Don't leak `work` if `__mlx4_register_vlan` fails in `mlx4_master_immediate_activate_vlan_qos`

[ngie]

Summary

Don't leak work if __mlx4_register_vlan fails in mlx4_master_immediate_activate_vlan_qos

MFC after: 3 days
Submitted by: Miles Olrich <[e-mail]>
Sponsored by: EMC / Isilon Storage Division

[markj]

Looks correct to me, but I'd suggest using an err goto label instead - that's how pretty much all the error-handling is done in the OFED code.

[ngie]

In D4203#88493, @markj wrote:

Looks correct to me, but I'd suggest using an err goto label instead - that's how pretty much all the error-handling is done in the OFED code.

work is usable in the positive case though.

A few other spots in the code do the same thing (just return instead of goto'ing to a label):

 932                                 if (!err) {
 933                                         for (vidx = index * 32; vidx < (index + 1) * 32; ++vidx) {
 934                                                 pidx = priv->virt2phys_pkey[slave][port - 1][vidx];
 935                                                 outtab[vidx % 32] = cpu_to_be16(table[pidx]);
 936                                         }
 937                                 }
 938                                 kfree(table);
 939                                 return err;
 940                         }
 ...
1614                 if (ret) {
1615                         mlx4_err(dev, "%s:Failed reading vhcr"
1616                                  "ret: 0x%x\n", __func__, ret);
1617                         kfree(vhcr);
1618                         return ret;
1619                 }
 ...
2405         if (!mailbox->buf) {
2406                 kfree(mailbox);
2407                 return ERR_PTR(-ENOMEM);
2408         }

[markj]

I'm suggesting having:

return 0;

err:

kfree(work);
return err;

There are counterexamples, but this is the most common and best pattern. But it's not a big deal in this case.