ESP_FAULT_ASSERT(C) was silently deleted by the optimizer when C is a cached
flag/status already proven by a preceding `if (!C) return/goto`: the compiler
folds C to a constant and drops all three checks, removing the fault-injection
protection with no warning.
ECDSA based Secure Boot V2 is not functional for certain input vectors on
ESP32-C5/C61/H2/P4 and on the preview targets ESP32-H4/H21. RSA based Secure
Boot V2 is the recommended scheme where the SoC supports it. This issue will be
fixed in a future hardware ECO revision; more details will be shared through the
hardware errata document.
A new hidden Kconfig option SECURE_BOOT_V2_ECDSA_INSECURE marks the affected
mass-production SoCs (ESP32-C5/C61/H2/P4). On these SoCs, when hardware Secure
Boot V2 is enabled, the ECDSA (V2) signing scheme is no longer offered by
default; it must be turned on explicitly via SECURE_BOOT_V2_FORCE_ENABLE_ECDSA
under "Allow potentially insecure options" (CONFIG_SECURE_BOOT_INSECURE). App
signing without hardware Secure Boot is not affected. Note that ESP32-C61 has no
RSA based Secure Boot V2, so it has no Secure Boot scheme enabled by default.
The preview targets ESP32-H4 and ESP32-H21 mark ECDSA Secure Boot V2 as not
supported in their SoC capabilities instead of using the option above. As
ESP32-H4 has no other Secure Boot V2 scheme, Secure Boot is disabled entirely on
it; ESP32-H21 retains RSA based Secure Boot V2.
The security documentation keeps the ECDSA Secure Boot V2 content visible and
adds a warning describing the limitation (including that ECDSA Secure Boot V2 on
ESP32-C61 is not recommended for production). CI apps that exercise ECDSA Secure
Boot V2 on the affected SoCs set CONFIG_SECURE_BOOT_V2_FORCE_ENABLE_ECDSA
accordingly.
- `bootloader_flash_execute_command_common`: whitelist the flash command
opcodes the REE actually uses; reject the rest
- `spi_flash_hal_* services`: a forged `host->driver` could hijack TEE
control flow since the HAL dispatches through it, so swap
`host->driver` to a TEE-rodata vtable around each HAL call
- Deny partition table and bootloader writes by default and permit
bootloader writes only when explicitly enabled via
`CONFIG_SPI_FLASH_DANGEROUS_WRITE_ALLOWED` option
- Protect the TEE-assigned interrupt pin configuration against REE
- Validate nested DS context pointers in start/finish_sign and bound
the result copy to the SoC max signature size
- Fix the stack usage in service dispatcher argument parsing
A typical scenario is: when XIP on PSRAM enabled, compiler optimization level is Os. Under certain binary layout, boot hangs and backtrace points to `esp_sleep_config_gpio_isolate`.
The root cause is that, during PSRAM initialization, it calls esp_gpio_reserve, which happens to place before the reported function. However, after call, there is no barrier before the clock adjustment in `mspi_timing_enter_low_speed_mode`. The clock gets changed when the cache is still fetching data, resulting in the corrupted data in the end of the cache line.
This commits add spi_flash_disable_cache as a barrier to make sure the cache transactions is finished before the clock switch.
TEE secure-service handlers had two bugs letting REE bypass
pointer-region validation:
1. valid_addr = instead of valid_addr &= in AEAD encrypt/decrypt
and DS sign handlers, clobbering prior failed checks.
Impact: REE writes to TEE DRAM via DS signature output, or reads
TEE DRAM via AEAD output.
2. data->rsa_length dereferenced before data is validated in DS sign
and DS start_sign handlers.
Fix: use &= for subsequent checks, add early return after initial
data pointer check in DS handlers.
- Using PMA, the TEE IRAM is marked as R/X while TEE DRAM is marked as R/W.
Moving the internal memory secure service call table from DRAM to IRAM
makes it immutable.
Currently, REE SPI flash HAL operations are routed as service calls to TEE,
but the TEE implementation incorrectly uses ROM APIs instead of HAL APIs.
This leads to issues and is not the recommended approach.
- Previously, only the U-mode interrupt threshold was raised in REE critical sections,
leaving M-mode at the lowest level.
- As a result, when a service call transitioned to M-mode, all interrupts were still
allowed to fire, including those that should have been masked.
- Ensure bound checks correctly handle all scenarios, including
when a requested operation's (SPI0/1) range fully contains the
TEE-protected region.
- Disable delegation of INTWDT timeout and Cache error interrupts as they reset
the device after the panic handler
- Fix incorrect setting in the edge interrupt acknowledgement API
- Avoid executing the service call dispatcher in the U-mode ecall,
rather execute `mret` to jump it
- Avoid `t1` register corruption when processing `ecall`
- Switch back to the bootloader stack from TEE stack after the
execution of the entire TEE initialization routine
- Rename `tee_test_fw` app configs for better CI tracking
- Decrease the lower bound of TEE I/DRAM config options
- Trim the TEE test-apps build
- Improve the TEE/REE OTA pytest script with additional checks
- Fix build issues when `tee_sec_storage`/`tee_ota_ops` are a
a part of the project build but ESP-TEE is disabled
Ashish Sharma
Laukik Hase
harshal.patil
Jiang Jiang Jian
Aditya Patwardhan
Alexey Lapshin
Xiao Xufeng
Fu Hanxi
Alexey Gerenkov
Mahavir Jain