Re: [PATCHv1 0/6] rdma controller support

[Parav Pandit]

On Fri, Jan 8, 2016 at 2:19 AM, Tejun Heo <tj@xxxxxxxxxx> wrote:
> Hello, Parav.
>
> On Fri, Jan 08, 2016 at 02:16:59AM +0530, Parav Pandit wrote:
>> Let me think through it. Its been late night for me currently. So dont
>> want to conclude in hurry.
>
> Sure thing.
>
>> At high level it looks doable by maintaining hash table head on per
>> device basis, that further reduces hash contention by one level.
>> I will get back on this tomorrow.
>
> Hmmm... why would it need a hash table?  Let's say there's a struct
> rdma_device for each rdma_device and then that stuct can simply have
> rdma_device->res_table[] or whatever to track limits and consumptions
> and rdma_device->res_enabled mask to tell which resources are enabled
> on the device.
>

That table won't be sufficient, because rdma_device is shared among
multiple rdma_cgroups each such cgroup has different individual
resource limit and usage count. This is currently rpool structure.
For res_table[] needs to be per cgroup basis.


> Thanks.
>
> --
> tejun