[PATCH net v3 0/2] taprio: Some fixes

[PATCH net v3 0/2] taprio: Some fixes

[Vinicius Costa Gomes]

Hi,

Changes from v2:
  - Squashed commits 2/3 and 3/3 into a single one (I think a single
    commit is going to be easier to review);
  - Removed an "improvement" that was causing changes in user visible
    behavior;

Changes from v1:
  - Fixed ignoring the 'flags' argument when adding a new
    instance (Vladimir Oltean);
  - Changed the order of commits;

Updated cover letter:

Some fixes for taprio:

Patch 1/2: Reported by Po Liu, is more of a improvement of usability for
drivers implementing offloading features, now they can rely on the
value of dev->num_tc, instead of going through some hops to get this
value.

Patch 2/2: Use 'q->flags' as the source of truth for the offloading
flags. Tries to solidify the current behavior, while avoiding going
into invalid states, one of which was causing a "rcu stall" (more
information in the commit message).

@Vladimir: If possible, I would appreciate your Ack on patch 2/2. I
have been looking at this code for so long that I might have missed
something obvious (and my growing dislike for the word 'flags' may be
affecting my judgement :-).


Vinicius Costa Gomes (2):
  taprio: Fix enabling offload with wrong number of traffic classes
  taprio: Fix still allowing changing the flags during runtime

 net/sched/sch_taprio.c | 87 ++++++++++++++++++++++++++----------------
 1 file changed, 54 insertions(+), 33 deletions(-)

-- 
2.25.0

[PATCH net v3 1/2] taprio: Fix enabling offload with wrong number of traffic classes

[Vinicius Costa Gomes]

If the driver implementing taprio offloading depends on the value of
the network device number of traffic classes (dev->num_tc) for
whatever reason, it was going to receive the value zero. The value was
only set after the offloading function is called.

So, moving setting the number of traffic classes to before the
offloading function is called fixes this issue. This is safe because
this only happens when taprio is instantiated (we don't allow this
configuration to be changed without first removing taprio).

Fixes: 9c66d1564676 ("taprio: Add support for hardware offloading")
Reported-by: Po Liu <po.liu@nxp.com>
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Acked-by: Vladimir Oltean <vladimir.oltean@nxp.com>
---
 net/sched/sch_taprio.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/net/sched/sch_taprio.c b/net/sched/sch_taprio.c
index c609373c8661..ad0dadcfcdba 100644
--- a/net/sched/sch_taprio.c
+++ b/net/sched/sch_taprio.c
@@ -1444,6 +1444,19 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 
 	taprio_set_picos_per_byte(dev, q);
 
+	if (mqprio) {
+		netdev_set_num_tc(dev, mqprio->num_tc);
+		for (i = 0; i < mqprio->num_tc; i++)
+			netdev_set_tc_queue(dev, i,
+					    mqprio->count[i],
+					    mqprio->offset[i]);
+
+		/* Always use supplied priority mappings */
+		for (i = 0; i <= TC_BITMASK; i++)
+			netdev_set_prio_tc_map(dev, i,
+					       mqprio->prio_tc_map[i]);
+	}
+
 	if (FULL_OFFLOAD_IS_ENABLED(taprio_flags))
 		err = taprio_enable_offload(dev, mqprio, q, new_admin, extack);
 	else
@@ -1471,19 +1484,6 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 		q->advance_timer.function = advance_sched;
 	}
 
-	if (mqprio) {
-		netdev_set_num_tc(dev, mqprio->num_tc);
-		for (i = 0; i < mqprio->num_tc; i++)
-			netdev_set_tc_queue(dev, i,
-					    mqprio->count[i],
-					    mqprio->offset[i]);
-
-		/* Always use supplied priority mappings */
-		for (i = 0; i <= TC_BITMASK; i++)
-			netdev_set_prio_tc_map(dev, i,
-					       mqprio->prio_tc_map[i]);
-	}
-
 	if (FULL_OFFLOAD_IS_ENABLED(taprio_flags)) {
 		q->dequeue = taprio_dequeue_offload;
 		q->peek = taprio_peek_offload;
-- 
2.25.0

[PATCH net v3 2/2] taprio: Fix still allowing changing the flags during runtime

[Vinicius Costa Gomes]

Because 'q->flags' starts as zero, and zero is a valid value, we
aren't able to detect the transition from zero to something else
during "runtime".

The solution is to initialize 'q->flags' with an invalid value, so we
can detect if 'q->flags' was set by the user or not.

To better solidify the behavior, 'flags' handling is moved to a
separate function. The behavior is:
 - 'flags' if unspecified by the user, is assumed to be zero;
 - 'flags' cannot change during "runtime" (i.e. a change() request
 cannot modify it);

Allowing flags to be changed was causing the following RCU stall:

[ 1730.558249] rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
[ 1730.558258] rcu: 	  6-...0: (190 ticks this GP) idle=922/0/0x1 softirq=25580/25582 fqs=16250
[ 1730.558264] 		  (detected by 2, t=65002 jiffies, g=33017, q=81)
[ 1730.558269] Sending NMI from CPU 2 to CPUs 6:
[ 1730.559277] NMI backtrace for cpu 6
[ 1730.559277] CPU: 6 PID: 0 Comm: swapper/6 Tainted: G            E     5.5.0-rc6+ #35
[ 1730.559278] Hardware name: Gigabyte Technology Co., Ltd. Z390 AORUS ULTRA/Z390 AORUS ULTRA-CF, BIOS F7 03/14/2019
[ 1730.559278] RIP: 0010:__hrtimer_run_queues+0xe2/0x440
[ 1730.559278] Code: 48 8b 43 28 4c 89 ff 48 8b 75 c0 48 89 45 c8 e8 f4 bb 7c 00 0f 1f 44 00 00 65 8b 05 40 31 f0 68 89 c0 48 0f a3 05 3e 5c 25 01 <0f> 82 fc 01 00 00 48 8b 45 c8 48 89 df ff d0 89 45 c8 0f 1f 44 00
[ 1730.559279] RSP: 0018:ffff9970802d8f10 EFLAGS: 00000083
[ 1730.559279] RAX: 0000000000000006 RBX: ffff8b31645bff38 RCX: 0000000000000000
[ 1730.559280] RDX: 0000000000000000 RSI: ffffffff9710f2ec RDI: ffffffff978daf0e
[ 1730.559280] RBP: ffff9970802d8f68 R08: 0000000000000000 R09: 0000000000000000
[ 1730.559280] R10: 0000018336d7944e R11: 0000000000000001 R12: ffff8b316e39f9c0
[ 1730.559281] R13: ffff8b316e39f940 R14: ffff8b316e39f998 R15: ffff8b316e39f7c0
[ 1730.559281] FS:  0000000000000000(0000) GS:ffff8b316e380000(0000) knlGS:0000000000000000
[ 1730.559281] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1730.559281] CR2: 00007f1105303760 CR3: 0000000227210005 CR4: 00000000003606e0
[ 1730.559282] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 1730.559282] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 1730.559282] Call Trace:
[ 1730.559282]  <IRQ>
[ 1730.559283]  ? taprio_dequeue_soft+0x2d0/0x2d0 [sch_taprio]
[ 1730.559283]  hrtimer_interrupt+0x104/0x220
[ 1730.559283]  ? irqtime_account_irq+0x34/0xa0
[ 1730.559283]  smp_apic_timer_interrupt+0x6d/0x230
[ 1730.559284]  apic_timer_interrupt+0xf/0x20
[ 1730.559284]  </IRQ>
[ 1730.559284] RIP: 0010:cpu_idle_poll+0x35/0x1a0
[ 1730.559285] Code: 88 82 ff 65 44 8b 25 12 7d 73 68 0f 1f 44 00 00 e8 90 c3 89 ff fb 65 48 8b 1c 25 c0 7e 01 00 48 8b 03 a8 08 74 0b eb 1c f3 90 <48> 8b 03 a8 08 75 13 8b 05 be a8 a8 00 85 c0 75 ed e8 75 48 84 ff
[ 1730.559285] RSP: 0018:ffff997080137ea8 EFLAGS: 00000202 ORIG_RAX: ffffffffffffff13
[ 1730.559285] RAX: 0000000000000001 RBX: ffff8b316bc3c580 RCX: 0000000000000000
[ 1730.559286] RDX: 0000000000000001 RSI: 000000002819aad9 RDI: ffffffff978da730
[ 1730.559286] RBP: ffff997080137ec0 R08: 0000018324a6d387 R09: 0000000000000000
[ 1730.559286] R10: 0000000000000400 R11: 0000000000000001 R12: 0000000000000006
[ 1730.559286] R13: ffff8b316bc3c580 R14: 0000000000000000 R15: 0000000000000000
[ 1730.559287]  ? cpu_idle_poll+0x20/0x1a0
[ 1730.559287]  ? cpu_idle_poll+0x20/0x1a0
[ 1730.559287]  do_idle+0x4d/0x1f0
[ 1730.559287]  ? complete+0x44/0x50
[ 1730.559288]  cpu_startup_entry+0x1b/0x20
[ 1730.559288]  start_secondary+0x142/0x180
[ 1730.559288]  secondary_startup_64+0xb6/0xc0
[ 1776.686313] nvme nvme0: I/O 96 QID 1 timeout, completion polled

Fixes: 4cfd5779bd6e ("taprio: Add support for txtime-assist mode")
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
---
 net/sched/sch_taprio.c | 61 ++++++++++++++++++++++++++++--------------
 1 file changed, 41 insertions(+), 20 deletions(-)

diff --git a/net/sched/sch_taprio.c b/net/sched/sch_taprio.c
index ad0dadcfcdba..7290de6cc0bc 100644
--- a/net/sched/sch_taprio.c
+++ b/net/sched/sch_taprio.c
@@ -31,6 +31,7 @@ static DEFINE_SPINLOCK(taprio_list_lock);
 
 #define TXTIME_ASSIST_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST)
 #define FULL_OFFLOAD_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD)
+#define TAPRIO_FLAGS_INVALID U32_MAX
 
 struct sched_entry {
 	struct list_head list;
@@ -1367,6 +1368,33 @@ static int taprio_mqprio_cmp(const struct net_device *dev,
 	return 0;
 }
 
+/* The semantics of the 'flags' argument in relation to 'change()'
+ * requests, are interpreted following two rules (which are applied in
+ * this order): (1) an omitted 'flags' argument is interpreted as
+ * zero; (2) the 'flags' of a "running" taprio instance cannot be
+ * changed.
+ */
+static int taprio_new_flags(const struct nlattr *attr, u32 old,
+			    struct netlink_ext_ack *extack)
+{
+	u32 new = 0;
+
+	if (attr)
+		new = nla_get_u32(attr);
+
+	if (old != TAPRIO_FLAGS_INVALID && old != new) {
+		NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
+		return -ENOTSUPP;
+	}
+
+	if (!taprio_flags_valid(new)) {
+		NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
+		return -EINVAL;
+	}
+
+	return new;
+}
+
 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 			 struct netlink_ext_ack *extack)
 {
@@ -1375,7 +1403,6 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 	struct taprio_sched *q = qdisc_priv(sch);
 	struct net_device *dev = qdisc_dev(sch);
 	struct tc_mqprio_qopt *mqprio = NULL;
-	u32 taprio_flags = 0;
 	unsigned long flags;
 	ktime_t start;
 	int i, err;
@@ -1388,21 +1415,14 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 	if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
 		mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
 
-	if (tb[TCA_TAPRIO_ATTR_FLAGS]) {
-		taprio_flags = nla_get_u32(tb[TCA_TAPRIO_ATTR_FLAGS]);
-
-		if (q->flags != 0 && q->flags != taprio_flags) {
-			NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
-			return -EOPNOTSUPP;
-		} else if (!taprio_flags_valid(taprio_flags)) {
-			NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
-			return -EINVAL;
-		}
+	err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
+			       q->flags, extack);
+	if (err < 0)
+		return err;
 
-		q->flags = taprio_flags;
-	}
+	q->flags = err;
 
-	err = taprio_parse_mqprio_opt(dev, mqprio, extack, taprio_flags);
+	err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
 	if (err < 0)
 		return err;
 
@@ -1457,7 +1477,7 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 					       mqprio->prio_tc_map[i]);
 	}
 
-	if (FULL_OFFLOAD_IS_ENABLED(taprio_flags))
+	if (FULL_OFFLOAD_IS_ENABLED(q->flags))
 		err = taprio_enable_offload(dev, mqprio, q, new_admin, extack);
 	else
 		err = taprio_disable_offload(dev, q, extack);
@@ -1477,14 +1497,14 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 		q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
 	}
 
-	if (!TXTIME_ASSIST_IS_ENABLED(taprio_flags) &&
-	    !FULL_OFFLOAD_IS_ENABLED(taprio_flags) &&
+	if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
+	    !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
 	    !hrtimer_active(&q->advance_timer)) {
 		hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
 		q->advance_timer.function = advance_sched;
 	}
 
-	if (FULL_OFFLOAD_IS_ENABLED(taprio_flags)) {
+	if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
 		q->dequeue = taprio_dequeue_offload;
 		q->peek = taprio_peek_offload;
 	} else {
@@ -1501,7 +1521,7 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 		goto unlock;
 	}
 
-	if (TXTIME_ASSIST_IS_ENABLED(taprio_flags)) {
+	if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 		setup_txtime(q, new_admin, start);
 
 		if (!oper) {
@@ -1528,7 +1548,7 @@ static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
 
 		spin_unlock_irqrestore(&q->current_entry_lock, flags);
 
-		if (FULL_OFFLOAD_IS_ENABLED(taprio_flags))
+		if (FULL_OFFLOAD_IS_ENABLED(q->flags))
 			taprio_offload_config_changed(q);
 	}
 
@@ -1597,6 +1617,7 @@ static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
 	 * and get the valid one on taprio_change().
 	 */
 	q->clockid = -1;
+	q->flags = TAPRIO_FLAGS_INVALID;
 
 	spin_lock(&taprio_list_lock);
 	list_add(&q->taprio_list, &taprio_list);
-- 
2.25.0

Re: [PATCH net v3 1/2] taprio: Fix enabling offload with wrong number of traffic classes

[Jakub Kicinski]

On Wed, 29 Jan 2020 17:37:20 -0800, Vinicius Costa Gomes wrote:
> If the driver implementing taprio offloading depends on the value of
> the network device number of traffic classes (dev->num_tc) for
> whatever reason, it was going to receive the value zero. The value was
> only set after the offloading function is called.
> 
> So, moving setting the number of traffic classes to before the
> offloading function is called fixes this issue. This is safe because
> this only happens when taprio is instantiated (we don't allow this
> configuration to be changed without first removing taprio).
> 
> Fixes: 9c66d1564676 ("taprio: Add support for hardware offloading")
> Reported-by: Po Liu <po.liu@nxp.com>
> Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
> Acked-by: Vladimir Oltean <vladimir.oltean@nxp.com>

What about Dave's question about resetting the tc state with
netdev_reset_tc()?

Re: [PATCH net v3 2/2] taprio: Fix still allowing changing the flags during runtime

[Jakub Kicinski]

On Wed, 29 Jan 2020 17:37:21 -0800, Vinicius Costa Gomes wrote:
> +static int taprio_new_flags(const struct nlattr *attr, u32 old,
> +			    struct netlink_ext_ack *extack)
> +{
> +	u32 new = 0;

TCA_TAPRIO_ATTR_FLAGS doesn't seem to be in the netlink policy 😖

> +	if (attr)
> +		new = nla_get_u32(attr);
> +
> +	if (old != TAPRIO_FLAGS_INVALID && old != new) {
> +		NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
> +		return -ENOTSUPP;

-EOPNOTSUPP

> +	}
> +
> +	if (!taprio_flags_valid(new)) {
> +		NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
> +		return -EINVAL;
> +	}
> +
> +	return new;
> +}

Re: [PATCH net v3 2/2] taprio: Fix still allowing changing the flags during runtime

[Vinicius Costa Gomes]

Jakub Kicinski <kuba@kernel.org> writes:

> On Wed, 29 Jan 2020 17:37:21 -0800, Vinicius Costa Gomes wrote:
>> +static int taprio_new_flags(const struct nlattr *attr, u32 old,
>> +			    struct netlink_ext_ack *extack)
>> +{
>> +	u32 new = 0;
>
> TCA_TAPRIO_ATTR_FLAGS doesn't seem to be in the netlink policy 😖

Will add it.

>
>> +	if (attr)
>> +		new = nla_get_u32(attr);
>> +
>> +	if (old != TAPRIO_FLAGS_INVALID && old != new) {
>> +		NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
>> +		return -ENOTSUPP;
>
> -EOPNOTSUPP

Will fix.

>
>> +	}
>> +
>> +	if (!taprio_flags_valid(new)) {
>> +		NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
>> +		return -EINVAL;
>> +	}
>> +
>> +	return new;
>> +}

Re: [PATCH net v3 1/2] taprio: Fix enabling offload with wrong number of traffic classes

[Vinicius Costa Gomes]

Jakub Kicinski <kuba@kernel.org> writes:

> On Wed, 29 Jan 2020 17:37:20 -0800, Vinicius Costa Gomes wrote:
>> If the driver implementing taprio offloading depends on the value of
>> the network device number of traffic classes (dev->num_tc) for
>> whatever reason, it was going to receive the value zero. The value was
>> only set after the offloading function is called.
>> 
>> So, moving setting the number of traffic classes to before the
>> offloading function is called fixes this issue. This is safe because
>> this only happens when taprio is instantiated (we don't allow this
>> configuration to be changed without first removing taprio).
>> 
>> Fixes: 9c66d1564676 ("taprio: Add support for hardware offloading")
>> Reported-by: Po Liu <po.liu@nxp.com>
>> Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
>> Acked-by: Vladimir Oltean <vladimir.oltean@nxp.com>
>
> What about Dave's question about resetting the tc state with
> netdev_reset_tc()?

I missed that question.