netdev - Re: [PATCH net-next 4/5] devlink: embed driver's priv data callback param into devlink

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

Message-ID: <2c38d704-f018-4bc6-b688-a7a7ce4adc0a@intel.com>
Date: Tue, 13 Aug 2024 05:45:47 +0200
From: Przemek Kitszel <przemyslaw.kitszel@...el.com>
To: Jiri Pirko <jiri@...nulli.us>, Jakub Kicinski <kuba@...nel.org>
CC: <netdev@...r.kernel.org>, Ido Schimmel <idosch@...dia.com>, Petr Machata
	<petrm@...dia.com>, Andrew Lunn <andrew@...n.ch>, Florian Fainelli
	<f.fainelli@...il.com>, Vladimir Oltean <olteanv@...il.com>, "David S.
 Miller" <davem@...emloft.net>, Eric Dumazet <edumazet@...gle.com>, "Paolo
 Abeni" <pabeni@...hat.com>, Saeed Mahameed <saeedm@...dia.com>, "Leon
 Romanovsky" <leon@...nel.org>, Tariq Toukan <tariqt@...dia.com>, Tony Nguyen
	<anthony.l.nguyen@...el.com>, <nex.sw.ncis.osdt.itp.upstreaming@...el.com>,
	Wojciech Drewek <wojciech.drewek@...el.com>
Subject: Re: [PATCH net-next 4/5] devlink: embed driver's priv data callback
 param into devlink_resource

On 8/12/24 17:00, Jiri Pirko wrote:
> Mon, Aug 12, 2024 at 01:50:06PM CEST, przemyslaw.kitszel@...el.com wrote:
>> On 8/9/24 13:02, Jiri Pirko wrote:
>>> Fri, Aug 09, 2024 at 04:41:50AM CEST, kuba@...nel.org wrote:
>>>> On Wed, 7 Aug 2024 08:49:57 +0200 Jiri Pirko wrote:
>>>>>> 	lockdep_assert_held(&devlink->lock);
>>>>>>
>>>>>> 	resource = devlink_resource_find(devlink, NULL, resource_id);
>>>>>> -	if (WARN_ON(!resource))
>>>>>> +	if (WARN_ON(!resource || occ_priv_size > resource->priv_size))
>>>>>
>>>>> Very odd. You allocate a mem in devl_resource_register() and here you
>>>>> copy data to it. Why the void pointer is not enough for you? You can
>>>>> easily alloc struct in the driver and pass a pointer to it.
>>>>>
>>>>> This is quite weird. Please don't.
>>>>
>>>> The patch is a bit of a half measure, true.
>>
>> Another option to suit my wants would be to just pass resource_id to the
>> callbacks, would you accept that?
> 
> Why, the callback is registered for particular resource. Passing ID is
> just redundant.

Yet enables one to nicely combine all occ getters/setters for given
resource group. It is also straightforward (compared to this series).
You are right it is not absolutely necessary, but does not hurt and
improves thing (this time I will don't update mlxsw just to have
consumer though, will just post later - as this is not so controversial,
I hope).

> 
> 
>>
>>>>
>>>> Could you shed more light on the design choices for the resource API,
>>>> tho? Why the tying of objects by driver-defined IDs? It looks like
>>>
>>> The ids are exposed all the way down to the user. They are the same
>>> across the reboots and allow user to use the same scripts. Similar to
>>> port index for example.
>>>
>>>
>>>> the callback for getting resources occupancy is "added" later once
>>>> the resource is registered? Is this some legacy of the old locking
>>>> scheme? It's quite unusual.
>>
>> I did such review last month, many decisions really bother me :F, esp:
>> - whole thing is about limiting resources, driver asks HW for occupancy.
> 
> Can you elaborate what's exactly wrong with that?

Typical way to think about resources is "there are X foos" (resource
register time), "give me one foo" (later, on user request). Users could
be heterogeneous, such as VFs and PFs, and resource pool shared over.
This is what I have for (different sizes of) RSS contexes.
(Limit is constant, need to "get*" resources by one at a time, so driver
knows occupancy and arbitrages usage requests).

"get*" == set usage to be increased by one

> 
> 
>>
>> Some minor things:
>> - resizing request validation: parent asks children for permission;
>> - the function to commit the size after the reload is named
>>   devl_resource_size_get().
>>
>>>From the user perspective, I'm going to add a setter, that will be
>> another mode of operation (if compared to the first thing on my complain
>> list):
>> + there is a limit that is constant, and driver/user allocates resource
>>   from such pool.
>>
>>>
>>> It's been some while since I reviewed this, but afaik the reason is that
>>> the occupancy was not possible to obtain during reload, yet the resource
>>> itself stayed during reload. This is now not a problem, since
>>> devlink->lock protects it. I don't see why occupancy getter cannot be
>>> put during resource register, you are correct.
>>>
>> I could add that to my todo list
> 
> Cool.

I guess no one cared about it yet, as resource register and occ getter
register is much separated in code space (to the point of being in
different file).