Appendix B Errata of the first edition

This appendix lists the content errors of the first edition (the PDF built in June 2020) that were corrected in the second edition, so that readers of the old PDF can check their copy, and contributors can see which reported issues were closed. Typo-only reports are not listed individually. Numbers in parentheses are GitHub issue or pull-request numbers in the tuhdo/os01 repository.

B.1 Preface and chapter 1

• The “What you will learn” list promised “bare metal x86_64”; everything in the book is 32-bit. The second edition says so and explains 64-bit mode in chapter 15.
• Chapter 1 claimed that the book’s website hosts copies of the Intel manuals and the System V ABI documents; it never did. The text now says where to search for them (#37).
• A citation of Kovitz was placed after the sentence’s full stop (#97).

B.2 Chapter 2

• The example machine code said that 10100000 means “add” and 000000101 (nine bits) means “halt”. The second edition uses two real one-byte x86 opcodes, 00000000 (add) and 11110100 (hlt).
• Example 2.5 said “domain-agnostic language” where the paragraph describes a domain-specific one.
• Figure 2.3.4, the four-chip OR gate layout, labeled pin 1B of the first 74HC00 as A2; it is A1 like on the other three chips (#115, PR #158).
• The three links to the 74HC00 datasheet were dead; they now point to the manufacturer’s current page (#113).
• A margin note recommending Ben Eater’s video on how semiconductors work had been lost from the text (#155).
• Grammar in several sentences (#183, #184, PR #185, PR #174, PR #159).

B.3 Chapter 3

• The photo captioned “Raspberry Pi B+ Rev 1.2” is a Raspberry Pi 2 Model B.
• “Field Gate Programmable Array” for FPGA.
• The Q35 chipset was described as “relatively modern”; the second edition dates it (2007) and explains what changed since.
• “mount” for “amount” (#129).

B.4 Chapter 4

• Example 4.5.7 used jmp [eax*4 + eax + esi], which nasm rejects (invalid effective address): a scaled index needs a base and one index, so the example is now jmp [eax*4 + esi] (#108, #148).
• Exercise 4.3.1 referred to a section number of the Intel manual that no longer exists; it now names the section by title (#177).
• Exercise 4.9.1 referred to the wrong manual; it is the Optimization Reference Manual (#112).
• The listing for add(1,1) did not match the source code shown (#114, #139).
• Figure 4.9.1, the shl/sal illustration: the row labeled “after 10-bit shift” showed an 11-bit shift (#156).
• Figure 4.9.6, the stack frame: the return address was drawn below the saved ebp; it sits above it, at a higher address (#152).
• In the conditional example, the registers holding i and j were swapped in the explanation (i is in edx, j in eax) (#179).
• The byte table of the jmp encoding example had a spurious tenth byte.
• The far-jump text described the in-memory order of an m16:32 pointer wrongly; it is the 32-bit offset first (34 12 00 00), then the 16-bit segment (78 56).
• The SIB table’s third column was labeled R/M; it is Index.
• “Produce Calls” for “Procedure Calls”; ESP/SP/AHMM4 and MM/XMM1 garbled register names; a margin note on cat /proc/cpuinfo was clipped out of the page (#109); other wording issues (#110, #111, #121, #166).
• The stack offsets in “Examine compiled code” were inconsistent (two variables at the same offset). gcc 14 allocates locals in declaration order, and the listings and prose now agree.

B.5 Chapter 5

• The ELF entry point was said to be main; it is _start, which calls the C runtime that calls main.
• The ELF header’s Flags field was described as the EFLAGS register; it is a processor-specific flags word, and no flags are defined for x86.
• “Size of program headers” is 56 bytes in a 64-bit file, not 64.
• Exercises 5.3 and 5.4 had the Link and Info fields of .rel.text swapped.
• Example 5.18 placed both arrays in .preinit_array while the text discusses .init_array; the source now uses .init_array for init_arr.
• A readelf -x 25 command dumped a section that was not .data in the listing shown.
• “HIREVERSE” for “HIPROC”/“HIUSER” range names; a hello2.c file compiled as hello.

B.6 Chapter 6

• The “Single stepping” section did not explain the mechanism; it now describes the trap flag and the debug exception (#145).
• The up command’s explanation said frame 1 was in function b at line 3; the session shows it in c at line 2.
• The until example’s line numbers did not match its own listing, and add1000 was called with arguments it does not take.
• The breakpoint figure labeled the patched bytes as sub esp,0x4; the instruction is sub esp,0xc.
• Examples were numbered out of order (6.9 before 6.8, 6.11 before 6.10, 6.25 before 6.24).
• The DWARF section and the DIE tree figure showed DWARF 4 output; the second edition shows DWARF 5, which current compilers emit by default.

B.7 Chapter 7

• A bootloader that prints a string would print nothing (#33): the flat binary is assembled at offset 0 and never sets DS, so mov si, msg points into the interrupt vector table once the BIOS has loaded the code at 0x7C00. The second edition adds the section “Two traps when writing a bootloader” (org 0x7c00, DS/ES/SS setup, and the leave-without-prologue trap reported in #154).
• The first flat Makefile (bootdisk: bootloader.o kernel.o) never worked; its targets are directories.
• The text said the bootloader reads 30 sectors; the code reads 2.
• The Makefile’s DISK_IMG path was wrong (#151), -T os.lds lacked a space (#163), the Make summary had $PATH for $(PATH), the wrong result for +=, a wrong $$ example, DISK_IMG=disk.img.o, “starts with .asm” for “ends with”, and the claim that Make starts with the all target rather than the first target (#176, #168, #167).
• The sample .gdbinit of chapter 7 contained two lines (symbol-file, b main) that belong to chapter 8 and made gdb abort the script.

B.8 Chapter 8

• The final linker script of the first edition is rejected by current versions of ld (PHDR segment not covered by LOAD segment). The section “A segment for the program header table” shows the error and the fix.
• Example 8.10 showed an implausible LOAD layout; it was replaced by a reproducible AT() demonstration.
• Figures no_FILEHDR and with_FILEHDR labeled the loaded segment .text .eh_frame with sizes of the 2016 toolchain.
• With gcc 14, uninitialized globals are no longer COMMON symbols unless -fcommon is given; the relocation example says so.

B.9 Code and tooling

• The committed sample code included build outputs and a stale debug.elf; all build products are now generated.
• Modern compilers default to position-independent executables; every listing in chapters 4 to 8 was regenerated with the flags of chapter 0, and the differences are explained where they matter (chapter 0, chapter 4 and chapter 5).
• set architecture i8086 in gdb no longer changes the disassembly when attached to QEMU’s gdb stub; chapters 7 and 9 explain how to read 16-bit code anyway.