Day 25 evidence

Date: 2026-09-20

Source baseline

  • Implementation commit: df4c9944997eebf65c67cf40d68419d690622e31
  • Text kernel SHA-256: 1c6c5d3f5858f1c7e627d7762e596e8c8e757f23d15aad81c6f583f82a06d8f1
  • Graphics kernel SHA-256: a689373d1c66320fef602628c3214ae4cd82b7eeae2646b9b2544ed2e3dd5c3f
  • Graphics disk image SHA-256: b3bf46c1c682967ff47eed1eb3ac2491cb7f7ba5326cb464171368714fe2329d
  • Screenshot SHA-256: 5a9fb76ec28c3403f8cba750c1028bd7bccbb0f5d0ad56625fa720be2f79d53e

The implementation commit contains the separate graphics stage2 and kernel
image, BootInfo v2, bounded pixel writer, forced video failure image, and the
QEMU/GDB check. It leaves all historical text images on BootInfo v1.

Environment

  • QEMU 8.2.2 (Debian 1:8.2.2+ds-0ubuntu1.18)
  • SeaBIOS 1.16.3 package image
  • GDB 15.1
  • GCC 13.3.0 and GNU Binutils 2.42
  • NASM 2.16.01
  • Machine: pc-i440fx-7.2, TCG, qemu32, 1 CPU, 64 MiB

The unpacked toolchain under /tmp/build-an-os-tools is an execution
dependency of this environment and is not included in the attachment.

Specialized command

1
2
3
4
5
6
7
8
9
PATH=/tmp/build-an-os-tools/bin:$PATH \
LD_LIBRARY_PATH=/tmp/build-an-os-tools/root2/usr/lib/x86_64-linux-gnu:\
/tmp/build-an-os-tools/root2/lib/x86_64-linux-gnu \
QEMU_MODULE_DIR=/tmp/build-an-os-tools/root2/usr/lib/x86_64-linux-gnu/qemu \
make check-graphics \
  QEMU=/tmp/build-an-os-tools/bin/qemu-system-i386 \
  GDB=/tmp/build-an-os-tools/bin/gdb-multiarch \
  BIOS=/tmp/build-an-os-tools/root2/usr/share/seabios/bios-256k.bin \
  QEMU_DATA=/tmp/build-an-os-tools/root2/usr/share/qemu

Exit status: 0.

Runtime observations

The normal graphics image reports that stage2 selected Mode 13h, entered
protected mode with BootInfo v2, and reached the bounded drawing demo:

1
2
3
4
D25 VIDEO MODE 13h
D04 PM OK - BootInfo v2
D25 GRAPHICS base=0xa0000 320x200 pitch=320 bpp=8 pixels=185120 rejected=2 hash=4130a99d
GRAPHICS OK framebuffer reserved and bounds checked

The complete log is docs/day25-graphics-serial.txt, SHA-256
81be47ca6f5a2c53d2177dbced6d30e186674bad3db76e2314ba3bd885bbb404.
The pixel count includes the initial full-frame pattern, six rectangles, and
the four border passes. It is an operation count rather than a distinct-pixel
count. The FNV-1a hash covers exactly 64,000 framebuffer bytes.

GDB stopped at graphics_demo_done and read the kernel symbols, VGA memory,
and allocator bitmap:

1
2
STATE base=0xa0000 size=320x200 pitch=320 mode=0x13 bpp=8 pixels=185120 rejected=2 guard=0 hash=0x4130a99d
PASS: BootInfo v2, framebuffer pixels, bounds and allocator exclusion verified

The check asserts the four border corners, one inner yellow pixel, and a center
white pixel directly from guest address 0xA0000. It also checks that page
number 0xA0 is absent from the allocator’s eligible bitmap. Two calls at
(320, 0) and (0, 200) were rejected, and the byte immediately after the
64,000-byte frame retained its original value. The GDB record is
docs/day25-graphics-gdb.txt, SHA-256
5ffa77e02b744b2a65b7f8eb1add4e30c774d776993b605b5075aea9e6437e67.

QEMU screendump produced docs/day25-screen.png. Its PNG dimensions are
640x400 because QEMU doubles the legacy 320x200 scanout on export. The check
does not infer guest geometry from that host image: guest width, height, pitch,
and key bytes are verified separately through BootInfo and GDB.

Failure path

mbr-graphics-fail.img uses the same stage2 code with a build-time fault that
makes the mode readback fail. Stage2 restores BIOS mode 3, writes the diagnostic
to text VGA and COM1, and halts before protected-mode entry:

1
D25 VIDEO FAIL

The failure log contains neither D04 PM OK nor D25 GRAPHICS. This proves
that the injected setup failure is fail-closed rather than a normal graphics
boot followed by a warning. The record is
docs/day25-graphics-fail-serial.txt.

Implemented contract

  • Historical text images publish the original 40-byte BootInfo v1 contract.
  • The separate graphics image publishes a 40-byte BootInfo v2 contract with
    physical base 0xA0000, width 320, height 200, pitch 320, mode 0x13, and
    depth 8 bits per pixel.
  • Stage2 calls BIOS video services only in real mode. It reads the current mode
    back with INT 10h/AH=0Fh before publishing v2.
  • The graphics kernel accepts only the exact v2 contract before touching VGA
    memory. Text kernels continue to require v1.
  • Pixel coordinates are checked against width and height before address
    calculation. The current rectangles are known to fit; later clipping is a
    Day 26 concern.
  • VGA memory below 1 MiB remains identity mapped and excluded from the physical
    page allocator. Both properties are checked at runtime.
  • COM1 remains the diagnostic truth after switching away from text mode.

Boundaries

Mode 13h is 320x200 with one byte per pixel. The byte is a palette index, not
an eight-bit RGB encoding. This implementation uses the BIOS default palette
and does not program the DAC. It has no VBE, high-resolution linear
framebuffer, double buffering, clipping of arbitrary signed rectangles, font,
text layout, vertical synchronization, GPU acceleration, or real-hardware
coverage.

The forced failure is a deterministic test of this project’s readback branch.
It does not prove that INT 10h/AH=00h exposes every possible BIOS failure;
the fixed SeaBIOS implementation does not propagate its internal set-mode
failure as a general status code.

Cumulative regression

After the specialized run, the same tree passed check-pm, check-pipe,
check-shell, check-fd, check-fat, check-block, check-process,
check-syscall, check-elf, check-memory, check-runtime-limit, and
check-keyboard. These are QEMU/GDB runtime checks, not compile-only targets.
The full output is docs/day25-cumulative.txt.

Standalone attachment verification

The frozen os-day-25.zip has SHA-256
57e512a88ef1490d47fb21c82f388670e0d1f684a696bfb9ef18411ed18b53a5.
It contains no build/ directory. It was extracted into a fresh temporary
directory, rebuilt from scratch, and passed check-graphics plus all twelve
cumulative targets listed above. Rebuilt kernel.bin, kernel-graphics.bin,
and mbr-graphics.img hashes match the source-tree values. The full receipt is
docs/day25-zipcheck.txt; it was created after the archive was frozen and is
intentionally not inserted back into the ZIP whose hash it describes.