summaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorLukas Wunner <lukas@wunner.de>2018-02-10 19:27:12 +0100
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>2018-02-27 18:10:42 +0100
commitead18c23c263374ed098a7d955b29b4a466d4573 (patch)
tree17e0951e486bd5ba4b757028c0ffef88c4d34715 /kernel
parentda997b22c40473b7db60bde6ea188d35565d10c8 (diff)
downloadlinux-ead18c23c263374ed098a7d955b29b4a466d4573.tar.bz2
driver core: Introduce device links reference counting
If device_link_add() is invoked multiple times with the same supplier and consumer combo, it will create the link on first addition and return a pointer to the already existing link on all subsequent additions. The semantics for device_link_del() are quite different, it deletes the link unconditionally, so multiple invocations are not allowed. In other words, this snippet ... struct device *dev1, *dev2; struct device_link *link1, *link2; link1 = device_link_add(dev1, dev2, 0); link2 = device_link_add(dev1, dev2, 0); device_link_del(link1); device_link_del(link2); ... causes the following crash: WARNING: CPU: 4 PID: 2686 at drivers/base/power/runtime.c:1611 pm_runtime_drop_link+0x40/0x50 [...] list_del corruption, 0000000039b800a4->prev is LIST_POISON2 (00000000ecf79852) kernel BUG at lib/list_debug.c:50! The issue isn't as arbitrary as it may seem: Imagine a device link which is added in both the supplier's and the consumer's ->probe hook. The two drivers can't just call device_link_del() in their ->remove hook without coordination. Fix by counting multiple additions and dropping the device link only when the last addition is unwound. Signed-off-by: Lukas Wunner <lukas@wunner.de> [ rjw: Subject ] Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Diffstat (limited to 'kernel')
0 files changed, 0 insertions, 0 deletions