Files
ww/selfhost/cmd/w6l/out.ww
Hojun-Cho a376ec89eb lib/rt: rename rt_alloc → rt_malloc; rt.alloc → rt.malloc
Hare's canonical runtime allocator is rt::malloc with linker symbol
rt.malloc (ref/hare/rt/malloc.ha:27,78). ww kept the dot→underscore
Plan 9 convention (CLAUDE.md rule 4) so the linker symbol becomes
rt_malloc; the lib/rt exported function name becomes malloc; ww
callers say rt.malloc(...).

The language builtin keyword stays `alloc(T)!` — unchanged from Hare
(ref/hare/hare/lex/token.ha:21 ltok::ALLOC, parse/expr.ha:398
builtin()). The rename only touches the lowered linker symbol and the
exported function name behind it; the user-facing syntax for
heap-allocation is identical to Hare.

Surface:
- rt/alloc.s: TEXT rt_alloc → TEXT rt_malloc, labels updated
- lib/rt/malloc.ww: @symbol("rt_malloc") fn malloc(...) (was rt_alloc/alloc)
- rt/ensure.ww: local FFI decl + call site updated to malloc; `!` dropped
  on the direct FFI call (rt_malloc returns *void, not a tagged union)
- 18 .ww callers: rt.alloc(...) → rt.malloc(...)
- cstage cmd/wcc/check.c + wwstage selfhost/cmd/wcc/check.ww
  alloc-builtin suppression gate routes through ffi_resolve("malloc")
  for the lowering; the user-shadow check still keys on the BUILTIN
  KEYWORD "alloc" since that is what `alloc(...)` parses as. Adding
  "malloc" to the user-shadow check was unnecessary and was reverted
  during pre-commit review.
- cstage cmd/w6c/cgen.c: 2× ffi_resolve("alloc") → ffi_resolve("malloc")
- wwstage cgenexpr/cgenstmt: 2× ffiresolve(c, "alloc") → ffiresolve(c, "malloc")
- Test fixtures (700_e2e, 758_cgalloc_str_field, 990_selfhost, 992_w6l_ww,
  selfhost/test/tagged_ptr_ret.ww): updated inline ww sources to the new
  decl + call form

This is commit 2 of 3 in the lib/rt extraction (#38). Commit 3 closes
the OOM contract — return type becomes nullable *void and the builtin
lowering null-checks + propagates nomem.

Verified 132/132 + 995_self_rebuild byte-identity (5 wwstage tools
round-trip identical) + make clean cold rebuild.
2026-05-20 22:11:34 +09:00

180 lines
5.5 KiB
Plaintext

// selfhost/cmd/w6l/out.ww — port of cmd/w6l/out.c.
//
// Emit a static ELF64 executable. File layout (per the C original):
// [0..64) Ehdr
// [64..120) Phdr (one PT_LOAD)
// [120..0x1000) zero pad
// [0x1000..) .text bytes
// Single PT_LOAD covers the whole file, R+X. No interpreter, no .bss.
package w6l;
import os;
import rt;
import sym;
import dynout;
def ET_EXEC: u16 = 2u16;
def EM_X86_64_W: u16 = 62u16;
def EV_CURRENT: u32 = 1u32;
def ELFCLASS64: u8 = 2u8;
def ELFDATA2LSB: u8 = 1u8;
def PT_LOAD: u32 = 1u32;
def PF_X: u32 = 1u32;
def PF_W: u32 = 2u32;
def PF_R: u32 = 4u32;
def TEXT_OFF: u64 = 4096u64; // 0x1000
def PAGE_SZ: u64 = 4096u64;
// ---- little-endian byte writers ----------------------------------------
fn wru16(buf: *u8, off: u64, v: u16) void = {
buf[off] = (v & 255u16): u8;
buf[off + 1u64] = ((v >> 8u16) & 255u16): u8;
};
fn wru32(buf: *u8, off: u64, v: u32) void = {
buf[off] = (v & 255u32): u8;
buf[off + 1u64] = ((v >> 8u32) & 255u32): u8;
buf[off + 2u64] = ((v >> 16u32) & 255u32): u8;
buf[off + 3u64] = ((v >> 24u32) & 255u32): u8;
};
fn wru64(buf: *u8, off: u64, v: u64) void = {
wru32(buf, off, (v & 4294967295u64): u32);
wru32(buf, off + 4u64, ((v >> 32u64) & 4294967295u64): u32);
};
// ---- emit ---------------------------------------------------------------
export fn emitelf(l: *lnk, fd: i32, base: u64, entry: u64) i32 = {
// Dispatch: any loaded shared object plus any dynamic ref means
// we owe the loader a real PT_INTERP/PT_DYNAMIC binary.
if (l.sos != nil) {
if (l.dynn > 0) {
return emitdynelf(l, fd, base, entry);
};
};
let hasdata: bool = l.datalen > 0u64;
let rxend: u64 = TEXT_OFF + l.textlen;
// .data lands at the next page boundary so the loader can give
// it fresh R+W permissions without overlapping the R+X mapping.
let dataoff: u64 = 0u64;
let datava: u64 = 0u64;
if (hasdata) {
dataoff = (rxend + PAGE_SZ - 1u64) & ~(PAGE_SZ - 1u64);
datava = base + dataoff;
};
// Apply relocations now that the layout's textva/datava are
// known. Deferred from main.ww so the dyn path uses its own
// datava.
if (relocate(l, base + TEXT_OFF, datava) != 0) { return -1; };
// BSS optimisation: trailing zero bytes in .data can be left
// out of the file. The loader zero-fills the gap between
// p_filesz and p_memsz. Scan after l_relocate has applied any
// DATAR patches — anything still zero at the tail genuinely is
// zero-init. Matches cmd/w6l/out.c byte-for-byte.
let bsslen: u64 = 0u64;
if (hasdata) {
for (bsslen < l.datalen) {
let b: u8 = l.data[l.datalen - 1u64 - bsslen];
if (b != 0u8) { break; };
bsslen += 1u64;
};
};
let datafilelen: u64 = l.datalen - bsslen;
// One contiguous header buffer covering [0..0x1000), then .text.
let hdr: *u8 = rt.malloc(TEXT_OFF): *u8; // zero-initialised by mmap
// --- Ehdr (64 bytes) ---
hdr[0u64] = 127u8; // 0x7f
hdr[1u64] = 69u8; // 'E'
hdr[2u64] = 76u8; // 'L'
hdr[3u64] = 70u8; // 'F'
hdr[4u64] = ELFCLASS64;
hdr[5u64] = ELFDATA2LSB;
hdr[6u64] = EV_CURRENT: u8;
wru16(hdr, 16u64, ET_EXEC); // e_type
wru16(hdr, 18u64, EM_X86_64_W); // e_machine
wru32(hdr, 20u64, EV_CURRENT); // e_version
wru64(hdr, 24u64, entry); // e_entry
wru64(hdr, 32u64, 64u64); // e_phoff = sizeof(Ehdr)
wru64(hdr, 40u64, 0u64); // e_shoff
wru32(hdr, 48u64, 0u32); // e_flags
wru16(hdr, 52u64, 64u16); // e_ehsize
wru16(hdr, 54u64, 56u16); // e_phentsize
if (hasdata) { wru16(hdr, 56u64, 2u16); }
else { wru16(hdr, 56u64, 1u16); };
wru16(hdr, 58u64, 0u16); // e_shentsize
wru16(hdr, 60u64, 0u16); // e_shnum
wru16(hdr, 62u64, 0u16); // e_shstrndx
// --- Phdr #1 (R+X) at offset 64 ---
wru32(hdr, 64u64, PT_LOAD);
wru32(hdr, 68u64, PF_R | PF_X);
wru64(hdr, 72u64, 0u64); // p_offset
wru64(hdr, 80u64, base); // p_vaddr
wru64(hdr, 88u64, base); // p_paddr
wru64(hdr, 96u64, rxend); // p_filesz
wru64(hdr, 104u64, rxend); // p_memsz
wru64(hdr, 112u64, TEXT_OFF); // p_align
if (hasdata) {
// --- Phdr #2 (R+W) at offset 64+56=120 ---
wru32(hdr, 120u64, PT_LOAD);
wru32(hdr, 124u64, PF_R | PF_W);
wru64(hdr, 128u64, dataoff); // p_offset
wru64(hdr, 136u64, base + dataoff); // p_vaddr
wru64(hdr, 144u64, base + dataoff); // p_paddr
wru64(hdr, 152u64, datafilelen); // p_filesz
wru64(hdr, 160u64, l.datalen); // p_memsz
wru64(hdr, 168u64, PAGE_SZ); // p_align
};
// Write [0..0x1000) then .text.
let r1: (i64 | os.oserror) = os.writeall(fd, hdr, TEXT_OFF);
let n1: i64 = 0i64;
match (r1) {
case let v: i64 => n1 = v;
case let e: os.oserror => return -1;
};
if (n1 != TEXT_OFF: i64) { return -1; };
if (l.textlen > 0u64) {
let r2: (i64 | os.oserror) = os.writeall(fd, l.text, l.textlen);
let n2: i64 = 0i64;
match (r2) {
case let v: i64 => n2 = v;
case let e: os.oserror => return -1;
};
if (n2 != l.textlen: i64) { return -1; };
};
if (hasdata && datafilelen > 0u64) {
// Pad to the page-aligned data offset, then write only
// the non-zero prefix of .data. The rest is BSS — the
// loader zero-fills from p_filesz to p_memsz.
let here: u64 = TEXT_OFF + l.textlen;
let zero: u8 = 0u8;
for (here < dataoff) {
let r3: (i64 | os.oserror) = os.writeall(fd, &zero, 1u64);
match (r3) {
case let v: i64 => { };
case let e: os.oserror => return -1;
};
here += 1u64;
};
let r4: (i64 | os.oserror) = os.writeall(fd, l.data, datafilelen);
let n4: i64 = 0i64;
match (r4) {
case let v: i64 => n4 = v;
case let e: os.oserror => return -1;
};
if (n4 != datafilelen: i64) { return -1; };
};
return 0;
};