Community-maintained, AI-assisted, heavily patched Linux driver for the AVerMedia GC573 (PCI 1461:0054, subsystem 1461:5730). Modernized for recent kernels. Experimental — development and testing only.

Last aligned with code: 2026-05-18 · Phase 3 (color / byte-order validation)

Kernel compatibility: Builds and runs on 6.19.x–7.x with matching headers (tested on CachyOS / Arch-style Clang kernels). There is no portable prebuilt .ko in the repo — always run ./build.sh LLVM=1 CC=clang on the machine (and kernel) you use.

*Phase 2: End-to-end streaming uses the FPGA MMIO path only (0x10, 0x304, 0x1040). ITE6805 register writes that the Windows driver performs in stream_on are disabled.

Load from driver/ (or pass parameters to insmod):

cd driver
sudo insmod cx511h.ko force_input_mode=1
# Runtime: echo 1 | sudo tee /sys/module/cx511h/parameters/force_input_mode

ibt=off iommu=pt

• Kernel headers for running kernel (/lib/modules/$(uname -r)/build)
• base-devel, llvm, clang
• AverMediaLib_64.a in repository root (included in clone)

# Build (output: driver/cx511h.ko, copy to project root)
./build.sh LLVM=1 CC=clang

# Verify vermagic
modinfo cx511h.ko | grep vermagic

# Load
sudo ./insmod.sh

build.sh copies cx511h.ko to the project root; insmod.sh loads driver/cx511h.ko (always the freshest build after build.sh).

sudo ./install.sh   # requires cx511h.ko in project root
sudo ./insmod.sh    # or: modprobe cx511h (after depmod)

Install path: /lib/modules/$(uname -r)/kernel/drivers/media/avermedia/

Recommended: use ffplay / ffmpeg with explicit UYVY — avoid v4l2-ctl for streaming tests (see Known Issues).

sudo ./insmod.sh

# UYVY = FPGA native output in auto mode
ffplay -f v4l2 -input_format uyvy422 -video_size 1920x1080 -framerate 60 /dev/video0

sudo ./unload.sh

./gst_1.0_raw_video.sh 0
./gst_1.0_raw_video_audio.sh 0

Source: driver/board/cx511h/board_v4l2.c → cx511h_stream_on().

1. I2C reads — ite6805_get_frameinfo(), get_workingmode(), get_colorspace(), get_sampingmode() (reads only; safe today)
2. CPU — fill vip_cfg (resolution, framerate, colorspace, bypass, pixel format; optional force_input_mode)
3. MMIO — aver_xilinx_enable_video_streaming(FALSE); msleep(50)
4. MMIO — aver_xilinx_config_video_process(&vip_cfg)
5. MMIO — pci_model_mmio_write(0x1040, csc_value) — FPGA CSC
6. MMIO — msleep(200); optional pixel-format debug (debug_pixel_format / auto_test_byteorder)
7. MMIO — aver_xilinx_enable_video_streaming(TRUE)
8. MMIO — pci_model_mmio_write(0x304, 0x01) — initial doorbell

YUV422 from userspace is mapped to FPGA UYVY unless debug_pixel_format overrides.

Reason: hdmirxwr() during streaming can deadlock the ITE6805 I2C bus. CSC and stream control use MMIO instead.

1. v4l2_model_buffer_done()
2. pci_model_mmio_write(0x304, 0x01) — doorbell
3. pci_model_mmio_write(0x10, 0x02) — IRQ ACK
4. wmb()
5. KERN_DEBUG frame count once per minute (~3600 frames @ 60 fps)

aver_xilinx_enable_video_streaming(FALSE) only — no I2C teardown.

• LLVM=1 CC=clang on Clang-kernel distros
• -fcf-protection=none, -fno-stack-protector, MODULE_INFO(ibt, "N")
• Blob: cp AverMediaLib_64.a → AverMediaLib_64.o via custom kbuild rule

1. PCI handle · 2. I2C manager · 3. GPIO manager · 4. Memory manager · 5. Task manager
2. aver_xilinx_init + aver_xilinx_init_registers · 7. I2C bus I2C_BUS_COM · 8. Board GPIO
3. board_i2c_init · 10. ITE6805 attach (with linker-section fallback) · 11. Bitmap overlay
4. ALSA · 13. V4L2

• Name: AVerMedia CL511H (subsystem 0x5730)
• Formats: S16_LE, S24_LE
• Rates: 32 / 44.1 / 48 / 96 / 192 kHz
• Channels: 2 (stereo PCM device; alsa_model.c)
• Buffer: period_size = 7680×4 (30720 bytes), up to 128 periods (board_alsa.c)

Legacy board_suspend / board_resume wired from PCI setup — not migrated to dev_pm_ops (see Known Issue #10).

• Status: workaround in stream_on
• Issue: hdmirxwr() (I2C write) to ITE6805 during/around streaming can freeze the system.
• Impact: TTL, unmute, streaming, CSC I2C sequences skipped; MMIO path used instead.
• Reads (ite6805_get_*) still run in stream_on and event handlers.

• Status: needs hardware validation
• Issue: YUYV vs UYVY mismatch between FPGA and userspace.
• Try: debug_pixel_format=0..3, or ffplay with each of uyvy422 / yuyv422 / yvyu422 / vyuy422.
• Workaround: -input_format uyvy422 in ffmpeg/ffplay.

• Driver advertises many pixel formats; hardware path uses UYVY for YUV422 in auto mode.
• Always set capture format explicitly in applications.

• Status: mitigated by unload.sh (PipeWire / WirePlumber / ffplay).

• Status: known
• Issue: v4l2-ctl (especially --stream-mmap, format probing, or scripts that call it) can trigger I2C traffic and hang.
• Prefer: ffplay / ffmpeg with explicit formats; raw dd only while streaming.

• Status: unchanged — use LLVM=1 CC=clang on Clang-built kernels.

• Status: under investigation
• Location: board_v4l2.c ~506–507
• Issue: vactive = width, hactive = height (swapped vs usual timing names). Bypass tables use the same convention — test before “fixing”.

• Status: operational note (not a code bug)
• Issue: Module must be rebuilt after each kernel update; vermagic mismatch prevents load.
• Fix: ./build.sh LLVM=1 CC=clang then reload.

• Hardware realistically: YUV422 (UYVY). OBS/GStreamer may pick an unsupported FourCC.

• PCI driver uses legacy suspend hooks; should move to dev_pm_ops for long-term 7.x maintenance.

• Status: misleading on its own
• Issue: dump_first_pixels() reads the DMA physical address via PCI MMIO offset, not CPU-mapped buffer memory — dmesg output may not reflect real frame bytes. Prefer userspace dd + xxd (below).

• Status: broken / risky
• Issue: wrong height parsing; calls v4l2-ctl; video caps use YV12/YUY2 without matching capture format.

• Status: software limited
• On lock, widths > 1920 are forced to 1920×1080@60 in cx511h_ite6805_event() — true 4K capture not implemented.

First step for color problems: dump one frame while HDMI source is active and compare hex patterns.

sudo ./insmod.sh
# Start capture in another terminal, e.g. ffplay ... &
sleep 2
sudo dd if=/dev/video0 of=frame_raw.bin bs=4147200 count=1
xxd frame_raw.bin | head -4

# Kernel log
dmesg | grep -iE 'cx511h-csc|cx511h-phase2|cx511h-color|cx511h-dma|cx511h-pixfmt'

# Module parameters (from driver/)
cd driver && sudo rmmod cx511h 2>/dev/null; sudo insmod cx511h.ko debug_pixel_format=1

# Force input colorspace interpretation (from driver/)
cd driver && sudo rmmod cx511h 2>/dev/null; sudo insmod cx511h.ko force_input_mode=1

# ffplay format sweep (safe)
for fmt in uyvy422 yuyv422 yvyu422 vyuy422; do
  ffplay -f v4l2 -input_format $fmt -video_size 1920x1080 -framerate 60 /dev/video0
done

• Register probing and Windows-driver comparison
• V4L2 / videobuf2 callback tracing
• Iterative testing on real GC573 hardware
• AI-assisted analysis (clearly experimental)

Interoperability-focused community project (EU Directive 2009/24/EC Art. 6). AVerMedia trademarks and the vendor blob belong to their respective owners.

Community project, not supported by AVerMedia. Use at your own risk.

Repository: github.com/Everlite/Avermedia-GC573-Linux
Maintained by Everlite · Thanks to derrod for earlier work.