Skip to contentSkip to conversation

Add a new Qdisc, ETS

PATCH In review Petr Machata 4 participants 17 messages Dec 18, 2019 – Dec 19, 2019

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

v2 cover Add a new Qdisc, ETS

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal transmission selection algorithms: strict priority, credit-based shaper, ETS (bandwidth sharing), and vendor-specific. All these have their corresponding knobs in DCB. But DCB does not have interfaces to configure RED and ECN, unlike Qdiscs.

In the Qdisc land, strict priority is implemented by PRIO. Credit-based transmission selection algorithm can then be modeled by having e.g. TBF or CBS Qdisc below some of the PRIO bands. ETS would then be modeled by placing a DRR Qdisc under the last PRIO band.

The problem with this approach is that DRR on its own, as well as the combination of PRIO and DRR, are tricky to configure and tricky to offload to 802.1Qaz-compliant hardware. This is due to several reasons:

- As any classful Qdisc, DRR supports adding classifiers to decide in which class to enqueue packets. Unlike PRIO, there's however no fallback in the form of priomap. A way to achieve classification based on packet priority is e.g. like this:

    # tc filter add dev swp1 root handle 1: \
		basic match 'meta(priority eq 0)' flowid 1:10

Expressing the priomap in this manner however forces drivers to deep dive into the classifier block to parse the individual rules.

A possible solution would be to extend the classes with a "defmap" a la split / defmap mechanism of CBQ, and introduce this as a last resort classification. However, unlike priomap, this doesn't have the guarantee of covering all priorities. Traffic whose priority is not covered is dropped by DRR as unclassified. But ASICs tend to implement dropping in the ACL block, not in scheduling pipelines. The need to treat these configurations correctly (if only to decide to not offload at all) complicates a driver.

It's not clear how to retrofit priomap with all its benefits to DRR without changing it beyond recognition.

- The interplay between PRIO and DRR is also causing problems. 802.1Qaz has all ETS TCs as a last resort. Switch ASICs that support ETS at all are likely to handle ETS traffic this way as well. However, the Linux model is more generic, allowing the DRR block in any band. Drivers would need to be careful to handle this case correctly, otherwise the offloaded model might not match the slow-path one.

In a similar vein, PRIO and DRR need to agree on the list of priorities assigned to DRR. This is doubly problematic--the user needs to take care to keep the two in sync, and the driver needs to watch for any holes in DRR coverage and treat the traffic correctly, as discussed above.

Note that at the time that DRR Qdisc is added, it has no classes, and thus any priorities assigned to that PRIO band are not covered. Thus this case is surprisingly rather common, and needs to be handled gracefully by the driver.

- Similarly due to DRR flexibility, when a Qdisc (such as RED) is attached below it, it is not immediately clear which TC the class represents. This is unlike PRIO with its straightforward classid scheme. When DRR is combined with PRIO, the relationship between classes and TCs gets even more murky.

This is a problem for users as well: the TC mapping is rather important for (devlink) shared buffer configuration and (ethtool) counters.

So instead, this patch set introduces a new Qdisc, which is based on 802.1Qaz 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 ETS, 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.

The chosen interface makes the overall system both reasonably easy to configure, and reasonably easy to offload. The extra code to support ETS in mlxsw (which already supports PRIO) is about 150 lines, of which perhaps 20 lines is bona fide new business logic.

Credit-based shaping transmission selection algorithm can be configured by adding a CBS Qdisc under one of the strict bands (e.g. TBF can be used to a similar effect as well). As a non-work-conserving Qdisc, CBS can't be hooked under the ETS bands. This is detected and handled identically to DRR Qdisc at runtime. Note that offloading CBS is not subject of this patchset.

The patchset proceeds in four stages:

- Patches #1-#3 are cleanups.
- Patches #4 and #5 contain the new Qdisc.
- Patches #6 and #7 update mlxsw to offload the new Qdisc.
- Patches #8-#10 add selftests for ETS.

Examples:

- Add a Qdisc with 6 bands, 3 strict and 3 ETS with 45%-30%-25% weights:

    # tc qdisc add dev swp1 root handle 1: \
	ets strict 3 quanta 4500 3000 2500 priomap 0 1 1 1 2 3 4 5
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 6 strict 3 quanta 4500 3000 2500 priomap 0 1 1 1 2 3 4 5 5 5 5 5 5 5 5 5 

- Tweak quantum of one of the classes of the previous Qdisc:

    # tc class ch dev swp1 classid 1:4 ets quantum 1000
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 6 strict 3 quanta 1000 3000 2500 priomap 0 1 1 1 2 3 4 5 5 5 5 5 5 5 5 5 
    # tc class ch dev swp1 classid 1:3 ets quantum 1000
    Error: Strict bands do not have a configurable quantum.

- Purely strict Qdisc with 1:1 mapping between priorities and TCs:

    # tc qdisc add dev swp1 root handle 1: \
	ets strict 8 priomap 7 6 5 4 3 2 1 0
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 8 strict 8 priomap 7 6 5 4 3 2 1 0 7 7 7 7 7 7 7 7 

- Use "bands" to specify number of bands explicitly. Underspecified bands are implicitly ETS and their quantum is taken from MTU. The following thus gives each band the same weight:

    # tc qdisc add dev swp1 root handle 1: \
	ets bands 8 priomap 7 6 5 4 3 2 1 0
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 8 quanta 1514 1514 1514 1514 1514 1514 1514 1514 priomap 7 6 5 4 3 2 1 0 7 7 7 7 7 7 7 7 

v2:
- This addresses points raised by David Miller.
- Patch #4:
- sch_ets.c: Add a comment with description of the Qdisc and the dequeuing algorithm.
- Kconfig: Add a high-level description to the help blurb.

v1:
- No changes, first upstream submission after RFC.

v3 (internal):
- This addresses review from Jiri Pirko.
- Patch #3:
- Rename to _HR_ instead of to _HIERARCHY_.
- Patch #4:
- pkt_sched.h: Keep all the TCA_ETS_ constants in one enum.
- pkt_sched.h: Rename TCA_ETS_BANDS to _NBANDS, _STRICT to _NSTRICT, _BAND_QUANTUM to _QUANTA_BAND and _PMAP_BAND to _PRIOMAP_BAND.
- sch_ets.c: Update to reflect the above changes. Add a new policy, ets_class_policy, which is used when parsing class changes. Currently that policy is the same as the quanta policy, but that might change.
- sch_ets.c: Move MTU handling from ets_quantum_parse() to the one caller that makes use of it.
- sch_ets.c: ets_qdisc_priomap_parse(): WARN_ON_ONCE on invalid attribute instead of returning an extack.
- Patch #6:
- __mlxsw_sp_qdisc_ets_replace(): Pass the weights argument to this function in this patch already. Drop the weight computation.
- mlxsw_sp_qdisc_prio_replace(): Rename "quanta" to "zeroes" and pass for the abovementioned "weights".
- mlxsw_sp_qdisc_prio_graft(): Convert to a wrapper around __mlxsw_sp_qdisc_ets_graft(), instead of invoking the latter directly from mlxsw_sp_setup_tc_prio().
- Update to follow the _HIERARCHY_ -> _HR_ renaming.
- Patch #7:
- __mlxsw_sp_qdisc_ets_replace(): The "weights" argument passing and weight computation removal are now done in a previous patch.
- mlxsw_sp_setup_tc_ets(): Drop case TC_ETS_REPLACE, which is handled earlier in the function.
- Patch #3 (iproute2):
- Add an example output to the commit message.
- tc-ets.8: Fix output of two examples.
- tc-ets.8: Describe default values of "bands", "quanta".
- q_ets.c: A number of fixes in error messages.
- q_ets.c: Comment formatting: /*padding*/ -> /* padding */
- q_ets.c: parse_nbands: Move duplicate checking to callers.
- q_ets.c: Don't accept both "quantum" and "quanta" as equivalent.

v2 (internal):
- This addresses review from Ido Schimmel and comments from Alexander Kushnarov.
- Patch #2:
- s/coment/comment in the commit message.
- Patch #4:
- sch_ets: ets_class_is_strict(), ets_class_id(): Constify an argument
- ets_class_find(): RXTify
- Patch #3 (iproute2):
- tc-ets.8: some spelling fixes
- tc-ets.8: add another example
- tc.8: add an ETS to "CLASSFUL QDISCS" section

v1 (internal):
- This addresses RFC reviews from Ido Schimmel and Roman Mashak, bugs found by Alexander Petrovskiy and myself, and other improvements.
- Patch #2:
- Expand the explanation with an explicit example.
- Patch #4:
- Kconfig: s/sch_drr/sch_ets/
- sch_ets: Reorder includes to be in alphabetical order
- sch_ets: ets_quantum_parse(): Rename the return-pointer argument from pquantum to quantum, and use it directly, not going through a local temporary.
- sch_ets: ets_qdisc_quanta_parse(): Convert syntax of function argument "quanta" from an array to a pointer.
- sch_ets: ets_qdisc_priomap_parse(): Likewise with "priomap".
- sch_ets: ets_qdisc_quanta_parse(), ets_qdisc_priomap_parse(): Invoke __nla_validate_nested directly instead of nl80211_validate_nested().
- sch_ets: ets_qdisc_quanta_parse(): WARN_ON_ONCE on invalid attribute instead of returning an extack.
- sch_ets: ets_qdisc_change(): Make the last band the default one for unmentioned priomap priorities.
- sch_ets: Fix a panic when an offloaded child in a bandwidth-sharing band notified its ETS parent.
- sch_ets: When ungrafting, add the newly-created invisible FIFO to the Qdisc hash
- Patch #5:
- pkt_cls.h: Note that quantum=0 signifies a strict band.
- Fix error path handling when ets_offload_dump() fails.
- Patch #6:
- __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function arguments "quanta" and "priomap" from arrays to pointers.
- Patch #7:
- __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function argument "weights" from an array to a pointer.
- Patch #9:
- mlxsw/sch_ets.sh: Add a comment explaining packet prioritization.
- Adjust the whole suite to allow testing of traffic classifiers in addition to testing priomap.
- Patch #10:
- Add a number of new tests to test default priomap band, overlarge number of bands, zeroes in quanta, and altogether missing quanta.
- Patch #1 (iproute2):
- State motivation for inclusion of this patch in the patcheset in the commit message.
- Patch #3 (iproute2):
- tc-ets.8: it is now December
- tc-ets.8: explain inactivity WRT using non-WC Qdiscs under ETS band
- tc-ets.8: s/flow/band in explanation of quantum
- tc-ets.8: explain what happens with priorities not covered by priomap
- tc-ets.8: default priomap band is now the last one
- q_ets.c: ets_parse_opt(): Remove unnecessary initialization of priomap and quanta.

Petr Machata (10):
net: pkt_cls: Clarify a comment
mlxsw: spectrum_qdisc: Clarify a comment
mlxsw: spectrum: Rename MLXSW_REG_QEEC_HIERARCY_* enumerators
net: sch_ets: Add a new Qdisc
net: sch_ets: Make the ETS qdisc offloadable
mlxsw: spectrum_qdisc: Generalize PRIO offload to support ETS
mlxsw: spectrum_qdisc: Support offloading of ETS Qdisc
selftests: forwarding: Move start_/stop_traffic from mlxsw to lib.sh
selftests: forwarding: sch_ets: Add test coverage for ETS Qdisc
selftests: qdiscs: Add test coverage for ETS Qdisc

 drivers/net/ethernet/mellanox/mlxsw/reg.h     |  11 +-
 .../net/ethernet/mellanox/mlxsw/spectrum.c    |  21 +-
 .../net/ethernet/mellanox/mlxsw/spectrum.h    |   2 +
 .../ethernet/mellanox/mlxsw/spectrum_dcb.c    |   8 +-
 .../ethernet/mellanox/mlxsw/spectrum_qdisc.c  | 219 +++-
 include/linux/netdevice.h                     |   1 +
 include/net/pkt_cls.h                         |  36 +-
 include/uapi/linux/pkt_sched.h                |  17 +
 net/sched/Kconfig                             |  17 +
 net/sched/Makefile                            |   1 +
 net/sched/sch_ets.c                           | 828 +++++++++++++++
 .../selftests/drivers/net/mlxsw/qos_lib.sh    |  46 +-
 .../selftests/drivers/net/mlxsw/sch_ets.sh    |  67 ++
 tools/testing/selftests/net/forwarding/lib.sh |  18 +
 .../selftests/net/forwarding/sch_ets.sh       |  44 +
 .../selftests/net/forwarding/sch_ets_core.sh  | 300 ++++++
 .../selftests/net/forwarding/sch_ets_tests.sh | 227 +++++
 .../tc-testing/tc-tests/qdiscs/ets.json       | 940 ++++++++++++++++++
 18 files changed, 2732 insertions(+), 71 deletions(-)
 create mode 100644 net/sched/sch_ets.c
 create mode 100755 tools/testing/selftests/drivers/net/mlxsw/sch_ets.sh
 create mode 100755 tools/testing/selftests/net/forwarding/sch_ets.sh
 create mode 100644 tools/testing/selftests/net/forwarding/sch_ets_core.sh
 create mode 100644 tools/testing/selftests/net/forwarding/sch_ets_tests.sh
 create mode 100644 tools/testing/selftests/tc-testing/tc-tests/qdiscs/ets.json
Signature · 1 line
2.20.1
#1 lore raw

PATCH v2 1/10 net: pkt_cls: Clarify a comment

The bit about negating HW backlog left me scratching my head. Clarify the comment.

---
 include/net/pkt_cls.h | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
include/net/pkt_cls.h +2 -3
index e553fc80eb23..a7c5d492bc04 100644
@@ -791,9 +791,8 @@ enum tc_prio_command {
 struct tc_prio_qopt_offload_params {
 	int bands;
 	u8 priomap[TC_PRIO_MAX + 1];
-	/* In case that a prio qdisc is offloaded and now is changed to a
-	 * non-offloadedable config, it needs to update the backlog & qlen
-	 * values to negate the HW backlog & qlen values (and only them).
+	/* At the point of un-offloading the Qdisc, the reported backlog and
+	 * qlen need to be reduced by the portion that is in HW.
 	 */
 	struct gnet_stats_queue *qstats;
 };
Signature · 1 line
2.20.1
#2 lore raw

PATCH v2 2/10 mlxsw: spectrum_qdisc: Clarify a comment

In reply to Petr Machata

Expand the comment at mlxsw_sp_qdisc_prio_graft() to make the problem that this function is trying to handle clearer.

---

Notes:
v2 (internal):
- s/coment/comment in the commit message.

v1 (internal):
- Expand the explanation with an explicit example.

 .../ethernet/mellanox/mlxsw/spectrum_qdisc.c  | 31 ++++++++++++++-----
 1 file changed, 24 insertions(+), 7 deletions(-)
drivers/net/ethernet/mellanox/mlxsw/spectrum_qdisc.c +24 -7
index 68cc6737d45c..135fef6c54b1 100644
@@ -631,10 +631,30 @@ static struct mlxsw_sp_qdisc_ops mlxsw_sp_qdisc_ops_prio = {
 	.clean_stats = mlxsw_sp_setup_tc_qdisc_prio_clean_stats,
 };
 
-/* Grafting is not supported in mlxsw. It will result in un-offloading of the
- * grafted qdisc as well as the qdisc in the qdisc new location.
- * (However, if the graft is to the location where the qdisc is already at, it
- * will be ignored completely and won't cause un-offloading).
+/* Linux allows linking of Qdiscs to arbitrary classes (so long as the resulting
+ * graph is free of cycles). These operations do not change the parent handle
+ * though, which means it can be incomplete (if there is more than one class
+ * where the Qdisc in question is grafted) or outright wrong (if the Qdisc was
+ * linked to a different class and then removed from the original class).
+ *
+ * E.g. consider this sequence of operations:
+ *
+ *  # tc qdisc add dev swp1 root handle 1: prio
+ *  # tc qdisc add dev swp1 parent 1:3 handle 13: red limit 1000000 avpkt 10000
+ *  RED: set bandwidth to 10Mbit
+ *  # tc qdisc link dev swp1 handle 13: parent 1:2
+ *
+ * At this point, both 1:2 and 1:3 have the same RED Qdisc instance as their
+ * child. But RED will still only claim that 1:3 is its parent. If it's removed
+ * from that band, its only parent will be 1:2, but it will continue to claim
+ * that it is in fact 1:3.
+ *
+ * The notification for child Qdisc replace (e.g. TC_RED_REPLACE) comes before
+ * the notification for parent graft (e.g. TC_PRIO_GRAFT). We take the replace
+ * notification to offload the child Qdisc, based on its parent handle, and use
+ * the graft operation to validate that the class where the child is actually
+ * grafted corresponds to the parent handle. If the two don't match, we
+ * unoffload the child.
  */
 static int
 mlxsw_sp_qdisc_prio_graft(struct mlxsw_sp_port *mlxsw_sp_port,
@@ -644,9 +664,6 @@ mlxsw_sp_qdisc_prio_graft(struct mlxsw_sp_port *mlxsw_sp_port,
 	int tclass_num = MLXSW_SP_PRIO_BAND_TO_TCLASS(p->band);
 	struct mlxsw_sp_qdisc *old_qdisc;
 
-	/* Check if the grafted qdisc is already in its "new" location. If so -
-	 * nothing needs to be done.
-	 */
 	if (p->band < IEEE_8021QAZ_MAX_TCS &&
 	    mlxsw_sp_port->tclass_qdiscs[tclass_num].handle == p->child_handle)
 		return 0;
Signature · 1 line
2.20.1
#3 lore raw

PATCH v2 3/10 mlxsw: spectrum: Rename MLXSW_REG_QEEC_HIERARCY_* enumerators

In reply to Petr Machata

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.

---

Notes:
v3 (internal):
- Rename to _HR_ instead of to _HIERARCHY_.

 drivers/net/ethernet/mellanox/mlxsw/reg.h     | 11 +++++------
 .../net/ethernet/mellanox/mlxsw/spectrum.c    | 19 +++++++++----------
 .../ethernet/mellanox/mlxsw/spectrum_dcb.c    |  8 ++++----
 3 files changed, 18 insertions(+), 20 deletions(-)
drivers/net/ethernet/mellanox/mlxsw/reg.h +5 -6
index 5294a1622643..86a2d575ae73 100644
@@ -3477,10 +3477,10 @@ MLXSW_REG_DEFINE(qeec, MLXSW_REG_QEEC_ID, MLXSW_REG_QEEC_LEN);
 MLXSW_ITEM32(reg, qeec, local_port, 0x00, 16, 8);
 
 enum mlxsw_reg_qeec_hr {
-	MLXSW_REG_QEEC_HIERARCY_PORT,
-	MLXSW_REG_QEEC_HIERARCY_GROUP,
-	MLXSW_REG_QEEC_HIERARCY_SUBGROUP,
-	MLXSW_REG_QEEC_HIERARCY_TC,
+	MLXSW_REG_QEEC_HR_PORT,
+	MLXSW_REG_QEEC_HR_GROUP,
+	MLXSW_REG_QEEC_HR_SUBGROUP,
+	MLXSW_REG_QEEC_HR_TC,
 };
 
 /* reg_qeec_element_hierarchy
@@ -3618,8 +3618,7 @@ static inline void mlxsw_reg_qeec_ptps_pack(char *payload, u8 local_port,
 {
 	MLXSW_REG_ZERO(qeec, payload);
 	mlxsw_reg_qeec_local_port_set(payload, local_port);
-	mlxsw_reg_qeec_element_hierarchy_set(payload,
-					     MLXSW_REG_QEEC_HIERARCY_PORT);
+	mlxsw_reg_qeec_element_hierarchy_set(payload, MLXSW_REG_QEEC_HR_PORT);
 	mlxsw_reg_qeec_ptps_set(payload, ptps);
 }
 
drivers/net/ethernet/mellanox/mlxsw/spectrum.c +9 -10
index 556dca328bb5..0d8fce749248 100644
@@ -3602,26 +3602,25 @@ static int mlxsw_sp_port_ets_init(struct mlxsw_sp_port *mlxsw_sp_port)
 	 * one subgroup, which are all member in the same group.
 	 */
 	err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
-				    MLXSW_REG_QEEC_HIERARCY_GROUP, 0, 0, false,
-				    0);
+				    MLXSW_REG_QEEC_HR_GROUP, 0, 0, false, 0);
 	if (err)
 		return err;
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
-					    MLXSW_REG_QEEC_HIERARCY_SUBGROUP, i,
+					    MLXSW_REG_QEEC_HR_SUBGROUP, i,
 					    0, false, 0);
 		if (err)
 			return err;
 	}
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
-					    MLXSW_REG_QEEC_HIERARCY_TC, i, i,
+					    MLXSW_REG_QEEC_HR_TC, i, i,
 					    false, 0);
 		if (err)
 			return err;
 
 		err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
-					    MLXSW_REG_QEEC_HIERARCY_TC,
+					    MLXSW_REG_QEEC_HR_TC,
 					    i + 8, i,
 					    true, 100);
 		if (err)
@@ -3633,13 +3632,13 @@ static int mlxsw_sp_port_ets_init(struct mlxsw_sp_port *mlxsw_sp_port)
 	 * for the initial configuration.
 	 */
 	err = mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
-					    MLXSW_REG_QEEC_HIERARCY_PORT, 0, 0,
+					    MLXSW_REG_QEEC_HR_PORT, 0, 0,
 					    MLXSW_REG_QEEC_MAS_DIS);
 	if (err)
 		return err;
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		err = mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
-						    MLXSW_REG_QEEC_HIERARCY_SUBGROUP,
+						    MLXSW_REG_QEEC_HR_SUBGROUP,
 						    i, 0,
 						    MLXSW_REG_QEEC_MAS_DIS);
 		if (err)
@@ -3647,14 +3646,14 @@ static int mlxsw_sp_port_ets_init(struct mlxsw_sp_port *mlxsw_sp_port)
 	}
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		err = mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
-						    MLXSW_REG_QEEC_HIERARCY_TC,
+						    MLXSW_REG_QEEC_HR_TC,
 						    i, i,
 						    MLXSW_REG_QEEC_MAS_DIS);
 		if (err)
 			return err;
 
 		err = mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
-						    MLXSW_REG_QEEC_HIERARCY_TC,
+						    MLXSW_REG_QEEC_HR_TC,
 						    i + 8, i,
 						    MLXSW_REG_QEEC_MAS_DIS);
 		if (err)
@@ -3664,7 +3663,7 @@ static int mlxsw_sp_port_ets_init(struct mlxsw_sp_port *mlxsw_sp_port)
 	/* Configure the min shaper for multicast TCs. */
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		err = mlxsw_sp_port_min_bw_set(mlxsw_sp_port,
-					       MLXSW_REG_QEEC_HIERARCY_TC,
+					       MLXSW_REG_QEEC_HR_TC,
 					       i + 8, i,
 					       MLXSW_REG_QEEC_MIS_MIN);
 		if (err)
drivers/net/ethernet/mellanox/mlxsw/spectrum_dcb.c +4 -4
index 21296fa7f7fb..fe3bbba90659 100644
@@ -160,7 +160,7 @@ static int __mlxsw_sp_dcbnl_ieee_setets(struct mlxsw_sp_port *mlxsw_sp_port,
 		u8 weight = ets->tc_tx_bw[i];
 
 		err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
-					    MLXSW_REG_QEEC_HIERARCY_SUBGROUP, i,
+					    MLXSW_REG_QEEC_HR_SUBGROUP, i,
 					    0, dwrr, weight);
 		if (err) {
 			netdev_err(dev, "Failed to link subgroup ETS element %d to group\n",
@@ -198,7 +198,7 @@ static int __mlxsw_sp_dcbnl_ieee_setets(struct mlxsw_sp_port *mlxsw_sp_port,
 		u8 weight = my_ets->tc_tx_bw[i];
 
 		err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
-					    MLXSW_REG_QEEC_HIERARCY_SUBGROUP, i,
+					    MLXSW_REG_QEEC_HR_SUBGROUP, i,
 					    0, dwrr, weight);
 	}
 	return err;
@@ -507,7 +507,7 @@ static int mlxsw_sp_dcbnl_ieee_setmaxrate(struct net_device *dev,
 
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		err = mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
-						    MLXSW_REG_QEEC_HIERARCY_SUBGROUP,
+						    MLXSW_REG_QEEC_HR_SUBGROUP,
 						    i, 0,
 						    maxrate->tc_maxrate[i]);
 		if (err) {
@@ -523,7 +523,7 @@ static int mlxsw_sp_dcbnl_ieee_setmaxrate(struct net_device *dev,
 err_port_ets_maxrate_set:
 	for (i--; i >= 0; i--)
 		mlxsw_sp_port_ets_maxrate_set(mlxsw_sp_port,
-					      MLXSW_REG_QEEC_HIERARCY_SUBGROUP,
+					      MLXSW_REG_QEEC_HR_SUBGROUP,
 					      i, 0, my_maxrate->tc_maxrate[i]);
 	return err;
 }
Signature · 1 line
2.20.1
#4 lore raw

PATCH v2 4/10 net: sch_ets: Add a new Qdisc

In reply to Petr Machata

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.

---

Notes:
v2 (upstream):
- sch_ets.c: Add a comment with description of the Qdisc and the dequeuing algorithm.
- Kconfig: Add a high-level description to the help blurb.

v3 (internal):
- pkt_sched.h: Keep all the TCA_ETS_ constants in one enum.
- pkt_sched.h: Rename TCA_ETS_BANDS to _NBANDS, _STRICT to _NSTRICT, _BAND_QUANTUM to _QUANTA_BAND and _PMAP_BAND to _PRIOMAP_BAND.
- sch_ets.c: Update to reflect the above changes. Add a new policy, ets_class_policy, which is used when parsing class changes. Currently that policy is the same as the quanta policy, but that might change.
- sch_ets.c: Move MTU handling from ets_quantum_parse() to the one caller that makes use of it.
- sch_ets.c: ets_qdisc_priomap_parse(): WARN_ON_ONCE on invalid attribute instead of returning an extack.

v2 (internal):
- sch_ets: ets_class_is_strict(), ets_class_id(): Constify an argument
- ets_class_find(): RXTify

v1 (internal):
- Kconfig: s/sch_drr/sch_ets/ in description
- sch_ets: Reorder includes to be in alphabetical order
- sch_ets: ets_quantum_parse(): Rename the return-pointer argument from pquantum to quantum, and use it directly, not going through a local temporary.
- sch_ets: ets_qdisc_quanta_parse(): Convert syntax of function argument "quanta" from an array to a pointer.
- sch_ets: ets_qdisc_priomap_parse(): Likewise with "priomap".
- sch_ets: ets_qdisc_quanta_parse(), ets_qdisc_priomap_parse(): Invoke __nla_validate_nested directly instead of nl80211_validate_nested().
- sch_ets: ets_qdisc_quanta_parse(): WARN_ON_ONCE on invalid attribute instead of returning an extack.
- sch_ets: ets_qdisc_change(): Make the last band the default one for unmentioned priomap priorities.
- sch_ets: Fix a panic when an offloaded child in a bandwidth-sharing band notified its ETS parent. (Reported by Alexander Petrovskiy.)
- sch_ets: When ungrafting, add the newly-created invisible FIFO to the Qdisc hash

 include/uapi/linux/pkt_sched.h |  17 +
 net/sched/Kconfig              |  17 +
 net/sched/Makefile             |   1 +
 net/sched/sch_ets.c            | 733 +++++++++++++++++++++++++++++++++
 4 files changed, 768 insertions(+)
 create mode 100644 net/sched/sch_ets.c
include/uapi/linux/pkt_sched.h +17 -0
index 9f1a72876212..bf5a5b1dfb0b 100644
@@ -1187,4 +1187,21 @@ enum {
 
 #define TCA_TAPRIO_ATTR_MAX (__TCA_TAPRIO_ATTR_MAX - 1)
 
+/* ETS */
+
+#define TCQ_ETS_MAX_BANDS 16
+
+enum {
+	TCA_ETS_UNSPEC,
+	TCA_ETS_NBANDS,		/* u8 */
+	TCA_ETS_NSTRICT,	/* u8 */
+	TCA_ETS_QUANTA,		/* nested TCA_ETS_QUANTA_BAND */
+	TCA_ETS_QUANTA_BAND,	/* u32 */
+	TCA_ETS_PRIOMAP,	/* nested TCA_ETS_PRIOMAP_BAND */
+	TCA_ETS_PRIOMAP_BAND,	/* u8 */
+	__TCA_ETS_MAX,
+};
+
+#define TCA_ETS_MAX (__TCA_ETS_MAX - 1)
+
 #endif
net/sched/Kconfig +17 -0
index 2985509147a2..b1e7ec726958 100644
@@ -409,6 +409,23 @@ config NET_SCH_PLUG
 	  To compile this code as a module, choose M here: the
 	  module will be called sch_plug.
 
+config NET_SCH_ETS
+	tristate "Enhanced transmission selection scheduler (ETS)"
+	help
+          The Enhanced Transmission Selection scheduler is a classful
+          queuing discipline that merges functionality of PRIO and DRR
+          qdiscs in one scheduler. ETS makes it easy to configure a set of
+          strict and bandwidth-sharing bands to implement the transmission
+          selection described in 802.1Qaz.
+
+	  Say Y here if you want to use the ETS packet scheduling
+	  algorithm.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called sch_ets.
+
+	  If unsure, say N.
+
 menuconfig NET_SCH_DEFAULT
 	bool "Allow override default queue discipline"
 	---help---
net/sched/Makefile +1 -0
index 415d1e1f237e..bc8856b865ff 100644
@@ -48,6 +48,7 @@ obj-$(CONFIG_NET_SCH_ATM)	+= sch_atm.o
 obj-$(CONFIG_NET_SCH_NETEM)	+= sch_netem.o
 obj-$(CONFIG_NET_SCH_DRR)	+= sch_drr.o
 obj-$(CONFIG_NET_SCH_PLUG)	+= sch_plug.o
+obj-$(CONFIG_NET_SCH_ETS)	+= sch_ets.o
 obj-$(CONFIG_NET_SCH_MQPRIO)	+= sch_mqprio.o
 obj-$(CONFIG_NET_SCH_SKBPRIO)	+= sch_skbprio.o
 obj-$(CONFIG_NET_SCH_CHOKE)	+= sch_choke.o
net/sched/sch_ets.c new file +733 -0

733 lines not on this page · raw message

Signature · 1 line
2.20.1
maintainer Acked-by on v2 4/10
#5 lore raw

Re: [PATCH net-next mlxsw v2 04/10] net: sch_ets: Add a new Qdisc

Quoted 11 lines · Wed, Dec 18, 2019 at 03:55:13PM CET, [email protected] wrote:

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.

#6 lore raw

PATCH v2 5/10 net: sch_ets: Make the ETS qdisc offloadable

In reply to Petr Machata

Add hooks at appropriate points to make it possible to offload the ETS Qdisc.

---

Notes:
v1 (internal):
- pkt_cls.h: Note that quantum=0 signifies a strict band.
- Fix error path handling when ets_offload_dump() fails.

 include/linux/netdevice.h |  1 +
 include/net/pkt_cls.h     | 31 +++++++++++++
 net/sched/sch_ets.c       | 95 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 127 insertions(+)
include/linux/netdevice.h +1 -0
index 30745068fb39..7a8ed11f5d45 100644
@@ -849,6 +849,7 @@ enum tc_setup_type {
 	TC_SETUP_QDISC_GRED,
 	TC_SETUP_QDISC_TAPRIO,
 	TC_SETUP_FT,
+	TC_SETUP_QDISC_ETS,
 };
 
 /* These structures hold the attributes of bpf state that are being passed
include/net/pkt_cls.h +31 -0
index a7c5d492bc04..47b115e2012a 100644
@@ -823,4 +823,35 @@ struct tc_root_qopt_offload {
 	bool ingress;
 };
 
+enum tc_ets_command {
+	TC_ETS_REPLACE,
+	TC_ETS_DESTROY,
+	TC_ETS_STATS,
+	TC_ETS_GRAFT,
+};
+
+struct tc_ets_qopt_offload_replace_params {
+	unsigned int bands;
+	u8 priomap[TC_PRIO_MAX + 1];
+	unsigned int quanta[TCQ_ETS_MAX_BANDS];	/* 0 for strict bands. */
+	unsigned int weights[TCQ_ETS_MAX_BANDS];
+	struct gnet_stats_queue *qstats;
+};
+
+struct tc_ets_qopt_offload_graft_params {
+	u8 band;
+	u32 child_handle;
+};
+
+struct tc_ets_qopt_offload {
+	enum tc_ets_command command;
+	u32 handle;
+	u32 parent;
+	union {
+		struct tc_ets_qopt_offload_replace_params replace_params;
+		struct tc_qopt_offload_stats stats;
+		struct tc_ets_qopt_offload_graft_params graft_params;
+	};
+};
+
 #endif
net/sched/sch_ets.c +95 -0

131 lines not on this page · raw message

Signature · 1 line
2.20.1
#7 lore raw

PATCH v2 6/10 mlxsw: spectrum_qdisc: Generalize PRIO offload to support ETS

In reply to Petr Machata

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.

---

Notes:
v3 (internal):
- __mlxsw_sp_qdisc_ets_replace(): Pass the weights argument to this function in this patch already. Drop the weight computation.
- mlxsw_sp_qdisc_prio_replace(): Rename "quanta" to "zeroes" and pass for the abovementioned "weights".
- mlxsw_sp_qdisc_prio_graft(): Convert to a wrapper around __mlxsw_sp_qdisc_ets_graft(), instead of invoking the latter directly from mlxsw_sp_setup_tc_prio().
- Update to follow the _HIERARCHY_ -> _HR_ renaming.

v1 (internal):
- __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function arguments "quanta" and "priomap" from arrays to pointers.

 .../ethernet/mellanox/mlxsw/spectrum_qdisc.c  | 104 ++++++++++++++----
 1 file changed, 81 insertions(+), 23 deletions(-)
drivers/net/ethernet/mellanox/mlxsw/spectrum_qdisc.c +81 -23
index 135fef6c54b1..d513af49c0a8 100644
@@ -471,14 +471,16 @@ int mlxsw_sp_setup_tc_red(struct mlxsw_sp_port *mlxsw_sp_port,
 }
 
 static int
-mlxsw_sp_qdisc_prio_destroy(struct mlxsw_sp_port *mlxsw_sp_port,
-			    struct mlxsw_sp_qdisc *mlxsw_sp_qdisc)
+__mlxsw_sp_qdisc_ets_destroy(struct mlxsw_sp_port *mlxsw_sp_port)
 {
 	int i;
 
 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 		mlxsw_sp_port_prio_tc_set(mlxsw_sp_port, i,
 					  MLXSW_SP_PORT_DEFAULT_TCLASS);
+		mlxsw_sp_port_ets_set(mlxsw_sp_port,
+				      MLXSW_REG_QEEC_HR_SUBGROUP,
+				      i, 0, false, 0);
 		mlxsw_sp_qdisc_destroy(mlxsw_sp_port,
 				       &mlxsw_sp_port->tclass_qdiscs[i]);
 		mlxsw_sp_port->tclass_qdiscs[i].prio_bitmap = 0;
@@ -487,6 +489,22 @@ mlxsw_sp_qdisc_prio_destroy(struct mlxsw_sp_port *mlxsw_sp_port,
 	return 0;
 }
 
+static int
+mlxsw_sp_qdisc_prio_destroy(struct mlxsw_sp_port *mlxsw_sp_port,
+			    struct mlxsw_sp_qdisc *mlxsw_sp_qdisc)
+{
+	return __mlxsw_sp_qdisc_ets_destroy(mlxsw_sp_port);
+}
+
+static int
+__mlxsw_sp_qdisc_ets_check_params(unsigned int nbands)
+{
+	if (nbands > IEEE_8021QAZ_MAX_TCS)
+		return -EOPNOTSUPP;
+
+	return 0;
+}
+
 static int
 mlxsw_sp_qdisc_prio_check_params(struct mlxsw_sp_port *mlxsw_sp_port,
 				 struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
@@ -494,30 +512,36 @@ mlxsw_sp_qdisc_prio_check_params(struct mlxsw_sp_port *mlxsw_sp_port,
 {
 	struct tc_prio_qopt_offload_params *p = params;
 
-	if (p->bands > IEEE_8021QAZ_MAX_TCS)
-		return -EOPNOTSUPP;
-
-	return 0;
+	return __mlxsw_sp_qdisc_ets_check_params(p->bands);
 }
 
 static int
-mlxsw_sp_qdisc_prio_replace(struct mlxsw_sp_port *mlxsw_sp_port,
-			    struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
-			    void *params)
+__mlxsw_sp_qdisc_ets_replace(struct mlxsw_sp_port *mlxsw_sp_port,
+			     unsigned int nbands,
+			     const unsigned int *quanta,
+			     const unsigned int *weights,
+			     const u8 *priomap)
 {
-	struct tc_prio_qopt_offload_params *p = params;
 	struct mlxsw_sp_qdisc *child_qdisc;
 	int tclass, i, band, backlog;
 	u8 old_priomap;
 	int err;
 
-	for (band = 0; band < p->bands; band++) {
+	for (band = 0; band < nbands; band++) {
 		tclass = MLXSW_SP_PRIO_BAND_TO_TCLASS(band);
 		child_qdisc = &mlxsw_sp_port->tclass_qdiscs[tclass];
 		old_priomap = child_qdisc->prio_bitmap;
 		child_qdisc->prio_bitmap = 0;
+
+		err = mlxsw_sp_port_ets_set(mlxsw_sp_port,
+					    MLXSW_REG_QEEC_HR_SUBGROUP,
+					    tclass, 0, !!quanta[band],
+					    weights[band]);
+		if (err)
+			return err;
+
 		for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
-			if (p->priomap[i] == band) {
+			if (priomap[i] == band) {
 				child_qdisc->prio_bitmap |= BIT(i);
 				if (BIT(i) & old_priomap)
 					continue;
@@ -540,21 +564,46 @@ mlxsw_sp_qdisc_prio_replace(struct mlxsw_sp_port *mlxsw_sp_port,
 		child_qdisc = &mlxsw_sp_port->tclass_qdiscs[tclass];
 		child_qdisc->prio_bitmap = 0;
 		mlxsw_sp_qdisc_destroy(mlxsw_sp_port, child_qdisc);
+		mlxsw_sp_port_ets_set(mlxsw_sp_port,
+				      MLXSW_REG_QEEC_HR_SUBGROUP,
+				      tclass, 0, false, 0);
 	}
 	return 0;
 }
 
+static int
+mlxsw_sp_qdisc_prio_replace(struct mlxsw_sp_port *mlxsw_sp_port,
+			    struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
+			    void *params)
+{
+	struct tc_prio_qopt_offload_params *p = params;
+	unsigned int zeroes[TCQ_ETS_MAX_BANDS] = {0};
+
+	return __mlxsw_sp_qdisc_ets_replace(mlxsw_sp_port, p->bands,
+					    zeroes, zeroes, p->priomap);
+}
+
+static void
+__mlxsw_sp_qdisc_ets_unoffload(struct mlxsw_sp_port *mlxsw_sp_port,
+			       struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
+			       struct gnet_stats_queue *qstats)
+{
+	u64 backlog;
+
+	backlog = mlxsw_sp_cells_bytes(mlxsw_sp_port->mlxsw_sp,
+				       mlxsw_sp_qdisc->stats_base.backlog);
+	qstats->backlog -= backlog;
+}
+
 static void
 mlxsw_sp_qdisc_prio_unoffload(struct mlxsw_sp_port *mlxsw_sp_port,
 			      struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
 			      void *params)
 {
 	struct tc_prio_qopt_offload_params *p = params;
-	u64 backlog;
 
-	backlog = mlxsw_sp_cells_bytes(mlxsw_sp_port->mlxsw_sp,
-				       mlxsw_sp_qdisc->stats_base.backlog);
-	p->qstats->backlog -= backlog;
+	__mlxsw_sp_qdisc_ets_unoffload(mlxsw_sp_port, mlxsw_sp_qdisc,
+				       p->qstats);
 }
 
 static int
@@ -657,22 +706,22 @@ static struct mlxsw_sp_qdisc_ops mlxsw_sp_qdisc_ops_prio = {
  * unoffload the child.
  */
 static int
-mlxsw_sp_qdisc_prio_graft(struct mlxsw_sp_port *mlxsw_sp_port,
-			  struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
-			  struct tc_prio_qopt_offload_graft_params *p)
+__mlxsw_sp_qdisc_ets_graft(struct mlxsw_sp_port *mlxsw_sp_port,
+			   struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
+			   u8 band, u32 child_handle)
 {
-	int tclass_num = MLXSW_SP_PRIO_BAND_TO_TCLASS(p->band);
+	int tclass_num = MLXSW_SP_PRIO_BAND_TO_TCLASS(band);
 	struct mlxsw_sp_qdisc *old_qdisc;
 
-	if (p->band < IEEE_8021QAZ_MAX_TCS &&
-	    mlxsw_sp_port->tclass_qdiscs[tclass_num].handle == p->child_handle)
+	if (band < IEEE_8021QAZ_MAX_TCS &&
+	    mlxsw_sp_port->tclass_qdiscs[tclass_num].handle == child_handle)
 		return 0;
 
 	/* See if the grafted qdisc is already offloaded on any tclass. If so,
 	 * unoffload it.
 	 */
 	old_qdisc = mlxsw_sp_qdisc_find_by_handle(mlxsw_sp_port,
-						  p->child_handle);
+						  child_handle);
 	if (old_qdisc)
 		mlxsw_sp_qdisc_destroy(mlxsw_sp_port, old_qdisc);
 
@@ -681,6 +730,15 @@ mlxsw_sp_qdisc_prio_graft(struct mlxsw_sp_port *mlxsw_sp_port,
 	return -EOPNOTSUPP;
 }
 
+static int
+mlxsw_sp_qdisc_prio_graft(struct mlxsw_sp_port *mlxsw_sp_port,
+			  struct mlxsw_sp_qdisc *mlxsw_sp_qdisc,
+			  struct tc_prio_qopt_offload_graft_params *p)
+{
+	return __mlxsw_sp_qdisc_ets_graft(mlxsw_sp_port, mlxsw_sp_qdisc,
+					  p->band, p->child_handle);
+}
+
 int mlxsw_sp_setup_tc_prio(struct mlxsw_sp_port *mlxsw_sp_port,
 			   struct tc_prio_qopt_offload *p)
 {
Signature · 1 line
2.20.1
#8 lore raw

PATCH v2 7/10 mlxsw: spectrum_qdisc: Support offloading of ETS Qdisc

In reply to Petr Machata

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.

---

Notes:
v3 (internal):
- __mlxsw_sp_qdisc_ets_replace(): The "weights" argument passing and weight computation removal are now done in a previous patch.
- mlxsw_sp_setup_tc_ets(): Drop case TC_ETS_REPLACE, which is handled earlier in the function.

v1 (internal):
- __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function argument "weights" from an array to a pointer.

 .../net/ethernet/mellanox/mlxsw/spectrum.c    |  2 +
 .../net/ethernet/mellanox/mlxsw/spectrum.h    |  2 +
 .../ethernet/mellanox/mlxsw/spectrum_qdisc.c  | 84 +++++++++++++++++++
 3 files changed, 88 insertions(+)
drivers/net/ethernet/mellanox/mlxsw/spectrum.c +2 -0
index 0d8fce749248..ea632042e609 100644
@@ -1796,6 +1796,8 @@ static int mlxsw_sp_setup_tc(struct net_device *dev, enum tc_setup_type type,
 		return mlxsw_sp_setup_tc_red(mlxsw_sp_port, type_data);
 	case TC_SETUP_QDISC_PRIO:
 		return mlxsw_sp_setup_tc_prio(mlxsw_sp_port, type_data);
+	case TC_SETUP_QDISC_ETS:
+		return mlxsw_sp_setup_tc_ets(mlxsw_sp_port, type_data);
 	default:
 		return -EOPNOTSUPP;
 	}
drivers/net/ethernet/mellanox/mlxsw/spectrum.h +2 -0
index 347bec9d1ecf..948ef4720d40 100644
@@ -852,6 +852,8 @@ int mlxsw_sp_setup_tc_red(struct mlxsw_sp_port *mlxsw_sp_port,
 			  struct tc_red_qopt_offload *p);
 int mlxsw_sp_setup_tc_prio(struct mlxsw_sp_port *mlxsw_sp_port,
 			   struct tc_prio_qopt_offload *p);
+int mlxsw_sp_setup_tc_ets(struct mlxsw_sp_port *mlxsw_sp_port,
+			  struct tc_ets_qopt_offload *p);
 
 /* spectrum_fid.c */
 bool mlxsw_sp_fid_is_dummy(struct mlxsw_sp *mlxsw_sp, u16 fid_index);
drivers/net/ethernet/mellanox/mlxsw/spectrum_qdisc.c +84 -0

102 lines not on this page · raw message

Signature · 1 line
2.20.1
#9 lore raw

PATCH v2 8/10 selftests: forwarding: Move start_/stop_traffic from mlxsw to lib.sh

In reply to Petr Machata

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.

---
 .../selftests/drivers/net/mlxsw/qos_lib.sh     | 18 ------------------
 tools/testing/selftests/net/forwarding/lib.sh  | 18 ++++++++++++++++++
 2 files changed, 18 insertions(+), 18 deletions(-)
tools/testing/selftests/drivers/net/mlxsw/qos_lib.sh +0 -18
index e80be65799ad..75a3fb3b5663 100644
@@ -24,24 +24,6 @@ rate()
 	echo $((8 * (t1 - t0) / interval))
 }
 
-start_traffic()
-{
-	local h_in=$1; shift    # Where the traffic egresses the host
-	local sip=$1; shift
-	local dip=$1; shift
-	local dmac=$1; shift
-
-	$MZ $h_in -p 8000 -A $sip -B $dip -c 0 \
-		-a own -b $dmac -t udp -q &
-	sleep 1
-}
-
-stop_traffic()
-{
-	# Suppress noise from killing mausezahn.
-	{ kill %% && wait %%; } 2>/dev/null
-}
-
 check_rate()
 {
 	local rate=$1; shift
tools/testing/selftests/net/forwarding/lib.sh +18 -0
index 1f64e7348f69..a0b09bb6995e 100644
@@ -1065,3 +1065,21 @@ flood_test()
 	flood_unicast_test $br_port $host1_if $host2_if
 	flood_multicast_test $br_port $host1_if $host2_if
 }
+
+start_traffic()
+{
+	local h_in=$1; shift    # Where the traffic egresses the host
+	local sip=$1; shift
+	local dip=$1; shift
+	local dmac=$1; shift
+
+	$MZ $h_in -p 8000 -A $sip -B $dip -c 0 \
+		-a own -b $dmac -t udp -q &
+	sleep 1
+}
+
+stop_traffic()
+{
+	# Suppress noise from killing mausezahn.
+	{ kill %% && wait %%; } 2>/dev/null
+}
Signature · 1 line
2.20.1
#10 lore raw

PATCH v2 9/10 selftests: forwarding: sch_ets: Add test coverage for ETS Qdisc

In reply to Petr Machata

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.

---

Notes:
v1 (internal):
- mlxsw/sch_ets.sh: Add a comment explaining packet prioritization.
- Adjust the whole suite to allow testing of traffic classifiers in addition to testing priomap.

 .../selftests/drivers/net/mlxsw/qos_lib.sh    |  28 ++
 .../selftests/drivers/net/mlxsw/sch_ets.sh    |  67 ++++
 .../selftests/net/forwarding/sch_ets.sh       |  44 +++
 .../selftests/net/forwarding/sch_ets_core.sh  | 300 ++++++++++++++++++
 .../selftests/net/forwarding/sch_ets_tests.sh | 227 +++++++++++++
 5 files changed, 666 insertions(+)
 create mode 100755 tools/testing/selftests/drivers/net/mlxsw/sch_ets.sh
 create mode 100755 tools/testing/selftests/net/forwarding/sch_ets.sh
 create mode 100644 tools/testing/selftests/net/forwarding/sch_ets_core.sh
 create mode 100644 tools/testing/selftests/net/forwarding/sch_ets_tests.sh
tools/testing/selftests/drivers/net/mlxsw/qos_lib.sh +28 -0
index 75a3fb3b5663..a5937069ac16 100644
@@ -78,3 +78,31 @@ measure_rate()
 	echo $ir $er
 	return $ret
 }
+
+bail_on_lldpad()
+{
+	if systemctl is-active --quiet lldpad; then
+
+		cat >/dev/stderr <<-EOF
+		WARNING: lldpad is running
+
+			lldpad will likely configure DCB, and this test will
+			configure Qdiscs. mlxsw does not support both at the
+			same time, one of them is arbitrarily going to overwrite
+			the other. That will cause spurious failures (or,
+			unlikely, passes) of this test.
+		EOF
+
+		if [[ -z $ALLOW_LLDPAD ]]; then
+			cat >/dev/stderr <<-EOF
+
+				If you want to run the test anyway, please set
+				an environment variable ALLOW_LLDPAD to a
+				non-empty string.
+			EOF
+			exit 1
+		else
+			return
+		fi
+	fi
+}
tools/testing/selftests/drivers/net/mlxsw/sch_ets.sh new file +67 -0

67 lines not on this page · raw message

tools/testing/selftests/net/forwarding/sch_ets.sh new file +44 -0

44 lines not on this page · raw message

tools/testing/selftests/net/forwarding/sch_ets_core.sh new file +300 -0

300 lines not on this page · raw message

tools/testing/selftests/net/forwarding/sch_ets_tests.sh new file +227 -0

227 lines not on this page · raw message

Signature · 1 line
2.20.1
#11 lore raw

PATCH v2 10/10 selftests: qdiscs: Add test coverage for ETS Qdisc

In reply to Petr Machata

Add TDC coverage for the new ETS Qdisc.

---

Notes:
v1 (internal):
- Add a number of new tests to test default priomap band, overlarge number of bands, zeroes in quanta, and altogether missing quanta.

 .../tc-testing/tc-tests/qdiscs/ets.json       | 940 ++++++++++++++++++
 1 file changed, 940 insertions(+)
 create mode 100644 tools/testing/selftests/tc-testing/tc-tests/qdiscs/ets.json
tools/testing/selftests/tc-testing/tc-tests/qdiscs/ets.json new file +940 -0
new file mode 100644
index 000000000000..180593010675
@@ -0,0 +1,940 @@
+[
+    {
+        "id": "e90e",
+        "name": "Add ETS qdisc using bands",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 2",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .* bands 2",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "b059",
+        "name": "Add ETS qdisc using quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1000 900 800 700",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 4 quanta 1000 900 800 700",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "e8e7",
+        "name": "Add ETS qdisc using strict",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 3",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 3 strict 3",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "233c",
+        "name": "Add ETS qdisc using bands + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 4 quanta 1000 900 800 700",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 4 quanta 1000 900 800 700 priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "3d35",
+        "name": "Add ETS qdisc using bands + strict",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 3 strict 3",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 3 strict 3 priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "7f3b",
+        "name": "Add ETS qdisc using strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 3 quanta 1500 750",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 5 strict 3 quanta 1500 750 priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "4593",
+        "name": "Add ETS qdisc using strict 0 + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 0 quanta 1500 750",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 2 quanta 1500 750 priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "8938",
+        "name": "Add ETS qdisc using bands + strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 5 strict 3 quanta 1500 750",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 5 .*strict 3 quanta 1500 750 priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "0782",
+        "name": "Add ETS qdisc with more bands than quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 2 quanta 1000",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 2 .*quanta 1000 [1-9][0-9]* priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "501b",
+        "name": "Add ETS qdisc with more bands than strict",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 3 strict 1",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 3 strict 1 quanta ([1-9][0-9]* ){2}priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "671a",
+        "name": "Add ETS qdisc with more bands than strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 3 strict 1 quanta 1000",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 3 strict 1 quanta 1000 [1-9][0-9]* priomap",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "2a23",
+        "name": "Add ETS qdisc with 16 bands",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 16",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .* bands 16",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "8daf",
+        "name": "Add ETS qdisc with 17 bands",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 17",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "7f95",
+        "name": "Add ETS qdisc with 17 strict",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 17",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "837a",
+        "name": "Add ETS qdisc with 16 quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .* bands 16",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "65b6",
+        "name": "Add ETS qdisc with 17 quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17",
+        "expExitCode": "2",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "b9e9",
+        "name": "Add ETS qdisc with 16 strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 8 quanta 1 2 3 4 5 6 7 8",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .* bands 16",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "9877",
+        "name": "Add ETS qdisc with 17 strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 9 quanta 1 2 3 4 5 6 7 8",
+        "expExitCode": "2",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "c696",
+        "name": "Add ETS qdisc with priomap",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 5 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "30c4",
+        "name": "Add ETS qdisc with quanta + priomap",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1000 2000 3000 4000 5000 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*quanta 1000 2000 3000 4000 5000 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "e8ac",
+        "name": "Add ETS qdisc with strict + priomap",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 5 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*bands 5 strict 5 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "5a7e",
+        "name": "Add ETS qdisc with quanta + strict + priomap",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 2 quanta 1000 2000 3000 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*strict 2 quanta 1000 2000 3000 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "cb8b",
+        "name": "Show ETS class :1",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 4000 3000 2000",
+        "expExitCode": "0",
+        "verifyCmd": "$TC class show dev $DUMMY classid 1:1",
+        "matchPattern": "class ets 1:1 root quantum 4000",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "1b4e",
+        "name": "Show ETS class :2",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 4000 3000 2000",
+        "expExitCode": "0",
+        "verifyCmd": "$TC class show dev $DUMMY classid 1:2",
+        "matchPattern": "class ets 1:2 root quantum 3000",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "f642",
+        "name": "Show ETS class :3",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 4000 3000 2000",
+        "expExitCode": "0",
+        "verifyCmd": "$TC class show dev $DUMMY classid 1:3",
+        "matchPattern": "class ets 1:3 root quantum 2000",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "0a5f",
+        "name": "Show ETS strict class",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 3",
+        "expExitCode": "0",
+        "verifyCmd": "$TC class show dev $DUMMY classid 1:1",
+        "matchPattern": "class ets 1:1 root $",
+        "matchCount": "1",
+        "teardown": [
+            "$TC qdisc del dev $DUMMY handle 1: root",
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "f7c8",
+        "name": "Add ETS qdisc with too many quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 2 quanta 1000 2000 3000",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "2389",
+        "name": "Add ETS qdisc with too many strict",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 2 strict 3",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "fe3c",
+        "name": "Add ETS qdisc with too many strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 4 strict 2 quanta 1000 2000 3000",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "cb04",
+        "name": "Add ETS qdisc with excess priomap elements",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 5 priomap 0 0 1 0 1 2 0 1 2 3 0 1 2 3 4 0 1 2",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "c32e",
+        "name": "Add ETS qdisc with priomap above bands",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 2 priomap 0 1 2",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "744c",
+        "name": "Add ETS qdisc with priomap above quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1000 500 priomap 0 1 2",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "7b33",
+        "name": "Add ETS qdisc with priomap above strict",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 2 priomap 0 1 2",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "dbe6",
+        "name": "Add ETS qdisc with priomap above strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets strict 1 quanta 1000 500 priomap 0 1 2 3",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "bdb2",
+        "name": "Add ETS qdisc with priomap within bands with strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 4 strict 1 quanta 1000 500 priomap 0 1 2 3",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "39a3",
+        "name": "Add ETS qdisc with priomap above bands with strict + quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 4 strict 1 quanta 1000 500 priomap 0 1 2 3 4",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "557c",
+        "name": "Unset priorities default to the last band",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 4 priomap 0 0 0 0",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets .*priomap 0 0 0 0 3 3 3 3 3 3 3 3 3 3 3 3",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "a347",
+        "name": "Unset priorities default to the last band -- no priomap",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 4",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets .*priomap 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "39c4",
+        "name": "Add ETS qdisc with too few bands",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 0",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "930b",
+        "name": "Add ETS qdisc with too many bands",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets bands 17",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "406a",
+        "name": "Add ETS qdisc without parameters",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "e51a",
+        "name": "Zero element in quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1000 0 800 700",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "e7f2",
+        "name": "Sole zero element in quanta",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta 0",
+        "expExitCode": "1",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "d6e6",
+        "name": "No values after the quanta keyword",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true"
+        ],
+        "cmdUnderTest": "$TC qdisc add dev $DUMMY handle 1: root ets quanta",
+        "expExitCode": "255",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets",
+        "matchCount": "0",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "28c6",
+        "name": "Change ETS band quantum",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true",
+            "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1000 2000 3000"
+        ],
+        "cmdUnderTest": "$TC class change dev $DUMMY classid 1:1 ets quantum 1500",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*quanta 1500 2000 3000 priomap ",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "4714",
+        "name": "Change ETS band without quantum",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true",
+            "$TC qdisc add dev $DUMMY handle 1: root ets quanta 1000 2000 3000"
+        ],
+        "cmdUnderTest": "$TC class change dev $DUMMY classid 1:1 ets",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets 1: root .*quanta 1000 2000 3000 priomap ",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "6979",
+        "name": "Change quantum of a strict ETS band",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true",
+            "$TC qdisc add dev $DUMMY handle 1: root ets strict 5"
+        ],
+        "cmdUnderTest": "$TC class change dev $DUMMY classid 1:2 ets quantum 1500",
+        "expExitCode": "2",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets .*bands 5 .*strict 5",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    },
+    {
+        "id": "9a7d",
+        "name": "Change ETS strict band without quantum",
+        "category": [
+            "qdisc",
+            "ets"
+        ],
+        "setup": [
+            "$IP link add dev $DUMMY type dummy || /bin/true",
+            "$TC qdisc add dev $DUMMY handle 1: root ets strict 5"
+        ],
+        "cmdUnderTest": "$TC class change dev $DUMMY classid 1:2 ets",
+        "expExitCode": "0",
+        "verifyCmd": "$TC qdisc show dev $DUMMY",
+        "matchPattern": "qdisc ets .*bands 5 .*strict 5",
+        "matchCount": "1",
+        "teardown": [
+            "$IP link del dev $DUMMY type dummy"
+        ]
+    }
+]
Signature · 1 line
2.20.1
#12 lore raw
In reply to Petr Machata
Quoted 5 lines · Petr Machata

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal transmission selection algorithms: strict priority, credit-based shaper, ETS (bandwidth sharing), and vendor-specific. All these have their corresponding knobs in DCB. But DCB does not have interfaces to configure RED and ECN, unlike Qdiscs.

So the idea here (way back when I did this years ago) is that marking ECN traffic was not paticularly CPU intensive on any metrics I came up with. And I don't recall anyone ever wanting to do RED here. The configuration I usually recommended was to use mqprio + SO_PRIORITY + fq per qdisc. Then once we got the BPF egress hook we replaced SO_PRIORITY configurations with the more dynamic BPF action to set it. There was never a compelling perf reason to offload red/ecn.

But these use cases were edge nodes. I believe this series is mostly about control path and maybe some light control traffic? This is for switches not for edge nodes right? I'm guessing because I don't see any performance analaysis on why this is useful, intuitively it makes sense if there is a small CPU sitting on a 48 port 10gbps box or something like that.

Quoted 9 lines

In the Qdisc land, strict priority is implemented by PRIO. Credit-based transmission selection algorithm can then be modeled by having e.g. TBF or CBS Qdisc below some of the PRIO bands. ETS would then be modeled by placing a DRR Qdisc under the last PRIO band.

The problem with this approach is that DRR on its own, as well as the combination of PRIO and DRR, are tricky to configure and tricky to offload to 802.1Qaz-compliant hardware. This is due to several reasons:

I would argue the trick to configure part could be hid behind tooling to simplify setup. The more annoying part is it was stuck behind the qdisc lock. I was hoping this would implement a lockless ETS qdisc seeing we have the infra to do lockless qdiscs now. But seems not. I guess software perf analysis might show prio+drr and ets here are about the same performance wise.

offload is tricky with stacked qdiscs though ;)

Quoted 24 lines

- As any classful Qdisc, DRR supports adding classifiers to decide in which class to enqueue packets. Unlike PRIO, there's however no fallback in the form of priomap. A way to achieve classification based on packet priority is e.g. like this:

    # tc filter add dev swp1 root handle 1: \
		basic match 'meta(priority eq 0)' flowid 1:10

Expressing the priomap in this manner however forces drivers to deep dive into the classifier block to parse the individual rules.

A possible solution would be to extend the classes with a "defmap" a la split / defmap mechanism of CBQ, and introduce this as a last resort classification. However, unlike priomap, this doesn't have the guarantee of covering all priorities. Traffic whose priority is not covered is dropped by DRR as unclassified. But ASICs tend to implement dropping in the ACL block, not in scheduling pipelines. The need to treat these configurations correctly (if only to decide to not offload at all) complicates a driver.

It's not clear how to retrofit priomap with all its benefits to DRR without changing it beyond recognition.

Quoted 6 lines

- The interplay between PRIO and DRR is also causing problems. 802.1Qaz has all ETS TCs as a last resort. Switch ASICs that support ETS at all are likely to handle ETS traffic this way as well. However, the Linux model is more generic, allowing the DRR block in any band. Drivers would need to be careful to handle this case correctly, otherwise the offloaded model might not match the slow-path one.

Yep, although cases already exist all over the offload side.

Quoted 30 lines

In a similar vein, PRIO and DRR need to agree on the list of priorities assigned to DRR. This is doubly problematic--the user needs to take care to keep the two in sync, and the driver needs to watch for any holes in DRR coverage and treat the traffic correctly, as discussed above.

Note that at the time that DRR Qdisc is added, it has no classes, and thus any priorities assigned to that PRIO band are not covered. Thus this case is surprisingly rather common, and needs to be handled gracefully by the driver.

- Similarly due to DRR flexibility, when a Qdisc (such as RED) is attached below it, it is not immediately clear which TC the class represents. This is unlike PRIO with its straightforward classid scheme. When DRR is combined with PRIO, the relationship between classes and TCs gets even more murky.

This is a problem for users as well: the TC mapping is rather important for (devlink) shared buffer configuration and (ethtool) counters.

So instead, this patch set introduces a new Qdisc, which is based on 802.1Qaz 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 ETS, 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.

Quoted 4 lines

The chosen interface makes the overall system both reasonably easy to configure, and reasonably easy to offload. The extra code to support ETS in mlxsw (which already supports PRIO) is about 150 lines, of which perhaps 20 lines is bona fide new business logic.

Sorry maybe obvious question but I couldn't sort it out. When the qdisc is offloaded if packets are sent via software stack do they also hit the sw side qdisc enqueue logic? Or did I miss something in the graft logic that then skips adding the qdisc to software side? For example taprio has dequeue logic for both offload and software cases but I don't see that here.

Quoted 6 lines

Credit-based shaping transmission selection algorithm can be configured by adding a CBS Qdisc under one of the strict bands (e.g. TBF can be used to a similar effect as well). As a non-work-conserving Qdisc, CBS can't be hooked under the ETS bands. This is detected and handled identically to DRR Qdisc at runtime. Note that offloading CBS is not subject of this patchset.

Any performance data showing how accurate we get on software side? The advantage of hardware always to me seemed to be precision in the WRR algorithm. Also data showing how much overhead we get hit with from basic mq case would help me understand if this is even useful for software or just a exercise in building some offload logic.

FWIW I like the idea I meant to write an ETS sw qdisc for years with the expectation that it could get close enough to hardware offload case for most use cases, all but those that really need <5% tolerance or something.

Thanks!
John

Quoted 196 lines

The patchset proceeds in four stages:

- Patches #1-#3 are cleanups.
- Patches #4 and #5 contain the new Qdisc.
- Patches #6 and #7 update mlxsw to offload the new Qdisc.
- Patches #8-#10 add selftests for ETS.

Examples:

- Add a Qdisc with 6 bands, 3 strict and 3 ETS with 45%-30%-25% weights:

    # tc qdisc add dev swp1 root handle 1: \
	ets strict 3 quanta 4500 3000 2500 priomap 0 1 1 1 2 3 4 5
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 6 strict 3 quanta 4500 3000 2500 priomap 0 1 1 1 2 3 4 5 5 5 5 5 5 5 5 5 

- Tweak quantum of one of the classes of the previous Qdisc:

    # tc class ch dev swp1 classid 1:4 ets quantum 1000
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 6 strict 3 quanta 1000 3000 2500 priomap 0 1 1 1 2 3 4 5 5 5 5 5 5 5 5 5 
    # tc class ch dev swp1 classid 1:3 ets quantum 1000
    Error: Strict bands do not have a configurable quantum.

- Purely strict Qdisc with 1:1 mapping between priorities and TCs:

    # tc qdisc add dev swp1 root handle 1: \
	ets strict 8 priomap 7 6 5 4 3 2 1 0
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 8 strict 8 priomap 7 6 5 4 3 2 1 0 7 7 7 7 7 7 7 7 

- Use "bands" to specify number of bands explicitly. Underspecified bands are implicitly ETS and their quantum is taken from MTU. The following thus gives each band the same weight:

    # tc qdisc add dev swp1 root handle 1: \
	ets bands 8 priomap 7 6 5 4 3 2 1 0
    # tc qdisc sh dev swp1
    qdisc ets 1: root refcnt 2 bands 8 quanta 1514 1514 1514 1514 1514 1514 1514 1514 priomap 7 6 5 4 3 2 1 0 7 7 7 7 7 7 7 7 

v2:
- This addresses points raised by David Miller.
- Patch #4:
- sch_ets.c: Add a comment with description of the Qdisc and the dequeuing algorithm.
- Kconfig: Add a high-level description to the help blurb.

v1:
- No changes, first upstream submission after RFC.

v3 (internal):
- This addresses review from Jiri Pirko.
- Patch #3:
- Rename to _HR_ instead of to _HIERARCHY_.
- Patch #4:
- pkt_sched.h: Keep all the TCA_ETS_ constants in one enum.
- pkt_sched.h: Rename TCA_ETS_BANDS to _NBANDS, _STRICT to _NSTRICT, _BAND_QUANTUM to _QUANTA_BAND and _PMAP_BAND to _PRIOMAP_BAND.
- sch_ets.c: Update to reflect the above changes. Add a new policy, ets_class_policy, which is used when parsing class changes. Currently that policy is the same as the quanta policy, but that might change.
- sch_ets.c: Move MTU handling from ets_quantum_parse() to the one caller that makes use of it.
- sch_ets.c: ets_qdisc_priomap_parse(): WARN_ON_ONCE on invalid attribute instead of returning an extack.
- Patch #6:
- __mlxsw_sp_qdisc_ets_replace(): Pass the weights argument to this function in this patch already. Drop the weight computation.
- mlxsw_sp_qdisc_prio_replace(): Rename "quanta" to "zeroes" and pass for the abovementioned "weights".
- mlxsw_sp_qdisc_prio_graft(): Convert to a wrapper around __mlxsw_sp_qdisc_ets_graft(), instead of invoking the latter directly from mlxsw_sp_setup_tc_prio().
- Update to follow the _HIERARCHY_ -> _HR_ renaming.
- Patch #7:
- __mlxsw_sp_qdisc_ets_replace(): The "weights" argument passing and weight computation removal are now done in a previous patch.
- mlxsw_sp_setup_tc_ets(): Drop case TC_ETS_REPLACE, which is handled earlier in the function.
- Patch #3 (iproute2):
- Add an example output to the commit message.
- tc-ets.8: Fix output of two examples.
- tc-ets.8: Describe default values of "bands", "quanta".
- q_ets.c: A number of fixes in error messages.
- q_ets.c: Comment formatting: /*padding*/ -> /* padding */
- q_ets.c: parse_nbands: Move duplicate checking to callers.
- q_ets.c: Don't accept both "quantum" and "quanta" as equivalent.

v2 (internal):
- This addresses review from Ido Schimmel and comments from Alexander Kushnarov.
- Patch #2:
- s/coment/comment in the commit message.
- Patch #4:
- sch_ets: ets_class_is_strict(), ets_class_id(): Constify an argument
- ets_class_find(): RXTify
- Patch #3 (iproute2):
- tc-ets.8: some spelling fixes
- tc-ets.8: add another example
- tc.8: add an ETS to "CLASSFUL QDISCS" section

v1 (internal):
- This addresses RFC reviews from Ido Schimmel and Roman Mashak, bugs found by Alexander Petrovskiy and myself, and other improvements.
- Patch #2:
- Expand the explanation with an explicit example.
- Patch #4:
- Kconfig: s/sch_drr/sch_ets/
- sch_ets: Reorder includes to be in alphabetical order
- sch_ets: ets_quantum_parse(): Rename the return-pointer argument from pquantum to quantum, and use it directly, not going through a local temporary.
- sch_ets: ets_qdisc_quanta_parse(): Convert syntax of function argument "quanta" from an array to a pointer.
- sch_ets: ets_qdisc_priomap_parse(): Likewise with "priomap".
- sch_ets: ets_qdisc_quanta_parse(), ets_qdisc_priomap_parse(): Invoke __nla_validate_nested directly instead of nl80211_validate_nested().
- sch_ets: ets_qdisc_quanta_parse(): WARN_ON_ONCE on invalid attribute instead of returning an extack.
- sch_ets: ets_qdisc_change(): Make the last band the default one for unmentioned priomap priorities.
- sch_ets: Fix a panic when an offloaded child in a bandwidth-sharing band notified its ETS parent.
- sch_ets: When ungrafting, add the newly-created invisible FIFO to the Qdisc hash
- Patch #5:
- pkt_cls.h: Note that quantum=0 signifies a strict band.
- Fix error path handling when ets_offload_dump() fails.
- Patch #6:
- __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function arguments "quanta" and "priomap" from arrays to pointers.
- Patch #7:
- __mlxsw_sp_qdisc_ets_replace(): Convert syntax of function argument "weights" from an array to a pointer.
- Patch #9:
- mlxsw/sch_ets.sh: Add a comment explaining packet prioritization.
- Adjust the whole suite to allow testing of traffic classifiers in addition to testing priomap.
- Patch #10:
- Add a number of new tests to test default priomap band, overlarge number of bands, zeroes in quanta, and altogether missing quanta.
- Patch #1 (iproute2):
- State motivation for inclusion of this patch in the patcheset in the commit message.
- Patch #3 (iproute2):
- tc-ets.8: it is now December
- tc-ets.8: explain inactivity WRT using non-WC Qdiscs under ETS band
- tc-ets.8: s/flow/band in explanation of quantum
- tc-ets.8: explain what happens with priorities not covered by priomap
- tc-ets.8: default priomap band is now the last one
- q_ets.c: ets_parse_opt(): Remove unnecessary initialization of priomap and quanta.

Petr Machata (10):
net: pkt_cls: Clarify a comment
mlxsw: spectrum_qdisc: Clarify a comment
mlxsw: spectrum: Rename MLXSW_REG_QEEC_HIERARCY_* enumerators
net: sch_ets: Add a new Qdisc
net: sch_ets: Make the ETS qdisc offloadable
mlxsw: spectrum_qdisc: Generalize PRIO offload to support ETS
mlxsw: spectrum_qdisc: Support offloading of ETS Qdisc
selftests: forwarding: Move start_/stop_traffic from mlxsw to lib.sh
selftests: forwarding: sch_ets: Add test coverage for ETS Qdisc
selftests: qdiscs: Add test coverage for ETS Qdisc

 drivers/net/ethernet/mellanox/mlxsw/reg.h     |  11 +-
 .../net/ethernet/mellanox/mlxsw/spectrum.c    |  21 +-
 .../net/ethernet/mellanox/mlxsw/spectrum.h    |   2 +
 .../ethernet/mellanox/mlxsw/spectrum_dcb.c    |   8 +-
 .../ethernet/mellanox/mlxsw/spectrum_qdisc.c  | 219 +++-
 include/linux/netdevice.h                     |   1 +
 include/net/pkt_cls.h                         |  36 +-
 include/uapi/linux/pkt_sched.h                |  17 +
 net/sched/Kconfig                             |  17 +
 net/sched/Makefile                            |   1 +
 net/sched/sch_ets.c                           | 828 +++++++++++++++
 .../selftests/drivers/net/mlxsw/qos_lib.sh    |  46 +-
 .../selftests/drivers/net/mlxsw/sch_ets.sh    |  67 ++
 tools/testing/selftests/net/forwarding/lib.sh |  18 +
 .../selftests/net/forwarding/sch_ets.sh       |  44 +
 .../selftests/net/forwarding/sch_ets_core.sh  | 300 ++++++
 .../selftests/net/forwarding/sch_ets_tests.sh | 227 +++++
 .../tc-testing/tc-tests/qdiscs/ets.json       | 940 ++++++++++++++++++
 18 files changed, 2732 insertions(+), 71 deletions(-)
 create mode 100644 net/sched/sch_ets.c
 create mode 100755 tools/testing/selftests/drivers/net/mlxsw/sch_ets.sh
 create mode 100755 tools/testing/selftests/net/forwarding/sch_ets.sh
 create mode 100644 tools/testing/selftests/net/forwarding/sch_ets_core.sh
 create mode 100644 tools/testing/selftests/net/forwarding/sch_ets_tests.sh
 create mode 100644 tools/testing/selftests/tc-testing/tc-tests/qdiscs/ets.json
Signature · 1 line
2.20.1
#13 lore raw
Quoted 15 lines · John Fastabend
Quoted 5 lines · Petr Machata

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal transmission selection algorithms: strict priority, credit-based shaper, ETS (bandwidth sharing), and vendor-specific. All these have their corresponding knobs in DCB. But DCB does not have interfaces to configure RED and ECN, unlike Qdiscs.

So the idea here (way back when I did this years ago) is that marking ECN traffic was not paticularly CPU intensive on any metrics I came up with. And I don't recall anyone ever wanting to do RED here. The configuration I usually recommended was to use mqprio + SO_PRIORITY + fq per qdisc. Then once we got the BPF egress hook we replaced SO_PRIORITY configurations with the more dynamic BPF action to set it. There was never a compelling perf reason to offload red/ecn.

Quoted 5 lines

But these use cases were edge nodes. I believe this series is mostly about control path and maybe some light control traffic? This is for switches not for edge nodes right? I'm guessing because I don't see any performance analaysis on why this is useful, intuitively it makes sense if there is a small CPU sitting on a 48 port 10gbps box or something like that.

Yes.

Our particular use case is a switch that has throughput in Tbps. There simply isn't enough bandwidth to even get all this traffic to the CPU, let alone process it on the CPU. You need to offload, or it doesn't make sense. 48 x 10Gbps with a small CPU is like that as well, yeah.

From what I hear, RED / ECN was not used very widely in these sorts of deployments, rather the deal was to have more bandwidth than you need and not worry about QoS. This is changing, and people experiment with this stuff more. So there is interest in strict vs. DWRR TCs, shapers, and RED / ECN.

Quoted 9 lines
Quoted 8 lines

In the Qdisc land, strict priority is implemented by PRIO. Credit-based transmission selection algorithm can then be modeled by having e.g. TBF or CBS Qdisc below some of the PRIO bands. ETS would then be modeled by placing a DRR Qdisc under the last PRIO band.

The problem with this approach is that DRR on its own, as well as the combination of PRIO and DRR, are tricky to configure and tricky to offload to 802.1Qaz-compliant hardware. This is due to several reasons:

Quoted 6 lines

I would argue the trick to configure part could be hid behind tooling to simplify setup. The more annoying part is it was stuck behind the qdisc lock. I was hoping this would implement a lockless ETS qdisc seeing we have the infra to do lockless qdiscs now. But seems not. I guess software perf analysis might show prio+drr and ets here are about the same performance wise.

Pretty sure. It's the same algorithm, and I would guess that the one extra virtual call will not throw it off.

Quoted 1 line

offload is tricky with stacked qdiscs though ;)

Offload and configuration both.

Of course there could be a script to somehow generate and parse the configuration on the front end, and some sort of library to consolidate on the driver side, but it's far cleaner and easier to understand for all involved if it's a Qdisc. Qdiscs are tricky, but people still understand them well in comparison.

Quoted 66 lines
Quoted 23 lines

- As any classful Qdisc, DRR supports adding classifiers to decide in which class to enqueue packets. Unlike PRIO, there's however no fallback in the form of priomap. A way to achieve classification based on packet priority is e.g. like this:

    # tc filter add dev swp1 root handle 1: \
		basic match 'meta(priority eq 0)' flowid 1:10

Expressing the priomap in this manner however forces drivers to deep dive into the classifier block to parse the individual rules.

A possible solution would be to extend the classes with a "defmap" a la split / defmap mechanism of CBQ, and introduce this as a last resort classification. However, unlike priomap, this doesn't have the guarantee of covering all priorities. Traffic whose priority is not covered is dropped by DRR as unclassified. But ASICs tend to implement dropping in the ACL block, not in scheduling pipelines. The need to treat these configurations correctly (if only to decide to not offload at all) complicates a driver.

It's not clear how to retrofit priomap with all its benefits to DRR without changing it beyond recognition.

Quoted 6 lines

- The interplay between PRIO and DRR is also causing problems. 802.1Qaz has all ETS TCs as a last resort. Switch ASICs that support ETS at all are likely to handle ETS traffic this way as well. However, the Linux model is more generic, allowing the DRR block in any band. Drivers would need to be careful to handle this case correctly, otherwise the offloaded model might not match the slow-path one.

Yep, although cases already exist all over the offload side.

Quoted 30 lines

In a similar vein, PRIO and DRR need to agree on the list of priorities assigned to DRR. This is doubly problematic--the user needs to take care to keep the two in sync, and the driver needs to watch for any holes in DRR coverage and treat the traffic correctly, as discussed above.

Note that at the time that DRR Qdisc is added, it has no classes, and thus any priorities assigned to that PRIO band are not covered. Thus this case is surprisingly rather common, and needs to be handled gracefully by the driver.

- Similarly due to DRR flexibility, when a Qdisc (such as RED) is attached below it, it is not immediately clear which TC the class represents. This is unlike PRIO with its straightforward classid scheme. When DRR is combined with PRIO, the relationship between classes and TCs gets even more murky.

This is a problem for users as well: the TC mapping is rather important for (devlink) shared buffer configuration and (ethtool) counters.

So instead, this patch set introduces a new Qdisc, which is based on 802.1Qaz 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 ETS, 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.

Quoted 4 lines

The chosen interface makes the overall system both reasonably easy to configure, and reasonably easy to offload. The extra code to support ETS in mlxsw (which already supports PRIO) is about 150 lines, of which perhaps 20 lines is bona fide new business logic.

Quoted 5 lines

Sorry maybe obvious question but I couldn't sort it out. When the qdisc is offloaded if packets are sent via software stack do they also hit the sw side qdisc enqueue logic? Or did I miss something in the graft logic that then skips adding the qdisc to software side? For example taprio has dequeue logic for both offload and software cases but I don't see that here.

You mean the graft logic in the driver? All that stuff is in there just to figure out how to configure the device. SW datapath packets are still handled as usual.

There even is a selftest for the SW datapath that uses veth pairs to implement interconnect and TBF to throttle it (so that the scheduling kicks in).

Quoted 9 lines
Quoted 6 lines

Credit-based shaping transmission selection algorithm can be configured by adding a CBS Qdisc under one of the strict bands (e.g. TBF can be used to a similar effect as well). As a non-work-conserving Qdisc, CBS can't be hooked under the ETS bands. This is detected and handled identically to DRR Qdisc at runtime. Note that offloading CBS is not subject of this patchset.

Any performance data showing how accurate we get on software side? The advantage of hardware always to me seemed to be precision in the WRR algorithm.

Quantum is specified as a number of bytes allowed to dequeue before a queue loses the medium. Over time, the amount of traffic dequeued from individual queues should average out to be the quanta your specified. At any point in time, size of the packets matters: if I push 1000B packets into a 10000B-quantum queue, it will use 100% of its allocation. If they are 800B packets, there will be some waste (and it will compensate next round).

As far as the Qdisc is defined, the SW side is as accurate as possible under given traffic patterns. For HW, we translate to %, and rounding might lead to artifacts. You kinda get the same deal with DCB, where there's no way to split 100% among 8 TCs perfectly fairly.

Quoted 3 lines

Also data showing how much overhead we get hit with from basic mq case would help me understand if this is even useful for software or just a exercise in building some offload logic.

So the Qdisc is written to do something reasonable in the SW datapath. In that respect it's as useful as PRIO and DRR are. Not sure that as a switch operator you really want to handle this much traffic on the CPU though.

Quoted 3 lines

FWIW I like the idea I meant to write an ETS sw qdisc for years with the expectation that it could get close enough to hardware offload case for most use cases, all but those that really need <5% tolerance or something.

#14 lore raw
Quoted 24 lines · Petr Machata
Quoted 15 lines · John Fastabend
Quoted 5 lines · Petr Machata

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal transmission selection algorithms: strict priority, credit-based shaper, ETS (bandwidth sharing), and vendor-specific. All these have their corresponding knobs in DCB. But DCB does not have interfaces to configure RED and ECN, unlike Qdiscs.

So the idea here (way back when I did this years ago) is that marking ECN traffic was not paticularly CPU intensive on any metrics I came up with. And I don't recall anyone ever wanting to do RED here. The configuration I usually recommended was to use mqprio + SO_PRIORITY + fq per qdisc. Then once we got the BPF egress hook we replaced SO_PRIORITY configurations with the more dynamic BPF action to set it. There was never a compelling perf reason to offload red/ecn.

Quoted 5 lines

But these use cases were edge nodes. I believe this series is mostly about control path and maybe some light control traffic? This is for switches not for edge nodes right? I'm guessing because I don't see any performance analaysis on why this is useful, intuitively it makes sense if there is a small CPU sitting on a 48 port 10gbps box or something like that.

Quoted 6 lines

Yes.

Our particular use case is a switch that has throughput in Tbps. There simply isn't enough bandwidth to even get all this traffic to the CPU, let alone process it on the CPU. You need to offload, or it doesn't make sense. 48 x 10Gbps with a small CPU is like that as well, yeah.

Got it so I suspect primary usage will be offload then at least for the initial usage.

Quoted 23 lines

From what I hear, RED / ECN was not used very widely in these sorts of deployments, rather the deal was to have more bandwidth than you need and not worry about QoS. This is changing, and people experiment with this stuff more. So there is interest in strict vs. DWRR TCs, shapers, and RED / ECN.

Quoted 9 lines
Quoted 8 lines

In the Qdisc land, strict priority is implemented by PRIO. Credit-based transmission selection algorithm can then be modeled by having e.g. TBF or CBS Qdisc below some of the PRIO bands. ETS would then be modeled by placing a DRR Qdisc under the last PRIO band.

The problem with this approach is that DRR on its own, as well as the combination of PRIO and DRR, are tricky to configure and tricky to offload to 802.1Qaz-compliant hardware. This is due to several reasons:

Quoted 6 lines

I would argue the trick to configure part could be hid behind tooling to simplify setup. The more annoying part is it was stuck behind the qdisc lock. I was hoping this would implement a lockless ETS qdisc seeing we have the infra to do lockless qdiscs now. But seems not. I guess software perf analysis might show prio+drr and ets here are about the same performance wise.

Quoted 2 lines

Pretty sure. It's the same algorithm, and I would guess that the one extra virtual call will not throw it off.

Yeah small in comparison to other performance issues I would guess.

Quoted 10 lines
Quoted 2 lines

offload is tricky with stacked qdiscs though ;)

Offload and configuration both.

Of course there could be a script to somehow generate and parse the configuration on the front end, and some sort of library to consolidate on the driver side, but it's far cleaner and easier to understand for all involved if it's a Qdisc. Qdiscs are tricky, but people still understand them well in comparison.

At one point I wrote an app to sit on top of the tc netlink interface and create common (at least for the customers at the time) setups. But that tool is probably lost to history at this point.

I don't think its paticularly difficult to build this type of tool on top of the API but also not against a new qdisc like this that folds in a more concrete usage and aligns with a spec. And Dave already merged it so good to see ;)

[...]

Quoted 11 lines
Quoted 10 lines
Quoted 4 lines

The chosen interface makes the overall system both reasonably easy to configure, and reasonably easy to offload. The extra code to support ETS in mlxsw (which already supports PRIO) is about 150 lines, of which perhaps 20 lines is bona fide new business logic.

Sorry maybe obvious question but I couldn't sort it out. When the qdisc is offloaded if packets are sent via software stack do they also hit the sw side qdisc enqueue logic? Or did I miss something in the graft logic that then skips adding the qdisc to software side? For example taprio has dequeue logic for both offload and software cases but I don't see that here.

Quoted 3 lines

You mean the graft logic in the driver? All that stuff is in there just to figure out how to configure the device. SW datapath packets are still handled as usual.

Got it just wasn't clear to me when viewing it from the software + smartnic use case. So is there a bug or maybe just missing feature, where if I offloaded this on a NIC that both software and hardware would do the ETS algorithm? How about on the switch would traffic from the CPU be both ETS classified in software and in hardware? Or maybe CPU uses different interface without offload on?

Quoted 4 lines

There even is a selftest for the SW datapath that uses veth pairs to implement interconnect and TBF to throttle it (so that the scheduling kicks in).

+1

Quoted 24 lines
Quoted 10 lines
Quoted 7 lines

Credit-based shaping transmission selection algorithm can be configured by adding a CBS Qdisc under one of the strict bands (e.g. TBF can be used to a similar effect as well). As a non-work-conserving Qdisc, CBS can't be hooked under the ETS bands. This is detected and handled identically to DRR Qdisc at runtime. Note that offloading CBS is not subject of this patchset.

Any performance data showing how accurate we get on software side? The advantage of hardware always to me seemed to be precision in the WRR algorithm.

Quantum is specified as a number of bytes allowed to dequeue before a queue loses the medium. Over time, the amount of traffic dequeued from individual queues should average out to be the quanta your specified. At any point in time, size of the packets matters: if I push 1000B packets into a 10000B-quantum queue, it will use 100% of its allocation. If they are 800B packets, there will be some waste (and it will compensate next round).

As far as the Qdisc is defined, the SW side is as accurate as possible under given traffic patterns. For HW, we translate to %, and rounding might lead to artifacts. You kinda get the same deal with DCB, where there's no way to split 100% among 8 TCs perfectly fairly.

Quoted 8 lines
Quoted 3 lines

Also data showing how much overhead we get hit with from basic mq case would help me understand if this is even useful for software or just a exercise in building some offload logic.

So the Qdisc is written to do something reasonable in the SW datapath. In that respect it's as useful as PRIO and DRR are. Not sure that as a switch operator you really want to handle this much traffic on the CPU though.

I was more thinking of using it in the smart nic case.

Quoted 4 lines
Quoted 4 lines

FWIW I like the idea I meant to write an ETS sw qdisc for years with the expectation that it could get close enough to hardware offload case for most use cases, all but those that really need <5% tolerance or something.

Anyways thanks for the answers clears it up on my side. One remaining question is if software does send packets if they get both classified via software and hardware. Might be worth thinking about fixing if that is the case or probably more likely switch knows not to do this.

Thanks,
John

#15 lore raw
Quoted 27 lines · John Fastabend
Quoted 26 lines · Petr Machata
Quoted 15 lines · John Fastabend
Quoted 5 lines · Petr Machata

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal transmission selection algorithms: strict priority, credit-based shaper, ETS (bandwidth sharing), and vendor-specific. All these have their corresponding knobs in DCB. But DCB does not have interfaces to configure RED and ECN, unlike Qdiscs.

So the idea here (way back when I did this years ago) is that marking ECN traffic was not paticularly CPU intensive on any metrics I came up with. And I don't recall anyone ever wanting to do RED here. The configuration I usually recommended was to use mqprio + SO_PRIORITY + fq per qdisc. Then once we got the BPF egress hook we replaced SO_PRIORITY configurations with the more dynamic BPF action to set it. There was never a compelling perf reason to offload red/ecn.

Quoted 5 lines

But these use cases were edge nodes. I believe this series is mostly about control path and maybe some light control traffic? This is for switches not for edge nodes right? I'm guessing because I don't see any performance analaysis on why this is useful, intuitively it makes sense if there is a small CPU sitting on a 48 port 10gbps box or something like that.

Yes.

Quoted 7 lines
Quoted 4 lines

Our particular use case is a switch that has throughput in Tbps. There simply isn't enough bandwidth to even get all this traffic to the CPU, let alone process it on the CPU. You need to offload, or it doesn't make sense. 48 x 10Gbps with a small CPU is like that as well, yeah.

Got it so I suspect primary usage will be offload then at least for the initial usage.

Yes, particularly configuration of offloaded forwarding path.

Quoted 36 lines
Quoted 9 lines
Quoted 1 line

offload is tricky with stacked qdiscs though ;)

Offload and configuration both.

Of course there could be a script to somehow generate and parse the configuration on the front end, and some sort of library to consolidate on the driver side, but it's far cleaner and easier to understand for all involved if it's a Qdisc. Qdiscs are tricky, but people still understand them well in comparison.

At one point I wrote an app to sit on top of the tc netlink interface and create common (at least for the customers at the time) setups. But that tool is probably lost to history at this point.

I don't think its paticularly difficult to build this type of tool on top of the API but also not against a new qdisc like this that folds in a more concrete usage and aligns with a spec. And Dave already merged it so good to see ;)

[...]

Quoted 11 lines
Quoted 10 lines
Quoted 4 lines

The chosen interface makes the overall system both reasonably easy to configure, and reasonably easy to offload. The extra code to support ETS in mlxsw (which already supports PRIO) is about 150 lines, of which perhaps 20 lines is bona fide new business logic.

Sorry maybe obvious question but I couldn't sort it out. When the qdisc is offloaded if packets are sent via software stack do they also hit the sw side qdisc enqueue logic? Or did I miss something in the graft logic that then skips adding the qdisc to software side? For example taprio has dequeue logic for both offload and software cases but I don't see that here.

Quoted 3 lines

You mean the graft logic in the driver? All that stuff is in there just to figure out how to configure the device. SW datapath packets are still handled as usual.

Quoted 6 lines

Got it just wasn't clear to me when viewing it from the software + smartnic use case. So is there a bug or maybe just missing feature, where if I offloaded this on a NIC that both software and hardware would do the ETS algorithm? How about on the switch would traffic from the CPU be both ETS classified in software and in hardware? Or maybe CPU uses different interface without offload on?

You would get SW scheduling if there's more traffic than the host interface can handle.

In the HW, control traffic gets TC 16, which the chip hardcodes as the highest priority and handles in a dedicated set of queues. So there's no second classification.

Quoted 26 lines
Quoted 26 lines
Quoted 8 lines
Quoted 5 lines

Credit-based shaping transmission selection algorithm can be configured by adding a CBS Qdisc under one of the strict bands (e.g. TBF can be used to a similar effect as well). As a non-work-conserving Qdisc, CBS can't be hooked under the ETS bands. This is detected and handled identically to DRR Qdisc at runtime. Note that offloading CBS is not subject of this patchset.

Any performance data showing how accurate we get on software side? The advantage of hardware always to me seemed to be precision in the WRR algorithm.

Quantum is specified as a number of bytes allowed to dequeue before a queue loses the medium. Over time, the amount of traffic dequeued from individual queues should average out to be the quanta your specified. At any point in time, size of the packets matters: if I push 1000B packets into a 10000B-quantum queue, it will use 100% of its allocation. If they are 800B packets, there will be some waste (and it will compensate next round).

As far as the Qdisc is defined, the SW side is as accurate as possible under given traffic patterns. For HW, we translate to %, and rounding might lead to artifacts. You kinda get the same deal with DCB, where there's no way to split 100% among 8 TCs perfectly fairly.

Quoted 3 lines

Also data showing how much overhead we get hit with from basic mq case would help me understand if this is even useful for software or just a exercise in building some offload logic.

Quoted 6 lines
Quoted 4 lines

So the Qdisc is written to do something reasonable in the SW datapath. In that respect it's as useful as PRIO and DRR are. Not sure that as a switch operator you really want to handle this much traffic on the CPU though.

I was more thinking of using it in the smart nic case.

I'm not really familiar with this.

I can imagine some knobs that map the individual bands to NIC queues for example. I think that's something that mlxsw_spectrum could actually use. We do have several queues to the chip, and currently round-robin them by hand in the driver. Logic to determine which queues to use for which traffic seems to make sense. But currently we simply don't see these use cases at all.

Quoted 9 lines
Quoted 3 lines
Quoted 3 lines

FWIW I like the idea I meant to write an ETS sw qdisc for years with the expectation that it could get close enough to hardware offload case for most use cases, all but those that really need <5% tolerance or something.

Anyways thanks for the answers clears it up on my side. One remaining question is if software does send packets if they get both classified via software and hardware. Might be worth thinking about fixing if that is the case or probably more likely switch knows not to do this.

Yeah, traffic from the CPU is handled specially.

maintainer
#16 lore raw
In reply to Petr Machata

From: Petr Machata <[email protected]>
Date: Wed, 18 Dec 2019 14:55:06 +0000

Quoted 14 lines

The IEEE standard 802.1Qaz (and 802.1Q-2014) specifies four principal transmission selection algorithms: strict priority, credit-based shaper, ETS (bandwidth sharing), and vendor-specific. All these have their corresponding knobs in DCB. But DCB does not have interfaces to configure RED and ECN, unlike Qdiscs.

In the Qdisc land, strict priority is implemented by PRIO. Credit-based transmission selection algorithm can then be modeled by having e.g. TBF or CBS Qdisc below some of the PRIO bands. ETS would then be modeled by placing a DRR Qdisc under the last PRIO band.

The problem with this approach is that DRR on its own, as well as the combination of PRIO and DRR, are tricky to configure and tricky to offload to 802.1Qaz-compliant hardware. This is due to several reasons:

...

Quoted 4 lines

So instead, this patch set introduces a new Qdisc, which is based on 802.1Qaz 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 ETS, quanta for the latter, and priomap.

...

Series applied, thanks.