svn commit: r341578 - head/sys/dev/mlx5/mlx5_en

[Gary Jennejohn]

On Mon, 17 Dec 2018 14:50:04 -0500
Andrew Gallatin <gallatin at cs.duke.edu> wrote:

> On 12/17/18 2:08 PM, Bruce Evans wrote:

[snip]
> > iflib uses queuing techniques to significantly pessimize em NICs with 1
> > hardware queue. On fast machines, it attempts to do 1 context switch per  
> 
> This can happen even w/o contention when "abdicate" is enabled in mp
> ring. I complained about this as well, and the default was changed in
> mp ring to not always "abdicate" (eg, switch to the tq to handle the
> packet). Abdication substantially pessimizes Netflix style web 
> uncontended workloads, but it generally helps small packet forwarding.
> 
> It is interesting that you see the opposite.  I should try benchmarking
> with just a single ring.
> 

Why are iflib and ifdi compiled into EVERY kernel with device
ether and/or device pci when only a few NICs actually use iflib? 
This is really unnecessary bloat in an already bloated kernel.

I use if_re which does not use iflib.

I removed iflib and ifdi from /sys/conf/files and my network
still works just fine.  It seems to me like these iflib entries
need finer-grained options, e.g. one of the NICs which use
iflib is enabled, before pulling them into the kernel build.

-- 
Gary Jennejohn