[ofa-general] [PATCH RFC] RDMA: New Memory Extensions.

[Steve Wise]

Ralph Campbell wrote:
> Do we have any expected consumers for this interface?
> I would guess ib_srp, ib_iser as likely candidates.
> 

NFSRDMA
RDS


> detailed comments inline below.
>

I followed up on Roland's answers to your questions, and added a few 
replies inline below:

<snip>

>> Usage Model:
>>
>> - MR allocated with ib_alloc_mr()
>>
>> - Page lists allocated via ib_alloc_fast_reg_page_list().
>>
>> - MR made VALID and bound to a specific page list via
>> ib_post_send(IB_WR_FAST_REG_MR)
> 
> Can the same ib_alloc_fast_reg_page_list() page list be
> bound to more than one MR?
> What happens if a user tries to issue a
> ib_post_send(IB_WR_FAST_REG_MR) to a VALID MR?
> 
> How can the memory be read/written?
> If the MR allows remote operations, then RDMA writes could be
> used. An RDMA READ could be used. What about local access
> by the host CPU?
>

LOCAL_WRITE can be supplied allowing the device to do local IO.

> 
> What does pbl_depth actually control?

It allows the device to pre-allocate the page_list resources in HW.

> Is it the maximum size page list that can be used in a
> ib_post_send(IB_WR_FAST_REG_MR) work request?

Yes.

> 
> pbl_depth should be unsigned since I don't think negative values
> make sense.
>

Ok.


>> @@ -676,6 +682,17 @@ struct ib_send_wr {
>>  			u16	pkey_index; /* valid for GSI only */
>>  			u8	port_num;   /* valid for DR SMPs on switch only */
>>  		} ud;
>> +		struct {
>> +			u64				iova_start;
>> +			struct ib_fast_reg_page_list	*page_list;
>> +			int				fbo;
> 
> What is fbo? First byte offset?
> I assume fbo can't be negative so it should be "unsigned"
>

Ok.