Re: [RFC net-next 1/2] ethtool: add support for controling the type of adaptive coalescing

[tanhuazhong <tanhuazhong@huawei.com>]

On 2020/11/21 5:22, Michal Kubecek wrote:
> On Fri, Nov 20, 2020 at 02:39:38PM +0100, Andrew Lunn wrote:
>> On Fri, Nov 20, 2020 at 08:23:22AM +0100, Michal Kubecek wrote:
>>> On Fri, Nov 20, 2020 at 10:59:59AM +0800, tanhuazhong wrote:
>>>> On 2020/11/20 6:02, Michal Kubecek wrote:
>>>>> We could use a similar approach as struct ethtool_link_ksettings, e.g.
>>>>>
>>>>> 	struct kernel_ethtool_coalesce {
>>>>> 		struct ethtool_coalesce base;
>>>>> 		/* new members which are not part of UAPI */
>>>>> 	}
>>>>>
>>>>> get_coalesce() and set_coalesce() would get pointer to struct
>>>>> kernel_ethtool_coalesce and ioctl code would be modified to only touch
>>>>> the base (legacy?) part.
>>>>>   > While already changing the ops arguments, we could also add extack
>>>>> pointer, either as a separate argument or as struct member (I slightly
>>>>> prefer the former).
>>>> If changing the ops arguments, each driver who implement
>>>> set_coalesce/get_coalesce of ethtool_ops need to be updated. Is it
>>>> acceptable adding two new ops to get/set ext_coalesce info (like
>>>> ecc31c60240b ("ethtool: Add link extended state") does)? Maybe i can send V2
>>>> in this way, and then could you help to see which one is more suitable?
>>> If it were just this one case, adding an extra op would be perfectly
>>> fine. But from long term point of view, we should expect extending also
>>> other existing ethtool requests and going this way for all of them would
>>> essentially double the number of callbacks in struct ethtool_ops.
>> coccinella might be useful here.
> I played with spatch a bit and it with the spatch and patch below, I got
> only three build failures (with allmodconfig) that would need to be
> addressed manually - these drivers call the set_coalesce() callback on
> device initialization.
> 
> I also tried to make the structure argument const in ->set_coalesce()
> but that was more tricky as adjusting other functions that the structure
> is passed to required either running the spatch three times or repeating
> the same two rules three times in the spatch (or perhaps there is
> a cleaner way but I'm missing relevant knowledge of coccinelle). Then
> there was one more problem in i40e driver which modifies the structure
> before passing it on to its helpers. It could be worked around but I'm
> not sure if constifying the argument is worth these extra complications.
> 
> Michal

will implement it like this in V3.

Regards,
Huazhong.