http://www.hallmark.com/ http://www.hallmark.com/wcsstore/HallmarkStore/images/globalNav/gnav_logo.gif
http://www.hallmark.com/webapp/wcs/stores/servlet/category1|10001|10051|-2|-2|pr... Online?lid=unShopOnline http://www.hallmark.com/wcsstore/HallmarkStore/images/globalNav/gnav_shop.gif
http://www.hallmark.com/webapp/wcs/stores/servlet/article|10001|10051|/HallmarkS... http://www.hallmark.com/wcsstore/HallmarkStore/images/globalNav/gnav_hmkmag.gif
http://www.hallmark.com/webapp/wcs/stores/servlet/category1|10001|10051|-1020! 01|-102001|ecards|unEcardandMore|E-Cards?lid=unEcardandMore http://www.hallmark.com/wcsstore/HallmarkStore/images/globalNav/gnav_ecards.gif
http://www.hallmark.com/webapp/wcs/sto...

On Sun, Mar 02, 2008 at 12:23:57AM +0000, Stuart Henderson wrote:

Yeah, wading through diffs that large can be a real pain :(

By the way, I would also suggest basing your work on the OpenBSD driver,
not the other way round, if this is feasible. Doing so should certainly
narrow down the diff you are creating for the OpenBSD driver.

On Sat, Mar 01, 2008 at 03:44:28PM -0700, Philip Guenther wrote:

Exactly.
I'm playing with TenDRA and compiling the OpenBSD userland makes a good
testcase.

On 2008/03/01 14:36, Stefan Sperling wrote:

Personally, I'm mostly trying to fix what's quite a nasty problem with
the existing driver, i.e. a nic where tx stops working after certain
media conditions.

Maybe it's just me (particularly now faced with a 3k+ line diff with a
bunch of changes in that are unrelated to the problems that it's been
reported to fix), but I think it's quite important to do things step by
step, the diffs to implement different things should be separated as
much as possible so it's easier to follow what changes have been made.

Surely the infrastructure would need to come first otherwise we'd
just be adding code to drivers that can't be tested?

On Sat, Mar 1, 2008 at 3:25 PM, Dasn lavabit.com> wrote:

FTR, C90 didn't permit a trailing comma, while C99 does. Many
compilers accepted the extra comma as an extension before C99 came
out. (The last compiler I had to use that didn't support it was the
AIX 4.x compiler. Blech!)

Philip Guenther

On 01/03/08 00:01 +0100, Tobias Ulmer wrote:

Does it matter? I think the comma here is convenient for extending the
enum in future.

PORNOLATION RELEASE

Hey Stuart,

On Sat, Mar 01, 2008 at 12:06:00PM +0000, Stuart Henderson wrote:

Yongari's version of that driver also has wake on lan support.

Would you be interested in trying to port the wake on lan support
over as well, instead of removing it from the driver?

I have an old diff that adds the necessary ioctls and teaches
ifconfig about them:
http://www.stsp.name/openbsd/OpenBSD-current-wol-2007-10-14.diff

The diff also has some driver changes (if_xl, if_vr, if_sis).

It's based on diffs I made for FreeBSD.

The diff is untested, but did compile back in October last year when
I wrote this on a train (literally). I never got around to see whether
it actually works.

The ifconfig changes should be straightforward.

If the section they appear in is protected by splnet(), probably yes.

Otherwise, you probably need to sprinkle in some splnet()/splx() pairs
yourself.

We have workq, but it's probably better to avoid this if you can?
What are the taskqueue's used for?

On Sat, Mar 01, 2008 at 12:06:00PM +0000, Stuart Henderson wrote:

Most probably you need splnet()/splx() calls as replacement (or probably a
splassert() to ensure correct spl levels). If you feel adventuresomely you
could use rwlocks and make dlg@ happy :)

Sounds like something similar to workq but not sure if I'm right.
Our workq may have different scheduling properties.
After glancing at the code -- the taskqueue is only used for mii link
checking -- I guess you can use a simple timeout callback for that.