Re: [PATCH rdma-next v6 03/15] RDMA/core: Introduce generic buffer descriptor infrastructure for umem

[Jiri Pirko <jiri@resnulli.us>]

Wed, May 20, 2026 at 06:08:44PM +0200, jgg@ziepe.ca wrote:
>On Wed, May 20, 2026 at 12:11:17PM +0200, Jiri Pirko wrote:
>> +int uverbs_attr_get_buffer_desc(const struct uverbs_attr_bundle *attrs,
>> +				u16 attr_id,
>> +				struct ib_uverbs_buffer_desc *desc)
>> +{
>> +	int ret;
>> +
>> +	ret = uverbs_copy_from(desc, attrs, attr_id);
>> +	if (ret)
>> +		return ret;
>> +	if (desc->reserved[0] || desc->reserved[1])
>> +		return -EINVAL;
>
>I'd like to carve out some space for optional kernel ignored flags
>here.
>
>Basically a flags field that kernel just ignores. If userspace
>requests a flag through this field that the kernel does not support it
>ignores the request and does nothing. 
>
>Then a seperate flags that the kernel does enforce every set bit is
>understood.
>
>We've not included this a few times in the past and were sad about
>it..

No problem. Will replace reverved[2] by "flags" and "optional_flag".


>
>I also want to think about how we can discover if the kernel supports
>this new path from userspace..

This is problematic. This is up to the driver if it implements
specific ATTR (may be generic or driver specific) or not. Not sure how
to do this. Per driver list of supported ATTRs? Idk. Ideas? One way or
another, this could be easily a follow-up.