Thread (116 messages) 116 messages, 7 authors, 2012-06-05

Re: [ 07/91] net/wireless: ipw2200: Fix WARN_ON occurring in wiphy_register called by ipw_pci_probe

From: Ben Hutchings <hidden>
Date: 2012-05-27 19:50:19
Also in: lkml

On Sun, 2012-05-27 at 16:24 -0300, Herton Ronaldo Krzesinski wrote:
On Sun, May 27, 2012 at 09:25:13AM +0900, Greg KH wrote:
quoted
3.4-stable review patch.  If anyone has any objections, please let me know.

------------------
As I see, this is not needed on 3.4.x or 3.3.x, since the commit
"ipw2200: Fix order of device registration", wasn't included in 3.4 or
any earlier kernels.

This is fixing an issue introduced on that commit ("ipw2200: Fix order of
device registration"), that moved register_netdev after ipw_wdev_init,
but ipw_wdev_init needs initialization made on ndo_init callback.

Unless there is any plan to include any of the "... Fix order of device
registration" commits on 3.4 stable, I don't see any reason to include
the fix for them.

Besides that, I noted that ipw2100 seems to need a similar fix as below
after the change "ipw2100: Fix order of device registration". ipw2100
will have the same issue I expect: because ipw2100_wdev_init was moved
before register_netdev, the ndo_init callback that runs ipw2100_up will
only execute later with required initialization, after wiphy_register
inside ipw2100_wdev_init runs.
[...]

Right.

As it happens, ipw2100 also has a comment explaining why .ndo_init is
used now:

	 * If we called ipw2100_up before we registered the device, then the
	 * device name wasn't registered.  So, we instead use the net_dev->init
	 * member to call a function that then just turns and calls ipw2100_up.
	 * net_dev->init is called after name allocation but before the
	 * notifier chain is called */

So this can result in log messages referring to "wlan%d" rather than the
yet-to-be-determined device name.  (That can be avoided by using
netdev_printk, which will log the PCI address.)

If userland depends on the "phy80211" symlinks for wireless net devices
in sysfs then all 4 fixes would be worth applying to stable.  Otherwise,
none of them should be applied.

Ben.

-- 
Ben Hutchings
The obvious mathematical breakthrough [to break modern encryption] would be
development of an easy way to factor large prime numbers. - Bill Gates

Attachments

Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help