crossbow discuss - Dec 2008 - crossbow change causing NIC driver problems

I was talking to Sunay in the hall (convenience of having his office by mine;-) 
and he suggested sending a message to this mailing list.

There was changes that were putback to 105 with crossbow, for the sfe driver. I 
have not been able to confirm that these changes work on Solaris yet, and 
having a hard time getting them installed and the driver loaded.

This looks like the changeset that changes went back into sfe_util.c and 
sfe_utils.h, AFAICT.

changeset:   8275:7c223a798022

I have changes that I had been testing, prior to the crossbow putback, and 
which I would like to get a confirmation test so I can putback. The problem is 
that when I build the driver, I get an error on dma_alloc_handle error:

Dec 16 11:49:49 westlake sfe: [ID 599844 kern.warning] WARNING: sfe0: 
gem_alloc_memory ddi_dma_alloc_handle for bounce buffer failed: err=-8, i=0

I have not been able to confirm yet that the driver works properly with the 
changeset above, so not sure they work either.

I have the device in 2399, if Eric might want to look at it, or if someone 
knows what needs to be done to build.

I BFU''d my build and test machienes, but even the resulting nightly
package
gives this error.

I have seem similar problems before, where getting out of sync with a major 
putback like crossbow has caused similar issues, but not sure if I need to have 
a fresh install of 105 with crossbow or what.

I know this is not very finite information, but wanted to start this discussion 
to understand what I should do to build a working driver.

--

Alan DuBoff - Solaris x86 IHV/OEM Group

Hi Alan,

On Dec 16, 2008, at 5:16 PM, Alan DuBoff wrote:
> I was talking to Sunay in the hall (convenience of having his office  
> by mine;-)
> and he suggested sending a message to this mailing list.
>
> There was changes that were putback to 105 with crossbow, for the  
> sfe driver. I
> have not been able to confirm that these changes work on Solaris  
> yet, and
> having a hard time getting them installed and the driver loaded.
>
> This looks like the changeset that changes went back into sfe_util.c  
> and
> sfe_utils.h, AFAICT.
>
> changeset:   8275:7c223a798022
>
>
> I have changes that I had been testing, prior to the crossbow  
> putback, and
> which I would like to get a confirmation test so I can putback. The  
> problem is
> that when I build the driver, I get an error on dma_alloc_handle  
> error:
>
> Dec 16 11:49:49 westlake sfe: [ID 599844 kern.warning] WARNING: sfe0:
> gem_alloc_memory ddi_dma_alloc_handle for bounce buffer failed:  
> err=-8, i=0
I looked at our changes again and they don''t seem to be related to  
this part of the code.
>
>
> I have not been able to confirm yet that the driver works properly  
> with the
> changeset above, so not sure they work either.
>
> I have the device in 2399, if Eric might want to look at it, or if  
> someone
> knows what needs to be done to build.
>
> I BFU''d my build and test machienes, but even the resulting
nightly
> package
> gives this error.
So you tried a plain bfu of build 105 and you see the same issue? even  
without touching the driver after the bfu?

Could you try bfu''ing the onnv daily snapshot before our putback to  
see if you still see the problem? E.g. /ws/onnv-gate/archives/ 
{sparc,i386}/daily.1204

Nicolas.
>
>
> I have seem similar problems before, where getting out of sync with  
> a major
> putback like crossbow has caused similar issues, but not sure if I  
> need to have
> a fresh install of 105 with crossbow or what.
>
> I know this is not very finite information, but wanted to start this  
> discussion
> to understand what I should do to build a working driver.
>
> --
>
> Alan DuBoff - Solaris x86 IHV/OEM Group
> _______________________________________________
> crossbow-discuss mailing list
> crossbow-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/crossbow-discuss
-- 
Nicolas Droux - Solaris Kernel Networking - Sun Microsystems, Inc.
nicolas.droux at sun.com - http://blogs.sun.com/droux

On Wed, 17 Dec 2008, Nicolas Droux wrote:
> I looked at our changes again and they don''t seem to be related to
this part
> of the code.
I don''t think my changes are related at all.

It seems that something surfaced with the crossbow changes that cause the 
sfe to not successfully allocate.

Eric Cheng is looking into this some more, and trying to issolate it.
> So you tried a plain bfu of build 105 and you see the same issue? even 
> without touching the driver after the bfu?
Yes, that is the case.
> Could you try bfu''ing the onnv daily snapshot before our putback
to see if
> you still see the problem? E.g. 
> /ws/onnv-gate/archives/{sparc,i386}/daily.1204
Yes, and I can try to netinstall 105, I see it came out now.

--

Alan DuBoff - Solaris x86 IHV/OEM Group

On Wed, 17 Dec 2008, Nicolas Droux wrote:
> So you tried a plain bfu of build 105 and you see the same issue? even
> without touching the driver after the bfu?
>
> Could you try bfu''ing the onnv daily snapshot before our putback
to
> see if you still see the problem? E.g. /ws/onnv-gate/archives/
> {sparc,i386}/daily.1204
I do have an update to this problem, but not with your daily, but the 
official build 105 release which was posted last night.

105 as a fresh install works.

This is very similar as I was trying to describe before, that when I 
installed 104, somehow in how it created the device node is not compatible 
with the tree after it is BFU''d, and it won''t load the driver.

I saw something similar when clearview was putback, and the interfaces 
changed.

I tried to explain this, the other day to Dan Mick on irc, but he
didn''t
think that was the case. I am not sure if it is or not, but if I can build 
and pkgrm/pkgadd the sfe driver, I don''t have any other explaination.

If I can do that I will close the CR I created, but I need to build and do 
that from current bits.

The odd thing that I can''t explain, is that the bge driver works fine 
across the interface changes...hmmm...

I''ll let you know if my driver loads properly after I can build/test
it.

--

Alan DuBoff - Solaris x86 IHV/OEM Group