feat(libbboeos): link clang-compiled exports into the shared blob

docs/CHANGELOG.md

@@ -11,6 +11,18 @@ time.
 
 ## [Unreleased](https://github.com/bboe/BBoeOS/compare/0.11.0...main)
 
+- **Wire user/libbboeos clang exports into the shared blob.**
+  `user/libbboeos/libbboeos.a` (clang-compiled) now links into `build/libbboeos`
+  alongside the asm helpers via the existing ld pipeline; `-ffunction-sections`
+  + `--gc-sections` drop everything the pointer table doesn't reference, so the
+  blob pays only for what it actually exports.  First entry: `strcmp`, with a
+  new `FUNCTION_STRCMP_PTR` constant at `FUNCTION_POINTER_TABLE + 52`.
+  Cc.py-side wiring (extern-call fallback that emits `call
+  [FUNCTION_<name>_PTR]` for unknown names) lands in a follow-up — this PR is
+  the build-pipeline plumbing.  `VDSO_SIGRETURN_OFFSET` moves from `0x460` →
+  `0xFE0` so the helper region can grow past 1 KB without colliding with the
+  sigreturn trampoline; sigreturn now lives near the end of page 0 (past the
+  pointer table at `0x800..0x83C`).
 - **cc.py: fold byte-immediate store + movzx reload through local.**  When a
   one-shot struct literal local is read via ``*(uint8_t *)&local`` — the driver
   port-I/O idiom — the compiler was emitting ``mov byte [ebp-N],
@@ -22,7 +34,6 @@ time.
   ~5 bytes per fold and trims the kernel's per-function frame traffic for every
   bitfield-struct + ``kernel_outb`` site.  Total kernel kasm reduction: 217
   bytes.
-
 - **Switch libbboeos build to nasm-elf + ld + linker script.**  vdso.asm is now
   `nasm -f elf32`-assembled with section directives
   (`.libbboeos.function_table`, `.libbboeos.text`, `.libbboeos.rodata`,

kernel/include/constants.asm

@@ -102,6 +102,18 @@
         %assign FUNCTION_PRINT_STRING_PTR       FUNCTION_POINTER_TABLE + 40
         %assign FUNCTION_PRINTF_PTR             FUNCTION_POINTER_TABLE + 44
         %assign FUNCTION_WRITE_STDOUT_PTR       FUNCTION_POINTER_TABLE + 48
+        ;; Clang-compiled libbboeos exports (Phase 3+).  Sorted
+        ;; alphabetically — and stays that way until the cc.py extern-call
+        ;; fallback ships and the first program with a baked-in
+        ;; FUNCTION_<name>_PTR offset gets written to disk.  After that
+        ;; the block becomes append-only (same ABI contract as the
+        ;; legacy block above) because reordering would shift offsets
+        ;; under existing on-disk bin/* binaries.  Until then, treat
+        ;; new entries like a freely-reorderable sorted list — insert
+        ;; in alphabetical position and renumber accordingly.  Order
+        ;; must match the LONG()s after the legacy block in
+        ;; user/libbboeos/libbboeos.ld.
+        %assign FUNCTION_STRCMP_PTR             FUNCTION_POINTER_TABLE + 52
         %assign IPPROTO_ICMP 1          ; Protocol argument to net_open for SOCK_DGRAM ICMP sockets
         %assign IPPROTO_UDP 17          ; Protocol argument to net_open for SOCK_DGRAM UDP sockets
         %assign KERNEL_VIRT_BASE 0FF800000h     ; Lowest kernel-virt address.  User pointers + lengths must stay strictly below this; idt.asm's user-fault triage and access_ok both gate on it.  Equals USER_STACK_TOP and DIRECT_MAP_BASE — all three move in lockstep.
@@ -280,7 +292,7 @@
         %assign USER_DATA_SELECTOR 23h  ; GDT[4] | RPL=3: ring-3 data segment (flat 4 GB)
         %assign USER_STACK_TOP 0FF800000h       ; Ring-3 stack top (one past last user-virt page); 64 KB stack at 0xFF7F0000-0xFF800000, 64 KB guard at 0xFF7E0000-0xFF7F0000.  Top sits exactly at the user/kernel boundary so ESP=USER_STACK_TOP can push 4 B into [0xFF7FFFFC, 0xFF800000) without crossing into the kernel half.
         %assign VDSO_PAGE_COUNT_MAX 4           ; Compile-time ceiling on how many 4 KB pages the kernel will map for the shared libbboeos image starting at FUNCTION_TABLE.  vdso_install reads `lib/libbboeos` at boot and rounds up to ceil(size / 4096); this constant sizes the per-page phys-frame array (vdso_code_phys) and bounds the per-program map loop in build_child_program_state.  Bumping it costs (4 * VDSO_PAGE_COUNT_MAX) bytes of kernel BSS.
-        %assign VDSO_SIGRETURN_OFFSET 0460h     ; offset within the vDSO page (FUNCTION_TABLE) of the __kernel_sigreturn trampoline that ends every signal handler — `mov ah, SYS_SYS_SIGRETURN; int 30h`.
+        %assign VDSO_SIGRETURN_OFFSET 0FE0h     ; offset within the vDSO page (FUNCTION_TABLE) of the __kernel_sigreturn trampoline that ends every signal handler — `mov ah, SYS_SYS_SIGRETURN; int 30h`.  Anchored near the end of page 0 (past the pointer table at 0x800..0x83C) so libbboeos.text can grow well past 0x460 as clang-compiled exports accumulate.
 
         ;; PIT constants used by entry.asm's IRQ 0 hookup and rtc.c's
         ;; PIT-driven sleep / tick counter.  PIC_EOI lives above with

make_os.sh

@@ -38,9 +38,14 @@ done
 # anchor; objcopy -O binary flattens the linked ELF into the on-disk
 # blob.
 mkdir -p build
+make -C user/libbboeos libbboeos.a >/dev/null || exit 1
 nasm -f elf32 -i kernel/include/ -o build/libbboeos.o user/vdso/vdso.asm || exit 1
+# --gc-sections drops every clang-compiled libbboeos function the
+# pointer-table LONG()s don't reference, so the blob pays only for
+# what it actually exports.
 ld -m elf_i386 -T user/libbboeos/libbboeos.ld -Map=build/libbboeos.map \
-    -o build/libbboeos.elf build/libbboeos.o || exit 1
+    --gc-sections -o build/libbboeos.elf \
+    build/libbboeos.o user/libbboeos/libbboeos.a || exit 1
 objcopy -O binary build/libbboeos.elf build/libbboeos || exit 1
 
 # Two-pass build: assemble kernel.bin first, measure its sector count,

user/libbboeos/Makefile

@@ -26,6 +26,7 @@ CC      := clang      # force clang — `?=` would lose to Make's built-in CC=cc
 CFLAGS  := --target=i386-pc-none-elf -m32 -march=i386 -mno-mmx -mno-sse -mno-sse2 \
            -mno-implicit-float -fno-vectorize -fno-slp-vectorize \
            -ffreestanding -fno-pic -fno-stack-protector \
+           -ffunction-sections -fdata-sections \
            -nostdlib -nostdinc -O2 -Wall -Wextra -Werror -Iinclude
 C_SRCS  := builtins.c ctype.c dirent.c errno.c math.c signal.c stdio.c stdlib.c string.c syscall.c
 OBJS     = $(C_SRCS:.c=.o) $(S_SRCS:.S=.o)

user/libbboeos/libbboeos.ld

@@ -10,34 +10,47 @@
  *   0x10046  .libbboeos.text             shared_* helper bodies (ported
  *                                        from src/lib/{proc,print}.asm).
  *   ...      .libbboeos.rodata           print_datetime_month_lengths.
- *   0x10460  .libbboeos.sigreturn        __kernel_sigreturn trampoline
+ *   0x10FE0  .libbboeos.sigreturn        __kernel_sigreturn trampoline
  *                                        (matches VDSO_SIGRETURN_OFFSET).
- *   0x10800  .libbboeos.pointer_table    FUNCTION_POINTER_TABLE — 13
- *                                        absolute LONG(shared_*) entries
- *                                        the linker resolves at link
- *                                        time, replacing the previous
- *                                        gen_libbboeos_pointers.py +
- *                                        python-concat dance.
+ *   0x10800  .libbboeos.pointer_table    FUNCTION_POINTER_TABLE.  13
+ *                                        legacy LONG(shared_*) entries
+ *                                        + a growing tail of clang-
+ *                                        compiled libbboeos exports.
+ *                                        Entry order MUST match the
+ *                                        FUNCTION_*_PTR offsets in
+ *                                        kernel/include/constants.asm.
  *
- * Future libbboeos C bodies (Phase 3+) land in .libbboeos.text, growing
- * the helper region until it bumps into the 0x10460 sigreturn anchor.
- * objcopy -O binary flattens the linked ELF into build/libbboeos, which
+ * Clang-compiled libbboeos C bodies (string.c, etc.) land in
+ * .libbboeos.text via `*(.text .text.*)`; -ffunction-sections + ld
+ * --gc-sections drop any function the LONG()s below don't reference,
+ * so only what the blob actually exports gets paid for.  objcopy -O
+ * binary flattens the linked ELF into build/libbboeos, which
  * make_os.sh adds to the disk image as `lib/libbboeos`. */
 
 ENTRY(function_table)
 
+/* Force ld to pull these archive members in (and seed --gc-sections).
+ * Without EXTERN, the LONG() references inside SECTIONS aren't enough
+ * to trigger archive resolution. */
+EXTERN(strcmp)
+
 SECTIONS {
     . = 0x00010000;
 
     .libbboeos.function_table : { KEEP(*(.libbboeos.function_table)) }
-    .libbboeos.text           : { *(.libbboeos.text) }
-    .libbboeos.rodata         : { *(.libbboeos.rodata) }
-
-    . = 0x00010460;
-    .libbboeos.sigreturn      : { KEEP(*(.libbboeos.sigreturn)) }
+    .libbboeos.text           : {
+        *(.libbboeos.text)
+        *(.text .text.*)
+    }
+    .libbboeos.rodata         : {
+        *(.libbboeos.rodata)
+        *(.rodata .rodata.*)
+    }
 
     . = 0x00010800;
     .libbboeos.pointer_table  : {
+        /* Legacy vDSO helpers (asm).  Order pinned by FUNCTION_*_PTR in
+         * kernel/include/constants.asm — entries here align 1:1. */
         LONG(shared_die);
         LONG(shared_exit);
         LONG(shared_get_character);
@@ -51,8 +64,19 @@ SECTIONS {
         LONG(shared_print_string);
         LONG(shared_printf);
         LONG(shared_write_stdout);
+        /* Clang-compiled libbboeos exports.  Sorted alphabetically.
+         * Stays freely-reorderable until the cc.py extern-call fallback
+         * ships and the first on-disk bin/* program bakes in an
+         * offset; from then on the block becomes append-only.  Until
+         * then, insert new entries in alphabetical position and update
+         * the matching FUNCTION_<name>_PTR offsets in
+         * kernel/include/constants.asm together. */
+        LONG(strcmp);
     }
 
+    . = 0x00010FE0;
+    .libbboeos.sigreturn      : { KEEP(*(.libbboeos.sigreturn)) }
+
     /DISCARD/ : {
         *(.note*)
         *(.comment)