luke/esp32-poe-scrypted-viewport/main/display.c, branch main

main: code-review fixes — brightness units, OTA stall cap, stats + locking

2026-07-17T23:56:03+00:00

- display: s_last_pwm cached the raw 0-100 percentage at init but display_wake writes it straight to REG_PWM (0-255 duty) — first wake ran visibly dim until a /config brightness change. Shared pct_to_duty helper now converts in both paths. - http_api: cap consecutive OTA recv timeouts (a stalled client spun forever holding s_ota_in_progress, wedging OTA until reboot); cap viewport name at 54 chars so viewport- fits the 63-byte mDNS label; log mdns_service_refresh failures; /state builds JSON from a snapshot instead of holding the state lock across ~25 cJSON allocs; respond_400 delegates to respond_status. - stream_server: bytes_in_window now counts painted frames only (frames discarded while asleep inflated the first post-wake window's MB/s, avg-jpeg and chunk/wire averages; recv_bytes folded in); so_rcvbuf carried into the stats snapshot under the mux; drop dead HEADER_BYTES; merge read_body_instrumented into read_n. - state_machine: transition mutex serializes concurrent wake/sleep (display side-effects ran after the state lock dropped, so racing callers could leave the backlight contradicting st->state); wake-path placeholder paint now takes the decoder lock like the stream path (concurrent esp_lcd_panel_draw_bitmap from two tasks isn't safe). - local_screens: overlay paint takes the decoder lock too; check the overlay timer create; panel dims from viewport_state.h. - jpeg_decoder: try_lock before init returns busy instead of passing a NULL semaphore to xSemaphoreTake. - net_eth: clear cached IP string on link down (stale /state + info). - dead code: display_present_bgr888, TOUCH_FT5426_ADDR, touch s_task, JPEG_DECODER_MAX_OUTPUT_BYTES; doc drift in nvs_config.h / jpeg_decoder.h / app_main flag legend.

display: tear-free frame path via triple buffering + scan tracking

2026-07-16T00:10:57+00:00

draw_bitmap on a direct fb pointer only updates the driver's cur_fb_index; the DPI DMA reloads that index at the END of the in-progress frame scan (~21ms period at ~47Hz). Under double buffering, flipping and immediately decoding the next frame into the other fb writes a buffer the DMA may still be scanning out — a torn frame. This regime is common now that the TCP window fix delivers frames back-to-back (decode starts ~6ms after flip). Fix with zero added latency: num_fbs 2 -> 3 (+1.15MB PSRAM of 25MB free), track the actually-scanning fb via on_refresh_done (fires in the DMA-done ISR exactly when the DMA reloads cur_fb_index), and pick the decode target as the fb that is neither pending display nor scanning. Three buffers minus at most two excluded roles = always a free one; no waiting on vsync anywhere. Instrumented: /state tear_guard_engaged counts back-buffer picks made while the previous fb was still mid-scan — each one is a frame that would have torn under double buffering.

firmware: double-buffer the panel + zero-copy decode → ~22 fps ceiling

2026-06-15T12:50:23+00:00

Per-frame paint cost drops from ~24 ms to ~45 µs (≈500× faster) by enabling num_fbs=2 on the DPI panel and decoding straight into the back framebuffer. Measured on the bench: before: lock=7us ttfb=370us body=40ms dec=6ms paint=24ms post=35us = ~70ms / ~14fps after : lock=7us ttfb=330us body=38ms dec=6ms paint=42us post=25us = ~45ms / ~22fps How the win actually lands: - num_fbs=2 in the esp_lcd_dpi_panel_config_t makes the IDF driver allocate two framebuffers and stream from one while we fill the other. - display_back_buffer() returns the inactive fb pointer + its size. - jpeg_decoder_decode() now accepts a caller-provided destination buffer instead of owning its own scratch. http_api passes the panel back-fb so the hardware JPEG decoder writes BGR888 pixels straight into where the DSI will eventually scan from. Zero memcpy in the hot path. - display_flip_back_buffer() calls esp_lcd_panel_draw_bitmap with the fb pointer. Because the buffer is inside the panel's own fb range, the IDF driver skips its memcpy and just does a cache writeback + swaps cur_fb_index. The actual flip happens on the next vsync, asynchronously — the call returns in microseconds. The remaining ceiling is network body time (~38 ms for ~210 KB JPEGs) and the hardware decoder (~6 ms). Per-viewport JPEG quality (smaller files = shorter body) is the next lever; everything firmware-side is already at or near floor. Also drop the old static jpeg output scratch + JPEG_DECODER_MAX_OUTPUT_BYTES constant — nothing references them anymore.

firmware: zero-copy JPEG → BGR888 → DSI hot path; Scrypted pre-rotates

2026-06-15T12:23:47+00:00

Architectural rework of the /frame hot path. Scrypted now ships every JPEG already scaled + rotated to the panel's native dimensions (read from the firmware's /state response so nothing is hardcoded on the Scrypted side); the firmware decodes the JPEG straight into a BGR888 buffer that's directly draw_bitmap'able by the DSI driver, with zero CPU pixel work and zero rotation work in between. Firmware - jpeg_decoder now uses JPEG_DECODE_OUT_FORMAT_RGB888 + JPEG_DEC_RGB_ELEMENT_ORDER_BGR. Output buffer sized for 800*480*3. - New display_present_bgr888() is a one-liner that hands the decoder's output straight to esp_lcd_panel_draw_bitmap. - /frame handler validates dimensions against the panel-native VIEWPORT_PANEL_WIDTH x VIEWPORT_PANEL_HEIGHT (was effective_dims branching on orientation). Returns 400 if it's anything else. - /state JSON adds panel_width + panel_height so Scrypted can read them without hardcoding board-specific knowledge. - display_present_rgb565 + s_rot_buf stay for the local-screens cold path (info screen, loading) which still does its own CPU conversion + rotation — infrequent enough that it's not worth the rewrite. Scrypted - startStream() GETs /state at stream-start time, caches panel_width and panel_height in viewport storage, and uses them as the ffmpeg scale target. Falls back to cached or 800x480 if /state is mid-reboot. - For portrait viewports the ffmpeg pipeline now does scale=H:W:flags=lanczos,transpose=1 so the JPEG arrives pre-rotated 90° CW into panel-native dimensions. Landscape is just scale=W:H. - No more in-firmware rotation; Scrypted is the single source of truth for "how do I get this camera frame into a panel-shaped JPEG". Expected ceiling lift: ~5 fps → ~10 fps, gated by the JPEG decoder hardware throughput instead of the CPU rgb565→bgr888 + rotation loop.

Default name = MAC; brief wake on boot; drop dead touch defences

2026-06-15T01:11:21+00:00

- Seed viewport_name with the full base MAC, colons stripped (e8:f6:0a:e0:90:94 → "e8f60ae09094"). Stable across reboots, globally unique, 12 alphanumeric chars — well inside the mDNS hostname limit even with the "viewport-" prefix. - Add a mac_str field to viewport_state and expose it as the new "mac" key in GET /state and as a "mac" line on the info screen. - "Configured" no longer requires a viewport_name (it always has the MAC default); the scrypted URL alone gates outbound POSTs and the loading-vs-info screen choice. - Strip viewport_name[0] fallbacks in http_api / mdns / local_screens now that the field is never empty. - Replace the wake-on-boot-stays-on behaviour with a ~600 ms flash followed by sleep — long enough for the FT5426 touch IC to come out of its initial unresponsive state, short enough that an idle device doesn't burn the backlight. - Drop the touch defensive cruft added when we were chasing a PoE-power theory: the DEV_MODE=0x00 write at init, the touch_reset_pulse() call at init, and the runtime wedge-self-heal in the polling loop didn't actually fix anything — the wake-on-boot flash did. Also delete the display_touch_reset_pulse() helper since it now has zero callers.

touch: retry + hard-reset when FT5426 boots into wedged 0xff state

2026-06-15T00:43:13+00:00

After a long bench session of taps + power cycles, the FT5426 on this unit started returning dev_mode=0xff on every read. The polling loop interpreted that as 15 simultaneous touches forever (td_status bits 0..3 all set), wedged `was_down=true` permanently, and no tap or long-press ever fired — the screen looked dead. Two layers of defence in touch_init: 1. retry the dev_mode read up to 10 × 30 ms in case the chip is just slow to come up after PC_RST_TP_N is released. 2. if still stuck, pulse PC_RST_TP_N via a new display_touch_reset_pulse() helper (drop reset for 20 ms then release + 50 ms settle), then retry the read another 10 × 30 ms. Plus a polling-loop sanity check: an FT5x06 supports max 5 simultaneous points, so a TD_STATUS lower-nibble value > 5 is bogus and the poll cycle is skipped. That keeps a wedged chip from producing phantom "15 touches" reports if it slips into 0xff during runtime. If both attempts still report 0xff, we log loudly and let the polling task run — it harmlessly drops cycles via the >5 sanity check. Wake still works via POST /state from Scrypted; this only disables touch gestures on a hardware-stuck boot.

Tidy: delete dead code, inline single-use helpers, fix double-lock

2026-06-14T23:30:44+00:00

Code-review pass after the M5 colour fix. -126 net lines (-215 / +89). Dead code removed: - display_fill, display_test_pattern — M3 bring-up self-tests, no callers - net_eth_is_up — no callers - nvs_config_reset — no callers (factory-reset gesture was removed earlier; NVS wipe now goes through `idf.py erase-flash`) Helpers inlined or collapsed (each had one call site): - state_name / orientation_name — ternaries at the cJSON site - fmt_bytes / fmt_uptime — inline scope in local_screens_show_info - expected_dims (http_api) + effective_dims (local_screens) — merged into viewport_state_effective_dims() in viewport_state, one canonical source state_machine cleanup: - arm_idle_timer_unlocked / disarm_idle_timer collapsed to one arm_idle_timer(ms) that takes the snapshotted timeout (ms==0 disables). Removes the misleading name and the second viewport_state lock acquisition per painted frame. - state_machine_set + state_machine_frame_painted now snapshot state + idle_ms under one lock. mdns_service cleanup: - snapshot_state / apply_hostname / apply_txt collapsed into a single apply_state(include_hostname) helper; mdns_service_start grabs hostname once for hostname_set + log instead of going under the lock three times. http_api cleanup: - POST /config brightness-changed path uses the local `bright` it already validated instead of re-locking + re-reading st->brightness. - Trimmed verbose bring-up comments (stack-size justification, M5 saga, DSI shadow struct) to one line each — kept the load-bearing facts. Smoke-tested on hardware: boot clean, display + JPEG + touch all up ([DJT]), info screen renders, no crashes.

M5 colors fixed: JPEG byte order + DPI memory layout

2026-06-14T23:08:50+00:00

Two channel-swap bugs were stacked, hiding each other on symmetric pixels (white text on black background) and showing up only on saturated colour JPEGs: 1. jpeg_decoder: rgb_order was JPEG_DEC_RGB_ELEMENT_ORDER_RGB which empirically emits the RGB565 word in big-endian byte order (the IDF header comment "color component in big endian" is correct; the enum's "RGB" name is misleading). Our painter reads native LE uint16_t, so the bytes were always swapped. Flipped to _BGR which emits LE — verified by dumping decoded bytes for solid red/green/blue JPEGs. 2. display: rgb565_to_rgb888 wrote bytes [R, G, B], but the ESP32-P4 DSI engine + TC358762 + Pi panel expect [B, G, R] in memory despite the "RGB888" label. (MIPI DSI defines bit order, not byte position in memory.) Verified with an embedded solid-blue test JPEG: with [R, G, B] the panel showed solid red; with [B, G, R] it shows solid blue. Diagnosed by embedding a 480x800 blue test JPEG in firmware, decoding + dumping the buffer at boot, and reading the actual bytes over USB serial (ethernet was disconnected mid-investigation). The temporary self-test + debug dump are removed; the fixes themselves are tiny.

M3: panel renders — TC358762 bridge init + non-burst DSI

2026-06-14T21:36:46+00:00

The Hosyond 5" panel uses Pi 7" v1.1 architecture (ATTINY ID 0xC3 + TC358762 DSI-to-DPI bridge). Getting it to paint needed three things that ESP-IDF doesn't expose: 1. TC358762 bridge configuration — 16 register writes via DSI Generic Long Write (DT=0x29) packets, transcribed from Linux's tc358762.c. 2. Non-burst video mode — IDF hardcodes BURST_WITH_SYNC_PULSES, but TC358762 requires NON_BURST_WITH_SYNC_PULSES. Overridden via the mipi_dsi_host_ll_* API after esp_lcd_new_panel_dpi() and before panel_init(). 3. ATTINY v1.1 power-on sequence + SPI-proxy bridge wake. The legacy REG_POWERON(0x85) path doesn't apply to 0xC3 firmware; instead write PORTC/PORTA/PORTB/PORTC in order, then later release bridge reset and proxy-write TC358762 SYSPMCTRL=0 through the ATTINY's ADDR_H/L + WR_DATA_H/L registers. Reaching the LL/HAL APIs requires shadowing esp_lcd_dsi_bus_t so we can pick the mipi_dsi_hal_context_t out of the private struct. Documented the layout dependency at the shadow definition. Tuned config (observed stable on ESP32-P4 per embenix's reference): - 1 data lane @ 600 Mbps - DPI 26 MHz (Linux modeline is 25.98) - timings HSW=2 HBP=46 HFP=210 / VSW=20 VBP=4 VFP=22 - RGB888 end-to-end, R,G,B byte order (BGR was wrong) - disable_lp=0 so LP windows are available for the bridge writes

M3 WIP: DSI bring-up + state machine + identity overlay

2026-06-14T20:53:33+00:00

DSI / panel: - LDO_VO3 acquired at 2500 mV (VDD_MIPI_DPHY) — without this the PHY PLL busy-waits forever inside esp_lcd_new_dsi_bus. - PSRAM bumped to 200 MHz (CONFIG_IDF_EXPERIMENTAL_FEATURES) to keep the DPI fed without underruns. - Pi-7"-style 800x480 panel init (REG_POWERON, PORTA/PORTB/PWM/PORTC) + 16-bit RGB565 framebuffer with portrait-mode rotation buffer in PSRAM. Currently shows garbled output — known issue, next commit switches to the TC358762-bridge-aware init sequence. State + UX: - viewport_state simplified to AWAKE/ASLEEP only; "unconfigured" is a flag, not a state. Tap always toggles; content choice (identity vs frame) is driven by the configured flag. - BOOT button arms a 15 s identity overlay via local_screens_overlay; expired callback returns to prior state. - Boot-done indicator: two backlight flashes (no usable LED GPIO on the Waveshare board — GPIO 35 BOOT is shared with EMAC TXD1). - Best-effort subsystem init in app_main; display init deferred to its own task so a panel hang can't block networking. Display test pattern: solid R/G/B for 2 s each on boot to characterize the garble independent of text rendering.