functions where the bridge is determined are:
- dsa_port_fdb_{add,del}
- dsa_port_host_fdb_{add,del}
- dsa_port_mdb_{add,del}
- dsa_port_host_mdb_{add,del}
aka the pre-crosschip-notifier functions.
Changing the API to pass a reference to a bridge is not superfluous, and
looking at the passed bridge argument is not the same as having the
driver look at dsa_to_port(ds, port)->bridge from the ->port_fdb_add()
method.
DSA installs FDB and MDB entries on shared (CPU and DSA) ports as well,
and those do not have any dp->bridge information to retrieve, because
they are not in any bridge - they are merely the pipes that serve the
user ports that are in one or multiple bridges.
The struct dsa_bridge associated with each FDB/MDB entry is encapsulated
in a larger "struct dsa_db" database. Although only databases associated
to bridges are notified for now, this API will be the starting point for
implementing IFF_UNICAST_FLT in DSA. There, the idea is to install FDB
entries on the CPU port which belong to the corresponding user port's
port database. These are supposed to match only when the port is
standalone.
It is better to introduce the API in its expected final form than to
introduce it for bridges first, then to have to change drivers which may
have made one or more assumptions.
Drivers can use the provided bridge.num, but they can also use a
different numbering scheme that is more convenient.
DSA must perform refcounting on the CPU and DSA ports by also taking
into account the bridge number. So if two bridges request the same local
address, DSA must notify the driver twice, once for each bridge.
In fact, if the driver supports FDB isolation, DSA must perform
refcounting per bridge, but if the driver doesn't, DSA must refcount
host addresses across all bridges, otherwise it would be telling the
driver to delete an FDB entry for a bridge and the driver would delete
it for all bridges. So introduce a bool fdb_isolation in drivers which
would make all bridge databases passed to the cross-chip notifier have
the same number (0). This makes dsa_mac_addr_find() -> dsa_db_equal()
say that all bridge databases are the same database - which is
essentially the legacy behavior.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The dsa_8021q_bridge_tx_fwd_offload_vid is no longer used just for
bridge TX forwarding offload, it is the private VLAN reserved for
VLAN-unaware bridging in a way that is compatible with FDB isolation.
So just rename it dsa_tag_8021q_bridge_vid.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In the old Shared VLAN Learning mode of operation that tag_8021q
previously used for forwarding, we needed to have distinct concepts for
an RX and a TX VLAN.
An RX VLAN could be installed on all ports that were members of a given
bridge, so that autonomous forwarding could still work, while a TX VLAN
was dedicated for precise packet steering, so it just contained the CPU
port and one egress port.
Now that tag_8021q uses Independent VLAN Learning and imprecise RX/TX
all over, those lines have been blurred and we no longer have the need
to do precise TX towards a port that is in a bridge. As for standalone
ports, it is fine to use the same VLAN ID for both RX and TX.
This patch changes the tag_8021q format by shifting the VLAN range it
reserves, and halving it. Previously, our DIR bits were encoding the
VLAN direction (RX/TX) and were set to either 1 or 2. This meant that
tag_8021q reserved 2K VLANs, or 50% of the available range.
Change the DIR bits to a hardcoded value of 3 now, which makes tag_8021q
reserve only 1K VLANs, and a different range now (the last 1K). This is
done so that we leave the old format in place in case we need to return
to it.
In terms of code, the vid_is_dsa_8021q_rxvlan and vid_is_dsa_8021q_txvlan
functions go away. Any vid_is_dsa_8021q is both a TX and an RX VLAN, and
they are no longer distinct. For example, felix which did different
things for different VLAN types, now needs to handle the RX and the TX
logic for the same VLAN.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The sja1105 switch can't populate the PORT field of the tag_8021q header
when sending a frame to the CPU with a non-zero VBID.
Similar to dsa_find_designated_bridge_port_by_vid() which performs
imprecise RX for VLAN-aware bridges, let's introduce a helper in
tag_8021q for performing imprecise RX based on the VLAN that it has
allocated for a VLAN-unaware bridge.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For VLAN-unaware bridging, tag_8021q uses something perhaps a bit too
tied with the sja1105 switch: each port uses the same pvid which is also
used for standalone operation (a unique one from which the source port
and device ID can be retrieved when packets from that port are forwarded
to the CPU). Since each port has a unique pvid when performing
autonomous forwarding, the switch must be configured for Shared VLAN
Learning (SVL) such that the VLAN ID itself is ignored when performing
FDB lookups. Without SVL, packets would always be flooded, since FDB
lookup in the source port's VLAN would never find any entry.
First of all, to make tag_8021q more palatable to switches which might
not support Shared VLAN Learning, let's just use a common VLAN for all
ports that are under the same bridge.
Secondly, using Shared VLAN Learning means that FDB isolation can never
be enforced. But if all ports under the same VLAN-unaware bridge share
the same VLAN ID, it can.
The disadvantage is that the CPU port can no longer perform precise
source port identification for these packets. But at least we have a
mechanism which has proven to be adequate for that situation: imprecise
RX (dsa_find_designated_bridge_port_by_vid), which is what we use for
termination on VLAN-aware bridges.
The VLAN ID that VLAN-unaware bridges will use with tag_8021q is the
same one as we were previously using for imprecise TX (bridge TX
forwarding offload). It is already allocated, it is just a matter of
using it.
Note that because now all ports under the same bridge share the same
VLAN, the complexity of performing a tag_8021q bridge join decreases
dramatically. We no longer have to install the RX VLAN of a newly
joining port into the port membership of the existing bridge ports.
The newly joining port just becomes a member of the VLAN corresponding
to that bridge, and the other ports are already members of it from when
they joined the bridge themselves. So forwarding works properly.
This means that we can unhook dsa_tag_8021q_bridge_{join,leave} from the
cross-chip notifier level dsa_switch_bridge_{join,leave}. We can put
these calls directly into the sja1105 driver.
With this new mode of operation, a port controlled by tag_8021q can have
two pvids whereas before it could only have one. The pvid for standalone
operation is different from the pvid used for VLAN-unaware bridging.
This is done, again, so that FDB isolation can be enforced.
Let tag_8021q manage this by deleting the standalone pvid when a port
joins a bridge, and restoring it when it leaves it.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When neighbour become invalid or destroyed, neigh_invalidate() will be
called. neigh->ops->error_report() will be called if the neighbour's
state is NUD_FAILED, and seems here is the only use of error_report().
So we can tell that the reason of skb drops in arp_error_report() is
SKB_DROP_REASON_NEIGH_FAILED.
Replace kfree_skb() used in arp_error_report() with kfree_skb_reason().
Reviewed-by: Mengen Sun <mengensun@tencent.com>
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
