Since Linux 4.5, and most certainly 25a84db15b3f3a24d3ea7d2baf90693bcff34b0c (HID: multitouch: enable palm rejection if device implements confidence usage), palm detection is done by the hardware, which is good news. However, after the palm is removed from the touchpad and a finger is put on the touchpad to move the pointer, the pointer first "jumps" from its previous location to another. This happens with both libinput and the synaptics module under Xorg, so I guess the issue is in the kernel (or touchpad firmware). Here's an evemu-record excerpt when a palm is detected: E: 6.801443 0003 0039 0258 # EV_ABS / ABS_MT_TRACKING_ID 258 E: 6.801443 0003 0035 0714 # EV_ABS / ABS_MT_POSITION_X 714 E: 6.801443 0003 0036 0391 # EV_ABS / ABS_MT_POSITION_Y 391 E: 6.801443 0001 014a 0001 # EV_KEY / BTN_TOUCH 1 E: 6.801443 0001 0145 0001 # EV_KEY / BTN_TOOL_FINGER 1 E: 6.801443 0003 0000 0714 # EV_ABS / ABS_X 714 E: 6.801443 0003 0001 0391 # EV_ABS / ABS_Y 391 E: 6.801443 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +5986ms E: 6.808003 0003 0036 0392 # EV_ABS / ABS_MT_POSITION_Y 392 E: 6.808003 0003 0001 0392 # EV_ABS / ABS_Y 392 E: 6.808003 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +7ms E: 6.814779 0003 0035 0713 # EV_ABS / ABS_MT_POSITION_X 713 E: 6.814779 0003 0000 0713 # EV_ABS / ABS_X 713 E: 6.814779 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +6ms E: 6.828478 0003 0035 0715 # EV_ABS / ABS_MT_POSITION_X 715 E: 6.828478 0003 0000 0715 # EV_ABS / ABS_X 715 E: 6.828478 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +14ms E: 6.835413 0003 0035 0717 # EV_ABS / ABS_MT_POSITION_X 717 E: 6.835413 0003 0036 0393 # EV_ABS / ABS_MT_POSITION_Y 393 E: 6.835413 0003 0000 0717 # EV_ABS / ABS_X 717 E: 6.835413 0003 0001 0393 # EV_ABS / ABS_Y 393 E: 6.835413 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +7ms E: 6.842195 0003 0035 0718 # EV_ABS / ABS_MT_POSITION_X 718 E: 6.842195 0003 0036 0395 # EV_ABS / ABS_MT_POSITION_Y 395 E: 6.842195 0003 0000 0718 # EV_ABS / ABS_X 718 E: 6.842195 0003 0001 0395 # EV_ABS / ABS_Y 395 E: 6.842195 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +7ms [palm moves on the touchpad, is filtered by the firmware] [there's no event when the palm is lifted up] [afterwards, I put my finger on the touchpad to make a move] E: 14.235784 0003 0035 0450 # EV_ABS / ABS_MT_POSITION_X 450 E: 14.235784 0003 0036 0396 # EV_ABS / ABS_MT_POSITION_Y 396 E: 14.235784 0003 0000 0450 # EV_ABS / ABS_X 450 E: 14.235784 0003 0001 0396 # EV_ABS / ABS_Y 396 E: 14.235784 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +7393ms E: 14.249453 0003 0035 0449 # EV_ABS / ABS_MT_POSITION_X 449 E: 14.249453 0003 0036 0397 # EV_ABS / ABS_MT_POSITION_Y 397 E: 14.249453 0003 0000 0449 # EV_ABS / ABS_X 449 E: 14.249453 0003 0001 0397 # EV_ABS / ABS_Y 397 E: 14.249453 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +14ms E: 14.290212 0003 0036 0398 # EV_ABS / ABS_MT_POSITION_Y 398 E: 14.290212 0003 0001 0398 # EV_ABS / ABS_Y 398 E: 14.290212 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +41ms E: 14.310613 0003 0036 0399 # EV_ABS / ABS_MT_POSITION_Y 399 E: 14.310613 0003 0001 0399 # EV_ABS / ABS_Y 399 E: 14.310613 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +20ms E: 14.330343 0003 0039 -001 # EV_ABS / ABS_MT_TRACKING_ID -1 E: 14.330343 0001 014a 0000 # EV_KEY / BTN_TOUCH 0 E: 14.330343 0001 0145 0000 # EV_KEY / BTN_TOOL_FINGER 0 E: 14.330343 0000 0000 0000 # ------------ SYN_REPORT (0) ---------- +20ms
A couple of precisions: the laptop model is 9350 (late 2015), and that issue doesn't happen under Windows 10.
Same laptop and same problem using Arch Linux and kernel 4.6rc2. The problem exists since early rc releases of 4.5.
Created attachment 220311 [details] Patch Hi Cyril, I am the author of 25a84db15b3f3a24d3ea7d2baf90693bcff34b0c. The attach is my patch to fix the bug you reported here. I have verified it on Dell XPS13. Could you help to verify the patch? If the result is positive, I will submit it to upstream. Allen
Hi Allen, Actually I cannot reproduce the issue on a vanilla kernel 4.6, so I guess something must have changed since 4.5? Do you still want me to test your patch on a 4.6?
Hi Cyril, I can still reproduce the issue on Ubuntu 16.04 with vanilla kernel 4.6.0 on XPS13. The file "drivers/hid/hid-multitouch.c" is where my patch applied on and it has no relevant changes between 4.5 and 4.6. Maybe the change has been made in some other where. The root cause is that the touch reports with (confidence=0) are directly dropped by hid-multitouch driver, without sending to upward layers. If there is a palm touch, says touch A, that is initially detected as a finger (confidence=1) and soon after is detected as a palm (conf=0) then the system is not aware of the latter change. The system still thinks there was a finger touch but then no reports were received. Once you make a new finger touch, says touch B, on a different position than where the palm has previously touched on, the system will treat touch A and touch B as in a continuous touch then the jumping occurs. As I saw the problem is still there in 4.6 I will try to get this fix upstream.
Dear Allen, as you can read from comment #2 in this report, I had these kind of problems with the XPS 13 (9350) touchpad since early rc releases of kernel 4.5. They are still present in kernel 4.6.2. As workaround I was used to switch on psmouse modules blacklisting i2c_hid one. This implied a different (less sensible) feedback but at least it worked. I'm using your patch on a 4.6.2 kernel built on my Arch Linux (Gnome desktop) with i2c_hid module. It is greatly working for me since at least 2 hours. I guess it fixes the regression introduced in kernel 4.5. Thank you for restoring the love for my touchpad! ;) Please, push this patch upstream ASAP.