| chadkelly | Upon trying to download the OpenRISC Ubuntu VirtualBox image (1.1G), the entire site stopped working from me. However, all other sites I visit work great. Is there a webserver limit on data per IP address? | 03:52 |
|---|---|---|
| stekern | llvm status: I'm now able to compile u-boot with it | 07:33 |
| stekern | only thing that doesn't work is the use of global register variables, since that gcc extension isn't supported in llvm | 07:34 |
| -!- Jia` is now known as Jia | 08:36 | |
| stekern | after hacking around that, it's close to working too :) | 10:18 |
| stekern | I get prompt, but then it crashes when trying to run any commands | 10:19 |
| stekern | looks like stuff gets linked in at wrong places | 10:19 |
| juliusb | nice one! | 12:14 |
| juliusb | how does that work? It still uses binutils' ld right? | 12:15 |
| stekern | yes | 12:23 |
| stekern | and as | 12:24 |
| stekern | even more correctly, it uses binutils for every binutilsy task ;) | 12:25 |
| stekern | it has an own assembler and linker, but I haven't even started to look into how that works | 12:27 |
| jemarch | hi | 12:27 |
| stekern | not many backends support it | 12:28 |
| stekern | but I don't think the fact that stuff get linked in at wrong places is ld's fault, more likely some asm attributes/directives that are not set correctly | 12:31 |
| stekern | i.e. a missing .section blabla or something similar | 12:32 |
| stekern | the problem I had prior to this was a similar one, the .weak attribute didn't get printed, so weak linked functions went missing | 12:33 |
| stekern | actually, seems very much related, the functions that get linked in at the wrong places are marked as weak | 12:49 |
| juliusb | ah right | 12:55 |
| stekern | the output from llvm has: .section .text.do_bootelf_exec,"axG",@progbits,do_bootelf_exec,comdat | 12:59 |
| stekern | output from gcc does not | 12:59 |
| jeremybennett | stekern: Remember the current binutils is known to have problems with weak symbols. I've never delved deeply into it, but it seems to be a problem that also affects the ARM linker, which may explain why we have it. | 13:01 |
| stekern | ok, I think the weak part is just a coincident, the real problem is that llvm pulls that weak function "out" of .text and marks it with .text.do_bootelf_exec | 13:05 |
| stekern | so the root problem is our u-boot .lds files, changing *(.text) to *(.text*) fixes it | 13:40 |
| stekern | http://pastebin.com/DJW6Ps99 | 14:51 |
| _franck_ | nice | 15:12 |
| stekern | still some problems when running 'help' though, something is getting unaligned | 15:16 |
| stekern | I've figured the unaligned accesses out, but I'm not sure how to fix it | 23:41 |
| stekern | yet | 23:42 |
| stekern | the problem is with dynamic stack allocations and outgoing calls arguments on the stack | 23:42 |
| stekern | I have to wedge the dynamic stack allocations in between | 23:44 |
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!