Skip to content

HavenDOS 0.9.6.2

Choose a tag to compare

@github-actions github-actions released this 16 Nov 10:49
f8db7a3

Which file do I want?

File For
HavenDOS-0.9.6.2.iso Booting or installing -- BIOS and UEFI, Secure Boot off
HavenDOS-0.9.6.2-bundle.zip Updating an installed system: the kernel, update.inf and the Windows/Linux staging scripts, plus the ISO, handoff and hardware-test docs
HavenDOS-0.9.6.2-source.zip Building it yourself -- make && make test

HavenDOS v0.9.6.2 Beta "Daedalus" — the monitor's buffer

The second maintenance update of 0.9.6, made from Kyan's gfx try photos
of 16 November. It contains fixes only (THS-RM-1004 section 12).

Fixed

The HDMI monitor showed one colour edge to edge, never the picture

gfx try on the HP settled it. Each of the four trials painted the
monitor in its own colour (green, red, white, blue-green) and none showed
a picture. The underrun rate hardly moved:

before deep watermark 64×64 plane both DBOX credits
underrun / 20000 1504 1346 1342 1485 1497

A plane one sixteenth the size starved just as badly, so the cause was
never bandwidth. The plane was reading a buffer that does not exist.

Alder Lake-P has four display-buffer slices of 1024 blocks:

  • pipe A (the panel) uses slices 1–2;
  • pipe B (the HDMI socket) uses slices 3–4.

On this display generation each pipe's PLANE_BUF_CFG is counted from
the start of its own half. Linux subtracts mbus_ddb_offset(), which
is 2048 for slices 3–4. 0.9.6.1 wrote pipe B's buffer as blocks
2048–4031, counted from slice 1. Pipe B read that as 2048 blocks past
the start of its half, which is past the end of the 2048 blocks it has.
Every fetch came back empty, whatever the plane's size or watermark.

gfx on now writes 0–1983 (PLANE_BUF_CFG 07BF0000): the same
1984 blocks, counted the way the hardware counts them. 64 blocks are
still left free for a cursor.

Changed

  • gfx try has a fifth trial, YELLOW, which is the fix itself. It
    moves the buffer to 0–1983 on the running pipe. Lighting the monitor
    again isn't needed. Trial 0 still puts everything back afterwards,
    the buffer included.
  • gfx b names the bug when it sees it. If pipe B's buffer reaches
    past block 2047, it prints four red lines saying why the plane
    underruns and what fixes it.
  • gfx on measures the underrun rate over 20,000 reads after it
    lights the monitor, instead of reading the register once. It says
    either "the plane keeps up" or "the plane is starving".

How to test

  1. Update and reboot. Plug the monitor into the HDMI socket.
  2. Run gfx on.
    • Expected: the HavenDOS screen in the middle of the monitor, and
      underrun in 0 of 20000 reads.
    • Photograph the monitor and the last lines on the laptop.
  3. If the monitor is still a flat colour, run gfx b and gfx try, and
    photograph them as before. Trial 5 is yellow.

A monitor that was lit by 0.9.6.1 in the same boot keeps the old buffer
until gfx off. gfx try trial 5 fixes it without that.

Tests

  • test_xemode 236 (floor 226 → 236): the buffer value, trial 5,
    xm_buf_outside(), and gfx b's new lines;
  • check_xemode_suite 33/33: three new injections, one of them
    0.9.6.1's absolute numbers put back;
  • 7258/7258 checks across 69 suites.