Re: if_wm.c workqueue

[Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>]

Hi,

On 2022/05/27 10:11, Emmanuel Dreyfus wrote:
> On Thu, May 26, 2022 at 05:46:31PM +0900, Kengo NAKAHARA wrote:
> > Could you use head of netbsd-9 branch?
>
> Tried that, workqueue now works fine, and this improves a lot
> the behavior against heavy flows.

I am grad to hear that.


> Is there a way to tune the priority of the workqueue against
> user processes?

I don't know to change kthread priority at run time.  Currently,
the workqueue's priority (PRI_SOFTNET) is higher than user processes.


> The wm(4) man page needs a word about this, I will try to update
> it. What about the other sysctls?
>
> hw.wm0.q0.txq_free = 4088
> hw.wm0.q0.txd_head = 1513
> hw.wm0.q0.txd_tail = 1514
> hw.wm0.q0.txq_next = 1514
> hw.wm0.q0.txq_sfree = 55
> hw.wm0.q0.txq_snext = 58
> hw.wm0.q0.txq_sdirty = 48
> hw.wm0.q0.txq_flags = 0
> hw.wm0.q0.txq_stopping = 0
> hw.wm0.q0.txq_sending = 1
> hw.wm0.q0.rxq_ptr = 82

I think they are not needed to be written in man, as they are debug
information strongly dependent on implementation.  About other device
drivers, not all sysctls are written in man.


Thanks,

--
//////////////////////////////////////////////////////////////////////
Internet Initiative Japan Inc.

Device Engineering Section,
Product Division,
Technology Unit

Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>