Re: [Xen-devel] xen-unstable stubdom build-failure when debug=n

[Sander Eikelenboom <linux@xxxxxxxxxxxxxx>]

Tuesday, July 22, 2014, 9:11:06 AM, you wrote:


> Monday, July 21, 2014, 6:24:33 PM, you wrote:

>> On Mon, 2014-07-21 at 17:13 +0100, Ian Campbell wrote:

>>> My guest is that turning off debug increases the optimisation level
>>> which somehow makes gcc decide this code is now wrong.

>> Well, I was right, but not how I thought...

>> Turns out there are two tpm_tis.c's in our source base and I was looking
>> at the wrong one ;-)

>> The code in extras/mini-os/tpm_tis.c rather than
>> stubdom/ioemu/hw/tpm_tis.c does look more like it might plausibly
>> generate the error I'm seeing this is:

>> tpm_tis.c:618:71: error: array subscript is below array bounds 
>> [-Werror=array-bounds]

>> Line 618 does appear to include the use of a signed variable as an array
>> offset, but I'm not sure how or why gcc is proving that it is negative.

>> The code in question appears to be identical in 4.4.0 . I'm not sure
>> what has changed.

>> Perhaps Daniel (CCd) has some idea what is going on here.

>> There's still the question of the error Sander is seeing which is
>> stubdom/lwip-x86_64/src/core/dhcp.c:1359:71: error: array subscript is above 
>> array bounds [-Werror=array-bounds]

>> Sander, did this used to work? If so can you identify when the
>> regression occurred?

> Hmm it's a pretty large window .. (essentially 4.4.0-release till now),
> it was only due to Konrad's reporting of the apparent major performance 
> regression that
> ended up being a difference between debug=y and debug=n build, that made me 
> wondering if
> i would notice that effect .. so i tried a debug=n build.

> If the build-test-system still has some slack, perhaps add a (weekly or only 
> on a push) test that does a debug=n build ?
> (instead of waiting for the end of the release cycle and somewhere in the 
> RC's to fixup the things that
> happen to turn up with debug=n (if my memory doesn't fail me too much, this 
> has happened before)

> (BTW i also use Debian Wheezy)
> --
> Sander

>> Ian.

Hi Ian,

Just tested some more with a strange outcome:

When doing:
~# git checkout  RELEASE-4.4.0
~# make mrproper && ./configure && make debug=n

It also fails with the error in dhcp i mentioned before.

However if i do:
~# git checkout  RELEASE-4.4.0
~# make mrproper && ./configure && make

It builds fine .. (and release-4.4.0 should have the implicit debug=n)
So what is different when explicitly giving debug=n ?
--
Sander








_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel