Merge branch 'master' into patch-screen-manager

.vscode/settings.json

@@ -114,7 +114,9 @@
         "errno.h": "c",
         "bug.h": "c",
         "apic_timer.h": "c",
-        "sched.h": "c"
+        "sched.h": "c",
+        "preempt.h": "c",
+        "softirq.h": "c"
     },
     "C_Cpp.errorSquiggles": "Enabled",
     "esbonio.sphinx.confDir": ""

Makefile

@@ -7,7 +7,7 @@ export ARCH=__x86_64__
 export ROOT_PATH=$(shell pwd)
 
 export DEBUG=DEBUG
-export GLOBAL_CFLAGS := -mcmodel=large -fno-builtin -m64  -O0 -fno-stack-protector -D $(ARCH) 
+export GLOBAL_CFLAGS := -mcmodel=large -fno-builtin -m64  -fno-stack-protector -D $(ARCH) -O1
 
 ifeq ($(DEBUG), DEBUG)
 GLOBAL_CFLAGS += -g 

README.md

@@ -11,7 +11,7 @@
 - 项目文档  **[docs.DragonOS.org](https://docs.dragonos.org)**
 - 开源论坛  **[bbs.DragonOS.org](https://bbs.dragonos.org)**
 - 开发交流QQ群 **115763565**
- 
+- 代码搜索引擎 [code.DragonOS.org](http://code.dragonos.org) 
 ## 开发环境
 
 GCC>=8.0

README_EN.md

@@ -10,7 +10,8 @@ This project is a operating system running on computer which is in X86_ 64 Archi
 - Home Page  **[DragonOS.org](https://dragonos.org)**
 - Documentation  **[docs.DragonOS.org](https://docs.dragonos.org)**
 - BBS  **[bbs.DragonOS.org](https://bbs.dragonos.org)**
- 
+- QQ group **115763565**
+- code search engine [code.DragonOS.org](http://code.dragonos.org) 
 
 ## Development Environment
 

kernel/common/cpu.c

@@ -1,7 +1,7 @@
 #include "cpu.h"
 #include "kprint.h"
 #include "printk.h"
-
+// #pragma GCC optimize("O0")
 // cpu支持的最大cpuid指令的基础主功能号
 uint Cpu_cpuid_max_Basic_mop;
 // cpu支持的最大cpuid指令的扩展主功能号

kernel/common/glib.h

@@ -93,6 +93,7 @@ struct List
 static inline void list_init(struct List *list)
 {
     list->next = list;
+    io_mfence();
     list->prev = list;
 }
 
@@ -106,8 +107,11 @@ static inline void list_add(struct List *entry, struct List *node)
 {
 
     node->next = entry->next;
+    barrier();
     node->prev = entry;
+    barrier();
     node->next->prev = node;
+    barrier();
     entry->next = node;
 }
 

kernel/common/printk.c

@@ -45,6 +45,8 @@ static uint *get_pos_VBE_FB_addr();
  */
 static int cls();
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 /**
  * @brief 滚动窗口（尚不支持向下滚动)
  *
@@ -53,7 +55,7 @@ static int cls();
  * @param animation 是否包含滑动动画
  */
 static int scroll(bool direction, int pixels, bool animation);
-
+#pragma GCC pop_options
 /**
  * @brief 将数字按照指定的要求转换成对应的字符串（2~36进制）
  *
@@ -184,7 +186,9 @@ static void auto_newline()
 #endif
         pos.y = pos.max_y;
         int lines_to_scroll = 1;
+        barrier();
         scroll(true, lines_to_scroll * pos.char_size_y, sw_show_scroll_animation);
+        barrier();
         pos.y -= (lines_to_scroll - 1);
     }
 }
@@ -835,7 +839,6 @@ static int scroll(bool direction, int pixels, bool animation)
     int md = pixels % pos.char_size_y;
     if (md)
         pixels = pixels + pos.char_size_y - md;
-
     if (animation == false)
         return do_scroll(direction, pixels);
     else
@@ -957,3 +960,4 @@ int sprintk(char *buf, const char *fmt, ...)
 
     return count;
 }
+

kernel/common/printk.h

@@ -2,7 +2,8 @@
 // Created by longjin on 2022/1/21.
 //
 #pragma once
-
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 #define PAD_ZERO 1 // 0填充
 #define LEFT 2     // 靠左对齐
 #define RIGHT 4    // 靠右对齐
@@ -124,3 +125,4 @@ void printk_disable_animation();
  * @return int 字符串长度
  */
 int sprintk(char *buf, const char *fmt, ...);
+#pragma GCC pop_options

kernel/debug/bug.h

@@ -2,6 +2,8 @@
 #include <common/compiler.h>
 #include <common/kprint.h>
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 /**
  * @brief 当condition为true时输出警告信息
  *
@@ -19,3 +21,4 @@
         goto to;                       \
     unlikely(__ret_warn_on);           \
 })
+#pragma GCC pop_options

kernel/driver/interrupt/apic/apic.c

@@ -10,6 +10,8 @@
 #include <process/process.h>
 #include <sched/sched.h>
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 // 导出定义在irq.c中的中段门表
 extern void (*interrupt_table[24])(void);
 
@@ -94,27 +96,7 @@ void apic_io_apic_init()
 
     // 不需要手动启动IO APIC，只要初始化了RTE寄存器之后，io apic就会自动启用了。
     // 而且不是每台电脑都有RCBA寄存器，因此不需要手动启用IO APIC
-    /*
-           // get RCBA address
-           io_out32(0xcf8, 0x8000f8f0);
-           uint x = io_in32(0xcfc);
-           uint *p;
-           printk_color(RED, BLACK, "Get RCBA Address:%#010x\n", x);
-           x = x & 0xffffc000UL;
-           printk_color(RED, BLACK, "Get RCBA Address:%#010x\n", x);
     
-           // get OIC address
-           if (x > 0xfec00000 && x < 0xfee00000)
-           {
-               p = (unsigned int *)(x + 0x31feUL-apic_ioapic_map.addr_phys+apic_ioapic_map.virtual_index_addr);
-           }
-
-           // enable IOAPIC
-           x = (*p & 0xffffff00) | 0x100;
-           io_mfence();
-           *p = x;
-           io_mfence();
-           */
 }
 
 /**
@@ -650,6 +632,7 @@ void apic_ioapic_edge_ack(ul irq_num) // 边沿触发
  *
  * @param irq_num
  */
+
 void apic_local_apic_edge_ack(ul irq_num)
 {
     // 向EOI寄存器写入0x00表示结束中断
@@ -660,6 +643,7 @@ void apic_local_apic_edge_ack(ul irq_num)
                              : "memory");
 }
 
+
 /**
  * @brief 读取指定类型的 Interrupt Control Structure
  *
@@ -735,3 +719,4 @@ void apic_make_rte_entry(struct apic_IO_APIC_RTE_entry *entry, uint8_t vector, u
         entry->destination.logical.reserved1 = 0;
     }
 }
+#pragma GCC pop_options

kernel/driver/interrupt/apic/apic.h

@@ -5,6 +5,9 @@
 #include <exception/irq.h>
 #include <mm/mm.h>
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
+
 #define APIC_SUCCESS 0
 #define APIC_E_NOTFOUND 1
 
@@ -319,3 +322,5 @@ void apic_local_apic_edge_ack(ul irq_num); // local apic边沿触发 应答
  */
 void apic_make_rte_entry(struct apic_IO_APIC_RTE_entry *entry, uint8_t vector, uint8_t deliver_mode, uint8_t dest_mode,
                          uint8_t deliver_status, uint8_t polarity, uint8_t irr, uint8_t trigger, uint8_t mask, uint8_t dest_apicID);
+
+#pragma GCC pop_options

kernel/driver/interrupt/apic/apic_timer.c

@@ -4,14 +4,20 @@
 #include <common/kprint.h>
 #include <sched/sched.h>
 
+// #pragma GCC push_options
+// #pragma GCC optimize("O0")
 uint64_t apic_timer_ticks_result = 0;
 
 void apic_timer_enable(uint64_t irq_num)
 {
     // 启动apic定时器
+    io_mfence();
     uint64_t val = apic_timer_get_LVT();
+    io_mfence();
     val &= (~APIC_LVT_INT_MASKED);
+    io_mfence();
     apic_timer_write_LVT(val);
+    io_mfence();
 }
 
 void apic_timer_disable(uint64_t irq_num)
@@ -29,18 +35,24 @@ void apic_timer_disable(uint64_t irq_num)
 uint64_t apic_timer_install(ul irq_num, void *arg)
 {
     // 设置div16
+    io_mfence();
     apic_timer_stop();
+    io_mfence();
     apic_timer_set_div(APIC_TIMER_DIVISOR);
+    io_mfence();
 
     // 设置初始计数
     apic_timer_set_init_cnt(*(uint64_t *)arg);
+    io_mfence();
     // 填写LVT
     apic_timer_set_LVT(APIC_TIMER_IRQ_NUM, 1, APIC_LVT_Timer_Periodic);
+    io_mfence();
 }
 
 void apic_timer_uninstall(ul irq_num)
 {
     apic_timer_write_LVT(APIC_LVT_INT_MASKED);
+    io_mfence();
 }
 
 hardware_intr_controller apic_timer_intr_controller =
@@ -63,6 +75,7 @@ void apic_timer_handler(uint64_t number, uint64_t param, struct pt_regs *regs)
 {
 
     sched_update_jiffies();
+    io_mfence();
 }
 
 /**

kernel/driver/interrupt/apic/apic_timer.h

@@ -10,6 +10,8 @@ extern uint64_t apic_timer_ticks_result;
 
 #define APIC_TIMER_IRQ_NUM 151
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 /**
  * @brief 设置apic定时器的分频计数
  *
@@ -79,3 +81,5 @@ extern uint64_t apic_timer_ticks_result;
  *
  */
 void apic_timer_init();
+
+#pragma GCC pop_options

kernel/driver/timers/HPET/HPET.c

@@ -12,6 +12,8 @@
 #include <driver/interrupt/apic/apic_timer.h>
 #include <common/spinlock.h>
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 static struct acpi_HPET_description_table_t *hpet_table;
 static uint64_t HPET_REG_BASE = 0;
 static uint32_t HPET_COUNTER_CLK_PERIOD = 0; // 主计数器时间精度（单位：飞秒）
@@ -158,9 +160,9 @@ void HPET_measure_freq()
     // 顺便测定tsc频率
     test_tsc_start = rdtsc();
     io_mfence();
-
     while (measure_apic_timer_flag == false)
         ;
+    kdebug("wait done");
 
     irq_unregister(34);
 
@@ -283,3 +285,4 @@ int HPET_init()
     // kdebug("HPET_freq=%ld", (long)HPET_freq);
     // kdebug("HPET_freq=%lf", HPET_freq);
 }
+#pragma GCC pop_options

kernel/driver/usb/usb.c

@@ -10,7 +10,7 @@ extern spinlock_t xhci_controller_init_lock; // xhci控制器初始化锁
 #define MAX_USB_NUM 8 // pci总线上的usb设备的最大数量
 
 // 在pci总线上寻找到的usb设备控制器的header
-struct pci_device_structure_header_t *usb_pdevs[MAX_USB_NUM];
+static struct pci_device_structure_header_t *usb_pdevs[MAX_USB_NUM];
 static int usb_pdevs_count = 0;
 
 /**
@@ -34,6 +34,7 @@ void usb_init()
     // 初始化每个usb控制器
     for (int i = 0; i < usb_pdevs_count; ++i)
     {
+        io_mfence();
         switch (usb_pdevs[i]->ProgIF)
         {
         case USB_TYPE_UHCI:
@@ -47,6 +48,7 @@ void usb_init()
         case USB_TYPE_XHCI:
             // 初始化对应的xhci控制器
             xhci_init((struct pci_device_structure_general_device_t *)usb_pdevs[i]);
+            io_mfence();
             break;
 
         default:

kernel/driver/usb/xhci/xhci.c

@@ -9,6 +9,10 @@
 #include <exception/irq.h>
 #include <driver/interrupt/apic/apic.h>
 
+// 由于xhci寄存器读取需要对齐，因此禁用GCC优化选项
+#pragma GCC push_options
+#pragma GCC optimize("O0")
+
 spinlock_t xhci_controller_init_lock = {0}; // xhci控制器初始化锁(在usb_init中被初始化)
 
 static int xhci_ctrl_count = 0; // xhci控制器计数
@@ -52,7 +56,6 @@ hardware_intr_controller xhci_hc_intr_controller =
     例子：不能在一个32bit的寄存器中的偏移量8的位置开始读取1个字节
             这种情况下，我们必须从32bit的寄存器的0地址处开始读取32bit，然后通过移位的方式得到其中的字节。
 */
-
 #define xhci_read_cap_reg8(id, offset) (*(uint8_t *)(xhci_hc[id].vbase + offset))
 #define xhci_get_ptr_cap_reg8(id, offset) ((uint8_t *)(xhci_hc[id].vbase + offset))
 #define xhci_write_cap_reg8(id, offset, value) (*(uint8_t *)(xhci_hc[id].vbase + offset) = (uint8_t)value)
@@ -195,11 +198,13 @@ static int xhci_hc_stop(int id)
     // 判断是否已经停止
     if (unlikely((xhci_read_op_reg32(id, XHCI_OPS_USBSTS) & (1 << 0)) == 1))
         return 0;
-
+    io_mfence();
     xhci_write_op_reg32(id, XHCI_OPS_USBCMD, 0x00000000);
+    io_mfence();
     char timeout = 17;
     while ((xhci_read_op_reg32(id, XHCI_OPS_USBSTS) & (1 << 0)) == 0)
     {
+        io_mfence();
         usleep(1000);
         if (--timeout == 0)
             return -ETIMEDOUT;
@@ -218,9 +223,11 @@ static int xhci_hc_reset(int id)
 {
     int retval = 0;
     kdebug("usbsts=%#010lx", xhci_read_op_reg32(id, XHCI_OPS_USBSTS));
+    io_mfence();
     // 判断HCHalted是否置位
     if ((xhci_read_op_reg32(id, XHCI_OPS_USBSTS) & (1 << 0)) == 0)
     {
+        io_mfence();
         kdebug("stopping usb hc...");
         // 未置位，需要先尝试停止usb主机控制器
         retval = xhci_hc_stop(id);
@@ -230,12 +237,16 @@ static int xhci_hc_reset(int id)
     int timeout = 500; // wait 500ms
     // reset
     uint32_t cmd = xhci_read_op_reg32(id, XHCI_OPS_USBCMD);
+    io_mfence();
     kdebug("cmd=%#010lx", cmd);
     cmd |= (1 << 1);
     xhci_write_op_reg32(id, XHCI_OPS_USBCMD, cmd);
+    io_mfence();
     kdebug("after rst, sts=%#010lx", xhci_read_op_reg32(id, XHCI_OPS_USBSTS));
+    io_mfence();
     while (xhci_read_op_reg32(id, XHCI_OPS_USBCMD) & (1 << 1))
     {
+        io_mfence();
         usleep(1000);
         if (--timeout == 0)
             return -ETIMEDOUT;
@@ -259,14 +270,15 @@ static int xhci_hc_stop_legacy(int id)
         // 判断当前entry是否为legacy support entry
         if (xhci_read_cap_reg8(id, current_offset) == XHCI_XECP_ID_LEGACY)
         {
-
+            io_mfence();
             // 接管控制权
             xhci_write_cap_reg32(id, current_offset, xhci_read_cap_reg32(id, current_offset) | XHCI_XECP_LEGACY_OS_OWNED);
-
+            io_mfence();
             // 等待响应完成
             int timeout = XHCI_XECP_LEGACY_TIMEOUT;
             while ((xhci_read_cap_reg32(id, current_offset) & XHCI_XECP_LEGACY_OWNING_MASK) != XHCI_XECP_LEGACY_OS_OWNED)
             {
+                io_mfence();
                 usleep(1000);
                 if (--timeout == 0)
                 {
@@ -277,9 +289,10 @@ static int xhci_hc_stop_legacy(int id)
             // 处理完成
             return 0;
         }
-
+        io_mfence();
         // 读取下一个entry的偏移增加量
         int next_off = ((xhci_read_cap_reg32(id, current_offset) & 0xff00) >> 8) << 2;
+        io_mfence();
         // 将指针跳转到下一个entry
         current_offset = next_off ? (current_offset + next_off) : 0;
     } while (current_offset);
@@ -296,7 +309,9 @@ static int xhci_hc_stop_legacy(int id)
  */
 static int xhci_hc_start_sched(int id)
 {
+    io_mfence();
     xhci_write_op_reg32(id, XHCI_OPS_USBCMD, (1 << 0) | (1 >> 2) | (1 << 3));
+    io_mfence();
     usleep(100 * 1000);
 }
 
@@ -308,7 +323,9 @@ static int xhci_hc_start_sched(int id)
  */
 static int xhci_hc_stop_sched(int id)
 {
+    io_mfence();
     xhci_write_op_reg32(id, XHCI_OPS_USBCMD, 0x00);
+    io_mfence();
 }
 
 /**
@@ -336,13 +353,14 @@ static uint32_t xhci_hc_get_protocol_offset(int id, uint32_t list_off, const int
     do
     {
         uint32_t dw0 = xhci_read_cap_reg32(id, list_off);
+        io_mfence();
         uint32_t next_list_off = (dw0 >> 8) & 0xff;
         next_list_off = next_list_off ? (list_off + (next_list_off << 2)) : 0;
 
         if ((dw0 & 0xff) == XHCI_XECP_ID_PROTOCOL && ((dw0 & 0xff000000) >> 24) == version)
         {
             uint32_t dw2 = xhci_read_cap_reg32(id, list_off + 8);
-
+            io_mfence();
             if (offset != NULL)
                 *offset = (uint32_t)(dw2 & 0xff) - 1; // 使其转换为zero based
             if (count != NULL)
@@ -368,8 +386,9 @@ static int xhci_hc_pair_ports(int id)
 {
 
     struct xhci_caps_HCSPARAMS1_reg_t hcs1;
+    io_mfence();
     memcpy(&hcs1, xhci_get_ptr_cap_reg32(id, XHCI_CAPS_HCSPARAMS1), sizeof(struct xhci_caps_HCSPARAMS1_reg_t));
-
+    io_mfence();
     // 从hcs1获取端口数量
     xhci_hc[id].port_num = hcs1.max_ports;
 
@@ -385,15 +404,18 @@ static int xhci_hc_pair_ports(int id)
     // 寻找所有的usb2端口
     while (next_off)
     {
+        io_mfence();
         next_off = xhci_hc_get_protocol_offset(id, next_off, 2, &offset, &cnt, &protocol_flags);
+        io_mfence();
 
         if (cnt)
         {
             for (int i = 0; i < cnt; ++i)
             {
+                io_mfence();
                 xhci_hc[id].ports[offset + i].offset = xhci_hc[id].port_num_u2++;
                 xhci_hc[id].ports[offset + i].flags = XHCI_PROTOCOL_USB2;
-
+                io_mfence();
                 // usb2 high speed only
                 if (protocol_flags & 2)
                     xhci_hc[id].ports[offset + i].flags |= XHCI_PROTOCOL_HSO;
@@ -405,12 +427,15 @@ static int xhci_hc_pair_ports(int id)
     next_off = xhci_hc[id].ext_caps_off;
     while (next_off)
     {
+        io_mfence();
         next_off = xhci_hc_get_protocol_offset(id, next_off, 3, &offset, &cnt, &protocol_flags);
+        io_mfence();
 
         if (cnt)
         {
             for (int i = 0; i < cnt; ++i)
             {
+                io_mfence();
                 xhci_hc[id].ports[offset + i].offset = xhci_hc[id].port_num_u3++;
                 xhci_hc[id].ports[offset + i].flags = XHCI_PROTOCOL_USB3;
             }
@@ -424,13 +449,13 @@ static int xhci_hc_pair_ports(int id)
         {
             if (unlikely(i == j))
                 continue;
-
+            io_mfence();
             if ((xhci_hc[id].ports[i].offset == xhci_hc[id].ports[j].offset) &&
                 ((xhci_hc[id].ports[i].flags & XHCI_PROTOCOL_INFO) != (xhci_hc[id].ports[j].flags & XHCI_PROTOCOL_INFO)))
             {
                 xhci_hc[id].ports[i].paired_port_num = j;
                 xhci_hc[id].ports[i].flags |= XHCI_PROTOCOL_HAS_PAIR;
-
+                io_mfence();
                 xhci_hc[id].ports[j].paired_port_num = i;
                 xhci_hc[id].ports[j].flags |= XHCI_PROTOCOL_HAS_PAIR;
             }
@@ -440,6 +465,7 @@ static int xhci_hc_pair_ports(int id)
     // 标记所有的usb3、单独的usb2端口为激活状态
     for (int i = 0; i < xhci_hc[id].port_num; ++i)
     {
+        io_mfence();
         if (XHCI_PORT_IS_USB3(id, i) ||
             (XHCI_PORT_IS_USB2(id, i) && (!XHCI_PORT_HAS_PAIR(id, i))))
             xhci_hc[id].ports[i].flags |= XHCI_PROTOCOL_ACTIVE;
@@ -481,11 +507,12 @@ static uint64_t xhci_create_ring(int trbs)
 {
     int total_size = trbs * sizeof(struct xhci_TRB_t);
     const uint64_t vaddr = (uint64_t)kmalloc(total_size, 0);
+    io_mfence();
     memset((void *)vaddr, 0, total_size);
-
+    io_mfence();
     // 设置最后一个trb为link trb
     xhci_TRB_set_link_cmd(vaddr + total_size - sizeof(struct xhci_TRB_t));
-
+    io_mfence();
     return vaddr;
 }
 
@@ -499,18 +526,19 @@ static uint64_t xhci_create_ring(int trbs)
 static uint64_t xhci_create_event_ring(int trbs, uint64_t *ret_ring_addr)
 {
     const uint64_t table_vaddr = (const uint64_t)kmalloc(64, 0); // table支持8个segment
+    io_mfence();
     if (unlikely(table_vaddr == NULL))
         return -ENOMEM;
     memset((void *)table_vaddr, 0, 64);
 
     // 暂时只创建1个segment
     const uint64_t seg_vaddr = (const uint64_t)kmalloc(trbs * sizeof(struct xhci_TRB_t), 0);
-
+    io_mfence();
     if (unlikely(seg_vaddr == NULL))
         return -ENOMEM;
 
     memset((void *)seg_vaddr, 0, trbs * sizeof(struct xhci_TRB_t));
-
+    io_mfence();
     // 将segment地址和大小写入table
     *(uint64_t *)(table_vaddr) = virt_2_phys(seg_vaddr);
     *(uint64_t *)(table_vaddr + 8) = trbs;
@@ -522,13 +550,17 @@ static uint64_t xhci_create_event_ring(int trbs, uint64_t *ret_ring_addr)
 void xhci_hc_irq_enable(uint64_t irq_num)
 {
     int cid = xhci_find_hcid_by_irq_num(irq_num);
+    io_mfence();
     if (WARN_ON(cid == -1))
         return;
     kdebug("start msi");
+    io_mfence();
     pci_start_msi(xhci_hc[cid].pci_dev_hdr);
     kdebug("start sched");
+    io_mfence();
     xhci_hc_start_sched(cid);
     kdebug("start ports");
+    io_mfence();
     xhci_hc_start_ports(cid);
     kdebug("enabled");
 }
@@ -536,29 +568,34 @@ void xhci_hc_irq_enable(uint64_t irq_num)
 void xhci_hc_irq_disable(uint64_t irq_num)
 {
     int cid = xhci_find_hcid_by_irq_num(irq_num);
+    io_mfence();
     if (WARN_ON(cid == -1))
         return;
 
     xhci_hc_stop_sched(cid);
+    io_mfence();
     pci_disable_msi(xhci_hc[cid].pci_dev_hdr);
+    io_mfence();
 }
 
 uint64_t xhci_hc_irq_install(uint64_t irq_num, void *arg)
 {
     int cid = xhci_find_hcid_by_irq_num(irq_num);
+    io_mfence();
     if (WARN_ON(cid == -1))
         return -EINVAL;
 
     struct xhci_hc_irq_install_info_t *info = (struct xhci_hc_irq_install_info_t *)arg;
     struct msi_desc_t msi_desc;
     memset(&msi_desc, 0, sizeof(struct msi_desc_t));
-
+    io_mfence();
     msi_desc.pci_dev = (struct pci_device_structure_header_t *)xhci_hc[cid].pci_dev_hdr;
     msi_desc.assert = info->assert;
     msi_desc.edge_trigger = info->edge_trigger;
     msi_desc.processor = info->processor;
     msi_desc.pci.msi_attribute.is_64 = 1;
     // todo: QEMU是使用msix的，因此要先在pci中实现msix
+    io_mfence();
     int retval = pci_enable_msi(&msi_desc);
     kdebug("pci retval = %d", retval);
     kdebug("xhci irq %d installed.", irq_num);
@@ -569,9 +606,11 @@ void xhci_hc_irq_uninstall(uint64_t irq_num)
 {
     // todo
     int cid = xhci_find_hcid_by_irq_num(irq_num);
+    io_mfence();
     if (WARN_ON(cid == -1))
         return;
     xhci_hc_stop(cid);
+    io_mfence();
 }
 /**
  * @brief xhci主机控制器的中断处理函数
@@ -599,11 +638,14 @@ static int xhci_reset_port(const int id, const int port)
     // 相对于op寄存器基地址的偏移量
     uint64_t port_status_offset = XHCI_OPS_PRS + port * 16;
     // kdebug("to reset %d, offset=%#018lx", port, port_status_offset);
+    io_mfence();
     // 检查端口电源状态
     if ((xhci_read_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC) & (1 << 9)) == 0)
     {
         kdebug("port is power off, starting...");
+        io_mfence();
         xhci_write_cap_reg32(id, port_status_offset + XHCI_PORT_PORTSC, (1 << 9));
+        io_mfence();
         usleep(2000);
         // 检测端口是否被启用, 若未启用，则报错
         if ((xhci_read_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC) & (1 << 9)) == 0)
@@ -613,10 +655,10 @@ static int xhci_reset_port(const int id, const int port)
         }
     }
     // kdebug("port:%d, power check ok", port);
-
+    io_mfence();
     // 确保端口的status被清0
     xhci_write_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC, (1 << 9) | XHCI_PORTUSB_CHANGE_BITS);
-
+    io_mfence();
     // 重置当前端口
     if (XHCI_PORT_IS_USB3(id, port))
         xhci_write_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC, (1 << 9) | (1 << 31));
@@ -629,7 +671,9 @@ static int xhci_reset_port(const int id, const int port)
     int timeout = 200;
     while (timeout)
     {
+        io_mfence();
         uint32_t val = xhci_read_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC);
+        io_mfence();
         if (XHCI_PORT_IS_USB3(id, port) && (val & (1 << 31)) == 0)
             break;
         else if (XHCI_PORT_IS_USB2(id, port) && (val & (1 << 4)) == 0)
@@ -647,12 +691,14 @@ static int xhci_reset_port(const int id, const int port)
         // 等待恢复
         usleep(USB_TIME_RST_REC * 1000);
         uint32_t val = xhci_read_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC);
-
+        io_mfence();
         // 如果reset之后，enable bit仍然是1，那么说明reset成功
         if (val & (1 << 1))
         {
+            io_mfence();
             // 清除status change bit
             xhci_write_op_reg32(id, port_status_offset + XHCI_PORT_PORTSC, (1 << 9) | XHCI_PORTUSB_CHANGE_BITS);
+            io_mfence();
         }
         retval = 0;
     }
@@ -691,6 +737,7 @@ static int xhci_hc_start_ports(int id)
     {
         if (XHCI_PORT_IS_USB3(id, i) && XHCI_PORT_IS_ACTIVE(id, i))
         {
+            io_mfence();
             // reset该端口
             if (likely(xhci_reset_port(id, i) == 0)) // 如果端口reset成功，就获取它的描述符
                                                      // 否则，reset函数会把它给设置为未激活，并且标志配对的usb2端口是激活的
@@ -731,14 +778,18 @@ static int xhci_hc_init_intr(int id)
 
     struct xhci_caps_HCSPARAMS1_reg_t hcs1;
     struct xhci_caps_HCSPARAMS2_reg_t hcs2;
+    io_mfence();
     memcpy(&hcs1, xhci_get_ptr_cap_reg32(id, XHCI_CAPS_HCSPARAMS1), sizeof(struct xhci_caps_HCSPARAMS1_reg_t));
+    io_mfence();
     memcpy(&hcs2, xhci_get_ptr_cap_reg32(id, XHCI_CAPS_HCSPARAMS2), sizeof(struct xhci_caps_HCSPARAMS2_reg_t));
+    io_mfence();
 
     uint32_t max_segs = (1 << (uint32_t)(hcs2.ERST_Max));
     uint32_t max_interrupters = hcs1.max_intrs;
 
     // 创建 event ring
     retval = xhci_create_event_ring(4096, &xhci_hc[id].event_ring_vaddr);
+    io_mfence();
     if (unlikely((int64_t)(retval) == -ENOMEM))
         return -ENOMEM;
     xhci_hc[id].event_ring_table_vaddr = retval;
@@ -747,15 +798,21 @@ static int xhci_hc_init_intr(int id)
     xhci_hc[id].current_event_ring_cycle = 1;
 
     // 写入第0个中断寄存器组
+    io_mfence();
     xhci_write_intr_reg32(id, 0, XHCI_IR_MAN, 0x3); // 使能中断并清除pending位（这个pending位是写入1就清0的）
+    io_mfence();
     xhci_write_intr_reg32(id, 0, XHCI_IR_MOD, 0); // 关闭中断管制
+    io_mfence();
     xhci_write_intr_reg32(id, 0, XHCI_IR_TABLE_SIZE, 1); // 当前只有1个segment
+    io_mfence();
     xhci_write_intr_reg64(id, 0, XHCI_IR_DEQUEUE, virt_2_phys(xhci_hc[id].event_ring_vaddr) | (1 << 3)); // 写入dequeue寄存器，并清除busy位（写1就会清除）
+    io_mfence();
     xhci_write_intr_reg64(id, 0, XHCI_IR_TABLE_ADDR, virt_2_phys(xhci_hc[id].event_ring_table_vaddr)); // 写入table地址
+    io_mfence();
 
     // 清除状态位
     xhci_write_op_reg32(id, XHCI_OPS_USBSTS, (1 << 10) | (1 << 4) | (1 << 3) | (1 << 2));
-
+    io_mfence();
     // 开启usb中断
     // 注册中断处理程序
     struct xhci_hc_irq_install_info_t install_info;
@@ -766,7 +823,9 @@ static int xhci_hc_init_intr(int id)
     char *buf = (char *)kmalloc(16, 0);
     memset(buf, 0, 16);
     sprintk(buf, "xHCI HC%d", id);
+    io_mfence();
     irq_register(xhci_controller_irq_num[id], &install_info, &xhci_hc_irq_handler, id, &xhci_hc_intr_controller, buf);
+    io_mfence();
     kfree(buf);
 
     kdebug("xhci host controller %d: interrupt registered. irq num=%d", id, xhci_controller_irq_num[id]);
@@ -790,7 +849,7 @@ void xhci_init(struct pci_device_structure_general_device_t *dev_hdr)
 
     spin_lock(&xhci_controller_init_lock);
     kinfo("Initializing xhci host controller: bus=%#02x, device=%#02x, func=%#02x, VendorID=%#04x, irq_line=%d, irq_pin=%d", dev_hdr->header.bus, dev_hdr->header.device, dev_hdr->header.func, dev_hdr->header.Vendor_ID, dev_hdr->Interrupt_Line, dev_hdr->Interrupt_PIN);
-
+    io_mfence();
     int cid = xhci_hc_find_available_id();
     if (cid < 0)
     {
@@ -801,13 +860,14 @@ void xhci_init(struct pci_device_structure_general_device_t *dev_hdr)
     memset(&xhci_hc[cid], 0, sizeof(struct xhci_host_controller_t));
     xhci_hc[cid].controller_id = cid;
     xhci_hc[cid].pci_dev_hdr = dev_hdr;
+    io_mfence();
     pci_write_config(dev_hdr->header.bus, dev_hdr->header.device, dev_hdr->header.func, 0x4, 0x0006); // mem I/O access enable and bus master enable
-
+    io_mfence();
     // 为当前控制器映射寄存器地址空间
     xhci_hc[cid].vbase = SPECIAL_MEMOEY_MAPPING_VIRT_ADDR_BASE + XHCI_MAPPING_OFFSET + 65536 * xhci_hc[cid].controller_id;
     // kdebug("dev_hdr->BAR0 & (~0xf)=%#018lx", dev_hdr->BAR0 & (~0xf));
     mm_map_phys_addr(xhci_hc[cid].vbase, dev_hdr->BAR0 & (~0xf), 65536, PAGE_KERNEL_PAGE | PAGE_PWT | PAGE_PCD, true);
-
+    io_mfence();
     // 读取xhci控制寄存器
     uint16_t iversion = *(uint16_t *)(xhci_hc[cid].vbase + XHCI_CAPS_HCIVERSION);
 
@@ -824,9 +884,11 @@ void xhci_init(struct pci_device_structure_general_device_t *dev_hdr)
     // kdebug("hcc1.xECP=%#010lx", hcc1.xECP);
     // 计算operational registers的地址
     xhci_hc[cid].vbase_op = xhci_hc[cid].vbase + xhci_read_cap_reg8(cid, XHCI_CAPS_CAPLENGTH);
-
+    io_mfence();
     xhci_hc[cid].db_offset = xhci_read_cap_reg32(cid, XHCI_CAPS_DBOFF) & (~0x3); // bits [1:0] reserved
+    io_mfence();
     xhci_hc[cid].rts_offset = xhci_read_cap_reg32(cid, XHCI_CAPS_RTSOFF) & (~0x1f); // bits [4:0] reserved.
+    io_mfence();
 
     xhci_hc[cid].ext_caps_off = 1UL * (hcc1.xECP) * 4;
     xhci_hc[cid].context_size = (hcc1.csz) ? 64 : 32;
@@ -845,25 +907,27 @@ void xhci_init(struct pci_device_structure_general_device_t *dev_hdr)
         pci_write_config(dev_hdr->header.bus, dev_hdr->header.device, dev_hdr->header.func, 0xd8, 0xffffffff);
         pci_write_config(dev_hdr->header.bus, dev_hdr->header.device, dev_hdr->header.func, 0xd0, 0xffffffff);
     }
-
+    io_mfence();
     // 关闭legacy支持
     FAIL_ON_TO(xhci_hc_stop_legacy(cid), failed);
-
+    io_mfence();
     // 重置xhci控制器
     FAIL_ON_TO(xhci_hc_reset(cid), failed);
+    io_mfence();
     // 端口配对
     FAIL_ON_TO(xhci_hc_pair_ports(cid), failed);
+    io_mfence();
 
     // ========== 设置USB host controller =========
     // 获取页面大小
     kdebug("ops pgsize=%#010lx", xhci_read_op_reg32(cid, XHCI_OPS_PAGESIZE));
     xhci_hc[cid].page_size = (xhci_read_op_reg32(cid, XHCI_OPS_PAGESIZE) & 0xffff) << 12;
     kdebug("page size=%d", xhci_hc[cid].page_size);
-
+    io_mfence();
     // 获取设备上下文空间
     xhci_hc[cid].dcbaap_vaddr = (uint64_t)kmalloc(2048, 0); // 分配2KB的设备上下文地址数组空间
     memset((void *)xhci_hc[cid].dcbaap_vaddr, 0, 2048);
-
+    io_mfence();
     kdebug("dcbaap_vaddr=%#018lx", xhci_hc[cid].dcbaap_vaddr);
     if (unlikely(!xhci_is_aligned64(xhci_hc[cid].dcbaap_vaddr))) // 地址不是按照64byte对齐
     {
@@ -872,7 +936,7 @@ void xhci_init(struct pci_device_structure_general_device_t *dev_hdr)
     }
     // 写入dcbaap
     xhci_write_op_reg64(cid, XHCI_OPS_DCBAAP, virt_2_phys(xhci_hc[cid].dcbaap_vaddr));
-
+    io_mfence();
     // 创建command ring
     xhci_hc[cid].cmd_ring_vaddr = xhci_create_ring(XHCI_CMND_RING_TRBS);
     if (unlikely(!xhci_is_aligned64(xhci_hc[cid].cmd_ring_vaddr))) // 地址不是按照64byte对齐
@@ -883,17 +947,21 @@ void xhci_init(struct pci_device_structure_general_device_t *dev_hdr)
 
     // 设置初始cycle bit为1
     xhci_hc[cid].cmd_trb_cycle = XHCI_TRB_CYCLE_ON;
-
+    io_mfence();
     // 写入command ring控制寄存器
     xhci_write_op_reg64(cid, XHCI_OPS_CRCR, virt_2_phys(xhci_hc[cid].cmd_ring_vaddr) | xhci_hc[cid].cmd_trb_cycle);
     // 写入配置寄存器
     uint32_t max_slots = hcs1.max_slots;
     kdebug("max slots = %d", max_slots);
+    io_mfence();
     xhci_write_op_reg32(cid, XHCI_OPS_CONFIG, max_slots);
+    io_mfence();
     // 写入设备通知控制寄存器
     xhci_write_op_reg32(cid, XHCI_OPS_DNCTRL, (1 << 1)); // 目前只有N1被支持
+    io_mfence();
 
     FAIL_ON_TO(xhci_hc_init_intr(cid), failed_free_dyn);
+    io_mfence();
     ++xhci_ctrl_count;
     spin_unlock(&xhci_controller_init_lock);
     return;
@@ -912,9 +980,10 @@ failed_free_dyn:; // 释放动态申请的内存
         kfree((void *)xhci_hc[cid].event_ring_vaddr);
 
 failed:;
+    io_mfence();
     // 取消地址映射
     mm_unmap(xhci_hc[cid].vbase, 65536);
-
+    io_mfence();
     // 清空数组
     memset((void *)&xhci_hc[cid], 0, sizeof(struct xhci_host_controller_t));
 
@@ -922,3 +991,4 @@ failed_exceed_max:;
     kerror("Failed to initialize controller: bus=%d, dev=%d, func=%d", dev_hdr->header.bus, dev_hdr->header.device, dev_hdr->header.func);
     spin_unlock(&xhci_controller_init_lock);
 }
+#pragma GCC pop_options

kernel/driver/usb/xhci/xhci.h

@@ -2,7 +2,7 @@
 #include <driver/usb/usb.h>
 #include <driver/pci/pci.h>
 #include <driver/pci/msi.h>
-
+// #pragma GCC optimize("O0")
 #define XHCI_MAX_HOST_CONTROLLERS 4 // 本驱动程序最大支持4个xhci root hub controller
 #define XHCI_MAX_ROOT_HUB_PORTS 128 // 本驱动程序最大支持127个root hub 端口（第0个保留）
 

kernel/exception/gate.h

@@ -12,6 +12,8 @@
 #include <common/kprint.h>
 #include <mm/mm.h>
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 //描述符表的结构体
 struct desc_struct
 {
@@ -185,3 +187,5 @@ void set_tss64(unsigned int *Table, unsigned long rsp0, unsigned long rsp1, unsi
     *(unsigned long *)(Table + 21) = ist7;
 }
 #endif
+
+#pragma GCC pop_options

kernel/exception/irq.c

@@ -1,7 +1,8 @@
+
 #include "irq.h"
 #include <common/errno.h>
 
-// 对进行
+
 #if _INTR_8259A_
 #include <driver/interrupt/8259A/8259A.h>
 #else
@@ -13,6 +14,8 @@
 #include "gate.h"
 #include <mm/slab.h>
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 // 保存函数调用现场的寄存器
 #define SAVE_ALL_REGS       \
     "cld; \n\t"             \
@@ -254,3 +257,4 @@ void irq_init()
 
 #endif
 }
+#pragma GCC optimize("O0")

kernel/exception/irq.h

@@ -10,10 +10,11 @@
  */
 
 #pragma once
-
 #include <common/glib.h>
 
 #include <process/ptrace.h>
+#pragma GCC push_options
+#pragma GCC optimize ("O0")
 
 #define IRQ_NUM 24
 #define SMP_IRQ_NUM 10
@@ -165,3 +166,4 @@ int irq_unregister(ul irq_num);
  * @brief 初始化中断模块
  */
 void irq_init();
+#pragma GCC pop_options

kernel/main.c

@@ -34,11 +34,12 @@
 #include <driver/video/video.h>
 
 #include <driver/interrupt/apic/apic_timer.h>
-
+#pragma GCC push_options
+#pragma GCC optimize("O3")
 unsigned int *FR_address = (unsigned int *)0xb8000; //帧缓存区的地址
 ul bsp_idt_size, bsp_gdt_size;
 
-struct memory_desc memory_management_struct = {{0}, 0};
+
 // struct Global_Memory_Descriptor memory_management_struct = {{0}, 0};
 void test_slab();
 
@@ -126,12 +127,19 @@ void system_initialize()
     current_pcb->preempt_count = 0;
     // 先初始化系统调用模块
     syscall_init();
+    io_mfence();
     //  再初始化进程模块。顺序不能调转
     sched_init();
+    io_mfence();
 
     timer_init();
 
+    // 这里必须加内存屏障，否则会出错
+    io_mfence();
     smp_init();
+    io_mfence();
+
+    
     cpu_init();
     ps2_keyboard_init();
     // ps2_mouse_init();
@@ -144,20 +152,24 @@ void system_initialize()
 
     // process_init();
     HPET_init();
+    io_mfence();
     HPET_measure_freq();
+    io_mfence();
     // current_pcb->preempt_count = 0;
     // kdebug("cpu_get_core_crysral_freq()=%ld", cpu_get_core_crysral_freq());
 
     process_init();
     // 对显示模块进行高级初始化，启用double buffer
     video_init(true);
+    io_mfence();
 
     // fat32_init();
     HPET_enable();
 
+    io_mfence();
     // 系统初始化到此结束，剩下的初始化功能应当放在初始内核线程中执行
     apic_timer_init();
-    
+    io_mfence();
 }
 
 //操作系统内核从这里开始执行
@@ -181,8 +193,9 @@ void Start_Kernel(void)
     mb2_magic &= 0xffffffff;
     multiboot2_magic = (uint)mb2_magic;
     multiboot2_boot_info_addr = mb2_info + PAGE_OFFSET;
-
+    io_mfence();
     system_initialize();
+    io_mfence();
 
     while (1)
         hlt();
@@ -191,5 +204,7 @@ void Start_Kernel(void)
 void ignore_int()
 {
     kwarn("Unknown interrupt or fault at RIP.\n");
-    while(1);
+    while (1)
+        ;
 }
+#pragma GCC pop_options

kernel/mm/mm.c

@@ -11,7 +11,10 @@
 static ul Total_Memory = 0;
 static ul total_2M_pages = 0;
 static ul root_page_table_phys_addr = 0; // 内核层根页表的物理地址
+// #pragma GCC push_options
+// #pragma GCC optimize("O3")
 
+struct memory_desc memory_management_struct = {{0}, 0};
 /**
  * @brief 虚拟地址长度所需要的entry数量
  *
@@ -82,9 +85,10 @@ void mm_init()
     int count;
 
     multiboot2_iter(multiboot2_get_memory, mb2_mem_info, &count);
-
+    io_mfence();
     for (int i = 0; i < count; ++i)
     {
+        io_mfence();
         //可用的内存
         if (mb2_mem_info->type == 1)
             Total_Memory += mb2_mem_info->len;
@@ -103,12 +107,12 @@ void mm_init()
     printk("[ INFO ] Total amounts of RAM : %ld bytes\n", Total_Memory);
 
     // 计算有效内存页数
-
+    io_mfence();
     for (int i = 0; i < memory_management_struct.len_e820; ++i)
     {
         if (memory_management_struct.e820[i].type != 1)
             continue;
-
+        io_mfence();
         // 将内存段的起始物理地址按照2M进行对齐
         ul addr_start = PAGE_2M_ALIGN(memory_management_struct.e820[i].BaseAddr);
         // 将内存段的终止物理地址的低2M区域清空，以实现对齐
@@ -117,7 +121,7 @@ void mm_init()
         // 内存段不可用
         if (addr_end <= addr_start)
             continue;
-
+        io_mfence();
         total_2M_pages += ((addr_end - addr_start) >> PAGE_2M_SHIFT);
     }
     kinfo("Total amounts of 2M pages : %ld.", total_2M_pages);
@@ -126,16 +130,18 @@ void mm_init()
     ul max_addr = memory_management_struct.e820[memory_management_struct.len_e820].BaseAddr + memory_management_struct.e820[memory_management_struct.len_e820].Length;
     // 初始化mms的bitmap
     // bmp的指针指向截止位置的4k对齐的上边界（防止修改了别的数据）
+    io_mfence();
     memory_management_struct.bmp = (unsigned long *)((memory_management_struct.start_brk + PAGE_4K_SIZE - 1) & PAGE_4K_MASK);
     memory_management_struct.bits_size = max_addr >> PAGE_2M_SHIFT;                                                                                         // 物理地址空间的最大页面数
     memory_management_struct.bmp_len = (((unsigned long)(max_addr >> PAGE_2M_SHIFT) + sizeof(unsigned long) * 8 - 1) / 8) & (~(sizeof(unsigned long) - 1)); // bmp由多少个unsigned long变量组成
+    io_mfence();
 
     // 初始化bitmap， 先将整个bmp空间全部置位。稍后再将可用物理内存页复位。
     memset(memory_management_struct.bmp, 0xff, memory_management_struct.bmp_len);
-
+    io_mfence();
+    kdebug("1212112");
     // 初始化内存页结构
     // 将页结构映射于bmp之后
-
     memory_management_struct.pages_struct = (struct Page *)(((unsigned long)memory_management_struct.bmp + memory_management_struct.bmp_len + PAGE_4K_SIZE - 1) & PAGE_4K_MASK);
 
     memory_management_struct.count_pages = max_addr >> PAGE_2M_SHIFT;
@@ -143,18 +149,25 @@ void mm_init()
     // 将pages_struct全部清空，以备后续初始化
     memset(memory_management_struct.pages_struct, 0x00, memory_management_struct.pages_struct_len); // init pages memory
 
+    kdebug("ffff");
+    io_mfence();
     // 初始化内存区域
     memory_management_struct.zones_struct = (struct Zone *)(((ul)memory_management_struct.pages_struct + memory_management_struct.pages_struct_len + PAGE_4K_SIZE - 1) & PAGE_4K_MASK);
+    io_mfence();
     // 由于暂时无法计算zone结构体的数量，因此先将其设为0
     memory_management_struct.count_zones = 0;
+    io_mfence();
     // zones-struct 成员变量暂时按照5个来计算
     memory_management_struct.zones_struct_len = (10 * sizeof(struct Zone) + sizeof(ul) - 1) & (~(sizeof(ul) - 1));
+    io_mfence();
     memset(memory_management_struct.zones_struct, 0x00, memory_management_struct.zones_struct_len);
 
     // ==== 遍历e820数组，完成成员变量初始化工作 ===
 
+    kdebug("ddd");
     for (int i = 0; i < memory_management_struct.len_e820; ++i)
     {
+        io_mfence();
         if (memory_management_struct.e820[i].type != 1) // 不是操作系统可以使用的物理内存
             continue;
         ul addr_start = PAGE_2M_ALIGN(memory_management_struct.e820[i].BaseAddr);
@@ -200,7 +213,7 @@ void mm_init()
 
     // 初始化0~2MB的物理页
     // 由于这个区间的内存由多个内存段组成，因此不会被以上代码初始化，需要我们手动配置page[0]。
-
+    io_mfence();
     memory_management_struct.pages_struct->zone = memory_management_struct.zones_struct;
     memory_management_struct.pages_struct->addr_phys = 0UL;
     set_page_attr(memory_management_struct.pages_struct, PAGE_PGT_MAPPED | PAGE_KERNEL_INIT | PAGE_KERNEL);
@@ -216,27 +229,13 @@ void mm_init()
     ZONE_NORMAL_INDEX = 0;
     ZONE_UNMAPPED_INDEX = 0;
 
-    /*
-    for (int i = 0; i < memory_management_struct.count_zones; ++i)
-    {
-        struct Zone *z = memory_management_struct.zones_struct + i;
-        // printk_color(ORANGE, BLACK, "zone_addr_start:%#18lx, zone_addr_end:%#18lx, zone_length:%#18lx, pages_group:%#18lx, count_pages:%#18lx\n",
-        //             z->zone_addr_start, z->zone_addr_end, z->zone_length, z->pages_group, z->count_pages);
     
-        // 1GB以上的内存空间不做映射
-        // if (z->zone_addr_start >= 0x100000000 && (!ZONE_UNMAPPED_INDEX))
-        //    ZONE_UNMAPPED_INDEX = i;
-    }
-    */
     // kdebug("ZONE_DMA_INDEX=%d\tZONE_NORMAL_INDEX=%d\tZONE_UNMAPPED_INDEX=%d", ZONE_DMA_INDEX, ZONE_NORMAL_INDEX, ZONE_UNMAPPED_INDEX);
     //  设置内存页管理结构的地址，预留了一段空间，防止内存越界。
     memory_management_struct.end_of_struct = (ul)((ul)memory_management_struct.zones_struct + memory_management_struct.zones_struct_len + sizeof(long) * 32) & (~(sizeof(long) - 1));
 
-    // printk_color(ORANGE, BLACK, "code_start:%#18lx, code_end:%#18lx, data_end:%#18lx, kernel_end:%#18lx, end_of_struct:%#18lx\n",
-    //              memory_management_struct.kernel_code_start, memory_management_struct.kernel_code_end, memory_management_struct.kernel_data_end, memory_management_struct.kernel_end, memory_management_struct.end_of_struct);
     
     // 初始化内存管理单元结构所占的物理页的结构体
-
     ul mms_max_page = (virt_2_phys(memory_management_struct.end_of_struct) >> PAGE_2M_SHIFT); // 内存管理单元所占据的序号最大的物理页
     // kdebug("mms_max_page=%ld", mms_max_page);
 
@@ -245,6 +244,7 @@ void mm_init()
     // 第0个page已经在上方配置
     for (ul j = 1; j <= mms_max_page; ++j)
     {
+        barrier();
         tmp_page = memory_management_struct.pages_struct + j;
         page_init(tmp_page, PAGE_PGT_MAPPED | PAGE_KERNEL | PAGE_KERNEL_INIT);
         page_num = tmp_page->addr_phys >> PAGE_2M_SHIFT;
@@ -278,6 +278,7 @@ unsigned long page_init(struct Page *page, ul flags)
     if ((!page->ref_counts) || (page->attr & PAGE_SHARED))
     {
         ++page->ref_counts;
+        barrier();
         ++page->zone->total_pages_link;
     }
     return 0;
@@ -977,3 +978,4 @@ int8_t mm_is_2M_page(uint64_t paddr)
         return 1;
     else return 0;
 }
+// #pragma GCC pop_options

kernel/mm/mm.h

@@ -145,6 +145,7 @@
     do                              \
     {                               \
         ul tmp;                     \
+        io_mfence();\
         __asm__ __volatile__(       \
             "movq %%cr3, %0\n\t"    \
             "movq %0, %%cr3\n\t"    \

kernel/mm/slab.c

@@ -348,9 +348,9 @@ ul slab_init()
     kinfo("Initializing SLAB...");
     // 将slab的内存池空间放置在mms的后方
     ul tmp_addr = memory_management_struct.end_of_struct;
-
     for (int i = 0; i < 16; ++i)
     {
+        io_mfence();
         spin_init(&kmalloc_cache_group[i].lock);
         // 将slab内存池对象的空间放置在mms的后面，并且预留4个unsigned long 的空间以防止内存越界
         kmalloc_cache_group[i].cache_pool_entry = (struct slab_obj *)memory_management_struct.end_of_struct;
@@ -370,7 +370,7 @@ ul slab_init()
 
         // bmp后方预留4个unsigned long的空间防止内存越界,且按照8byte进行对齐
         memory_management_struct.end_of_struct = (ul)(memory_management_struct.end_of_struct + kmalloc_cache_group[i].cache_pool_entry->bmp_len + (sizeof(ul) << 2)) & (~(sizeof(ul) - 1));
-
+        io_mfence();
         // @todo：此处可优化，直接把所有位设置为0，然后再对部分不存在对应的内存对象的位设置为1
         memset(kmalloc_cache_group[i].cache_pool_entry->bmp, 0xff, kmalloc_cache_group[i].cache_pool_entry->bmp_len);
         for (int j = 0; j < kmalloc_cache_group[i].cache_pool_entry->bmp_count; ++j)
@@ -378,6 +378,7 @@ ul slab_init()
 
         kmalloc_cache_group[i].count_total_using = 0;
         kmalloc_cache_group[i].count_total_free = kmalloc_cache_group[i].cache_pool_entry->count_free;
+        io_mfence();
     }
 
     struct Page *page = NULL;
@@ -388,13 +389,17 @@ ul slab_init()
     ul page_num = 0;
     for (int i = PAGE_2M_ALIGN(virt_2_phys(tmp_addr)) >> PAGE_2M_SHIFT; i <= tmp_page_mms_end; ++i)
     {
+
         page = memory_management_struct.pages_struct + i;
         page_num = page->addr_phys >> PAGE_2M_SHIFT;
         *(memory_management_struct.bmp + (page_num >> 6)) |= (1UL << (page_num % 64));
         ++page->zone->count_pages_using;
+        io_mfence();
         --page->zone->count_pages_free;
         page_init(page, PAGE_KERNEL_INIT | PAGE_KERNEL | PAGE_PGT_MAPPED);
+
     }
+    io_mfence();
 
     // 为slab内存池对象分配内存空间
     ul *virt = NULL;
@@ -406,8 +411,11 @@ ul slab_init()
         page = Virt_To_2M_Page(virt);
 
         page_num = page->addr_phys >> PAGE_2M_SHIFT;
+
         *(memory_management_struct.bmp + (page_num >> 6)) |= (1UL << (page_num % 64));
+
         ++page->zone->count_pages_using;
+        io_mfence();    // 该位置必须加一个mfence，否则O3优化运行时会报错
         --page->zone->count_pages_free;
         page_init(page, PAGE_PGT_MAPPED | PAGE_KERNEL | PAGE_KERNEL_INIT);
 

kernel/process/process.c

@@ -21,6 +21,9 @@
 
 #include <ktest/ktest.h>
 
+// #pragma GCC push_options
+// #pragma GCC optimize("O0")
+
 spinlock_t process_global_pid_write_lock; // 增加pid的写锁
 long process_global_pid = 1;              // 系统中最大的pid
 
@@ -100,6 +103,7 @@ uint64_t process_exit_mm(struct process_control_block *pcb);
 uint64_t process_copy_thread(uint64_t clone_flags, struct process_control_block *pcb, uint64_t stack_start, uint64_t stack_size, struct pt_regs *current_regs);
 
 void process_exit_thread(struct process_control_block *pcb);
+
 /**
  * @brief 切换进程
  *
@@ -108,7 +112,8 @@ void process_exit_thread(struct process_control_block *pcb);
  * 由于程序在进入内核的时候已经保存了寄存器，因此这里不需要保存寄存器。
  * 这里切换fs和gs寄存器
  */
-
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 void __switch_to(struct process_control_block *prev, struct process_control_block *next)
 {
     initial_tss[proc_current_cpu_id].rsp0 = next->thread->rbp;
@@ -124,6 +129,7 @@ void __switch_to(struct process_control_block *prev, struct process_control_bloc
     __asm__ __volatile__("movq	%0,	%%fs \n\t" ::"a"(next->thread->fs));
     __asm__ __volatile__("movq	%0,	%%gs \n\t" ::"a"(next->thread->gs));
 }
+#pragma GCC pop_options
 
 /**
  * @brief 打开要执行的程序文件
@@ -299,6 +305,8 @@ load_elf_failed:;
  * @param envp 环境变量
  * @return ul 错误码
  */
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 ul do_execve(struct pt_regs *regs, char *path, char *argv[], char *envp[])
 {
 
@@ -403,6 +411,7 @@ ul do_execve(struct pt_regs *regs, char *path, char *argv[], char *envp[])
 exec_failed:;
     process_do_exit(tmp);
 }
+#pragma GCC pop_options
 
 /**
  * @brief 内核init进程
@@ -410,6 +419,8 @@ exec_failed:;
  * @param arg
  * @return ul 参数
  */
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 ul initial_kernel_thread(ul arg)
 {
     // kinfo("initial proc running...\targ:%#018lx", arg);
@@ -446,6 +457,7 @@ ul initial_kernel_thread(ul arg)
     current_pcb->thread->rip = (ul)ret_from_system_call;
     current_pcb->thread->rsp = (ul)current_pcb + STACK_SIZE - sizeof(struct pt_regs);
     current_pcb->thread->fs = USER_DS | 0x3;
+    barrier();
     current_pcb->thread->gs = USER_DS | 0x3;
 
     // 主动放弃内核线程身份
@@ -467,7 +479,7 @@ ul initial_kernel_thread(ul arg)
 
     return 1;
 }
-
+#pragma GCC pop_options
 /**
  * @brief 当子进程退出后向父进程发送通知
  *
@@ -517,23 +529,29 @@ ul process_do_exit(ul code)
 int kernel_thread(unsigned long (*fn)(unsigned long), unsigned long arg, unsigned long flags)
 {
     struct pt_regs regs;
+    barrier();
     memset(&regs, 0, sizeof(regs));
-
+    barrier();
     // 在rbx寄存器中保存进程的入口地址
     regs.rbx = (ul)fn;
     // 在rdx寄存器中保存传入的参数
     regs.rdx = (ul)arg;
-
+    barrier();
     regs.ds = KERNEL_DS;
+    barrier();
     regs.es = KERNEL_DS;
+    barrier();
     regs.cs = KERNEL_CS;
+    barrier();
     regs.ss = KERNEL_DS;
+    barrier();
 
     // 置位中断使能标志位
     regs.rflags = (1 << 9);
-
+    barrier();
     // rip寄存器指向内核线程的引导程序
     regs.rip = (ul)kernel_thread_func;
+    barrier();
     // kdebug("kernel_thread_func=%#018lx", kernel_thread_func);
     // kdebug("&kernel_thread_func=%#018lx", &kernel_thread_func);
     // kdebug("1111\tregs.rip = %#018lx", regs.rip);
@@ -577,23 +595,34 @@ void process_init()
     for (int i = 256; i < 512; ++i)
     {
         uint64_t *tmp = idle_pml4t_vaddr + i;
+        barrier();
         if (*tmp == 0)
         {
             void *pdpt = kmalloc(PAGE_4K_SIZE, 0);
+            barrier();
             memset(pdpt, 0, PAGE_4K_SIZE);
+            barrier();
             set_pml4t(tmp, mk_pml4t(virt_2_phys(pdpt), PAGE_KERNEL_PGT));
         }
     }
+    barrier();
+
+    flush_tlb();
     /*
     kdebug("initial_thread.rbp=%#018lx", initial_thread.rbp);
     kdebug("initial_tss[0].rsp1=%#018lx", initial_tss[0].rsp1);
     kdebug("initial_tss[0].ist1=%#018lx", initial_tss[0].ist1);
 */
     // 初始化pid的写锁
+
     spin_init(&process_global_pid_write_lock);
+
     // 初始化进程的循环链表
     list_init(&initial_proc_union.pcb.list);
+    barrier();
     kernel_thread(initial_kernel_thread, 10, CLONE_FS | CLONE_SIGNAL); // 初始化内核线程
+    barrier();
+
     initial_proc_union.pcb.state = PROC_RUNNING;
     initial_proc_union.pcb.preempt_count = 0;
     initial_proc_union.pcb.cpu_id = 0;
@@ -617,6 +646,7 @@ unsigned long do_fork(struct pt_regs *regs, unsigned long clone_flags, unsigned
 
     // 为新的进程分配栈空间，并将pcb放置在底部
     tsk = (struct process_control_block *)kmalloc(STACK_SIZE, 0);
+    barrier();
 
     if (tsk == NULL)
     {
@@ -624,13 +654,17 @@ unsigned long do_fork(struct pt_regs *regs, unsigned long clone_flags, unsigned
         return retval;
     }
 
+    barrier();
     memset(tsk, 0, sizeof(struct process_control_block));
+    io_mfence();
     // 将当前进程的pcb复制到新的pcb内
     memcpy(tsk, current_pcb, sizeof(struct process_control_block));
+    io_mfence();
 
     // 初始化进程的循环链表结点
     list_init(&tsk->list);
 
+    io_mfence();
     // 判断是否为内核态调用fork
     if (current_pcb->flags & PF_KTHREAD && stack_start != 0)
         tsk->flags |= PF_KFORK;
@@ -641,11 +675,14 @@ unsigned long do_fork(struct pt_regs *regs, unsigned long clone_flags, unsigned
     // 增加全局的pid并赋值给新进程的pid
     spin_lock(&process_global_pid_write_lock);
     tsk->pid = process_global_pid++;
-
+    barrier();
     // 加入到进程链表中
     tsk->next_pcb = initial_proc_union.pcb.next_pcb;
+    barrier();
     initial_proc_union.pcb.next_pcb = tsk;
+    barrier();
     tsk->parent_pcb = current_pcb;
+    barrier();
 
     spin_unlock(&process_global_pid_write_lock);
 
@@ -654,7 +691,7 @@ unsigned long do_fork(struct pt_regs *regs, unsigned long clone_flags, unsigned
 
     tsk->parent_pcb = current_pcb;
     wait_queue_init(&tsk->wait_child_proc_exit, NULL);
-
+    barrier();
     list_init(&tsk->list);
 
     retval = -ENOMEM;
@@ -1047,6 +1084,7 @@ uint64_t process_copy_thread(uint64_t clone_flags, struct process_control_block
 
         child_regs = (struct pt_regs *)(((uint64_t)pcb) + STACK_SIZE - size);
         memcpy(child_regs, (void *)current_regs, size);
+        barrier();
         // 然后重写新的栈中，每个栈帧的rbp值
         process_rewrite_rbp(child_regs, pcb);
     }
@@ -1054,6 +1092,7 @@ uint64_t process_copy_thread(uint64_t clone_flags, struct process_control_block
     {
         child_regs = (struct pt_regs *)((uint64_t)pcb + STACK_SIZE - sizeof(struct pt_regs));
         memcpy(child_regs, current_regs, sizeof(struct pt_regs));
+        barrier();
         child_regs->rsp = stack_start;
     }
 
@@ -1088,3 +1127,5 @@ uint64_t process_copy_thread(uint64_t clone_flags, struct process_control_block
 void process_exit_thread(struct process_control_block *pcb)
 {
 }
+
+// #pragma GCC pop_options

kernel/process/process.h

@@ -9,7 +9,6 @@
  */
 
 #pragma once
-
 #include <common/cpu.h>
 #include <common/glib.h>
 #include <mm/mm.h>
@@ -19,6 +18,8 @@
 #include <filesystem/VFS/VFS.h>
 #include <common/wait_queue.h>
 
+
+
 // 进程最大可拥有的文件描述符数量
 #define PROC_MAX_FD_NUM 16
 
@@ -213,17 +214,21 @@ struct tss_struct
 		.io_map_base_addr = 0                                             \
 	}
 
+#pragma GCC push_options
+#pragma GCC optimize("O0")
 // 获取当前的pcb
 struct process_control_block *get_current_pcb()
 {
 	struct process_control_block *current = NULL;
 	// 利用了当前pcb和栈空间总大小为32k大小对齐，将rsp低15位清空，即可获得pcb的起始地址
+	barrier();
 	__asm__ __volatile__("andq %%rsp, %0   \n\t"
 						 : "=r"(current)
 						 : "0"(~32767UL));
+	barrier();
 	return current;
 };
-
+#pragma GCC pop_options
 #define current_pcb get_current_pcb()
 
 #define GET_CURRENT_PCB    \
@@ -350,7 +355,7 @@ int kernel_thread(unsigned long (*fn)(unsigned long), unsigned long arg, unsigne
 		asm volatile("movq %0, %%cr3	\n\t" ::"r"(next_pcb->mm->pgd) \
 					 : "memory");                                      \
 	} while (0)
-// flush_tlb();                                                   \
+// flush_tlb();                                                   
 
 // 获取当前cpu id
 #define proc_current_cpu_id (current_pcb->cpu_id)

kernel/sched/sched.c

@@ -3,6 +3,7 @@
 #include <driver/video/video.h>
 #include <common/spinlock.h>
 
+
 struct sched_queue_t sched_cfs_ready_queue[MAX_CPU_NUM]; // 就绪队列
 
 /**

kernel/sched/sched.h

@@ -46,3 +46,4 @@ void sched_init();
  * 
  */
 void sched_update_jiffies();
+

kernel/smp/smp.c

@@ -30,43 +30,51 @@ void smp_init()
 
     // kdebug("processor num=%d", total_processor_num);
     for (int i = 0; i < total_processor_num; ++i)
+    {
+        io_mfence();
         proc_local_apic_structs[i] = (struct acpi_Processor_Local_APIC_Structure_t *)(tmp_vaddr[i]);
+    }
 
     //*(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 *)phys_2_virt(0x20000), _apu_boot_start, (unsigned long)&_apu_boot_end - (unsigned long)&_apu_boot_start);
-
+    io_mfence();
     // 设置多核IPI中断门
     for (int i = 200; i < 210; ++i)
         set_intr_gate(i, 0, SMP_interrupt_table[i - 200]);
-
     memset((void *)SMP_IPI_desc, 0, sizeof(irq_desc_t) * SMP_IRQ_NUM);
     
+    io_mfence();
+
     // 注册接收bsp处理器的hpet中断转发的处理函数
     ipi_regiserIPI(0xc8, NULL, &ipi_0xc8_handler, NULL, NULL, "IPI 0xc8");
-
+    io_mfence();
     ipi_send_IPI(DEST_PHYSICAL, IDLE, ICR_LEVEL_DE_ASSERT, EDGE_TRIGGER, 0x00, ICR_INIT, ICR_ALL_EXCLUDE_Self, true, 0x00);
     kdebug("total_processor_num=%d", total_processor_num);
     for (int i = 1; i < total_processor_num; ++i) // i从1开始，不初始化bsp
     {
+        io_mfence();
         if (proc_local_apic_structs[i]->ACPI_Processor_UID == 0)
             --total_processor_num;
+        io_mfence();
         if (proc_local_apic_structs[i]->local_apic_id > total_processor_num)
         {
             --total_processor_num;
-                continue;}
+            continue;
+        }
         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]->local_apic_id, proc_local_apic_structs[i]->flags);
-
+        io_mfence();
         spin_lock(&multi_core_starting_lock);
         preempt_enable(); // 由于ap处理器的pcb与bsp的不同，因此ap处理器放锁时，bsp的自旋锁持有计数不会发生改变,需要手动恢复preempt count
         current_starting_cpu = proc_local_apic_structs[i]->local_apic_id;
-
+        io_mfence();
-
         // 为每个AP处理器分配栈空间
         cpu_core_info[current_starting_cpu].stack_start = (uint64_t)kmalloc(STACK_SIZE, 0) + STACK_SIZE;
         cpu_core_info[current_starting_cpu].ist_stack_start = (uint64_t)(kmalloc(STACK_SIZE, 0)) + STACK_SIZE;
+        io_mfence();
         memset((void *)cpu_core_info[current_starting_cpu].stack_start - STACK_SIZE, 0, STACK_SIZE);
         memset((void *)cpu_core_info[current_starting_cpu].ist_stack_start - STACK_SIZE, 0, STACK_SIZE);
+        io_mfence();
 
         // 设置ap处理器的中断栈及内核栈中的cpu_id
         ((struct process_control_block *)(cpu_core_info[current_starting_cpu].stack_start - STACK_SIZE))->cpu_id = proc_local_apic_structs[i]->local_apic_id;
@@ -77,18 +85,18 @@ void smp_init()
         memset(&initial_tss[current_starting_cpu], 0, sizeof(struct tss_struct));
 
         set_tss_descriptor(10 + (current_starting_cpu * 2), (void *)(cpu_core_info[current_starting_cpu].tss_vaddr));
-
+        io_mfence();
         set_tss64((uint *)cpu_core_info[current_starting_cpu].tss_vaddr, cpu_core_info[current_starting_cpu].stack_start, cpu_core_info[current_starting_cpu].stack_start, cpu_core_info[current_starting_cpu].stack_start,
                   cpu_core_info[current_starting_cpu].ist_stack_start, cpu_core_info[current_starting_cpu].ist_stack_start, cpu_core_info[current_starting_cpu].ist_stack_start, cpu_core_info[current_starting_cpu].ist_stack_start, cpu_core_info[current_starting_cpu].ist_stack_start, cpu_core_info[current_starting_cpu].ist_stack_start, cpu_core_info[current_starting_cpu].ist_stack_start);
-
+        io_mfence();
         // 连续发送两次start-up IPI
         ipi_send_IPI(DEST_PHYSICAL, IDLE, ICR_LEVEL_DE_ASSERT, EDGE_TRIGGER, 0x20, ICR_Start_up, ICR_No_Shorthand, true, proc_local_apic_structs[i]->local_apic_id);
+        io_mfence();
         ipi_send_IPI(DEST_PHYSICAL, IDLE, ICR_LEVEL_DE_ASSERT, EDGE_TRIGGER, 0x20, ICR_Start_up, ICR_No_Shorthand, true, proc_local_apic_structs[i]->local_apic_id);
     }
-
+    io_mfence();
     while (num_cpu_started != total_processor_num)
-        __asm__ __volatile__("pause" ::
+        pause();
-                                 : "memory");
 
     kinfo("Cleaning page table remapping...\n");
 
@@ -96,6 +104,7 @@ void smp_init()
     uint64_t *global_CR3 = get_CR3();
     for (int i = 0; i < 256; ++i)
     {
+        io_mfence();
         *(ul *)(phys_2_virt(global_CR3) + i) = 0UL;
     }
     kdebug("init proc's preempt_count=%ld", current_pcb->preempt_count);
@@ -121,7 +130,7 @@ void smp_ap_start()
         __asm__ __volatile__("movq %0, %%rsp \n\t" ::"m"(stack_start)
                              : "memory");*/
     ksuccess("AP core successfully started!");
-
+    io_mfence();
     ++num_cpu_started;
 
     kdebug("current cpu = %d", current_starting_cpu);
@@ -155,6 +164,7 @@ void smp_ap_start()
     // kdebug("IDT_addr = %#018lx", phys_2_virt(IDT_Table));
     spin_unlock(&multi_core_starting_lock);
     preempt_disable(); // 由于ap处理器的pcb与bsp的不同，因此ap处理器放锁时，需要手动恢复preempt count
+    io_mfence();
     sti();
 
     while (1)

kernel/syscall/syscall.c

@@ -22,9 +22,18 @@ extern uint64_t sys_clock(struct pt_regs *regs);
  * @brief 导出系统调用处理函数的符号
  *
  */
-#define SYSCALL_COMMON(syscall_num, symbol) extern unsigned long symbol(struct pt_regs *regs);
+
-SYSCALL_COMMON(0, system_call_not_exists); // 导出system_call_not_exists函数
+/**
-#undef SYSCALL_COMMON                      // 取消前述宏定义
+ * @brief 系统调用不存在时的处理函数
+ *
+ * @param regs 进程3特权级下的寄存器
+ * @return ul
+ */
+ul system_call_not_exists(struct pt_regs *regs)
+{
+    kerror("System call [ ID #%d ] not exists.", regs->rax);
+    return ESYSCALL_NOT_EXISTS;
+}                    // 取消前述宏定义
 
 /**
  * @brief 重新定义为：把系统调用函数加入系统调用表

kernel/syscall/syscall.h

@@ -36,11 +36,7 @@ long enter_syscall_int(ul syscall_id, ul arg0, ul arg1, ul arg2, ul arg3, ul arg
  * @param regs 进程3特权级下的寄存器
  * @return ul
  */
-ul system_call_not_exists(struct pt_regs *regs)
+ul system_call_not_exists(struct pt_regs *regs);
-{
-    kerror("System call [ ID #%d ] not exists.", regs->rax);
-    return ESYSCALL_NOT_EXISTS;
-}
 
 /**
  * @brief 打印字符串的系统调用

tools/batch_delete_loop.py

@@ -0,0 +1,9 @@
+import os
+
+start = int(input("Start from: "))
+end = int(input("End at: "))
+
+for i in range(start, end+1):
+    print("Deleting: " + str(i))
+    os.system("sudo losetup -d /dev/loop" + str(i))
+print("Done!")