bug: AP处理器发生异常时无法正确处理

2025-06-09 07:06:47 +00:00 · 2022-04-05 14:40:26 +08:00 · 2022-04-05 14:40:26 +08:00 · 32b8a163bb
commit 32b8a163bb
parent 5ea38e3b53
12 changed files with 143 additions and 51 deletions
--- a/kernel/common/cpu.h
+++ b/kernel/common/cpu.h
@ -2,7 +2,7 @@

 #include "glib.h"

-#define CPU_NUM 8 
+#define MAX_CPU_NUM 32 // 操作系统支持的最大处理器数量

 // cpu支持的最大cpuid指令的基础主功能号
 uint Cpu_cpuid_max_Basic_mop;
@ -53,3 +53,9 @@ void cpu_cpuid(uint mop, uint sop, uint *eax, uint*ebx, uint*ecx, uint*edx)
 * 
 */
 void cpu_init(void);
+
+struct cpu_core_info
+{
+    uint64_t stack_start;   // 栈基地址
+    uint64_t tss_vaddr; // tss地址
+}cpu_core_info[MAX_CPU_NUM];
--- a/kernel/common/glib.h
+++ b/kernel/common/glib.h
@ -16,6 +16,7 @@
 #define cli() __asm__ __volatile__("cli\n\t" :: \
                                       : "memory") //关闭外部中断
 #define nop() __asm__ __volatile__("nop\n\t")
+#define hlt() __asm__ __volatile__("hlt\n\t")

 //内存屏障
 #define io_mfence() __asm__ __volatile__("mfence\n\t" :: \
--- a/kernel/driver/acpi/acpi.c
+++ b/kernel/driver/acpi/acpi.c
@ -80,7 +80,7 @@ bool acpi_get_MADT(const struct acpi_system_description_table_header_t *_iter_da
    //*(struct acpi_Multiple_APIC_Description_Table_t *)_data = *(struct acpi_Multiple_APIC_Description_Table_t *)_iter_data;
    // 返回MADT的虚拟地址
    *(ul *)_data = (ul)_iter_data;
-    acpi_madt_vaddr = _iter_data;
+    acpi_madt_vaddr = (ul)_iter_data;
    return true;
 }

--- a/kernel/driver/interrupt/apic/apic.c
+++ b/kernel/driver/interrupt/apic/apic.c
@ -102,6 +102,46 @@ void apic_io_apic_init()
           */
 }

+/**
+ * @brief 初始化AP处理器的Local apic
+ *
+ */
+void apic_init_ap_core_local_apic()
+{
+    kinfo("Initializing AP-core's local apic...");
+    uint eax, edx;
+    // 启用xAPIC 和x2APIC
+    __asm__ __volatile__("movq  $0x1b, %%rcx   \n\t" // 读取IA32_APIC_BASE寄存器
+                         "rdmsr  \n\t"
+                         "bts $10,   %%rax  \n\t"
+                         "bts $11,   %%rax   \n\t"
+                         "wrmsr  \n\t"
+                         "movq $0x1b,    %%rcx   \n\t"
+                         "rdmsr  \n\t"
+                         : "=a"(eax), "=d"(edx)::"memory");
+
+    // kdebug("After enable xAPIC and x2APIC: edx=%#010x, eax=%#010x", edx, eax);
+
+    // 检测是否成功启用xAPIC和x2APIC
+    if (eax & 0xc00)
+        kinfo("xAPIC & x2APIC enabled!");
+    // 设置SVR寄存器，开启local APIC、禁止EOI广播
+
+    __asm__ __volatile__("movq $0x80f, %%rcx    \n\t"
+                         "rdmsr  \n\t"
+                         "bts $8, %%rax  \n\t"
+                         "bts $12, %%rax \n\t"
+                         "movq $0x80f, %%rcx    \n\t"
+                         "wrmsr  \n\t"
+                         "movq $0x80f , %%rcx   \n\t"
+                         "rdmsr \n\t"
+                         : "=a"(eax), "=d"(edx)::"memory", "rcx");
+
+    if (eax & 0x100)
+        kinfo("APIC Software Enabled.");
+    if (eax & 0x1000)
+        kinfo("EOI-Broadcast Suppression Enabled.");
+}
 /**
 * @brief 初始化local apic
 *
@ -174,7 +214,7 @@ void apic_local_apic_init()
    __asm__ __volatile__("movq $0x80f, %%rcx    \n\t"
                         "rdmsr  \n\t"
                         "bts $8, %%rax  \n\t"
-                         "bts $12, %%rax \n\t"
+                         //                         "bts $12, %%rax \n\t"
                         "movq $0x80f, %%rcx    \n\t"
                         "wrmsr  \n\t"
                         "movq $0x80f , %%rcx   \n\t"
@ -205,10 +245,11 @@ void apic_local_apic_init()
    //                          Table 10-6. Local APIC Register Address Map Supported by x2APIC
    // 获取 Local APIC ID
    // 0x802处是x2APIC ID 位宽32bits 的 Local APIC ID register
+    /*
    __asm__ __volatile__("movq $0x802, %%rcx    \n\t"
                         "rdmsr  \n\t"
                         : "=a"(eax), "=d"(edx)::"memory");
-
+    */
    // kdebug("get Local APIC ID: edx=%#010x, eax=%#010x", edx, eax);
    // kdebug("local_apic_id=%#018lx", *(uint *)(APIC_LOCAL_APIC_VIRT_BASE_ADDR + LOCAL_APIC_OFFSET_Local_APIC_ID));

--- a/kernel/driver/interrupt/apic/apic.h
+++ b/kernel/driver/interrupt/apic/apic.h
@ -269,6 +269,13 @@ ul apic_ioapic_read_rte(unsigned char index);
 */
 void apic_ioapic_write_rte(unsigned char index, ul value);

+
+/**
+ * @brief 初始化AP处理器的Local apic
+ * 
+ */
+void apic_init_ap_core_local_apic();
+
 /**
 * @brief 初始化apic控制器
 *
--- a/kernel/exception/gate.h
+++ b/kernel/exception/gate.h
@ -24,7 +24,6 @@ extern struct desc_struct GDT_Table[]; //GDT_Table是head.S中的GDT_Table
 extern struct gate_struct IDT_Table[]; // IDT_Table是head.S中的IDT_Table
 extern unsigned int TSS64_Table[26];

-
 /**
 * @brief 初始化中段描述符表内的门描述符（每个16B）
 * @param gate_selector_addr IDT表项的地址
@ -52,14 +51,16 @@ void set_gate(ul *gate_selector_addr, ul attr, unsigned char ist, ul *code_addr)

    __d1 = (0xffffffff & tmp_code_addr); //设置段内偏移[63:32]

-
    *gate_selector_addr = __d0;
    *(gate_selector_addr + 1) = __d1;
 }

+void set_tss_descriptor(unsigned int n, void *addr)
+{

-
-
+    *(unsigned long *)(GDT_Table + n) = (103UL & 0xffff) | (((unsigned long)addr & 0xffff) << 16) | (((unsigned long)addr >> 16 & 0xff) << 32) | ((unsigned long)0x89 << 40) | ((103UL >> 16 & 0xf) << 48) | (((unsigned long)addr >> 24 & 0xff) << 56); /////89 is attribute
+    *(unsigned long *)(GDT_Table + n + 1) = ((unsigned long)addr >> 32 & 0xffffffff) | 0;
+}

 /**
 * @brief 加载任务状态段寄存器
@ -69,7 +70,7 @@ void set_gate(ul *gate_selector_addr, ul attr, unsigned char ist, ul *code_addr)
 #define load_TR(n)                                      \
    do                                                  \
    {                                                   \
-        __asm__ __volatile__("ltr %%ax" ::"a"(n << 3)); \
+        __asm__ __volatile__("ltr %%ax" ::"a"((n)<< 3)); \
    } while (0)

 /**
--- a/kernel/head.S
+++ b/kernel/head.S
@ -381,7 +381,7 @@ switch_seg:

    .quad entry64

-.global entry64
+
 entry64:

    movq $0x10, %rax
@ -590,7 +590,7 @@ GDT_Table:
    .quad 0x0000f20000000000 // 6 用户64位数据段描述符 0x30
    .quad 0x00cf9a000000ffff // 7 内核32位代码段描述符 0x38
    .quad 0x00cf92000000ffff // 8 内核32位数据段描述符 0x40
-    .fill 10, 8, 0           // 10-11 TSS(跳过了第9段)  重复十次填充8字节的空间，赋值为0
+    .fill 100, 8, 0           // 10-11 TSS(跳过了第9段)  重复十次填充8字节的空间，赋值为0   长模式下，每个TSS长度为128bit
 GDT_END:

 GDT_POINTER:
--- a/kernel/main.c
+++ b/kernel/main.c
@ -12,6 +12,7 @@
 #include "mm/slab.h"
 #include "process/process.h"
 #include "syscall/syscall.h"
+#include "smp/smp.h"

 #include "driver/multiboot2/multiboot2.h"
 #include "driver/acpi/acpi.h"
@ -148,9 +149,12 @@ void system_initialize()
    load_TR(10); // 加载TR寄存器
    ul tss_item_addr = 0x7c00;

-    set_TSS64(_stack_start, _stack_start, _stack_start, tss_item_addr, tss_item_addr,
+    set_TSS64((ul)TSS64_Table, _stack_start, _stack_start, tss_item_addr, tss_item_addr,
              tss_item_addr, tss_item_addr, tss_item_addr, tss_item_addr, tss_item_addr);

+    cpu_core_info[0].stack_start = _stack_start;
+    cpu_core_info[0].tss_vaddr = (ul)TSS64_Table;
+
    // 初始化中断描述符表
    sys_vector_init();

@ -161,6 +165,9 @@ void system_initialize()
    // 初始化中断模块
    irq_init();

+    smp_init();
+
+    hlt();
    // 先初始化系统调用模块
    syscall_init();

@ -171,12 +178,12 @@ void system_initialize()
    pci_init();
    ahci_init();
    
-    smp_init();
    // test_slab();
    // test_mm();

    //  再初始化进程模块。顺序不能调转
    // process_init();
+    
 }

 //操作系统内核从这里开始执行
--- a/kernel/process/process.h
+++ b/kernel/process/process.h
@ -154,7 +154,7 @@ struct thread_struct initial_thread;
 // 初始化 初始进程的union ，并将其链接到.data.init_proc段内
 union proc_union initial_proc_union __attribute__((__section__(".data.init_proc_union"))) = {INITIAL_PROC(initial_proc_union.pcb)};

-struct process_control_block *initial_proc[CPU_NUM] = {&initial_proc_union.pcb, 0};
+struct process_control_block *initial_proc[MAX_CPU_NUM] = {&initial_proc_union.pcb, 0};

 struct mm_struct initial_mm = {0};
 struct thread_struct initial_thread =
@ -211,7 +211,7 @@ struct tss_struct
 		.io_map_base_addr = 0                                             \
 	}
 // 为每个核心初始化初始进程的tss
-struct tss_struct initial_tss[CPU_NUM] = {[0 ... CPU_NUM - 1] = INITIAL_TSS};
+struct tss_struct initial_tss[MAX_CPU_NUM] = {[0 ... MAX_CPU_NUM - 1] = INITIAL_TSS};

 // 获取当前的pcb
 struct process_control_block *get_current_pcb()
--- a/kernel/smp/smp.c
+++ b/kernel/smp/smp.c
@ -1,11 +1,14 @@
 #include "smp.h"
 #include "../common/kprint.h"
 #include "../driver/interrupt/apic/apic.h"
-
-extern void apic_local_apic_init();
+#include "../exception/gate.h"
+#include "../common/cpu.h"
+#include "../mm/slab.h"
+#include "../process/process.h"

 static struct acpi_Processor_Local_APIC_Structure_t *proc_local_apic_structs[MAX_SUPPORTED_PROCESSOR_NUM];
 static uint32_t total_processor_num = 0;
+int current_starting_cpu = 0;

 void smp_init()
 {
@ -17,21 +20,44 @@ void smp_init()
    for (int i = 0; i < total_processor_num; ++i)
        proc_local_apic_structs[i] = (struct acpi_Processor_Local_APIC_Structure_t *)(tmp_vaddr[i]);

-    for (int i = 0; i < total_processor_num; ++i)
-    {
-        kdebug("[core %d] acpi processor UID=%d, APIC ID=%d, flags=%#010lx", i, proc_local_apic_structs[i]->ACPI_Processor_UID, proc_local_apic_structs[i]->ACPI_ID, proc_local_apic_structs[i]->flags);
-    }
-
    //*(uchar *)0x20000 = 0xf4; // 在内存的0x20000处写入HLT指令(AP处理器会执行物理地址0x20000的代码)
    // 将引导程序复制到物理地址0x20000处
    memcpy((unsigned char *)0x20000, _apu_boot_start, (unsigned long)&_apu_boot_end - (unsigned long)&_apu_boot_start);
+    wrmsr(0x830, 0xc4500); // init IPI
+
+    struct INT_CMD_REG icr_entry;
+    icr_entry.dest_mode = DEST_PHYSICAL;
+    icr_entry.deliver_status = IDLE;
+    icr_entry.res_1 = 0;
+    icr_entry.level = ICR_LEVEL_DE_ASSERT;
+    icr_entry.trigger = EDGE_TRIGGER;
+    icr_entry.res_2 = 0;
+    icr_entry.res_3 = 0;
+
+    for (int i = 1; i < total_processor_num; ++i) // i从1开始，不初始化bsp
+    {
+        current_starting_cpu = i;
+        kdebug("[core %d] acpi processor UID=%d, APIC ID=%d, flags=%#010lx", i, proc_local_apic_structs[i]->ACPI_Processor_UID, proc_local_apic_structs[i]->ACPI_ID, proc_local_apic_structs[i]->flags);
+        // 为每个AP处理器分配栈空间、tss空间
+        cpu_core_info[i].stack_start = (uint64_t)kmalloc(STACK_SIZE, 0) + STACK_SIZE;
+        cpu_core_info[i].tss_vaddr = (uint64_t)kmalloc(128, 0);
+
+        set_tss_descriptor(10 + (i * 2), (void *)(cpu_core_info[i].tss_vaddr));
+        set_TSS64(cpu_core_info[i].tss_vaddr, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start, cpu_core_info[i].stack_start);
+        kdebug("GDT Table %#018lx, \t %#018lx", GDT_Table[10 + i * 2], GDT_Table[10 + i * 2 + 1]);
+
+        icr_entry.vector = 0x20;
+        icr_entry.deliver_mode = ICR_Start_up;
+        icr_entry.dest_shorthand = ICR_No_Shorthand;
+        icr_entry.destination.x2apic_destination = current_starting_cpu;

        // 先init ipi， 然后连续发送两次start-up IPI
        // x2APIC下，ICR寄存器地址为0x830
        // xAPIC下则为0xfee00300(31-0) 0xfee00310 (63-32)
-    wrmsr(0x830, 0xc4500);  // init IPI
-    wrmsr(0x830, 0xc4620);  // start-up IPI
-    wrmsr(0x830, 0xc4620);  // start-up IPI
+
+        wrmsr(0x830, *(ul *)&icr_entry); // start-up IPI
+        wrmsr(0x830, *(ul *)&icr_entry); // start-up IPI
+    }
 }

 /**
@ -41,7 +67,10 @@ void smp_init()
 void smp_ap_start()
 {
    ksuccess("AP core successfully started!");
-    kinfo("Initializing AP's local apic...");
-    apic_local_apic_init();
-    while(1);
+    kdebug("current=%d", current_starting_cpu);
+    load_TR(10 + current_starting_cpu * 2);
+    apic_init_ap_core_local_apic();
+    int a =1/0; // 在这儿会出现异常，cs fs gs ss寄存器会被改变
+    
+    hlt();
 }
--- a/kernel/smp/smp.h
+++ b/kernel/smp/smp.h
@ -4,7 +4,7 @@
 #include "../driver/acpi/acpi.h"
 #include "../driver/interrupt/apic/apic.h"

-#define MAX_SUPPORTED_PROCESSOR_NUM 1024    // 操作系统支持的最大处理器数量
+#define MAX_SUPPORTED_PROCESSOR_NUM 1024    

 extern uchar _apu_boot_start[];
 extern uchar _apu_boot_end[];
--- a/run.sh
+++ b/run.sh
@ -93,7 +93,7 @@ if [ $flag_can_run -eq 1 ]; then
        bochs -q -f ${bochsrc} -rc ./tools/bochsinit
    else
        qemu-system-x86_64 -cdrom ${iso} -m 512M -smp 2,cores=2,threads=1,sockets=1 \
-        -monitor stdio -s -S -cpu IvyBridge --enable-kvm \
+        -monitor stdio -d cpu_reset,guest_errors -s -S -cpu IvyBridge --enable-kvm \
        -drive id=disk,file=bin/disk.img,if=none \
        -device ahci,id=ahci \
        -device ide-hd,drive=disk,bus=ahci.0    \