CVE-2026-10644
MediumCVSS 4.2Exploitation Probability (EPSS)
Low risk4th percentile — higher than 4% of all known CVEs
Summary
The Microchip SERCOM-G1 UART driver for PIC32CM-JH SoCs has an out-of-bounds write vulnerability in the asynchronous (DMA) receive path. When uart_rx_enable() is called with a 1-byte receive buffer and CONFIG_UART_MCHP_ASYNC is enabled, the RX-complete ISR triggers a single-beat DMA transfer while a byte is already pending in the DATA register, causing a one-byte write past the buffer end (CWE-787).
Risk Assessment
The risk is limited single-byte memory corruption adjacent to the RX buffer, potentially causing a crash or denial of service. Exploitation requires adjacent network access (UART) and specific configuration (async UART, 1-byte buffer), limiting the attack surface.
Recommendation
Apply the patched driver version (v4.4.0+) immediately. As a workaround, disable CONFIG_UART_MCHP_ASYNC or avoid using 1-byte buffers with uart_rx_enable().
Original NVD description (English source)
The Microchip SERCOM-G1 UART driver (drivers/serial/uart_mchp_sercom_g1.c), used by the PIC32CM-JH SoC family, contains an out-of-bounds write in its asynchronous (DMA) receive path. When uart_rx_enable() is invoked with a one-byte receive buffer (len == 1) and CONFIG_UART_MCHP_ASYNC is enabled, the RX-complete ISR starts a single-beat DMA transfer while a received byte is already pending in the SERCOM DATA register. On this SoC the peripheral-triggered DMA start sequencing then writes one byte past the end of the caller-supplied buffer (CWE-787). The overflowed byte's value is the UART RX data supplied by the connected serial peer (adjacent attacker), while its size and location are fixed at one byte immediately after the buffer. Exploitation requires the async UART config (not enabled by default on the in-tree PIC32CM-JH boards) and a consumer that enables RX with a one-byte buffer; impact is limited single-byte memory corruption adjacent to the RX buffer (possible crash / denial of service). The defect shipped in v4.4.0. The fix reads the first byte with the CPU and, for one-byte buffers, performs no DMA at all; for larger buffers it sizes the DMA for the remaining len-1 bytes.