Yeah, I did a bunch of work on the iOS WKdm implementation.
Conversation
That gave me a chuckle when I came across it.
(In WKdm_decompress on arm64)
4
1
20
That is ABSOLUTELY NOT my doing. I am still angry about it.
1
16
I also would never use src_buf / dest_buf. I used to do src/dst, but now I vastly prefer to just use register names.
1
3
(Because when I go back to edit it, it's a royal pain to figure out what registers are available for scratch or which names are actually the same register. So now I use the register name for everything and put a decoder ring at each main label.)
2
2
does there exist tooling to help people like you? one can imagine all sorts of great assembly-assist things, but the market would be so small it would be hard to justify producing it
2
4
I wanted a register-allocating, macro assembler for A64 with pluggable scheduler backends and couldn't find one so I wrote "sketchy" a few years ago
2
14
On top of there being few people wanting this tool, those few people can also all write their own instead of taking time to learn someone else's
1
3
Joe y would I use yr Perl script when i can write as good a perl script of my own??? it will do register allocation using 13 regexes.
1
2
7
Have you seen Samsung's kernel CFI implementation from before they gave up and moved to Google's upstream Clang CFI?
It's the most horrifying thing and by that I mean they pass compiled kernel code through a Python script using objdump to add CFI instrumentation as a total hack.
i'm vibing on a previously-undiscovered plane of existence
5
The only thing they actually implemented in the compiler (still GCC) was adding NOP padding for them to use in their post-processing Python script along with reserving at least one register. Extremely coarse-grained indirect call CFI + ret addr XOR with random value in register.
1
1
So, Clang CFI figures out all the functions that aren't indirectly callable from the fact that their address isn't taken and they aren't exported (works well with -fvisibility=hidden) and then enforces type-based CFI for the indirect calls where that's not the case.
1
1
Show replies