Syslinux - Nov 2013 - Sysinux 6 will not boot ISOs on BIOS (i.e. pre-UEFI) systems

On 11/26/2013 01:14 AM, H. Peter Anvin wrote:
> On 11/25/2013 06:32 PM, H. Peter Anvin wrote:
>> On 11/25/2013 04:12 PM, Gerardo Exequiel Pozzi wrote:
>>>
>>> I made some more and more experiments beyond LBA 65535
>>>
>>> Works: * isolinux-4.07.bin (size 24574) * isolinux-5.00.bin (size
>>> 34816)
>>>
>>> Does not work: * isolinux-5.01.bin (size 34816) * isolinux-5.10.bin
>>> (size 36864) * isolinux-6.00.bin (size 38912) * isolinux-6.02.bin
>>> (size 43008)
>>>
>>> * I just tested exactly at LBA 65535 (works all versions) and LBA
>>> 65536 (fails all versions > 5.00).
>>>
>>
>> What the heck... there are just about zero plausible differences
>> between 5.00 and 5.01.
>>
>> I'm wondering if 17a36e2c91706512e6c425222c9a9a451d9db854 could
>> plausibly have anything to do with this, since that seems to be the
>> only possible patch that might affect isolinux.  Is there any way you
>> could try 5.01-pre1 and 5.01-pre2?  If -pre1 works and -pre2
doesn't,
>> it is likely that the problem lies in that commit.
>>
Yes this is the offending commit as I said in the previous message
[#1](*) reverting fixes it.
> 
> If you have the ability to rebuild Syslinux from source, perhaps you
> could try the following patch?
Yes. I done the rebuild, but I can not test just now (enconding video!).
I will post the result in ~10 hours ;)

If anyone that can reproduce the issue want to test, here is the
compressed (~167K) ISO image (~129M)[#2].
This is just and ISO image with isolinux.bin (at LBA 65536) patched +
"coco" (a file filled with \0), so if works propertly, you can see
message about missing ldlinux.c32.


Thanks for your time!


[#1] http://www.syslinux.org/archives/2013-November/021187.html
[#2]
https://dl.dropboxusercontent.com/u/88891866/test-isolinux-6.02.patched.iso.gz

https://dl.dropboxusercontent.com/u/88891866/test-isolinux-6.02.patched.iso.gz.sha1



(*) PS: Why messages sent to syslinux mailing-list takes too longer time
to become available (I am registered)?

-- 
Gerardo Exequiel Pozzi
\cos^2\alpha + \sin^2\alpha = 1

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 555 bytes
Desc: OpenPGP digital signature
URL:
<http://www.zytor.com/pipermail/syslinux/attachments/20131126/2d052063/attachment.sig>

On 11/26/2013 02:22 AM, Gerardo Exequiel Pozzi wrote:
> On 11/26/2013 01:14 AM, H. Peter Anvin wrote:
>> On 11/25/2013 06:32 PM, H. Peter Anvin wrote:
>>> On 11/25/2013 04:12 PM, Gerardo Exequiel Pozzi wrote:
>>>>
>>>> I made some more and more experiments beyond LBA 65535
>>>>
>>>> Works: * isolinux-4.07.bin (size 24574) * isolinux-5.00.bin
(size
>>>> 34816)
>>>>
>>>> Does not work: * isolinux-5.01.bin (size 34816) *
isolinux-5.10.bin
>>>> (size 36864) * isolinux-6.00.bin (size 38912) *
isolinux-6.02.bin
>>>> (size 43008)
>>>>
>>>> * I just tested exactly at LBA 65535 (works all versions) and
LBA
>>>> 65536 (fails all versions > 5.00).
>>>>
>>>
>>> What the heck... there are just about zero plausible differences
>>> between 5.00 and 5.01.
>>>
>>> I'm wondering if 17a36e2c91706512e6c425222c9a9a451d9db854 could
>>> plausibly have anything to do with this, since that seems to be the
>>> only possible patch that might affect isolinux.  Is there any way
you
>>> could try 5.01-pre1 and 5.01-pre2?  If -pre1 works and -pre2
doesn't,
>>> it is likely that the problem lies in that commit.
>>>
> 
> Yes this is the offending commit as I said in the previous message
> [#1](*) reverting fixes it.
> 
>>
>> If you have the ability to rebuild Syslinux from source, perhaps you
>> could try the following patch?
> 
> Yes. I done the rebuild, but I can not test just now (enconding video!).
> I will post the result in ~10 hours ;)
Yes! Fixed :)

I tested at LBA 65536 (~128M) and 655360 (~1280M) and works fine in both
cases.

The big mistery is why this works in some machines and why fails in
others. (maybe garbage in high word of "edx"?)

Thanks.

> 
> If anyone that can reproduce the issue want to test, here is the
> compressed (~167K) ISO image (~129M)[#2].
> This is just and ISO image with isolinux.bin (at LBA 65536) patched +
> "coco" (a file filled with \0), so if works propertly, you can
see
> message about missing ldlinux.c32.
> 
> 
> Thanks for your time!
> 
> 
> [#1] http://www.syslinux.org/archives/2013-November/021187.html
> [#2]
>
https://dl.dropboxusercontent.com/u/88891866/test-isolinux-6.02.patched.iso.gz
> 
>
https://dl.dropboxusercontent.com/u/88891866/test-isolinux-6.02.patched.iso.gz.sha1
> 
> 
> 
> (*) PS: Why messages sent to syslinux mailing-list takes too longer time
> to become available (I am registered)?
> 
> 
> 
> _______________________________________________
> Syslinux mailing list
> Submissions to Syslinux at zytor.com
> Unsubscribe or set options at:
> http://www.zytor.com/mailman/listinfo/syslinux
> Please do not send private replies to mailing list traffic.
> 

-- 
Gerardo Exequiel Pozzi
\cos^2\alpha + \sin^2\alpha = 1

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 555 bytes
Desc: OpenPGP digital signature
URL:
<http://www.zytor.com/pipermail/syslinux/attachments/20131126/22eaaa89/attachment.sig>

Hi,

hpa wrote:
> -               mov dx,cx
> +               movzx edx,cx
Gerardo Exequiel Pozzi:
> Yes! Fixed :)
> (maybe garbage in high word of "edx"?)
I am now pondering too, why my machine booted from high LBAs.

I understand that the new code zeros the upper 16 bit of EDX.
Was there remaining garbage from early BIOS activity before
isolinux.bin got started ?
Does my BIOS perhaps clean registers before handing over
control ?


Have a nice day :)

Thomas

On 11/26/2013 06:52 AM, Gerardo Exequiel Pozzi wrote:
>> 
>> Yes. I done the rebuild, but I can not test just now (enconding
>> video!). I will post the result in ~10 hours ;)
> 
> Yes! Fixed :)
> 
> I tested at LBA 65536 (~128M) and 655360 (~1280M) and works fine in
> both cases.
> 
> The big mistery is why this works in some machines and why fails
> in others. (maybe garbage in high word of "edx"?)
> 
Awesome, *thank you* for testing!

	-hpa