igorsh/kolibrios
1
0 Fork 0
forked from KolibriOS/kolibrios
Code Pull Requests Activity

drivers/i2chid: reset a pad stuck on one frame, bring-up ladder only in debug builds

Browse Source 
Assisted-by: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Leency authored and Burer committed 2026-09-24 10:01:16 +00:00
1 parent 873d4de379
commit 02916ed0ee
1 file changed
+91 -3
Download Patch File Download Diff File Expand all files Collapse all files
drivers/i2chid/i2chid.asm
+91 -3
View File
@@ -72,9 +72,26 @@ I2CHID_PWR_ON = 0x00
; single tick forever.
MAX_POLL_ERRORS = 100
; Polls of silence before the next recovery step is tried (20 ms each).
; Polls of silence before the next recovery step is tried (10 ms each).
; The ladder exists for bringing up a pad that never reports at all, and is
; compiled in only at debug level 1: on a working pad it would run merely
; because nobody touched it during the first seconds after boot, and the
; power cycle, the second reset and the poll-mode flips it consists of are
; exactly the kind of thing a fragile firmware answers with silence.
SILENT_LIMIT = 150
; A pad that has ever answered an idle poll with an empty length, yet keeps
; returning one and the same non-empty frame, has stopped producing frames:
; a finger on the surface changes the scan time in every report, and a pad
; that holds its last frame instead of answering empty (Elan) never counts
; here. Seen on the SIPODEV SP1064: a single frame came through and was
; then repeated on every poll for good. This many identical answers in a
; row, and the device is reset and switched into touchpad mode again; the
; limit doubles on each reset, up to the maximum, so a pad that stays wedged
; is retried, not hammered.
STUCK_DUPS = 100
STUCK_DUPS_MAX = 3000
; ---------------------------- Settings -------------------------------------
; Everything below has a built-in default and can be overridden from the
; [touchpad] section of /sys/settings/system.ini. The file is read once at
@@ -1686,6 +1703,56 @@ proc set_ptp_mode uses ebx esi edi
ret
endp
; The pad has wedged (see STUCK_DUPS): drop whatever touch the driver holds,
; then bring the device up again the way setup_device did - RESET, a quiet
; wait, one read to consume the ack, SET_POWER(ON), and the touchpad mode,
; which the reset cleared. Runs in the poll thread.
proc reinit_device uses ebx esi edi
inc [StReinit]
DEBUGF 2, "i2chid: pad stuck: the same frame %u polls in a row, resetting it (%u so far)\n", [DupRun], [StReinit]
mov [DupRun], 0
mov eax, [StuckLimit]
add eax, eax
cmp eax, STUCK_DUPS_MAX
jbe @f
mov eax, STUCK_DUPS_MAX
@@:
mov [StuckLimit], eax
; Whatever the driver was in the middle of is over: no finger, no button.
call ptp_forget
mov [PrevTip], 0
mov [BtnLatch], 0
mov [HeldBtn], 0
mov [TapHold], 0
mov [TapRelease], 0
mov [PendingPress], 0
mov [DragActive], 0
invoke SetMouseData, 0, 0, 0, 0, 0
stdcall hid_command, 0, I2CHID_OP_RESET
test eax, eax
jz @f
DEBUGF 2, "i2chid: RESET failed (%x)\n", eax
@@:
mov esi, 150
invoke Sleep
stdcall dw_read_block, [SlaveAddr], 0, 0, input_buf, [MaxInput]
stdcall hid_command, I2CHID_PWR_ON, I2CHID_OP_SET_POWER
test eax, eax
jz @f
DEBUGF 2, "i2chid: post-reset SET_POWER failed (%x)\n", eax
@@:
mov esi, 10
invoke Sleep
call set_ptp_mode
mov [PollMode], 0
; The held frame is stale now; the next one the pad sends must count as new.
mov edi, prev_rep
mov ecx, MAX_INPUT_BUF
xor eax, eax
rep stosb
ret
endp
; Asks the device for every report it declared, as both an Input and a
; Feature report, and logs each answer. A pad whose sensor is alive but
; whose reporting mode was never selected will answer at least one of
@@ -1883,8 +1950,8 @@ proc i2chid_thread
test edx, edx
jnz @f
.stats:
DEBUGF 2, "i2chid: after %u polls: %u reads, %u reports, %u duplicates, %u empty, at most %u per poll, %u lifts by silence\n", \
[StPolls], [StReads], [StReports], [StDups], [StEmpty], [StMaxDrain], [StForced]
DEBUGF 2, "i2chid: after %u polls: %u reads, %u reports, %u duplicates, %u empty, at most %u per poll, %u lifts by silence, %u resets\n", \
[StPolls], [StReads], [StReports], [StDups], [StEmpty], [StMaxDrain], [StForced], [StReinit]
@@:
; Syscall 5 is the delay that sleeps on an event and yields; the Sleep import
; drivers normally use is delay_ms, which busy-waits and would burn a core.
@@ -1965,6 +2032,8 @@ proc i2chid_poll_once uses ebx esi edi
mov [GotAny], 1
mov [Consumed], 1
inc [StReports]
mov [DupRun], 0
mov [StuckLimit], STUCK_DUPS
; The first few reports are dumped raw, before any interpretation, so that a
; touchpad which does send data but whose bytes the driver reads wrongly can
; be told apart from one that says nothing at all.
@@ -2068,9 +2137,17 @@ proc i2chid_poll_once uses ebx esi edi
.duplicate:
pop eax
inc [StDups]
cmp [StEmpty], 0
je .done ; holding the last frame is this pad's idle
inc [DupRun]
mov eax, [DupRun]
cmp eax, [StuckLimit]
jb .done
call reinit_device
jmp .done
.silent:
inc [StEmpty]
mov [DupRun], 0
; Nothing arrived. An idle answer is still evidence: if the device reacts
; to a finger at all, something in those bytes changes even when it never
; sets a length, so the first answers and every change are logged.
@@ -2080,6 +2157,8 @@ proc i2chid_poll_once uses ebx esi edi
; If any of them wakes the device, GotAny freezes the ladder for good.
cmp [GotAny], 0
jnz .done
cmp [LadderOn], 0
jz .done
inc [SilentPolls]
cmp [SilentPolls], SILENT_LIMIT
jb .done
@@ -2759,6 +2838,15 @@ TraceCount dd 8
DecodeTrace dd 8
GotAny dd 0
SilentPolls dd 0
DupRun dd 0
; The bring-up ladder runs only in a level-1 (chatty) build.
if __DEBUG_LEVEL__ > 1
LadderOn dd 0
else
LadderOn dd 1
end if
StuckLimit dd STUCK_DUPS
StReinit dd 0
; Parser state (see hidmin.inc).
p_usage_page dd 0