On 2/7/24 19:05, Ilya Maximets wrote:
On 2/5/24 13:02, Adrian Moreno wrote:
netdev_linux_get_speed needs to lock netdev_linux->mutex, and so do the
internal tc operations. Therefore, the former cannot be called from the
latter.

Create a lock-free version of netdev_linux_get_speed() and call it from
tc operations.

Also expand the unit test to cover queues where ceil is determined by
the maximum link speed.

Fixes: b8f8fad86435 ("netdev-linux: Use speed as max rate in tc classes.")
Reported-by: Daryl Wang <[email protected]>
Suggested-by: Ilya Maximets <[email protected]>
Signed-off-by: Adrian Moreno <[email protected]>
---
  lib/netdev-linux.c      | 32 +++++++++++++++----------
  tests/system-traffic.at | 53 ++++++++++++++++++++++++++++-------------
  2 files changed, 56 insertions(+), 29 deletions(-)

diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
index 1b2e5b6c2..00df7f634 100644
--- a/lib/netdev-linux.c
+++ b/lib/netdev-linux.c
@@ -2721,16 +2721,11 @@ exit:
  }
static int
-netdev_linux_get_speed(const struct netdev *netdev_, uint32_t *current,
-                       uint32_t *max)
+netdev_linux_get_speed_locked(struct netdev_linux *netdev,
+                              uint32_t *current, uint32_t *max)
  {
-    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
-    int error;
-
-    ovs_mutex_lock(&netdev->mutex);
      if (netdev_linux_netnsid_is_remote(netdev)) {
-        error = EOPNOTSUPP;
-        goto exit;
+        return EOPNOTSUPP;
      }
netdev_linux_read_features(netdev);
@@ -2740,9 +2735,18 @@ netdev_linux_get_speed(const struct netdev *netdev_, 
uint32_t *current,
          *max = MIN(UINT32_MAX,
                     netdev_features_to_bps(netdev->supported, 0) / 1000000ULL);
      }
-    error = netdev->get_features_error;
+    return netdev->get_features_error;
+}
-exit:
+static int
+netdev_linux_get_speed(const struct netdev *netdev_, uint32_t *current,
+                       uint32_t *max)
+{
+    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+    int error;
+
+    ovs_mutex_lock(&netdev->mutex);
+    error = netdev_linux_get_speed_locked(netdev, current, max);
      ovs_mutex_unlock(&netdev->mutex);
      return error;
  }
@@ -4954,8 +4958,10 @@ htb_parse_qdisc_details__(struct netdev *netdev, const 
struct smap *details,
      hc->max_rate = smap_get_ullong(details, "max-rate", 0) / 8;
      if (!hc->max_rate) {
          uint32_t current_speed;
+        uint32_t max_speed OVS_UNUSED;
- netdev_get_speed(netdev, &current_speed, NULL);
+        netdev_linux_get_speed_locked(netdev_linux_cast(netdev),
+                                      &current_speed, &max_speed);
          hc->max_rate = current_speed ? current_speed / 8 * 1000000ULL
                                       : NETDEV_DEFAULT_BPS / 8;
      }
@@ -5424,8 +5430,10 @@ hfsc_parse_qdisc_details__(struct netdev *netdev, const 
struct smap *details,
      uint32_t max_rate = smap_get_ullong(details, "max-rate", 0) / 8;
      if (!max_rate) {
          uint32_t current_speed;
+        uint32_t max_speed OVS_UNUSED;
- netdev_get_speed(netdev, &current_speed, NULL);

Oh, wow.  This thing not only deadlocks, it also crashes on the NULL, right?


netdev_get_speed() defined in lib/netdev.c does support the pointer being NULL.
netdev_linux_get_speed{,_locked}() do not. I didn't add such logic now precisely to avoid duplicating logic since it's similar to get_features_*.

I see we call netdev_get_speed(..., NULL) a lot in other palces.  Should
those be fixed as well?  Maybe the function should be fixed to allow NULL
instead?

+        netdev_linux_get_speed_locked(netdev_linux_cast(netdev),
+                                      &current_speed, &max_speed);
          max_rate = current_speed ? current_speed / 8 * 1000000ULL
                                   : NETDEV_DEFAULT_BPS / 8;
      }
diff --git a/tests/system-traffic.at b/tests/system-traffic.at
index f363a778c..d90717d1b 100644
--- a/tests/system-traffic.at
+++ b/tests/system-traffic.at
@@ -2458,34 +2458,53 @@ AT_BANNER([QoS])
AT_SETUP([QoS - basic configuration])
  OVS_CHECK_TC_QDISC()
+AT_SKIP_IF([test $HAVE_ETHTOOL = "no"])
  OVS_TRAFFIC_VSWITCHD_START()
-ADD_NAMESPACES(at_ns0, at_ns1)
+AT_CHECK([ip tuntap add tap0 mode tap])
+on_exit 'ip link del tap0'
+AT_CHECK([ip tuntap add tap1 mode tap])
+on_exit 'ip link del tap1'

tapN is a fairly common name.  Even though we run in a separate
namespace in GHA, it may not be always the case.  Maybe rename
the ports into ovs-tapN ?


Sure. I'll respin the patch with the changed name.


-ADD_VETH(p0, at_ns0, br0, "10.1.1.1/24")
-ADD_VETH(p1, at_ns1, br0, "10.1.1.2/24")
+dnl Set maximum link speed to 5Gb.
+AT_CHECK([ethtool -s tap0 speed 5000 duplex full])
+AT_CHECK([ip link set dev tap0 up])
+AT_CHECK([ethtool -s tap1 speed 5000 duplex full])
+AT_CHECK([ip link set dev tap1 up])
-dnl Adding a custom qdisc to ovs-p1, ovs-p0 will have the default qdisc.
-AT_CHECK([tc qdisc add dev ovs-p1 root noqueue])
-AT_CHECK([tc qdisc show dev ovs-p1 | grep -q noqueue])
+AT_CHECK([ovs-vsctl add-port br0 tap0 -- set int tap0 type=tap])
+AT_CHECK([ovs-vsctl add-port br0 tap1 -- set int tap1 type=tap])
-dnl Configure the same QoS for both ports.
-AT_CHECK([ovs-vsctl set port ovs-p0 qos=@qos -- set port ovs-p1 qos=@qos dnl
-            -- --id=@qos create qos dnl
-               type=linux-htb other-config:max-rate=3000000 queues:0=@queue dnl
-            -- --id=@queue create queue dnl
+dnl Adding a custom qdisc to tap1, tap0 will have the default qdisc.
+AT_CHECK([tc qdisc add dev tap1 root noqueue])
+AT_CHECK([tc qdisc show dev tap1 | grep -q noqueue])
+
+dnl Configure the same QoS for both ports:
+dnl queue0 uses fixed max-rate.
+dnl queue1 relies on underlying link speed.
+AT_CHECK([ovs-vsctl dnl
+            -- --id=@queue0 create queue dnl
                 other_config:min-rate=2000000 other_config:max-rate=3000000 dnl
-               other_config:burst=3000000],
+               other_config:burst=3000000 dnl
+            -- --id=@queue1 create queue dnl
+               other_config:min-rate=4000000 other_config:burst=4000000 dnl
+            -- --id=@qos create qos dnl
+               type=linux-htb queues:0=@queue0 dnl
+               queues:1=@queue1 -- dnl
+            -- set port tap0 qos=@qos -- set port tap1 qos=@qos],
           [ignore], [ignore])
dnl Wait for qdiscs to be applied.
-OVS_WAIT_UNTIL([tc qdisc show dev ovs-p0 | grep -q htb])
-OVS_WAIT_UNTIL([tc qdisc show dev ovs-p1 | grep -q htb])
+OVS_WAIT_UNTIL([tc qdisc show dev tap0 | grep -q htb])
+OVS_WAIT_UNTIL([tc qdisc show dev tap1 | grep -q htb])
dnl Check the configuration.
-m4_define([HTB_CONF], [rate 2Mbit ceil 3Mbit burst 375000b cburst 375000b])
-AT_CHECK([tc class show dev ovs-p0 | grep -q 'class htb .* HTB_CONF'])
-AT_CHECK([tc class show dev ovs-p1 | grep -q 'class htb .* HTB_CONF'])
+m4_define([HTB_CONF0], [rate 2Mbit ceil 3Mbit burst 375000b cburst 375000b])
+m4_define([HTB_CONF1], [rate 4Mbit ceil 5Gbit burst 500000b cburst 500000b])
+AT_CHECK([tc class show dev tap0 | grep -q 'class htb .* HTB_CONF0'])
+AT_CHECK([tc class show dev tap0 | grep -q 'class htb .* HTB_CONF1'])
+AT_CHECK([tc class show dev tap1 | grep -q 'class htb .* HTB_CONF0'])
+AT_CHECK([tc class show dev tap1 | grep -q 'class htb .* HTB_CONF1'])
OVS_TRAFFIC_VSWITCHD_STOP
  AT_CLEANUP


--
Adrián Moreno

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to