+#include <lunaix/types.h>
#include <lunaix/block.h>
#include <lunaix/boot_generic.h>
#include <lunaix/device.h>
#include <lunaix/foptions.h>
-#include <lunaix/fs/twifs.h>
#include <lunaix/input.h>
-#include <lunaix/lxconsole.h>
#include <lunaix/mm/cake.h>
-#include <lunaix/mm/mmio.h>
-#include <lunaix/mm/page.h>
#include <lunaix/mm/pmm.h>
+#include <lunaix/mm/page.h>
#include <lunaix/mm/valloc.h>
#include <lunaix/mm/vmm.h>
#include <lunaix/process.h>
#include <lunaix/sched.h>
#include <lunaix/spike.h>
#include <lunaix/trace.h>
-#include <lunaix/tty/tty.h>
-#include <lunaix/types.h>
+#include <lunaix/owloysius.h>
+#include <lunaix/hart_state.h>
+#include <lunaix/syslog.h>
+#include <lunaix/sections.h>
#include <hal/acpi/acpi.h>
-#include <hal/intc.h>
-#include <hal/pci.h>
+#include <hal/devtree.h>
#include <sys/abi.h>
-#include <sys/interrupts.h>
-#include <sys/mm/mempart.h>
+#include <sys/mm/mm_defs.h>
-#include <klibc/stdio.h>
+#include <klibc/strfmt.h>
#include <klibc/string.h>
+LOG_MODULE("kinit")
+
extern void
-__proc0(); /* proc0.c */
+lunad_main();
-void
-spawn_proc0();
+/**
+ * @brief 创建并运行Lunaix守护进程
+ *
+ */
+static void
+spawn_lunad()
+{
+ int has_error;
+ struct thread* kthread;
+
+ has_error = spawn_process(&kthread, (ptr_t)lunad_main, false);
+ assert_msg(!has_error, "failed to spawn lunad");
+
+ run(kthread);
+
+ fail("Unexpected Return");
+}
-void
-kmem_init(struct boot_handoff* bhctx);
+static void
+kmem_init(struct boot_handoff* bhctx)
+{
+ pte_t* ptep = mkptep_va(VMS_SELF, KERNEL_RESIDENT);
+
+ ptep = mkl0tep(ptep);
+
+ unsigned int i = ptep_vfn(ptep);
+ do {
+ if (l0tep_implie_vmnts(ptep)) {
+ ptep++;
+ continue;
+ }
+
+#if LnT_ENABLED(1)
+ assert(mkl1t(ptep++, 0, KERNEL_PGTAB));
+#elif LnT_ENABLED(2)
+ assert(mkl2t(ptep++, 0, KERNEL_PGTAB));
+#elif LnT_ENABLED(3)
+ assert(mkl3t(ptep++, 0, KERNEL_PGTAB));
+#else
+ assert(mklft(ptep++, 0, KERNEL_PGTAB));
+#endif
+ } while (++i < MAX_PTEN);
+
+ // allocators
+ cake_init();
+ valloc_init();
+}
+
+static void
+__remap_and_load_dtb(struct boot_handoff* bhctx)
+{
+#ifdef CONFIG_USE_DEVICETREE
+ ptr_t dtb = bhctx->kexec.dtb_pa;
+
+ if (!dtb) {
+ return;
+ }
+
+ if (va_offset(dtb)) {
+ WARN("bad-aligned dtb location, expect page aligned");
+ return;
+ }
+
+ pte_t *ptep, pte;
+ size_t nr_pages;
+ bool loaded;
+
+ pte = mkpte(dtb, KERNEL_DATA);
+ ptep = mkptep_va(VMS_SELF, dtb_start);
+ nr_pages = leaf_count(CONFIG_DTB_MAXSIZE);
+
+ pmm_onhold_range(dtb, nr_pages);
+ vmm_set_ptes_contig(ptep, pte, PAGE_SIZE, nr_pages);
+
+ loaded = dt_load(dtb_start);
+ if (!loaded) {
+ ERROR("dtb load failed");
+ }
+#endif
+
+ return;
+}
void
kernel_bootstrap(struct boot_handoff* bhctx)
{
- pmm_init(bhctx->mem.size);
vmm_init();
+ pmm_init(bhctx);
+ // now we can start reserving physical space
+
/* Begin kernel bootstrapping sequence */
boot_begin(bhctx);
/* Setup kernel memory layout and services */
kmem_init(bhctx);
+ __remap_and_load_dtb(bhctx);
+
+ boot_parse_cmdline(bhctx);
+
/* Prepare stack trace environment */
trace_modksyms_init(bhctx);
- // crt
- tty_init(ioremap(0xB8000, PG_SIZE));
- tty_set_theme(VGA_COLOR_WHITE, VGA_COLOR_BLACK);
- lxconsole_init();
+ device_scan_drivers();
- /* Get platform configuration */
- acpi_init();
-
- /* Let's get fs online as soon as possible, as things rely on them */
- vfs_init();
- fsm_init();
- input_init();
+ device_sysconf_load();
- /* Get intc online, this is the cornerstone when initing devices */
- intc_init();
+ invoke_init_function(on_earlyboot);
- /* System timing and clock support */
clock_init();
timer_init();
- block_init();
+ /*
+ TODO autoload these init function that do not have dependency between
+ them
+ */
- /* the bare metal are now happy, let's get software over with */
+ /* Let's get fs online as soon as possible, as things rely on them */
+ vfs_init();
+ fsm_init();
+ input_init();
+ block_init();
sched_init();
- int errno = 0;
- if ((errno = vfs_mount_root("ramfs", NULL))) {
- panickf("Fail to mount root. (errno=%d)", errno);
- }
+ device_onboot_load();
- /* Mount these system-wide pseudo-fs */
- vfs_mount("/dev", "devfs", NULL, 0);
- vfs_mount("/sys", "twifs", NULL, MNT_RO);
- vfs_mount("/task", "taskfs", NULL, MNT_RO);
+ /* the bare metal are now happy, let's get software over with */
- lxconsole_spawn_ttydev();
- device_init_builtin();
+ must_success(vfs_mount_root("ramfs", NULL));
+ must_success(vfs_mount("/dev", "devfs", NULL, 0));
+
+ invoke_init_function(on_boot);
/* Finish up bootstrapping sequence, we are ready to spawn the root process
* and start geting into uspace
*/
boot_end(bhctx);
- spawn_proc0();
-}
-
-/**
- * @brief 创建并运行proc0进程
- *
- */
-void
-spawn_proc0()
-{
- struct proc_info* proc0 = alloc_process();
-
- /**
- * @brief
- * 注意:这里和视频中说的不一样,属于我之后的一点微调。
- * 在视频中,spawn_proc0是在_kernel_post_init的末尾才调用的。并且是直接跳转到_proc0
- *
- * 但是我后来发现,上述的方法会产生竞态条件。这是因为spawn_proc0被调用的时候,时钟中断已经开启,
- * 而中断的产生会打乱栈的布局,从而使得下面的上下文设置代码产生未定义行为(Undefined
- * Behaviour)。 为了保险起见,有两种办法:
- * 1. 在创建proc0进程前关闭中断
- * 2. 将_kernel_post_init搬进proc0进程
- * (_kernel_post_init已经更名为init_platform)
- *
- * 目前的解决方案是2
- */
-
- proc0->parent = proc0;
-
- // 方案1:必须在读取eflags之后禁用。否则当进程被调度时,中断依然是关闭的!
- // cpu_disable_interrupt();
-
- /* Ok... 首先fork进我们的零号进程,而后由那里,我们fork进init进程。 */
-
- // 把当前虚拟地址空间(内核)复制一份。
- proc0->page_table = vmm_dup_vmspace(proc0->pid);
-
- // 直接切换到新的拷贝,进行配置。
- cpu_chvmspace(proc0->page_table);
-
- // 为内核创建一个专属栈空间。
- for (size_t i = 0; i < KERNEL_STACK_SIZE; i += PG_SIZE) {
- ptr_t pa = pmm_alloc_page(KERNEL_PID, 0);
- vmm_set_mapping(VMS_SELF, KERNEL_STACK + i, pa, PG_PREM_RW, VMAP_NULL);
- }
-
- proc_init_transfer(proc0, KERNEL_STACK_END, (ptr_t)__proc0, 0);
-
- // 向调度器注册进程。
- commit_process(proc0);
-
- // 由于时钟中断与APIC未就绪,我们需要手动进行第一次调度。这里也会同时隐式地恢复我们的eflags.IF位
- proc0->state = PS_RUNNING;
- switch_context(proc0);
-
- /* Should not return */
- assert_msg(0, "Unexpected Return");
+ spawn_lunad();
}
-void
-kmem_init(struct boot_handoff* bhctx)
-{
- extern u8_t __kexec_end;
- // 将内核占据的页,包括前1MB,hhk_init 设为已占用
- size_t pg_count = ((ptr_t)&__kexec_end - KERNEL_EXEC) >> PG_SIZE_BITS;
- pmm_mark_chunk_occupied(KERNEL_PID, 0, pg_count, PP_FGLOCKED);
-
- // reserve higher half
- for (size_t i = L1_INDEX(KERNEL_EXEC); i < 1023; i++) {
- assert(vmm_set_mapping(VMS_SELF, i << 22, 0, 0, VMAP_NOMAP));
- }
-
- // allocators
- cake_init();
- valloc_init();
-}