On 4/10/2018 2:26 PM, Stephen Hemminger wrote:
On Tue, 10 Apr 2018 11:59:50 -0700
Sridhar Samudrala <sridhar.samudrala@xxxxxxxxx> wrote:
Use the registration/notification framework supported by the generic
bypass infrastructure.
Signed-off-by: Sridhar Samudrala <sridhar.samudrala@xxxxxxxxx>
---
Thanks for doing this. Your current version has couple show stopper
issues.
First, the slave device is instantly taking over the slave.
This doesn't allow udev/systemd to do its device rename of the slave
device. Netvsc uses a delayed work to workaround this.
OK. I guess you are referring to the dev_set_mtu() and dev_open() calls that are
made in bypass_slave_register() and you want to defer them to be done after
a delay. I could avoid these calls in case of netvsc based on bypass_ops.
Secondly, the select queue needs to call queue selection in VF.
The bonding/teaming logic doesn't work well for UDP flows.
Commit b3bf5666a510 ("hv_netvsc: defer queue selection to VF")
fixed this performance problem.
netvsc should not be using bypass_select_queue() as that ndo op gets used
only with 3-netdev model.
Anyway, will look into updating bypass_select_queue() based on your fix.
Lastly, more indirection is bad in current climate.
I am not completely adverse to this but it needs to be fast, simple
and completely transparent.
Not sure we can avoid this indirection if we want to commonize the code, but use
different models for virtio-net and netvsc.
On the other hand, these patches avoid calls to get_netvsc_bymac() and
get_netvsc_by_ref() that go through all the devices for all the netdev events.
netvsc lookups should be much faster.
Thanks
Sridhar
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
