linux-kernel - Re: [PATCH for-next v3 2/2] RDMA/rxe: Enable asynchronous prefetch for ODP MRs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <1720c440-719b-4bc9-9685-148c20d4cdfe@gmail.com>
Date: Mon, 19 May 2025 20:58:14 +0900
From: Daisuke Matsuda <dskmtsd@...il.com>
To: Leon Romanovsky <leon@...nel.org>
Cc: linux-kernel@...r.kernel.org, linux-rdma@...r.kernel.org, jgg@...pe.ca,
 zyjzyj2000@...il.com, Zhu Yanjun <yanjun.zhu@...ux.dev>
Subject: Re: [PATCH for-next v3 2/2] RDMA/rxe: Enable asynchronous prefetch
 for ODP MRs


On 2025/05/18 14:54, Leon Romanovsky wrote:
> On Tue, May 13, 2025 at 05:04:05AM +0000, Daisuke Matsuda wrote:
>> Calling ibv_advise_mr(3) with flags other than IBV_ADVISE_MR_FLAG_FLUSH
>> invokes asynchronous request. It is best-effort, and thus can safely be
>> deferred to the system-wide workqueue.
>>
>> Signed-off-by: Daisuke Matsuda <dskmtsd@...il.com>
>> Reviewed-by: Zhu Yanjun <yanjun.zhu@...ux.dev>
>> ---
>>   drivers/infiniband/sw/rxe/rxe_odp.c | 84 ++++++++++++++++++++++++++++-
>>   1 file changed, 82 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/infiniband/sw/rxe/rxe_odp.c b/drivers/infiniband/sw/rxe/rxe_odp.c
>> index 4c98a02d572c..0f3b281a265f 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_odp.c
>> +++ b/drivers/infiniband/sw/rxe/rxe_odp.c
>> @@ -425,6 +425,73 @@ enum resp_states rxe_odp_do_atomic_write(struct rxe_mr *mr, u64 iova, u64 value)
>>   	return RESPST_NONE;
>>   }
> 
> <...>
> 
>> +static int rxe_init_prefetch_work(struct ib_pd *ibpd,
>> +				  enum ib_uverbs_advise_mr_advice advice,
>> +				  u32 pf_flags, struct prefetch_mr_work *work,
>> +				  struct ib_sge *sg_list, u32 num_sge)
> 
> There is no need one-time called function. It can be embedded into rxe_ib_advise_mr_prefetch().

Certainly.

> 
>> +{
> 
> <...>
> 
>> @@ -475,6 +542,8 @@ static int rxe_ib_advise_mr_prefetch(struct ib_pd *ibpd,
>>   				     u32 flags, struct ib_sge *sg_list, u32 num_sge)
>>   {
> 
> <...>
> 
>> +	queue_work(system_unbound_wq, &work->work);
> 
> How do you ensure that this work isn't running after RXE is destroyed?

I think we can use per-pd reference counter in struct rxe_pd.
I will fix it in v4.

Thanks,
Daisuke

> 
> Thanks
> 
>> +
>> +	return 0;
>>   }
>>   
>>   int rxe_ib_advise_mr(struct ib_pd *ibpd,
>> -- 
>> 2.43.0
>>