driver core: Fix lockdep warning on wfs_lock
authorSaravana Kannan <saravanak@google.com>
Wed, 4 Nov 2020 20:54:30 +0000 (12:54 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 9 Nov 2020 17:17:52 +0000 (18:17 +0100)
There's a potential deadlock with the following cycle:
wfs_lock --> device_links_lock --> kn->count

Fix this by simply dropping the lock around a list_empty() check that's
just exported to a sysfs file. The sysfs file output is an instantaneous
check anyway and the lock doesn't really add any protection.

Lockdep log:

[   48.808132]
[   48.808132] the existing dependency chain (in reverse order) is:
[   48.809069]
[   48.809069] -> #2 (kn->count){++++}:
[   48.809707]        __kernfs_remove.llvm.7860393000964815146+0x2d4/0x460
[   48.810537]        kernfs_remove_by_name_ns+0x54/0x9c
[   48.811171]        sysfs_remove_file_ns+0x18/0x24
[   48.811762]        device_del+0x2b8/0x5a8
[   48.812269]        __device_link_del+0x98/0xb8
[   48.812829]        device_links_driver_bound+0x210/0x2d8
[   48.813496]        driver_bound+0x44/0xf8
[   48.814000]        really_probe+0x340/0x6e0
[   48.814526]        driver_probe_device+0xb8/0x100
[   48.815117]        device_driver_attach+0x78/0xb8
[   48.815708]        __driver_attach+0xe0/0x194
[   48.816255]        bus_for_each_dev+0xa8/0x11c
[   48.816816]        driver_attach+0x24/0x30
[   48.817331]        bus_add_driver+0x100/0x1e0
[   48.817880]        driver_register+0x78/0x114
[   48.818427]        __platform_driver_register+0x44/0x50
[   48.819089]        0xffffffdbb3227038
[   48.819551]        do_one_initcall+0xd8/0x1e0
[   48.820099]        do_init_module+0xd8/0x298
[   48.820636]        load_module+0x3afc/0x44c8
[   48.821173]        __arm64_sys_finit_module+0xbc/0xf0
[   48.821807]        el0_svc_common+0xbc/0x1d0
[   48.822344]        el0_svc_handler+0x74/0x98
[   48.822882]        el0_svc+0x8/0xc
[   48.823310]
[   48.823310] -> #1 (device_links_lock){+.+.}:
[   48.824036]        __mutex_lock_common+0xe0/0xe44
[   48.824626]        mutex_lock_nested+0x28/0x34
[   48.825185]        device_link_add+0xd4/0x4ec
[   48.825734]        of_link_to_suppliers+0x158/0x204
[   48.826347]        of_fwnode_add_links+0x50/0x64
[   48.826928]        device_link_add_missing_supplier_links+0x90/0x11c
[   48.827725]        fw_devlink_resume+0x58/0x130
[   48.828296]        of_platform_default_populate_init+0xb4/0xd0
[   48.829030]        do_one_initcall+0xd8/0x1e0
[   48.829578]        do_initcall_level+0xb8/0xcc
[   48.830137]        do_basic_setup+0x60/0x7c
[   48.830662]        kernel_init_freeable+0x128/0x1ac
[   48.831275]        kernel_init+0x18/0x29c
[   48.831781]        ret_from_fork+0x10/0x18
[   48.832297]
[   48.832297] -> #0 (wfs_lock){+.+.}:
[   48.832922]        __lock_acquire+0xe04/0x2e20
[   48.833480]        lock_acquire+0xbc/0xec
[   48.833984]        __mutex_lock_common+0xe0/0xe44
[   48.834577]        mutex_lock_nested+0x28/0x34
[   48.835136]        waiting_for_supplier_show+0x3c/0x98
[   48.835781]        dev_attr_show+0x48/0xb4
[   48.836295]        sysfs_kf_seq_show+0xe8/0x184
[   48.836864]        kernfs_seq_show+0x48/0x8c
[   48.837401]        seq_read+0x1c8/0x600
[   48.837884]        kernfs_fop_read+0x68/0x204
[   48.838431]        __vfs_read+0x60/0x214
[   48.838925]        vfs_read+0xbc/0x15c
[   48.839397]        ksys_read+0x78/0xe4
[   48.839869]        __arm64_sys_read+0x1c/0x28
[   48.840416]        el0_svc_common+0xbc/0x1d0
[   48.840953]        el0_svc_handler+0x74/0x98
[   48.841490]        el0_svc+0x8/0xc
[   48.841917]
[   48.841917] other info that might help us debug this:
[   48.841917]
[   48.842920] Chain exists of:
[   48.842920]   wfs_lock --> device_links_lock --> kn->count
[   48.842920]
[   48.844152]  Possible unsafe locking scenario:
[   48.844152]
[   48.844895]        CPU0                    CPU1
[   48.845463]        ----                    ----
[   48.846032]   lock(kn->count);
[   48.846417]                                lock(device_links_lock);
[   48.847203]                                lock(kn->count);
[   48.847902]   lock(wfs_lock);
[   48.848276]
[   48.848276]  *** DEADLOCK ***

Reported-by: Cheng-Jui.Wang@mediatek.com
Signed-off-by: Saravana Kannan <saravanak@google.com>
Link: https://lore.kernel.org/r/20201104205431.3795207-1-saravanak@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/core.c

index d661ada1518fb49ceab77789043946abfa32d3d8..1165a80f8010ec174c4d9c62f1706245292319b0 100644 (file)
@@ -1062,10 +1062,8 @@ static ssize_t waiting_for_supplier_show(struct device *dev,
        bool val;
 
        device_lock(dev);
-       mutex_lock(&wfs_lock);
        val = !list_empty(&dev->links.needs_suppliers)
              && dev->links.need_for_probe;
-       mutex_unlock(&wfs_lock);
        device_unlock(dev);
        return sysfs_emit(buf, "%u\n", val);
 }