]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/commitdiff
net: dsa: Fix off-by-one number of calls to devlink_port_unregister
authorVladimir Oltean <olteanv@gmail.com>
Sat, 31 Aug 2019 12:46:19 +0000 (15:46 +0300)
committerDavid S. Miller <davem@davemloft.net>
Mon, 2 Sep 2019 18:59:29 +0000 (11:59 -0700)
When a function such as dsa_slave_create fails, currently the following
stack trace can be seen:

[    2.038342] sja1105 spi0.1: Probed switch chip: SJA1105T
[    2.054556] sja1105 spi0.1: Reset switch and programmed static config
[    2.063837] sja1105 spi0.1: Enabled switch tagging
[    2.068706] fsl-gianfar soc:ethernet@2d90000 eth2: error -19 setting up slave phy
[    2.076371] ------------[ cut here ]------------
[    2.080973] WARNING: CPU: 1 PID: 21 at net/core/devlink.c:6184 devlink_free+0x1b4/0x1c0
[    2.088954] Modules linked in:
[    2.092005] CPU: 1 PID: 21 Comm: kworker/1:1 Not tainted 5.3.0-rc6-01360-g41b52e38d2b6-dirty #1746
[    2.100912] Hardware name: Freescale LS1021A
[    2.105162] Workqueue: events deferred_probe_work_func
[    2.110287] [<c03133a4>] (unwind_backtrace) from [<c030d8cc>] (show_stack+0x10/0x14)
[    2.117992] [<c030d8cc>] (show_stack) from [<c10b08d8>] (dump_stack+0xb4/0xc8)
[    2.125180] [<c10b08d8>] (dump_stack) from [<c0349d04>] (__warn+0xe0/0xf8)
[    2.132018] [<c0349d04>] (__warn) from [<c0349e34>] (warn_slowpath_null+0x40/0x48)
[    2.139549] [<c0349e34>] (warn_slowpath_null) from [<c0f19d74>] (devlink_free+0x1b4/0x1c0)
[    2.147772] [<c0f19d74>] (devlink_free) from [<c1064fc0>] (dsa_switch_teardown+0x60/0x6c)
[    2.155907] [<c1064fc0>] (dsa_switch_teardown) from [<c1065950>] (dsa_register_switch+0x8e4/0xaa8)
[    2.164821] [<c1065950>] (dsa_register_switch) from [<c0ba7fe4>] (sja1105_probe+0x21c/0x2ec)
[    2.173216] [<c0ba7fe4>] (sja1105_probe) from [<c0b35948>] (spi_drv_probe+0x80/0xa4)
[    2.180920] [<c0b35948>] (spi_drv_probe) from [<c0a4c1cc>] (really_probe+0x108/0x400)
[    2.188711] [<c0a4c1cc>] (really_probe) from [<c0a4c694>] (driver_probe_device+0x78/0x1bc)
[    2.196933] [<c0a4c694>] (driver_probe_device) from [<c0a4a3dc>] (bus_for_each_drv+0x58/0xb8)
[    2.205414] [<c0a4a3dc>] (bus_for_each_drv) from [<c0a4c024>] (__device_attach+0xd0/0x168)
[    2.213637] [<c0a4c024>] (__device_attach) from [<c0a4b1d0>] (bus_probe_device+0x84/0x8c)
[    2.221772] [<c0a4b1d0>] (bus_probe_device) from [<c0a4b72c>] (deferred_probe_work_func+0x84/0xc4)
[    2.230686] [<c0a4b72c>] (deferred_probe_work_func) from [<c03650a4>] (process_one_work+0x218/0x510)
[    2.239772] [<c03650a4>] (process_one_work) from [<c03660d8>] (worker_thread+0x2a8/0x5c0)
[    2.247908] [<c03660d8>] (worker_thread) from [<c036b348>] (kthread+0x148/0x150)
[    2.255265] [<c036b348>] (kthread) from [<c03010e8>] (ret_from_fork+0x14/0x2c)
[    2.262444] Exception stack(0xea965fb0 to 0xea965ff8)
[    2.267466] 5fa0:                                     00000000 00000000 00000000 00000000
[    2.275598] 5fc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[    2.283729] 5fe0: 00000000 00000000 00000000 00000000 00000013 00000000
[    2.290333] ---[ end trace ca5d506728a0581a ]---

devlink_free is complaining right here:

WARN_ON(!list_empty(&devlink->port_list));

This happens because devlink_port_unregister is no longer done right
away in dsa_port_setup when a DSA_PORT_TYPE_USER has failed.
Vivien said about this change that:

    Also no need to call devlink_port_unregister from within dsa_port_setup
    as this step is inconditionally handled by dsa_port_teardown on error.

which is not really true. The devlink_port_unregister function _is_
being called unconditionally from within dsa_port_setup, but not for
this port that just failed, just for the previous ones which were set
up.

ports_teardown:
for (i = 0; i < port; i++)
dsa_port_teardown(&ds->ports[i]);

Initially I was tempted to fix this by extending the "for" loop to also
cover the port that failed during setup. But this could have potentially
unforeseen consequences unrelated to devlink_port or even other types of
ports than user ports, which I can't really test for. For example, if
for some reason devlink_port_register itself would fail, then
unconditionally unregistering it in dsa_port_teardown would not be a
smart idea. The list might go on.

So just make dsa_port_setup undo the setup it had done upon failure, and
let the for loop undo the work of setting up the previous ports, which
are guaranteed to be brought up to a consistent state.

Fixes: 955222ca5281 ("net: dsa: use a single switch statement for port setup")
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/dsa/dsa2.c

index f8445fa7344890255e8828bc1b5c86b793ca4d24..b501c90aabe45e16741f17ff663917bc82464a2e 100644 (file)
@@ -259,8 +259,11 @@ static int dsa_port_setup(struct dsa_port *dp)
        const unsigned char *id = (const unsigned char *)&dst->index;
        const unsigned char len = sizeof(dst->index);
        struct devlink_port *dlp = &dp->devlink_port;
+       bool dsa_port_link_registered = false;
+       bool devlink_port_registered = false;
        struct devlink *dl = ds->devlink;
-       int err;
+       bool dsa_port_enabled = false;
+       int err = 0;
 
        switch (dp->type) {
        case DSA_PORT_TYPE_UNUSED:
@@ -272,15 +275,19 @@ static int dsa_port_setup(struct dsa_port *dp)
                                       dp->index, false, 0, id, len);
                err = devlink_port_register(dl, dlp, dp->index);
                if (err)
-                       return err;
+                       break;
+               devlink_port_registered = true;
 
                err = dsa_port_link_register_of(dp);
                if (err)
-                       return err;
+                       break;
+               dsa_port_link_registered = true;
 
                err = dsa_port_enable(dp, NULL);
                if (err)
-                       return err;
+                       break;
+               dsa_port_enabled = true;
+
                break;
        case DSA_PORT_TYPE_DSA:
                memset(dlp, 0, sizeof(*dlp));
@@ -288,15 +295,19 @@ static int dsa_port_setup(struct dsa_port *dp)
                                       dp->index, false, 0, id, len);
                err = devlink_port_register(dl, dlp, dp->index);
                if (err)
-                       return err;
+                       break;
+               devlink_port_registered = true;
 
                err = dsa_port_link_register_of(dp);
                if (err)
-                       return err;
+                       break;
+               dsa_port_link_registered = true;
 
                err = dsa_port_enable(dp, NULL);
                if (err)
-                       return err;
+                       break;
+               dsa_port_enabled = true;
+
                break;
        case DSA_PORT_TYPE_USER:
                memset(dlp, 0, sizeof(*dlp));
@@ -304,18 +315,26 @@ static int dsa_port_setup(struct dsa_port *dp)
                                       dp->index, false, 0, id, len);
                err = devlink_port_register(dl, dlp, dp->index);
                if (err)
-                       return err;
+                       break;
+               devlink_port_registered = true;
 
                dp->mac = of_get_mac_address(dp->dn);
                err = dsa_slave_create(dp);
                if (err)
-                       return err;
+                       break;
 
                devlink_port_type_eth_set(dlp, dp->slave);
                break;
        }
 
-       return 0;
+       if (err && dsa_port_enabled)
+               dsa_port_disable(dp);
+       if (err && dsa_port_link_registered)
+               dsa_port_link_unregister_of(dp);
+       if (err && devlink_port_registered)
+               devlink_port_unregister(dlp);
+
+       return err;
 }
 
 static void dsa_port_teardown(struct dsa_port *dp)