ESP32-S2 disables the brownout detector before deep sleep; if sleep is
rejected, re-init it so BOD is not left disabled (PM-519).
Co-authored-by: Cursor <cursoragent@cursor.com>
A blocking transport read can return zero when no data arrives before
timeout. Passing that zero length to http_parser_execute signals EOF while
a response is incomplete, puts the parser in HPE_INVALID_EOF_STATE, and
causes later response bytes to be discarded. The shortened response can
then be treated as successful.
Skip parser execution for every zero-length transport read, not only async
reads. Also compare the raw esp_transport_read result against raw
ERR_TCP_TRANSPORT values so timeout and peer-close failures retain their
documented HTTP error classifications.
A standalone reproduction built with the unmodified HTTP parser showed the
blocking timeout transition to HPE_INVALID_EOF_STATE and loss of the
remaining 15 bytes. Skipping the zero-length parser call delivered the full
chunk and message-complete callback.
Disclosure: this fix was prepared with AI assistance (Claude) and reviewed by me before submission.
Constraint: esp_http_client_get_data returns raw transport result values before esp_transport_translate_error.
Rejected: Keep the async-only zero-length guard | blocking transport reads also return zero on timeout.
Confidence: high
Scope-risk: moderate
Directive: Do not pass a transient zero-length transport read to the HTTP parser as EOF.
Tested: standalone blocking mid-chunk timeout reproduction; source-only duplicate-comment cleanup; git diff --check
Not-tested: hardware TLS transport integration
Signed-off-by: yi chen <94xhn1@gmail.com>
These per-event traces fire on every async dispatch and flood BT logs
without aiding diagnosis; higher-layer traces remain for debugging.
(cherry picked from commit b9b9b8633f)
Co-authored-by: zhanghaipeng <zhanghaipeng@espressif.com>
When the bond list is full, drop the oldest disconnected device instead
of the oldest NVS entry, and allow a per-bond except flag so selected
devices are never auto-removed.
(cherry picked from commit bf86892eef)
Co-authored-by: zhanghaipeng <zhanghaipeng@espressif.com>
A peripheral's reset also resets the ones it occupies, so a lock has to cover
both. Gate the ECDSA MPI lock on SOC_ECDSA_USES_MPI rather than the runtime
ecdsa_ll_is_mpi_required() and set that capability on C5, lock the Key Manager
path in esp_key_mgr.c, clean HMAC after its reset, and enable DS before the
primitives its reset covers.
ECDSA enable pulses a reset that also holds SHA in reset, and SHA shares
its DMA with AES. Key Manager enable pulses a reset that also covers the
XTS-AES flash encryption key-usage selector. Neither path was serialized
against those victims, so a hardware ECDSA/HMAC/DS operation could
corrupt a concurrent SHA/AES transfer or an in-flight encrypted flash
read.
- Take the SHA/AES lock inside esp_crypto_ecdsa_lock_acquire(), before
MPI, matching the DS lock order (sha_aes < mpi)
- Add esp_crypto_key_mgr_enable_periph_clk_no_reset() and switch ECDSA,
HMAC and DS to it; they only need the key-usage selector writable
- Hold esp_crypto_key_manager_lock across those clock enable/disable
pairs so selector writes stay serialized without resetting KM
The loops waiting for the card to leave its busy state started their yield
backoff at 100 ms. A card is typically busy for a few milliseconds after a
write, so the backoff never fired and every write was followed by hundreds
of back-to-back CMD13 commands. Occupying the host controller like this
slows down unrelated work on both cores, not just the calling task.
Delay between polls instead, starting at CONFIG_SD_READY_POLL_PERIOD_START_US
(100 us) and doubling. Both the delay and the configured start period are
capped at one FreeRTOS tick period, where vTaskDelay() already yields and one
command per tick is not a storm. A typical wait now costs a handful of
commands instead of hundreds.
Applies to sdmmc_wait_for_idle(), sdmmc_init_sd_wait_data_ready() and
read_tuning_block().
Closes https://github.com/espressif/esp-idf/issues/19034
Restore GATT_NO_RESOURCES when sequential attributes have differing UUID
sizes so gatts_process_find_info() stops building the response PDU.
(cherry picked from commit 196cb39545)
Co-authored-by: zhanghaipeng <zhanghaipeng@espressif.com>
Add smp_repairing_is_allowed() behind BT_BLE_SMP_HARDENED_REPAIRING so a
peer cannot replace an existing bond with one that has less MITM
protection, no Secure Connections, or a shorter key. Compare a preceding
Security Request against the pairing command AuthReq, not the
association-model result, and always allow first pairing.
A refusal keeps the stored bond. Pairing-failure erase is split by link
role: default is erase as Central and keep as Peripheral.
Closes BLERP (NDSS 2026) V3, V4 and V6.
(cherry picked from commit 88ea45be73)
Co-authored-by: zhiweijian <zhiweijian@espressif.com>
Keep the existing bond until the new pairing is encrypted, and on encryption
failure drop the link instead of clearing keys. Recovering from a peer that
really deleted the bond is opt-in through BT_BLE_SMP_UNBOND_ON_KEY_MISSING.
Closes BLERP (NDSS 2026) V5, and stops an unauthenticated Pairing Request
from dropping the stored keys (V2 exploitation).
(cherry picked from commit f864615d7d)
Co-authored-by: zhiweijian <zhiweijian@espressif.com>
The Wi-Fi task deletes itself when esp_wifi_deinit() is called, and FreeRTOS
only reclaims its TCB and stack from the idle task afterwards. Test apps that
read the heap right after deinit therefore see that memory as still allocated
and report a leak, most visibly as the eloop unit tests failing on ESP32.
Wait for the idle task at every point where a test deinitialises Wi-Fi
before a leak check, replacing the single-tick delays that only matched what
esp_wifi_deinit() already waits for internally.
(cherry picked from commit bff264775a)
Low-RSSI roaming used determine_best_ap(0), so a 1 dB better AP was
enough and nearby APs could ping-pong. Add ESP_WIFI_ROAMING_LOW_RSSI_ROAM_DIFF
(default 5 dB) as hysteresis for that path.
zwx
yinqingzhao
wuzhenghuiandCursor
Astha Verma
yi chen
Zhang Hai Peng
Jin Cheng
Rahul Tank
Jiang Jiang Jian
harshal.patil
radek.tandler
Shen Wei Long
Martin Vychodil
morris
liuning
Island
Adam Múdry
cjin
Zhi Wei Jian
Nachiket Kukade
Laukik Hase
Sarvesh Bodakhe
Chen Jichang
Wang Meng Yang
tarun.kumar