CONFIG_NVME_MULTIPATH - Do we need this?

CONFIG_NVME_MULTIPATH - Do we need this?

[John Meneghini]

Now that RHEL-10 beta has shipped we no longer support anything but native nvme multi-pathing and I'd like to do something about 
the multipath module parameter:

bool multipath = true;
module_param(multipath, bool, 0444);
MODULE_PARM_DESC(multipath,
         "turn on native support for multiple controllers per subsystem");

I'd like to remove this parameter from nvme_core/multipath.c. The fact is, setting nvme_core.multipath=N never worked in 
anything but RHEL and keeping it around only causes confusion.

For that matter, I'd like to remove the CONFIG_NVME_MULTIPATH kconfig option altogether.

Does anybody actually use this?

# CONFIG_NVME_MULTIPATH is not set

Do we actually support running nvme without multipath.c ?

-- 
John A. Meneghini
Senior Principal Platform Storage Engineer
RHEL SST - Platform Storage Group
jmeneghi@redhat.com

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[Christoph Hellwig]

On Tue, Nov 12, 2024 at 07:23:06PM -0500, John Meneghini wrote:
> I'd like to remove this parameter from nvme_core/multipath.c. The fact is,
> setting nvme_core.multipath=N never worked in anything but RHEL and keeping
> it around only causes confusion.

I'm all in for that.

> For that matter, I'd like to remove the CONFIG_NVME_MULTIPATH kconfig option altogether.
> 
> Does anybody actually use this?

It might still make sense to keep the code footprint minimal for
embedded systems that just have an SD/CF Express card or a sodered on
BGA device that most certainly aren't CMIC/NMIC capable.  But I have
no really strong opinion.

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[Guixin Liu]

在 2024/11/13 08:23, John Meneghini 写道:
> Now that RHEL-10 beta has shipped we no longer support anything but 
> native nvme multi-pathing and I'd like to do something about the 
> multipath module parameter:
>
> bool multipath = true;
> module_param(multipath, bool, 0444);
> MODULE_PARM_DESC(multipath,
>         "turn on native support for multiple controllers per subsystem");
>
> I'd like to remove this parameter from nvme_core/multipath.c. The fact 
> is, setting nvme_core.multipath=N never worked in anything but RHEL 
> and keeping it around only causes confusion.
>
> For that matter, I'd like to remove the CONFIG_NVME_MULTIPATH kconfig 
> option altogether.
>
> Does anybody actually use this?
>
> # CONFIG_NVME_MULTIPATH is not set
>
> Do we actually support running nvme without multipath.c ?
>
If the user want use dm-multipath instead of nvme-multipath, but the 
kernel's CONFIG_NVME_MULTIPATH is y, they can set this parameter to 
disable nvme-multipath.

Best Regards,

Guixin Liu

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[Christoph Hellwig]

On Fri, Nov 15, 2024 at 09:40:47AM +0800, Guixin Liu wrote:
> If the user want use dm-multipath instead of nvme-multipath, but the
> kernel's CONFIG_NVME_MULTIPATH is y, they can set this parameter to disable
> nvme-multipath.

.. which has never been supported to start with.

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[Keith Busch]

On Tue, Nov 12, 2024 at 07:23:06PM -0500, John Meneghini wrote:
> Now that RHEL-10 beta has shipped we no longer support anything but native
> nvme multi-pathing and I'd like to do something about the multipath module
> parameter:
> 
> bool multipath = true;
> module_param(multipath, bool, 0444);
> MODULE_PARM_DESC(multipath,
>         "turn on native support for multiple controllers per subsystem");
> 
> I'd like to remove this parameter from nvme_core/multipath.c. The fact is,
> setting nvme_core.multipath=N never worked in anything but RHEL and keeping
> it around only causes confusion.

Not sure the about this being a RHEL specific parameter. I use this for
testing purposes, but I don't use RHEL. Admittedly I haven't needed it
recently, but I have found it useful in the past.

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[Nilay Shroff]

On 11/13/24 05:53, John Meneghini wrote:
> Now that RHEL-10 beta has shipped we no longer support anything but native nvme multi-pathing and I'd like to do something about the multipath module parameter:
> 
> bool multipath = true;
> module_param(multipath, bool, 0444);
> MODULE_PARM_DESC(multipath,
>         "turn on native support for multiple controllers per subsystem");
> 
> I'd like to remove this parameter from nvme_core/multipath.c. The fact is, setting nvme_core.multipath=N never worked in anything but RHEL and keeping it around only causes confusion.
> 
> For that matter, I'd like to remove the CONFIG_NVME_MULTIPATH kconfig option altogether.
> 
> Does anybody actually use this?
> 
> # CONFIG_NVME_MULTIPATH is not set
> 
> Do we actually support running nvme without multipath.c ?
> 
Yes I think it's supported but I don't foresee any side effect of removing 
CONFIG_NVME_MULTIPATH except that the following blktests would require some updates:

nvme/033
nvme/034
nvme/035
nvme/036
nvme/037
nvme/039

All above tests require CONFIG_NVME_MULTIPATH is unset. So if you're planning 
to get rid off this config option then we need to update above tests.

Thanks,
--Nilay

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[John Meneghini]

At this point I think we are going to remove the module_param(multipath) and leave the KCONFIG option alone.

That should be a little less disruptive and we can always submit a separate patch to remove the KCONFIG option later.

Red Hat is currently testing this change internally and will submit a patch for consideration sometime next week.

Thanks,

/John

On 11/19/24 05:18, Nilay Shroff wrote:
> 
> 
> On 11/13/24 05:53, John Meneghini wrote:
>> Now that RHEL-10 beta has shipped we no longer support anything but native nvme multi-pathing and I'd like to do something about the multipath module parameter:
>>
>> bool multipath = true;
>> module_param(multipath, bool, 0444);
>> MODULE_PARM_DESC(multipath,
>>          "turn on native support for multiple controllers per subsystem");
>>
>> I'd like to remove this parameter from nvme_core/multipath.c. The fact is, setting nvme_core.multipath=N never worked in anything but RHEL and keeping it around only causes confusion.
>>
>> For that matter, I'd like to remove the CONFIG_NVME_MULTIPATH kconfig option altogether.
>>
>> Does anybody actually use this?
>>
>> # CONFIG_NVME_MULTIPATH is not set
>>
>> Do we actually support running nvme without multipath.c ?
>>
> Yes I think it's supported but I don't foresee any side effect of removing
> CONFIG_NVME_MULTIPATH except that the following blktests would require some updates:
> 
> nvme/033
> nvme/034
> nvme/035
> nvme/036
> nvme/037
> nvme/039
> 
> All above tests require CONFIG_NVME_MULTIPATH is unset. So if you're planning
> to get rid off this config option then we need to update above tests.
> 
> Thanks,
> --Nilay
> 
> 
> 

Re: CONFIG_NVME_MULTIPATH - Do we need this?

[Nilay Shroff]

On 11/19/24 20:25, John Meneghini wrote:
> At this point I think we are going to remove the module_param(multipath) and leave the KCONFIG option alone.
> 
> That should be a little less disruptive and we can always submit a separate patch to remove the KCONFIG option later.
> 
Agreed!
> Red Hat is currently testing this change internally and will submit a patch for consideration sometime next week.
> 
OK makes sense. I would help review and test changes on PPC with multi controller nvme disk.
> Thanks,
> 
> /John
Thanks,
--Nilay