[dpdk-dev] [PATCH 03/21] vhost: protect virtio_net device struct

[Maxime Coquelin]

Hi Stephen,

On 09/06/2017 04:59 AM, Stephen Hemminger wrote:
>>>> This lock has currently two purposes:
>>>> 1. Prevent referencing freed virtio_dev struct in case of numa_realloc.
>>>> 2. Protect vring pages against invalidation.
>>>>
>>>> For 2., it can be fixed by using the per-vq IOTLB lock (it was not the
>>>> case in my early prototypes that had per device IOTLB cache).
>>>>
>>>> For 1., this is an existing problem, so we might consider it is
>>>> acceptable to keep current state. Maybe it could be improved by only
>>>> reallocating in case VQ0 is not on the right NUMA node, the other VQs
>>>> not being initialized at this point.
> 
> Something like RCU does a better job of protecting against freed virtio_dev.
> But using RCU requires quiescent callback in the main loop.

In my early prototypes, I used the urcu library for the IOTLB cache
implementation.

The problems are that:
1. The lib is LGPL, so only small functions can be inlined into the
code (maybe not a important, if "large functions" are only called in the
slow path).
2. It adds an external dependency, and the lib is not distributed with
every distributions (For RHEL, it is only distributed in EPEL repos).

For the virtio_dev protection, my understanding might be incorrect as
this is the first time I used RCU, but the struct has one field
that the processing threads can write (broadcast_rarp).
But it may not be a problem because at worst, only broadcast_rarp
clearing would be lost, so we would broadcast RARP one more time.

Maxime