[salikisazazil]

Target: BruceDevices/firmware (post-1.17)
Scope: 1 file changed, ~10 lines (src/core/radio_mem.h)
Related: #3078 (T-Embed CC1101 Plus: all BLE features fail right after boot)

---

Summary

Since 1.17, every BLE feature (BLE Suite, BLE Scan, BLE Spam, Ducky BLE)
fails immediately after boot on boards other than Cardputer with:

Low RAM: free WiFi/SD first
Failed to init BLE

even when WiFi is off. The error text is misleading: the gate is not about
WiFi being active, it is radioHasMemForBle() in src/core/radio_mem.h,
which requires the largest contiguous DMA-capable internal RAM block
to be at least RADIO_BLE_MIN_DMA_BLOCK before the BLE stack is touched.

Root cause

Two 1.17 changes interact:

1. Commit ec014ad1 ("fix ble cardputer") raised RADIO_BLE_MIN_DMA_BLOCK
   from 15 kB to 48 kB. That number was measured on Cardputer ADV,
   where the new SDR feature reserves a 64 KiB SRAM bank out of the heap at
   boot ("restrict SDR"), leaving only ~15.8 kB contiguous at the menu. The
   same commit added an SDR bank hand-off (sdrBankReleaseForRadio())
   that returns those 64 KiB to the heap when the radio needs them — this
   is what makes a 48 kB gate reachable on Cardputer.

2. Commit 3e9bc11f ("restrict SDR") added -D BRUCE_NO_SDR_JAMMER=1 to
   the global build_flags in platformio.ini, so the SDR bank (and
   therefore the hand-off) only exists on the Cardputer env, whose ini
   removes the define via build_unflags. On every other board
   (T-Embed, T-Deck, T-Stick, ...) the hand-off functions compile to no-ops.

Net effect: all non-Cardputer boards now require a 48 kB contiguous DMA
block with no mechanism to ever satisfy it, where 1.16 required 15 kB
(the value BLE shipped with and worked with everywhere). On T-Embed the
menu heap never reaches 48 kB contiguous, so BLE is dead from boot —
matching #3078.

Serial confirmation on an affected board (1.17):

[RAM] Low contiguous DMA memory for BLE, attempting to free WiFi...
[RAM] Still only <N> bytes DMA block, minimum 49152 needed

Fix

Make the BLE gate conditional on whether the SDR bank is actually
reserved out of the heap — the only configuration the 48 kB number was
calibrated for, and the only one where the hand-off can satisfy it:

#if !defined(LITE_VERSION) && !defined(BRUCE_NO_SDR_JAMMER)
constexpr size_t RADIO_BLE_MIN_DMA_BLOCK = 48 * 1024;
#else
constexpr size_t RADIO_BLE_MIN_DMA_BLOCK = RADIO_WIFI_MIN_DMA_BLOCK; // 15 kB, pre-1.17 gate
#endif

• The #if guard is deliberately identical to the guard that makes
  sdrBankReleaseForRadio() real (a few lines below), so the threshold
  and the hand-off can never disagree.
• Cardputer / Cardputer ADV (bank reserved): unchanged, still 48 kB +
  hand-off.
• LITE builds: unchanged vs 1.16 (they never had the bank).
• All other boards: exact 1.16 behavior restored (15 kB gate, same as the
  Wi-Fi threshold, which was never raised).
• WiFi path untouched (RADIO_WIFI_MIN_DMA_BLOCK unchanged).

Why this is safe

• Restores the exact gate 1.16 shipped and was field-tested with on all
  boards; the failure mode it re-introduces (BLE init aborting cleanly
  with an error instead of a half-initialized controller crash) is kept.
• Single file, no new code paths, no API changes, no behavior change for
  the boards the 48 kB gate was designed for.
• All four BLE call sites (BLE_Suite.cpp, ble_common.cpp,
  ble_spam.cpp, ducky_typer.cpp) go through radioHasMemForBle(),
  so one constant fixes all of them.

Test plan

1. pio run -e lilygo-t-embed-cc1101 (and one more non-Cardputer S3 env,
   e.g. lilygo-t-deck) — build must succeed (threshold is a compile-time
   constant, no runtime code change for these boards).
2. Flash 1.17 + this patch on T-Embed CC1101 Plus:
   • Boot → Bluetooth → BLE Suite: menu opens without the "Low RAM" error.
   • Quick Vulnerability Scan / target scan: no "Failed to init BLE".
   • BLE Scan, BLE Spam, BLE Keyboard: work.
   • Serial shows the old 15 kB minimum in the failure path
     (minimum 15360 needed) and no [RAM] Still only ... line at boot.
3. Cardputer regression (bank reserved, hand-off active):
   • BLE still starts (48 kB gate + hand-off intact).
   • SDR menu still present; if SDR was used, BLE refuses cleanly with
     "Low RAM: free WiFi/SD first" until reboot (unchanged).
4. LITE build (env_light): BLE behaves as in 1.16.

Commit message suggestion

fix(ble): restore BLE bring-up on boards without the SDR bank

1.17 raised RADIO_BLE_MIN_DMA_BLOCK from 15 kB to 48 kB, a value
measured on Cardputer where the SDR feature reserves a 64 KiB SRAM bank
out of the heap and the new hand-off releases it when a radio needs it.
"restrict SDR" then made the bank (and the hand-off) Cardputer-only,
leaving every other board with a 48 kB contiguous-DMA requirement that
no mechanism can satisfy; BLE fails right after boot on T-Embed etc.
even with WiFi off (#3078).

Keep the 48 kB gate only where the bank is reserved (same #if guard as
the hand-off itself) and fall back to the pre-1.17 gate elsewhere.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[derangedabbot]

Confirmed on LilyGO T-Embed CC1101 Plus (nRF24 menu present), built from tag 1.17 with this patch applied as-is:

• Before: every BLE entry (BLE Suite, Scan, Spam) → Low RAM: free WiFi/SD first immediately after boot, with Wi-Fi off and the SD card removed. free reported ~119 KB free heap and ~8 MB free PSRAM, so the 48 KB contiguous DMA gate was the only thing blocking init.
• After: BLE Suite opens and runs, no error. BLE Scan and Spam also start normally.
• T-Embed inherits the global -D BRUCE_NO_SDR_JAMMER=1 (only the Cardputer ini build_unflags it), so the #else branch (15 KB, the pre-1.17 gate) is what takes effect on this board — consistent with the root-cause analysis above.
• Build: pio run -e lilygo-t-embed-cc1101 -t build-firmware, PlatformIO 6.2.0, merged bin flashed at 0x0 over USB-Serial/JTAG.

Fixes #3078 for this board.