MacroSilicon MS91xx Linux DRM driver (modern-kernel fork)

Linux DRM driver for MacroSilicon MS9132 / MS9133 / MS9135 USB display chips (lsusb: 345f:9132 / 345f:9133 "MS USB Video"), based on MacroSilicon's official GPL-2.0 source code, fixed to build and run on modern kernels (6.8 → 7.0+) and packaged for DKMS.

These chips power countless brandless "lightweight" portable USB monitors and Chinese dual-screen / tri-screen laptop extenders sold on AliExpress, Temu, Amazon, etc., as well as generic USB 3.0 → HDMI/VGA adapters.

✅ Tested working on Ubuntu 24.04 (Noble), kernel 6.17.0-35-generic, GNOME 46 Wayland with a brandless dual-screen laptop monitor (MS9132 in single mode, 1920×1080@60, USB 3 SuperSpeed) — both mirror and extend modes.

Is this my device?

$ lsusb | grep 345f
Bus 004 Device 003: ID 345f:9132 MS           USB Video

If you see 345f:9132, 345f:9133 or 345f:9135, yes.

If you see 534d:6021 ("MacroSilicon USB Video" / "VGA Display Adapter"), also yes — that is the functional identity these dongles mode-switch to on their own after enumerating as 345f:9132. USB→VGA adapters in that family are usually MS912C chips (dmesg reports chip id:0x3) and need two extra module options to produce any picture at all; see docs/MS912C-VGA.md.

Typical symptoms before installing this driver:

• The monitor only shows a red/standby LED on Linux (works fine on Windows).
• The device re-enumerates every ~10 seconds (dmesg shows connect/disconnect loops) because no driver claims its video interface.
• With the reverse-engineered ms912x driver, the panel flickers and then shows "No Signal" (see rhgndf/ms912x#42), or shows corrupted/magenta images. This driver fixes that: the official enable sequence keeps the video output muted until the first frame has been transferred, which the dual-screen (MS9133-family) firmware requires.

Install (DKMS, recommended)

git clone https://github.com/bambinounos/ms91xx-linux-drm.git
cd ms91xx-linux-drm
sudo rsync -a --exclude .git ./ /usr/src/msdisp-3.0.3.13/
sudo dkms add msdisp/3.0.3.13
sudo dkms install msdisp/3.0.3.13

Plug the screen in (or unplug/replug it). The usbdisp_usb module auto-loads via the USB modalias, a new DRM card appears, and your desktop (tested on GNOME Wayland/Mutter) lights the panel up like any other monitor. DKMS rebuilds the modules automatically on every kernel update.

Manual build (without DKMS)

make                          # builds against the running kernel
sudo insmod drm/usbdisp_drm.ko
sudo insmod drm/usbdisp_usb.ko

⚠️ The tree must live in a path without spaces. kbuild copies M= into KBUILD_EXTMOD and rejects it with $(if $(word 2,$(KBUILD_EXTMOD)),...), which splits the path on whitespace inside make — after the shell already passed it as a single argument — and aborts with the misleading building multiple external modules is not supported. No amount of quoting in the recipe avoids it, and a space-free symlink does not help either: make resolves CURDIR to the physical path. drm/Makefile now fails early with a message that names the real cause. DKMS is unaffected — it builds in /usr/src/msdisp-3.0.3.13.

What this fork fixes (vs. official V3.0.3.13)

MacroSilicon's last public release targets kernels up to ~6.6. On newer kernels it does not compile — and even when it compiles, on 6.12+ the DRM node cannot be opened at all (every open() of /dev/dri/cardN fails with EINVAL), so compositors silently ignore the device. All fixes are guarded with LINUX_VERSION_CODE, preserving compatibility with old kernels:

Fix	Affected kernels
.fop_flags = FOP_UNSIGNED_OFFSET in fops — without this, drm_open_helper() rejects every open and the GPU is unusable	≥ 6.12
platform_driver.remove returns void	≥ 6.11
drm_helper_mode_fill_fb_struct() / fb_create take const struct drm_format_info *	≥ 6.15
drm_do_get_edid() removed → ported to drm_edid_read_custom() (raw EDID copy kept, rest of the flow unchanged)	≥ 6.13
connector->mode_valid() takes const struct drm_display_mode *	≥ 6.15
MODULE_IMPORT_NS("...") requires a string literal	≥ 6.13
from_timer() → timer_container_of(), del_timer() → timer_delete()	≥ 6.16 / 6.2
.date removed from struct drm_driver; PRIME fd/handle hooks now core defaults	≥ 6.14 / 6.6
drm_legacy.h deleted; missing vmalloc.h / scatterlist.h includes	≥ 6.8
struct_mutex removed from drm_device (GEM locking is internal); explicit drm_print.h includes; DRM_DEBUG_PRIME → drm_dbg_prime()	≥ 7.0
Makefiles: removed vestigial KERNELRELEASE/include Kbuild branch that broke DKMS builds; $(PWD) → $(CURDIR); added dkms.conf	all
drm_do_get_edid() guard lowered to ≥ 6.12 — Debian's 6.12.101+deb13 backports the 6.13 EDID API while still reporting 6.12, so the build failed	≥ 6.12 (Debian)
.fbdev_probe implemented + GEM .vmap/.vunmap + drm_client_setup() (adapted for drm_client_buffer_create_dumb() and pre-allocated fb_info on 7.0+) — provides /dev/fb0 and text console support	≥ 6.10
534d:6021 added to all three ID lookup tables; port_type= override; custom_mode= modes honoured with EDID fallback for standard HDMI — see docs/MS912C-VGA.md	all
Plane atomic update fix: uses plane->state->fb instead of old_state->fb — resolves 1-frame delayed screen updates (input lag when typing) and black screen on resume	all
Wayland mouse cursor fix: supports cursor sizes ≤ 64×64 (24×24, 32×32, 48×48) with transparent padding and fixes GEM refcount leak — cursor now visible in GNOME Wayland	all
Hotplug uevents: drm_kms_helper_hotplug_event() triggered on probe, disconnect, and reset-resume so compositors immediately detect and repaint the display	all

The protocol/init logic is untouched — it is exactly the manufacturer's code.

Known issues (inherited from the vendor code)

• Do not rmmod usbdisp_drm. The vendor teardown path has a use-after-free that oopses the kernel and leaves the module stuck (refcount -1). To reload the driver, reboot. In normal use the modules simply stay loaded.
• If the panel shows only the red LED and never enumerates (no dmesg events at all when plugging), the chip's firmware is latched: disconnect the USB cable and any other power source for ~2 minutes, then reconnect.
• A ~10 s connect/disconnect loop in dmesg is normal only while no driver is bound; it stops as soon as usbdisp_usb claims the device.

Credits & provenance

• MacroSilicon Technology Co., Ltd. — original driver, licensed GPL-2.0. Source obtained from their public download server: http://www.macrosilicon.com:9080/download/USBDisplay/Linux/SourceCode/ (MS91xx_Linux_Drm_SourceCode_V3.0.3.13.zip, 2025-07-10 — the pristine zip is kept in this repo for provenance).
• rhgndf/ms912x and contributors — the reverse-engineering work that kept these devices alive on Linux for years and whose issue tracker pointed the way here.
• Modern-kernel fixes and DKMS packaging: this fork (debugged live against a real dual-screen MS9132/MS9133 device, including USB protocol captures of the Windows driver).

License

GPL-2.0 (same as the original MacroSilicon code). See LICENSE.

---

Guía rápida en español

Driver Linux para los chips MacroSilicon MS9132/MS9133 (lsusb: 345f:9132) que usan las pantallas portátiles chinas sin marca y los monitores duales para laptop ("dual screen", "lightweight monitor"). Es el código oficial GPL del fabricante, corregido para kernels modernos (probado en Ubuntu 24.04 con kernel 6.17, GNOME Wayland, modos espejo y extendido).

git clone https://github.com/bambinounos/ms91xx-linux-drm.git
cd ms91xx-linux-drm
sudo rsync -a --exclude .git ./ /usr/src/msdisp-3.0.3.13/
sudo dkms add msdisp/3.0.3.13
sudo dkms install msdisp/3.0.3.13

Conecta la pantalla y listo: aparece como un monitor más. Consejos: nunca hagas rmmod usbdisp_drm (bug del fabricante: para recargar, reinicia); si la pantalla queda solo con el LED rojo y no aparece en lsusb, desconéctala de todo ~2 minutos y vuelve a conectarla.