net: forcedeth: Fix wake-on-lan regression
authorTobias Diedrich <ranma+kernel@tdiedrich.de>
Mon, 16 Feb 2009 08:13:20 +0000 (00:13 -0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 16 Feb 2009 08:13:20 +0000 (00:13 -0800)
Commit f55c21fd9a92a444e55ad1ca4e4732d56661bf2e ("forcedeth: call
restore mac addr in nv_shutdown path"), which was introduced to fix
the regression tracked at
http://bugzilla.kernel.org/show_bug.cgi?id=11358 causes the
wake-on-lan mac to be reversed in the shutdown path.  Apparently the
forcedeth situation is rather messy in that the mac we need to
writeback for a subsequent modprobe to work is exactly the reverse of
what is needed for proper wake-on-lan.

The following patch explains the situation in the comments and
makes the call to nv_restore_mac_addr() conditional (only called if
we are not really going for poweroff).

Tobias Diedrich wrote:
> Hmm, I had not tried WOL for some time.
> With 2.6.29-rc3 is see the following behaviour:
>
> State            WOL Behaviour
> ------------------------------
> shutdown         reversed MAC
> disk/shutdown    reversed MAC
> disk/platform    OK
>
> Apparently nv_restore_mac_addr() restores the MAC in the wrong order
> for WOL (at least for my PCI_DEVICE_ID_NVIDIA_NVENET_15).  platform
> works, because the MAC is not touched in the nv_suspend() path.
>
> A possible fix might be to only call nv_restore_mac_addr() if
> system_state != SYSTEM_POWER_OFF.

With the following patch:
shutdown         OK
disk/shutdown    OK
disk/platform    OK
kexec            OK

Signed-off-by: Tobias Diedrich <ranma+kernel@tdiedrich.de>
Tested-by: Philipp Matthias Hahn <pmhahn@titan.lahn.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/forcedeth.c

index 5b910cf..b8251e8 100644 (file)
@@ -6011,9 +6011,20 @@ static void nv_shutdown(struct pci_dev *pdev)
        if (netif_running(dev))
                nv_close(dev);
 
-       nv_restore_mac_addr(pdev);
+       /*
+        * Restore the MAC so a kernel started by kexec won't get confused.
+        * If we really go for poweroff, we must not restore the MAC,
+        * otherwise the MAC for WOL will be reversed at least on some boards.
+        */
+       if (system_state != SYSTEM_POWER_OFF) {
+               nv_restore_mac_addr(pdev);
+       }
 
        pci_disable_device(pdev);
+       /*
+        * Apparently it is not possible to reinitialise from D3 hot,
+        * only put the device into D3 if we really go for poweroff.
+        */
        if (system_state == SYSTEM_POWER_OFF) {
                if (pci_enable_wake(pdev, PCI_D3cold, np->wolenabled))
                        pci_enable_wake(pdev, PCI_D3hot, np->wolenabled);