[Click] e1000e-0.4.1.7 polling driver update.

[Joonwoo Park]

Hi Eddie,

You're very welcome.   I'm the one should thank you for maintaining
this brilliant project!

That's good idea to poll_on() returns buffer length.
In terms of commenting out there was added lines mistakenly in my
previous e1000e patch so people had to  comment out those lines from
patch to build successfully.
Those issue has resolved at this patch.
However that was not related to kernel oops.  So please let me know if
you are pointing out another issue.

The one thing about kernel oops was an issue with large mtu on 82571+ by Nuutti.
It's fixed with this patch.

Thanks,
Joonwoo

On Thu, Feb 25, 2010 at 6:57 PM, Eddie Kohler <kohler at cs.ucla.edu> wrote:
> Hi Joonwoo,
>
> I had a look at this patch; thanks so so so so so much for doing the work!
> There are aspects of it that I did not like and changed.  In particular,
> adding a new rx_buffer_len function is not the right way to go, since it
> would require changing all our existing kernel patches and any other patched
> drivers.  Instead I changed the poll_on() function to return the correct rx
> buffer length, and updated PollDevice to support this.
>
> This version of the patch is now checked in to click/drivers.
>
> I also note that we have a bunch of questions and comments from people
> complaining about crashes, where the right answer is to "uncomment some code
> that you had commented out."  Am I right that the 2010-02-21 patch you sent
> already has this code uncommented?  Is it good to go?
>
> All: Please use the e1000e-0.4.1.7-2010-02-25 patch that is currently in the
> tree and let me know how it goes.
>
> Eddie
>
>
> Joonwoo Park wrote:
>>
>> Hi,
>>
>> I've updated e1000e-0.4.1.7 polling extension patch.
>> Mainly this change set intends to fix kernel oops which happens on
>> Intel 82571+ with mtu > 1500. (Nuutti Varis <nvaris at cc.hut.fi>
>> reported this bug)
>> It also contains minor fixes to fix another kernel oops and code clean up.
>>
>> If you are using old version of driver, please try this patch and let
>> me know how it works. It will be very appreciated.
>>
>> Patch set contains followings
>> - e1000e-0.4.1.7-2010-02-21.polling.patch: cumulative patch to apply
>> to vanilla e1000e-0.4.1.7 driver.
>> - 0001-PollDevice-allocates-specified-size-of-receive-buffe.patch:
>> patch to apply to click to get proper rx buffer length.
>> - 0001-added-rx_buffer_len-to-net_device-structure.patch: patch to
>> apply to linux after linux-2.6.24.7-patch.
>>
>> Known issue: It seems to me we've been having race in the driver when
>> PollDevice() is already running if interface goes up & down (mtu
>> change also makes device up & down) but I couldn't fix it yet.
>>
>> Eddie,
>> Please review this and consider to apply.
>>
>> Thanks,
>> Joonwoo
>>
>