Re: [PATCH net-next V3 3/5] net/mlx4_en: Add HW timestamping (TS) support

[Amir Vadai <amirv@mellanox.com>]

On 28/04/2013 10:46, Richard Cochran wrote:
> On Sun, Apr 28, 2013 at 09:33:12AM +0300, Amir Vadai wrote:
>> On 25/04/2013 22:26, Richard Cochran wrote:
>>> On Tue, Apr 23, 2013 at 07:06:49PM +0300, Amir Vadai wrote:
>>>
>>>> +u64 mlx4_en_get_cqe_ts(struct mlx4_cqe *cqe)
>>>> +{
>>>> +	u64 hi, lo;
>>>> +	struct mlx4_ts_cqe *ts_cqe = (struct mlx4_ts_cqe *)cqe;
>>>> +
>>>> +	lo = (u64)be16_to_cpu(ts_cqe->timestamp_lo);
>>>> +	hi = ((u64)be32_to_cpu(ts_cqe->timestamp_hi) + !lo) << 16;
>>>                                                      ^^^^^
>>> That looks a bit strange. Can you explain?
>>
>> We need to increment the high order 32bit by one when the low order
>> 16bit value is zero, due to HW limitation.
> 
> So 'hi' increases by one, when 'lo' goes from 0x0000 to 0x0001?
Actually, 'hi' was supposed to be increased by one when 'lo' goes from
0xffff to 0x0000, and instead it is done by HW only when 'lo' goes from
0x0000 to 0x0001.
So the SW is filling this gap, and do the increase when 'lo' goes to
0x0000, later on the HW will make sure 'hi' is increased by one.

> 
> Thanks,
> Richard
> 

Amir