[syslinux] Strange bug since 3.83 version

[Gene Cumm]

On Fri, Nov 12, 2010 at 15:39, Eric Seigne <eric.seigne at ryxeo.com> wrote:
>  On 12/11/2010 20:22, H. Peter Anvin wrote:
>>
>> What are the actual machines?
>>
>> There are some machines with mainboard BIOS problems which cause this,
>> but I'm particularly interested in your case since you say 3.82 worked
>> but 3.83 didn't.  In particular, if you could try the 3.83-prereleases
>> (or even better, if you feel comfortable with building from source, use
>> "git bisect") it might be possible to narrow this to a specific change.
>>
>> However, first of all try "localboot 0" as Gene points out.
>>
>>        -hpa
>
> Theses computers are:
>  - lenovo S10-2 netbook
>  - gateway G630 laptop
> All of them are with BIOS/PXE network controler. I could play with S10-2
> (i've one in spare @work) but i don't have many time to try all 3.83-pre
> releases.

I'd suggest taking an approach similar to git, a binary search.
Seeing as they are numbered 1 through 12, start with 6 or 7 and choose
your next (3 or 10) based on if it acts like 3.82 or 3.83.  It should
only take about 4 tries to get it narrowed down to two prereleases.
As an example, I ran into an issue that had around 1000 commits
between a known broken release and known functional release and I
think it only took me 8 tries to find the commit that was related
(about 11 would have been worst case scenario).

> And i could try "localboot 0" tomorrow.

I think you should consider trying releases 3.82, 3.83, possibly 3.86
(the last of the 3.xx series), 4.03 and possibly 4.00 (if 4.03 yields
different results from 3.83+ and you have time) and also testing both
"LOCALBOOT 0" and "LOCALBOOT -1".  I've had a few machines that work
one way but don't work the other way.

If my online search is correct, the Lenovo IdeaPad S10-2 has the
RealTek RTL8101E while the Gateway G630 (is it a Gateway or eMachines
G630?) has the Atheros AR8132 (1969:1066, I think).

As far as gathering info from lspci, I've found the -nn option very
useful as it specifies the numbers and names on one line.

-- 
-Gene