docs/hardware-status.md

Hardware status

On this page

Lenovo Legion Tab Gen 5 / Legion Y700 5th Gen (TB323FU), mainline Linux 7.3 plus this project's patch series (the newest stable kernel release; rows name the kernel they were checked on where it matters, and most were checked on development kernels with the same patches). Tested mostly with Debian 13 and GNOME 48 on the device, the guided install with Ubuntu 26.04 and GNOME 50, and booted with the other distributions in distros.md; the kernel itself does not depend on any of them. How fast it is compared with Android: Performance. Android firmware: everything here is on ZUI 18.0.12.104; on 18.0.12.187 the kernel hangs during boot (#9).

How to read this page#

A driver that probes, a device node that exists, or a media element that shows up is not evidence that a feature works. Every row carries one of these evidence levels:

EvidenceMeaning
measuredchecked on the device with logs, counters or numbers (current, fps, CRC, frame counters, events, bit-exact output)
observeda person saw, heard or felt it (screen, sound, LEDs, vibration)
user-reporteda person reported it working, but no log or numbers were kept; will be re-checked with a record
probe onlydriver, node or element present; never actually used
untestednever tried

Status: ✅ works · 🟡 partial or conditional · ❌ does not work · ❓ not verified · — not applicable, not implemented, or a limit of the device itself that Android has as well (Device limits).

A feature that needs a human to judge (display, sound, LEDs, vibration) is never marked ✅ without an observation. Repeated or probabilistic results (suspend, mode sets) are counted as N/M.

Labels on other pages#

Other pages say in their own words how far something was tried. This is how those words relate to the evidence levels above:

PageLabelMeansClosest level here
install.md and the pages for each PCfollowed end to enda person ran the guide as written against a rooted tablet; how far (up to the first start, or further) and how often is said next to itmeasured / observed
install-manual.md, custom-kernel.md[verified]done on the development tablet with the command or tool shownmeasured / observed
[from records]done on the development tablet, but the commands are reconstructed from notes and were not re-run in this formuser-reported
[untested]not done on any tablet; follows from how the tools workuntested
distros.mdchecked on the deviceseen on the device with that system, not "should work"observed
distribution modulesverifiedthe guided install with that module was followed end to end on a tabletmeasured / observed
experimentaltried on a tablet, with known gaps listed next to it; the installer warns before using it🟡
customyour own module; nothing is known about ituntested
this page✅ 🟡 ❌ ❓the status of a feature; every row also carries one of the evidence levels above—
How a check is done

Each ✅ comes from reproducing the feature on the device the same way every time: a scripted run that measures something (frame counters and CRCs for display modes, bit-exact comparison against a software decoder for video, battery current and temperatures for charging and thermal, repeated suspend cycles counted as N/M, traffic soaks for Wi-Fi), or a person watching or listening while a script drives the feature. A kernel change resets confidence: after a large rebase the whole table is re-checked.

Display, graphics, video#

FeatureStatusEvidenceNotes
Internal panel (CSOT, Novatek NT36536 TDDI, dual DSI)✅measured + observedused daily
DSC 1.2, 10 bpc✅measured
120 Hz✅measured + observeddefault mode
Idle refresh 60 / 30 Hz✅measured + observeddone in the kernel: the mode stays 120 Hz and only the vertical porch is stretched, so the desktop sees no mode change. No flicker; back to 120 Hz before suspend and display off
90 Hz✅measured + observedvendor timing; screen and touch fine, no underruns
165 Hz (offered as 164 Hz)✅measured + observed164 Hz uses the 120 Hz horizontal timing; screen and touch fine, no underruns. SteamOS picks it by default; long sessions not watched. The vendor's 165 Hz timing has too short a horizontal blanking and underruns every frame, so it is not offered
144 Hz✅measured + observedkernel t53 (patch 0142): the 164 Hz horizontal timing with a longer vertical front porch; no underruns, screen and touch fine. One of six switches once showed the picture shifted up, not seen again. The vendor timing underruns
Backlight✅measured + observedthe top of the range (4095) is already the high-brightness (HBM) level: Android's HBM sets the backlight chip's LED current to 28.0 mA (register 0x02 = 0x2e), the value Linux keeps all the time; brightness in nits not measured
Display off / on (with and without an external display)✅measured
GPU OpenGL 4.6 / GLES 3.2 (Adreno 840, freedreno)✅measured + observedup to 1200 MHz
GPU idle power collapse (IFPC)✅measured
Vulkan (turnip)✅measuredVulkan 1.4; SteamOS's gamescope runs on it; numbers with Debian's Mesa and the project's Mesa build in Performance
GPU compute, OpenCL (rusticl)✅measuredwith the project's Mesa build from the helper; Debian's Mesa is much slower. Numbers in Performance
Video decode H.264 / HEVC / VP9 / AV1 (iris)✅measuredoutput bit-exact against software decoders, 1080p and 4K; at 8K (7680×4320) HEVC, VP9 and AV1 are bit-exact, H.264 stops after 2 of 30 frames
Video decode 10-bit (HEVC Main10, VP9 profile 2, AV1 Main 10-bit)✅measuredbit-exact, 1080p and 4K; needs this project's iris line-buffer fix. At 8K, AV1 10-bit is bit-exact, but HEVC Main10 at 8K does not decode: with kernels up to kernel-t60 it resets the tablet (an internal buffer lands in the IOVA range the codec's firmware stream cannot address); patch 0206 (upstream) keeps buffers out of that range, and the decode then fails cleanly for lack of IOVA space
Video encode H.264 / HEVC (8-bit)✅measuredNV12 input; up to 4K (at 8K, H.264 gives broken output and HEVC none). 10-bit (HEVC Main10) is not possible yet: the driver takes no 10-bit input format
AV1 in applications—untestedthe driver works, but Debian's GStreamer 1.26 and FFmpeg 7.1 have no stateful V4L2 AV1 decoder; newer distributions not tried
Browser hardware video decode—untestedFirefox does not use V4L2 stateful decoders

Input, audio, camera, indicators#

FeatureStatusEvidenceNotes
Touchscreen✅measured + observedoccasional firmware-reset message after resume. SteamOS Gaming Mode rotates only the picture, so the builder adds a touch calibration matrix
Pen (AES)✅user-reportedLenovo Tab Pen Plus (AP500U): tracking in all corners, pressure (4096 levels, the full range reached; patches 0143, 0146), both barrel buttons, mapped to the built-in display and turning with it in GNOME without libwacom (0144); reported in issue #16 on kernels t54–t56. GNOME's Graphics Tablets settings do not list it (libwacom has no SPI devices). At 164 Hz the pen stops tracking while touch keeps working (reported in #16; 120 and 60 Hz fine). Not tested here (no pen)
Speakers (2 × Awinic aw882xx)✅measured + observedleft/right heard clean; both amplifiers start on every distribution in distros.md; speaker protection is a PipeWire filter, not Android's DSP protection
Microphones✅measuredthe internal microphone records speaker playback (noise floor −62 dBFS, playback peaks −26 dBFS)
USB-C analog audio (headset adapter)—probe onlyno audio route yet
Rear camera (Samsung S5KJNS)✅measured + observed1080p and 4072×3072 at 30 fps via libcamera; auto-exposure is imprecise. The sensor driver has only its 4080×3072 (binned, 12.5 MP) mode; the full 50 MP mode is not in it
Front camera (GalaxyCore GC08A8)✅measured + observedfull resolution (3264×2448) at 30 fps
Rear camera focus (VCM)🟡measuredmanual focus works; no autofocus in libcamera
Camera streaming with the screen off✅measuredrestarting a capture with the display off used to hang the system; fixed by patch 0116 (repeated starts with the display off pass)
Torch✅observedalso from the helper's quick settings
Flash strobe✅measured + observed/sys/class/leds/white:flash: three 0.3 s strobes at 500 mA fired and ended on their timeout (2026-11-08); libcamera does not drive the flash yet
RGB ring light✅observed + measuredcharge indicator, solid colors (a nine-color palette in the helper; colors are gamma-corrected because the LEDs are linear) and breathing run by the chip itself (hw_pattern) watched on 2026-10-03. Known limitation: every brightness or color change goes dark for about 45 ms while the chip's program restarts (no update command found in the vendor material)
Haptics (2 motors)✅observed
Volume and power keys✅observedpower key suspends and resumes (seen in SteamOS)
Emergency key (volume up + down, 10 s → back to Android)✅measured + observedAndroid restore starts 10 s after the keys are detected and Android boots (1/1); switching back to Linux from Android works

Wireless, sensors, power#

FeatureStatusEvidenceNotes
Wi-Fi (WCN7860, ath12k)✅measuredused daily; a receive stall under sustained heavy traffic is fixed by this project's ath12k patch (30-minute 2.4 and 5 GHz soaks clean); the PCIe link trains at x2 but runs at x1; 6 GHz / 320 MHz / MLO not checked
Bluetooth LE input devices✅measured + observedthe controller keeps its address across power off/on and rfkill (patch 0130), so paired devices reconnect
Bluetooth audio (A2DP)✅observedthe first syllable of a stream can be clipped
Bluetooth HFP✅measured + observedBluetooth earbuds with a microphone: the headset profile with mSBC (16 kHz) in PipeWire; a tone played to them and the microphone recorded and played back were heard clearly (2026-11-08)
Accelerometer / auto-rotate✅measured + observedvia the sensor hub (SSC)
Ambient light / auto-brightness✅measured + observed
Proximity✅measured
Compass✅measured + observedrelative rotation is right (about 90° per quarter turn); the absolute heading is about 30° off compared with a phone, and Android's map apps are 90–180° off on this device too: the sensor is uncalibrated on the stock OS as well (Device limits). GNOME's sensor proxy only lets authorized clients read it
Gyroscope, SAR—untestedbehind the sensor hub, not wired to iio-sensor-proxy
Hall sensor (cover)❓measuredneeds the original folio case
GNSS / GPS—measuredthe modem, location service and a location session run, but no satellites are ever seen; Android on the same device gets no GPS fix either — most likely no antenna, although Lenovo's spec sheet lists GPS (Device limits)
Modem (MPSS)🟡measuredboots and stays up when rmtfs and tqftpserv start before it; if its initialization stalls, the SoC firmware resets the whole tablet and Linux cannot contain it. Kept off (rmtfs and tqftpserv masked): no cellular, no usable GNSS
Thermal throttling✅measuredCPU (95 °C) and GPU (105 °C) chip limits, plus board-temperature steps from 43 °C following Android's policy; checked under a 10-minute full load
Battery readings✅measuredwhile bypass charging with the Charge Limit method, status still reads "Charging" (display only); the firmware does not report the design or full capacity
USB PD charging, PPS✅measuredabout 40 W into the battery on PPS with a 65 W charger (about 9.25 V in). The USB-C controller reports the negotiated voltage and current as 0
Charge limit, bypass charging✅measured + observedbypass with the Charge Current method (patch 0132) holds the charger at about 9 V with about 85 mA into the battery; under full CPU load the input stops at about 2.4 A (about 21 W) and the battery makes up the rest. The Charge Limit method cuts the battery current (9 A → 185 mA) but lets the charger drop to 5 V; the kernel status then stays "Charging", so the helper derives its State from the current
Suspend (s2idle) and resume🟡measuredworks (9 of 9 RTC cycles with Wi-Fi, sensors, sound, USB and video decode back each time; 15 more with storage checked after patch 0128; with a USB-C monitor connected, 5 power-key resumes); suspend still uses the CPU cluster idle state that runtime idle leaves out (next rows)
Deep sleep (CX / DDR power collapse)✅measured
CPU cluster idle state in runtime idle🟡measuredleft out by default since patch 0118 (cpuidle_psci_domain.allow_cluster_off=0 on the built-in command line; switch at run time with /sys/module/cpuidle_psci_domain/parameters/allow_cluster_off, 1 = allowed). With it allowed the tablet reset when idle (2 resets in 78 min, no error message); refused, 3 h idle without one, at the same display-off power (658 / 659 / 621 mW allowed / refused / allowed). Why the cluster's power-down resets the SoC is not fixed; s2idle still uses the state (the system state needs it). The release kernels since t38 run with it refused
Wake sources✅measuredpower key, RTC; USB wake off by default

USB, storage, other#

FeatureStatusEvidenceNotes
USB host (SuperSpeed+, hubs)✅measuredthe second port is USB 2.0 only. A USB 3 flash drive directly on the bottom port: SuperSpeed 5 Gbps, stays attached across a suspend, 4/4 replugs
USB low-speed devices✅measuredthe low-speed device inside a USB-C to HDMI cable enumerates 4/4 with kernel t52; earlier kernels sometimes failed with -71/-110, not seen since
USB gadget (network, serial)✅measured
DisplayPort alt mode✅measured + observed4 lanes HBR3 with pin assignment C: a 5120×2160 USB-C monitor at 75 Hz, its USB 2.0 hub working after replugs and across suspend (sinks in pin assignment D, which keep USB 3, are untested). Earlier, with 2 lanes HBR2: a TV through a USB-C to HDMI adapter, extended desktop at 4K 30, no flicker on either screen while the internal panel idles down to 30 Hz, unplug/replug 3/3. The adapter's own low-speed USB device fails to enumerate (harmless)
DisplayPort MST (daisy chain / "extend" on MST hubs)🟡observed (kernel-t61)patches 0157-0205 (upstream MST v7, under review, and five fixes): a Realtek MST hub with two 1920×1200 panels shows two separate displays next to the built-in one, also after suspend (3 of 3) and replugs. Android (ZUI 18.0.12.104) drives both MST streams in the kernel, but its settings use only one of them as an extended display and mirror the other; Linux extends both. With this monitor a replug sometimes leaves one or both screens dark (its hub drops HPD or stops answering right after the plug; Android shows the same); unplug and plug again, or reconnect the monitor's power. Daisy chains, 4K and audio over MST untested
DisplayPort audio✅observed (development kernel, patch 0207)sound on a DisplayPort monitor and on an HDMI monitor through a USB-C to HDMI cable, at the same time as the speakers; the PipeWire device follows plug and unplug. Unplugging the display while it plays moves the sound to the speakers; only a stream pinned to the display output stops until the application is restarted (#53). Over MST: untested
UFS storage✅measuredmulti-queue mode (patch 0119), kept over suspend (0128); fixed a deadlock that rebooted the tablet after hours of use (0129) and the recovery after a controller reset (0141 and Linux 7.3); numbers in Performance
microSD✅measuredUHS-I SDR104; also used for the multiboot roots
Multiboot (systems on microSD partitions)✅measured + observedthe systems in distros.md boot, picked by the initramfs (falls back to the internal root)
Firewall (nftables, iptables-nft, firewalld)✅measuredfirewalld including IPv6 reverse-path filtering, NixOS's default firewall; legacy iptables tables are not built
Audio DSP (ADSP)✅measured
Compute DSP / NPU (Hexagon HTP v81)✅measuredQualcomm's QNN runtime and llama.cpp's Hexagon backend run with the open fastrpc library (unsigned PD); not packaged by the project. Numbers in Performance
CPU frequency scaling, boost✅measured6 cores up to 3.63 GHz and 2 up to 4.61 GHz (boost), the same maximum frequencies as Android; the helper sets limits per performance profile and can turn boost off
CPU capacity, energy-aware scheduling✅measuredpatch 0117 gives the scheduler the two core sizes (cpu_capacity 799 / 1024; before it all eight read 1024, so a busy single thread stayed on a small core) and energy-aware scheduling turns on (Geekbench 6 single-core without pinning: 2288 before 0117, at Android's level after; numbers in Performance)
Switching between Android and Linux✅measured + observedabout 40 s
Boot time✅measuredabout 6 s from kernel start to the desktop (Debian, automatic login); 7–15 s on the other distributions

Known issues#

  • CPU cluster idle state left out of runtime idle (patch 0118): with it the tablet reset when idle, without an error message. Leaving it out avoids that at the same idle power; the cause in the cluster's power-down path is not fixed, and suspend still uses the state (knob: /sys/module/cpuidle_psci_domain/parameters/allow_cluster_off).
  • 8K HEVC 10-bit video does not decode: with kernel-t60 and older, hardware decoding of an 8K (7680×4320) HEVC Main10 stream resets the tablet; from kernel-t61 (patch 0206) it fails with a decode error instead. 8K 8-bit HEVC, VP9 and AV1 (also 10-bit) decode fine; H.264 and encoding do not work at 8K.
  • Bypass charging works; with the Charge Limit method the kernel battery status still reads "Charging" and the charger drops to 5 V (the Charge Current method keeps 9 V).
  • Charging can stop below the limit: the firmware stops at 5 V and resumes only when the limit is moved below the current charge and back; the helper's Restart Stalled Charging does this automatically. Cause not found.
  • Modem: a stalled modem start resets the whole SoC, so rmtfs and tqftpserv stay masked on every root (see distros.md).
  • SteamOS (systemd 257.7) can hang a shutdown when the USB serial console is connected to a PC that is not reading it (the tty reset waits for output to drain, then the watchdog resets the tablet). The SteamOS builder turns that tty reset off; fixed in systemd 257.8 and later; other distributions are not affected.
  • Do not force higher display bandwidth votes: it resets the SoC.
  • Do not stop the audio DSP at runtime: it takes the SoC down.

Device limits (same on Android)#

These come from the hardware or its firmware; the stock Android on the same tablet behaves the same way, so they are not gaps in the Linux port.

  • GNSS: no satellites are seen on Linux, and Android gets no GPS fix either — most likely no antenna.
  • Compass: relative rotation is right, the absolute heading is off; Android's map apps are off on this device too.
  • Fast USB-C replugs: replugging a USB-C monitor within about a second sometimes ends without a power-delivery contract (no picture, no hub); unplug and plug again. Android shows "Check the charger connection" in the same case.

Still to test#

In rough order of value. Rows above marked user-reported, ❓ or 🟡, and rows last checked before a relevant kernel change, are here.

WhatHow it will be settledNeeds a person
164 Hz in long sessions, SteamOS orientation and UI size, touch on Fedora and NixOS10+ minutes in Gaming Mode with underrun counters; first taps on each desktopeyes, hands
Long idle with patch 0118hours of idle and display-off time without a reset, against the 78-minute baseline with the cluster state allowedno (scripted)
AV1 in applicationsnewer GStreamer / FFmpeg on Arch, Fedora or NixOS, compared bit-exact with a software decoderno (scripted)
144 Hzlong sessions, after suspend and with idle refresh; the shifted picture seen onceeyes, hands
From the spec sheetfull-resolution (50 MP) rear capture (needs a new sensor mode), touch report rate (up to 480 Hz), HDR, film-grain videomostly no; touch rate needs a finger
Flash strobetrigger the strobe control and watcheyes
Board thermal above 45 °C20–30 minutes of CPU + GPU load, steps and frequencies logged, desktop smoothness watchedeyes (optional)
Charging stall and the 2.4 A input capbattery and charger values logged until the next stall; the input limit on other chargersno (scripted)
Needs hardware not at handpen, Bluetooth LE Audio, USB audio adapters, the original folio case, the 68 W charger, a 6 GHz / 320 MHz / MLO access pointhands

Spec sheet vs status#

Lenovo / Qualcomm specifications compared with the status here

Sources: Lenovo PSREF "Legion Tab (8.8", 5)" and datasheet, Qualcomm Snapdragon 8 Elite Gen 5 product brief.

AreaLenovo / QualcommStatus here
Display8.8" 3040×1904, up to 165 Hz, HBM 800 nit, HDR120 / 90 / 60 / 30 / 144 Hz and 165 (as 164) ✅; HBM ✅ (the top brightness is the HBM LED current), HDR untested
External displayUSB-C DisplayPort; SoC up to 4K 120 / 8K 305120×2160 75 Hz, 4 lanes HBR3 ✅; MST 🟡 (two 1920×1200, development kernel)
Video decodeH.264, H.265, VP9, AV1, 10-bitall ✅ in the driver (1080p and 4K); 8K ✅ for HEVC 8-bit, VP9, AV1 (8- and 10-bit), ❌ for H.264 and HEVC 10-bit
Video encodeHEVC 10-bitH.264 / HEVC 8-bit ✅ up to 4K; 10-bit ❌ (no 10-bit input in the driver)
GPUGLES 3.2, Vulkan 1.3, OpenCL 3.0GL ✅, Vulkan ✅, OpenCL ✅ (the project's Mesa build)
NPUHexagon NPUQNN and llama.cpp run ✅
Audio2 speakers, 2 microphones✅
Cameras50 MP rear with AF and flash, 8 MP front✅ (rear up to 12.5 MP binned, front at its full 8 MP); 50 MP mode not in the driver; AF manual only
Sensorsaccelerometer + gyro, hall, proximity + light, compass, GPSaccelerometer, light, proximity ✅; compass ✅ (absolute heading off, as on Android); gyro not wired; GPS — (no fix on Android either)
Charging68 W PD 3.0 / PPS, bypass chargingabout 40 W on PPS ✅, bypass ✅ (charger kept at 9 V)
USBUSB-C 10 Gbps + DP, second port USB 2.0✅
StorageUFS 4.1, microSD up to 2 TB✅
Wi-Fi / BluetoothWi-Fi 7, Bluetooth 6.0Wi-Fi ✅ (2.4 / 5 GHz; 6 GHz untested), BT ✅ (HFP ✅; LE Audio untested)

View this page's Markdown on GitHub