arch/riscv: Defer locking of interrupt stack guard PMP entry

The PMP entry for the interrupt stack guard is initially configured in
`z_riscv_pmp_init()`. This entry was previously locked immediately using
the `PMP_L` (Lock) flag.

This immediate locking causes issues in complex boot scenarios, such as
systems that jump from a Read-Only (RO) image to a Read-Write (RW) image.
When switching images, the kernel must re-initialize PMP. If the
interrupt stack address, calculated as `(uintptr_t)z_interrupt_stacks[_current_cpu->id]`,
changes between the RO and RW stages, the already-locked PMP entry for the
original address cannot be reconfigured, leading to an incorrect or
inoperable memory protection setup.

This commit removes the premature `PMP_L` flag from the PMP entry setup
in `z_riscv_pmp_init()`.

The essential locking of the interrupt stack guard will now be deferred
and correctly applied within `z_riscv_pmp_stackguard_enable()`, which is
executed only when the system is ready and the conditions
`defined(CONFIG_PMP_STACK_GUARD) && defined(CONFIG_MULTITHREADING)` are met.

Signed-off-by: Firas Sammoura <[email protected]>

Author: fsammoura1980

arch/riscv/core/pmp.c

@@ -378,7 +378,7 @@ void z_riscv_pmp_init(void)
 	 * addresses inaccessible. This will never change so we do it here
 	 * and lock it too.
 	 */
-	set_pmp_entry(&index, PMP_NONE | PMP_L,
+	set_pmp_entry(&index, PMP_NONE,
 		      (uintptr_t)z_interrupt_stacks[_current_cpu->id],
 		      Z_RISCV_STACK_GUARD_SIZE,
 		      pmp_addr, pmp_cfg, ARRAY_SIZE(pmp_addr));