IRC channel logs

<stikonas>oriansj: can you merge this https://github.com/oriansj/M2libc/pull/45 ?

<stikonas>I'll switch all risc-v defines to lowercase so that it mathes more those new x86 define style and also the ones I'm using in riscv64/mescc

<stikonas>(more repos will need updates but M2libc needs to be merged first)

<oriansj>stikonas: merged

<stikonas>oriansj: part 2 https://github.com/oriansj/M2-Planet/pull/55

<stikonas>by the way, I ran stage0-posix-riscv64 with all of this to completion

<stikonas>riscv32 needs a bit more work, so I'll probably finish stage0-posix-riscv* changes tomorrow...

<stikonas>but those I can merge myself anyway

<oriansj>stikonas: merged

<stikonas>thanks!

<emilytrau[m]>janneke: thanks for merging that change!

<fossy>yes, of course stikonas[m], it was rickmasters who did all the hard work to get this in - huge props to rickmasters :D

<janneke>stikonas[m]: i've bisected the scaffold/hello.c problem using this silly script:

<janneke> https://gitlab.com/janneke/mes/-/tree/wip-riscv/bisect.sh

<janneke>the problem is in _write (# Defining function _write)

<janneke>This: ld_____%a0,-0x08(%fp)

<janneke>now looks like

<janneke>rd_a0 rs1_fp !0x08 ld

<janneke>note the negative sign has been dropped

<janneke>it should be something like

<janneke>rd_a0 rs1_fp !-0x08 ld

<janneke>but that doesn't compile

<janneke>hex2 says: Target label -0x08 is not valid

<stikonas[m]>janneke: oh I must have missed it yesterday

<janneke>ah, but this works...

<janneke>rd_a0 rs1_fp !-8 ld

<janneke>weird, that means we cannot use hex!

<stikonas[m]>I'll tesok, let's switch to decimal

<stikonas[m]>Hex should work

<stikonas[m]>But maybe not for negative numbers

<stikonas[m]>Let me check M1 source

<janneke>ACTION really likes how oriansj created something that can sometimes be bisected trivially

<janneke>stikonas[m]: yeah, that's for negative numbers

<janneke>stikonas[m]: the !-0x08 => !0x08 oversight also happened consistently elsewhere, of course

<janneke>lib/linux/riscv64-mes-m2/syscall.c, eg

<janneke>it would be great if we could do that (bisecting) mix-and-matching also with M2-Planet generated M1 and MesCC-generated M1; that's (much) more tricksy to do if it works at all

<stikonas[m]>Hmm yes...

<stikonas[m]>It fails to detect that number is hex

<stikonas[m]>As it starts with - and not 0x

<stikonas[m]>Might be fixable...

<stikonas[m]>janneke: decimal is in else block here https://github.com/oriansj/M2libc/blob/main/bootstrappable.c#L149

<janneke>stikonas[m]: this seems to work:

<janneke>rd_a0 rs1_fp !0x-08 ld

<stikonas[m]>Oh!

<janneke>not sure if that might be even "more correct"

<stikonas[m]>Maybe let's fix mescc-tools instead

<stikonas[m]>!0x-8 is just wrong even if it works

<janneke>stikonas, reading _write.c, what do you think about:

<janneke>- asm ("rd_a7 !64 addi"); // SYS_write

<janneke>+ asm ("rd_a7 !64 addi #SYS_write");

<stikonas>addi has to be at the end

<stikonas>if that works

<stikonas>but I don't know it it does

<stikonas>this is read by M2-Planet?

<stikonas>if yes, then oriansj suggested

<stikonas> asm ("rd_a7 SYS_write "addi")

<stikonas>though his SYS_write already includede quotes around the number

<janneke>it works, it's just a comment

<stikonas>oh yes

<stikonas>sorry missed that you are still passing !64

<stikonas>janneke: I'll take a look at negative hexes later, need to run to work soon

<janneke>stikonas, np and there's no hurry!

<janneke>yeah, still using !64; just making the M1 better to read -- dunno

<stikonas>also need to make sure which is the correct way

<stikonas>need to find some specs

<stikonas>it might be that 0x-8 is the correct

<stikonas>no, python accepts only -0x8

<janneke>yeah, i'm not sure either

<stikonas>but not 0x-8

<janneke>python! bah

<janneke>guile accepts #x-33

<janneke>(add a winky there)

<stikonas[m]>GCC is the one we need to test...

<janneke>stikonas[m]: yeah, gcc says it's -0x03 and rejects 0x-03

<oriansj>janneke: I don't see much advantage for hex offsets for risc-v as unlike x86, it has little mapping to the resulting instruction due to how the encoding is jumbled.

<janneke>oriansj: sure, fair enough

<stikonas[m]>And positive ones are supported...

<stikonas[m]>So you could still do something like 0xFFFF

<janneke>yeah

<Piraty>ACTION boatstrips

IRC channel logs

2023-05-22.log