[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openbsd-tech
Subject:    Re: patch axen(4) (WIP)
From:       Nils Frohberg <nilsf () tako ! de>
Date:       2019-02-28 6:53:39
Message-ID: 20190228065339.GA13031 () tako ! de
[Download RAW message or body]

On Tue, Feb 26, 2019 at 08:57:57PM +0200, Artturi Alm wrote:
> On Tue, Feb 26, 2019 at 03:00:15PM +0100, Nils Frohberg wrote:
> > On Mon, Feb 25, 2019 at 03:50:48PM -0300, Martin Pieuchot wrote:
> > > On 25/02/19(Mon) 14:52, Nils Frohberg wrote:
> > > > Hi,
> > > > 
> > > > as mentioned previously, I'm looking into axen(4). While searching
> > > > for the cause of a panic (fixed since, thanks mpi@) I started to
> > > > rewrite parts of the driver. References were mainly the FreeBSD and
> > > > Linux drivers.
> > > 
> > > Please try to isolate parts of your diff that fixes issues and cosmetic
> > > changes.  The simpler it gets the easier it is for us to review it.
> > 
> > Sure, I'll send separate patches.
> > 
> > (I cvs up'ed my src forder in order to test compile the individual
> > patches and found out that xhci.c,v1.91 breaks my USB devices. I'll
> > send a mail to bugs@ later.)
> > 
> > > > I didn't get around to much testing/debugging lately, therefore I
> > > > wanted to share the current state (diff below).
> > > > 
> > > > The current state works a lot better than previously (for me). I
> > > > used to have a huge amount of ierrs (aprrox. 1 ierr per ipkt) and
> > > > often no packets would be transferred at all (or stop being transferred
> > > > after some time).
> > > 
> > > Do you know why?  What were the problems?
> > 
> > I'm not 100% sure, since I did a lot of back and forth. It finally
> > got better once I disabled EEE and lowered the watermark levels.
> > 
> 
> Have you looked at what NetBSD has done with their axen(4)? there has
> been 20commits in 2019 so far[0], while some of them are possibly,
> idk., useless to us(thinking about hw checksum offloading), there was
> some bug fixes that did look relevant to me, but i succesfully
> installed&built kernels on nfs over axen(4) a couple of weeks ago,
> so the bugs it has didn't feel critical enough for me to make
> a branch for them. that was on arm64/dwctwo(4), tbh. i haven't been
> happy with axen(4) on amd64/{e,x}hci(4) myself in the past either. :]

I wrote the diff last December. I looked at NetBSD's code back then, 
but they didn't have any significant changes.

At a cursory glance, many changes are similar to mine. But there are 
a few things that should be worth looking at.

> I guess i'm trying to say maybe it wouldn't hurt to sync a bit before
> deviating as much as atleast your whole WIP diff did. I haven't read
> your separate patches yet, but i'll try to get around to also testing
> those before weekend:]

The separate patches are (more or less) the big patch split up via 
$(git add -p). There are a few things that might be worth looking 
into, such as the pause water levels, enabling EEE, axen_bulk_size 
values, buffer sizes, ...

More testing would be great. Especially since this is the only box 
I have where I can attach it to xhci.

> -Artturi
> 
> [0] https://github.com/NetBSD/src/commits/trunk/sys/dev/usb/if_axen.c
> 

[prev in list] [next in list] [prev in thread] [next in thread]