[lld/ELF][x86-64] Place large executable sections at the edges of binary (#70358)

So that when mixing small and large text, large text stays out of the
way of the rest of the binary.

Place large RX sections at the beginning rather than at the end so that
with `--no-rosegment`, the large text and rodata share a single PT_LOAD
segment. Place large RWX sections at the end to keep writable and
readonly sections separate.

Clang started emitting the large section flag for `.ltext` sections in
#73037.

Author: aeubanks

lld/ELF/Writer.cpp

@@ -653,15 +653,17 @@ enum RankFlags {
   RF_NOT_ADDR_SET = 1 << 27,
   RF_NOT_ALLOC = 1 << 26,
   RF_PARTITION = 1 << 18, // Partition number (8 bits)
+  RF_LARGE_EXEC_WRITE = 1 << 16,
   RF_LARGE_ALT = 1 << 15,
   RF_WRITE = 1 << 14,
   RF_EXEC_WRITE = 1 << 13,
   RF_EXEC = 1 << 12,
   RF_RODATA = 1 << 11,
-  RF_LARGE = 1 << 10,
-  RF_NOT_RELRO = 1 << 9,
-  RF_NOT_TLS = 1 << 8,
-  RF_BSS = 1 << 7,
+  RF_LARGE_EXEC = 1 << 10,
+  RF_LARGE = 1 << 9,
+  RF_NOT_RELRO = 1 << 8,
+  RF_NOT_TLS = 1 << 7,
+  RF_BSS = 1 << 6,
 };
 
 unsigned elf::getSectionRank(Ctx &ctx, OutputSection &osec) {
@@ -691,14 +693,15 @@ unsigned elf::getSectionRank(Ctx &ctx, OutputSection &osec) {
   // places.
   bool isExec = osec.flags & SHF_EXECINSTR;
   bool isWrite = osec.flags & SHF_WRITE;
+  bool isLarge = osec.flags & SHF_X86_64_LARGE && ctx.arg.emachine == EM_X86_64;
 
   if (!isWrite && !isExec) {
     // Among PROGBITS sections, place .lrodata further from .text.
     // For -z lrodata-after-bss, place .lrodata after .lbss like GNU ld. This
     // layout has one extra PT_LOAD, but alleviates relocation overflow
     // pressure for absolute relocations referencing small data from -fno-pic
     // relocatable files.
-    if (osec.flags & SHF_X86_64_LARGE && ctx.arg.emachine == EM_X86_64)
+    if (isLarge)
       rank |= ctx.arg.zLrodataAfterBss ? RF_LARGE_ALT : 0;
     else
       rank |= ctx.arg.zLrodataAfterBss ? 0 : RF_LARGE;
@@ -722,7 +725,13 @@ unsigned elf::getSectionRank(Ctx &ctx, OutputSection &osec) {
     else
       rank |= RF_RODATA;
   } else if (isExec) {
-    rank |= isWrite ? RF_EXEC_WRITE : RF_EXEC;
+    // Place readonly .ltext before .lrodata and writable .ltext after .lbss to
+    // keep writable and readonly segments separate.
+    if (isLarge) {
+      rank |= isWrite ? RF_LARGE_EXEC_WRITE : RF_LARGE_EXEC;
+    } else {
+      rank |= isWrite ? RF_EXEC_WRITE : RF_EXEC;
+    }
   } else {
     rank |= RF_WRITE;
     // The TLS initialization block needs to be a single contiguous block. Place
@@ -737,7 +746,7 @@ unsigned elf::getSectionRank(Ctx &ctx, OutputSection &osec) {
     // alleviates relocation overflow pressure.
     // For -z lrodata-after-bss, place .lbss/.lrodata/.ldata after .bss.
     // .bss/.lbss being adjacent reuses the NOBITS size optimization.
-    if (osec.flags & SHF_X86_64_LARGE && ctx.arg.emachine == EM_X86_64) {
+    if (isLarge) {
       rank |= ctx.arg.zLrodataAfterBss
                   ? (osec.type == SHT_NOBITS ? 1 : RF_LARGE_ALT)
                   : RF_LARGE;

lld/test/ELF/x86-64-section-layout.s

@@ -18,6 +18,10 @@
 # RUN: ld.lld --section-start=.note=0x200300 a.o -z lrodata-after-bss -o a3
 # RUN: llvm-readelf -S -l -sX a3 | FileCheck %s --check-prefix=CHECK3
 
+# RUN: llvm-mc -filetype=obj -triple=x86_64 c.s -o c.o
+# RUN: ld.lld c.o -o c
+# RUN: llvm-readelf -S -l c | FileCheck %s --check-prefix=CHECK4
+
 # CHECK:       Name              Type            Address          Off    Size   ES Flg Lk Inf Al
 # CHECK-NEXT:                    NULL            0000000000000000 000000 000000 00      0   0  0
 # CHECK-NEXT:  .note             NOTE            0000000000200300 000300 000001 00   A  0   0  1
@@ -116,6 +120,18 @@
 # CHECK3-NEXT:  0000000000203307     0 NOTYPE  GLOBAL DEFAULT [[#]] (.data)   _edata
 # CHECK3-NEXT:  0000000000207d0d     0 NOTYPE  GLOBAL DEFAULT [[#]] (.ldata2) _end
 
+# CHECK4:      .note      NOTE
+# CHECK4-NEXT: .ltext     PROGBITS
+# CHECK4-NEXT: .lrodata   PROGBITS
+# CHECK4-NEXT: .rodata    PROGBITS
+# CHECK4-NEXT: .text      PROGBITS
+# CHECK4-NEXT: .data      PROGBITS
+# CHECK4-NEXT: .bss       NOBITS
+# CHECK4-NEXT: .ldata     PROGBITS
+# CHECK4-NEXT: .lbss      NOBITS
+# CHECK4-NEXT: .ltext_w   PROGBITS
+# CHECK4-NEXT: .comment   PROGBITS
+
 #--- a.s
 .globl _start, _etext, _edata, _end
 _start:
@@ -155,3 +171,28 @@ SECTIONS {
   .ldata2 : {}
   .lbss : { *(.lbss .lbss.*) }
 }
+
+#--- c.s
+## Test .ltext layout
+.section .ltext,"axl",@progbits
+.globl f
+f:
+  ret
+
+.section .ltext_w,"awxl",@progbits
+.globl g
+g:
+  ret
+
+.section .text,"ax",@progbits
+.globl h
+h:
+  ret
+
+.section .note,"a",@note; .space 1
+.section .rodata,"a",@progbits; .space 1
+.section .data,"aw",@progbits; .space 1
+.section .bss,"aw",@nobits; .space 1
+.section .lrodata,"al"; .space 1
+.section .ldata,"awl"; .space 1
+.section .lbss,"awl",@nobits; .space 1