Time: SysTick, timers, PWM
Overview
This station has so far landed the first five pieces of the time theme. Piece one takes the TIM family as its main line: tim.hpp's three-piece set (TimBase period beats, TimOneShot one-shot windows, TimPwm duty-cycle output) has a consteval solver derive PSC and ARR from the period, the APB1×2 rule for timclk and the defense against the spurious UIF are wrapped up in two examples, 09_tim_heartbeat pairs a 1Hz heartbeat with a WFI main loop, and 10_button_confirm uses an EXTI edge to start a 20ms one-shot confirmation window, checked head-to-head against the polling debounce chain. Piece two puts configuration into the type: designated initializers used directly as template arguments, typos reported at compile time, the configuration folded into immediates in the instruction stream, viewed side by side with the HAL's handle route. Piece three computes the range: at 64MHz, a 16-bit PSC times a 16-bit ARR covers periods from 1µs to 67.108864s, the approximate path's error is capped at 15ppm, and what the 0.1% precision gate actually stops are configurations outside the range. Piece four dissects NVIC preemption priorities: tim.hpp's CMSIS single-argument form and the HAL two-argument form of exti and uart coexist in one firmware, with the library's stance giving all 16 levels to preemption; the Renode three-act probe pastes higher-interrupts nesting, same-level suspended catch-up, IRQ-number ordering, and an out-of-bounds 16 rolling back into the highest level, the grouping ladder is recomputed group by group from the NVIC_EncodePriority source, and SysTick parked by the HAL at the lowest 0x0F leads into that wipeout of an ISR busy-waiting on the beat and locking itself up. Piece five dissects the SysTick time base: the HAL's 1ms has HAL_InitTick compute the reload value on the spot from SystemCoreClock (64000@64MHz on this board, CLKSOURCE=1 on the processor clock), and not one place in the whole HAL Src tree reads SysTick->CALIB; CALIB's TENMS is by architecture definition a 10ms reload value, yet the 9000 ST loads on F1 computes to 1ms at 9MHz, and the Renode four-act probe pastes four log segments — the boot snapshot's TENMS=640000, a write of CLKSOURCE=0 bounced back to a read of 7, and 9000 carried to 64MHz where one beat shrinks to 0.140625ms — with the hand computation closing out the real-hardware deduction of 1.125ms and the computed right answer. The PWM breathing light waits until the material matures.