| 1 |
Kumba wrote: |
| 2 |
|
| 3 |
> Brian Koropoff wrote: |
| 4 |
> |
| 5 |
>> As far as I know, crossdev only builds a cross compiler which runs on |
| 6 |
>> your build system but generates executables for another |
| 7 |
>> processor/architecture. I've already used crossdev sucessfully to |
| 8 |
>> generate such a compiler for arm. What I want to do now is build a |
| 9 |
>> compiler that *runs* on arm and creates arm executables (that is, both |
| 10 |
>> the host and target platform is arm, not just the target). This should |
| 11 |
>> be as simple as using portage with my crossdev-built cross-compiler and |
| 12 |
>> emerging it, but the build fails. Is there any trick to getting portage |
| 13 |
>> to build a native gcc to be included in the arm root fs? |
| 14 |
> |
| 15 |
> |
| 16 |
> If I'm reading this right, then what you want to do is impossible. |
| 17 |
> You can't build a gcc that runs on arm, and builds for arm, on a |
| 18 |
> cross-host. gcc has a rather strange build phase, wherein it rebuilds |
| 19 |
> itself 2-3 times (if you watch a gcc build closely, you'll see an |
| 20 |
> "xgcc" and references to stage1 and stage2 bootstraps). As such, if |
| 21 |
> you want to build a native compiler, you have to build on the native |
| 22 |
> architecture. This is because when the initial bootstrap completes, |
| 23 |
> and gcc kicks off it's stage1 xgcc run, that xgcc is an arm binary, |
| 24 |
> and it tries to build a stage2 xgcc arm binary. If performed on a |
| 25 |
> fast x86 box, this will fail. |
| 26 |
> |
| 27 |
> |
| 28 |
> --Kumba |
| 29 |
> |
| 30 |
Sorry to but in here, however I've been building the armv5teb (IXP4xx |
| 31 |
chip) like this for months. Unfortunately, it's with open embedded (OE). |
| 32 |
|
| 33 |
Not only that, but I've natively built the following with the tools |
| 34 |
generated: |
| 35 |
Linux 2.6.11.2 |
| 36 |
Apache 2.0.5 |
| 37 |
Perl 5.8.7 ran make test Failed 1 test script out of 878, 99.89% |
| 38 |
okay which is a known coreutils bug |
| 39 |
Python 2.4.1 make test 244 OK 3 tests failed: test_mimetools |
| 40 |
test_random test_urllib2 |
| 41 |
|
| 42 |
OE is pretty libc neutral so this can usually be done for both glibc and |
| 43 |
uClibc. |
| 44 |
|
| 45 |
The hardened and PIE parts are missing from OE. Also the package are |
| 46 |
constantly changing and some packages build but don't work. It's a lot |
| 47 |
of chaos. And there are not nearly the number of packages as in Gentoo. |
| 48 |
Hopefully in the not to distant future will I'll have everything |
| 49 |
required to a stage1 build straight from the target. We may have |
| 50 |
everything now, but the finishing touches are being done to the native |
| 51 |
development environment. Currently, OE and the nslu2-linux project are |
| 52 |
in the middle of the SCM switch from BK to something else. Things are |
| 53 |
in flux right now and this squares the chaos. I'm sure it'll die down in |
| 54 |
a couple weeks. |
| 55 |
|
| 56 |
See |
| 57 |
http://www.nslu2-linux.org/wiki/HowTo/OpenSlugNativeCompileEnvironment. |
| 58 |
The following for all the packages that are cross built |
| 59 |
http://www.nslu2-linux.org/wiki/HowTo/OpenSlugDevInst |
| 60 |
|
| 61 |
Best Regards, |
| 62 |
|
| 63 |
Tommy B. |
| 64 |
[g2] |
| 65 |
|
| 66 |
|
| 67 |
|
| 68 |
|
| 69 |
|
| 70 |
|
| 71 |
|
| 72 |
|
| 73 |
|
| 74 |
|
| 75 |
|
| 76 |
-- |
| 77 |
gentoo-embedded@g.o mailing list |
Archives