You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Support for Synaptics 06cb:00b7 (Sensor Type 0xd51) #225
Hello, I'm trying to get a fingerprint sensor with USB ID 06cb:00b7 working. After extensive troubleshooting, it appears the device is unsupported but very close to being compatible. The initialization fails even after applying several patches to the driver source code.
Here is a summary of the manual patches applied in order:
Udev Rule: Edited /usr/lib/udev/rules.d/60-python-validity.rules to add a line for the 00b7 product ID to ensure the python3-validity.service is triggered for this
Device ID in Driver: Edited /usr/lib/python3.13/site-packages/validitysensor/usb.py to add DEV_b7 = (0x06cb, 0x00b7) to the SupportedDevices Enum, which fixed an initial "No matching devices found" error.
Firmware Blob Loading: Edited /usr/lib/python3.13/site-packages/validitysensor/blobs.py. The __load_blob function was failing with an UnboundLocalError because it did not have a case for the 00b7 product ID. I modified the line if usb.usb_dev().idProduct == 0x009a: to if usb.usb_dev().idProduct == 0x009a or usb.usb_dev().idProduct == 0x00b7: to make my device load the blobs_9a module.
Sensor Type Profile: Edited /usr/lib/python3.13/site-packages/validitysensor/sensor.py. In the open function, the sensor was correctly identified as sensor type 0xd51 but was then rejected. I modified the line if self.device_info.type == 0x199: to if self.device_info.type == 0x199 or self.device_info.type == 0xd51: to make the driver use the 0x199 sensor profile.
Capture Program: Still in sensor.py, the driver then failed with the error Can't find initial capture program. I modified the line self.hardcoded_prog = SensorCaptureProg.get(self.rom_info, self.device_info.type, ...) to self.hardcoded_prog = SensorCaptureProg.get(self.rom_info, 0x199, ...) to force it to load the capture program for the supported 0x199 sensor.
After this final change, the python3-validity.service still fails to start. The full journalctl log for this final failure is attached below.
lsusb output:
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
Bus 001 Device 002: ID 04d9:fc30 Holtek Semiconductor, Inc. Gaming Mouse [Redragon M711]
Bus 001 Device 003: ID 06cb:00b7 Synaptics, Inc. Fingerprint reader [HP G6]
Bus 001 Device 004: ID 04f2:b669 Chicony Electronics Co., Ltd HP HD Camera
Bus 001 Device 005: ID 8087:0029 Intel Corp. AX200 Bluetooth
Bus 002 Device 001: ID 1d6b:0003 Linux Foundation 3.0 root hub
Progress update: 06cb:00b7 support has been iterated on in PR #256 through several rounds.
Initial 0xd51 handling covers the HP G6 variants that report sensor type 0xd51 (@RamcharanRajpurohit confirmed working on Fedora 43).
@Karloss1234's Point 1 review added 0x969 support (the other silicon behind this same USB ID, reporting sensor type 0x969 — HP ZBook 17 G6, ZBook Studio G5, ProBook G6 etc.). Merged in 0c00f0a.
His Point 4 fix for post-suspend/resume regression on 0x969 chips (the chip re-enumerates reporting 0x199 directly rather than 0x969, so Sensor.open() was skipping the alias + interrupt fix). Landed in 3b23194.
His Point 6 diagnosis of flash-management (04b5 on enroll, del_record missing db_write_enable, no auto-compaction) — the del_record bug is fixed in 82584e1; the "chip doesn't auto-compact after deletion" facet is documented with erase_flash(4) as recovery.
Karloss's review of PR #256 is the reference thread for the 0x969 details on this chip family. Full end-to-end on Kubuntu 26.04 + HP ZBook 17 G6: enroll, verify, sudo, KDE lock screen, resume from suspend.
If you're on 0x969 hardware and can test after the next PPA build (0.16~hp8, incoming), that would help close this out.
Device Support Confirmation: HP ProBook 650 G7 (06cb:00b7 / Sensor Type 0xd51)
Hello, I can confirm that the 06cb:00b7 Synaptics fingerprint reader (Sensor Type 0xd51 ) is working perfectly on my HP ProBook 650 G7 using the changes in PR #256 (from branch feat/sensor-type-0xd51 on
SimpleX-T/python-validity ).
Here are the details of the testing:
Firmware Extraction: Successful. Running validity-sensors-firmware downloaded and extracted 6_07f_hp_cmit_mis_qm.xpfwext from the HP Softpaq package sp135736.exe .
Enrolling: fprintd-enroll runs, communicates correctly with the device, and completes enrollment:
Thanks @rsomonte — this is exactly the kind of validation we needed for this issue's specific 06cb:00b7 + 0xd51 combo, on a model (HP ProBook 650 G7) that hadn't been reported before. Fedora 44 + kernel 7.0.14 + GNOME 50.3 Wayland is also new ground.
For anyone else reading this on the same USB ID: 06cb:00b7 has been observed as two different silicon variants in the wild:
0x969 — HP ZBook 17 G6 (Karloss1234's report). Same PID, different chip, added to the PR in 0c00f0a from @ggiesen's SimpleX-T/python-validity#2.
Both variants are handled on the current PR #256 branch. 0.16~hp8 was uploaded to the PPA last night rolling up all the recent fixes — Launchpad's build queue should have the Fedora-equivalents (well, the Ubuntu builds — Fedora users would still install from source via COPR / setup.py, since we don't ship a .spec here).
If you have the time, a bench_capture.py run on your machine would give us a second empirical data point on capture quality for this chip family (my baseline is ~0.9 retries/verify on the 840 G5, 0xd51). The harness is in the source tree at bench_capture.py; runs 10 back-to-back fprintd-verify invocations and reports mean/median/p95 chip-side retries. Purely optional though — the enroll + PAM confirmation you posted is already the important bit.
Here are the capture quality metrics from my machine.
Note that I couldn't find the bench_capture.py script in the source tree of the repository, so I wrote a custom harness script that runs fprintd-verify 10 times and parses the journalctl -u python3-validity.service log output to calculate the stats based on the Chip capture retry-scan #X entries.
Here are the results:
Run
Retries
Match Outcome
Run 1
0
verify-no-match
Run 2
3
verify-no-match
Run 3
0
verify-no-match
Run 4
0
verify-no-match
Run 5
0
verify-match
Run 6
0
verify-no-match
Run 7
0
verify-no-match
Run 8
1
verify-match
Run 9
0
verify-no-match
Run 10
1
verify-no-match
Mean
0.50
Median
0.0
p95
3.0
Summary & Feedback
Mean Retries: 0.50 (very fast and clean compared to the G5 baseline of ~0.9).
Median: 0.0
p95: 3.0
Matching Observation: During the test runs, even when using the correct enrolled finger, the verification returned verify-no-match on 8 out of 10 attempts (only matching successfully on Run 5 and Run 8). It seems that while the capture itself is extremely fast with low retry counts, the matching threshold is quite sensitive, and slightly off-angle scans are registered as incorrect fingerprints rather than asking for a retry. Re-enrolling or scanning at the exact angle helps, but there is a noticeable rate of false negatives.
Quote: * Matching Observation:** During the test runs, even when using the correct enrolled finger, the verification returned `verify-no-match` on 8 out of 10 attempts (only matching successfully on Run 5 and Run 8). It seems that while the capture itself is extremely fast with low retry counts, the matching threshold is quite sensitive, and slightly off-angle scans are registered as incorrect fingerprints rather than asking for a retry. Re-enrolling or scanning at the exact angle helps, but there is a noticeable rate of false negatives. *
Does your system have a separate boot loaded Windows OS, with a finger enrolled in Windows Hello?
I had similar results to you when I had my right index finger enrolled in both Windows Hello and Linux fprintd-enrol.
In Linux fprintd-verify, all stored templates compete for verification and Windows Hello enroll template algorithm seems to be better at verifying, therefore the Windows Hello template wins fprintd-verify 80-90% of attempts, and therefore you get a 'verify-no-match'.
Delete/Switch your Windows Hello enrolled finger to your middle finger and see if that helps. It resulted in me getting a 95% Linux verify-match rate.
After switching the Windows Hello enrolled finger to a different finger and re-enrolling my right index finger under Linux fprintd-enroll, the template competition issue completely disappeared.
I re-ran the 10-invocation test harness, and all 10 out of 10 runs matched successfully on the first attempt (100% match rate).
Benchmark Results
Run
Retries
Outcome
Run 1
6
verify-match (exit 0)
Run 2
0
verify-match (exit 0)
Run 3
1
verify-match (exit 0)
Run 4
0
verify-match (exit 0)
Run 5
1
verify-match (exit 0)
Run 6
0
verify-match (exit 0)
Run 7
0
verify-match (exit 0)
Run 8
0
verify-match (exit 0)
Run 9
0
verify-match (exit 0)
Run 10
0
verify-match (exit 0)
Mean
0.80
100% Match Success
Median
0.0
p95
6.0
Summary & Takeaways
• Match Success Rate: 100% (10/10 successful verify-match).
• Mean Retries: 0.80 (7 out of 10 scans matched on 0 retries).
• Root Cause Confirmation: Having the exact same finger enrolled in both Windows Hello and Linux fprintd causes the on-chip Windows Hello template to win the match search ~80% of the time, causing Linux fprintd-verify to see a SID mismatch and return verify-no-match.
• Recommendation: Dual-boot users on 0xd51 / 0x969 silicon should ensure they do not enroll the exact same finger in both Windows Hello and Linux fprintd , or assign separate fingers for each OS.
06cb:00b7 / 0xd51 works natively via libfprint MR !626 — pairing recovered, no chip reset
DUT: HP EliteBook x360 1040 G6, Ubuntu 26.04.1, kernel 7.0.0-31, GNOME 50.3 Wayland. 06cb:00b7 rev 1.64, sensor type 0xd51, firmware Version 6.7 Product 48 Build 164,
left with an orphaned Windows Hello pairing. (Only the 0xd51 variant.)
I validated the validity driver from libfprint MR !626 on this
PID: probe, open (TLS + calibration), enroll, list and identify all succeed via the
normal libfprint API — details and review notes are in that MR.
Two data points for this repo:
Your hard-coded constants are enough for this chip. The MR driver needs a data
tree from libfprint-validity-data, which isn't published, so I generated it from
python-validity: init.bin / init_clean_slate.bin / reset.bin / db_write_enable.bin from blobs_9a.py (i.e. the 06cb:0090/009a set), plus tls.password_hardcoded, gwk_sign_hardcoded, crt_hardcoded, init_flash.partition_signature[_0090] and the two EC coords from tls.handle_ecdh
(byte-reversed — the driver wants little-endian). Handshake, calibration, enroll and
identify all worked on 0xd51 with those. Question: is there any genuinely
per-device content in those blobs, or are they firmware-family constants? If the
latter, the SupportedDevices gate in usb.py is the only real per-device piece and 00b7 could simply be added to it — which is what this thread's patches already do.
The old pairing was recovered, not reset. The driver decrypted the host private
key from the sensor's TLS flash block with the PSK and completed the handshake against
the pairing Windows had created — no factory reset, nothing wiped. Worth an FAQ note:
"reset the chip" is the usual first advice for 00b7 under Windows Hello, but it isn't
always needed and it irreversibly wipes stored prints.
Practical consequence: if !626 lands (with a published data package, which it currently
lacks), 06cb:00b7 is supported in mainline libfprint/fprintd and this python route
becomes legacy — cross-linking the two, and closing #225 as upstream-tracked after the
merge, would save the next person a lot of patching.
For me, on Pop! OS, I did get it working like this, thank you very much!
One problem was that it would only work once , then the second time fprintd-verify would immediately exit with a no-match. Only way to get it working (once more, just once) was to restart the services.
After some extensive Gemini troubleshooting the solution seems to be to edit dbus-service and in the try: block of def run():, add a line usb.cancel_event.clear()
before usrid, subtype, hsh = sensor.identify(update_cb)
So it reads
try:
usb.cancel_event.clear() # this is the fix
# TODO: pass down the user db record id and implement a proper Sensor.verify() method
usrid, subtype, hsh = sensor.identify(update_cb)
if usr.dbid == usrid:
self.VerifyStatus('verify-match', True)
else:
self.VerifyStatus('verify-no-match', True)
except usb_core.USBError as e:
logging.exception(e)
self.VerifyStatus('verify-no-match', True)
loop.quit()
except Exception as e:
logging.exception(e)
self.VerifyStatus('verify-no-match', True)```
Activity
SimpleX-T commentedon May 21, 2026
Hey @rsomonte — installation is now packaged so you can drop the manual patches:
Builds for Ubuntu 22.04, 24.04, 25.10, and 26.04. Let me know if you hit anything on your machine.
SimpleX-T commentedon Jul 7, 2026
Progress update:
06cb:00b7support has been iterated on in PR #256 through several rounds.0c00f0a.Sensor.open()was skipping the alias + interrupt fix). Landed in3b23194.04b5on enroll,del_recordmissingdb_write_enable, no auto-compaction) — thedel_recordbug is fixed in82584e1; the "chip doesn't auto-compact after deletion" facet is documented witherase_flash(4)as recovery.Karloss's review of PR #256 is the reference thread for the 0x969 details on this chip family. Full end-to-end on Kubuntu 26.04 + HP ZBook 17 G6: enroll, verify, sudo, KDE lock screen, resume from suspend.
If you're on 0x969 hardware and can test after the next PPA build (
0.16~hp8, incoming), that would help close this out.rsomonte commentedon Jul 8, 2026
Device Support Confirmation: HP ProBook 650 G7 (06cb:00b7 / Sensor Type 0xd51)
Hello, I can confirm that the 06cb:00b7 Synaptics fingerprint reader (Sensor Type 0xd51 ) is working perfectly on my HP ProBook 650 G7 using the changes in PR #256 (from branch feat/sensor-type-0xd51 on
SimpleX-T/python-validity ).
Here are the details of the testing:
fprintd-enrollruns, communicates correctly with the device, and completes enrollment:System Information ( fastfetch )
DMI System Hardware Info ( dmidecode )
USB Device Info ( lsusb -v )
SimpleX-T commentedon Jul 8, 2026
Thanks @rsomonte — this is exactly the kind of validation we needed for this issue's specific
06cb:00b7 + 0xd51combo, on a model (HP ProBook 650 G7) that hadn't been reported before. Fedora 44 + kernel 7.0.14 + GNOME 50.3 Wayland is also new ground.For anyone else reading this on the same USB ID:
06cb:00b7has been observed as two different silicon variants in the wild:0xd51— HP G6 series (this reporter's ProBook 650 G7, plus RamcharanRajpurohit's earlier Fedora 43 report, plus @bcoutts's ProBook 445R G6 — although his is stuck on a separate factory-init issue tracked on PR Add support for sensor type 0xd51 (138a:00ab HP 840 G5, 06cb:00b7 HP G6) #256).0x969— HP ZBook 17 G6 (Karloss1234's report). Same PID, different chip, added to the PR in0c00f0afrom @ggiesen's SimpleX-T/python-validity#2.Both variants are handled on the current PR #256 branch.
0.16~hp8was uploaded to the PPA last night rolling up all the recent fixes — Launchpad's build queue should have the Fedora-equivalents (well, the Ubuntu builds — Fedora users would still install from source via COPR / setup.py, since we don't ship a.spechere).If you have the time, a
bench_capture.pyrun on your machine would give us a second empirical data point on capture quality for this chip family (my baseline is ~0.9 retries/verify on the 840 G5, 0xd51). The harness is in the source tree atbench_capture.py; runs 10 back-to-backfprintd-verifyinvocations and reports mean/median/p95 chip-side retries. Purely optional though — the enroll + PAM confirmation you posted is already the important bit.rsomonte commentedon Jul 8, 2026
Capture Quality Benchmark & Feedback (HP ProBook 650 G7 / 06cb:00b7 / 0xd51)
Here are the capture quality metrics from my machine.
Note that I couldn't find the
bench_capture.pyscript in the source tree of the repository, so I wrote a custom harness script that runsfprintd-verify10 times and parses thejournalctl -u python3-validity.servicelog output to calculate the stats based on theChip capture retry-scan #Xentries.Here are the results:
verify-no-matchverify-no-matchverify-no-matchverify-no-matchverify-matchverify-no-matchverify-no-matchverify-matchverify-no-matchverify-no-matchSummary & Feedback
verify-no-matchon 8 out of 10 attempts (only matching successfully on Run 5 and Run 8). It seems that while the capture itself is extremely fast with low retry counts, the matching threshold is quite sensitive, and slightly off-angle scans are registered as incorrect fingerprints rather than asking for a retry. Re-enrolling or scanning at the exact angle helps, but there is a noticeable rate of false negatives.Karloss1234 commentedon Jul 21, 2026
Quote:
* Matching Observation:** During the test runs, even when using the correct enrolled finger, the verification returned `verify-no-match` on 8 out of 10 attempts (only matching successfully on Run 5 and Run 8). It seems that while the capture itself is extremely fast with low retry counts, the matching threshold is quite sensitive, and slightly off-angle scans are registered as incorrect fingerprints rather than asking for a retry. Re-enrolling or scanning at the exact angle helps, but there is a noticeable rate of false negatives. *Does your system have a separate boot loaded Windows OS, with a finger enrolled in Windows Hello?
I had similar results to you when I had my right index finger enrolled in both Windows Hello and Linux fprintd-enrol.
In Linux fprintd-verify, all stored templates compete for verification and Windows Hello enroll template algorithm seems to be better at verifying, therefore the Windows Hello template wins fprintd-verify 80-90% of attempts, and therefore you get a 'verify-no-match'.
Delete/Switch your Windows Hello enrolled finger to your middle finger and see if that helps. It resulted in me getting a 95% Linux verify-match rate.
rsomonte commentedon Jul 29, 2026
Your diagnosis was 100% correct!
After switching the Windows Hello enrolled finger to a different finger and re-enrolling my right index finger under Linux
fprintd-enroll, the template competition issue completely disappeared.I re-ran the 10-invocation test harness, and all 10 out of 10 runs matched successfully on the first attempt (100% match rate).
Benchmark Results
Summary & Takeaways
• Match Success Rate: 100% (10/10 successful
verify-match).• Mean Retries: 0.80 (7 out of 10 scans matched on 0 retries).
• Root Cause Confirmation: Having the exact same finger enrolled in both Windows Hello and Linux
fprintdcauses the on-chip Windows Hello template to win the match search ~80% of the time, causing Linuxfprintd-verifyto see a SID mismatch and returnverify-no-match.• Recommendation: Dual-boot users on 0xd51 / 0x969 silicon should ensure they do not enroll the exact same finger in both Windows Hello and Linux fprintd , or assign separate fingers for each OS.
AcckiyGerman commentedon Sep 25, 2026
06cb:00b7/0xd51works natively via libfprint MR !626 — pairing recovered, no chip resetDUT: HP EliteBook x360 1040 G6, Ubuntu 26.04.1, kernel 7.0.0-31, GNOME 50.3 Wayland.
06cb:00b7rev 1.64, sensor type0xd51, firmwareVersion 6.7 Product 48 Build 164,left with an orphaned Windows Hello pairing. (Only the
0xd51variant.)I validated the
validitydriver from libfprint MR!626 on this
PID: probe, open (TLS + calibration), enroll, list and identify all succeed via the
normal libfprint API — details and review notes are in that MR.
Two data points for this repo:
tree from
libfprint-validity-data, which isn't published, so I generated it frompython-validity:
init.bin/init_clean_slate.bin/reset.bin/db_write_enable.binfromblobs_9a.py(i.e. the 06cb:0090/009a set), plustls.password_hardcoded,gwk_sign_hardcoded,crt_hardcoded,init_flash.partition_signature[_0090]and the two EC coords fromtls.handle_ecdh(byte-reversed — the driver wants little-endian). Handshake, calibration, enroll and
identify all worked on
0xd51with those. Question: is there any genuinelyper-device content in those blobs, or are they firmware-family constants? If the
latter, the
SupportedDevicesgate inusb.pyis the only real per-device piece and00b7could simply be added to it — which is what this thread's patches already do.key from the sensor's TLS flash block with the PSK and completed the handshake against
the pairing Windows had created — no factory reset, nothing wiped. Worth an FAQ note:
"reset the chip" is the usual first advice for
00b7under Windows Hello, but it isn'talways needed and it irreversibly wipes stored prints.
Practical consequence: if !626 lands (with a published data package, which it currently
lacks),
06cb:00b7is supported in mainline libfprint/fprintd and this python routebecomes legacy — cross-linking the two, and closing #225 as upstream-tracked after the
merge, would save the next person a lot of patching.
🤖 — written by AI, validated by Dmytro
salomonsters commentedon Nov 15, 2026
For me, on Pop! OS, I did get it working like this, thank you very much!
One problem was that it would only work once , then the second time fprintd-verify would immediately exit with a no-match. Only way to get it working (once more, just once) was to restart the services.
After some extensive Gemini troubleshooting the solution seems to be to edit dbus-service and in the try: block of def run():, add a line
usb.cancel_event.clear()before
usrid, subtype, hsh = sensor.identify(update_cb)So it reads