[Click] Re: Click on AMD64

[Qinghua(Kevin) Ye]

Hi Felipe,

Are you sure that the driver you loaded is exact the polling driver? The
first thing I would rather to make sure is that you have correctly compiled
the right driver during kernel compiling. Acturally, I would prefer to load
the driver manually, so you can debug it.

During the initialization of PollDevice Element, it will call e1000_poll_on
function, which will print a message "e1000_poll_on\n" on system log. Could
you check it in your /var/log/message to make sure it was there?

Regards,
Qinghua(Kevin) Ye
----- Original Message ----- 
From: "Felipe Huici" <felipehuici at gmail.com>
To: "Qinghua Ye" <qinghua at cs.ualberta.ca>
Cc: "Adam Greenhalgh" <A.Greenhalgh at cs.ucl.ac.uk>
Sent: Tuesday, July 05, 2005 4:52 AM
Subject: Re: Click on AMD64


Hi Qinghua,

I've read the posting you made regarding patches to linux/click. Most
of them coincide...I did notice that a lot of the patches you made
were due to type problems, I think Eddie fixed those in the latest CVS
click (which version of click are you using?). Besides those, we've
had to make the following changes:

1) /usr/src/linux/asm-x86_64/mmu_context.h (line 57), missing :
(introduced by click patch)
2) /usr/src/linux/asm-x86_64/thread_info.h (line 63), change void * to
struct thread_info *
3) /usr/src/linux/asm-x86_64/system.h (line 249), undefined LOCK_PREFIX
4) /usr/src/click/include/click/sync.hh (line 48), changed to
#if((!defined(__i386__))&&(!defined(__x86_64__))) as per your posting
5) /usr/src/click/linuxmodule/proclikefs.c (line 26), commented line
out as per your posting

With these in place we get click to compile and run in kernel space
using interrupts. However, when we try to click-install a config that
uses PollDevice, the router fails to install complaining that the
interface does not support polling. What we tried doing was apply the
given 2.6.11 click patch and copy the driver files from
/usr/src/click/drivers/e1000-5.x to /usr/src/linux/drivers/net/e1000
before compiling the kernel (previously we had omitted this copy step
and it hadn't worked). This didn't work though....do you have any
suggestions? Thanks

Felipe

On 7/4/05, Qinghua Ye <qinghua at cs.ualberta.ca> wrote:
> Hi Felipe,
>
> I was also working on the polling these days. However, my NICs are
Broadcom
> 5703.There is no available polling support in Click CVS. Based on the
> polling extension of IntelPro 1000 code, I also extended the driver for
> polling support. Unfortunately, I am still have trouble getting it to work
> properly. If there is no comming packet in one Nic's RX, then the TX
> doesn't work. Otherwise, it works ok. I guess I missed some status flush
> operation. What's your problem? Hope we can work together to get it run.
>
> for the kernel space click on AMD-64, have you read my code in the mail
> archive? Are there any inconsistency between your modification and mine?
> Hope we can compare it and avoid any bugs. Thanks.
>
> Qinghua
>
> On Mon, 4 Jul 2005, Felipe Huici wrote:
>
> > Dear Qinghua,
> >
> > We're currently trying to get Click working on a Dual-opteron AMD-64
> > box with linux 2.6.11.6. Although we have gotten this to work in
> > kernel space using interrupts (with a few minor modification to the
> > linux/click source code), we're having trouble getting it to work with
> > polling using IntelPro 1000/MT (dual Gigabit card, controller
> > 82546GB). Have you been able to get click to work using polling?
> > Thanks for the help
> >
> > Felipe Huici
> > UCL, research student
> >
>
>