Ok, I have already submitted the patches to remove the sysfs entry, and the
comments about it from the commit in igbvf to Jeff. Hopefully we can get
the pushed through sometime soon.
In the meantime I have been working on the rtnl_link_ops approach and I
think I have a few things going but I wanted to get some input before I go
much further.
First, is it ok for me to call rtnl_unlock prior to doing my settings
changes on the sriov config space, followed by rtnl_lock afterwards in my
newlink and dellink operations? I ask because I had to do this in order to
prevent a deadlock when the pci-hotplug events fired for the vfs and called
unregister/register_netdev.
Second is it acceptable for me to just free the netdev at the end of
newlink and call delete on the PF interface directly? I ask because I
don't have any use for the netdevs that are generated and I cannot call
delete on specific VFs anyway since they are allocated/freed in LIFO order
so I would always have to free the last one I allocated.
I have included a patch for review below that implements these changes
against the current driver. Please feel free to comment.
Thanks,
Alex
--
This patch converts igb from using a sysfs interface to allocate vfs to
using the rtnl_link interface.
Basically the new interface work like this to add a VF:
ip link add eth3 type igbpf
To delete a VF it would be:
ip link del eth3 type igbpf
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
---
drivers/net/igb/igb_main.c | 120 +++++++++++++++++++++++++++++---------------
1 files changed, 80 insertions(+), 40 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 7d3ac76..d1e9425 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -154,11 +154,17 @@ static void igb_netpoll(struct net_device *);
#endif
#ifdef CONFIG_PCI_IOV
-static ssize_t igb_set_num_vfs(struct device *, struct device_attribute *,
- ...
