MFC requests for 6.3

[Goran Lowkrantz]

<brian at hyperreal.org> wrote:

> On Mon, 29 Oct 2007, Goran Lowkrantz wrote:
>>> On Sun, 28 Oct 2007, Brett Glass wrote:
>>>> I would like to request that some useful work on networking be MFCed
>>>> from -CURRENT to -STABLE in time for the release of FreeBSD 6.3.
>>>
>>> The thing that's worried me, following only stable at freebsd.org traffic,
>>> were the reports about the new em driver (6.6.6?) causing hangs and
>>> other problems with people who have been updating RELENG_6 - have those
>>> who had those problems seen their issues resolved?  I've got remote
>>> boxes using em that I can't risk making inaccessible and can't test
>>> locally.
>>
>> All my problems with watchdog timeouts and em 6.6.6 occurred when  em
>> shared  interrupt with USB but I don't know if it's MB, em or usb that's
>> the problem.  Removing USB from the kernel or switching to polling and
>> the driver works  just fine.
>>
>> The fact that one of my test machines didn't have any problems was
>> because it  had no USB in it's kernel. I had removed USB as I had
>> problems with watchdog  timeouts with the bge driver when I first
>> upgraded to the D915GAV MB ages  ago. The bge also shared interrupt with
>> USB.
>
> Thanks.  /var/run/dmesg.boot says my em0 and em1 don't share an IRQ with
> USB or anyone else, so sounds like I'm in the clear, yay.  But it still
> seems like something the vendor (Intel?) should be looking into before a
> another release on RELENG_6 is cut - even if USB is the culprit, it
> sounded like the problems started for em users when the driver was
> updated from 6.2.9 to 6.6.6.
>

Yes, that's correct.  But then 6.6.6 is faster, so it may be that we have a 
timing problem in any driver when they share interrupt with a Gigant locked 
driver. Isn't there a new non-Gigant USB driver that is waiting for 
testers? I have two workstations with em that needs usb that I could test 
the combination on.

/glz