/* * Copyright (c) 2006-2023, RT-Thread Development Team * * SPDX-License-Identifier: Apache-2.0 * * Change Logs: * Date Author Notes * 2023-02-21 GuEe-GUI first version */ #include #define DBG_TAG "cpu.aa64" #define DBG_LVL DBG_INFO #include #include #include #include #include #include #include #include #include #include #include #include #define SIZE_KB 1024 #define SIZE_MB (1024 * SIZE_KB) #define SIZE_GB (1024 * SIZE_MB) extern rt_ubase_t _start, _end; extern void _secondary_cpu_entry(void); extern size_t MMUTable[]; extern void *system_vectors; static void *fdt_ptr = RT_NULL; static rt_size_t fdt_size = 0; static rt_uint64_t initrd_ranges[3] = { }; #ifdef RT_USING_SMP extern struct cpu_ops_t cpu_psci_ops; extern struct cpu_ops_t cpu_spin_table_ops; #else extern int rt_hw_cpu_id(void); #endif rt_uint64_t rt_cpu_mpidr_table[] = { [RT_CPUS_NR] = 0, }; static struct cpu_ops_t *cpu_ops[] = { #ifdef RT_USING_SMP &cpu_psci_ops, &cpu_spin_table_ops, #endif }; static struct rt_ofw_node *cpu_np[RT_CPUS_NR] = { }; void rt_hw_fdt_install_early(void *fdt) { if (fdt != RT_NULL && !fdt_check_header(fdt)) { fdt_ptr = fdt; fdt_size = fdt_totalsize(fdt); } } #ifdef RT_USING_HWTIMER static rt_ubase_t loops_per_tick[RT_CPUS_NR]; static rt_ubase_t cpu_get_cycles(void) { rt_ubase_t cycles; rt_hw_sysreg_read(cntpct_el0, cycles); return cycles; } static void cpu_loops_per_tick_init(void) { rt_ubase_t offset; volatile rt_ubase_t freq, step, cycles_end1, cycles_end2; volatile rt_uint32_t cycles_count1 = 0, cycles_count2 = 0; rt_hw_sysreg_read(cntfrq_el0, freq); step = freq / RT_TICK_PER_SECOND; cycles_end1 = cpu_get_cycles() + step; while (cpu_get_cycles() < cycles_end1) { __asm__ volatile ("nop"); __asm__ volatile ("add %0, %0, #1":"=r"(cycles_count1)); } cycles_end2 = cpu_get_cycles() + step; while (cpu_get_cycles() < cycles_end2) { __asm__ volatile ("add %0, %0, #1":"=r"(cycles_count2)); } if ((rt_int32_t)(cycles_count2 - cycles_count1) > 0) { offset = cycles_count2 - cycles_count1; } else { /* Impossible, but prepared for any eventualities */ offset = cycles_count2 / 4; } loops_per_tick[rt_hw_cpu_id()] = offset; } static void cpu_us_delay(rt_uint32_t us) { volatile rt_base_t start = cpu_get_cycles(), cycles; cycles = ((us * 0x10c7UL) * loops_per_tick[rt_hw_cpu_id()] * RT_TICK_PER_SECOND) >> 32; while ((cpu_get_cycles() - start) < cycles) { rt_hw_cpu_relax(); } } #endif /* RT_USING_HWTIMER */ rt_weak void rt_hw_idle_wfi(void) { __asm__ volatile ("wfi"); } static void system_vectors_init(void) { rt_hw_set_current_vbar((rt_ubase_t)&system_vectors); } rt_inline void cpu_info_init(void) { int i = 0; rt_uint64_t mpidr; struct rt_ofw_node *np; /* get boot cpu info */ rt_hw_sysreg_read(mpidr_el1, mpidr); rt_ofw_foreach_cpu_node(np) { rt_uint64_t hwid = rt_ofw_get_cpu_hwid(np, 0); if ((mpidr & MPIDR_AFFINITY_MASK) != hwid) { /* Only save affinity and res make smp boot can check */ hwid |= 1ULL << 31; } else { hwid = mpidr; } cpu_np[i] = np; rt_cpu_mpidr_table[i] = hwid; rt_ofw_data(np) = (void *)hwid; for (int idx = 0; idx < RT_ARRAY_SIZE(cpu_ops); ++idx) { struct cpu_ops_t *ops = cpu_ops[idx]; if (ops->cpu_init) { ops->cpu_init(i, np); } } if (++i >= RT_CPUS_NR) { break; } } rt_hw_cpu_dcache_ops(RT_HW_CACHE_FLUSH, rt_cpu_mpidr_table, sizeof(rt_cpu_mpidr_table)); #ifdef RT_USING_HWTIMER cpu_loops_per_tick_init(); if (!rt_device_hwtimer_us_delay) { rt_device_hwtimer_us_delay = &cpu_us_delay; } #endif /* RT_USING_HWTIMER */ } rt_inline rt_bool_t is_kernel_aspace(const char *name) { static char * const names[] = { "kernel", "memheap", }; if (!name) { return RT_FALSE; } for (int i = 0; i < RT_ARRAY_SIZE(names); ++i) { if (!rt_strcmp(names[i], name)) { return RT_TRUE; } } return RT_FALSE; } void rt_hw_common_setup(void) { rt_size_t mem_region_nr; rt_region_t *mem_region; rt_size_t page_best_start; rt_region_t platform_mem_region; static struct mem_desc platform_mem_desc; void *kernel_start, *kernel_end, *memheap_start = RT_NULL, *memheap_end = RT_NULL; system_vectors_init(); #ifdef RT_USING_SMART rt_hw_mmu_map_init(&rt_kernel_space, (void*)0xfffffffff0000000, 0x10000000, MMUTable, PV_OFFSET); #else rt_hw_mmu_map_init(&rt_kernel_space, (void*)0xffffd0000000, 0x10000000, MMUTable, 0); #endif kernel_start = rt_kmem_v2p((void *)&_start) - 64; kernel_end = rt_kmem_v2p((void *)&_end); if (!rt_fdt_commit_memregion_request(&mem_region, &mem_region_nr, RT_TRUE)) { const char *name = "memheap"; while (mem_region_nr --> 0) { if (mem_region->name == name || !rt_strcmp(mem_region->name, name)) { memheap_start = (void *)mem_region->start; memheap_end = (void *)mem_region->end; break; } mem_region++; } } page_best_start = (rt_size_t)(memheap_end ? : kernel_end); if (memheap_end && fdt_ptr > kernel_start) { rt_memmove(memheap_end - PV_OFFSET, fdt_ptr - PV_OFFSET, fdt_size); fdt_ptr = memheap_end; page_best_start = (rt_size_t)fdt_ptr + fdt_size; } rt_fdt_commit_memregion_early(&(rt_region_t) { .name = "fdt", .start = (rt_size_t)fdt_ptr, .end = (rt_size_t)(fdt_ptr + fdt_size), }, RT_TRUE); fdt_ptr -= PV_OFFSET; rt_fdt_commit_memregion_early(&(rt_region_t) { .name = "kernel", .start = (rt_size_t)kernel_start, .end = (rt_size_t)kernel_end, }, RT_TRUE); #ifndef RT_USING_SMART rt_fdt_commit_memregion_early(&(rt_region_t) { .name = "null", .start = (rt_size_t)RT_NULL, .end = (rt_size_t)RT_NULL + ARCH_PAGE_SIZE, }, RT_TRUE); #endif /* !RT_USING_SMART */ if (rt_fdt_prefetch(fdt_ptr)) { /* Platform cannot be initialized */ RT_ASSERT(0); } rt_fdt_scan_chosen_stdout(); rt_fdt_scan_initrd(initrd_ranges); rt_fdt_scan_memory(); if (memheap_start && memheap_end) { rt_system_heap_init(memheap_start - PV_OFFSET, memheap_end - PV_OFFSET); } platform_mem_region.start = ~0UL; platform_mem_region.end = 0; if (!rt_fdt_commit_memregion_request(&mem_region, &mem_region_nr, RT_TRUE)) { LOG_I("Reserved memory:"); while (mem_region_nr --> 0) { if (is_kernel_aspace(mem_region->name)) { if (platform_mem_region.start > mem_region->start) { platform_mem_region.start = mem_region->start; } if (platform_mem_region.end < mem_region->end) { platform_mem_region.end = mem_region->end; } } LOG_I(" %-*.s [%p, %p]", RT_NAME_MAX, mem_region->name, mem_region->start, mem_region->end); ++mem_region; } } if (!rt_fdt_commit_memregion_request(&mem_region, &mem_region_nr, RT_FALSE)) { rt_ubase_t best_offset = ~0UL; rt_region_t *usable_mem_region = mem_region, *page_region = RT_NULL; rt_region_t init_page_region = { 0 }; rt_region_t defer_hi = { 0 }; rt_err_t error; LOG_I("Usable memory:"); for (int i = 0; i < mem_region_nr; ++i, ++mem_region) { if (!mem_region->name) { continue; } if (platform_mem_region.start > mem_region->start) { platform_mem_region.start = mem_region->start; } if (platform_mem_region.end < mem_region->end) { platform_mem_region.end = mem_region->end; } if (mem_region->start >= page_best_start && mem_region->start - page_best_start < best_offset && /* MUST >= 1MB */ mem_region->end - mem_region->start >= SIZE_MB) { page_region = mem_region; best_offset = page_region->start - page_best_start; } LOG_I(" %-*.s [%p, %p]", RT_NAME_MAX, mem_region->name, mem_region->start, mem_region->end); } RT_ASSERT(page_region != RT_NULL); /* don't map more than ARCH_EARLY_MAP_SIZE */ if (page_region->end - page_region->start > ARCH_PAGE_INIT_THRESHOLD) { defer_hi.name = page_region->name; defer_hi.end = page_region->end; defer_hi.start = RT_ALIGN_DOWN(page_region->start + ARCH_PAGE_INIT_THRESHOLD, ARCH_SECTION_SIZE); page_region->end = defer_hi.start; } init_page_region.start = page_region->start - PV_OFFSET; init_page_region.end = page_region->end - PV_OFFSET; rt_page_init(init_page_region); platform_mem_region.start = RT_ALIGN(platform_mem_region.start, ARCH_PAGE_SIZE); platform_mem_region.end = RT_ALIGN_DOWN(platform_mem_region.end, ARCH_PAGE_SIZE); RT_ASSERT(platform_mem_region.end - platform_mem_region.start != 0); platform_mem_desc.paddr_start = platform_mem_region.start; platform_mem_desc.vaddr_start = platform_mem_region.start - PV_OFFSET; platform_mem_desc.vaddr_end = platform_mem_region.end - PV_OFFSET - 1; platform_mem_desc.attr = NORMAL_MEM; rt_hw_mmu_setup(&rt_kernel_space, &platform_mem_desc, 1); rt_fdt_earlycon_kick(FDT_EARLYCON_KICK_UPDATE); mem_region = usable_mem_region; if (defer_hi.start) { /* to virt address */ init_page_region.start = defer_hi.start - PV_OFFSET; init_page_region.end = defer_hi.end - PV_OFFSET; error = rt_page_install(init_page_region); if (error) { LOG_W("Deferred page installation FAILED:"); LOG_W(" %-*.s [%p, %p]", RT_NAME_MAX, defer_hi.name, defer_hi.start, defer_hi.end); } else { LOG_I("Deferred page installation SUCCEED:"); LOG_I(" %-*.s [%p, %p]", RT_NAME_MAX, defer_hi.name, defer_hi.start, defer_hi.end); } } for (int i = 0; i < mem_region_nr; ++i, ++mem_region) { if (mem_region != page_region && mem_region->name) { init_page_region.start = mem_region->start - PV_OFFSET; init_page_region.end = mem_region->end - PV_OFFSET; rt_page_install(init_page_region); } } } rt_fdt_unflatten(); cpu_info_init(); #ifdef RT_USING_PIC rt_pic_init(); rt_pic_irq_init(); #else /* initialize hardware interrupt */ rt_hw_interrupt_init(); /* initialize uart */ rt_hw_uart_init(); #endif #ifndef RT_HWTIMER_ARM_ARCH /* initialize timer for os tick */ rt_hw_gtimer_init(); #endif /* !RT_HWTIMER_ARM_ARCH */ #ifdef RT_USING_COMPONENTS_INIT rt_components_board_init(); #endif #if defined(RT_USING_CONSOLE) && defined(RT_USING_DEVICE) rt_ofw_console_setup(); #endif rt_thread_idle_sethook(rt_hw_idle_wfi); #ifdef RT_USING_SMP /* Install the IPI handle */ rt_hw_ipi_handler_install(RT_SCHEDULE_IPI, rt_scheduler_ipi_handler); rt_hw_ipi_handler_install(RT_STOP_IPI, rt_scheduler_ipi_handler); rt_hw_interrupt_umask(RT_SCHEDULE_IPI); rt_hw_interrupt_umask(RT_STOP_IPI); #endif } #ifdef RT_USING_SMP rt_weak void rt_hw_secondary_cpu_up(void) { int cpu_id = rt_hw_cpu_id(); rt_uint64_t entry = (rt_uint64_t)rt_kmem_v2p(_secondary_cpu_entry); if (!entry) { LOG_E("Failed to translate '_secondary_cpu_entry' to physical address"); RT_ASSERT(0); } /* Maybe we are no in the first cpu */ for (int i = 0; i < RT_ARRAY_SIZE(cpu_np); ++i) { int err; const char *enable_method; if (!cpu_np[i] || i == cpu_id) { continue; } err = rt_ofw_prop_read_string(cpu_np[i], "enable-method", &enable_method); for (int idx = 0; !err && idx < RT_ARRAY_SIZE(cpu_ops); ++idx) { struct cpu_ops_t *ops = cpu_ops[idx]; if (ops->method && !rt_strcmp(ops->method, enable_method) && ops->cpu_boot) { err = ops->cpu_boot(i, entry); break; } } if (err) { LOG_W("Call cpu %d on %s", i, "failed"); } } } rt_weak void rt_hw_secondary_cpu_bsp_start(void) { int cpu_id = rt_hw_cpu_id(); system_vectors_init(); rt_hw_spin_lock(&_cpus_lock); /* Save all mpidr */ rt_hw_sysreg_read(mpidr_el1, rt_cpu_mpidr_table[cpu_id]); rt_hw_mmu_ktbl_set((unsigned long)MMUTable); #ifdef RT_USING_PIC rt_pic_irq_init(); #else /* initialize vector table */ rt_hw_vector_init(); arm_gic_cpu_init(0, 0); #ifdef BSP_USING_GICV3 arm_gic_redist_init(0, 0); #endif /* BSP_USING_GICV3 */ #endif #ifndef RT_HWTIMER_ARM_ARCH /* initialize timer for os tick */ rt_hw_gtimer_local_enable(); #endif /* !RT_HWTIMER_ARM_ARCH */ rt_dm_secondary_cpu_init(); rt_hw_interrupt_umask(RT_SCHEDULE_IPI); rt_hw_interrupt_umask(RT_STOP_IPI); LOG_I("Call cpu %d on %s", cpu_id, "success"); #ifdef RT_USING_HWTIMER if (rt_device_hwtimer_us_delay == &cpu_us_delay) { cpu_loops_per_tick_init(); } #endif rt_system_scheduler_start(); } rt_weak void rt_hw_secondary_cpu_idle_exec(void) { rt_hw_wfe(); } #endif void rt_hw_console_output(const char *str) { rt_fdt_earlycon_output(str); }