feat(ram): Add flat ELF loading capabilities (#485)

This adds the ELF format as an additional pre-loadable file format.

Makefile

@@ -166,6 +166,12 @@ else
 SIM_LDFLAGS  += -lzstd
 endif
 
+# Elf image support
+IMAGE_ELF ?= 1
+ifeq ($(IMAGE_ELF),0)
+SIM_CXXFLAGS += -DNO_IMAGE_ELF
+endif
+
 # spike-dasm plugin
 WITH_SPIKE_DASM ?= 1
 ifeq ($(WITH_SPIKE_DASM),1)

src/test/csrc/common/elfloader.cpp

@@ -0,0 +1,124 @@
+/***************************************************************************************
+* Copyright (c) 2024 Axelera AI
+*
+* DiffTest is licensed under Mulan PSL v2.
+* You can use this software according to the terms and conditions of the Mulan PSL v2.
+* You may obtain a copy of Mulan PSL v2 at:
+*          http://license.coscl.org.cn/MulanPSL2
+*
+* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
+* EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
+* MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
+*
+* See the Mulan PSL v2 for more details.
+***************************************************************************************/
+
+#include "elfloader.h"
+
+void ElfBinary::load() {
+  assert(size >= sizeof(Elf64_Ehdr));
+  eh64 = (const Elf64_Ehdr *)raw;
+  assert(IS_ELF32(*eh64) || IS_ELF64(*eh64));
+
+  if (IS_ELF32(*eh64))
+    parse(data32);
+  else
+    parse(data64);
+}
+
+template <typename ehdr_t, typename phdr_t, typename shdr_t, typename sym_t>
+void ElfBinary::parse(ElfBinaryData<ehdr_t, phdr_t, shdr_t, sym_t> &data) {
+  data.eh = (const ehdr_t *)raw;
+  data.ph = (const phdr_t *)(raw + data.eh->e_phoff);
+  entry = data.eh->e_entry;
+  assert(size >= data.eh->e_phoff + data.eh->e_phnum * sizeof(*data.ph));
+  for (unsigned i = 0; i < data.eh->e_phnum; i++) {
+    if (data.ph[i].p_type == PT_LOAD && data.ph[i].p_memsz) {
+      if (data.ph[i].p_filesz) {
+        assert(size >= data.ph[i].p_offset + data.ph[i].p_filesz);
+        sections.push_back({
+          .data_src = (const uint8_t *)raw + data.ph[i].p_offset,
+          .data_dst = data.ph[i].p_paddr,
+          .data_len = data.ph[i].p_filesz,
+          .zero_dst = data.ph[i].p_paddr + data.ph[i].p_filesz,
+          .zero_len = data.ph[i].p_memsz - data.ph[i].p_filesz,
+        });
+      }
+    }
+  }
+}
+
+ElfBinaryFile::ElfBinaryFile(const char *filename) : filename(filename) {
+  int fd = open(filename, O_RDONLY);
+  struct stat s;
+  assert(fd != -1);
+  assert(fstat(fd, &s) >= 0);
+  size = s.st_size;
+
+  raw = (uint8_t *)mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
+  assert(raw != MAP_FAILED);
+  close(fd);
+
+  load();
+}
+
+ElfBinaryFile::~ElfBinaryFile() {
+  if (raw)
+    munmap((void *)raw, size);
+}
+
+bool isElfFile(const char *filename) {
+  int fd = -1;
+
+#ifdef NO_IMAGE_ELF
+  return false;
+#endif
+
+  fd = open(filename, O_RDONLY);
+  assert(fd);
+
+  uint8_t buf[4];
+
+  size_t sz = read(fd, buf, 4);
+  if (!IS_ELF(*((const Elf64_Ehdr *)buf))) {
+    close(fd);
+    return false;
+  }
+
+  close(fd);
+  return true;
+}
+
+long readFromElf(void *ptr, const char *file_name, long buf_size) {
+  auto elf_file = ElfBinaryFile(file_name);
+
+  if (elf_file.sections.size() < 1) {
+    printf("The requested elf '%s' contains zero sections\n", file_name);
+    return -1;
+  }
+
+  uint64_t len_written = 0;
+  auto base_addr = elf_file.sections[0].data_dst;
+
+  if (base_addr != PMEM_BASE) {
+    printf(
+        "The first address in the elf does not match the base of the physical memory. It is likely that execution "
+        "leads to unexpected behaviour.");
+  }
+
+  for (auto section: elf_file.sections) {
+    auto len = section.data_len + section.zero_len;
+    auto offset = section.data_dst - base_addr;
+
+    if (offset + len > buf_size) {
+      printf("The size (%ld bytes) of the section at address 0x%lx is larger than buf_size!\n", len, section.data_dst);
+      return -1;
+    }
+
+    printf("Loading %ld bytes at address 0x%lx at offset 0x%lx\n", len, section.data_dst, offset);
+    std::memset((uint8_t *)ptr + offset, 0, len);
+    std::memcpy((uint8_t *)ptr + offset, section.data_src, section.data_len);
+    len_written += len;
+  }
+  return len_written;
+}

src/test/csrc/common/elfloader.h

@@ -0,0 +1,189 @@
+/***************************************************************************************
+* Copyright (c) 2024 Axelera AI
+*
+* DiffTest is licensed under Mulan PSL v2.
+* You can use this software according to the terms and conditions of the Mulan PSL v2.
+* You may obtain a copy of Mulan PSL v2 at:
+*          http://license.coscl.org.cn/MulanPSL2
+*
+* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
+* EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
+* MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
+*
+* See the Mulan PSL v2 for more details.
+***************************************************************************************/
+
+#ifndef __ELFLOADER_H
+#define __ELFLOADER_H
+
+#include "config.h"
+#include <assert.h>
+#include <cstring>
+#include <fcntl.h>
+#include <iostream>
+#include <map>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <vector>
+
+#define IS_ELF(hdr) \
+  ((hdr).e_ident[0] == 0x7f && (hdr).e_ident[1] == 'E' && (hdr).e_ident[2] == 'L' && (hdr).e_ident[3] == 'F')
+
+#define IS_ELF32(hdr) (IS_ELF(hdr) && (hdr).e_ident[4] == 1)
+#define IS_ELF64(hdr) (IS_ELF(hdr) && (hdr).e_ident[4] == 2)
+
+#define PT_LOAD      1
+#define SHT_NOBITS   8
+#define SHT_PROGBITS 0x1
+#define SHT_GROUP    0x11
+
+typedef struct {
+  uint8_t e_ident[16];
+  uint16_t e_type;
+  uint16_t e_machine;
+  uint32_t e_version;
+  uint32_t e_entry;
+  uint32_t e_phoff;
+  uint32_t e_shoff;
+  uint32_t e_flags;
+  uint16_t e_ehsize;
+  uint16_t e_phentsize;
+  uint16_t e_phnum;
+  uint16_t e_shentsize;
+  uint16_t e_shnum;
+  uint16_t e_shstrndx;
+} Elf32_Ehdr;
+
+typedef struct {
+  uint32_t sh_name;
+  uint32_t sh_type;
+  uint32_t sh_flags;
+  uint32_t sh_addr;
+  uint32_t sh_offset;
+  uint32_t sh_size;
+  uint32_t sh_link;
+  uint32_t sh_info;
+  uint32_t sh_addralign;
+  uint32_t sh_entsize;
+} Elf32_Shdr;
+
+typedef struct {
+  uint32_t p_type;
+  uint32_t p_offset;
+  uint32_t p_vaddr;
+  uint32_t p_paddr;
+  uint32_t p_filesz;
+  uint32_t p_memsz;
+  uint32_t p_flags;
+  uint32_t p_align;
+} Elf32_Phdr;
+
+typedef struct {
+  uint32_t st_name;
+  uint32_t st_value;
+  uint32_t st_size;
+  uint8_t st_info;
+  uint8_t st_other;
+  uint16_t st_shndx;
+} Elf32_Sym;
+
+typedef struct {
+  uint8_t e_ident[16];
+  uint16_t e_type;
+  uint16_t e_machine;
+  uint32_t e_version;
+  uint64_t e_entry;
+  uint64_t e_phoff;
+  uint64_t e_shoff;
+  uint32_t e_flags;
+  uint16_t e_ehsize;
+  uint16_t e_phentsize;
+  uint16_t e_phnum;
+  uint16_t e_shentsize;
+  uint16_t e_shnum;
+  uint16_t e_shstrndx;
+} Elf64_Ehdr;
+
+typedef struct {
+  uint32_t sh_name;
+  uint32_t sh_type;
+  uint64_t sh_flags;
+  uint64_t sh_addr;
+  uint64_t sh_offset;
+  uint64_t sh_size;
+  uint32_t sh_link;
+  uint32_t sh_info;
+  uint64_t sh_addralign;
+  uint64_t sh_entsize;
+} Elf64_Shdr;
+
+typedef struct {
+  uint32_t p_type;
+  uint32_t p_flags;
+  uint64_t p_offset;
+  uint64_t p_vaddr;
+  uint64_t p_paddr;
+  uint64_t p_filesz;
+  uint64_t p_memsz;
+  uint64_t p_align;
+} Elf64_Phdr;
+
+typedef struct {
+  uint32_t st_name;
+  uint8_t st_info;
+  uint8_t st_other;
+  uint16_t st_shndx;
+  uint64_t st_value;
+  uint64_t st_size;
+} Elf64_Sym;
+
+template <typename ehdr_t, typename phdr_t, typename shdr_t, typename sym_t> struct ElfBinaryData {
+  const ehdr_t *eh = nullptr;
+  const phdr_t *ph = nullptr;
+};
+
+struct ElfSection {
+  const uint8_t *data_src;
+  uint64_t data_dst;
+  size_t data_len;
+  uint64_t zero_dst;
+  size_t zero_len;
+};
+
+struct ElfBinary {
+  const uint8_t *raw = nullptr;
+  size_t size = 0;
+
+  const Elf64_Ehdr *eh64 = nullptr;
+  ElfBinaryData<Elf32_Ehdr, Elf32_Phdr, Elf32_Shdr, Elf32_Sym> data32;
+  ElfBinaryData<Elf64_Ehdr, Elf64_Phdr, Elf64_Shdr, Elf64_Sym> data64;
+
+  uint64_t entry;
+  std::vector<ElfSection> sections;
+
+  void load();
+
+  template <typename ehdr_t, typename phdr_t, typename shdr_t, typename sym_t>
+  void parse(ElfBinaryData<ehdr_t, phdr_t, shdr_t, sym_t> &data);
+};
+
+struct ElfBinaryFile : public ElfBinary {
+  const std::string filename;
+
+  ElfBinaryFile(const char *filename);
+
+  ~ElfBinaryFile();
+};
+
+// Is the file at the given path an Elf file
+bool isElfFile(const char *filename);
+// load binary content at `file_name` into ptr. Returns the number of bytes
+// written.
+long readFromElf(void *ptr, const char *file_name, long buf_size);
+
+#endif // __ELFLOADER_H

src/test/csrc/common/ram.cpp

@@ -17,6 +17,7 @@
 #include "ram.h"
 #include "common.h"
 #include "compress.h"
+#include "elfloader.h"
 #include <iostream>
 #include <sys/mman.h>
 #ifdef CONFIG_DIFFTEST_PERFCNT
@@ -284,6 +285,10 @@ MmapMemory::MmapMemory(const char *image, uint64_t n_bytes) : SimMemory(n_bytes)
     Info("Zstd file detected and loading image from extracted zstd file\n");
     img_size = readFromZstd(ram, image, memory_size, LOAD_RAM);
     assert(img_size >= 0);
+  } else if (isElfFile(image)) {
+    Info("ELF file detected and loading image from extracted elf file\n");
+    img_size = readFromElf(ram, image, memory_size);
+    assert(img_size >= 0);
   } else {
     InputReader *reader = createInputReader(image);
     img_size = reader->read_all(ram, memory_size);