diff --git a/linker/linker_main.cpp b/linker/linker_main.cpp
index 69355d3..f65f82d 100644
--- a/linker/linker_main.cpp
+++ b/linker/linker_main.cpp
@@ -46,6 +46,7 @@
 #include "linker_tls.h"
 #include "linker_utils.h"
 
+#include "platform/bionic/macros.h"
 #include "private/KernelArgumentBlock.h"
 #include "private/bionic_call_ifunc_resolver.h"
 #include "private/bionic_globals.h"
@@ -367,6 +368,8 @@
   init_link_map_head(*solinker);
 
 #if defined(__aarch64__)
+  __libc_init_mte(somain->memtag_dynamic_entries(), somain->phdr, somain->phnum, somain->load_bias);
+
   if (exe_to_load == nullptr) {
     // Kernel does not add PROT_BTI to executable pages of the loaded ELF.
     // Apply appropriate protections here if it is needed.
@@ -467,7 +470,6 @@
 #if defined(__aarch64__)
   // This has to happen after the find_libraries, which will have collected any possible
   // libraries that request memtag_stack in the dynamic section.
-  __libc_init_mte(somain->memtag_dynamic_entries(), somain->phdr, somain->phnum, somain->load_bias);
   __libc_init_mte_stack(args.argv);
 #endif
 
@@ -627,8 +629,13 @@
     // Apply RELR relocations first so that the GOT is initialized for ifunc
     // resolvers.
     if (relr && relrsz) {
+      // Nothing has tagged the memtag globals here, so it is pointless either
+      // way to handle them, the tags will be zero anyway.
+      // That is moot though, because the linker does not use memtag_globals
+      // in the first place.
       relocate_relr(reinterpret_cast<ElfW(Relr*)>(ehdr + relr),
-                    reinterpret_cast<ElfW(Relr*)>(ehdr + relr + relrsz), ehdr);
+                    reinterpret_cast<ElfW(Relr*)>(ehdr + relr + relrsz), ehdr,
+                    /*has_memtag_globals=*/ false);
     }
     if (pltrel && pltrelsz) {
       call_ifunc_resolvers_for_section(reinterpret_cast<RelType*>(ehdr + pltrel),
@@ -648,6 +655,16 @@
   }
 }
 
+// Remapping MTE globals segments happens before the linker relocates itself, and so can't use
+// memcpy() from string.h. This function is compiled with -ffreestanding.
+void linker_memcpy(void* dst, const void* src, size_t n) {
+  char* dst_bytes = reinterpret_cast<char*>(dst);
+  const char* src_bytes = reinterpret_cast<const char*>(src);
+  for (size_t i = 0; i < n; ++i) {
+    dst_bytes[i] = src_bytes[i];
+  }
+}
+
 // Detect an attempt to run the linker on itself. e.g.:
 //   /system/bin/linker64 /system/bin/linker64
 // Use priority-1 to run this constructor before other constructors.