Windows内核-缓存

缓存

知识点

WC（写组合/写合并）

同一时刻只能保证4个地址;

WB（写回绕/回写）

WT（直写）

通常是做多媒体

实验代码

课程中test1比test2快, 实验是test2比test1快???

// 10_19_Cache(缓存).cpp : 此文件包含 "main" 函数。程序执行将在此处开始并结束。
//

#include <iostream>
#include <Windows.h>

#define ALLOCATE_SIZE 0x10000
#define MASK (ALLOCATE_SIZE - 1)

void test1()
{
	PUCHAR x1 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x2 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x3 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x4 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x5 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x6 = (PUCHAR)malloc(ALLOCATE_SIZE);

	int startTime = GetTickCount();
	int i = INT_MAX;
	while (i--)
	{
		int index = i & MASK;
		char b = i;
		x1[index] = b;
		x2[index] = b;
		x3[index] = b;
	}

	int j = INT_MAX;
	while (j--)
	{
		int index = j & MASK;
		char b = j;
		x4[index] = b;
		x5[index] = b;
		x6[index] = b;
	}
	int endTime = GetTickCount();
	printf("1. 一共耗时%d毫秒\r\n", endTime - startTime);
	free(x1);
	free(x2);
	free(x3);
	free(x4);
	free(x5);
	free(x6);
}

void test2()
{
	PUCHAR x1 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x2 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x3 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x4 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x5 = (PUCHAR)malloc(ALLOCATE_SIZE);
	PUCHAR x6 = (PUCHAR)malloc(ALLOCATE_SIZE);

	int startTime = GetTickCount();
	int i = INT_MAX;
	while (i--)
	{
		int index = i & MASK;
		char b = i;
		x1[index] = b;
		x2[index] = b;
		x3[index] = b;
		x4[index] = b;
		x5[index] = b;
		x6[index] = b;
	}

	int endTime = GetTickCount();
	printf("2. 一共耗时%d毫秒\r\n", endTime - startTime);
	free(x1);
	free(x2);
	free(x3);
	free(x4);
	free(x5);
	free(x6);
}


int main()
{
	test1();
	test2();
	test1();
	test2();
	test1();
	test2();

	system("pause");
	return 0;
}

PAT

TLB

每个核有4个TLB, 2个数据核（dTLB）, 2个指令TLB（iTLB）;

实验步骤

申请2个地址, 分别赋值;
把第一个地址挂在0地址上, 然后访问, 取出结果保存变量;
又把第二个地址挂在0地址上, 然后访问, 取出结果也保存变量;
观察变量1, 变量2有什么不同;

401080
gdtr: 0040ec00`00081080

实验代码

代码出现问题, 获取0地址时为0;

// 10_21_TLB.cpp : 此文件包含 "main" 函数。程序执行将在此处开始并结束。
//

#include <iostream>
#include <Windows.h>

PUCHAR addr1 = NULL;
PUCHAR addr2 = NULL;
ULONG temp1 = 0, temp2 = 0;

void  __declspec(naked) test()
{
	__asm
	{
		pushad;
		pushfd;
		push 0x30;
		pop fs;


		mov eax, [addr1]; // 线形地址1, 使用原始汇编代码, 注释代码会导致系统蓝屏
		shr eax, 0x9;
		and eax, 0x7ffff8;
		mov ecx, [eax-0x3FFFFFFC];
		sub eax, 0x40000000;
		mov ecx, [eax];
		mov edx, [eax + 4];

		//and eax, 0xC0000000;
		//int 3;
		//mov ecx, [eax];		// 取低4字节
		//mov edx, [eax + 4];	// 取高4字节

		mov dword ptr ds:[0xC0000000], ecx;
		mov dword ptr ds:[0xC0000004], edx;
		mov eax, dword ptr ds:[0];
		int 3;
		mov [temp1], eax;

		mov eax, cr3;	//刷新cr3
		mov cr3, eax;	// 刷新cr3

		//mov eax, [addr2]; // 线形地址2
		//shr eax, 0x9;
		//and eax, 0x7ffff8;
		//and eax, 0xC0000000;
		//mov ecx, [eax];		// 取低4字节
		//mov edx, [eax + 4];	// 取高4字节

		//mov dword ptr ds : [0xC0000000], ecx;
		//mov dword ptr ds : [0xC0000004], edx;
		//mov eax, dword ptr ds : [0];
		//mov [temp2], eax;

		push 0x3b;
		pop fs;
		popfd;
		popad;
		retf;
	}
}

int main()
{
	addr1 = (PUCHAR)VirtualAlloc(NULL, 0x1000, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
	addr2 = (PUCHAR)VirtualAlloc(NULL, 0x1000, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

	*(PUCHAR)addr1 = 0x1000;
	*(PUCHAR)addr2 = 0x2000;

	temp1 = 0;
	temp2 = 0;

	printf("addr1 = %x, addr2 = %x, test = %x\r\n", addr1, addr2, test);
	system("pause");

	char bufcode[6] = { 0,0,0,0,0x4b,0 };
	__asm
	{
		// push fs;
		call fword ptr bufcode;
		// pop fs;
	}
	printf("temp1 = %x, temp2 = %x\r\n", temp1, temp2);

	system("pause");
	return 0;
}

刷新CR3

当切换CR3的时候, CPU就认为在切换不同的页表; 当页表发生变化的时候, TLB就会产生一个刷新; 凡是G位等于0的通通刷新掉; 刷新除G位为1的情况;
切换线程的时候, 判断当前被切换与切换是否是一个进程, 如果不是, 会切换页表（CR3）;

1 2	mov eax, cr3; //刷新cr3 mov cr3, eax; // 刷新cr3

把地址变为无效缓存; 刷新一条

1	invlpg dword ptr ds:[0]; // 强制刷新TLB中某一行

CR4控制寄存器中有一个PGE位, 这个是控制所有的G位是否有效; 刷新所有;