1 |
On Sunday 31 Jul 2016 23:18:00 Andrew Savchenko wrote: |
2 |
> On Sun, 31 Jul 2016 19:40:37 +0100 Mick wrote: |
3 |
> > Hi All, |
4 |
> > |
5 |
> > I am dipping my toe into cross-compile territory, in order to build i686 |
6 |
> > binaries for a 32bit box, which is too old to do its own emerges. I am |
7 |
> > using an amd64 box which is significantly faster to do all the heavy |
8 |
> > lifting and started applying this page: |
9 |
> > |
10 |
> > https://wiki.gentoo.org/wiki/Embedded_Handbook/General/Creating_a_cross-co |
11 |
> > mpiler |
12 |
> > |
13 |
> > which I followed up with: |
14 |
> > |
15 |
> > https://wiki.gentoo.org/wiki/Cross_build_environment |
16 |
> |
17 |
> And here comes this misconception again... Please, tell me, why on |
18 |
> the earth cross-compiling is needed for amd64 to produce i686 |
19 |
> binaries?! |
20 |
|
21 |
I thought it did. From what you're saying I got this wrong. When I read the |
22 |
first use case bullet point, on the 2nd URL above, I thought I had arrived at |
23 |
the right place. :-/ |
24 |
|
25 |
|
26 |
> amd64 CPU _natively_ supports x86 instructions, amd64 kernel |
27 |
> natively supports x86 code (this can be disabled during kernel |
28 |
> config, but usually it isn't), amd64 gcc *can* produce x86 binaries. |
29 |
|
30 |
I thought amd64 can run x86 binaries, but I wasn't aware that it can compile |
31 |
them too, or what is needed to achieve this. My knowledge on gcc is pretty |
32 |
much minimal. I did search the Wiki, gentoo.org and Google for it, but all I |
33 |
could come across was cross-compiling. |
34 |
|
35 |
|
36 |
> There are two ways to help older x86 boxes to build packages faster: |
37 |
> |
38 |
> 1. Set up distcc to produce x86 code on your amd64 processors. Just |
39 |
> add -m32 to your *FLAGS. |
40 |
|
41 |
I read somewhere in these unsuccessful searches of mine that distcc is |
42 |
deprecated and it is better to use cross-compiling instead ... |
43 |
|
44 |
|
45 |
> 2. Copy old box system to a chroot dir on amd64. Run setarch i686 |
46 |
> and chroot to that directory, and build 32-bit packages as usual! |
47 |
> There are two ways to deliver them: |
48 |
> |
49 |
> 2.a. Generate binary packages on new box and install them on old |
50 |
> boxes. |
51 |
|
52 |
OK, I'll uninstall crossdev and try 2.a in the first instance. Is there a Wiki |
53 |
page explaining what parts of the x86 system are needed to carry across to the |
54 |
amd64 guest_root_fs? I wouldn't think I will need the whole x86 fs? Anything |
55 |
else I need to pay attention to? |
56 |
|
57 |
|
58 |
> 2.b. Instead of copying old box's root, mount it over NFS. |
59 |
|
60 |
I'll look into this later, after I get 2.a going. |
61 |
|
62 |
|
63 |
> I'm currently using 1, but planning to switch to 2.a, because |
64 |
> distcc can't help with everything (execution of java, python, |
65 |
> autotools and other stuff can't be helped with distcc). |
66 |
> |
67 |
> I used 2.b earlier on very old box (it is dead now). |
68 |
> |
69 |
> 3. Well, one can do full cross-compilation as you proposed, but |
70 |
> this is ridiculous. Cross-compilation is always a pain and if it |
71 |
> can be avoided, it should be avoided. |
72 |
|
73 |
Thanks for this advice. I am not particularly interested to use crossdev if |
74 |
it is not the best suited tool for the job, but I wasn't aware of the |
75 |
alternatives you suggested and haven't as yet found any HOWTOs on it. |
76 |
|
77 |
-- |
78 |
Regards, |
79 |
Mick |