Re: [PATCH V3 rdma-next 4/4] RDMA/hns: Implement the disassociate_ucontext API

From: Wei Hu (Xavier)
Date: Tue May 29 2018 - 07:37:26 EST




On 2018/5/29 0:53, Jason Gunthorpe wrote:
> On Sat, May 26, 2018 at 04:41:47PM +0800, Wei Hu (Xavier) wrote:
>> This patch implemented the IB core disassociate_ucontext API.
>>
>> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@xxxxxxxxxx>
>>
>> v2->v3: Addressed the comments from Jason. The related link:
>> https://lkml.org/lkml/2018/5/22/967
>> v1->v2: no change.
>> drivers/infiniband/hw/hns/hns_roce_device.h | 8 ++++
>> drivers/infiniband/hw/hns/hns_roce_main.c | 70 ++++++++++++++++++++++++++++-
>> 2 files changed, 77 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
>> index da8512b..31221d5 100644
>> +++ b/drivers/infiniband/hw/hns/hns_roce_device.h
>> @@ -217,11 +217,19 @@ struct hns_roce_uar {
>> unsigned long logic_idx;
>> };
>>
>> +struct hns_roce_vma_data {
>> + struct list_head list;
>> + struct vm_area_struct *vma;
>> + struct mutex *vma_list_mutex;
>> +};
> This stuff is basically shared in all the drivers too - would it make
> sense to maintain this information and do the zap_pte in the core code
> as well?
>
> There is no way to implement dis-associate without also doing the
> zaps..
>
> Jason
>
> .
Hi, Jason
You may have noticed there are difference about the process of
zap_vmp_ptes
between mlx4_ib_disassociate_ucontext and mlx5_ib_disassociate_ucontext
If I hoist zap_vmp_ptes process into the core code, Maybe need to
modify more mlx4 code.
Maybe I am not the right person to do this thing, Could you give
more suggestion?
Thanks

Regards
Wei Hu
>