[PATCH] [AArch64] Support AArch64 MTE memory tag dumps in core files

[Luis Machado]

v3:

- Update the segment name from PT_ARM_MEMTAG_MTE to PT_AARCH64_MEMTAG_MTE
  and change its value to PT_LOPROC + 0x2, as PT_LOPROC + 0x1 was taken.

v2:

- Drop arch-specific fields from the generic section data structure.
- Use the rawsize field of the section structure to store the original
  memory range of the memory-tagged area.
- Implement the bfd_elf_modify_headers for aarch64, to adjust the values
  for the memory tag segment according to the Linux Kernel's format.

--

The Linux kernel can dump memory tag segments to a core file, one segment
per mapped range. The format and documentation can be found in the Linux
kernel tree [1].

The following patch adjusts bfd and binutils so they can handle this new
segment type and display it accordingly. It also adds code required so GDB
can properly read/dump core file data containing memory tags.

Upon reading, each segment that contains memory tags gets mapped to a
section named "memtag". These sections will be used by GDB to lookup the tag
data. There can be multiple such sections with the same name, and they are not
numbered to simplify GDB's handling and lookup.

There is another patch for GDB that enables both reading
and dumping of memory tag segments.

Tested on aarch64-linux Ubuntu 20.04.

[1] Documentation/arm64/memory-tagging-extension.rst (Core Dump Support)
---
 bfd/elfnn-aarch64.c   | 89 +++++++++++++++++++++++++++++++++++++++++++
 binutils/readelf.c    |  1 +
 include/elf/aarch64.h |  3 ++
 3 files changed, 93 insertions(+)

diff --git a/bfd/elfnn-aarch64.c b/bfd/elfnn-aarch64.c
index 4926bab9cf2..3eee97b7ada 100644
--- a/bfd/elfnn-aarch64.c
+++ b/bfd/elfnn-aarch64.c
@@ -8158,6 +8158,89 @@ elfNN_aarch64_section_from_shdr (bfd *abfd,
   return true;
 }
 
+/* Process any AArch64-specific program segment types.  */
+
+static bool
+elfNN_aarch64_section_from_phdr (bfd *abfd ATTRIBUTE_UNUSED,
+				 Elf_Internal_Phdr *hdr,
+				 int hdr_index ATTRIBUTE_UNUSED,
+				 const char *name ATTRIBUTE_UNUSED)
+{
+  /* Right now we only handle the PT_AARCH64_MEMTAG_MTE segment type.  */
+  if (hdr == NULL || hdr->p_type != PT_AARCH64_MEMTAG_MTE)
+    return false;
+
+  if (hdr->p_filesz > 0)
+    {
+      /* Sections created from memory tag p_type's are always named
+	 "memtag".  This makes it easier for tools (for example, GDB)
+	 to find them.  */
+      asection *newsect = bfd_make_section_anyway (abfd, "memtag");
+
+      if (newsect == NULL)
+	return false;
+
+      unsigned int opb = bfd_octets_per_byte (abfd, NULL);
+
+      /* p_vaddr holds the original start address of the tagged memory
+	 range.  */
+      newsect->vma = hdr->p_vaddr / opb;
+
+      /* p_filesz holds the storage size of the packed tags.  */
+      newsect->size = hdr->p_filesz;
+      newsect->filepos = hdr->p_offset;
+
+      /* p_memsz holds the size of the memory range that contains tags.  The
+	 section's rawsize field is reused for this purpose.  */
+      newsect->rawsize = hdr->p_memsz;
+
+      /* Make sure the section's flags has SEC_HAS_CONTENTS set, otherwise
+	 BFD will return all zeroes when attempting to get contents from this
+	 section.  */
+      newsect->flags |= SEC_HAS_CONTENTS;
+    }
+
+  return true;
+}
+
+/* Implements the bfd_elf_modify_headers hook for aarch64.  */
+
+static bool
+elfNN_aarch64_modify_headers (bfd *abfd,
+			      struct bfd_link_info *info)
+{
+  struct elf_segment_map *m;
+  unsigned int segment_count = 0;
+  Elf_Internal_Phdr *p;
+
+  for (m = elf_seg_map (abfd); m != NULL; m = m->next, segment_count++)
+    {
+      /* We are only interested in the memory tag segment that will be dumped
+	 to a core file.  If we have no memory tags or this isn't a core file we
+	 are dealing with, just skip this segment.  */
+      if (m->p_type != PT_AARCH64_MEMTAG_MTE
+	  || bfd_get_format (abfd) != bfd_core)
+	continue;
+
+      /* For memory tag segments in core files, the size of the file contents
+	 is smaller than the size of the memory range.  Adjust the memory size
+	 accordingly.  The real memory size is held in the section's rawsize
+	 field.  */
+      if (m->count > 0)
+	{
+	  p = elf_tdata (abfd)->phdr;
+	  p += m->idx;
+	  p->p_memsz = m->sections[0]->rawsize;
+	  p->p_flags = 0;
+	  p->p_paddr = 0;
+	  p->p_align = 0;
+	}
+    }
+
+  /* Give the generic code a chance to handle the headers.  */
+  return _bfd_elf_modify_headers (abfd, info);
+}
+
 /* A structure used to record a list of sections, independently
    of the next and prev fields in the asection structure.  */
 typedef struct section_list
@@ -10059,6 +10142,12 @@ const struct elf_size_info elfNN_aarch64_size_info =
 #define elf_backend_section_from_shdr		\
   elfNN_aarch64_section_from_shdr
 
+#define elf_backend_section_from_phdr		\
+  elfNN_aarch64_section_from_phdr
+
+#define elf_backend_modify_headers		\
+  elfNN_aarch64_modify_headers
+
 #define elf_backend_size_dynamic_sections	\
   elfNN_aarch64_size_dynamic_sections
 
diff --git a/binutils/readelf.c b/binutils/readelf.c
index d45e0920788..ea20479daef 100644
--- a/binutils/readelf.c
+++ b/binutils/readelf.c
@@ -4499,6 +4499,7 @@ get_aarch64_segment_type (unsigned long type)
   switch (type)
     {
     case PT_AARCH64_ARCHEXT:  return "AARCH64_ARCHEXT";
+    case PT_AARCH64_MEMTAG_MTE:	return "AARCH64_MEMTAG_MTE";
     default:                  return NULL;
     }
 }
diff --git a/include/elf/aarch64.h b/include/elf/aarch64.h
index 703a62baeca..e368ff2dffe 100644
--- a/include/elf/aarch64.h
+++ b/include/elf/aarch64.h
@@ -27,6 +27,9 @@
 /* Processor specific program header types.  */
 #define PT_AARCH64_ARCHEXT	(PT_LOPROC + 0)
 
+/* MTE memory tag segment type.  */
+#define PT_AARCH64_MEMTAG_MTE     (PT_LOPROC + 0x2)
+
 /* Additional section types.  */
 #define SHT_AARCH64_ATTRIBUTES	0x70000003  /* Section holds attributes.  */
 
-- 
2.25.1