an elf loader prototype

[notzed]

I have grander plans but to nut out some details I threw together a basic elf loader this morning. I've only tested it on core 0,0 in a group and it wont handle multi-core elf files but it's still useful in it's current (incomplete) form.

The source drop is available here:


It includes a simple example that works on parallella-16 (at least). It's setup for compiling on-host.

The loader itself is simpler & faster than the srec loader and allows for symbolic resolution as demonstrated by the example. TBH i'm kind of baffled why the current sdk even includes an srec thing since elf is designed specifically to allow for minimal loader code. Last (and only) time i used srec was 20 years ago at uni ...

edit: I moved the location of the file.

[ysapir]

Thanks, @notzed, this is great. I actually started working on a similar loader a couple of days ago, but trying to follow the libelf-by-example tutorial utilizing the libelf library (). The obvious motivation for using this is to have a more future-proof solution than hard-coded pointer references, etc. (think 64-bit Epiphany, for example).

To your question - the reason why we use SREC's is purely legacy. Our previous, PC based EMEK systems used a different e-loader/e-server system and the original developer used the SREC as an intermediate step. When we developed the e-host library (the former incarnation of e-hal), the goal was to get a working interface ASAP - which we thought would be easiest by reusing the bulk of the loader code. At the time I was totally unfamiliar with the ELF system, not knowing the depth of the water, so SREC seems like a good enough compromise.

One thing I am not sure of - in your program you ignore the section headers table. My understanding is that the sections contain data used by the compiler and linker and not by the loader. If this is true, then ignoring those at load time is OK. Well - is this true? Are all loadable data (string tables, initialized static arrays, etc.) located in the program segments pointed to by program headers?

[G33KatWork]

[ysapir]

Thanks. Seems like we understand the "theory" the same way. I already used readelf to inspect simple Epiphany programs, which seem to adhere to these concepts. Strictly speaking, though, the fact that a couple of examples work well, does not mean that there will be no odd situation where using specific compiler and linker options, or a language other than C, will end up placing loadable data in the sections tables. This was the source of my concern. But I guess it is alright.

BTW, if we decide to go with the direct approach for elf parsing, can I re-use your code?

[G33KatWork]

[notzed]