Skip to content

X64: Use Struct for ELF Ehdr and Phdr headers #1694

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 11 commits into from
Apr 14, 2023
Merged

X64: Use Struct for ELF Ehdr and Phdr headers #1694

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
b766d8a
X64: Define elf ehdr, phdr structs for x86 and x64
ubaidsk Apr 10, 2023
dab6068
X64: Define header constructing functions
ubaidsk Apr 10, 2023
bd19300
WASM: X64: Emit alignment bytes before segment end
ubaidsk Apr 13, 2023
a05b71c
WASM_X64: Define and use common get_seg_header()
ubaidsk Apr 13, 2023
150a9c8
WASM_X64: Support elf headers, footer in asm text
ubaidsk Apr 13, 2023
9ffc308
X86Assembler: Isolate get_asm() and save_bin() for 32 and 64bit
ubaidsk Apr 13, 2023
06ed809
WASM: Fix neg val assign to global_var_idx
ubaidsk Apr 12, 2023
f12e42c
WASM_X64: Remove unused functions
ubaidsk Apr 13, 2023
14b1197
WASM_X64: Create and return only header binary
ubaidsk Apr 14, 2023
0630d23
WASM_X64: Pass only required params to create_elf64_x86_header()
ubaidsk Apr 14, 2023
1a99f9a
WASM_X64: Assign mod value directly as constant
ubaidsk Apr 14, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions src/libasr/codegen/asr_to_wasm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@
#include <chrono>
#include <iomanip>
#include <fstream>
#include <climits>

#include <libasr/asr.h>
#include <libasr/exception.h>
Expand Down Expand Up @@ -526,7 +527,7 @@ class ASRToWASMVisitor : public ASR::BaseVisitor<ASRToWASMVisitor> {

using namespace wasm;
int kind = ASRUtils::extract_kind_from_ttype_t(v->m_type);
uint32_t global_var_idx = -1;
uint32_t global_var_idx = UINT_MAX;
switch (v->m_type->type){
case ASR::ttypeType::Integer: {
uint64_t init_val = 0;
Expand Down Expand Up @@ -583,7 +584,7 @@ class ASRToWASMVisitor : public ASR::BaseVisitor<ASRToWASMVisitor> {
global_var_idx = m_wa.declare_global_var(i32, 0);
}
}
LCOMPILERS_ASSERT(global_var_idx >= 0);
LCOMPILERS_ASSERT(global_var_idx < UINT_MAX);
m_global_var_idx_map[get_hash((ASR::asr_t *)v)] = global_var_idx;
}

Expand Down
11 changes: 3 additions & 8 deletions src/libasr/codegen/wasm_to_x64.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -581,8 +581,6 @@ class X64Visitor : public WASMDecoder<X64Visitor>,
void visit_F32Sqrt() { visit_F64Sqrt(); }

void gen_x64_bytes() {
emit_elf64_header(m_a);

// declare compile-time strings
std::string base_memory = " "; /* in wasm backend, memory starts after 4 bytes*/
for (uint32_t i = 0; i < data_segments.size(); i++) {
Expand All @@ -592,7 +590,6 @@ class X64Visitor : public WASMDecoder<X64Visitor>,

NO_OF_IMPORTS = imports.size();

m_a.align_by_byte(0x1000);
m_a.add_label("text_segment_start");
for (uint32_t idx = 0; idx < type_indices.size(); idx++) {
m_a.add_label(exports[idx + 1].name);
Expand All @@ -616,9 +613,9 @@ class X64Visitor : public WASMDecoder<X64Visitor>,
emit_double_const(m_a, d.first, d.second);
}

m_a.align_by_byte(0x1000);
m_a.add_label("text_segment_end");

m_a.align_by_byte(0x1000);
m_a.add_label("data_segment_start");
for (auto &s : label_to_str) {
emit_data_string(m_a, s.first, s.second);
Expand Down Expand Up @@ -647,8 +644,6 @@ class X64Visitor : public WASMDecoder<X64Visitor>,
}
}
m_a.add_label("data_segment_end");

emit_elf64_footer(m_a);
}
};

Expand Down Expand Up @@ -700,15 +695,15 @@ Result<int> wasm_to_x64(Vec<uint8_t> &wasm_bytes, Allocator &al,

{
auto t1 = std::chrono::high_resolution_clock::now();
m_a.save_binary(filename);
m_a.save_binary64(filename);
auto t2 = std::chrono::high_resolution_clock::now();
time_save =
std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1)
.count();
}

//! Helpful for debugging
// std::cout << x64_visitor.m_a.get_asm() << std::endl;
// std::cout << x64_visitor.m_a.get_asm64() << std::endl;

if (time_report) {
std::cout << "Codegen Time report:" << std::endl;
Expand Down
261 changes: 179 additions & 82 deletions src/libasr/codegen/x86_assembler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,25 @@

namespace LCompilers {

void X86Assembler::save_binary64(const std::string &filename) {
Vec<uint8_t> header = create_elf64_x86_header(
m_al, origin(), get_defined_symbol("_start").value,
compute_seg_size("text_segment_start", "text_segment_end"),
compute_seg_size("data_segment_start", "data_segment_end"));
{
std::ofstream out;
out.open(filename);
out.write((const char*) header.p, header.size());
out.write((const char*) m_code.p, m_code.size());
}
#ifdef LFORTRAN_LINUX
int mod = 0755;
if (chmod(filename.c_str(),mod) < 0) {
throw AssemblerError("chmod failed");
}
#endif
}

void X86Assembler::save_binary(const std::string &filename) {
{
std::ofstream out;
Expand All @@ -25,6 +44,36 @@ void X86Assembler::save_binary(const std::string &filename) {
#endif
}

// ELF header structure for 32-bit
struct Elf32_Ehdr {
uint8_t ident[16];
uint16_t type;
uint16_t machine;
uint32_t version;
uint32_t entry;
uint32_t phoff;
uint32_t shoff;
uint32_t flags;
uint16_t ehsize;
uint16_t phentsize;
uint16_t phnum;
uint16_t shentsize;
uint16_t shnum;
uint16_t shstrndx;
};

// Program header structure for 32-bit
struct Elf32_Phdr {
uint32_t type;
uint32_t offset;
uint32_t vaddr;
uint32_t paddr;
uint32_t filesz;
uint32_t memsz;
uint32_t flags;
uint32_t align;
};

void emit_elf32_header(X86Assembler &a, uint32_t p_flags) {
/* Elf32_Ehdr */
a.add_label("ehdr");
Expand Down Expand Up @@ -283,94 +332,142 @@ void emit_print_float(X86Assembler &a, const std::string &name) {

/************************* 64-bit functions **************************/

void emit_elf64_header(X86Assembler &a) {
/* Elf64_Ehdr */
a.add_label("ehdr");
// e_ident
a.asm_db_imm8(0x7F);
a.asm_db_imm8('E');
a.asm_db_imm8('L');
a.asm_db_imm8('F');
a.asm_db_imm8(2);
a.asm_db_imm8(1);
a.asm_db_imm8(1);
a.asm_db_imm8(0);
// ELF header structure for 64-bit
struct Elf64_Ehdr {
uint8_t ident[16];
uint16_t type;
uint16_t machine;
uint32_t version;
uint64_t entry;
uint64_t phoff;
uint64_t shoff;
uint32_t flags;
uint16_t ehsize;
uint16_t phentsize;
uint16_t phnum;
uint16_t shentsize;
uint16_t shnum;
uint16_t shstrndx;
};

// Program header structure for 64-bit
struct Elf64_Phdr {
uint32_t type;
uint32_t flags;
uint64_t offset;
uint64_t vaddr;
uint64_t paddr;
uint64_t filesz;
uint64_t memsz;
uint64_t align;
};

Elf64_Ehdr get_elf_header(uint64_t asm_entry) {
Elf64_Ehdr e;
e.ident[0] = 0x7f; // magic number
e.ident[1] = 'E';
e.ident[2] = 'L';
e.ident[3] = 'F';
e.ident[4] = 2; // file class (64-bit)
e.ident[5] = 1; // data encoding (little endian)
e.ident[6] = 1; // ELF version
e.ident[7] = 0; // padding
e.ident[8] = 0;
e.ident[9] = 0;
e.ident[10] = 0;
e.ident[11] = 0;
e.ident[12] = 0;
e.ident[13] = 0;
e.ident[14] = 0;
e.ident[15] = 0;
e.type = 2;
e.machine = 0x3e;
e.version = 1;
e.entry = asm_entry;
e.phoff = sizeof(Elf64_Ehdr);
e.shoff = 0;
e.flags = 0;
e.ehsize = sizeof(Elf64_Ehdr);
e.phentsize = sizeof(Elf64_Phdr);
e.phnum = 3;
e.shentsize = 0;
e.shnum = 0;
e.shstrndx = 0;
return e;
}

a.asm_db_imm8(0);
a.asm_db_imm8(0);
a.asm_db_imm8(0);
a.asm_db_imm8(0);
Elf64_Phdr get_seg_header(uint32_t flags, uint64_t origin_addr,
uint64_t seg_size, uint64_t prev_seg_offset, uint64_t prev_seg_size) {
Elf64_Phdr p;
p.type = 1;
p.flags = flags;
p.offset = prev_seg_offset + prev_seg_size;
p.vaddr = origin_addr + p.offset;
p.paddr = p.vaddr;
p.filesz = seg_size;
p.memsz = p.filesz;
p.align = 0x1000;
return p;
}

a.asm_db_imm8(0);
a.asm_db_imm8(0);
a.asm_db_imm8(0);
a.asm_db_imm8(0);
template <typename T>
void append_header_bytes(Allocator &al, T src, Vec<uint8_t> &des) {
char *byteArray = (char *)&src;
for (size_t i = 0; i < sizeof(src); i++) {
des.push_back(al, byteArray[i]);
}
}

a.asm_dw_imm16(2); // e_type
a.asm_dw_imm16(0x3e); // e_machine
a.asm_dd_imm32(1); // e_version
a.asm_dq_label("_start"); // e_entry
a.asm_dq_label("e_phoff"); // e_phoff
a.asm_dq_imm64(0); // e_shoff
a.asm_dd_imm32(0); // e_flags
a.asm_dw_label("ehdrsize"); // e_ehsize
a.asm_dw_label("phdrsize"); // e_phentsize
a.asm_dw_imm16(3); // e_phnum
a.asm_dw_imm16(0); // e_shentsize
a.asm_dw_imm16(0); // e_shnum
a.asm_dw_imm16(0); // e_shstrndx

/* Elf64_Phdr */
a.add_label("phdr");
a.asm_dd_imm32(1); // p_type
a.asm_dd_imm32(4); // p_flags (permission to read only)
a.asm_dq_imm64(0); // p_offset
a.asm_dq_imm64(a.origin()); // p_vaddr
a.asm_dq_imm64(a.origin()); // p_paddr
a.asm_dq_label("phdr_size"); // p_filesz
a.asm_dq_label("phdr_size"); // p_memsz
a.asm_dq_imm64(0x1000); // p_align

/* text_segment_phdr */
a.add_label("text_phdr");
a.asm_dd_imm32(1); // p_type
a.asm_dd_imm32(5); // p_flags (permission to read and execute)
a.asm_dq_label("text_segment_offset"); // p_offset
a.asm_dq_label("text_segment_start"); // p_vaddr
a.asm_dq_label("text_segment_start"); // p_paddr
a.asm_dq_label("text_segment_size"); // p_filesz
a.asm_dq_label("text_segment_size"); // p_memsz
a.asm_dq_imm64(0x1000); // p_align

/* data_segment_phdr */
a.add_label("data_phdr");
a.asm_dd_imm32(1); // p_type
a.asm_dd_imm32(6); // p_flags (permission to read and write)
a.asm_dq_label("data_segment_offset"); // p_offset
a.asm_dq_label("data_segment_start"); // p_vaddr
a.asm_dq_label("data_segment_start"); // p_paddr
a.asm_dq_label("data_segment_size"); // p_filesz
a.asm_dq_label("data_segment_size"); // p_memsz
a.asm_dq_imm64(0x1000); // p_align
}
void align_by_byte(Allocator &al, Vec<uint8_t> &code, uint64_t alignment) {
uint64_t code_size = code.size() ;
uint64_t padding_size = (alignment * ceil(code_size / (double)alignment)) - code_size;
for (size_t i = 0; i < padding_size; i++) {
code.push_back(al, 0);
}
}

void emit_elf64_footer(X86Assembler &a) {
a.add_var("ehdrsize", "ehdr", "phdr");
a.add_var("phdrsize", "phdr", "text_phdr");
a.add_var64("e_phoff", "ehdr", "phdr");
a.add_var64("phdr_size", "ehdr", "text_segment_start");
a.add_var64("text_segment_offset", "ehdr", "text_segment_start");
a.add_var64("text_segment_size", "text_segment_start", "text_segment_end");
a.add_var64("data_segment_offset", "ehdr", "data_segment_start");
a.add_var64("data_segment_size", "data_segment_start", "data_segment_end");
}
Vec<uint8_t> create_elf64_x86_header(Allocator &al, uint64_t origin, uint64_t entry,
uint64_t text_seg_size, uint64_t data_seg_size) {

void emit_exit_64(X86Assembler &a, std::string name, int exit_code) {
a.add_label(name);
// void exit(int status);
a.asm_mov_r64_imm64(LCompilers::X64Reg::rax, 60); // sys_exit
a.asm_mov_r64_imm64(LCompilers::X64Reg::rdi, exit_code); // exit code
a.asm_syscall(); // syscall
/*
The header segment is a segment which holds the elf and program headers.
Its size currently is
sizeof(Elf64_Ehdr) + 3 * sizeof(Elf64_Phdr)
that is, 64 + 3 * 56 = 232
Since, it is a segment, it needs to be aligned by boundary 0x1000
(we add temporary zero bytes as padding to accomplish this alignment)

Thus, the header segment size for us currently is 0x1000.

For now, we are hardcoding this size here.

TODO: Later compute this header segment size dynamically depending
on the different segments present
*/
const int HEADER_SEGMENT_SIZE = 0x1000;

// adjust/offset the origin address as per the extra bytes of HEADER_SEGMENT_SIZE
uint64_t origin_addr = origin - HEADER_SEGMENT_SIZE;

Elf64_Ehdr e = get_elf_header(entry);
Elf64_Phdr p_program = get_seg_header(4, origin_addr, HEADER_SEGMENT_SIZE, 0, 0);
Elf64_Phdr p_text_seg = get_seg_header(5, origin_addr, text_seg_size, p_program.offset, p_program.filesz);
Elf64_Phdr p_data_seg = get_seg_header(6, origin_addr, data_seg_size, p_text_seg.offset, p_text_seg.filesz);

Vec<uint8_t> header;
header.reserve(al, HEADER_SEGMENT_SIZE);

{
append_header_bytes(al, e, header);
append_header_bytes(al, p_program, header);
append_header_bytes(al, p_text_seg, header);
append_header_bytes(al, p_data_seg, header);

LCompilers::align_by_byte(al, header, 0x1000);
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this copying the (potentially big) binary code byte by byte?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, it is copying the binary code byte by byte. Another approach that we can do is to just return the ELF and Phdr headers in create_elf64_x86_binary() (we can rename it as create_elf64_x86_header()).

For example:

void X86Assembler::save_binary64(const std::string &filename) {
    Vec<uint8_t> header = create_elf64_x86_header(m_al, *this);
    {
        std::ofstream out;
        out.open(filename);
        out.write((const char*) header.p, header.size());
        out.write((const char*) m_code.p, m_code.size());
        out.close();
    }
#ifdef LFORTRAN_LINUX
    std::string mode = "0755";
    int mod = strtol(mode.c_str(), 0, 8);
    if (chmod(filename.c_str(),mod) < 0) {
        throw AssemblerError("chmod failed");
    }
#endif
}

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I currently updated it to return the header only.


return header;
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would keep this function outside of X86Assembler. Eventually once we add Arm64Assembler, we will figure out how to generalize this function to also work for ELF ARM binaries.

Also I would rename it to something like create_elf64_x86_binary and I would pass m_code as an argument.

I would make the function return a string.

The caller will then save the string to a file.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would make the function return a string.

I currently used a Vec<uint8_t>. I would like to know if it is fine or if I shall update it.


void emit_print_64(X86Assembler &a, const std::string &msg_label, uint64_t size)
Expand Down
Loading