[ELF] Set the .dynamic address to the first entry of .got

glibc for ARM64 prior to August 2021 arguably wrongly assumes that
the GOT first entry contains the address of the .dynamic section.
That's been removed in the following commit, but without satisfying
that assumption, mold can't create working static position-independent
execuables for ARM64.

https://sourceware.org/git/?p=glibc.git;h=43d06ed218fc8be58987bdfd00e21e5720f0b862

elf/mold.h

@@ -484,6 +484,7 @@ class GotSection : public Chunk<E> {
     this->shdr.sh_type = SHT_PROGBITS;
     this->shdr.sh_flags = SHF_ALLOC | SHF_WRITE;
     this->shdr.sh_addralign = sizeof(Word<E>);
+    this->shdr.sh_size = sizeof(Word<E>);
   }
 
   void add_got_symbol(Context<E> &ctx, Symbol<E> *sym);
@@ -495,7 +496,6 @@ class GotSection : public Chunk<E> {
   u64 get_tlsld_addr(Context<E> &ctx) const;
   bool has_tlsld(Context<E> &ctx) const { return tlsld_idx != -1; }
   i64 get_reldyn_size(Context<E> &ctx) const;
-  void update_shdr(Context<E> &ctx) override;
   void copy_buf(Context<E> &ctx) override;
 
   void compute_symtab_size(Context<E> &ctx) override;

elf/output-chunks.cc

@@ -1259,20 +1259,21 @@ std::vector<GotEntry<E>> GotSection<E>::get_entries(Context<E> &ctx) const {
   return entries;
 }
 
-template <typename E>
-void GotSection<E>::update_shdr(Context<E> &ctx) {
-  // We always create a .got so that _GLOBAL_OFFSET_TABLE_ refers
-  // a GOT section.
-  if (this->shdr.sh_size == 0)
-    this->shdr.sh_size = sizeof(Word<E>);
-}
-
 // Fill .got and .rel.dyn.
 template <typename E>
 void GotSection<E>::copy_buf(Context<E> &ctx) {
   Word<E> *buf = (Word<E> *)(ctx.buf + this->shdr.sh_offset);
   memset(buf, 0, this->shdr.sh_size);
 
+  // glibc before 2021-08 assumes GOT[0] refers _DYNAMIC. We need to
+  // satisfy the assumption until older glibc completely retires.
+  // Since it is only required for static PIE on ARM64, we do this only
+  // for that case, so that no new code would accidentally depend on it.
+  //
+  // https://sourceware.org/git/?p=glibc.git;h=43d06ed218fc8be58987bdfd00e21e5720f0b862
+  if (std::is_same_v<E, ARM64> && ctx.arg.is_static && ctx.arg.pie && ctx.dynamic)
+    buf[0] = ctx.dynamic->shdr.sh_addr;
+
   ElfRel<E> *rel = (ElfRel<E> *)(ctx.buf + ctx.reldyn->shdr.sh_offset);
 
   for (GotEntry<E> &ent : get_entries(ctx)) {

test/elf/ifunc-static-pie.sh

@@ -3,7 +3,6 @@
 
 test_cflags -static-pie || skip
 supports_ifunc || skip
-[ $MACHINE = aarch64 ] && skip
 
 cat <<EOF | $CC -o $t/a.o -c -xc - -fPIC
 #include <stdio.h>

test/elf/static-pie.sh

@@ -2,7 +2,6 @@
 . $(dirname $0)/common.inc
 
 test_cflags -static-pie || skip
-[ $MACHINE = aarch64 ] && skip
 
 cat <<EOF | $CC -o $t/a.o -c -xc - -fPIE
 #include <stdio.h>