[dpdk-dev] [PATCH] net/thunderx: sync mailbox definitions with Linux PF driver

[Jerin Jacob]

On Thu, Mar 23, 2017 at 01:26:15PM +0000, Ferruh Yigit wrote:
> On 3/23/2017 1:20 PM, Jerin Jacob wrote:
> > On Wed, Mar 22, 2017 at 10:33:14AM +0000, Ferruh Yigit wrote:
> >> On 3/22/2017 9:39 AM, Jerin Jacob wrote:
> >>> On Tue, Mar 21, 2017 at 02:53:29PM +0000, Ferruh Yigit wrote:
> >>>> On 3/21/2017 2:38 PM, Jerin Jacob wrote:
> >>>>> On Tue, Mar 21, 2017 at 02:31:41PM +0000, Ferruh Yigit wrote:
> >>>>>> On 3/20/2017 2:10 PM, Jerin Jacob wrote:
> >>>>>>> - bgx_link_status mbox definition was changed in Linux
> >>>>>>> commit 1cc702591bae ("net: thunderx: Add ethtool support")
> >>>>>>> - NIC_MBOX_MSG_RES_BIT related changes were never part of Linux PF driver
> >>>>>>>
> >>>>>>> Signed-off-by: Jerin Jacob <jerin.jacob at caviumnetworks.com>
> >>>>>>
> >>>>>> <...>
> >>>>>>
> >>>>>>> @@ -157,6 +151,7 @@ struct rss_cfg_msg {
> >>>>>>>  /* Physical interface link status */
> >>>>>>>  struct bgx_link_status {
> >>>>>>>  	uint8_t    msg;
> >>>>>>> +	uint8_t    mac_type;
> >>>>>>
> >>>>>> Hi Jerin,
> >>>>>>
> >>>>>> Is this modification related to this patch?
> >>>>>
> >>>>> Yes Ferruh.
> >>>>>
> >>>>> This was related to the following section in git log comment.
> >>>>> ----
> >>>>> - bgx_link_status mbox definition was changed in Linux
> >>>>> commit 1cc702591bae ("net: thunderx: Add ethtool support") 
> >>>>> ---
> >>>>
> >>>> I see now, thanks. Since this is to sync with Linux PF, shouldn't it be
> >>>> used in driver, perhaps something like in Linux driver:
> >>>> "nic->mac_type = mbx.link_status.mac_type"
> >>>>
> >>>> What is the point of just adding definition without using it?
> >>>
> >>> That is to keep "link_up"(next element in the struct bgx_link_status) points
> >>> to correct location after the kernel change. 
> >>
> >> I see, thanks. It is pity that new field added into middle (or beginning
> >> if you exclude msg) of the struct.
> >>
> >>> I thought about, the backward
> >>> compatibility with older kernel, Is it OK to use linux/version.h in PMD drivers
> >>> to detect the kernel version?
> >>
> >> Technically possible, but kernel version check has its problems:
> >> 1) Adds maintenance cost, which gets worse by time.
> >> 2) If added a compile time check, it creates binary distribution
> >> problems, distro guys may not like PMD has dependency to kernel.
> >> 3) It will create dependency to Linux, assuming Thunderx PMD supported
> >> in FreeBSD
> >>
> >>> drivers/net/mlx5/mlx5_ethdev.c has similar
> >>> kernel detection mechanism to make it backward compatible.
> >>
> >> Yes, at least mlx added a dynamic check which prevents above issue 2.
> >>
> >>> If there are no issue with such approach, I will roll out a new revision.
> >>
> >> Is there any way to detect this dynamically, first things I can think of:
> >> - Any mbx versioning to rely on?
> >> - Any common NIC user header file provided by kernel, to use in DPDK, to
> >> prevent any this kind of issues in the future?
> >> - Any msg size value to detect if mac_type exists or not?
> >> - Heuristic approach to other values to detect mac_type existence?
> > 
> > Thanks for sharing the options.
> > First three options wont work. Heuristic option is pretty limited,
> > mac_type value can also be 0 or 1(same as link_up)
> > So I think, The best option now to use the patch as is(without kernel
> > check).
> > What do you think?
> 
> If this won't cause a problem for you with older kernels, I would prefer
> no kernel check option.

OK. I will ask our kernel team to backport the offending the kernel change
to old versions based on the requirement. Please merge this patch if you don't
see any other issue.
I can post a separate patch add to the kernel version check if it
is really required.


> 
> > 
> >>
> >>>
> >>>
> >>>>
> >>>>>
> >>>>>
> >>>>>>
> >>>>>>>  	uint8_t    link_up;
> >>>>>>>  	uint8_t    duplex;
> >>>>>>>  	uint32_t   speed;
> >>>>>>>
> >>>>>>
> >>>>
> >>
>