From da54a0ced4502dc2a25df034f218463a2a50488d Mon Sep 17 00:00:00 2001 From: Jiri Kosina Date: Thu, 25 Mar 2010 15:34:34 +0100 Subject: HID: update BKL comment in hiddev Update comment explaining BKL usage in legacy hiddev driver. Signed-off-by: Jiri Kosina --- drivers/hid/usbhid/hiddev.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) (limited to 'drivers/hid/usbhid/hiddev.c') diff --git a/drivers/hid/usbhid/hiddev.c b/drivers/hid/usbhid/hiddev.c index 433602aed468..c24d2fa3e3b6 100644 --- a/drivers/hid/usbhid/hiddev.c +++ b/drivers/hid/usbhid/hiddev.c @@ -267,6 +267,7 @@ static int hiddev_open(struct inode *inode, struct file *file) struct hiddev_list *list; int res, i; + /* See comment in hiddev_connect() for BKL explanation */ lock_kernel(); i = iminor(inode) - HIDDEV_MINOR_BASE; @@ -894,8 +895,22 @@ int hiddev_connect(struct hid_device *hid, unsigned int force) hiddev->hid = hid; hiddev->exist = 1; - /* when lock_kernel() usage is fixed in usb_open(), - * we could also fix it here */ + /* + * BKL here is used to avoid race after usb_register_dev(). + * Once the device node has been created, open() could happen on it. + * The code below will then fail, as hiddev_table hasn't been + * updated. + * + * The obvious fix -- introducing mutex to guard hiddev_table[] + * doesn't work, as usb_open() and usb_register_dev() both take + * minor_rwsem, thus we'll have ABBA deadlock. + * + * Before BKL pushdown, usb_open() had been acquiring it in right + * order, so _open() was safe to use it to protect from this race. + * Now the order is different, but AB-BA deadlock still doesn't occur + * as BKL is dropped on schedule() (i.e. while sleeping on + * minor_rwsem). Fugly. + */ lock_kernel(); retval = usb_register_dev(usbhid->intf, &hiddev_class); if (retval) { -- cgit v1.2.3