Re: [PATCH 0/8 v1] Remove FMR support from RDMA drivers

From: Dennis Dalessandro <dennis.dalessandro@intel.com>
To: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Leon Romanovsky <leon@kernel.org>,
	Max Gurtovoy <maxg@mellanox.com>,
	bvanassche@acm.org, linux-rdma@vger.kernel.org,
	dledford@redhat.com, sagi@grimberg.me, israelr@mellanox.com,
	shlomin@mellanox.com, "Marciniszyn,
	Mike" <mike.marciniszyn@intel.com>
Subject: Re: [PATCH 0/8 v1] Remove FMR support from RDMA drivers
Date: Tue, 26 May 2020 11:38:26 -0400	[thread overview]
Message-ID: <2254dc01-541c-3576-06c3-2a145bc04157@intel.com> (raw)
In-Reply-To: <20200523220821.GB744@ziepe.ca>

On 5/23/2020 6:08 PM, Jason Gunthorpe wrote:
> On Tue, May 19, 2020 at 10:37:07AM -0400, Dennis Dalessandro wrote:
>> On 5/19/2020 10:30 AM, Jason Gunthorpe wrote:
>>> On Tue, May 19, 2020 at 10:26:37AM -0400, Dennis Dalessandro wrote:
>>>> On 5/19/2020 10:19 AM, Leon Romanovsky wrote:
>>>>> On Tue, May 19, 2020 at 10:53:52AM -0300, Jason Gunthorpe wrote:
>>>>>> On Tue, May 19, 2020 at 09:43:14AM -0400, Dennis Dalessandro wrote:
>>>>>>> On 5/18/2020 2:10 PM, Jason Gunthorpe wrote:
>>>>>>>> On Mon, May 18, 2020 at 11:20:04AM -0400, Dennis Dalessandro wrote:
>>>>>>>>> On 5/14/2020 8:02 AM, Max Gurtovoy wrote:
>>>>>>>>>> This series removes the support for FMR mode to register memory. This ancient
>>>>>>>>>> mode is unsafe and not maintained/tested in the last few years. It also doesn't
>>>>>>>>>> have any reasonable advantage over other memory registration methods such as
>>>>>>>>>> FRWR (that is implemented in all the recent RDMA adapters). This series should
>>>>>>>>>> be reviewed and approved by the maintainer of the effected drivers and I
>>>>>>>>>> suggest to test it as well.
>>>>>>>>>>
>>>>>>>>>> The tests that I made for this series (fio benchmarks and fio verify data):
>>>>>>>>>> 1. iSER initiator on ConnectX-4
>>>>>>>>>> 2. iSER initiator on ConnectX-3
>>>>>>>>>> 3. SRP initiator on ConnectX-4 (loopback to SRP target)
>>>>>>>>>> 4. SRP initiator on ConnectX-3
>>>>>>>>>>
>>>>>>>>>> Not tested:
>>>>>>>>>> 1. RDS
>>>>>>>>>> 2. mthca
>>>>>>>>>> 3. rdmavt
>>>>>>>>>
>>>>>>>>> This will effectively kill qib which uses rdmavt. It's gonna have to be a
>>>>>>>>> NAK from me.
>>>>>>>>
>>>>>>>> Are you objecting the SRP and iSER changes too?
>>>>>>>
>>>>>>> No, just want to keep basic verbs support at least. NFS already dropped,
>>>>>>> similarly we are ok with dropping it from SRP/iSER as a next step.
>>>>>>
>>>>>> So you see a major user in RDS for qib?
>>>>>
>>>>> Didn't we agree to drop it from RDS too?
>>>>>
>>>>
>>>> Just basic verbs application support is enough for qib I think. I don't see
>>>> any major use of RDS.
>>>
>>> Well, once the in-kernel users of an API are gone that API will be
>>> purged. This is standard kernel policy.
>>>
>>> So you can't NAK this series on the grounds you want to keep an API
>>> without users, presumably for out of tree modules...
>>>
>>
>> Maybe I need to look at this again. I thought it would kill off user access
>> as well. We don't need any kernel ULPs.
> 
> Did you make a conclusion? Seems like everyone else is in agreement
> here, if Max resends a v2 I'm inclined to take it unless RDS objects.
> 
> I did not think FMR or FRWR were available from userspace at all.

Yeah, looked it over again and agree it's OK. No issues here now. Thanks 
for checking.

-Denny