fix:在smp模块中增加mfence

kernel/Makefile

@@ -10,7 +10,7 @@ LIB_FILES := $(foreach DIR,$(DIR_LIB),$(addprefix $(DIR)/,$(lib_patterns)))
 
 # 控制操作系统使用的中断控制器 _INTR_8259A_ _INTR_APIC_
 PIC := _INTR_APIC_
-CFLAGS = $(GLOBAL_CFLAGS) -D $(PIC) -I $(shell pwd)
+CFLAGS = $(GLOBAL_CFLAGS) -D $(PIC) -I $(shell pwd)  -O3
 
 export ASFLAGS := --64
 

kernel/driver/usb/usb.c

@@ -8,8 +8,7 @@
 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
-#pragma GCC push_options
+
-#pragma GCC optimize("O0")
 // 在pci总线上寻找到的usb设备控制器的header
 static struct pci_device_structure_header_t *usb_pdevs[MAX_USB_NUM];
 static int usb_pdevs_count = 0;
@@ -35,6 +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:
@@ -48,6 +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:
@@ -58,4 +59,3 @@ void usb_init()
     }
     kinfo("Successfully initialized all usb host controllers!");
 }
-#pragma GCC pop_options

kernel/driver/usb/xhci/xhci.c

@@ -56,7 +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)
@@ -199,11 +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;
@@ -222,9 +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);
@@ -234,12 +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;
@@ -263,14 +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)
                 {
@@ -281,9 +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);
@@ -300,7 +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);
 }
 
@@ -312,7 +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();
 }
 
 /**
@@ -340,13 +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)
@@ -372,8 +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;
 
@@ -389,15 +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;
@@ -409,12 +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;
             }
@@ -428,13 +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;
             }
@@ -444,6 +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;
@@ -485,11 +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;
 }
 
@@ -503,18 +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;
@@ -526,13 +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");
 }
@@ -540,29 +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);
@@ -573,9 +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
@@ -603,11 +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)
@@ -617,10 +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));
@@ -633,7 +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)
@@ -651,12 +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;
     }
@@ -695,6 +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
@@ -735,14 +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;
@@ -751,15 +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;
@@ -770,7 +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]);
@@ -794,7 +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)
     {
@@ -805,13 +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);
 
@@ -828,9 +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;
@@ -849,25 +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
     {
@@ -876,7 +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
@@ -887,17 +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;
@@ -916,9 +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));
 
@@ -926,4 +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 optimize("O0")
+#pragma GCC pop_options

kernel/exception/trap.c

@@ -1,4 +1,3 @@
-// #pragma GCC optimize ("O0")
 #include "trap.h"
 #include "gate.h"
 #include <process/ptrace.h>

kernel/mm/slab.c

@@ -1,8 +1,6 @@
 #include "slab.h"
 #include <common/compiler.h>
 
-#pragma GCC push_options
-#pragma GCC optimize("O0")
 struct slab kmalloc_cache_group[16] =
     {
         {32, 0, 0, NULL, NULL, NULL, NULL},
@@ -707,4 +705,3 @@ unsigned long kfree(void *address)
     kBUG("kfree(): Can't free memory.");
     return ECANNOT_FREE_MEM;
 }
-#pragma GCC pop_options

kernel/process/process.h

@@ -18,8 +18,7 @@
 #include <filesystem/VFS/VFS.h>
 #include <common/wait_queue.h>
 
-// #pragma GCC push_options
+
-// #pragma GCC optimize("O0")
 
 // 进程最大可拥有的文件描述符数量
 #define PROC_MAX_FD_NUM 16

kernel/smp/apu_boot.S

@@ -1,4 +1,3 @@
-//#pragma GCC optimize("O0")
 #include "../common/asm.h"
 
 

kernel/smp/smp.c

@@ -10,8 +10,7 @@
 #include <sched/sched.h>
 
 #include "ipi.h"
-#pragma GCC push_options
+
-#pragma GCC optimize("O1")
 void ipi_0xc8_handler(uint64_t irq_num, uint64_t param, struct pt_regs *regs); // 由BSP转发的HPET中断处理函数
 
 static spinlock_t multi_core_starting_lock; // 多核启动锁
@@ -31,21 +30,25 @@ 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
@@ -53,22 +56,25 @@ void smp_init()
         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;
         }
         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;
@@ -82,15 +88,15 @@ void smp_init()
         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");
 
@@ -98,6 +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);
@@ -123,7 +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);
@@ -157,6 +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)
@@ -179,5 +187,3 @@ void ipi_0xc8_handler(uint64_t irq_num, uint64_t param, struct pt_regs *regs)
 {
     sched_update_jiffies();
 }
-
-#pragma GCC optimize("O0")