i.MX93: can a watchdog reset only the Cortex-A55 and leave the Cortex-M33 running? I am using a custom i.mx93 board but mostly follows evk's design and PMIC. On software, I am using NXP BSP-based Yocto (walnascar-6.12.34-2.1.0 branch). I want a watchdog timeout to reset only the Cortex-A55 and leave the Cortex-M33 (Zephyr) running, so the M33 can supervise the A55 and buffer data during the outage. I found in the reference manual that - Section 32.2.10 lists wdog4_ca55_rst_b and wdog5_ca55_rst_b as cold reset sources. - Section 32.2.12.2 says the CA55 cannot be warm reset and describes a power down, CPUWAIT, power up, release sequence. - The WDOG chapter 69 does not say what a timeout resets. I have dump the registers of SRC base 0x44460000 by using devmem2 +0x18 (reset mask): 0x00000FBF. Bit 6 (CSU) is clear, and the WDOG bits, including WDOG5 (bit 4), are set, so those sources are masked. Is it supported to unmask WDOG4 or WDOG5 and set its reset mode to the CA55 slices only, so the A55 resets while the M33 keeps running? If so, is there anything else required for the A55 to restart? Is there an application note or example for it? Thank you for any pointers in advance. Best Regards Linux Re: i.MX93: can a watchdog reset only the Cortex-A55 and leave the Cortex-M33 running? The SRC can route a WDOG4/5 reset to A55 domains without selecting the M33 domain, but that alone does not establish that the A55 will power-cycle, restart, and boot Linux while the M33 keeps running. The A55 restart sequence and any watchdog-triggered PMIC reset must also be accounted for. I could not find an NXP example implementing that complete flow for your BSP.
Also, your mask-bit reading appears off: in SRMASK , WDOG4 and WDOG5 are bits 19 and 20 , not bits 3 and 4. Recheck the register fields before changing them. Re: i.MX93: can a watchdog reset only the Cortex-A55 and leave the Cortex-M33 running? Thanks you for the quick reply. My SRMASK diagram (screenshot attached) has WDOG4 and WDOG5 at bits 3 and 4, and WDOG1 to WDOG3 at bits 0 to 2, with reset value 0xFBF. That matches what I read at boot. I am reading IMX93RM.pdf Rev. 7, 2026-02-10. Could you confirm which document and revision you are reading? My open question is the restart: after this reset, with CPUWAIT held and the reset vector set, what is the supported sequence to get the A55 running again? The RM says the CA55 cannot be warm reset and needs a power cycle. Is there a supported way to power-cycle it from the M33? Thank you. Best Regards Re: i.MX93: can a watchdog reset only the Cortex-A55 and leave the Cortex-M33 running? You are right about SRMASK . I misread the diagram’s bit-position labels.
For the restart, an A55 reset alone is not the documented sequence. Section 32.2.12.2 says the CA55 cannot be warm reset and specifies: power down the CA55; set its CA55_CPUWAIT[CPUn_WAIT] bit to 1; power the CA55 up; then release WAIT so the core can run. Setting the reset vector and holding CPUWAIT therefore do not, by themselves, perform the required power cycle.
There is a documented software mechanism to power-cycle SRC slices , potentially usable by M33 firmware: §32.2.8 describes enabling MEM_CTRL[MEM_LP_EN] , setting the slice’s SLICE_SW_CTRL[PDN_SOFT] to 1 for power-down, and clearing it to 0 for power-up. It permits software slice control only with AUTHEN_CTRL[LPM_MODE] = 0 ; the slice’s AUTHEN_CTRL[WHITE_LIST] also determines which TRDC domains may write its registers. The RM identifies separate A55 core and platform slices, so this must be applied to the appropriate CA55 power domains—not assumed to happen when a WDOG reset is routed to them.
Use the RM’s CA55 power-down → CPUWAIT → power-up → WAIT-release sequence; watchdog reset routing alone does not establish a complete A55 restart.
View full article