Release Cosmopolitan v3.3

This change upgrades to GCC 12.3 and GNU binutils 2.42. The GNU linker appears to have changed things so that only a single de-duplicated str table is present in the binary, and it gets placed wherever the linker wants, regardless of what the linker script says. To cope with that we need to stop using .ident to embed licenses. As such, this change does significant work to revamp how third party licenses are defined in the codebase, using `.section .notice,"aR",@progbits`. This new GCC 12.3 toolchain has support for GNU indirect functions. It lets us support __target_clones__ for the first time. This is used for optimizing the performance of libc string functions such as strlen and friends so far on x86, by ensuring AVX systems favor a second codepath that uses VEX encoding. It shaves some latency off certain operations. It's a useful feature to have for scientific computing for the reasons explained by the test/libcxx/openmp_test.cc example which compiles for fifteen different microarchitectures. Thanks to the upgrades, it's now also possible to use newer instruction sets, such as AVX512FP16, VNNI. Cosmo now uses the %gs register on x86 by default for TLS. Doing it is helpful for any program that links `cosmo_dlopen()`. Such programs had to recompile their binaries at startup to change the TLS instructions. That's not great, since it means every page in the executable needs to be faulted. The work of rewriting TLS-related x86 opcodes, is moved to fixupobj.com instead. This is great news for MacOS x86 users, since we previously needed to morph the binary every time for that platform but now that's no longer necessary. The only platforms where we need fixup of TLS x86 opcodes at runtime are now Windows, OpenBSD, and NetBSD. On Windows we morph TLS to point deeper into the TIB, based on a TlsAlloc assignment, and on OpenBSD/NetBSD we morph %gs back into %fs since the kernels do not allow us to specify a value for the %gs register. OpenBSD users are now required to use APE Loader to run Cosmo binaries and assimilation is no longer possible. OpenBSD kernel needs to change to allow programs to specify a value for the %gs register, or it needs to stop marking executable pages loaded by the kernel as mimmutable(). This release fixes __constructor__, .ctor, .init_array, and lastly the .preinit_array so they behave the exact same way as glibc. We no longer use hex constants to define math.h symbols like M_PI.
2025-07-21 01:50:30 +00:00 · 2024-02-20 11:12:09 -08:00 · 2024-02-20 11:12:09 -08:00 · 957c61cbbf
commit 957c61cbbf
parent d3ff48c63f
736 changed files with 13726 additions and 9445 deletions
--- a/tool/build/fixupobj.c
+++ b/tool/build/fixupobj.c
@ -29,7 +29,6 @@
 #include "libc/errno.h"
 #include "libc/fmt/itoa.h"
 #include "libc/fmt/magnumstrs.internal.h"
-#include "libc/intrin/kprintf.h"
 #include "libc/limits.h"
 #include "libc/log/log.h"
 #include "libc/macros.internal.h"
@ -226,8 +225,56 @@ static void CheckPrivilegedCrossReferences(void) {
  }
 }

+// Change AMD code to use %gs:0x30 instead of %fs:0
+// We assume -mno-tls-direct-seg-refs has been used
+static void ChangeTlsFsToGs(unsigned char *p, size_t n) {
+  unsigned char *e = p + n - 9;
+  while (p <= e) {
+    // we're checking for the following expression:
+    //   0144 == p[0] &&           // %fs
+    //   0110 == (p[1] & 0373) &&  // rex.w (and ignore rex.r)
+    //   (0213 == p[2] ||          // mov reg/mem → reg (word-sized)
+    //   0003 == p[2]) &&          // add reg/mem → reg (word-sized)
+    //   0004 == (p[3] & 0307) &&  // mod/rm (4,reg,0) means sib → reg
+    //   0045 == p[4] &&           // sib (5,4,0) → (rbp,rsp,0) → disp32
+    //   0000 == p[5] &&           // displacement (von Neumann endian)
+    //   0000 == p[6] &&           // displacement
+    //   0000 == p[7] &&           // displacement
+    //   0000 == p[8]              // displacement
+    uint64_t w = READ64LE(p) & READ64LE("\377\373\377\307\377\377\377\377");
+    if ((w == READ64LE("\144\110\213\004\045\000\000\000") ||
+         w == READ64LE("\144\110\003\004\045\000\000\000")) &&
+        !p[8]) {
+      p[0] = 0145;  // change %fs to %gs
+      p[5] = 0x30;  // change 0 to 0x30
+      p += 9;
+    } else {
+      ++p;
+    }
+  }
+}
+
+static void RewriteTlsCodeAmd64(void) {
+  int i;
+  uint8_t *p;
+  Elf64_Shdr *shdr;
+  for (i = 0; i < elf->e_shnum; ++i) {
+    if (!(shdr = GetElfSectionHeaderAddress(elf, esize, i))) {
+      Die("elf header overflow #1");
+    }
+    if (shdr->sh_type == SHT_PROGBITS &&  //
+        (shdr->sh_flags & SHF_ALLOC) &&   //
+        (shdr->sh_flags & SHF_EXECINSTR)) {
+      if (!(p = GetElfSectionAddress(elf, esize, shdr))) {
+        Die("elf header overflow #2");
+      }
+      ChangeTlsFsToGs(p, shdr->sh_size);
+    }
+  }
+}
+
 // Modify ARM64 code to use x28 for TLS rather than tpidr_el0.
-static void RewriteTlsCode(void) {
+static void RewriteTlsCodeArm64(void) {
  int i;
  Elf64_Shdr *shdr;
  uint32_t *p, *pe;
@ -594,10 +641,11 @@ static void FixupObject(void) {
    CheckPrivilegedCrossReferences();
    if (mode == O_RDWR) {
      if (elf->e_machine == EM_NEXGEN32E) {
+        RewriteTlsCodeAmd64();
        OptimizePatchableFunctionEntries();
        GenerateIfuncInit();
      } else if (elf->e_machine == EM_AARCH64) {
-        RewriteTlsCode();
+        RewriteTlsCodeArm64();
        if (elf->e_type != ET_REL) {
          UseFreebsdOsAbi();
        }
--- a/tool/cosmocc/README.md
+++ b/tool/cosmocc/README.md
@ -9,12 +9,13 @@ reach a broader audience from the platform(s) of your choosing.

 ## What's Included

-This toolchain bundles GCC 11.2.0, Cosmopolitan Libc, LLVM LIBCXX, and
-LLVM compiler-rt. Additional libraries were provided by Musl Libc, and
-the venerable BSDs OSes. This lets you benefit from the awesome modern
-GCC compiler with the strongest GPL barrier possible. The preprocessor
-advertises cross compilers as both `__COSMOCC__` and `__COSMOPOLITAN__`
-whereas `cosmocc` additionally defines `__FATCOSMOCC__`.
+This toolchain bundles GCC 12.3.0, Cosmopolitan Libc, LLVM LIBCXX, LLVM
+compiler-rt, and LLVM OpenMP. Additional libraries were provided by Musl
+Libc, and the venerable BSDs OSes. This lets you benefit from the
+awesome modern GCC compiler with the strongest GPL barrier possible. The
+preprocessor advertises cross compilers as both `__COSMOCC__` and
+`__COSMOPOLITAN__` whereas `cosmocc` additionally defines
+`__FATCOSMOCC__`.

 ## Getting Started

@ -296,9 +297,9 @@ EINVAL: ... }` in cases where constants like `EINVAL` are linkable
 symbols. Your code will be rewritten in such cases to use a series of if
 statements instead, so that Cosmopolitan Libc's system constants will
 work as expected. Our modifications to GNU GCC are published under the
-ISC license at <https://github.com/ahgamut/gcc/tree/portcosmo-11.2>. The
+ISC license at <https://github.com/ahgamut/gcc/tree/portcosmo-12.3>. The
 binaries you see here were first published at
-<https://github.com/ahgamut/superconfigure/releases/tag/z0.0.30> which
+<https://github.com/ahgamut/superconfigure/releases/tag/z0.0.32> which
 is regularly updated.

 ## Legal
--- a/tool/cosmocc/bin/cosmocc
+++ b/tool/cosmocc/bin/cosmocc
@ -239,7 +239,7 @@ PLATFORM="-D__COSMOPOLITAN__ -D__COSMOCC__ -D__FATCOSMOCC__"
 PREDEF="-include libc/integral/normalize.inc"
 CPPFLAGS="-fno-pie -nostdinc -fno-math-errno -isystem $BIN/../include"
 CFLAGS="-fportcosmo -fno-dwarf2-cfi-asm -fno-unwind-tables -fno-asynchronous-unwind-tables -fno-semantic-interposition"
-LDFLAGS="-static -nostdlib -no-pie -fuse-ld=bfd -Wl,-z,norelro -Wl,--gc-sections"
+LDFLAGS="-static -nostdlib -no-pie -fuse-ld=bfd -Wl,-z,noexecstack -Wl,-z,norelro -Wl,--gc-sections"
 PRECIOUS="-fno-omit-frame-pointer"

 if [ x"$OPT" != x"-Os" ] && [ x"$MODE" != x"tiny" ]; then
--- a/tool/cosmocc/bin/cosmocross
+++ b/tool/cosmocc/bin/cosmocross
@ -49,7 +49,7 @@ PLATFORM="-D__COSMOPOLITAN__ -D__COSMOCC__"
 PREDEF="-include libc/integral/normalize.inc"
 CFLAGS="-fportcosmo -fno-dwarf2-cfi-asm -fno-unwind-tables -fno-asynchronous-unwind-tables -fno-semantic-interposition"
 CPPFLAGS="-fno-pie -nostdinc -fno-math-errno -isystem $BIN/../include"
-LDFLAGS="-static -no-pie -nostdlib -fuse-ld=bfd"
+LDFLAGS="-static -no-pie -nostdlib -fuse-ld=bfd -Wl,-z,noexecstack"
 APEFLAGS="-Wl,--gc-sections"
 PRECIOUS="-fno-omit-frame-pointer"

--- a/tool/cosmocc/package.sh
+++ b/tool/cosmocc/package.sh
@ -19,6 +19,7 @@ OUTDIR=${1:-cosmocc}
 APELINK=o/$(mode)/tool/build/apelink.com
 AMD64=${2:-x86_64}
 ARM64=${3:-aarch64}
+GCCVER=12.3.0

 make -j32 m= \
  $APELINK
@ -89,10 +90,10 @@ fetch() {
 OLD=$PWD
 cd "$OUTDIR/"
 if [ ! -x bin/x86_64-linux-cosmo-gcc ]; then
-  fetch https://github.com/ahgamut/superconfigure/releases/download/z0.0.30/aarch64-gcc.zip
+  fetch https://github.com/ahgamut/superconfigure/releases/download/z0.0.32/aarch64-gcc.zip
  unzip aarch64-gcc.zip
  rm -f aarch64-gcc.zip
-  fetch https://github.com/ahgamut/superconfigure/releases/download/z0.0.30/x86_64-gcc.zip
+  fetch https://github.com/ahgamut/superconfigure/releases/download/z0.0.32/x86_64-gcc.zip
  unzip x86_64-gcc.zip
  rm -f x86_64-gcc.zip
 fi
@ -113,14 +114,14 @@ for arch in aarch64 x86_64; do
  ln -sf $arch-linux-cosmo-objdump bin/$arch-unknown-cosmo-objdump
  ln -sf $arch-linux-cosmo-readelf bin/$arch-unknown-cosmo-readelf
  ln -sf $arch-linux-cosmo-strip bin/$arch-unknown-cosmo-strip
-  cmp -s libexec/gcc/$arch-linux-cosmo/11.2.0/ld.bfd libexec/gcc/$arch-linux-cosmo/11.2.0/ld
-  ln -sf ld.bfd libexec/gcc/$arch-linux-cosmo/11.2.0/ld
-  cmp -s libexec/gcc/$arch-linux-cosmo/11.2.0/ld.bfd bin/$arch-linux-cosmo-ld
-  ln -sf ../libexec/gcc/$arch-linux-cosmo/11.2.0/ld.bfd bin/$arch-linux-cosmo-ld
-  cmp -s libexec/gcc/$arch-linux-cosmo/11.2.0/as bin/$arch-linux-cosmo-as
-  ln -sf ../libexec/gcc/$arch-linux-cosmo/11.2.0/as bin/$arch-linux-cosmo-as
-  cmp -s libexec/gcc/$arch-linux-cosmo/11.2.0/ld.bfd bin/$arch-linux-cosmo-ld.bfd
-  ln -sf ../libexec/gcc/$arch-linux-cosmo/11.2.0/ld.bfd bin/$arch-linux-cosmo-ld.bfd
+  cmp -s libexec/gcc/$arch-linux-cosmo/$GCCVER/ld.bfd libexec/gcc/$arch-linux-cosmo/$GCCVER/ld
+  ln -sf ld.bfd libexec/gcc/$arch-linux-cosmo/$GCCVER/ld
+  cmp -s libexec/gcc/$arch-linux-cosmo/$GCCVER/ld.bfd bin/$arch-linux-cosmo-ld
+  ln -sf ../libexec/gcc/$arch-linux-cosmo/$GCCVER/ld.bfd bin/$arch-linux-cosmo-ld
+  cmp -s libexec/gcc/$arch-linux-cosmo/$GCCVER/as bin/$arch-linux-cosmo-as
+  ln -sf ../libexec/gcc/$arch-linux-cosmo/$GCCVER/as bin/$arch-linux-cosmo-as
+  cmp -s libexec/gcc/$arch-linux-cosmo/$GCCVER/ld.bfd bin/$arch-linux-cosmo-ld.bfd
+  ln -sf ../libexec/gcc/$arch-linux-cosmo/$GCCVER/ld.bfd bin/$arch-linux-cosmo-ld.bfd
 done
 cd "$OLD"

--- a/tool/emacs/c.lang
+++ b/tool/emacs/c.lang
@ -121,7 +121,6 @@ Keywords={
 "hasatleast",
 "nodebuginfo",
 "noreturn",
-"initarray",
 "mayalias",
 "dontinstrument",
 "interruptfn",
--- a/tool/emacs/cosmo-asm-mode.el
+++ b/tool/emacs/cosmo-asm-mode.el
@ -65,6 +65,7 @@
           "pltoff"
           "gotpcrel"
           "progbits"
+           "note"
           "nobits"
           "init_array"
           "fini_array"
--- a/tool/emacs/cosmo-c-builtins.el
+++ b/tool/emacs/cosmo-c-builtins.el
@ -204,6 +204,7 @@
         '("DebugBreak"
           "__veil"
           "__conceal"
+           "__notice"
           "__expropriate"
           "__yoink"
           "__dll_import"
--- a/tool/emacs/cosmo-c-keywords.el
+++ b/tool/emacs/cosmo-c-keywords.el
@ -87,7 +87,6 @@
           "nomsan"
           "dontubsan"
           "nostackprotector"
-           "initarray"
           "mayalias"
           "dontinstrument"
           "interruptfn"
--- a/tool/emacs/key.py
+++ b/tool/emacs/key.py
@ -335,7 +335,6 @@ cosmo_kws = frozenset([
  "forcealignargpointer",
  "forceinline",
  "hasatleast",
-  "initarray",
  "interruptfn",
  "mallocesque",
  "mayalias",
@ -394,7 +393,6 @@ cosmo_kws = frozenset([
  "forcealignargpointer",
  "forceinline",
  "hasatleast",
-  "initarray",
  "interruptfn",
  "mallocesque",
  "mayalias",
--- a/tool/hello/BUILD.mk
+++ b/tool/hello/BUILD.mk
@ -39,7 +39,7 @@ o/$(MODE)/tool/hello/hello.com.dbg:				\

 # uses apelink to turn it into an ape executable
 # support vector is set to all operating systems
-o/$(MODE)/tool/hello/hello.com:					\
+o/$(MODE)/tool/hello/hello.ape:					\
 		o/$(MODE)/tool/hello/hello.com.dbg		\
 		o/$(MODE)/tool/build/apelink.com		\
 		o/$(MODE)/tool/build/pecheck.com		\
@ -49,7 +49,7 @@ o/$(MODE)/tool/hello/hello.com:					\

 # uses apelink to generate elf-only executable
 # support vector = linux/freebsd/openbsd/netbsd/metal
-o/$(MODE)/tool/hello/hello-elf.com:				\
+o/$(MODE)/tool/hello/hello-elf.ape:				\
 		o/$(MODE)/tool/hello/hello.com.dbg		\
 		o/$(MODE)/tool/build/apelink.com		\
 		o/$(MODE)/ape/ape.elf
@ -59,7 +59,7 @@ o/$(MODE)/tool/hello/hello-elf.com:				\
 # support vector = macos/linux/freebsd/openbsd/netbsd
 # - great way to avoid attention from bad virus scanners
 # - creates tinier executable by reducing alignment requirement
-o/$(MODE)/tool/hello/hello-unix.com:				\
+o/$(MODE)/tool/hello/hello-unix.ape:				\
 		o/$(MODE)/tool/hello/hello.com.dbg		\
 		o/$(MODE)/tool/build/apelink.com		\
 		o/$(MODE)/ape/ape.elf
@ -71,7 +71,7 @@ o/$(MODE)/tool/hello/hello-unix.com:				\
 o/$(MODE)/tool/hello/hello-pe.com.dbg:				\
 		o/$(MODE)/tool/hello/hello-pe.o
 	@$(COMPILE) -ALINK.elf $(LINK) $(LINKARGS) $(OUTPUT_OPTION) -q -e WinMain
-o/$(MODE)/tool/hello/hello-pe.com:				\
+o/$(MODE)/tool/hello/hello-pe.ape:				\
 		o/$(MODE)/tool/hello/hello-pe.com.dbg		\
 		o/$(MODE)/tool/build/elf2pe.com
 	@$(COMPILE) -AELF2PE o/$(MODE)/tool/build/elf2pe.com -o $@ $<
@ -80,7 +80,7 @@ o/$(MODE)/tool/hello/hello-pe.com:				\
 o/$(MODE)/tool/hello/life-pe.com.dbg:				\
 		o/$(MODE)/tool/hello/life-pe.o
 	@$(COMPILE) -ALINK.elf $(LINK) $(LINKARGS) $(OUTPUT_OPTION) -q -e WinMain
-o/$(MODE)/tool/hello/life-pe.com:				\
+o/$(MODE)/tool/hello/life-pe.ape:				\
 		o/$(MODE)/tool/hello/life-pe.com.dbg		\
 		o/$(MODE)/tool/build/elf2pe.com
 	@$(COMPILE) -AELF2PE o/$(MODE)/tool/build/elf2pe.com -o $@ $<
@ -89,7 +89,7 @@ o/$(MODE)/tool/hello/life-pe.com:				\
 o/$(MODE)/tool/hello/wait-pe.com.dbg:				\
 		o/$(MODE)/tool/hello/wait-pe.o
 	@$(COMPILE) -ALINK.elf $(LINK) $(LINKARGS) $(OUTPUT_OPTION) -q -e WinMain
-o/$(MODE)/tool/hello/wait-pe.com:				\
+o/$(MODE)/tool/hello/wait-pe.ape:				\
 		o/$(MODE)/tool/hello/wait-pe.com.dbg		\
 		o/$(MODE)/tool/build/elf2pe.com
 	@$(COMPILE) -AELF2PE o/$(MODE)/tool/build/elf2pe.com -R 64kb -S 4kb -o $@ $<
--- a/tool/net/largon2.c
+++ b/tool/net/largon2.c
@ -32,10 +32,9 @@
 #include "third_party/lua/lua.h"
 #include "third_party/lua/lualib.h"

-asm(".ident\t\"\\n\\n\
-largon2 (MIT License)\\n\
-Copyright 2016 Thibault Charbonnier\"");
-asm(".include \"libc/disclaimer.inc\"");
+__notice(largon2_notice, "\
+largon2 (MIT License)\n\
+Copyright 2016 Thibault Charbonnier");

 // clang-format off
 /***
--- a/tool/net/lsqlite3.c
+++ b/tool/net/lsqlite3.c
@ -38,10 +38,9 @@
 #include "third_party/sqlite3/sqlite3.h"
 // clang-format off

-asm(".ident\t\"\\n\\n\
-lsqlite3 (MIT License)\\n\
-Copyright 2002-2016 Tiago Dionizio, Doug Currie\"");
-asm(".include \"libc/disclaimer.inc\"");
+__notice(lsqlite3_notice, "\
+lsqlite3 (MIT License)\n\
+Copyright 2002-2016 Tiago Dionizio, Doug Currie");

 // LOCAL CHANGES
 //