net: dsa: ocelot: call dsa_tag_8021q_unregister() under rtnl_lock() on driver remove
authorVladimir Oltean <vladimir.oltean@nxp.com>
Thu, 3 Aug 2023 13:42:53 +0000 (16:42 +0300)
committerJakub Kicinski <kuba@kernel.org>
Fri, 4 Aug 2023 22:34:16 +0000 (15:34 -0700)
When the tagging protocol in current use is "ocelot-8021q" and we unbind
the driver, we see this splat:

$ echo '0000:00:00.2' > /sys/bus/pci/drivers/fsl_enetc/unbind
mscc_felix 0000:00:00.5 swp0: left promiscuous mode
sja1105 spi2.0: Link is Down
DSA: tree 1 torn down
mscc_felix 0000:00:00.5 swp2: left promiscuous mode
sja1105 spi2.2: Link is Down
DSA: tree 3 torn down
fsl_enetc 0000:00:00.2 eno2: left promiscuous mode
mscc_felix 0000:00:00.5: Link is Down
------------[ cut here ]------------
RTNL: assertion failed at net/dsa/tag_8021q.c (409)
WARNING: CPU: 1 PID: 329 at net/dsa/tag_8021q.c:409 dsa_tag_8021q_unregister+0x12c/0x1a0
Modules linked in:
CPU: 1 PID: 329 Comm: bash Not tainted 6.5.0-rc3+ #771
pc : dsa_tag_8021q_unregister+0x12c/0x1a0
lr : dsa_tag_8021q_unregister+0x12c/0x1a0
Call trace:
 dsa_tag_8021q_unregister+0x12c/0x1a0
 felix_tag_8021q_teardown+0x130/0x150
 felix_teardown+0x3c/0xd8
 dsa_tree_teardown_switches+0xbc/0xe0
 dsa_unregister_switch+0x168/0x260
 felix_pci_remove+0x30/0x60
 pci_device_remove+0x4c/0x100
 device_release_driver_internal+0x188/0x288
 device_links_unbind_consumers+0xfc/0x138
 device_release_driver_internal+0xe0/0x288
 device_driver_detach+0x24/0x38
 unbind_store+0xd8/0x108
 drv_attr_store+0x30/0x50
---[ end trace 0000000000000000 ]---
------------[ cut here ]------------
RTNL: assertion failed at net/8021q/vlan_core.c (376)
WARNING: CPU: 1 PID: 329 at net/8021q/vlan_core.c:376 vlan_vid_del+0x1b8/0x1f0
CPU: 1 PID: 329 Comm: bash Tainted: G        W          6.5.0-rc3+ #771
pc : vlan_vid_del+0x1b8/0x1f0
lr : vlan_vid_del+0x1b8/0x1f0
 dsa_tag_8021q_unregister+0x8c/0x1a0
 felix_tag_8021q_teardown+0x130/0x150
 felix_teardown+0x3c/0xd8
 dsa_tree_teardown_switches+0xbc/0xe0
 dsa_unregister_switch+0x168/0x260
 felix_pci_remove+0x30/0x60
 pci_device_remove+0x4c/0x100
 device_release_driver_internal+0x188/0x288
 device_links_unbind_consumers+0xfc/0x138
 device_release_driver_internal+0xe0/0x288
 device_driver_detach+0x24/0x38
 unbind_store+0xd8/0x108
 drv_attr_store+0x30/0x50
DSA: tree 0 torn down

This was somewhat not so easy to spot, because "ocelot-8021q" is not the
default tagging protocol, and thus, not everyone who tests the unbinding
path may have switched to it beforehand. The default
felix_tag_npi_teardown() does not require rtnl_lock() to be held.

Fixes: 7c83a7c539ab ("net: dsa: add a second tagger for Ocelot switches based on tag_8021q")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20230803134253.2711124-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/dsa/ocelot/felix.c

index 8da46d2..bef879c 100644 (file)
@@ -1625,8 +1625,10 @@ static void felix_teardown(struct dsa_switch *ds)
        struct felix *felix = ocelot_to_felix(ocelot);
        struct dsa_port *dp;
 
+       rtnl_lock();
        if (felix->tag_proto_ops)
                felix->tag_proto_ops->teardown(ds);
+       rtnl_unlock();
 
        dsa_switch_for_each_available_port(dp, ds)
                ocelot_deinit_port(ocelot, dp->index);