Why does FreeRTOS (on Cortex-M) perform the switch in PendSV rather than directly in SysTick?

Because PendSV is set to the lowest exception priority, which guarantees the context switch happens only after every other pending interrupt has completed, you never switch tasks in the middle of servicing a higher-priority ISR, which would be unsafe and would add latency to that interrupt. SysTick (and other ISRs) just request a switch by setting PendSV pending; the core then tail-chains into PendSV once all higher-priority handlers have finished, and PendSV does the actual register save/restore. This decoupling matters for two reasons. First, correctness and latency: an interrupt that unblocks a high-priority task can request a switch, but the switch is deferred until the interrupt (and any nested ones) return, so interrupt servicing isn't delayed by the switch and the switch isn't interleaved with interrupt code. Second, it provides a single, well-defined place where switches occur regardless of what triggered them (the tick or any ISR), simplifying the port. The flip side is that task-switch latency includes any in-flight interrupt time, which is one more reason to keep ISRs short.