Add a selftest for RED early_drop and mark qevents when a trap action is
attached at the associated block.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When offloading action trap on a qevent, pass to_dev of NULL to the SPAN
module to trigger the mirror to the CPU port. Query the buffer drops
policer and use it for policing of the trapped traffic.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When unsetting policer base, the SPAN code currently uses refcount_dec().
However that function splats when the counter reaches zero, because
reaching zero without actually testing is in general indicative of a
missing cleanup. There is no cleanup to be done here, but nonetheless, use
refcount_dec_and_test() as required.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Mirred currently does not mix well with blocks executed after the qdisc
root lock is taken. This includes classification blocks (such as in PRIO,
ETS, DRR qdiscs) and qevents. The locking caused by the packet mirrored by
mirred can cause deadlocks: either when the thread of execution attempts to
take the lock a second time, or when two threads end up waiting on each
other's locks.
The qevent patchset attempted to not introduce further badness of this
sort, and dropped the lock before executing the qevent block. However this
lead to too little locking and races between qdisc configuration and packet
enqueue in the RED qdisc.
Before the deadlock issues are solved in a way that can be applied across
many qdiscs reasonably easily, do for qevents what is done for the
classification blocks and just keep holding the root lock.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Jakub Kicinski <[email protected]>
Add a selftest for offloading a mirror action attached to the block
associated with RED early_drop qevent.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The RED qevents early_drop and mark can be offloaded under the following
fairly strict conditions:
- At most one filter is configured at the qevent block
- The protocol is "any"
- The classifier is matchall
- The action is trap, sample, or mirror with the same conditions as
with other SPAN offloads
- The hw_counters type is none
In this patchset, implement offload of mirror for early_drop qevent.
The ECN trigger is currently not implemented in the FW and therefore
the mark qevent is not supported.
The qevent notifications look exactly like regular block binding
notifications with a binder type that identifies them as qevents.
Therefore the details of processing this binding are fairly similar
to the matchall offload.
struct flow_block_offload.sch points at the qdisc in question. Use it to
figure out if the qdisc is offloaded at all and what TC it configures.
Bounce bindings on not-offloaded qdiscs.
Individual bindings are kept in a list so that several qevents can share
the same block and all binding points get configured as the configured
filters change.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Two RED qevents have been introduced recently. From the point of view of a
driver, qevents are simply blocks with unusual binder types. However they
need to be handled by different logic than ACL-like flows.
Thus rename mlxsw_sp_setup_tc_block() to mlxsw_sp_setup_tc_block_clsact()
and move the binder-type dispatch from there to spectrum.c into a new
function of the original name. The new dispatcher is easier to extend with
new binder types.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
A following patch introduces offloading of filters attached to blocks bound
to the RED tail_drop qevent. The only classifier that mlxsw will permit in
this role is matchall. mlxsw currently offloads matchall filters used with
clsact qdisc. The data structures used for that offload will come handy for
the qevent offload as well. Publish them in spectrum.h.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The field "dev" in struct mlxsw_sp_flow_block_binding is not used. Drop it.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
No clean-up is performed at the target label of this goto. Convert it to a
direct return.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Previously, shared blocks were only relevant for the pseudo-qdiscs ingress
and clsact. Recently, a qevent facility was introduced, which allows to
bind blocks to well-defined slots of a qdisc instance. RED in particular
got two qevents: early_drop and mark. Drivers that wish to offload these
blocks will be sent the usual notification, and need to know which qdisc it
is related to.
To that end, extend flow_block_offload with a "sch" pointer, and initialize
as appropriate. This prompts changes in the indirect block facility, which
now tracks the scheduler in addition to the netdevice. Update signatures of
several functions similarly.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This test is inspired by the mlxsw RED selftest. It is much simpler to set
up (also because there is no point in testing PRIO / RED encapsulation). It
tests bare RED, ECN and ECN+nodrop modes of operation. On top of that it
tests RED early_drop and mark qevents.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In order to allow acting on dropped and/or ECN-marked packets, add two new
qevents to the RED qdisc: "early_drop" and "mark". Filters attached at
"early_drop" block are executed as packets are early-dropped, those
attached at the "mark" block are executed as packets are ECN-marked.
Two new attributes are introduced: TCA_RED_EARLY_DROP_BLOCK with the block
index for the "early_drop" qevent, and TCA_RED_MARK_BLOCK for the "mark"
qevent. Absence of these attributes signifies "don't care": no block is
allocated in that case, or the existing blocks are left intact in case of
the change callback.
For purposes of offloading, blocks attached to these qevents appear with
newly-introduced binder types, FLOW_BLOCK_BINDER_TYPE_RED_EARLY_DROP and
FLOW_BLOCK_BINDER_TYPE_RED_MARK.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In the following patches, RED will get two qevents. The implementation will
be clearer if the callback for change is not a pure subset of the callback
for init. Split the two and promote attribute parsing to the callbacks
themselves from the common code, because it will be handy there.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Qevents are attach points for TC blocks, where filters can be put that are
executed when "interesting events" take place in a qdisc. The data to keep
and the functions to invoke to maintain a qevent will be largely the same
between qevents. Therefore introduce sched-wide helpers for qevent
management.
Currently, similarly to ingress and egress blocks of clsact pseudo-qdisc,
blocks attachment cannot be changed after the qdisc is created. To that
end, add a helper tcf_qevent_validate_change(), which verifies whether
block index attribute is not attached, or if it is, whether its value
matches the current one (i.e. there is no material change).
The function tcf_qevent_handle() should be invoked when qdisc hits the
"interesting event" corresponding to a block. This function releases root
lock for the duration of executing the attached filters, to allow packets
generated through user actions (notably mirred) to be reinserted to the
same qdisc tree.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
A following patch introduces qevents, points in qdisc algorithm where
packet can be processed by user-defined filters. Should this processing
lead to a situation where a new packet is to be enqueued on the same port,
holding the root lock would lead to deadlocks. To solve the issue, qevent
handler needs to unlock and relock the root lock when necessary.
To that end, add the root lock argument to the qdisc op enqueue, and
propagate throughout.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In a fashion similar to the other Spectrum systems, enforce a specific
firmware version for Spectrum-3 so that the driver and firmware are
always in sync with regards to new features.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This version comes with fixes to the following problems, among others:
- Wrong shaper configuration on Spectrum-1
- Bogus temperature reading on Spectrum-2
- Problems in setting egress buffer size after MTU change on Spectrum-2
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a test that checks that pedit adjusts port numbers of tcp and udp
packets.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Spectrum-2 supports an ACL action L4_PORT, which allows TCP and UDP source
and destination port number change. Offload suitable mangles to this
action.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add fields related to L4_PORT_ACTION, which is used for changing of TCP and
UDP port numbers.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Certain ACL actions are only available on some Spectrum revisions. In
particular, L4_PORT_ACTION is not available on Spectrum-1. Introduce a
new ops struct intended to hold these differences, mlxsw_sp_rulei_ops.
Prime it with a sole member, act_mangle_field, meant for handling of
pedit mangles.
Create two ops structures, one for Spectrum-1, the other for Spectrum-2
and above. Add callbacks for act_mangle_field and dispatch to the common
handler.
Invoke mlxsw_sp_rulei_ops.act_mangle_field from the field mangler
instead of calling the common handler directly.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
A missing stats_update callback was recently added to act_pedit. Now that
iproute2 supports JSON dumping for pedit, extend the pedit_dsfield selftest
with a check that would have caught the fact that the callback was missing.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Using ping in tests is error-prone, because ping is too smart. On a
flaky system (notably in a simulator), when packets don't come quickly
enough, more pings are sent, and that throws off counters. Instead use
mausezahn to generate ICMP echo request packets. That allows us to
send them in quicker succession as well, because the reason the ping
was made slow in the first place was to make the tests work on
simulated systems.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a self-test for the IPv6 dsfield munge that iproute2 will support.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Extend the pedit_dsfield forwarding selftest with coverage of "pedit ex
munge ip6 dsfield set".
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The handler for FLOW_ACTION_VLAN_MANGLE ends by returning whatever the
lower-level function that it calls returns. If there are more actions lined
up after this action, those are never offloaded. Fix by only bailing out
when the called function returns an error.
Fixes: a150201a70 ("mlxsw: spectrum: Add support for vlan modify TC action")
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The handler for FLOW_ACTION_PRIORITY ends by returning whatever the
lower-level function that it calls returns. If there are more actions lined
up after this action, those are never offloaded. Fix by only bailing out
when the called function returns an error.
Fixes: 463957e3fb ("mlxsw: spectrum_flower: Offload FLOW_ACTION_PRIORITY")
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Currently the test checks the observable effect of skbedit priority:
queueing of packets at the correct qdisc band. It therefore misses the fact
that the counters for offloaded rules are not updated. Add an extra check
for the counter.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Implement this callback in order to get the offloaded stats added to the
kernel stats.
Reported-by: Alexander Petrovskiy <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Implement this callback in order to get the offloaded stats added to the
kernel stats.
Reported-by: Alexander Petrovskiy <[email protected]>
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When DSCP is updated through an offloaded pedit action, DSCP rewrite on
egress should be disabled. Add a test that check that it is so.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a test that runs packets with dsfield set, and test that pedit adjusts
the DSCP or ECN parts or the whole field.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Offload action pedit ex munge when used with a flower classifier. Only
allow setting of DSCP, ECN, or the whole DSField in IPv4 and IPv6 packets.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The QOS_ACTION is used for manipulating the QOS attributes of the packet.
Add the defines and helpers related to DSCP and ECN fields, and dscp_rw.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The original idea was to reuse this set of actions for ECN rewrite as well,
but on second look, it's not such a great idea. These two items should each
have its own command. Rename the existing enum to make it obvious that it
belongs to switch_prio_cmd.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This field references FLOW_ACTION_PACKET_EDIT. Such action does not exist
though. Instead the field is used for FLOW_ACTION_MANGLE and _ADD.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a test that runs traffic through a port such that skbedit priority
action acts on it during forwarding. Test that at egress, it is classified
correctly according to the new priority at a PRIO qdisc.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Offload action skbedit priority when keyed to a flower classifier. The
skb->priority field in Linux is very generic, so only allow setting the
bottom 8 priorities and bounce anything else.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The QOS_ACTION is used for manipulating the QoS attributes of a packet.
Add the corresponding defines and helpers, in particular for the
switch_priority override.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The skbedit action "priority" is used for adjusting SKB priority. Allow
drivers to offload the action by introducing two new skbedit getters and a
new flow action, and initializing appropriately in tc_setup_flow_action().
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The two functions is_tcf_skbedit_mark() and is_tcf_skbedit_ptype() have a
very similar structure. A follow-up patch will add one more such function.
Instead of more cut'n'pasting, extract a helper function that checks
whether a TC action is an skbedit with the required flag. Convert the two
existing functions into thin wrappers around the helper.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In the commit referenced below, hw_stats_type of an entry is set for every
entry that corresponds to a pedit action. However, the assignment is only
done after the entry pointer is bumped, and therefore could overwrite
memory outside of the entries array.
The reason for this positioning may have been that the current entry's
hw_stats_type is already set above, before the action-type dispatch.
However, if there are no more actions, the assignment is wrong. And if
there are, the next round of the for_each_action loop will make the
assignment before the action-type dispatch anyway.
Therefore fix this issue by simply reordering the two lines.
Fixes: 74522e7baa ("net: sched: set the hw_stats_type in pedit loop")
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The fix referenced below causes a crash when an ERSPAN tunnel is created
without passing IFLA_INFO_DATA. Fix by validating passed-in data in the
same way as ipgre does.
Fixes: e1f8f78ffe ("net: ip_gre: Separate ERSPAN newlink / changelink callbacks")
Reported-by: [email protected]
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
ERSPAN shares most of the code path with GRE and gretap code. While that
helps keep the code compact, it is also error prone. Currently a broken
userspace can turn a gretap tunnel into a de facto ERSPAN one by passing
IFLA_GRE_ERSPAN_VER. There has been a similar issue in ip6gretap in the
past.
To prevent these problems in future, split the newlink and changelink code
paths. Split the ERSPAN code out of ipgre_netlink_parms() into a new
function erspan_netlink_parms(). Extract a piece of common logic from
ipgre_newlink() and ipgre_changelink() into ipgre_newlink_encap_setup().
Add erspan_newlink() and erspan_changelink().
Fixes: 84e54fe0a5 ("gre: introduce native tunnel support for ERSPAN")
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Extend RED testsuite to cover the new nodrop mode of RED-ECN. This test is
really similar to ECN test, diverging only in the last step, where UDP
traffic should go to backlog instead of being dropped. Thus extract a
common helper, ecn_test_common(), make do_ecn_test() into a relatively
simple wrapper, and add another one, do_ecn_nodrop_test().
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
RED ECN nodrop mode means that non-ECT traffic should not be early-dropped,
but enqueued normally instead. In Spectrum systems, this is achieved by
disabling CWTPM.ew (enable WRED) for a given traffic class.
So far CWTPM.ew was unconditionally enabled. Instead disable it when the
RED qdisc is in nodrop mode.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When the RED Qdisc is currently configured to enable ECN, the RED algorithm
is used to decide whether a certain SKB should be marked. If that SKB is
not ECN-capable, it is early-dropped.
It is also possible to keep all traffic in the queue, and just mark the
ECN-capable subset of it, as appropriate under the RED algorithm. Some
switches support this mode, and some installations make use of it.
To that end, add a new RED flag, TC_RED_NODROP. When the Qdisc is
configured with this flag, non-ECT traffic is enqueued instead of being
early-dropped.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jakub Kicinski <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The qdiscs RED, GRED, SFQ and CHOKE use different subsets of the same pool
of global RED flags. These are passed in tc_red_qopt.flags. However none of
these qdiscs validate the flag field, and just copy it over wholesale to
internal structures, and later dump it back. (An exception is GRED, which
does validate for VQs -- however not for the main setup.)
A broken userspace can therefore configure a qdisc with arbitrary
unsupported flags, and later expect to see the flags on qdisc dump. The
current ABI therefore allows storage of several bits of custom data to
qdisc instances of the types mentioned above. How many bits, depends on
which flags are meaningful for the qdisc in question. E.g. SFQ recognizes
flags ECN and HARDDROP, and the rest is not interpreted.
If SFQ ever needs to support ADAPTATIVE, it needs another way of doing it,
and at the same time it needs to retain the possibility to store 6 bits of
uninterpreted data. Likewise RED, which adds a new flag later in this
patchset.
To that end, this patch adds a new function, red_get_flags(), to split the
passed flags of RED-like qdiscs to flags and user bits, and
red_validate_flags() to validate the resulting configuration. It further
adds a new attribute, TCA_RED_FLAGS, to pass arbitrary flags.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jakub Kicinski <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a handful of tests for creating RED with different flags.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Roman Mashak <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Currently the SW-datapath ETS selftests use "ip link" stats to obtain the
number of packets that went through a given band. mlxsw then uses ethtool
per-priority counters.
Instead, change both to use qdiscs. In SW datapath this is the obvious
choice, and now that mlxsw offloads FIFO, this should work on the offloaded
datapath as well. This has the effect of verifying that the FIFO offload
works.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
There are two peculiarities about offloading FIFO:
- sometimes the qdisc has an unspecified handle (it is "invisible")
- it may be created before the qdisc that it will be a child of
These features make the offload a bit more tricky. The approach chosen in
this patch is to make note of all the FIFOs that needed to be rejected
because their parents were not known. Later when the parent is created,
they are offloaded
FIFO is only offloaded for its counters, queue length is ignored.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
PRIO and ETS will need to check the value of qdisc handle in their
handlers. Add it to the callback and propagate through.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In order to have a tidy structure where to put information related to Qdisc
offloads, introduce a new structure. Move there the two existing pieces of
data: root_qdisc and tclass_qdiscs. Embed them directly, because there's no
reason to go through pointer anymore. Convert users, update init/fini
functions.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Invoke ndo_setup_tc() as appropriate to signal init / replacement,
destroying and dumping of pFIFO / bFIFO Qdisc.
A lot of the FIFO logic is used for pFIFO_head_drop as well, but that's a
semantically very different Qdisc that isn't really in the same boat as
pFIFO / bFIFO. Split some of the functions to keep the Qdisc intact.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Instead of hand-coding the busywait() predicate, use the until_counter_is()
introduced recently.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Amit Cohen <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
A function busywait() was recently added based on the logic in
__tc_check_packets(). Convert the code in tc_common to use the new
function.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Amit Cohen <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
until_counter_is() currently takes as an argument a number and the
condition holds when the current counter value is >= that number. Make the
function more generic by taking a partial expression instead of just the
number.
Convert the two existing users.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Amit Cohen <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The function tc_rule_stats_get() fetches a given statistic of a TC rule
given the rule preference. Another common way to reference a rule is using
its handle. Introduce a dual to the aforementioned function that gets a
statistic given rule handle.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Amit Cohen <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This tests that below the queue minimum length, there is no dropping /
marking, and above max, everything is dropped / marked.
The test is structured as a core file with topology and test code, and
three wrappers: one for RED used as a root Qdisc, and two for
testing (W)RED under PRIO and ETS.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Extract a helper __start_traffic() configurable by protocol type. Allow
passing through extra mausezahn arguments. Add a wrapper,
start_tcp_traffic().
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The buffer factor on Spectrum-3 is larger than on Spectrum-2. Add a new
callback and use it for mlxsw_sp->span_ops on Spectrum-3.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Spectrum-1 and Spectrum-2 do not have a per-TC counter of number of packets
marked by ECN. The value reported currently is the total number of marked
packets. Showing this value at individual TC Qdiscs is misleading.
Move the counter to ethtool instead.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a test that runs traffic across a port throttled with TBF. The test
checks that the observed throughput is within +-5% from the installed
shaper.
To allow checking both the software datapath and the offloaded one, make
the test suitable for inclusion from driver-specific wrapper. Introduce
such wrappers for mlxsw.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The function tc_rule_stats_get() fetches a packet counter of a given TC
rule. Extend it to support byte counters as well by adding an optional
argument with selector.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The function busywait() is handy as a safety-latched variant of a while
loop. Many selftests deal specifically with counter values, and busywaiting
on them is likely to be rather common (it is not quite common now, but
busywait() has not been around for very long). To facilitate expressing
simply what is tested, introduce two helpers:
- until_counter_is(), which can be used as a predicate passed to
busywait(), which holds when expression, which is itself passed as an
argument to until_counter_is(), reaches a desired value.
- busywait_for_counter(), which is useful for waiting until a given counter
changes "by" (as opposed to "to") a certain amount.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The function humanize() is used for converting value in bits/s to a
human-friendly approximate value in Kbps, Mbps or Gbps. There is nothing
hardware-specific in that, so move the function to lib.sh.
Similarly for the rate() function, which just does a bit of math to
calculate a rate, given two counter values and a time interval.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
React to the TC messages that were introduced in a preceding patch and
configure egress maximum shaper as appropriate. TBF can be used as a root
qdisc or under one of PRIO or strict ETS bands.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In order to allow configuration of burst size together with shaper rate,
extend mlxsw_sp_port_ets_maxrate_set() with a burst_size argument. Convert
call sites to pass 0 (for default).
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Lower limit of burst size configuration is dependent on system type. Add a
datum to track the value. Initialize as appropriate in mlxsw_spX_init().
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
As the port speeds grow, the current value of "unlimited shaper",
200000000Kbps, might become lower than the actually supported speeds. Bump
it to the maximum value that fits in the corresponding QEEC field, which is
about 2.1Tbps.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The QEEC register configures scheduling elements. One of the bits of
configuration is the burst size to use for the shaper installed on the
element. Add the necessary fields to support this configuration.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When the RED Qdisc is unoffloaded, it needs to reduce the reported backlog
by the amount that is in the HW, so that only the SW backlog is contained
in the counter. The same thing will need to be done by TBF, and likely any
other leaf Qdisc as well.
Extract a helper mlxsw_sp_qdisc_leaf_unoffload() and call it from
mlxsw_sp_qdisc_red_unoffload().
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add a wrapper around mlxsw_sp_qdisc_collect_tc_stats() and
mlxsw_sp_qdisc_update_stats() for the simple case of doing both in one go:
mlxsw_sp_qdisc_get_class_stats(). Dispatch to that function from
mlxsw_sp_qdisc_get_red_stats(). This new function will be useful for other
leaf Qdiscs as well.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Extract from mlxsw_sp_qdisc_get_prio_stats() two new functions:
mlxsw_sp_qdisc_collect_tc_stats() to accumulate stats for that one TC only,
and mlxsw_sp_qdisc_update_stats() that makes the stats relative to base
values stored earlier. Use them from mlxsw_sp_qdisc_get_red_stats().
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Invoke ndo_setup_tc as appropriate to signal init / replacement, destroying
and dumping of TBF Qdisc.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
In 2011, in commit b0460e4484 ("sch_tbf: report backlog information"),
TBF started copying backlog depth from the child Qdisc before dumping, with
the motivation that the backlog was otherwise not visible in "tc -s qdisc
show".
Later, in 2016, in commit 8d5958f424 ("sch_tbf: update backlog as well"),
TBF got a full-blown backlog tracking. However it kept copying the child's
backlog over before dumping.
That line is now unnecessary, so remove it.
As shown in the following example, backlog is still reported correctly:
# tc -s qdisc show dev veth0 invisible
qdisc tbf 1: root refcnt 2 rate 1Mbit burst 128Kb lat 82.8s
Sent 505475370 bytes 406985 pkt (dropped 0, overlimits 812544 requeues 0)
backlog 81972b 66p requeues 0
qdisc bfifo 0: parent 1:1 limit 10Mb
Sent 505475370 bytes 406985 pkt (dropped 0, overlimits 0 requeues 0)
backlog 81972b 66p requeues 0
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
mlxsw configures Spectrum in such a way that BUM traffic is passed not
through its nominal traffic class TC, but through its MC counterpart TC+8.
However, when collecting statistics, Qdiscs only look at the nominal TC and
ignore the MC TC.
Add two helpers to compute the value for logical TC from the constituents,
one for backlog, the other for tail drops. Use them throughout instead of
going through the xstats pointer directly.
Counters for TX bytes and packets are deduced from packet priority
counters, and therefore already include BUM traffic. wred_drop counter is
irrelevant on MC TCs, because RED is not enabled on them.
Fixes: 7b81953066 ("mlxsw: spectrum: Configure MC-aware mode on mlxsw ports")
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Per-port counter cache used by Qdiscs is updated periodically, unless the
port is down. The fact that the cache is not updated for down ports is no
problem for most counters, which are relative in nature. However, backlog
is absolute in nature, and if there is a non-zero value in the cache around
the time that the port goes down, that value just stays there. This value
then leaks to offloaded Qdiscs that report non-zero backlog even if
there (obviously) is no traffic.
The HW does not keep backlog of a downed port, so do likewise: as the port
goes down, wipe the backlog value from xstats.
Fixes: 075ab8adaf ("mlxsw: spectrum: Collect tclass related stats periodically")
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Mausezahn does not recognize "own" as a keyword on source IP address. As a
result, the MC stream is not running at all, and therefore no UC
degradation can be observed even in principle.
Fix the invocation, and tighten the test: due to the minimum shaper
configured at the MC TCs, we always expect about 20% degradation. Fail the
test if it is lower.
Fixes: 573363a68f ("selftests: mlxsw: Add qos_lib.sh")
Signed-off-by: Petr Machata <[email protected]>
Reported-by: Amit Cohen <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When a child Qdisc is removed from one of the PRIO Qdisc's bands, it is
replaced unconditionally by a NOOP qdisc. As a result, any traffic hitting
that band gets dropped. That is incorrect--no Qdisc was explicitly added
when PRIO was created, and after removal, none should have to be added
either.
Fix PRIO by first attempting to create a default Qdisc and only falling
back to noop when that fails. This pattern of attempting to create an
invisible FIFO, using NOOP only as a fallback, is also seen in other
Qdiscs.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The following patch will change PRIO to replace a removed Qdisc with an
invisible FIFO, instead of NOOP. mlxsw will see this replacement due to the
graft message that is generated. But because FIFO does not issue its own
REPLACE message, when the graft operation takes place, the Qdisc that mlxsw
tracks under the indicated band is still the old one. The child
handle (0:0) therefore does not match, and mlxsw rejects the graft
operation, which leads to an extack message:
Warning: Offloading graft operation failed.
Fix by ignoring the invisible children in the PRIO graft handler. The
DESTROY message of the removed Qdisc is going to follow shortly and handle
the removal.
Fixes: 32dc5efc6c ("mlxsw: spectrum: qdiscs: prio: Handle graft command")
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Send non-IP traffic to a port and observe that it gets prioritized
according to the lldptool app=$prio,1,0 rules.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When APP TLV selector 1 (EtherType) is used with PID of 0, the
corresponding entry specifies "default application priority [...] when
application priority is not otherwise specified."
mlxsw currently supports this type of APP entry, but uses it only as a
fallback for unspecified DSCP rules. However non-IP traffic is prioritized
according to port-default priority, not according to the DSCP-to-prio
tables, and thus it's currently not possible to prioritize such traffic
correctly.
Extend the use of the abovementioned APP entry to also set default port
priority.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add QPDP. This register controls the port default Switch Priority and
Color. The default Switch Priority and Color are used for frames where the
trust state uses default values. Currently there are two cases where this
applies: a port is in trust-PCP state, but a packet arrives untagged; and a
port is in trust-DSCP state, but a non-IP packet arrives.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This tests the newly-added ETS Qdisc. It runs two to three streams of
traffic, each with a different priority. ETS Qdisc is supposed to allocate
bandwidth according to the DRR algorithm and given weights. After running
the traffic for a while, counters are compared for each stream to check
that the expected ratio is in fact observed.
In order for the DRR process to kick in, a traffic bottleneck must exist in
the first place. In slow path, such bottleneck can be implemented by
wrapping the ETS Qdisc inside a TBF or other shaper. This might however
make the configuration unoffloadable. Instead, on HW datapath, the
bottleneck would be set up by lowering port speed and configuring shared
buffer suitably.
Therefore the test is structured as a core component that implements the
testing, with two wrapper scripts that implement the details of slow path
resp. fast path configuration.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
These two functions are used for starting several streams of traffic, and
then stopping them later. They will be handy for the test coverage of ETS
Qdisc. Move them from mlxsw-specific qos_lib.sh to the generic lib.sh.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Handle TC_SETUP_QDISC_ETS, add a new ops structure for the ETS Qdisc.
Invoke the extended prio handlers implemented in the previous patch. For
stats ops, invoke directly the prio callbacks, which are not sensitive to
differences between PRIO and ETS.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Thanks to the similarity between PRIO and ETS it is possible to simply
reuse most of the code for offloading PRIO Qdisc. Extract the common
functionality into separate functions, making the current PRIO handlers
thin API adapters.
Extend the new functions to pass quanta for individual bands, which allows
configuring a subset of bands as WRR. Invoke mlxsw_sp_port_ets_set() as
appropriate to de/configure WRR-ness and weight of individual bands.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Introduces a new Qdisc, which is based on 802.1Q-2014 wording. It is
PRIO-like in how it is configured, meaning one needs to specify how many
bands there are, how many are strict and how many are dwrr, quanta for the
latter, and priomap.
The new Qdisc operates like the PRIO / DRR combo would when configured as
per the standard. The strict classes, if any, are tried for traffic first.
When there's no traffic in any of the strict queues, the ETS ones (if any)
are treated in the same way as in DRR.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
These enums want to be named MLXSW_REG_QEEC_HIERARCHY_, but due to a typo
lack the second H. That is confusing and complicates searching.
But actually the enumerators should be named _HR_, because that is how
their enum type is called. So rename them as appropriate.
Signed-off-by: Petr Machata <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Expand the comment at mlxsw_sp_qdisc_prio_graft() to make the problem that
this function is trying to handle clearer.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
When a GRE tunnel is bound to an underlay netdevice and that netdevice is
moved to a different VRF, that could cause two tunnels to have the same
underlay local address in the same VRF. Linux in this situation dispatches
the traffic according to the tunnel key (or lack thereof), but that cannot
be offloaded to Spectrum devices.
Detect this situation and unoffload the two impacted tunnels when it
happens.
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: Jakub Kicinski <[email protected]>
In commit a82055af59 ("netfilter: nft_payload: add VLAN offload
support"), VLAN fields in struct flow_dissector_key_vlan were unionized
with the intention of introducing another field that covered the whole TCI
header. However without a wrapping struct the subfields end up sharing the
same bits. As a result, "tc filter add ... flower vlan_id 14" specifies not
only vlan_id, but also vlan_priority.
Fix by wrapping the individual VLAN fields in a struct.
Fixes: a82055af59 ("netfilter: nft_payload: add VLAN offload support")
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The helper mlxsw_sp_ipip_dev_ul_tb_id() determines the underlay VRF of a
GRE tunnel. For a tunnel without a bound device, it uses the same VRF that
the tunnel is in. However in Linux, a GRE tunnel without a bound device
uses the main VRF as the underlay. Fix the function accordingly.
mlxsw further assumed that moving a tunnel to a different VRF could cause
conflict in local tunnel endpoint address, which cannot be offloaded.
However, the only way that an underlay could be changed by moving the
tunnel device itself is if the tunnel device does not have a bound device.
But in that case the underlay is always the main VRF, so there is no
opportunity to introduce a conflict by moving such device. Thus this check
constitutes a dead code, and can be removed, which do.
Fixes: 6ddb7426a7 ("mlxsw: spectrum_router: Introduce loopback RIFs")
Signed-off-by: Petr Machata <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Petr Machata