Lecture 18 Virtual Memory Systems.ass

[Script Info]
; Script generated by Aegisub 3.2.2
; http://www.aegisub.org/
Title: Default Aegisub file
ScriptType: v4.00+
WrapStyle: 0
ScaledBorderAndShadow: yes
YCbCr Matrix: TV.601
PlayResX: 1280
PlayResY: 720

[Aegisub Project Garbage]
Last Style Storage: Default
Audio File: csapp/Lecture 18  Virtual Memory  Systems.mp4
Video File: csapp/Lecture 18  Virtual Memory  Systems.mp4
Video AR Mode: 4
Video AR Value: 1.777778
Video Zoom Percent: 1.000000
Scroll Position: 1612
Active Line: 1618
Video Position: 138786

[V4+ Styles]
Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
Style: English,Noto Sans CJK SC Black,30,&H00FFFFFF,&H00412A2C,&H00412A2C,&H00412A2C,0,0,0,0,100,100,0,0,1,1.5,1.2,2,10,10,10,1
Style: Chinese,Noto Sans CJK SC Black,40,&H00FFFF00,&H00FFFFFF,&H00000000,&H00000000,0,0,0,0,100,100,0,0,1,1.2,1.2,2,10,10,10,1

[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:00.56,0:00:05.50,English,,0,0,0,,So today we're going to continue our study of virtual memory
Dialogue: 0,0:00:00.56,0:00:05.50,Chinese,,0,0,0,,今天，我们要继续学习虚拟内存
Dialogue: 0,0:00:06.14,0:00:11.40,English,,0,0,0,,By looking at more detail at how address translation works
Dialogue: 0,0:00:06.14,0:00:11.40,Chinese,,0,0,0,,学习更多关于地址翻译的细节
Dialogue: 0,0:00:13.32,0:00:20.20,English,,0,0,0,,And then we'll learn about how virtual memory works on real systems on Linux and x86-64 systems
Dialogue: 0,0:00:13.32,0:00:20.20,Chinese,,0,0,0,,然后我们将要学习虚拟内存在真实系统，Linux，x86-64 下的工作情况
Dialogue: 0,0:00:20.96,0:00:26.54,English,,0,0,0,,And that the payoff for that is that you'll really be able to now understand how fork and execve really work
Dialogue: 0,0:00:20.96,0:00:26.54,Chinese,,0,0,0,,学习虚拟内存后，你会真正地了解系统调用 fork 和 execve 是如何工作的
Dialogue: 0,0:00:26.80,0:00:33.70,English,,0,0,0,,And how vital the virtual memory system is due to these sort of the entire working of the system
Dialogue: 0,0:00:26.80,0:00:33.70,Chinese,,0,0,0,,虚拟内存对于整个系统的工作是多么重要
Dialogue: 0,0:00:38.54,0:00:43.70,English,,0,0,0,,So let's look at some specific examples of how we do address translation
Dialogue: 0,0:00:38.54,0:00:43.70,Chinese,,0,0,0,,我们首先看一些具体的地址翻译的例子
Dialogue: 0,0:00:44.00,0:00:46.02,English,,0,0,0,,And here's that reference slide
Dialogue: 0,0:00:44.00,0:00:46.02,Chinese,,0,0,0,,参考我们的 PPT
Dialogue: 0,0:00:49.06,0:00:54.10,English,,0,0,0,,So let's suppose we have this simple memory system with 14-bit virtual addresses
Dialogue: 0,0:00:49.06,0:00:54.10,Chinese,,0,0,0,,假设我们有一个简单内存系统，它的虚拟地址是 14 位
Dialogue: 0,0:00:54.80,0:00:56.64,English,,0,0,0,,12-bit physical addresses
Dialogue: 0,0:00:54.80,0:00:56.64,Chinese,,0,0,0,,物理地址是 12 位
Dialogue: 0,0:00:57.04,0:00:58.82,English,,0,0,0,,Page size of 64 bytes
Dialogue: 0,0:00:57.04,0:00:58.82,Chinese,,0,0,0,,页面大小是 64 字节
Dialogue: 0,0:01:00.14,0:01:06.30,English,,0,0,0,,So in our we have so we need 6 VPO bits offset bits
Dialogue: 0,0:01:00.14,0:01:06.30,Chinese,,0,0,0,,所以我们需要 6 比特（64 = 2^6）的 VPO（虚拟页面偏移量）
Dialogue: 0,0:01:07.24,0:01:10.02,English,,0,0,0,,And then the remaining bits are the virtual page number
Dialogue: 0,0:01:07.24,0:01:10.02,Chinese,,0,0,0,,剩下的比特作为虚拟页号
Dialogue: 0,0:01:10.66,0:01:14.50,English,,0,0,0,,And similarly for our physical addresses we need
Dialogue: 0,0:01:10.66,0:01:14.50,Chinese,,0,0,0,,物理地址的划分也是类似的
Dialogue: 0,0:01:15.20,0:01:20.40,English,,0,0,0,,We need 6 offset bits and the remaining bits form the physical page number
Dialogue: 0,0:01:15.20,0:01:20.40,Chinese,,0,0,0,,我们需要 6 比特的物理页偏移量然后剩下的作为物理页号
Dialogue: 0,0:01:21.90,0:01:28.58,English,,0,0,0,,Okay now our that the TLB in this system has 16 entries
Dialogue: 0,0:01:21.90,0:01:28.58,Chinese,,0,0,0,,嗯，我们的 TLB 有 16 个条目
Dialogue: 0,0:01:28.94,0:01:30.98,English,,0,0,0,,And it's 4-way set associative
Dialogue: 0,0:01:28.94,0:01:30.98,Chinese,,0,0,0,,并且它是 4 路组相连的
Dialogue: 0,0:01:32.28,0:01:36.06,English,,0,0,0,,So remember that the TLB caches page table entries
Dialogue: 0,0:01:32.28,0:01:36.06,Chinese,,0,0,0,,TLB 缓存的是页表条目
Dialogue: 0,0:01:36.56,0:01:40.56,English,,0,0,0,,Okay and those page table entries are uniquely identified by the virtual page number
Dialogue: 0,0:01:36.56,0:01:40.56,Chinese,,0,0,0,,并且缓存页条目被虚拟页号唯一标识
Dialogue: 0,0:01:41.82,0:01:48.66,English,,0,0,0,,So we only need to use the VPN to to access entries in the TLB
Dialogue: 0,0:01:41.82,0:01:48.66,Chinese,,0,0,0,,所以，我们只需要用虚拟页号去访问 TLB
Dialogue: 0,0:01:49.84,0:01:55.66,English,,0,0,0,,We have 16 entries 4-way set associative so there's a total of 4 sets
Dialogue: 0,0:01:49.84,0:01:55.66,Chinese,,0,0,0,,TLB 有 16 个条目，并且是 4 路组相连的，所有我们有 4 组
Dialogue: 0,0:01:57.76,0:02:04.52,English,,0,0,0,,And so we use the low order bits and the VPN as the index
Dialogue: 0,0:01:57.76,0:02:04.52,Chinese,,0,0,0,,所以我们使用 VPN 的低两位(2^2=4)作为组索引
Dialogue: 0,0:02:05.58,0:02:09.18,English,,0,0,0,,And then the remaining bits just like any other caches, the tag bits
Dialogue: 0,0:02:05.58,0:02:09.18,Chinese,,0,0,0,,和其他缓存一样，剩下的比特作为标记位
Dialogue: 0,0:02:11.06,0:02:14.78,English,,0,0,0,,Okay and then this we've initialized this with some some various values
Dialogue: 0,0:02:11.06,0:02:14.78,Chinese,,0,0,0,,然后我们用不同的值来初始化 TLB
Dialogue: 0,0:02:14.78,0:02:24.98,English,,0,0,0,,Notice that the actual contents of the TLB are in this red in this region here right
Dialogue: 0,0:02:14.78,0:02:24.98,Chinese,,0,0,0,,注意到左边的红色区域（第一个列）并不是 TLB 的条目
Dialogue: 0,0:02:25.22,0:02:32.68,English,,0,0,0,,So the set this is just to help identify what set there there is no entry called set with a value of zero
Dialogue: 0,0:02:25.22,0:02:32.68,Chinese,,0,0,0,,组（set）列仅仅是为了帮助我们标识是哪一组，事实上第一行并没有值为 0 的组项
Dialogue: 0,0:02:32.76,0:02:37.38,English,,0,0,0,,Right we're just identifying that as it
Dialogue: 0,0:02:32.76,0:02:37.38,Chinese,,0,0,0,,我们仅仅是标识这一组
Dialogue: 0,0:02:38.50,0:02:42.24,English,,0,0,0,,We're just explicitly listening the index
Dialogue: 0,0:02:38.50,0:02:42.24,Chinese,,0,0,0,,我们只根据索引来查找组
Dialogue: 0,0:02:42.94,0:02:46.43,English,,0,0,0,,And so each of our entries in the TLB consists of this tag
Dialogue: 0,0:02:42.94,0:02:46.43,Chinese,,0,0,0,,每一个条目都有一个标记位
Dialogue: 0,0:02:47.92,0:02:52.50,English,,0,0,0,,And if the TLB entry is valid a physical page number
Dialogue: 0,0:02:47.92,0:02:52.50,Chinese,,0,0,0,,一个 TLB 条目如果有效，它就含有一个物理地址
Dialogue: 0,0:02:55.42,0:02:56.02,English,,0,0,0,,Okay now...
Dialogue: 0,0:02:55.42,0:02:56.02,Chinese,,0,0,0,,现在...
Dialogue: 0,0:02:57.82,0:03:00.12,English,,0,0,0,,We also need a page table so we'll have...
Dialogue: 0,0:02:57.82,0:03:00.12,Chinese,,0,0,0,,我们还需要页表
Dialogue: 0,0:03:03.46,0:03:08.36,English,,0,0,0,,We'll assume that we have this page table with...we're just showing the first 16 entries
Dialogue: 0,0:03:03.46,0:03:08.36,Chinese,,0,0,0,,假设这是我们页表的前 16 个条目
Dialogue: 0,0:03:09.34,0:03:13.58,English,,0,0,0,,And the each page table entry consists of a physical page number
Dialogue: 0,0:03:09.34,0:03:13.58,Chinese,,0,0,0,,每一个页表有一个物理页号
Dialogue: 0,0:03:14.40,0:03:15.60,English,,0,0,0,,And a valid bit
Dialogue: 0,0:03:14.40,0:03:15.60,Chinese,,0,0,0,,和一个有效位
Dialogue: 0,0:03:16.08,0:03:22.12,English,,0,0,0,,If valid bit is on then that indicates that the page is in memory and
Dialogue: 0,0:03:16.08,0:03:22.12,Chinese,,0,0,0,,如果有效位有效，则表示那个虚拟页面对应的物理页面在内存中
Dialogue: 0,0:03:22.60,0:03:26.20,English,,0,0,0,,And the PPN field gives the physical page number
Dialogue: 0,0:03:22.60,0:03:26.20,Chinese,,0,0,0,,并且 PPN 项给出了对应的物理页号
Dialogue: 0,0:03:27.04,0:03:32.64,English,,0,0,0,,Okay and again there's no actual this VPN column doesn't actually exist in the page table
Dialogue: 0,0:03:27.04,0:03:32.64,Chinese,,0,0,0,,和前面类似，在页表中并没有 VPN 这一项，VPN 只是一个标识
Dialogue: 0,0:03:35.02,0:03:41.95,English,,0,0,0,,Okay and now we have a simple direct mapped cache
Dialogue: 0,0:03:35.02,0:03:41.95,Chinese,,0,0,0,,好，这是我们的简单直接映射高速缓存
Dialogue: 0,0:03:42.66,0:03:44.68,English,,0,0,0,,It's addressed with physical addresses
Dialogue: 0,0:03:42.66,0:03:44.68,Chinese,,0,0,0,,它是通过物理地址来查找的
Dialogue: 0,0:03:46.76,0:03:51.62,English,,0,0,0,,Like and it contains 16 sets each with one line
Dialogue: 0,0:03:46.76,0:03:51.62,Chinese,,0,0,0,,它有 16 组，每组有一行
Dialogue: 0,0:03:52.38,0:03:54.90,English,,0,0,0,,And we'll assume a four byte block size
Dialogue: 0,0:03:52.38,0:03:54.90,Chinese,,0,0,0,,我们假设它是 4 字节的块
Dialogue: 0,0:03:56.22,0:03:59.28,English,,0,0,0,,So we have
Dialogue: 0,0:03:56.22,0:03:59.28,Chinese,,0,0,0,,我们的
Dialogue: 0,0:04:00.32,0:04:04.62,English,,0,0,0,,So we have a four byte block size so we need two offset bits for the cache
Dialogue: 0,0:04:00.32,0:04:04.62,Chinese,,0,0,0,,我们的缓存块的大小是 4 字节，所以我们需要用两比特用于表示偏移
Dialogue: 0,0:04:06.74,0:04:11.08,English,,0,0,0,,16 sets we means we need four cache index bits
Dialogue: 0,0:04:06.74,0:04:11.08,Chinese,,0,0,0,,缓存有 16 个组，意味着我们需要 4 比特表示索引
Dialogue: 0,0:04:11.78,0:04:16.28,English,,0,0,0,,And then the remaining bits are for the tags
Dialogue: 0,0:04:11.78,0:04:16.28,Chinese,,0,0,0,,剩下的比特作为标记位
Dialogue: 0,0:04:17.28,0:04:22.18,English,,0,0,0,,Now it's just an accident in the way that I define this the system that
Dialogue: 0,0:04:17.28,0:04:22.18,Chinese,,0,0,0,,我定义的这个系统有一个巧合
Dialogue: 0,0:04:22.92,0:04:27.54,English,,0,0,0,,The cash tags are exactly the same as the physical page number so this
Dialogue: 0,0:04:22.92,0:04:27.54,Chinese,,0,0,0,,缓存的标记位和物理页号是一样的
Dialogue: 0,0:04:28.04,0:04:33.32,English,,0,0,0,,This is just the coincidence and it doesn't isn't necessarily that way
Dialogue: 0,0:04:28.04,0:04:33.32,Chinese,,0,0,0,,这仅仅是一个巧合，对于虚拟内存系统来说这不是必须的
Dialogue: 0,0:04:35.34,0:04:40.19,English,,0,0,0,,Okay now with the magic of AV audio-visual systems
Dialogue: 0,0:04:35.34,0:04:40.19,Chinese,,0,0,0,,好，这是我们的视听教学系统
Dialogue: 0,0:04:40.80,0:04:42.36,English,,0,0,0,,Switch good
Dialogue: 0,0:04:40.80,0:04:42.36,Chinese,,0,0,0,,切换到投影设备
Dialogue: 0,0:04:45.82,0:04:51.66,English,,0,0,0,,Okay so let's we're going to look at this a couple examples using this little system
Dialogue: 0,0:04:45.82,0:04:51.66,Chinese,,0,0,0,,我们来看看这个小系统的几个例子
Dialogue: 0,0:04:53.50,0:04:59.28,English,,0,0,0,,Okay now so let's suppose the CPU executes is an instruction that generates an effective address
Dialogue: 0,0:04:53.50,0:04:59.28,Chinese,,0,0,0,,所以假设 CPU 执行了一条指令，它产生了一个有效地址
Dialogue: 0,0:04:59.32,0:05:03.46,English,,0,0,0,,It's a virtual address of 0x03d4 and then
Dialogue: 0,0:04:59.32,0:05:03.46,Chinese,,0,0,0,,这个地址的虚拟地址是 0x3d4 (例子和书 574 页的例子一样)
Dialogue: 0,0:05:04.06,0:05:06.52,English,,0,0,0,,It passes that to the MMU
Dialogue: 0,0:05:04.06,0:05:06.52,Chinese,,0,0,0,,它把这个地址传递给了 MMU
Dialogue: 0,0:05:07.10,0:05:10.38,English,,0,0,0,,Which is needs to figure out the the
Dialogue: 0,0:05:07.10,0:05:10.38,Chinese,,0,0,0,,我们需要找出
Dialogue: 0,0:05:10.90,0:05:16.48,English,,0,0,0,,Corresponding physical address and then it needs to fetch the data  from cache or memory
Dialogue: 0,0:05:10.90,0:05:16.48,Chinese,,0,0,0,,对应的物理地址，然后从缓存或内存中取出数据
Dialogue: 0,0:05:18.44,0:05:21.14,English,,0,0,0,,Ok so let's the first thing we do
Dialogue: 0,0:05:18.44,0:05:21.14,Chinese,,0,0,0,,第一件我们做的事...
Dialogue: 0,0:05:22.90,0:05:26.64,English,,0,0,0,,Is we just write out the bits for this virtual address
Dialogue: 0,0:05:22.90,0:05:26.64,Chinese,,0,0,0,,使用二进制表示这个虚拟地址
Dialogue: 0,0:05:28.44,0:05:30.78,English,,0,0,0,,Okay and then we identify the different fields
Dialogue: 0,0:05:28.44,0:05:30.78,Chinese,,0,0,0,,然后我们找出它的虚拟页号和虚拟页面偏移
Dialogue: 0,0:05:31.10,0:05:36.16,English,,0,0,0,,Ok so we have the offset virtual page offset bits
Dialogue: 0,0:05:31.10,0:05:36.16,Chinese,,0,0,0,,这是表示虚拟页面的偏移量的比特
Dialogue: 0,0:05:36.88,0:05:39.30,English,,0,0,0,,And then the remaining bits is the virtual page number
Dialogue: 0,0:05:36.88,0:05:39.30,Chinese,,0,0,0,,剩下的比特是虚拟页号
Dialogue: 0,0:05:40.40,0:05:45.76,English,,0,0,0,,Okay the offset in this case is
Dialogue: 0,0:05:40.40,0:05:45.76,Chinese,,0,0,0,,在这个例子中，虚拟页面偏移是
Dialogue: 0,0:05:47.54,0:05:51.06,English,,0,0,0,,2,4 0x24
Dialogue: 0,0:05:47.54,0:05:51.06,Chinese,,0,0,0,,2，4，0x24
Dialogue: 0,0:05:51.82,0:05:54.46,English,,0,0,0,,The VPN is 0xf
Dialogue: 0,0:05:51.82,0:05:54.46,Chinese,,0,0,0,,虚拟页号是 0xf
Dialogue: 0,0:05:56.70,0:06:00.15,English,,0,0,0,,And then we have that the 2 TLBI bits
Dialogue: 0,0:05:56.70,0:06:00.15,Chinese,,0,0,0,,TLB 索引是虚拟页号的低两位
Dialogue: 0,0:06:00.98,0:06:03.46,English,,0,0,0,,Which are 0b11 so 0x3
Dialogue: 0,0:06:00.98,0:06:03.46,Chinese,,0,0,0,,是 0b11，也就是 0x3
Dialogue: 0,0:06:04.48,0:06:08.56,English,,0,0,0,,And then we have the tag bits which are also equal to 3
Dialogue: 0,0:06:04.48,0:06:08.56,Chinese,,0,0,0,,这是我们的虚拟页面标记位，也是 3
Dialogue: 0,0:06:10.30,0:06:15.56,English,,0,0,0,,So the first thing that the MMU does given this address is it checks the TLB
Dialogue: 0,0:06:10.30,0:06:15.56,Chinese,,0,0,0,,MMU 做的第一件是就是查询 TLB
Dialogue: 0,0:06:16.12,0:06:20.90,English,,0,0,0,,On hoping that the page table entry that it needs is cached in the TLB
Dialogue: 0,0:06:16.12,0:06:20.90,Chinese,,0,0,0,,看虚拟地址对应的页表条目是否在 TLB 中
Dialogue: 0,0:06:22.10,0:06:26.66,English,,0,0,0,,So it goes extracts the index bits 3
Dialogue: 0,0:06:22.10,0:06:26.66,Chinese,,0,0,0,,所以我们先取出索引位，值为 3
Dialogue: 0,0:06:26.66,0:06:27.90,English,,0,0,0,,And that that gives us...
Dialogue: 0,0:06:26.66,0:06:27.90,Chinese,,0,0,0,,然后...
Dialogue: 0,0:06:31.06,0:06:31.70,English,,0,0,0,,So now that
Dialogue: 0,0:06:31.06,0:06:31.70,Chinese,,0,0,0,,现在
Dialogue: 0,0:06:33.26,0:06:38.60,English,,0,0,0,,That says that if this page table entry is in the TLB it's going to be in set 3
Dialogue: 0,0:06:33.26,0:06:38.60,Chinese,,0,0,0,,3 表明如果页表条目在 TLB 中的话，它应该在第 3 组
Dialogue: 0,0:06:39.56,0:06:41.92,English,,0,0,0,,Ok so we go to set 3
Dialogue: 0,0:06:39.56,0:06:41.92,Chinese,,0,0,0,,我们找到第 3 组
Dialogue: 0,0:06:43.82,0:06:50.60,English,,0,0,0,,And now we're looking for an entry in set 3 with a tag of 3
Dialogue: 0,0:06:43.82,0:06:50.60,Chinese,,0,0,0,,我们在第 3 组中找标记位为 3 的表项
Dialogue: 0,0:06:52.34,0:06:56.12,English,,0,0,0,,Okay so we go across this one there's 4 entries in this
Dialogue: 0,0:06:52.34,0:06:56.12,Chinese,,0,0,0,,我们遍历这 4 个条目
Dialogue: 0,0:06:56.82,0:07:01.16,English,,0,0,0,,Here's a entry with the tag of 7 so that's not end and plus the valid bit zero
Dialogue: 0,0:06:56.82,0:07:01.16,Chinese,,0,0,0,,这里有一个标记位为 7 的项，但它不是我们想要的，并且它的有效位为 0
Dialogue: 0,0:07:01.75,0:07:06.80,English,,0,0,0,,Oh here's an entry with a tag of three and  a in a valid bit of one
Dialogue: 0,0:07:01.75,0:07:06.80,Chinese,,0,0,0,,找到了，这里有一个项，标记位为 3 并且有效位为 1
Dialogue: 0,0:07:06.80,0:07:09.90,English,,0,0,0,,So we lucked out the page table entries in the TLB
Dialogue: 0,0:07:06.80,0:07:09.90,Chinese,,0,0,0,,所以，我们在 TLB 中找到了页表条目
Dialogue: 0,0:07:11.42,0:07:14.40,English,,0,0,0,,And so the the TLB returns the value
Dialogue: 0,0:07:11.42,0:07:14.40,Chinese,,0,0,0,,所以页表条目返回这个值
Dialogue: 0,0:07:15.72,0:07:20.36,English,,0,0,0,,The physical page number which is 0x0D back to the MMU
Dialogue: 0,0:07:15.72,0:07:20.36,Chinese,,0,0,0,,MMU 返回的物理页号是 0x0D
Dialogue: 0,0:07:20.36,0:07:23.04,English,,0,0,0,,Which can now construct the physical address
Dialogue: 0,0:07:20.36,0:07:23.04,Chinese,,0,0,0,,现在我们可以构造物理地址
Dialogue: 0,0:07:23.96,0:07:26.94,English,,0,0,0,,So it constructs that physical address PA
Dialogue: 0,0:07:23.96,0:07:26.94,Chinese,,0,0,0,,构造物理地址
Dialogue: 0,0:07:27.42,0:07:31.16,English,,0,0,0,,By copying directly copying the VPO bits
Dialogue: 0,0:07:27.42,0:07:31.16,Chinese,,0,0,0,,仅仅需要直接拷贝 PVO 的值
Dialogue: 0,0:07:32.10,0:07:33.14,English,,0,0,0,,To the PPO bits okay
Dialogue: 0,0:07:32.10,0:07:33.14,Chinese,,0,0,0,,到 PPO 的位置
Dialogue: 0,0:07:33.44,0:07:40.50,English,,0,0,0,,Right as we discuss the last time that the virtual page offs that's always identical to the physical page offset
Dialogue: 0,0:07:33.44,0:07:40.50,Chinese,,0,0,0,,我们上次课讨论了，VPO 的值永远等于 VPO 的值
Dialogue: 0,0:07:40.84,0:07:45.24,English,,0,0,0,,Because the block sizes in the virtual address space and the physical address space are the same size
Dialogue: 0,0:07:40.84,0:07:45.24,Chinese,,0,0,0,,因为虚拟页和物理页有相同的块大小
Dialogue: 0,0:07:46.50,0:07:49.98,English,,0,0,0,,Okay so here so we get the low order PPO bits and then
Dialogue: 0,0:07:46.50,0:07:49.98,Chinese,,0,0,0,,我们得到了 PPO 的值
Dialogue: 0,0:07:50.64,0:07:58.40,English,,0,0,0,,The physical page number which comes from the PTE that's cached in the TLB is 0xd
Dialogue: 0,0:07:50.64,0:07:58.40,Chinese,,0,0,0,,和从 TLB 缓存的 PTE 中得到了物理页号
Dialogue: 0,0:07:58.42,0:08:03.22,English,,0,0,0,,And so that forms the PPN bits of the physical address
Dialogue: 0,0:07:58.42,0:08:03.22,Chinese,,0,0,0,,合在一起构成了物理地址（书 575 页,地址为 0x354）
Dialogue: 0,0:08:03.22,0:08:07.28,English,,0,0,0,,So these now constitute a physical address
Dialogue: 0,0:08:03.22,0:08:07.28,Chinese,,0,0,0,,现在我们得到了物理地址
Dialogue: 0,0:08:08.78,0:08:11.88,English,,0,0,0,,So now the next step once we have the physical address is to look
Dialogue: 0,0:08:08.78,0:08:11.88,Chinese,,0,0,0,,下一步是使用这个物理地址去看高速缓存中有没有这个物理地址的缓存
Dialogue: 0,0:08:12.76,0:08:17.18,English,,0,0,0,,To send that to the cache and ask the cache to return the value at that address
Dialogue: 0,0:08:12.76,0:08:17.18,Chinese,,0,0,0,,把这个值送入高速缓存，请求高速缓存返回对应物理地址上的值
Dialogue: 0,0:08:18.60,0:08:22.42,English,,0,0,0,,And in this case we're just doing one byte accesses okay
Dialogue: 0,0:08:18.60,0:08:22.42,Chinese,,0,0,0,,在这个例子中，我们只需要返回一个字节
Dialogue: 0,0:08:22.70,0:08:27.74,English,,0,0,0,,So we're asking the cache to return the byte that's in this at this physical address
Dialogue: 0,0:08:22.70,0:08:27.74,Chinese,,0,0,0,,我们请求高速缓存返回这个物理地址的第一个字节
Dialogue: 0,0:08:29.22,0:08:30.60,English,,0,0,0,,Okay so the the cache
Dialogue: 0,0:08:29.22,0:08:30.60,Chinese,,0,0,0,,收到请求后
Dialogue: 0,0:08:32.92,0:08:35.30,English,,0,0,0,,Of course it first checks to see
Dialogue: 0,0:08:32.92,0:08:35.30,Chinese,,0,0,0,,高速缓存首先去...
Dialogue: 0,0:08:35.84,0:08:40.60,English,,0,0,0,,If the byte that's requested is contained in some block in the cache
Dialogue: 0,0:08:35.84,0:08:40.60,Chinese,,0,0,0,,检查高速缓存中是否有块缓存了该字节
Dialogue: 0,0:08:41.80,0:08:42.56,English,,0,0,0,,Okay so we...
Dialogue: 0,0:08:41.80,0:08:42.56,Chinese,,0,0,0,,嗯...
Dialogue: 0,0:08:43.58,0:08:47.12,English,,0,0,0,,So it takes in it extracts the cache index bits
Dialogue: 0,0:08:43.58,0:08:47.12,Chinese,,0,0,0,,高速缓存先取出物理地址的索引位
Dialogue: 0,0:08:48.06,0:08:51.22,English,,0,0,0,,Which is 0b00101 so 5
Dialogue: 0,0:08:48.06,0:08:51.22,Chinese,,0,0,0,,是 0b00101，也就是 0x5
Dialogue: 0,0:08:52.62,0:08:59.08,English,,0,0,0,,And so if this byte that's being requested is in the cache.It's going to be in set 5.So we go to set 5
Dialogue: 0,0:08:52.62,0:08:59.08,Chinese,,0,0,0,,所以，如果所请求的字节在高速缓存中的话，它应该在第 5 组，所以我们去第 5 组找
Dialogue: 0,0:09:01.48,0:09:04.82,English,,0,0,0,,And then we're looking for a tag of 0xd
Dialogue: 0,0:09:01.48,0:09:04.82,Chinese,,0,0,0,,找标记位为 0xd 的项
Dialogue: 0,0:09:06.46,0:09:12.22,English,,0,0,0,,And long behold here's we have a matching tag and matching valid bit
Dialogue: 0,0:09:06.46,0:09:12.22,Chinese,,0,0,0,,注意到，这里有一个匹配的标记位且有效位为 1
Dialogue: 0,0:09:12.68,0:09:16.30,English,,0,0,0,,So the good the line that we're that we request that is in the cash
Dialogue: 0,0:09:12.68,0:09:16.30,Chinese,,0,0,0,,这就是我们要在高速缓存中找的项
Dialogue: 0,0:09:17.24,0:09:22.28,English,,0,0,0,,And that the word that we requested is at offset 2
Dialogue: 0,0:09:17.24,0:09:22.28,Chinese,,0,0,0,,物理页面的偏移量为 2
Dialogue: 0,0:09:24.42,0:09:25.80,English,,0,0,0,,Okay so we go to...
Dialogue: 0,0:09:24.42,0:09:25.80,Chinese,,0,0,0,,好，我们找...
Dialogue: 0,0:09:27.28,0:09:28.66,English,,0,0,0,,So this is offset one
Dialogue: 0,0:09:27.28,0:09:28.66,Chinese,,0,0,0,,这是偏移量为 1 的项
Dialogue: 0,0:09:30.66,0:09:32.10,English,,0,0,0,,This is offset 2
Dialogue: 0,0:09:30.66,0:09:32.10,Chinese,,0,0,0,,这是偏移量为 2 的项
Dialogue: 0,0:09:34.22,0:09:35.54,English,,0,0,0,,Oh no I'm sorry
Dialogue: 0,0:09:34.22,0:09:35.54,Chinese,,0,0,0,,欧，抱歉，我弄错了
Dialogue: 0,0:09:49.14,0:09:50.72,English,,0,0,0,,Let's see...
Dialogue: 0,0:09:49.14,0:09:50.72,Chinese,,0,0,0,,我再看看...
Dialogue: 0,0:09:54.76,0:09:56.20,English,,0,0,0,,Oh I'm sorry it's...
Dialogue: 0,0:09:54.76,0:09:56.20,Chinese,,0,0,0,,抱歉，偏移量是...
Dialogue: 0,0:09:57.46,0:10:01.70,English,,0,0,0,,I was getting confused the into these are the index bits,so the offset is actually 0
Dialogue: 0,0:09:57.46,0:10:01.70,Chinese,,0,0,0,,我把表的位位置(看成了上一行)看成了偏移量，偏移量真实的值是 0
Dialogue: 0,0:10:03.48,0:10:07.50,English,,0,0,0,,Here's the this is a physical address
Dialogue: 0,0:10:03.48,0:10:07.50,Chinese,,0,0,0,,这是我们的物理地址
Dialogue: 0,0:10:08.20,0:10:13.42,English,,0,0,0,,Okay and the value of that offset is zero
Dialogue: 0,0:10:08.20,0:10:13.42,Chinese,,0,0,0,,它的偏移量是 0
Dialogue: 0,0:10:13.42,0:10:19.72,English,,0,0,0,,So we're asking for the in the fifth set we're asking for the the byte at offset zero
Dialogue: 0,0:10:13.42,0:10:19.72,Chinese,,0,0,0,,所以我们去请求第五组偏移量为 0 的字节
Dialogue: 0,0:10:19.72,0:10:21.98,English,,0,0,0,,And that's 0x36
Dialogue: 0,0:10:19.72,0:10:21.98,Chinese,,0,0,0,,值为 0x36
Dialogue: 0,0:10:23.12,0:10:31.22,English,,0,0,0,,So we have a hit the cache returns that byte back to the MMU which which passes it back to the processor
Dialogue: 0,0:10:23.12,0:10:31.22,Chinese,,0,0,0,,缓存命中，高速缓存把这个字节返回给 MMU，然后 MMU 把它传递给处理器
Dialogue: 0,0:10:32.06,0:10:34.74,English,,0,0,0,,Which stores it in a register most likely
Dialogue: 0,0:10:32.06,0:10:34.74,Chinese,,0,0,0,,然后处理器可能把这个字节存储在一个寄存器里
Dialogue: 0,0:10:37.72,0:10:39.76,English,,0,0,0,,Ok so let's look at another example
Dialogue: 0,0:10:37.72,0:10:39.76,Chinese,,0,0,0,,好了，我们来看下一个例子
Dialogue: 0,0:10:44.70,0:10:48.06,English,,0,0,0,,And I've recorded these here for you if you want to go back in and look at them
Dialogue: 0,0:10:44.70,0:10:48.06,Chinese,,0,0,0,,我把这个例子放在这里了，如果你想复习的话，可以看看这几页 PPT
Dialogue: 0,0:10:48.84,0:10:50.62,English,,0,0,0,,Let's look at another example this would
Dialogue: 0,0:10:48.84,0:10:50.62,Chinese,,0,0,0,,来看看我们的这个例子
Dialogue: 0,0:10:51.10,0:10:58.54,English,,0,0,0,,So this time the cpu sends a virtual address of 0x0020 to the MMU
Dialogue: 0,0:10:51.10,0:10:58.54,Chinese,,0,0,0,,这次 CPU 发送给 MMU 的虚拟地址是 0x0020
Dialogue: 0,0:10:59.96,0:11:06.16,English,,0,0,0,,Ok and so if we write out the bits for that virtual address
Dialogue: 0,0:10:59.96,0:11:06.16,Chinese,,0,0,0,,我们使用二进制写下这个虚拟地址
Dialogue: 0,0:11:09.38,0:11:16.26,English,,0,0,0,,We get this result and then we mark off the virtual page offset,the virtual page number
Dialogue: 0,0:11:09.38,0:11:16.26,Chinese,,0,0,0,,我们得到了这个结果，然后我们标记处 VPN（13-6 位）和 VPN（5-0 位）
Dialogue: 0,0:11:17.28,0:11:24.84,English,,0,0,0,,And the TLB index low order bits of the virtual page number and then the tag for the the TLBT
Dialogue: 0,0:11:17.28,0:11:24.84,Chinese,,0,0,0,,VPN 的低两位（7-6）位是 TLB 的索引位，VPN 剩下的部分是 TLB 的标记位
Dialogue: 0,0:11:26.04,0:11:32.60,English,,0,0,0,,So as step 1 same as before check with the TLB let's see if that page table entries in the TLBT
Dialogue: 0,0:11:26.04,0:11:32.60,Chinese,,0,0,0,,和前面一样，第一步是检查 TLB 看是否有页表条目的缓存
Dialogue: 0,0:11:33.18,0:11:37.70,English,,0,0,0,,In the TLB
Dialogue: 0,0:11:33.18,0:11:37.70,Chinese,,0,0,0,,在 TLB 中
Dialogue: 0,0:11:40.32,0:11:41.94,English,,0,0,0,,So we're asking in this case
Dialogue: 0,0:11:40.32,0:11:41.94,Chinese,,0,0,0,,在这个例子中
Dialogue: 0,0:11:45.38,0:11:53.32,English,,0,0,0,,This set if this page table entries in the TLB it will be in set 0
Dialogue: 0,0:11:45.38,0:11:53.32,Chinese,,0,0,0,,如果缓存存在，它应该在第一组
Dialogue: 0,0:11:56.06,0:11:57.96,English,,0,0,0,,And it will have a tag of 0
Dialogue: 0,0:11:56.06,0:11:57.96,Chinese,,0,0,0,,并且它的标记位应该为 0
Dialogue: 0,0:12:00.14,0:12:04.76,English,,0,0,0,,So we look in set 0 for a tag of 0
Dialogue: 0,0:12:00.14,0:12:04.76,Chinese,,0,0,0,,所以，我们在第 0 组内找标记位为 0 的项
Dialogue: 0,0:12:05.46,0:12:07.54,English,,0,0,0,,Now this doesn't match 0x03
Dialogue: 0,0:12:05.46,0:12:07.54,Chinese,,0,0,0,,第一项是 0x03，不匹配
Dialogue: 0,0:12:08.72,0:12:10.32,English,,0,0,0,,Here's 0x09 that doesn't match
Dialogue: 0,0:12:08.72,0:12:10.32,Chinese,,0,0,0,,第二项是 0x09，不匹配
Dialogue: 0,0:12:11.00,0:12:14.30,English,,0,0,0,,0x00 matches but the valid bits 0
Dialogue: 0,0:12:11.00,0:12:14.30,Chinese,,0,0,0,,第三项是 0x00，匹配，但是有效位为 0
Dialogue: 0,0:12:14.30,0:12:19.04,English,,0,0,0,,So this isn't a valid entry,it's just a coincidence that this tag was 0
Dialogue: 0,0:12:14.30,0:12:19.04,Chinese,,0,0,0,,所以，这不是一个有效值，它的标记位为 0 仅仅是一个巧合
Dialogue: 0,0:12:19.74,0:12:21.24,English,,0,0,0,,So this is a TLB miss
Dialogue: 0,0:12:19.74,0:12:21.24,Chinese,,0,0,0,,所以，这次 TLB 缓存不命中
Dialogue: 0,0:12:22.20,0:12:26.78,English,,0,0,0,,Ok so that's a bummer because now we have to go off the chip
Dialogue: 0,0:12:22.20,0:12:26.78,Chinese,,0,0,0,,查找缓存失败了，我们不得不放弃 TLB
Dialogue: 0,0:12:26.98,0:12:34.60,English,,0,0,0,,And make an expensive memory access to read that page table entry from the page table
Dialogue: 0,0:12:26.98,0:12:34.60,Chinese,,0,0,0,,通过开销较大的内存访问，去读取页表中对应的页表条目
Dialogue: 0,0:12:35.72,0:12:36.72,English,,0,0,0,,So now we have to...
Dialogue: 0,0:12:35.72,0:12:36.72,Chinese,,0,0,0,,我们要...
Dialogue: 0,0:12:37.84,0:12:41.89,English,,0,0,0,,So now we have to check with the page table
Dialogue: 0,0:12:37.84,0:12:41.89,Chinese,,0,0,0,,所以我们要去查看页表
Dialogue: 0,0:12:49.92,0:12:57.44,English,,0,0,0,,And we're looking for virtual page virtual page zero
Dialogue: 0,0:12:49.92,0:12:57.44,Chinese,,0,0,0,,寻找虚拟页号为 0 的项
Dialogue: 0,0:13:02.38,0:13:05.76,English,,0,0,0,,So we check the entry at virtual page zero to see if that page
Dialogue: 0,0:13:02.38,0:13:05.76,Chinese,,0,0,0,,我们检查对应的页表条目
Dialogue: 0,0:13:07.76,0:13:09.20,English,,0,0,0,,If that page is in memory
Dialogue: 0,0:13:07.76,0:13:09.20,Chinese,,0,0,0,,看虚拟页是否在内存中
Dialogue: 0,0:13:10.00,0:13:11.70,English,,0,0,0,,And if so if it has a valid
Dialogue: 0,0:13:10.00,0:13:11.70,Chinese,,0,0,0,,在的话有效位为 1
Dialogue: 0,0:13:11.92,0:13:15.70,English,,0,0,0,,And it is in memory because the valid bits one
Dialogue: 0,0:13:11.92,0:13:15.70,Chinese,,0,0,0,,它在内存中，因为它的有效位为 1
Dialogue: 0,0:13:16.90,0:13:20.16,English,,0,0,0,,So this...so we have a valid physical page number so
Dialogue: 0,0:13:16.90,0:13:20.16,Chinese,,0,0,0,,所以，我们有了一个有效的物理地址
Dialogue: 0,0:13:20.84,0:13:29.52,English,,0,0,0,,The memory returns the page table entry or the PPN back to the back to the MMU
Dialogue: 0,0:13:20.84,0:13:29.52,Chinese,,0,0,0,,所以，内存返回物理页条目或者 PPN 给 MMU
Dialogue: 0,0:13:30.28,0:13:34.75,English,,0,0,0,,Which is a it's a 0x28
Dialogue: 0,0:13:30.28,0:13:34.75,Chinese,,0,0,0,,它是 0x28
Dialogue: 0,0:13:35.16,0:13:36.94,English,,0,0,0,,And now the MMU can use that
Dialogue: 0,0:13:35.16,0:13:36.94,Chinese,,0,0,0,,现在 MMU
Dialogue: 0,0:13:37.56,0:13:40.72,English,,0,0,0,,That physical page number to construct the physical address
Dialogue: 0,0:13:37.56,0:13:40.72,Chinese,,0,0,0,,可以使用物理页号去构造物理地址
Dialogue: 0,0:13:41.52,0:13:47.32,English,,0,0,0,,So as before we copy the virtual page offset directly into the physical page offset
Dialogue: 0,0:13:41.52,0:13:47.32,Chinese,,0,0,0,,和前面类似，构造物理地址只需要拷贝虚拟页偏移到物理页偏移的位置
Dialogue: 0,0:13:48.30,0:13:55.90,English,,0,0,0,,And then a PPN is 0x28,so we have 0b10 it's 0x2, 0b1000,0x8
Dialogue: 0,0:13:48.30,0:13:55.90,Chinese,,0,0,0,,PPN 是 0x28，写成二进制就是 0x011000
Dialogue: 0,0:13:56.90,0:14:00.50,English,,0,0,0,,So now concatenated that those form the physical address
Dialogue: 0,0:13:56.90,0:14:00.50,Chinese,,0,0,0,,我们把物理页号和虚拟页偏移连起来就是我们的物理地址
Dialogue: 0,0:14:01.46,0:14:05.20,English,,0,0,0,,And now the MMU has a physical address that it can hand off to the cache
Dialogue: 0,0:14:01.46,0:14:05.20,Chinese,,0,0,0,,现在 MMU 拥有了物理地址，所以它可以将其发送到高速缓存
Dialogue: 0,0:14:07.40,0:14:13.04,English,,0,0,0,,And request that the cache return that value at that physical address
Dialogue: 0,0:14:07.40,0:14:13.04,Chinese,,0,0,0,,并请求高速缓存返回对应的物理地址上的一个字节
Dialogue: 0,0:14:15.32,0:14:18.60,English,,0,0,0,,So now the cache gets that physical address
Dialogue: 0,0:14:15.32,0:14:18.60,Chinese,,0,0,0,,现在高速缓存得到了这个物理地址
Dialogue: 0,0:14:20.22,0:14:23.20,English,,0,0,0,,And it extracts the cache index bits
Dialogue: 0,0:14:20.22,0:14:23.20,Chinese,,0,0,0,,它取出对应的索引位
Dialogue: 0,0:14:23.98,0:14:29.08,English,,0,0,0,,Which in this case are 0x8 okay 0x8
Dialogue: 0,0:14:23.98,0:14:29.08,Chinese,,0,0,0,,在这个例子中是 0x8
Dialogue: 0,0:14:30.52,0:14:34.98,English,,0,0,0,,So if this the block if the word that we're looking for is contained in the cache
Dialogue: 0,0:14:30.52,0:14:34.98,Chinese,,0,0,0,,如果有缓存中缓存有我们要找的这个字的话
Dialogue: 0,0:14:35.40,0:14:39.80,English,,0,0,0,,I'm sorry,the byte that we're looking for is contained in the cache,it'll be in set 8
Dialogue: 0,0:14:35.40,0:14:39.80,Chinese,,0,0,0,,错了，是字节，它应该在第八组
Dialogue: 0,0:14:40.80,0:14:43.46,English,,0,0,0,,Ok so we go...so the cache goes to set 8
Dialogue: 0,0:14:40.80,0:14:43.46,Chinese,,0,0,0,,好的，所以我们去高速缓存的第八组
Dialogue: 0,0:14:45.56,0:14:49.64,English,,0,0,0,,And it looks for a matching tag which in this case is
Dialogue: 0,0:14:45.56,0:14:49.64,Chinese,,0,0,0,,然后寻找对应的标记位，在这个例子中是
Dialogue: 0,0:14:50.60,0:14:54.40,English,,0,0,0,,It turns out that it's 28 just like the PPN just by coincidence
Dialogue: 0,0:14:50.60,0:14:54.40,Chinese,,0,0,0,,0x28，恰好和 PPN 一样
Dialogue: 0,0:14:56.70,0:15:04.02,English,,0,0,0,,So in set 8 we have a tag of we have an entry the block has a tag of 0x24 so there's a miss
Dialogue: 0,0:14:56.70,0:15:04.02,Chinese,,0,0,0,,第八组有一个条目，它的标记位是 24，所以这里是一次缓存不命中
Dialogue: 0,0:15:05.75,0:15:13.82,English,,0,0,0,,Okay so this is a cache miss and now the cache has to requests past that physical address to the memory to fetch that byte
Dialogue: 0,0:15:05.75,0:15:13.82,Chinese,,0,0,0,,缓存不命中，所以缓存不得不向内存传递那个物理地址去得到所需要的字节
Dialogue: 0,0:15:13.82,0:15:14.50,English,,0,0,0,,Yes
Dialogue: 0,0:15:13.82,0:15:14.50,Chinese,,0,0,0,,请讲
Dialogue: 0,0:15:14.50,0:15:20.60,English,,0,0,0,,[student speaking]
Dialogue: 0,0:15:14.50,0:15:20.60,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:15:20.62,0:15:25.00,English,,0,0,0,,How do...yeah the question is how do you know,how do you determine the size of the virtual page offset
Dialogue: 0,0:15:20.62,0:15:25.00,Chinese,,0,0,0,,怎样...这个学生的问题是，如何确定虚拟页面偏移的大小
Dialogue: 0,0:15:25.00,0:15:26.50,English,,0,0,0,,And yes it's based on the page size
Dialogue: 0,0:15:25.00,0:15:26.50,Chinese,,0,0,0,,它是根据页面大小来确定的
Dialogue: 0,0:15:26.60,0:15:29.50,English,,0,0,0,,[student speaking]
Dialogue: 0,0:15:26.60,0:15:29.50,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:15:29.50,0:15:31.66,English,,0,0,0,,It was 64 bytes
Dialogue: 0,0:15:29.50,0:15:31.66,Chinese,,0,0,0,,页面大小是 64 字节
Dialogue: 0,0:15:34.96,0:15:39.92,English,,0,0,0,,So 64 bytes so that's why we have 6 VPO and PPO bits
Dialogue: 0,0:15:34.96,0:15:39.92,Chinese,,0,0,0,,所以 VPO 和 PPO 都是 6 比特（2^6 = 64）
Dialogue: 0,0:15:43.96,0:15:45.88,English,,0,0,0,,Okay so is that clear to everybody that
Dialogue: 0,0:15:43.96,0:15:45.88,Chinese,,0,0,0,,好了，大家应该都清楚了
Dialogue: 0,0:15:47.08,0:15:51.70,English,,0,0,0,,That's a kind of thing that sometimes turns up on exams you know just
Dialogue: 0,0:15:47.08,0:15:51.70,Chinese,,0,0,0,,这种类型的题可能会出现在考试中，你知道的
Dialogue: 0,0:15:56.26,0:16:03.92,English,,0,0,0,,[student speaking]Yes any relationship between the length on physical page number and the cash tag
Dialogue: 0,0:15:56.26,0:16:03.92,Chinese,,0,0,0,,[学生提问]物理页号和缓存的标记位有什么关系吗
Dialogue: 0,0:16:03.92,0:16:09.72,English,,0,0,0,,Okay so the question is there any relationship between the length of the physical page number and the cash tag
Dialogue: 0,0:16:03.92,0:16:09.72,Chinese,,0,0,0,,好的，这位学生的问题是，物理页号和缓存标记位之间有什么关系吗
Dialogue: 0,0:16:10.08,0:16:11.68,English,,0,0,0,,This was just a complete coincidence
Dialogue: 0,0:16:10.08,0:16:11.68,Chinese,,0,0,0,,没有关系，这里仅仅是一个巧合
Dialogue: 0,0:16:12.04,0:16:19.04,English,,0,0,0,,Okay it just happened it was just a coincidence they normally they don't necessarily have to line up like that
Dialogue: 0,0:16:12.04,0:16:19.04,Chinese,,0,0,0,,这仅仅是一个巧合，物理页号和缓存标记位并不是一定需要满足相等关系
Dialogue: 0,0:16:21.34,0:16:22.18,English,,0,0,0,,Any other questions
Dialogue: 0,0:16:21.34,0:16:22.18,Chinese,,0,0,0,,还有其他问题吗
Dialogue: 0,0:16:24.52,0:16:24.98,English,,0,0,0,,Yes
Dialogue: 0,0:16:24.52,0:16:24.98,Chinese,,0,0,0,,请讲
Dialogue: 0,0:16:25.32,0:16:32.66,English,,0,0,0,,[student speaking]
Dialogue: 0,0:16:25.32,0:16:32.66,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:16:32.66,0:16:37.66,English,,0,0,0,,For instance if I'd had more sets in my cache then I need more bits and they'd spill over right
Dialogue: 0,0:16:32.66,0:16:37.66,Chinese,,0,0,0,,例如，如果缓存有更多组，则需要更多的比特来表示索引位，所以表示它们的比特会向左边扩展，对吧
Dialogue: 0,0:16:37.68,0:16:43.83,English,,0,0,0,,If my...it was just a coincidence because I had 6 PPO bits
Dialogue: 0,0:16:37.68,0:16:43.83,Chinese,,0,0,0,,这里相等的原因是因为我的 PPO 位 6 比特
Dialogue: 0,0:16:44.66,0:16:49.36,English,,0,0,0,,And just a block size of 4 and 4 sets
Dialogue: 0,0:16:44.66,0:16:49.36,Chinese,,0,0,0,,并且它们是 4 路组相连的，每组有 4 个条目
Dialogue: 0,0:16:49.38,0:16:54.52,English,,0,0,0,,So if my cache had more sets the cache index would spill over and then the cache tag
Dialogue: 0,0:16:49.38,0:16:54.52,Chinese,,0,0,0,,所以，如果我的缓存有更多的索引位，它就会使用现在标记位的比特
Dialogue: 0,0:16:54.88,0:16:57.32,English,,0,0,0,,Wouldn't match up with the physical page number okay
Dialogue: 0,0:16:54.88,0:16:57.32,Chinese,,0,0,0,,就会导致标记位的大小和物理页号是不相等的
Dialogue: 0,0:17:00.07,0:17:04.18,English,,0,0,0,,[student speaking]
Dialogue: 0,0:17:00.07,0:17:04.18,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:17:04.18,0:17:09.30,English,,0,0,0,,That's right,it's similar to the question before and it just happened to be a coincidence
Dialogue: 0,0:17:04.18,0:17:09.30,Chinese,,0,0,0,,对的，这个问题和上一个类似，这仅仅是一个巧合
Dialogue: 0,0:17:09.30,0:17:14.62,English,,0,0,0,,Maybe I should have chosen a different example,but they're completely independent
Dialogue: 0,0:17:09.30,0:17:14.62,Chinese,,0,0,0,,或许我应该换一个例子，但是物理页号和缓存标记位之间的确是独立的，没有关联
Dialogue: 0,0:17:19.04,0:17:20.18,English,,0,0,0,,Everybody else good okay
Dialogue: 0,0:17:19.04,0:17:20.18,Chinese,,0,0,0,,还有人要提问吗？
Dialogue: 0,0:17:27.94,0:17:32.22,English,,0,0,0,,Okay so now I want to talk to you about how...
Dialogue: 0,0:17:27.94,0:17:32.22,Chinese,,0,0,0,,好的，现在我要讲...
Dialogue: 0,0:17:32.80,0:17:39.58,English,,0,0,0,,How this stuff all works in a real system in this case it's a core i7 system from intel x86-64
Dialogue: 0,0:17:32.80,0:17:39.58,Chinese,,0,0,0,,虚拟内存在真实系统上是如何工作的，这里举的例子是 Intel x86-64 下的 core i7
Dialogue: 0,0:17:39.74,0:17:45.64,English,,0,0,0,,It's a family of high-end desktop systems x86-64
Dialogue: 0,0:17:39.74,0:17:45.64,Chinese,,0,0,0,,它属于高端桌面系统 x86-64 家族
Dialogue: 0,0:17:45.64,0:17:52.28,English,,0,0,0,,they are similar to the the shark machines that you use to do your labs
Dialogue: 0,0:17:45.64,0:17:52.28,Chinese,,0,0,0,,它们和你在实验中使用的 shark 机器是相似的
Dialogue: 0,0:17:55.78,0:18:01.58,English,,0,0,0,,So here's the way the memory system looks in the core i7
Dialogue: 0,0:17:55.78,0:18:01.58,Chinese,,0,0,0,,这是 core i7 内存系统的结构图
Dialogue: 0,0:18:02.52,0:18:04.64,English,,0,0,0,,The processor package is the chip
Dialogue: 0,0:18:02.52,0:18:04.64,Chinese,,0,0,0,,这是芯片的处理器封装
Dialogue: 0,0:18:05.12,0:18:07.36,English,,0,0,0,,Okay so that's like a single
Dialogue: 0,0:18:05.12,0:18:07.36,Chinese,,0,0,0,,看上去这是一个单核
Dialogue: 0,0:18:07.90,0:18:12.86,English,,0,0,0,,And then...and then there's four cores in this package
Dialogue: 0,0:18:07.90,0:18:12.86,Chinese,,0,0,0,,实际上...实际上在这个处理器封装里面有 4 核
Dialogue: 0,0:18:13.48,0:18:20.44,English,,0,0,0,,Each core is a separate as a separate cpu and can executes instructions separately
Dialogue: 0,0:18:13.48,0:18:20.44,Chinese,,0,0,0,,每一核都是一个独立的 cpu，可以独立的执行指令
Dialogue: 0,0:18:21.94,0:18:25.40,English,,0,0,0,,Okay so each of these cores has a register file
Dialogue: 0,0:18:21.94,0:18:25.40,Chinese,,0,0,0,,每一核都有寄存器文件
Dialogue: 0,0:18:27.50,0:18:30.10,English,,0,0,0,,And then a some hardware that fetches instructions
Dialogue: 0,0:18:27.50,0:18:30.10,Chinese,,0,0,0,,还有一些用于取指令的硬件
Dialogue: 0,0:18:31.36,0:18:36.48,English,,0,0,0,,It has 2 L1 caches,there's a data cache called the d-cache
Dialogue: 0,0:18:31.36,0:18:36.48,Chinese,,0,0,0,,这里有两个 L1 缓存，这里数据缓存，称为 d-cache
Dialogue: 0,0:18:36.58,0:18:40.68,English,,0,0,0,,Which is used to fetch data
Dialogue: 0,0:18:36.58,0:18:40.68,Chinese,,0,0,0,,主要是用于缓存数据
Dialogue: 0,0:18:41.88,0:18:47.58,English,,0,0,0,,And which holds data fetched from memory and then there's an instruction cache called an i-cache
Dialogue: 0,0:18:41.88,0:18:47.58,Chinese,,0,0,0,,即保存从内存中读取的数据，另一个是指令缓存，称为 i-cache
Dialogue: 0,0:18:48.78,0:18:52.40,English,,0,0,0,,Which holds instructions fetched from the code the code region
Dialogue: 0,0:18:48.78,0:18:52.40,Chinese,,0,0,0,,它保存从代码段中获取的指令
Dialogue: 0,0:18:55.12,0:19:01.54,English,,0,0,0,,So data this the d-cache has data only the i-cache has  instructions only
Dialogue: 0,0:18:55.12,0:19:01.54,Chinese,,0,0,0,,数据缓存只缓存数据，指令缓存只缓存指令
Dialogue: 0,0:19:01.90,0:19:06.50,English,,0,0,0,,There each 32 k bytes 8 way set associative so they're very small
Dialogue: 0,0:19:01.90,0:19:06.50,Chinese,,0,0,0,,它们的大小都是 32k 字节，并且都是 8 路组相连的，可以看出，它们都很小
Dialogue: 0,0:19:07.50,0:19:09.62,English,,0,0,0,,But they have fairly high associativity
Dialogue: 0,0:19:07.50,0:19:09.62,Chinese,,0,0,0,,但是它们的集成度很高
Dialogue: 0,0:19:12.76,0:19:16.32,English,,0,0,0,,And then the next level in the hierarchy is a L2
Dialogue: 0,0:19:12.76,0:19:16.32,Chinese,,0,0,0,,在缓存的层次结构中下一级是 L2
Dialogue: 0,0:19:16.32,0:19:20.10,English,,0,0,0,,So called unified cache because they can hold both instructions and data
Dialogue: 0,0:19:16.32,0:19:20.10,Chinese,,0,0,0,,它被称为统一缓存，因为它可以缓存指令和数据
Dialogue: 0,0:19:23.60,0:19:28.72,English,,0,0,0,,Okay it's a little bit bigger 256 k bytes and also eight-way associative
Dialogue: 0,0:19:23.60,0:19:28.72,Chinese,,0,0,0,,它的大小比 L1 大一点，256k 字节，也是 8 路组相连的
Dialogue: 0,0:19:28.72,0:19:32.88,English,,0,0,0,,And so both of these caches are on the core itself
Dialogue: 0,0:19:28.72,0:19:32.88,Chinese,,0,0,0,,L1 和 L2 缓存都是位于芯片内部的
Dialogue: 0,0:19:33.28,0:19:33.68,English,,0,0,0,,Ok
Dialogue: 0,0:19:33.28,0:19:33.68,Chinese,,0,0,0,,好的
Dialogue: 0,0:19:35.08,0:19:39.44,English,,0,0,0,,And then outside and enters a L3 cache which is shared by all cores
Dialogue: 0,0:19:35.08,0:19:39.44,Chinese,,0,0,0,,在芯片的外面有 L3 缓存，它是被所有核共享的
Dialogue: 0,0:19:40.26,0:19:43.48,English,,0,0,0,,Which is 8 megabytes and 16 way associative
Dialogue: 0,0:19:40.26,0:19:43.48,Chinese,,0,0,0,,它的大小是 8M，是 16 路组相连的
Dialogue: 0,0:19:44.92,0:19:48.46,English,,0,0,0,,Okay so the access is because the L1 cache is closest to the processor
Dialogue: 0,0:19:44.92,0:19:48.46,Chinese,,0,0,0,,然后谈谈访问时间，因为 L1 缓存最靠近处理器
Dialogue: 0,0:19:48.46,0:19:53.58,English,,0,0,0,,This is fast,this is like four cycles to access L1
Dialogue: 0,0:19:48.46,0:19:53.58,Chinese,,0,0,0,,它是最快的，访问时间大概是 4 个时钟周期
Dialogue: 0,0:19:54.54,0:19:59.60,English,,0,0,0,,L2 is bigger and a little further away so it
Dialogue: 0,0:19:54.54,0:19:59.60,Chinese,,0,0,0,,L2 的大小要大一点，并且离处理器也要远一点
Dialogue: 0,0:20:00.58,0:20:03.68,English,,0,0,0,,The it's about 10,10 cycles to access this
Dialogue: 0,0:20:00.58,0:20:03.68,Chinese,,0,0,0,,所以访问 L2 大概需要 10 个时钟周期
Dialogue: 0,0:20:04.36,0:20:09.54,English,,0,0,0,,And L3 is off the core right so it has to go over a connection
Dialogue: 0,0:20:04.36,0:20:09.54,Chinese,,0,0,0,,L3 是在处理器外边的，需要建立
Dialogue: 0,0:20:11.20,0:20:15.72,English,,0,0,0,,Some connection off the chip into this cache
Dialogue: 0,0:20:11.20,0:20:15.72,Chinese,,0,0,0,,芯片外部到芯片内的连接
Dialogue: 0,0:20:16.32,0:20:24.52,English,,0,0,0,,And so the the access time for this L3 cache is like (50) 30 to 50 cycles
Dialogue: 0,0:20:16.32,0:20:24.52,Chinese,,0,0,0,,所以 L3 的访问时间是 30 到 50 个时钟周期
Dialogue: 0,0:20:26.86,0:20:32.66,English,,0,0,0,,Now the MMU has it also has a hierarchy of TLBs
Dialogue: 0,0:20:26.86,0:20:32.66,Chinese,,0,0,0,,现在的 MMU 也有 TLB 的层次结构
Dialogue: 0,0:20:33.60,0:20:37.28,English,,0,0,0,,And I was talking with a student after class the last lecture night
Dialogue: 0,0:20:33.60,0:20:37.28,Chinese,,0,0,0,,上次课我和一个学生课后交谈的时候
Dialogue: 0,0:20:37.58,0:20:44.04,English,,0,0,0,,I mistakenly said that the system didn't have a hierarchy of TLBs
Dialogue: 0,0:20:37.58,0:20:44.04,Chinese,,0,0,0,,我说错了，我说系统的 TLB 没有层次结构
Dialogue: 0,0:20:44.20,0:20:45.18,English,,0,0,0,,But it does
Dialogue: 0,0:20:44.20,0:20:45.18,Chinese,,0,0,0,,但实际上是有的
Dialogue: 0,0:20:46.06,0:20:52.86,English,,0,0,0,,There's a a small L1 data TLB and a separate instruction TLB
Dialogue: 0,0:20:46.06,0:20:52.86,Chinese,,0,0,0,,这里有一个小的 L1 数据 TLB 缓存和小的 L1 指令缓存
Dialogue: 0,0:20:54.82,0:21:00.04,English,,0,0,0,,So the this has 64 total entries and it's 4-way set associative
Dialogue: 0,0:20:54.82,0:21:00.04,Chinese,,0,0,0,,它们有 64 个条目，是 4 路组相连的
Dialogue: 0,0:21:01.44,0:21:02.96,English,,0,0,0,,So how many sets are there then in this
Dialogue: 0,0:21:01.44,0:21:02.96,Chinese,,0,0,0,,所以，这里面有多少组
Dialogue: 0,0:21:05.72,0:21:09.48,English,,0,0,0,,How many sets in the L1 data TLB
Dialogue: 0,0:21:05.72,0:21:09.48,Chinese,,0,0,0,,L1 数据 TLB 里面有多少组
Dialogue: 0,0:21:13.76,0:21:16.66,English,,0,0,0,,16 right good
Dialogue: 0,0:21:13.76,0:21:16.66,Chinese,,0,0,0,,16 组，对的
Dialogue: 0,0:21:17.08,0:21:21.54,English,,0,0,0,,And then the instruction TLB has more entries interestingly
Dialogue: 0,0:21:17.08,0:21:21.54,Chinese,,0,0,0,,指令 TLB 的大小更大，这是比较有意思的
Dialogue: 0,0:21:23.90,0:21:27.04,English,,0,0,0,,It's an interesting decision I guess the thinking
Dialogue: 0,0:21:23.90,0:21:27.04,Chinese,,0,0,0,,我猜，这是一个有趣的设计
Dialogue: 0,0:21:27.70,0:21:30.84,English,,0,0,0,,I don't really know I'm just trying to conjecturing that the reason
Dialogue: 0,0:21:27.70,0:21:30.84,Chinese,,0,0,0,,我没有查过文献，我只是说一说我的猜测
Dialogue: 0,0:21:30.84,0:21:40.58,English,,0,0,0,,They would make the instruction TLB bigger is that the penalty for missing on instructions would be much larger
Dialogue: 0,0:21:30.84,0:21:40.58,Chinese,,0,0,0,,指令 TLB 比较大的原因可能是对指令 TLB 缓存不命中的惩罚要比数据 TLB 缓存不命中大
Dialogue: 0,0:21:48.92,0:21:53.56,English,,0,0,0,,But yeah I'm not sure,it's an interesting decision and then there's a there's a unified TLB below that
Dialogue: 0,0:21:48.92,0:21:53.56,Chinese,,0,0,0,,我不确定，但是确实是一个有趣的设计，好了，在 L1 TLB 下面是统一 TLB
Dialogue: 0,0:21:54.26,0:21:57.92,English,,0,0,0,,Which is a larger, larger still
Dialogue: 0,0:21:54.26,0:21:57.92,Chinese,,0,0,0,,它比 L1 大
Dialogue: 0,0:21:59.28,0:22:04.08,English,,0,0,0,,So the I think if you're wondering...I mean it's interesting to think like why do they have these
Dialogue: 0,0:21:59.28,0:22:04.08,Chinese,,0,0,0,,我想，你们可能比较疑惑...为什么要有 L2 缓存
Dialogue: 0,0:22:04.96,0:22:11.54,English,,0,0,0,,Like like these second-level caches why didn't they just make these out the L1 cache is bigger right
Dialogue: 0,0:22:04.96,0:22:11.54,Chinese,,0,0,0,,为什么不直接把 L1 缓存设计得大一些呢
Dialogue: 0,0:22:13.02,0:22:18.64,English,,0,0,0,,And in the case of the d-cache and the i-cache
Dialogue: 0,0:22:13.02,0:22:18.64,Chinese,,0,0,0,,以数据缓存和指令缓存为例
Dialogue: 0,0:22:19.48,0:22:21.24,English,,0,0,0,,Over here I'll show you in
Dialogue: 0,0:22:19.48,0:22:21.24,Chinese,,0,0,0,,我要说的是
Dialogue: 0,0:22:21.78,0:22:25.10,English,,0,0,0,,It just a little bit if you hang on there's a really interesting reason why
Dialogue: 0,0:22:21.78,0:22:25.10,Chinese,,0,0,0,,如果你认真去找原因，你会找到一个有趣的理由
Dialogue: 0,0:22:25.74,0:22:29.28,English,,0,0,0,,They can't make these things bigger or much bigger right there
Dialogue: 0,0:22:25.74,0:22:29.28,Chinese,,0,0,0,,它们的大小并非是不受限的
Dialogue: 0,0:22:29.96,0:22:35.82,English,,0,0,0,,There's a so there's there's a there's a property of the
Dialogue: 0,0:22:29.96,0:22:35.82,Chinese,,0,0,0,,因为它们有一个性质
Dialogue: 0,0:22:37.18,0:22:43.10,English,,0,0,0,,They're sort of they're constrained in the size of the offset and index bits  in these caches
Dialogue: 0,0:22:37.18,0:22:43.10,Chinese,,0,0,0,,在缓存中，它们的索引位和偏移位的大小是受限的
Dialogue: 0,0:22:43.62,0:22:45.70,English,,0,0,0,,And it's interesting I'll show you in a second
Dialogue: 0,0:22:43.62,0:22:45.70,Chinese,,0,0,0,,后面我要说明原因
Dialogue: 0,0:22:46.16,0:22:48.70,English,,0,0,0,,For that...for the TLB
Dialogue: 0,0:22:46.16,0:22:48.70,Chinese,,0,0,0,,对于 TLB 来说
Dialogue: 0,0:22:49.36,0:22:54.84,English,,0,0,0,,I mean one reason that comes to mind for why they didn't why they have the second level
Dialogue: 0,0:22:49.36,0:22:54.84,Chinese,,0,0,0,,为什么需要第二级缓存，我想到的第一个原因就是
Dialogue: 0,0:22:55.36,0:23:02.32,English,,0,0,0,,is that if they decided just to make the L1 data DTLB and ITLBs bigger
Dialogue: 0,0:22:55.36,0:23:02.32,Chinese,,0,0,0,,如果增加 L1 TLB 的大小，包括 DTLB 和 ITLB
Dialogue: 0,0:23:02.32,0:23:06.50,English,,0,0,0,,You know take the take the transistors that they used for this L2
Dialogue: 0,0:23:02.32,0:23:06.50,Chinese,,0,0,0,,可以通过把用于 L2 的资源
Dialogue: 0,0:23:07.08,0:23:10.22,English,,0,0,0,,And just give them to the level one TLBs
Dialogue: 0,0:23:07.08,0:23:10.22,Chinese,,0,0,0,,用于第一级的 TLB
Dialogue: 0,0:23:13.18,0:23:20.16,English,,0,0,0,,It you're kind of by partitioning those, by partitioning those those that storage
Dialogue: 0,0:23:13.18,0:23:20.16,Chinese,,0,0,0,,通过把第二级 TLB 的资源分割，然后用于第一级 TLB
Dialogue: 0,0:23:20.76,0:23:24.06,English,,0,0,0,,I ahead of time you're kind of locking yourself in, so if
Dialogue: 0,0:23:20.76,0:23:24.06,Chinese,,0,0,0,,这样是有问题的
Dialogue: 0,0:23:25.04,0:23:29.24,English,,0,0,0,,If you didn't have this level two TLB and and
Dialogue: 0,0:23:25.04,0:23:29.24,Chinese,,0,0,0,,如果没有了第二级的 TLB
Dialogue: 0,0:23:29.94,0:23:37.42,English,,0,0,0,,You just increase say you gave half and so it made L1 the data TLB an instruction TLB twice as big
Dialogue: 0,0:23:29.94,0:23:37.42,Chinese,,0,0,0,,你可以通过使用 L2 的资源把 L1 的数据 TLB 和指令 TLB 的大小增加一倍
Dialogue: 0,0:23:38.96,0:23:41.20,English,,0,0,0,,Then you're sort of locking yourself in right
Dialogue: 0,0:23:38.96,0:23:41.20,Chinese,,0,0,0,,然后你会发现有一个问题
Dialogue: 0,0:23:41.84,0:23:51.74,English,,0,0,0,,If you run,you might run out of,you still might run out of instructions and have capacity misses in that L1 cache
Dialogue: 0,0:23:41.84,0:23:51.74,Chinese,,0,0,0,,如果系统运行，仍然可能出现指令工作集大于缓存的情况，就会出现容量不命中
Dialogue: 0,0:23:52.48,0:23:54.36,English,,0,0,0,,I'm saying similarly for data right
Dialogue: 0,0:23:52.48,0:23:54.36,Chinese,,0,0,0,,同样的情况也可能发生在数据 TLB 上
Dialogue: 0,0:23:54.46,0:24:00.80,English,,0,0,0,,So you're not sure,you may in a particular program there mate there may be a lot more data
Dialogue: 0,0:23:54.46,0:24:00.80,Chinese,,0,0,0,,而且，你并不能对情况作出假设，你有可能运行一个程序，它的数据页表条目
Dialogue: 0,0:24:01.14,0:24:05.42,English,,0,0,0,,Page table entries than instruction page table entries in or vice versa
Dialogue: 0,0:24:01.14,0:24:05.42,Chinese,,0,0,0,,大于指令页表条目，或者相反
Dialogue: 0,0:24:06.26,0:24:11.88,English,,0,0,0,,So by creating the second level of a cache
Dialogue: 0,0:24:06.26,0:24:11.88,Chinese,,0,0,0,,所以，通过增加第二级缓存
Dialogue: 0,0:24:11.90,0:24:16.30,English,,0,0,0,,You're kind of hedging your bets right so you may the miss penalty
Dialogue: 0,0:24:11.90,0:24:16.30,Chinese,,0,0,0,,虽然大部分情况下指令缓存大于数据缓存，但是仍然可能出现缓存不命中的情况
Dialogue: 0,0:24:17.82,0:24:26.24,English,,0,0,0,,When you have this L2 cache the miss penalty in L1 isn't nearly as big as it didn't
Dialogue: 0,0:24:17.82,0:24:26.24,Chinese,,0,0,0,,通过使用二级缓存可以减少不命中惩罚
Dialogue: 0,0:24:26.72,0:24:29.42,English,,0,0,0,,Okay so that
Dialogue: 0,0:24:26.72,0:24:29.42,Chinese,,0,0,0,,所以
Dialogue: 0,0:24:30.44,0:24:33.44,English,,0,0,0,,That that's sort of the thinking that goes on
Dialogue: 0,0:24:30.44,0:24:33.44,Chinese,,0,0,0,,这就是设计二级缓存的想法
Dialogue: 0,0:24:34.82,0:24:40.66,English,,0,0,0,,Now there's a memory controller which count,which fetches data from memory
Dialogue: 0,0:24:34.82,0:24:40.66,Chinese,,0,0,0,,这里有一个内存控制器，它从内存中提取数据
Dialogue: 0,0:24:41.28,0:24:45.28,English,,0,0,0,,And then there's links to other cores and to the I/O bridge
Dialogue: 0,0:24:41.28,0:24:45.28,Chinese,,0,0,0,,这里是一个 QuickPath 互联，连接到其他核心和 I/O 桥
Dialogue: 0,0:24:49.84,0:24:52.30,English,,0,0,0,,Now n2n so it's really interesting so
Dialogue: 0,0:24:49.84,0:24:52.30,Chinese,,0,0,0,,端到端的地址翻译是有趣的，所以
Dialogue: 0,0:24:52.78,0:24:55.88,English,,0,0,0,,It's interesting to see how this all fits together end to end
Dialogue: 0,0:24:52.78,0:24:55.88,Chinese,,0,0,0,,接下来，看看这些缓存如何协作，完成端到端的翻译
Dialogue: 0,0:24:57.92,0:24:59.38,English,,0,0,0,,So let's look at it
Dialogue: 0,0:24:57.92,0:24:59.38,Chinese,,0,0,0,,让我们开始吧
Dialogue: 0,0:24:59.38,0:25:02.66,English,,0,0,0,,So the cpu generates a virtual address
Dialogue: 0,0:24:59.38,0:25:02.66,Chinese,,0,0,0,,CPU 产生一个虚拟地址
Dialogue: 0,0:25:03.74,0:25:07.74,English,,0,0,0,,In intel systems the virtual addresses are 48 bits
Dialogue: 0,0:25:03.74,0:25:07.74,Chinese,,0,0,0,,在 Intel 系统中虚拟地址长度为 48 比特
Dialogue: 0,0:25:08.94,0:25:12.99,English,,0,0,0,,We have 4k sized blocks so 12 offset bits
Dialogue: 0,0:25:08.94,0:25:12.99,Chinese,,0,0,0,,我们的页大小是 4k 的，所以 VPO 为 12 比特
Dialogue: 0,0:25:13.72,0:25:15.58,English,,0,0,0,,And then 36 VPN bits
Dialogue: 0,0:25:13.72,0:25:15.58,Chinese,,0,0,0,,VPN 需要 36 比特
Dialogue: 0,0:25:19.70,0:25:23.08,English,,0,0,0,,So first we look in the cache I mean in the TLB
Dialogue: 0,0:25:19.70,0:25:23.08,Chinese,,0,0,0,,首先，我们查看 TLB
Dialogue: 0,0:25:23.78,0:25:27.46,English,,0,0,0,,And there's as we said before there's 16 TLB sets
Dialogue: 0,0:25:23.78,0:25:27.46,Chinese,,0,0,0,,前面说过了，TLB 有 16 组
Dialogue: 0,0:25:27.74,0:25:31.14,English,,0,0,0,,This is the L1 I'm showing the L1 data TLB
Dialogue: 0,0:25:27.74,0:25:31.14,Chinese,,0,0,0,,我指的是 L1 的数据 TLB
Dialogue: 0,0:25:32.54,0:25:37.78,English,,0,0,0,,So we break it up into 4 TLBI index bits and 32 tag bits
Dialogue: 0,0:25:32.54,0:25:37.78,Chinese,,0,0,0,,所以把它拆分为 4 比特 TLBI 和 32 位标记位
Dialogue: 0,0:25:38.64,0:25:39.98,English,,0,0,0,,So we do a look up on the...
Dialogue: 0,0:25:38.64,0:25:39.98,Chinese,,0,0,0,,所以，我们在
Dialogue: 0,0:25:40.32,0:25:48.06,English,,0,0,0,,In the TLB to see if we can find that the PTE that contains that the corresponding physical page number for this virtual address
Dialogue: 0,0:25:40.32,0:25:48.06,Chinese,,0,0,0,,TLB 中寻找是否有对应虚拟地址的物理页号
Dialogue: 0,0:25:49.20,0:25:52.30,English,,0,0,0,,If we have a hit then the MMU can just construct
Dialogue: 0,0:25:49.20,0:25:52.30,Chinese,,0,0,0,,如果缓存命中，MMU 就构造物理地址
Dialogue: 0,0:25:53.06,0:25:58.14,English,,0,0,0,,The physical address directly by copying as we saw before copying
Dialogue: 0,0:25:53.06,0:25:58.14,Chinese,,0,0,0,,和前面类似，直接把
Dialogue: 0,0:25:58.62,0:26:03.16,English,,0,0,0,,The VPO to the PPO and then using the PPN that was returned from the TLB
Dialogue: 0,0:25:58.62,0:26:03.16,Chinese,,0,0,0,,VPO 拷贝到 PPO 的位置，然后连接上从 TLB 返回的 PPN
Dialogue: 0,0:26:03.86,0:26:05.78,English,,0,0,0,,How's that for an out acronym
Dialogue: 0,0:26:03.86,0:26:05.78,Chinese,,0,0,0,,注意缩略词的含义
Dialogue: 0,0:26:09.96,0:26:11.64,English,,0,0,0,,So if there's a TLB miss
Dialogue: 0,0:26:09.96,0:26:11.64,Chinese,,0,0,0,,如果 TLB 缓存不命中
Dialogue: 0,0:26:12.58,0:26:19.88,English,,0,0,0,,Then the system has to fetch the corresponding PPN from the page table
Dialogue: 0,0:26:12.58,0:26:19.88,Chinese,,0,0,0,,系统就去页表取出对应的 PPN
Dialogue: 0,0:26:20.28,0:26:25.32,English,,0,0,0,,Using that multi way lookup we looked multi level lookup we looked at before out
Dialogue: 0,0:26:20.28,0:26:25.32,Chinese,,0,0,0,,我们使用多路查表，使用前面说过的多级查表
Dialogue: 0,0:26:25.34,0:26:29.16,English,,0,0,0,,And I'll show you how this works in a little more detail coming up
Dialogue: 0,0:26:25.34,0:26:29.16,Chinese,,0,0,0,,接下来我将详细介绍一下这是如何工作的
Dialogue: 0,0:26:29.76,0:26:37.24,English,,0,0,0,,But the end result is that a page table entry is located and
Dialogue: 0,0:26:29.76,0:26:37.24,Chinese,,0,0,0,,但是结果就是一条页表条目
Dialogue: 0,0:26:37.62,0:26:44.96,English,,0,0,0,,The PPN is extracted from that and concatenated with the PPO to form the physical address
Dialogue: 0,0:26:37.62,0:26:44.96,Chinese,,0,0,0,,从页表条目中取出 PPN，然后和 PPO 连起来构成物理地址
Dialogue: 0,0:26:46.00,0:26:49.66,English,,0,0,0,,Then the MMU passes that physical address to the cache
Dialogue: 0,0:26:46.00,0:26:49.66,Chinese,,0,0,0,,MMU 将物理地址传递给高速缓存
Dialogue: 0,0:26:51.54,0:26:54.34,English,,0,0,0,,The L1 data cache has 64 sets
Dialogue: 0,0:26:51.54,0:26:54.34,Chinese,,0,0,0,,L1 数据缓存有 64 组
Dialogue: 0,0:26:55.42,0:26:58.00,English,,0,0,0,,So we need six cache index bits
Dialogue: 0,0:26:55.42,0:26:58.00,Chinese,,0,0,0,,所以我们需要 6 比特的索引位
Dialogue: 0,0:26:59.08,0:27:06.96,English,,0,0,0,,Now here's what I was alluding to before notice that the number of cache index and cache offset bits
Dialogue: 0,0:26:59.08,0:27:06.96,Chinese,,0,0,0,,这里引出了我前面说的 L1 的缓存不能太大的原因，注意缓存索引位和偏移位的长度
Dialogue: 0,0:27:08.28,0:27:12.90,English,,0,0,0,,Is exactly identical to the VPO in the virtual address
Dialogue: 0,0:27:08.28,0:27:12.90,Chinese,,0,0,0,,它们是和 VPO 一样的
Dialogue: 0,0:27:14.18,0:27:19.16,English,,0,0,0,,Okay so the cache offset and index bits in the physical address
Dialogue: 0,0:27:14.18,0:27:19.16,Chinese,,0,0,0,,所以，物理地址的索引位和偏移位
Dialogue: 0,0:27:20.24,0:27:26.50,English,,0,0,0,,Are identical to the VPO the offset bits in the virtual address
Dialogue: 0,0:27:20.24,0:27:26.50,Chinese,,0,0,0,,和 VPO 的偏移是一样的
Dialogue: 0,0:27:27.24,0:27:32.06,English,,0,0,0,,Ok and this is not a coincidence ok unlike that other example I showed you this is not a coincidence
Dialogue: 0,0:27:27.24,0:27:32.06,Chinese,,0,0,0,,不同与前面我所讲的例子，这不是一个巧合
Dialogue: 0,0:27:32.98,0:27:36.26,English,,0,0,0,,And this is sort,this is at the root
Dialogue: 0,0:27:32.98,0:27:36.26,Chinese,,0,0,0,,这是缓存设计的时候刻意设计的
Dialogue: 0,0:27:37.10,0:27:40.40,English,,0,0,0,,This is at the root of why the L1 cache is so small
Dialogue: 0,0:27:37.10,0:27:40.40,Chinese,,0,0,0,,这也是 L1 缓存为什么如此小的原因
Dialogue: 0,0:27:41.22,0:27:45.66,English,,0,0,0,,That the way intel implements their cache lookups
Dialogue: 0,0:27:41.22,0:27:45.66,Chinese,,0,0,0,,这是 Intel 实现缓存查找表的方式
Dialogue: 0,0:27:46.46,0:27:52.38,English,,0,0,0,,Which we'll see in a second depends on the cache index and cache offset bits and the physical address
Dialogue: 0,0:27:46.46,0:27:52.38,Chinese,,0,0,0,,缓存的索引位和偏移位合起来和
Dialogue: 0,0:27:52.46,0:27:55.50,English,,0,0,0,,Being identical to the offset bits and the virtual address
Dialogue: 0,0:27:52.46,0:27:55.50,Chinese,,0,0,0,,虚拟地址的 VPO 一样
Dialogue: 0,0:27:56.80,0:27:58.30,English,,0,0,0,,Anyway so then we do a cache
Dialogue: 0,0:27:56.80,0:27:58.30,Chinese,,0,0,0,,然后我们进行缓存查找
Dialogue: 0,0:27:58.74,0:28:05.32,English,,0,0,0,,We do a cache look up using the cache doesn't look up using the using this physical address
Dialogue: 0,0:27:58.74,0:28:05.32,Chinese,,0,0,0,,高速缓存使用物理地址进行查找
Dialogue: 0,0:28:05.90,0:28:08.33,English,,0,0,0,,Takes the index bits to identify set
Dialogue: 0,0:28:05.90,0:28:08.33,Chinese,,0,0,0,,使用索引位确定是哪一组
Dialogue: 0,0:28:09.02,0:28:11.32,English,,0,0,0,,Uses the tag to see if there's a match
Dialogue: 0,0:28:09.02,0:28:11.32,Chinese,,0,0,0,,使用标记位看是否匹配
Dialogue: 0,0:28:12.16,0:28:19.10,English,,0,0,0,,If there is then we have a hit which returns the resulting word back to the CPU
Dialogue: 0,0:28:12.16,0:28:19.10,Chinese,,0,0,0,,如果缓存命中，就返回对应的字给 CPU
Dialogue: 0,0:28:19.56,0:28:27.10,English,,0,0,0,,Otherwise there's a miss,so then the cash has to request the data from L2,L3 and main memory eventually
Dialogue: 0,0:28:19.56,0:28:27.10,Chinese,,0,0,0,,如果缓存不命中，我们只能去 L2，L3 去取数据，最后可能去内存中取数据
Dialogue: 0,0:28:28.18,0:28:31.96,English,,0,0,0,,Worst case that that data comes from main memory
Dialogue: 0,0:28:28.18,0:28:31.96,Chinese,,0,0,0,,最坏的情况是去主存中取数据
Dialogue: 0,0:28:33.20,0:28:37.50,English,,0,0,0,,So I guess worst case it comes from disk right if you have page miss page hope
Dialogue: 0,0:28:33.20,0:28:37.50,Chinese,,0,0,0,,嗯，最坏的情况应该是从磁盘中取数据，如果出现了缺页的话
Dialogue: 0,0:28:38.80,0:28:44.08,English,,0,0,0,,So eventually though the data comes back and is returned back to the CPU
Dialogue: 0,0:28:38.80,0:28:44.08,Chinese,,0,0,0,,但最后，都能取得数据，然后返回给 CPU
Dialogue: 0,0:28:46.80,0:28:53.24,English,,0,0,0,,Now that the page table entries on in intel systems have the following structure
Dialogue: 0,0:28:46.80,0:28:53.24,Chinese,,0,0,0,,下面我们看看 Intel 系统页表条目的结构（和书上 578 页一样）
Dialogue: 0,0:28:54.52,0:28:58.28,English,,0,0,0,,You can see that the level one page
Dialogue: 0,0:28:54.52,0:28:58.28,Chinese,,0,0,0,,我们先看看第一级的结构
Dialogue: 0,0:28:59.86,0:29:05.94,English,,0,0,0,,Which levels the level 1 through 3 pages
Dialogue: 0,0:28:59.86,0:29:05.94,Chinese,,0,0,0,,第一级、第二级、和第三级的页表条目结构是相同的
Dialogue: 0,0:29:06.60,0:29:13.70,English,,0,0,0,,Remember that page table entry points to the address of the next level page table ok
Dialogue: 0,0:29:06.60,0:29:13.70,Chinese,,0,0,0,,记住页表条目指向的是下一级页表的地址
Dialogue: 0,0:29:14.58,0:29:17.64,English,,0,0,0,,So it contains the level 1
Dialogue: 0,0:29:14.58,0:29:17.64,Chinese,,0,0,0,,所以，第一级的
Dialogue: 0,0:29:18.44,0:29:27.10,English,,0,0,0,,The level 1 PTE contains the base address the physical address of the level 2 table and so on
Dialogue: 0,0:29:18.44,0:29:27.10,Chinese,,0,0,0,,第一级的 PTE 含有第二级的页表物理基地址，后面的类似
Dialogue: 0,0:29:29.00,0:29:34.48,English,,0,0,0,,There's a bit,this is the valid bit they call it the p,p for present
Dialogue: 0,0:29:29.00,0:29:34.48,Chinese,,0,0,0,,这里有一个有效位，叫做 p，p 是 present 的首字母
Dialogue: 0,0:29:35.34,0:29:39.52,English,,0,0,0,,So this identifies whether that the page is in memory or not if it's not then it's
Dialogue: 0,0:29:35.34,0:29:39.52,Chinese,,0,0,0,,它标识子页表是否在物理内存中，如果在，值为 1，不在，值为 0
Dialogue: 0,0:29:40.60,0:29:46.40,English,,0,0,0,,Then there's a the location of it of that page table on disk
Dialogue: 0,0:29:40.60,0:29:46.40,Chinese,,0,0,0,,如果为 0，前面的就表示页表在磁盘上的位置
Dialogue: 0,0:29:48.12,0:29:58.12,English,,0,0,0,,There's bits that control whether the that page table can be read as it is read-only or it can be written
Dialogue: 0,0:29:48.12,0:29:58.12,Chinese,,0,0,0,,这里有一个 bit（R/W） 标识页表是只读的，还是可读可写的
Dialogue: 0,0:29:59.92,0:30:02.24,English,,0,0,0,,This corresponds to all reachable pages
Dialogue: 0,0:29:59.92,0:30:02.24,Chinese,,0,0,0,,这个标识对所有可达的页表都有效
Dialogue: 0,0:30:02.52,0:30:09.22,English,,0,0,0,,So you know you can...so all reachable pages,so all all of the portion of the address space
Dialogue: 0,0:30:02.52,0:30:09.22,Chinese,,0,0,0,,当前页表条目的所有可达的页表
Dialogue: 0,0:30:09.92,0:30:13.84,English,,0,0,0,,That's covered by this particular page table entry
Dialogue: 0,0:30:09.92,0:30:13.84,Chinese,,0,0,0,,所有覆盖的地址空间
Dialogue: 0,0:30:14.46,0:30:22.52,English,,0,0,0,,There's also a bit that indicates whether users can access that or if they need to be running in kernel mode
Dialogue: 0,0:30:14.46,0:30:22.52,Chinese,,0,0,0,,还有一个比特（U/S）指示是用户访问权限还是内核访问权限
Dialogue: 0,0:30:22.76,0:30:28.16,English,,0,0,0,,So this,this is how the kernel protects its code and data from from user programs
Dialogue: 0,0:30:22.76,0:30:28.16,Chinese,,0,0,0,,这就是内核保护它的数据和代码不受用户程序的破坏
Dialogue: 0,0:30:29.16,0:30:35.32,English,,0,0,0,,There's a bit here to indicate whether we should be using write back or write through for this
Dialogue: 0,0:30:29.16,0:30:35.32,Chinese,,0,0,0,,还有一个比特（WT）表示子页表使用直写还是写回缓存的策略
Dialogue: 0,0:30:37.14,0:30:42.02,English,,0,0,0,,For these page tables and all the systems I know of use use right back just because of the miss penalty
Dialogue: 0,0:30:37.14,0:30:42.02,Chinese,,0,0,0,,我所了解的所有系统都是使用的写回策略，因为缓存不命中的
Dialogue: 0,0:30:42.74,0:30:44.52,English,,0,0,0,,The enormous miss penalty
Dialogue: 0,0:30:42.74,0:30:44.52,Chinese,,0,0,0,,惩罚太高了（感觉应该是直写的代价太大了）
Dialogue: 0,0:30:48.42,0:30:49.90,English,,0,0,0,,Forget what cd does
Dialogue: 0,0:30:48.42,0:30:49.90,Chinese,,0,0,0,,CD 这个标识不用掌握
Dialogue: 0,0:30:50.36,0:30:55.06,English,,0,0,0,,There's a reference bit which is set when the MMU reads or writes
Dialogue: 0,0:30:50.36,0:30:55.06,Chinese,,0,0,0,,还有一个引用位（A），由 MMU 读和写时设置
Dialogue: 0,0:30:58.18,0:31:03.78,English,,0,0,0,,Reads or writes that the page table at this points to
Dialogue: 0,0:30:58.18,0:31:03.78,Chinese,,0,0,0,,表示读写对应的页表
Dialogue: 0,0:31:04.48,0:31:10.46,English,,0,0,0,,And then there's a there's a bit that indicates whether that the page size is either 4 kbytes or 4 megabytes
Dialogue: 0,0:31:04.48,0:31:10.46,Chinese,,0,0,0,,还有一个比特（PS）表示页面的大小为 4KB 或 4MB
Dialogue: 0,0:31:12.06,0:31:17.82,English,,0,0,0,,Okay and then there's this new disabled
Dialogue: 0,0:31:12.06,0:31:17.82,Chinese,,0,0,0,,还有一个标识位
Dialogue: 0,0:31:18.78,0:31:24.80,English,,0,0,0,,It's the xd bit which which disables allows you to disable or enable execution
Dialogue: 0,0:31:18.78,0:31:24.80,Chinese,,0,0,0,,名为 XD，标识能不能从这个 PTE 可访问的所有页中
Dialogue: 0,0:31:25.14,0:31:28.40,English,,0,0,0,,From all the pages reachable from that page table entry
Dialogue: 0,0:31:25.14,0:31:28.40,Chinese,,0,0,0,,去指令和执行指令
Dialogue: 0,0:31:28.74,0:31:34.90,English,,0,0,0,,So this is how the stack modern systems protect the stack from code injection attacks
Dialogue: 0,0:31:28.74,0:31:34.90,Chinese,,0,0,0,,这就是现代栈系统降低受到缓存区溢出攻击风险的一个方式
Dialogue: 0,0:31:36.38,0:31:42.62,English,,0,0,0,,Ok the last level page table entry points to a 4...it points to not another page table
Dialogue: 0,0:31:36.38,0:31:42.62,Chinese,,0,0,0,,最后一级的页表条目的页表物理基地址不再指向另一个页表
Dialogue: 0,0:31:42.90,0:31:44.92,English,,0,0,0,,But actually a page in memory ok
Dialogue: 0,0:31:42.90,0:31:44.92,Chinese,,0,0,0,,而是指令物理内存中某一页的基地址
Dialogue: 0,0:31:45.90,0:31:50.16,English,,0,0,0,,And so the same,we have the same bits here the same 40 bits that
Dialogue: 0,0:31:45.90,0:31:50.16,Chinese,,0,0,0,,同样的，还是使用 40 比特表示
Dialogue: 0,0:31:50.62,0:31:54.70,English,,0,0,0,,That give the physical address of that page in memory
Dialogue: 0,0:31:50.62,0:31:54.70,Chinese,,0,0,0,,但它的内容变成了物理内存中某一页的基地址
Dialogue: 0,0:31:55.46,0:32:02.34,English,,0,0,0,,And then we have the indication of whether that page is present in memory or not
Dialogue: 0,0:31:55.46,0:32:02.34,Chinese,,0,0,0,,也有一个比特表示物理页是不是在内存中
Dialogue: 0,0:32:02.44,0:32:07.18,English,,0,0,0,,And the rewrite and permission execute permission bits
Dialogue: 0,0:32:02.44,0:32:07.18,Chinese,,0,0,0,,还有表示读写和执行权限的标志
Dialogue: 0,0:32:09.00,0:32:10.22,English,,0,0,0,,Right through a write back
Dialogue: 0,0:32:09.00,0:32:10.22,Chinese,,0,0,0,,直写还是写回
Dialogue: 0,0:32:11.08,0:32:17.86,English,,0,0,0,,There's a reference,there's this reference bit and then there's a dirty bit which the MMU sets
Dialogue: 0,0:32:11.08,0:32:17.86,Chinese,,0,0,0,,一个引用位（A），和一个修改位
Dialogue: 0,0:32:18.40,0:32:20.50,English,,0,0,0,,When it writes to that page ok
Dialogue: 0,0:32:18.40,0:32:20.50,Chinese,,0,0,0,,由 MMU 读写时设置
Dialogue: 0,0:32:20.52,0:32:24.30,English,,0,0,0,,So this is how,this is how the os knows that it needs to
Dialogue: 0,0:32:20.52,0:32:24.30,Chinese,,0,0,0,,修改位用于告诉操作系统
Dialogue: 0,0:32:24.58,0:32:28.26,English,,0,0,0,,When it selects this if it selects this page as a victim it
Dialogue: 0,0:32:24.58,0:32:28.26,Chinese,,0,0,0,,当操作系统打算将这一页作为牺牲页时
Dialogue: 0,0:32:29.14,0:32:32.46,English,,0,0,0,,It looks at the dirty bit to see if it needs to write that page back
Dialogue: 0,0:32:29.14,0:32:32.46,Chinese,,0,0,0,,需要根据这一位来决定是否写回牺牲页
Dialogue: 0,0:32:33.52,0:32:34.38,English,,0,0,0,,Did you have a question
Dialogue: 0,0:32:33.52,0:32:34.38,Chinese,,0,0,0,,这位同学，你要提问吗
Dialogue: 0,0:32:34.46,0:32:41.20,English,,0,0,0,,[student speaking]
Dialogue: 0,0:32:34.46,0:32:41.20,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:32:41.22,0:32:47.72,English,,0,0,0,,Execute meaning that you can't execute any instructions
Dialogue: 0,0:32:41.22,0:32:47.72,Chinese,,0,0,0,,执行位（XD）无效说明你不能执行指令
Dialogue: 0,0:32:48.26,0:32:51.34,English,,0,0,0,,You can't load any instructions from that page okay
Dialogue: 0,0:32:48.26,0:32:51.34,Chinese,,0,0,0,,你不能从那个页表中加载指令
Dialogue: 0,0:32:54.88,0:33:02.06,English,,0,0,0,,The this user or supervisor mode it determines
Dialogue: 0,0:32:54.88,0:33:02.06,Chinese,,0,0,0,,用户或超级用户模式访问权限（U/S）
Dialogue: 0,0:33:02.06,0:33:06.56,English,,0,0,0,,Whether you can access that page,so if it's set to supervisor mode only the kernel
Dialogue: 0,0:33:02.06,0:33:06.56,Chinese,,0,0,0,,决定你是否能够访问那一页，如果它设置为超级用户，那么只有内核模式
Dialogue: 0,0:33:07.14,0:33:09.94,English,,0,0,0,,Can access that page anything on that page
Dialogue: 0,0:33:07.14,0:33:09.94,Chinese,,0,0,0,,才能访问那一页
Dialogue: 0,0:33:10.68,0:33:15.64,English,,0,0,0,,If it's set to U then anybody user code and kernel code can access that page
Dialogue: 0,0:33:10.68,0:33:15.64,Chinese,,0,0,0,,如果设置为用户模式，那么用户代码和内核代码都可以访问那一页
Dialogue: 0,0:33:17.90,0:33:18.66,English,,0,0,0,,Any other questions
Dialogue: 0,0:33:17.90,0:33:18.66,Chinese,,0,0,0,,还有其他问题吗
Dialogue: 0,0:33:23.16,0:33:26.88,English,,0,0,0,,Okay so then the way it does address translation then is
Dialogue: 0,0:33:23.16,0:33:26.88,Chinese,,0,0,0,,好的，这是地址翻译的图表
Dialogue: 0,0:33:28.02,0:33:32.14,English,,0,0,0,,The physical for this process,so the kernel for each process maintains
Dialogue: 0,0:33:28.02,0:33:32.14,Chinese,,0,0,0,,这是进程 PT 的物理地址，内核为每个进程都维护了一个 PT 的物理地址
Dialogue: 0,0:33:33.74,0:33:35.26,English,,0,0,0,,The you know set of page tables of course
Dialogue: 0,0:33:33.74,0:33:35.26,Chinese,,0,0,0,,当然，你还必须要知道页表
Dialogue: 0,0:33:35.42,0:33:40.84,English,,0,0,0,,And then it in it maintains the address of the first L1 one table
Dialogue: 0,0:33:35.42,0:33:40.84,Chinese,,0,0,0,,内核记录了 L1 的物理页表的首地址
Dialogue: 0,0:33:41.40,0:33:43.20,English,,0,0,0,,In a register called CR3
Dialogue: 0,0:33:41.40,0:33:43.20,Chinese,,0,0,0,,它被存放在 CR3 寄存器中
Dialogue: 0,0:33:44.08,0:33:45.54,English,,0,0,0,,Ok and this is a physical address
Dialogue: 0,0:33:44.08,0:33:45.54,Chinese,,0,0,0,,好的，这是一个物理地址
Dialogue: 0,0:33:47.74,0:33:51.26,English,,0,0,0,,Points to the base of the level 1 page table
Dialogue: 0,0:33:47.74,0:33:51.26,Chinese,,0,0,0,,指向 L1 页表的基地址
Dialogue: 0,0:33:52.32,0:34:00.62,English,,0,0,0,,The the VPN is broken up into 36 divided by 4  equals 9 bits
Dialogue: 0,0:33:52.32,0:34:00.62,Chinese,,0,0,0,,VPN 的 36 个比特划分为 4 组，每组 9 比特
Dialogue: 0,0:34:01.28,0:34:06.98,English,,0,0,0,,That give offsets into each page table entry absolutely
Dialogue: 0,0:34:01.28,0:34:06.98,Chinese,,0,0,0,,表示在页表条目中的绝对偏移
Dialogue: 0,0:34:11.78,0:34:20.18,English,,0,0,0,,Okay so there's 2^9 potential page table entries in each in each entries of these different page tables
Dialogue: 0,0:34:11.78,0:34:20.18,Chinese,,0,0,0,,在每个页表中都有 2^9 潜在的（可能没有创建）页表条目
Dialogue: 0,0:34:21.34,0:34:29.56,English,,0,0,0,,So the this VPN one is used the high order nine bits are used to compute an offset into that table
Dialogue: 0,0:34:21.34,0:34:29.56,Chinese,,0,0,0,,所以 VPN1 用来计算在 L1 PT 页中的偏移
Dialogue: 0,0:34:32.84,0:34:40.06,English,,0,0,0,,And you can see each level one PTE covers a swath of memory that's 512 gigabytes in size
Dialogue: 0,0:34:32.84,0:34:40.06,Chinese,,0,0,0,,你可以看到，每一个 L1 的 PTE 条目包含了 512G 的内存区域
Dialogue: 0,0:34:41.84,0:34:45.82,English,,0,0,0,,So you can see for most programs that exist today
Dialogue: 0,0:34:41.84,0:34:45.82,Chinese,,0,0,0,,所以，对于今天的大多数程序来说
Dialogue: 0,0:34:46.42,0:34:51.38,English,,0,0,0,,You only need one level one page table entry that to be initialized that would cover
Dialogue: 0,0:34:46.42,0:34:51.38,Chinese,,0,0,0,,只需要一个 L1 的页表
Dialogue: 0,0:34:52.24,0:34:55.12,English,,0,0,0,,You know the vast majority of your programs
Dialogue: 0,0:34:52.24,0:34:55.12,Chinese,,0,0,0,,绝大部分程序都只需要一个 L1 的页表
Dialogue: 0,0:34:56.72,0:34:59.76,English,,0,0,0,,The next nine bits it
Dialogue: 0,0:34:56.72,0:34:59.76,Chinese,,0,0,0,,接下来的 9 比特
Dialogue: 0,0:35:00.14,0:35:13.34,English,,0,0,0,,So the page table entry in the level 1,the page table points to the it gives the physical address of the level 2 page table
Dialogue: 0,0:35:00.14,0:35:13.34,Chinese,,0,0,0,,L1 的页表条目含有 L2 页表的物理地址
Dialogue: 0,0:35:14.28,0:35:22.78,English,,0,0,0,,And then the VPN bits in the virtual address are used to compute an offset into that table and so on
Dialogue: 0,0:35:14.28,0:35:22.78,Chinese,,0,0,0,,VPN2 用于计算在 L2 中的偏移，后面的 L3，L4 也是类似的
Dialogue: 0,0:35:22.80,0:35:29.60,English,,0,0,0,,It just cascades and then finally the last the low-order bits of the nine bits of the VPN
Dialogue: 0,0:35:22.80,0:35:29.60,Chinese,,0,0,0,,它们只是级联在了一起，VPN 的最后 9 比特（VPN4）
Dialogue: 0,0:35:30.06,0:35:34.62,English,,0,0,0,,Are used to form an offset into the level 4 page table
Dialogue: 0,0:35:30.06,0:35:34.62,Chinese,,0,0,0,,用于表示在 VPN4 中的偏移
Dialogue: 0,0:35:35.14,0:35:41.78,English,,0,0,0,,Which in which each each PTE points to an actual page
Dialogue: 0,0:35:35.14,0:35:41.78,Chinese,,0,0,0,,L4 中的每一个 PTE 都表示一个真实页
Dialogue: 0,0:35:42.88,0:35:44.60,English,,0,0,0,,It gives a physical page number
Dialogue: 0,0:35:42.88,0:35:44.60,Chinese,,0,0,0,,里面含有物理页号
Dialogue: 0,0:35:46.02,0:35:48.84,English,,0,0,0,,So then that physical page number then is extracted
Dialogue: 0,0:35:46.02,0:35:48.84,Chinese,,0,0,0,,然后 MMU 取出物理页号
Dialogue: 0,0:35:49.58,0:35:56.22,English,,0,0,0,,And concatenated with the VPO which we copied down to form a physical address
Dialogue: 0,0:35:49.58,0:35:56.22,Chinese,,0,0,0,,连上 VPO，构成物理地址
Dialogue: 0,0:36:02.40,0:36:04.08,English,,0,0,0,,Okay so that everybody good
Dialogue: 0,0:36:02.40,0:36:04.08,Chinese,,0,0,0,,大家都跟上了吗
Dialogue: 0,0:36:07.30,0:36:09.38,English,,0,0,0,,Okay now there's this cool trick which
Dialogue: 0,0:36:07.30,0:36:09.38,Chinese,,0,0,0,,好的，现在这里有一个很酷的技巧
Dialogue: 0,0:36:10.24,0:36:13.98,English,,0,0,0,,I was alluding to earlier that limits the size of L1 cache
Dialogue: 0,0:36:10.24,0:36:13.98,Chinese,,0,0,0,,就是我前面提到的关于限制 L1 缓存大小的原因
Dialogue: 0,0:36:15.22,0:36:17.90,English,,0,0,0,,Now to this point we've been we've...
Dialogue: 0,0:36:15.22,0:36:17.90,Chinese,,0,0,0,,现在我要说明原因
Dialogue: 0,0:36:18.96,0:36:26.46,English,,0,0,0,,Using a model where that the MMU takes the verge does address translation and
Dialogue: 0,0:36:18.96,0:36:26.46,Chinese,,0,0,0,,MMU 的地址翻译分为两个步骤
Dialogue: 0,0:36:26.86,0:36:30.42,English,,0,0,0,,Creates a complete virtual address a physical address
Dialogue: 0,0:36:26.86,0:36:30.42,Chinese,,0,0,0,,将物理地址翻译成物理地址
Dialogue: 0,0:36:30.42,0:36:32.50,English,,0,0,0,,And then it sends that physical address to the cache
Dialogue: 0,0:36:30.42,0:36:32.50,Chinese,,0,0,0,,然后把物理地址送到缓存
Dialogue: 0,0:36:33.42,0:36:40.26,English,,0,0,0,,Okay but in reality intel does this cute trick to speed up L1 cache accesses
Dialogue: 0,0:36:33.42,0:36:40.26,Chinese,,0,0,0,,事实上 Intel 使用了一些技巧加速了 L1 缓存的访问
Dialogue: 0,0:36:41.24,0:36:44.80,English,,0,0,0,,So what in reality what happens so we're given a virtual address
Dialogue: 0,0:36:41.24,0:36:44.80,Chinese,,0,0,0,,事实上，我们这里的虚拟地址
Dialogue: 0,0:36:47.72,0:36:52.60,English,,0,0,0,,And in this virtual address the index and of the physical
Dialogue: 0,0:36:47.72,0:36:52.60,Chinese,,0,0,0,,虚拟地址的索引位
Dialogue: 0,0:36:53.74,0:37:02.66,English,,0,0,0,,The index and offset bits in the physical address are identical exactly correspond to the PPO in the physical address
Dialogue: 0,0:36:53.74,0:37:02.66,Chinese,,0,0,0,,物理地址的索引位和偏移位和虚拟地址的 PPO 是一样的
Dialogue: 0,0:37:02.66,0:37:06.60,English,,0,0,0,,Which is exactly identical to the VPO in the virtual address
Dialogue: 0,0:37:02.66,0:37:06.60,Chinese,,0,0,0,,要强调的是，和虚拟地址的 VPO 是完全一样的
Dialogue: 0,0:37:08.08,0:37:13.06,English,,0,0,0,,So that means is that when the MMU is given a virtual address
Dialogue: 0,0:37:08.08,0:37:13.06,Chinese,,0,0,0,,那意味着，当 MMU 得到了虚拟地址之后
Dialogue: 0,0:37:14.20,0:37:18.20,English,,0,0,0,,It can send the VPO off to the L1 cache
Dialogue: 0,0:37:14.20,0:37:18.20,Chinese,,0,0,0,,它可以吧 VPO 送到 L1 缓存中
Dialogue: 0,0:37:18.82,0:37:23.06,English,,0,0,0,,Even though it's of even though L1 is physically addressed
Dialogue: 0,0:37:18.82,0:37:23.06,Chinese,,0,0,0,,尽管 L1 是通过物理地址寻址的
Dialogue: 0,0:37:23.78,0:37:28.34,English,,0,0,0,,We can send the VPO in the virtual address to the L1 cache because of this
Dialogue: 0,0:37:23.78,0:37:28.34,Chinese,,0,0,0,,但我们可以送入虚拟地址的 VPO，是因为
Dialogue: 0,0:37:29.60,0:37:34.32,English,,0,0,0,,Because of this phenomenon that the the PPO is identical to the VPO
Dialogue: 0,0:37:29.60,0:37:34.32,Chinese,,0,0,0,,PPO 和 VPO 相等
Dialogue: 0,0:37:35.06,0:37:38.70,English,,0,0,0,,So even before the MMU is doing any address translation
Dialogue: 0,0:37:35.06,0:37:38.70,Chinese,,0,0,0,,在 MMU 做地址翻译之前
Dialogue: 0,0:37:39.10,0:37:43.26,English,,0,0,0,,It can send this these these VPO bits to the cache
Dialogue: 0,0:37:39.10,0:37:43.26,Chinese,,0,0,0,,它可以把 VPO 送入缓存
Dialogue: 0,0:37:44.14,0:37:49.32,English,,0,0,0,,And then the the cache can get busy extracting the index the cache index bits
Dialogue: 0,0:37:44.14,0:37:49.32,Chinese,,0,0,0,,缓存可以取出索引位
Dialogue: 0,0:37:50.30,0:37:54.38,English,,0,0,0,,Looking up you know looking up all of the the lines
Dialogue: 0,0:37:50.30,0:37:54.38,Chinese,,0,0,0,,在所有的组里面去查找
Dialogue: 0,0:37:54.88,0:37:56.96,English,,0,0,0,,In that in that set
Dialogue: 0,0:37:54.88,0:37:56.96,Chinese,,0,0,0,,找到需要的条目
Dialogue: 0,0:37:57.82,0:38:00.94,English,,0,0,0,,And then have everything ready for the tag check
Dialogue: 0,0:37:57.82,0:38:00.94,Chinese,,0,0,0,,为标记位的检查做好准备
Dialogue: 0,0:38:01.74,0:38:05.40,English,,0,0,0,,Which can only occur after the address translation happens
Dialogue: 0,0:38:01.74,0:38:05.40,Chinese,,0,0,0,,标记位的检查只有在地址翻译结束之后进行
Dialogue: 0,0:38:05.40,0:38:10.32,English,,0,0,0,,So it can only occur once there's a physical address with a
Dialogue: 0,0:38:05.40,0:38:10.32,Chinese,,0,0,0,,因为拥有了物理地址
Dialogue: 0,0:38:11.04,0:38:13.76,English,,0,0,0,,From which we can extract the cash tag
Dialogue: 0,0:38:11.04,0:38:13.76,Chinese,,0,0,0,,才能取出其中的标记位
Dialogue: 0,0:38:15.10,0:38:18.88,English,,0,0,0,,So this...so there's a little bit of parallelism now in the L1 cache accesses
Dialogue: 0,0:38:15.10,0:38:18.88,Chinese,,0,0,0,,这个技巧可以使得在地址翻译的同时
Dialogue: 0,0:38:19.28,0:38:23.20,English,,0,0,0,,So we can be the MMU you can be doing address translation while
Dialogue: 0,0:38:19.28,0:38:23.20,Chinese,,0,0,0,,对 L1 的缓存进行访问
Dialogue: 0,0:38:23.74,0:38:27.82,English,,0,0,0,,The the L1 cache gets going on it's it's lookup
Dialogue: 0,0:38:23.74,0:38:27.82,Chinese,,0,0,0,,L1 将会查找缓存
Dialogue: 0,0:38:32.02,0:38:37.88,English,,0,0,0,,Okay so now let's look at how Linux implements virtual memory
Dialogue: 0,0:38:32.02,0:38:37.88,Chinese,,0,0,0,,好的，现在我们来看 Linux 的虚拟内存系统
Dialogue: 0,0:38:38.54,0:38:45.56,English,,0,0,0,,And in doing so will now get a clearer understanding of how fork and exec and loading really work
Dialogue: 0,0:38:38.54,0:38:45.56,Chinese,,0,0,0,,了解 fork 和 exec 和加载器是如何工作的
Dialogue: 0,0:38:46.48,0:38:49.84,English,,0,0,0,,Okay so we've seen this pictures...picture like this several times before
Dialogue: 0,0:38:46.48,0:38:49.84,Chinese,,0,0,0,,好，看这张图...这张图和我们以前看过多次的图差不多
Dialogue: 0,0:38:49.84,0:38:52.94,English,,0,0,0,,This is the virtual address space of the Linux process
Dialogue: 0,0:38:49.84,0:38:52.94,Chinese,,0,0,0,,这是 Linux 进程的虚拟地址空间
Dialogue: 0,0:38:53.40,0:38:59.46,English,,0,0,0,,Because we have...because of the way the virtual memory system works every process has a very similar-looking
Dialogue: 0,0:38:53.40,0:38:59.46,Chinese,,0,0,0,,因为...因为虚拟内存系统的工作方式，每个进程的虚拟地址空间看上去都差不多
Dialogue: 0,0:39:00.96,0:39:14.08,English,,0,0,0,,Address space that program text the code is always loaded at the same  0x400000 address
Dialogue: 0,0:39:00.96,0:39:14.08,Chinese,,0,0,0,,程序的代码，即 .text 段，总是位于 0x4000000 这个虚拟地址上
Dialogue: 0,0:39:14.58,0:39:22.12,English,,0,0,0,,And that's followed by initialized data which comes from the .data  section of the executable binary
Dialogue: 0,0:39:14.58,0:39:22.12,Chinese,,0,0,0,,接下来是来自可执行二进制文件的初始化数据，也就是 .data 段
Dialogue: 0,0:39:22.46,0:39:27.54,English,,0,0,0,,And then there's a BSS section which contains uninitialized data
Dialogue: 0,0:39:22.46,0:39:27.54,Chinese,,0,0,0,,紧接着是未初始化数据，
Dialogue: 0,0:39:28.24,0:39:30.62,English,,0,0,0,,That was defined in the binary and
Dialogue: 0,0:39:28.24,0:39:30.62,Chinese,,0,0,0,,即可执行二进制文件中所定义的 .bss 段
Dialogue: 0,0:39:32.22,0:39:38.76,English,,0,0,0,,Then the heap grows up from the initialized data
Dialogue: 0,0:39:32.22,0:39:38.76,Chinese,,0,0,0,,紧接着是堆，它从未初始化数据段开始向上生长
Dialogue: 0,0:39:39.52,0:39:46.54,English,,0,0,0,,There it's pointed to by a global variable called in a global variable in this process context called brk
Dialogue: 0,0:39:39.52,0:39:46.54,Chinese,,0,0,0,,有一个进程的全局变量 brk 指向堆顶
Dialogue: 0,0:39:46.62,0:39:51.84,English,,0,0,0,,So the kernel keeps track of where the top of the heap is for this process
Dialogue: 0,0:39:46.62,0:39:51.84,Chinese,,0,0,0,,所以内核能够知道进程的堆顶
Dialogue: 0,0:39:52.40,0:39:54.68,English,,0,0,0,,There's a memory mapped region for shared libraries
Dialogue: 0,0:39:52.40,0:39:54.68,Chinese,,0,0,0,,这里是共享库的内存映射区域
Dialogue: 0,0:39:55.46,0:40:02.70,English,,0,0,0,,And then at the top of the users,the user level,user user-accessible memory
Dialogue: 0,0:39:55.46,0:40:02.70,Chinese,,0,0,0,,在用户可以访问内存的顶部
Dialogue: 0,0:40:02.70,0:40:06.32,English,,0,0,0,,There's this user stack which grows down and which is pointed to by %rsp
Dialogue: 0,0:40:02.70,0:40:06.32,Chinese,,0,0,0,,有一个向下生长的用户栈，寄存器 %rsp 指向它的栈顶
Dialogue: 0,0:40:07.32,0:40:13.70,English,,0,0,0,,And then the kernel code and data live up in the upper portion of the address space
Dialogue: 0,0:40:07.32,0:40:13.70,Chinese,,0,0,0,,内核代码和数据在地址空间的上部
Dialogue: 0,0:40:14.90,0:40:20.64,English,,0,0,0,,Now this picture isn't completely accurate there's actually like a big gap
Dialogue: 0,0:40:14.90,0:40:20.64,Chinese,,0,0,0,,这张图不是完全正确的，实际上在
Dialogue: 0,0:40:21.32,0:40:26.18,English,,0,0,0,,Between the top of the stack and the beginning of the the kernel code and data
Dialogue: 0,0:40:21.32,0:40:26.18,Chinese,,0,0,0,,用户栈底和内核代码开始直接，有空白的地址空间
Dialogue: 0,0:40:26.84,0:40:30.96,English,,0,0,0,,I'm the reason for that is that the intel architecture says that that
Dialogue: 0,0:40:26.84,0:40:30.96,Chinese,,0,0,0,,原因就是在 Intel 的体系结构中
Dialogue: 0,0:40:31.80,0:40:34.22,English,,0,0,0,,So there's 48 virtual address bits right
Dialogue: 0,0:40:31.80,0:40:34.22,Chinese,,0,0,0,,虚拟地址是 48 位的
Dialogue: 0,0:40:35.30,0:40:41.06,English,,0,0,0,,So if so that's...it's the high-order bit of that 48 bit address  is 0 then
Dialogue: 0,0:40:35.30,0:40:41.06,Chinese,,0,0,0,,所以...如果 48 位的地址的最高位是 0 的话
Dialogue: 0,0:40:41.38,0:40:44.50,English,,0,0,0,,All the all the remaining bits have to be 0
Dialogue: 0,0:40:41.38,0:40:44.50,Chinese,,0,0,0,,那么剩下的比特也应该是 0
Dialogue: 0,0:40:45.06,0:40:50.38,English,,0,0,0,,All of the all of the height the remaining 16 the remainder of the 64 high order bits need to be 0
Dialogue: 0,0:40:45.06,0:40:50.38,Chinese,,0,0,0,,也就是未使用的 16 比特需要设置为 0
Dialogue: 0,0:40:51.10,0:40:59.00,English,,0,0,0,,So it's kind of like sign extension right and if that high order bit and the 48 bit address is 1
Dialogue: 0,0:40:51.10,0:40:59.00,Chinese,,0,0,0,,这有一点像符号扩展，如果 48 位虚拟地址的最高位为 1
Dialogue: 0,0:40:59.22,0:41:03.42,English,,0,0,0,,Then you extend the the 1 all the way up to the remaining high order bits
Dialogue: 0,0:40:59.22,0:41:03.42,Chinese,,0,0,0,,那么扩展出来的高位也是 1
Dialogue: 0,0:41:04.16,0:41:08.48,English,,0,0,0,,Okay so those are the only feasible bits so what that does is it creates a
Dialogue: 0,0:41:04.16,0:41:08.48,Chinese,,0,0,0,,所以，最高位是和 48 位地址保持一致的，所以它造成了...
Dialogue: 0,0:41:09.16,0:41:16.16,English,,0,0,0,,That the kerne's living,the kernel's living in in a portion of the virtual address space where that the 16
Dialogue: 0,0:41:09.16,0:41:16.16,Chinese,,0,0,0,,内核所在的虚拟地址空间的最高 16 位（说的 12，但应该属于口误）
Dialogue: 0,0:41:17.62,0:41:19.66,English,,0,0,0,,The 16 high order bits are all ones okay
Dialogue: 0,0:41:17.62,0:41:19.66,Chinese,,0,0,0,,全部为 1
Dialogue: 0,0:41:22.10,0:41:22.86,English,,0,0,0,,So that creates
Dialogue: 0,0:41:22.10,0:41:22.86,Chinese,,0,0,0,,这就造成了
Dialogue: 0,0:41:23.60,0:41:30.50,English,,0,0,0,,So you can think of it this is the kernel lives in the very top of the 64-bit address space
Dialogue: 0,0:41:23.60,0:41:30.50,Chinese,,0,0,0,,所以，你可以认 64 位虚拟地址空间的最顶部
Dialogue: 0,0:41:33.56,0:41:36.90,English,,0,0,0,,Okay another way to think of it in kernel addresses always start with one
Dialogue: 0,0:41:33.56,0:41:36.90,Chinese,,0,0,0,,你也可以认为内核的地址空间的最高位
Dialogue: 0,0:41:36.90,0:41:41.24,English,,0,0,0,,The most significant bit is 1 and user addresses always have a most significant bit of zero
Dialogue: 0,0:41:36.90,0:41:41.24,Chinese,,0,0,0,,或者说符号位是 1，用户的地址空间的符号位是 0
Dialogue: 0,0:41:42.16,0:41:45.74,English,,0,0,0,,Ok that's just the detail doesn't really affect things
Dialogue: 0,0:41:42.16,0:41:45.74,Chinese,,0,0,0,,好的，不理解这个细节也不影响我们对虚拟内存系统的理解
Dialogue: 0,0:41:45.74,0:41:48.20,English,,0,0,0,,But what is important is that
Dialogue: 0,0:41:45.74,0:41:48.20,Chinese,,0,0,0,,好的，下面要讲一个重点
Dialogue: 0,0:41:49.62,0:41:55.24,English,,0,0,0,,This is the process virtual address space of a process right and so
Dialogue: 0,0:41:49.62,0:41:55.24,Chinese,,0,0,0,,这是进程的虚拟地址空间
Dialogue: 0,0:41:55.94,0:42:00.16,English,,0,0,0,,The kernel exists in the virtual address space of each process in it
Dialogue: 0,0:41:55.94,0:42:00.16,Chinese,,0,0,0,,内核也在虚拟地址空间中
Dialogue: 0,0:42:01.54,0:42:03.66,English,,0,0,0,,So there's kernel code and data
Dialogue: 0,0:42:01.54,0:42:03.66,Chinese,,0,0,0,,这是内核的代码和数据
Dialogue: 0,0:42:04.32,0:42:08.90,English,,0,0,0,,And then it also maps it has a region of its virtual address space
Dialogue: 0,0:42:04.32,0:42:08.90,Chinese,,0,0,0,,内核也将一组连续的虚拟页面
Dialogue: 0,0:42:08.96,0:42:17.92,English,,0,0,0,,That map's Each element of this of this region with an address in physical memory
Dialogue: 0,0:42:08.96,0:42:17.92,Chinese,,0,0,0,,映射到一组相应的连续物理页面
Dialogue: 0,0:42:18.58,0:42:24.22,English,,0,0,0,,Ok so that this mapping this region is corresponds to the amount of DRAM on the chip
Dialogue: 0,0:42:18.58,0:42:24.22,Chinese,,0,0,0,,并且，页面的大小等同于系统中 DRAM 的总量
Dialogue: 0,0:42:24.84,0:42:28.50,English,,0,0,0,,And so it's a mechanism that the kernel uses to get access to physical addresses
Dialogue: 0,0:42:24.84,0:42:28.50,Chinese,,0,0,0,,这为内核提供了一种便利的方法来访问物理内存
Dialogue: 0,0:42:29.32,0:42:36.78,English,,0,0,0,,Ok and this is important because remember the kernel there you can never you can't turn off address translation
Dialogue: 0,0:42:29.32,0:42:36.78,Chinese,,0,0,0,,这是很重要的，因为你不能关闭地址翻译
Dialogue: 0,0:42:37.04,0:42:38.30,English,,0,0,0,,Right it's always happening
Dialogue: 0,0:42:37.04,0:42:38.30,Chinese,,0,0,0,,地址翻译总在进行
Dialogue: 0,0:42:41.02,0:42:46.76,English,,0,0,0,,Ok even when the kernel runs the address translation is happening the kernel is generating virtual addresses
Dialogue: 0,0:42:41.02,0:42:46.76,Chinese,,0,0,0,,内核代码运行的时候，地址翻译也在进行，内核产生的也是虚拟地址
Dialogue: 0,0:42:47.46,0:42:56.66,English,,0,0,0,,So this region this is map to a region of physical memory
Dialogue: 0,0:42:47.46,0:42:56.66,Chinese,,0,0,0,,这个虚拟地址空间就映射到物理内存
Dialogue: 0,0:42:56.66,0:43:03.20,English,,0,0,0,,So that if the kernel accesses bytes 0 of this block
Dialogue: 0,0:42:56.66,0:43:03.20,Chinese,,0,0,0,,如果内核访问这个区域的第 0 个字节
Dialogue: 0,0:43:03.52,0:43:11.46,English,,0,0,0,,It will actually be the address that the corresponding physical address will be 0
Dialogue: 0,0:43:03.52,0:43:11.46,Chinese,,0,0,0,,实际上内核访问的是物理地址的第 0 个字节
Dialogue: 0,0:43:11.94,0:43:16.46,English,,0,0,0,,And if it accesses a byte at an offset of 1 in this region
Dialogue: 0,0:43:11.94,0:43:16.46,Chinese,,0,0,0,,如果访问的是这个区域的第 1 个字节
Dialogue: 0,0:43:17.00,0:43:22.00,English,,0,0,0,,The physical address the corresponding physical address from the address translation process will be 1
Dialogue: 0,0:43:17.00,0:43:22.00,Chinese,,0,0,0,,实际上访问的是物理地址的第 1 个字节
Dialogue: 0,0:43:23.20,0:43:28.12,English,,0,0,0,,Ok so basically by reading and writing into this region the kernel is reading and writing into physical memory
Dialogue: 0,0:43:23.20,0:43:28.12,Chinese,,0,0,0,,所以，内核在这个虚拟内存上的读写，实际上是对物理内存的读写
Dialogue: 0,0:43:30.18,0:43:32.38,English,,0,0,0,,And this so there's portions of the kernel
Dialogue: 0,0:43:30.18,0:43:32.38,Chinese,,0,0,0,,所以，内核的这一部分空间
Dialogue: 0,0:43:33.06,0:43:35.14,English,,0,0,0,,That are identical for every process right
Dialogue: 0,0:43:33.06,0:43:35.14,Chinese,,0,0,0,,对所有进程来说都是一样的
Dialogue: 0,0:43:35.56,0:43:37.80,English,,0,0,0,,The codes the same it's not in
Dialogue: 0,0:43:35.56,0:43:37.80,Chinese,,0,0,0,,内核的代码对每个进程来说也是一样的
Dialogue: 0,0:43:38.30,0:43:43.26,English,,0,0,0,,But then there's process specific data structures that
Dialogue: 0,0:43:38.30,0:43:43.26,Chinese,,0,0,0,,但是，这里还有内核为每个进程保存的
Dialogue: 0,0:43:43.86,0:43:49.04,English,,0,0,0,,That the kernel maintains for each process that the context for that process
Dialogue: 0,0:43:43.86,0:43:49.04,Chinese,,0,0,0,,与进程相关的数据结构，进程的上下文
Dialogue: 0,0:43:49.30,0:43:54.94,English,,0,0,0,,So we refer to these refer to all these with all these data structures as the context
Dialogue: 0,0:43:49.30,0:43:54.94,Chinese,,0,0,0,,这些进程相关的数据结构作为进程的上下文
Dialogue: 0,0:43:57.36,0:44:00.50,English,,0,0,0,,And these of course have to be these would be different for each process
Dialogue: 0,0:43:57.36,0:44:00.50,Chinese,,0,0,0,,当然，对每个进程来说，进程上下文是不一样的
Dialogue: 0,0:44:02.72,0:44:07.96,English,,0,0,0,,Now Linux or organizes this virtual address region
Dialogue: 0,0:44:02.72,0:44:07.96,Chinese,,0,0,0,,现在，Linux 将内存组织成
Dialogue: 0,0:44:09.14,0:44:11.42,English,,0,0,0,,As a collection of what it calls areas
Dialogue: 0,0:44:09.14,0:44:11.42,Chinese,,0,0,0,,一些区域的集合
Dialogue: 0,0:44:12.42,0:44:20.64,English,,0,0,0,,An area is like a segment think of it as just a chunk of a contiguous chunk of related memory items
Dialogue: 0,0:44:12.42,0:44:20.64,Chinese,,0,0,0,,一个区域就是已经存在着的虚拟内存的连续片，这些页以某种方式相关联
Dialogue: 0,0:44:22.22,0:44:29.64,English,,0,0,0,,So the there's an area for code, there's an area for data
Dialogue: 0,0:44:22.22,0:44:29.64,Chinese,,0,0,0,,如，这是代码段，数据段
Dialogue: 0,0:44:31.90,0:44:36.44,English,,0,0,0,,Shared libraries have areas there's an area for the stack
Dialogue: 0,0:44:31.90,0:44:36.44,Chinese,,0,0,0,,共享库段，栈
Dialogue: 0,0:44:38.18,0:44:47.46,English,,0,0,0,,And it contains there's a structure for each process called the task struct
Dialogue: 0,0:44:38.18,0:44:47.46,Chinese,,0,0,0,,内核为系统中的每个进程维护一个单独的任务结构
Dialogue: 0,0:44:48.52,0:44:55.54,English,,0,0,0,,And this contains a pointer to the mm_struct which has a bunch of stuff in it but the
Dialogue: 0,0:44:48.52,0:44:55.54,Chinese,,0,0,0,,它包含一个指针，指向 mm_struct，mm_struct 有很多字段，但是我们只介绍几个
Dialogue: 0,0:44:56.62,0:45:00.96,English,,0,0,0,,It contains the address of the level one page table
Dialogue: 0,0:44:56.62,0:45:00.96,Chinese,,0,0,0,,mm_struct 有一个指向第一级页表的指针
Dialogue: 0,0:45:01.68,0:45:04.44,English,,0,0,0,,Okay so that's part of the context when this process runs
Dialogue: 0,0:45:01.68,0:45:04.44,Chinese,,0,0,0,,第一级页表指针是进程上下文的一部分，当进程
Dialogue: 0,0:45:06.28,0:45:12.14,English,,0,0,0,,When this process is scheduled the kernel will take this entry this PGD entry and he'll copy it into CR3
Dialogue: 0,0:45:06.28,0:45:12.14,Chinese,,0,0,0,,当进程被调度的时候，内核会把 pgd 条目拷贝到 CR3 中
Dialogue: 0,0:45:12.70,0:45:19.30,English,,0,0,0,,Okay so that's how it switches just by changing the CR3 register the kernel changes the address space
Dialogue: 0,0:45:12.70,0:45:19.30,Chinese,,0,0,0,,这就是切换地址空间的方法，仅仅需要改变寄存器 CR3 的值，就改变了地址空间
Dialogue: 0,0:45:20.84,0:45:27.30,English,,0,0,0,,So that's how...and once that once that CR3 value is changed then
Dialogue: 0,0:45:20.84,0:45:27.30,Chinese,,0,0,0,,好...一旦 CR3 的值改变了
Dialogue: 0,0:45:27.94,0:45:33.14,English,,0,0,0,,The process no longer has access to the previous processes page tables right so
Dialogue: 0,0:45:27.94,0:45:33.14,Chinese,,0,0,0,,当前进程就不能访问上一个进程的页表
Dialogue: 0,0:45:34.76,0:45:40.00,English,,0,0,0,,And then there's a and then there's a pointer to a list of these so-called area_struct
Dialogue: 0,0:45:34.76,0:45:40.00,Chinese,,0,0,0,,mm_struct 中还有一个指向一个名为 area_struct 的结构体
Dialogue: 0,0:45:41.14,0:45:46.04,English,,0,0,0,,The area_struct identifies the start at the beginning and end of the region
Dialogue: 0,0:45:41.14,0:45:46.04,Chinese,,0,0,0,,area_struct 定义了这个区域的起始处和结束处
Dialogue: 0,0:45:46.84,0:45:49.58,English,,0,0,0,,Any protections like is this read-only region
Dialogue: 0,0:45:46.84,0:45:49.58,Chinese,,0,0,0,,还有许可权限，像是不是只读的
Dialogue: 0,0:45:51.02,0:45:55.80,English,,0,0,0,,For example so the code section would be set as read-only or is it read right
Dialogue: 0,0:45:51.02,0:45:55.80,Chinese,,0,0,0,,例如，代码段就被设置为只读的
Dialogue: 0,0:45:56.36,0:46:00.08,English,,0,0,0,,And enter some other some other flags we'll see
Dialogue: 0,0:45:56.36,0:46:00.08,Chinese,,0,0,0,,area_struct 还定义了一些其他的标志位
Dialogue: 0,0:46:02.16,0:46:04.08,English,,0,0,0,,Which we'll see later
Dialogue: 0,0:46:02.16,0:46:04.08,Chinese,,0,0,0,,我们后面会看到
Dialogue: 0,0:46:08.30,0:46:13.78,English,,0,0,0,,In particular when we look at sharing and mapping for our purposes
Dialogue: 0,0:46:08.30,0:46:13.78,Chinese,,0,0,0,,当我们需要内存共享和映射时
Dialogue: 0,0:46:13.78,0:46:22.16,English,,0,0,0,,These flag bits tell us whether the page is shared with other processes or whether it's private to this process
Dialogue: 0,0:46:13.78,0:46:22.16,Chinese,,0,0,0,,这个标志位会告诉我们，这个页面是和其他进程共享的，还是这个进程私有的
Dialogue: 0,0:46:24.18,0:46:26.80,English,,0,0,0,,So normally the default is that pages are all private
Dialogue: 0,0:46:24.18,0:46:26.80,Chinese,,0,0,0,,所以，默认的情况下，页面是进程私有的
Dialogue: 0,0:46:27.18,0:46:34.82,English,,0,0,0,,But you have the option if you do a fair amount of work you can get processes to share memory
Dialogue: 0,0:46:27.18,0:46:34.82,Chinese,,0,0,0,,但是，如果你的程序需要共享内存，你可以设置这个标志
Dialogue: 0,0:46:35.96,0:46:38.28,English,,0,0,0,,Okay so let's look at what happens when we have a page fault
Dialogue: 0,0:46:35.96,0:46:38.28,Chinese,,0,0,0,,好的，接下来学习缺页异常处理
Dialogue: 0,0:46:41.62,0:46:51.66,English,,0,0,0,,So let's say so there's you know the processor issues and instruction the address translation process determines that the corresponding page
Dialogue: 0,0:46:41.62,0:46:51.66,Chinese,,0,0,0,,缺页异常是这样一种情况，在 MMU 翻译某个地址时
Dialogue: 0,0:46:53.30,0:46:54.56,English,,0,0,0,,Isn't contained in memory
Dialogue: 0,0:46:53.30,0:46:54.56,Chinese,,0,0,0,,发现地址对应的页表不在内存中
Dialogue: 0,0:46:54.90,0:46:56.14,English,,0,0,0,,So it triggers a page fault
Dialogue: 0,0:46:54.90,0:46:56.14,Chinese,,0,0,0,,就触发了一个缺页异常
Dialogue: 0,0:46:57.04,0:47:02.48,English,,0,0,0,,And so the fault handler there's several up there are several things that might have happened
Dialogue: 0,0:46:57.04,0:47:02.48,Chinese,,0,0,0,,默认的错误处理需要处理可能发生的多种情况
Dialogue: 0,0:47:03.20,0:47:10.06,English,,0,0,0,,One maybe we were reading memory it's faulted because that memory, that area doesn't even exist
Dialogue: 0,0:47:03.20,0:47:10.06,Chinese,,0,0,0,,一种情况是，当我们读内存的时候，发现地址是不合法的，因为访问了一个不存在的页面
Dialogue: 0,0:47:10.56,0:47:18.98,English,,0,0,0,,We haven't even created the kernel hasn't even created allocated that that page in virtual memory yet
Dialogue: 0,0:47:10.56,0:47:18.98,Chinese,,0,0,0,,我们还没有创建该页，内核没有在虚拟内存中创建分配该页。
Dialogue: 0,0:47:19.70,0:47:26.18,English,,0,0,0,,Okay so that's an error so that's accessing a non-existing page elicits a segment fault
Dialogue: 0,0:47:19.70,0:47:26.18,Chinese,,0,0,0,,这一一个错误，访问了一个不存在的页面，这会触发一个段错误
Dialogue: 0,0:47:26.18,0:47:31.86,English,,0,0,0,,So the kernel figures that out because it runs down this list of area_struct
Dialogue: 0,0:47:26.18,0:47:31.86,Chinese,,0,0,0,,内核能够识别出来，因为内核可以遍历 area_struct 这个链表
Dialogue: 0,0:47:32.40,0:47:37.64,English,,0,0,0,,And it it doesn't find this address anywhere within the ranges of those areas
Dialogue: 0,0:47:32.40,0:47:37.64,Chinese,,0,0,0,,然后内核发现它并不能找到地址在某个 area_struct 定义的区域之内
Dialogue: 0,0:47:38.40,0:47:42.90,English,,0,0,0,,So that's that's a segment fault caused by accessing a non-existing page
Dialogue: 0,0:47:38.40,0:47:42.90,Chinese,,0,0,0,,所以，这是一个因为访问不存在的页面触发的段错误
Dialogue: 0,0:47:44.06,0:47:55.02,English,,0,0,0,,Another possibility is that the instruction is attempting to write into a read-only segment of the virtual address space
Dialogue: 0,0:47:44.06,0:47:55.02,Chinese,,0,0,0,,另外一个可能的情况，就是指令尝试对一个虚拟地址空间的只读页面进行写操作
Dialogue: 0,0:47:56.18,0:48:04.00,English,,0,0,0,,Well they...there's nothing in the page table the MMU typically would check the permission bits
Dialogue: 0,0:47:56.18,0:48:04.00,Chinese,,0,0,0,,嗯...地址是合法的，MMU 通常会检查页表的权限位
Dialogue: 0,0:48:04.26,0:48:11.62,English,,0,0,0,,In the page table entry and throw an exception if the write is trying to to write to a read-only page
Dialogue: 0,0:48:04.26,0:48:11.62,Chinese,,0,0,0,,发现对一个只读页进行了写操作，然后抛出了一个异常
Dialogue: 0,0:48:11.62,0:48:14.10,English,,0,0,0,,But in this case that there was no page table entry
Dialogue: 0,0:48:11.62,0:48:14.10,Chinese,,0,0,0,,但是这里有一个问题，现在我们还没有页表条目
Dialogue: 0,0:48:14.23,0:48:17.80,English,,0,0,0,,Right at the trick that's why I triggered a page fault
Dialogue: 0,0:48:14.23,0:48:17.80,Chinese,,0,0,0,,因为我们触发了一个缺页异常
Dialogue: 0,0:48:18.54,0:48:21.56,English,,0,0,0,,So the MMU doesn't know that this is an illegal right
Dialogue: 0,0:48:18.54,0:48:21.56,Chinese,,0,0,0,,所以 MMU 并不知道这个写操作是异常的（并不知道那个页表是只读的）
Dialogue: 0,0:48:22.14,0:48:27.42,English,,0,0,0,,The kernel can check the protection for this area
Dialogue: 0,0:48:22.14,0:48:27.42,Chinese,,0,0,0,,但是内核可以检查这个区域的保护位
Dialogue: 0,0:48:28.22,0:48:32.34,English,,0,0,0,,And if in the case of text that will be set to read only
Dialogue: 0,0:48:28.22,0:48:32.34,Chinese,,0,0,0,,在这个例子中，是代码段，所以是只读的
Dialogue: 0,0:48:32.78,0:48:34.88,English,,0,0,0,,So that triggers a protection exception
Dialogue: 0,0:48:32.78,0:48:34.88,Chinese,,0,0,0,,所以会触发一个保护错误
Dialogue: 0,0:48:36.12,0:48:38.92,English,,0,0,0,,Which actually Linux reports as a segment fault
Dialogue: 0,0:48:36.12,0:48:38.92,Chinese,,0,0,0,,在 Linux 的报告中，是一个段错误
Dialogue: 0,0:48:42.58,0:48:49.76,English,,0,0,0,,And then the other option is that maybe we're trying to read data from a valid from a valid area
Dialogue: 0,0:48:42.58,0:48:49.76,Chinese,,0,0,0,,还有一种情况是我们打算从一个有效区域读取数据
Dialogue: 0,0:48:49.76,0:48:56.04,English,,0,0,0,,So then the kernel faults that the requested page in
Dialogue: 0,0:48:49.76,0:48:56.04,Chinese,,0,0,0,,然后内核陷入异常，换入需求的页表
Dialogue: 0,0:48:56.96,0:49:02.48,English,,0,0,0,,And and returns the requested data back to the back to the CPU
Dialogue: 0,0:48:56.96,0:49:02.48,Chinese,,0,0,0,,然后取出对应的数据返回给 CPU
Dialogue: 0,0:49:05.74,0:49:08.60,English,,0,0,0,,Okay the last thing I want to look at today is it yes
Dialogue: 0,0:49:05.74,0:49:08.60,Chinese,,0,0,0,,好的，今天我要讲的最后一个内容是，请讲
Dialogue: 0,0:49:08.60,0:49:17.94,English,,0,0,0,,[student speaking]
Dialogue: 0,0:49:08.60,0:49:17.94,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:49:17.94,0:49:19.02,English,,0,0,0,,Okay sorry
Dialogue: 0,0:49:17.94,0:49:19.02,Chinese,,0,0,0,,抱歉，我没听清楚
Dialogue: 0,0:49:19.22,0:49:26.30,English,,0,0,0,,[student speaking]
Dialogue: 0,0:49:19.22,0:49:26.30,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:49:26.32,0:49:30.46,English,,0,0,0,,Yeah so all of the segments are contained in the executable file
Dialogue: 0,0:49:26.32,0:49:30.46,Chinese,,0,0,0,,对的，所有的段是在可执行文件中的
Dialogue: 0,0:49:31.54,0:49:35.62,English,,0,0,0,,Okay there's a...at the beginning of the elf binary
Dialogue: 0,0:49:31.54,0:49:35.62,Chinese,,0,0,0,,嗯...在 elf 二进制文件的开始
Dialogue: 0,0:49:35.64,0:49:38.70,English,,0,0,0,,There's a list of all the different segments
Dialogue: 0,0:49:35.64,0:49:38.70,Chinese,,0,0,0,,有一个表，列出了所有的段
Dialogue: 0,0:49:40.06,0:49:50.86,English,,0,0,0,,Okay so when kernel loads the binary creates areas for each of those each of those segments
Dialogue: 0,0:49:40.06,0:49:50.86,Chinese,,0,0,0,,当内核加载可执行文件的时候，它为每一个段都创建了一个独立的区域
Dialogue: 0,0:49:53.80,0:49:54.88,English,,0,0,0,,Good yes
Dialogue: 0,0:49:53.80,0:49:54.88,Chinese,,0,0,0,,好，请讲
Dialogue: 0,0:49:54.88,0:50:05.38,English,,0,0,0,,[student speaking]
Dialogue: 0,0:49:54.88,0:50:05.38,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:50:05.38,0:50:06.90,English,,0,0,0,,This is after the page fault
Dialogue: 0,0:50:05.38,0:50:06.90,Chinese,,0,0,0,,这是在段错误之后
Dialogue: 0,0:50:06.94,0:50:12.34,English,,0,0,0,,Yes
Dialogue: 0,0:50:06.94,0:50:12.34,Chinese,,0,0,0,,是的
Dialogue: 0,0:50:12.34,0:50:14.88,English,,0,0,0,,The MMU doesn't know about these
Dialogue: 0,0:50:12.34,0:50:14.88,Chinese,,0,0,0,,MMU 并不知道这个
Dialogue: 0,0:50:15.50,0:50:18.24,English,,0,0,0,,Okay so yeah the question is exactly how are these area_struct use
Dialogue: 0,0:50:15.50,0:50:18.24,Chinese,,0,0,0,,好的，这个学生的问题是 area_struct 是怎么使用的
Dialogue: 0,0:50:18.24,0:50:21.66,English,,0,0,0,,And they're used,they're used
Dialogue: 0,0:50:18.24,0:50:21.66,Chinese,,0,0,0,,在这个例子中
Dialogue: 0,0:50:21.68,0:50:25.52,English,,0,0,0,,In this content in this example anyway they're used during the page faults
Dialogue: 0,0:50:21.68,0:50:25.52,Chinese,,0,0,0,,它们在缺页处理的时候使用
Dialogue: 0,0:50:26.43,0:50:32.32,English,,0,0,0,,Okay and also just as an aside
Dialogue: 0,0:50:26.43,0:50:32.32,Chinese,,0,0,0,,说一句题外话
Dialogue: 0,0:50:32.52,0:50:35.40,English,,0,0,0,,I've shown these actually are implemented as list
Dialogue: 0,0:50:32.52,0:50:35.40,Chinese,,0,0,0,,我这里展示的 area_struct 结构体是一个链表
Dialogue: 0,0:50:35.40,0:50:41.68,English,,0,0,0,,But in the real system they use some kind of tree,red black tree or something like that
Dialogue: 0,0:50:35.40,0:50:41.68,Chinese,,0,0,0,,但是在真实的系统中，这是某种树，红黑树或其他类似的树
Dialogue: 0,0:50:47.22,0:50:49.84,English,,0,0,0,,Okay so I want to talk about the last thing we're going to look at today is
Dialogue: 0,0:50:47.22,0:50:49.84,Chinese,,0,0,0,,今天我要讲的最后一个内容是
Dialogue: 0,0:50:50.64,0:50:54.94,English,,0,0,0,,This is interesting idea called mapping
Dialogue: 0,0:50:50.64,0:50:54.94,Chinese,,0,0,0,,关于内存映射的
Dialogue: 0,0:50:55.88,0:51:01.24,English,,0,0,0,,It will help us understand a lot about what's going on in the system when it executes fork and exec
Dialogue: 0,0:50:55.88,0:51:01.24,Chinese,,0,0,0,,内存映射可以帮助我们理解执行 fork 和 exec 系统调用的时候，系统干了什么事
Dialogue: 0,0:51:03.46,0:51:10.76,English,,0,0,0,,So VM areas are initialized by associating them with disk objects
Dialogue: 0,0:51:03.46,0:51:10.76,Chinese,,0,0,0,,虚拟内存区域初始化的时候和磁盘上的对象关联起来
Dialogue: 0,0:51:12.45,0:51:18.40,English,,0,0,0,,Okay and so this process in a using a process known as memory mapping right
Dialogue: 0,0:51:12.45,0:51:18.40,Chinese,,0,0,0,,所以，这个过程叫做内存映射
Dialogue: 0,0:51:19.24,0:51:30.24,English,,0,0,0,,So the every area and thus every page within that area is associated with some portion of a file
Dialogue: 0,0:51:19.24,0:51:30.24,Chinese,,0,0,0,,每一个区域，即每个区域包含的页，都和文件中的某些部分关联起来
Dialogue: 0,0:51:31.50,0:51:38.64,English,,0,0,0,,And it's that and that's the initial contents of the pages in that area come from that file
Dialogue: 0,0:51:31.50,0:51:38.64,Chinese,,0,0,0,,并且初始的内容来类从那个文件
Dialogue: 0,0:51:40.24,0:51:48.38,English,,0,0,0,,Okay so the an area can be backed by what we call or get its initial values from either a regular file on disk
Dialogue: 0,0:51:40.24,0:51:48.38,Chinese,,0,0,0,,也就是说，当我们第一次引用到某个区域，它的初始值来自于磁盘上的普通文件
Dialogue: 0,0:51:49.38,0:51:56.72,English,,0,0,0,,So this we saw this...so in the case of pages that are containing code
Dialogue: 0,0:51:49.38,0:51:56.72,Chinese,,0,0,0,,所以...例如，如果该区域是代码区
Dialogue: 0,0:51:57.72,0:52:04.28,English,,0,0,0,,That area that has a code is mapped to a portion of the executable binary
Dialogue: 0,0:51:57.72,0:52:04.28,Chinese,,0,0,0,,那么这个区域会被映射到一个可执行文件的某个部分
Dialogue: 0,0:52:05.26,0:52:10.36,English,,0,0,0,,Okay and then the initial values of that area come from that executable binary file
Dialogue: 0,0:52:05.26,0:52:10.36,Chinese,,0,0,0,,也就是，这个区域的初始值就来自哪个可执行文件
Dialogue: 0,0:52:11.88,0:52:16.34,English,,0,0,0,,Okay so that's how we get that's how we get data copied
Dialogue: 0,0:52:11.88,0:52:16.34,Chinese,,0,0,0,,这就是我们如何
Dialogue: 0,0:52:17.14,0:52:21.48,English,,0,0,0,,Back and forth from or from executables into memory
Dialogue: 0,0:52:17.14,0:52:21.48,Chinese,,0,0,0,,来回的从磁盘拷贝数据到内存
Dialogue: 0,0:52:23.08,0:52:26.50,English,,0,0,0,,The file can also be an anonymous file meaning
Dialogue: 0,0:52:23.08,0:52:26.50,Chinese,,0,0,0,,文件也可以是一个匿名文件
Dialogue: 0,0:52:27.34,0:52:31.58,English,,0,0,0,,It's a file,it's some anonymous file that consists of all zeros
Dialogue: 0,0:52:27.34,0:52:31.58,Chinese,,0,0,0,,匿名文件是有内核创建，包含的全是二进制 0
Dialogue: 0,0:52:31.84,0:52:36.74,English,,0,0,0,,It's an anonymous file of arbitrary size that consists of all zeros
Dialogue: 0,0:52:31.84,0:52:36.74,Chinese,,0,0,0,,匿名文件的大小是任意的，但只含有 0
Dialogue: 0,0:52:37.06,0:52:40.60,English,,0,0,0,,And so of course it doesn't exist,this is just a trick
Dialogue: 0,0:52:37.06,0:52:40.60,Chinese,,0,0,0,,当然，这个文件并不真实存在，它是一个技巧
Dialogue: 0,0:52:40.60,0:52:48.28,English,,0,0,0,,So this allows us to create pages that are initialized to all zeros
Dialogue: 0,0:52:40.60,0:52:48.28,Chinese,,0,0,0,,允许我们创建一个全为 0 的页
Dialogue: 0,0:52:49.52,0:52:55.88,English,,0,0,0,,And then so the if a page is associated with the anonymous file
Dialogue: 0,0:52:49.52,0:52:55.88,Chinese,,0,0,0,,如果一个页面和匿名文件关联在了一起
Dialogue: 0,0:52:56.24,0:53:00.04,English,,0,0,0,,Then the first fault will allocate a physical page of all zeros
Dialogue: 0,0:52:56.24,0:53:00.04,Chinese,,0,0,0,,当第一次引用页面的时候，会分配一个物理页，并初始化为全 0
Dialogue: 0,0:53:00.88,0:53:03.10,English,,0,0,0,,So this is called the demand zero page
Dialogue: 0,0:53:00.88,0:53:03.10,Chinese,,0,0,0,,所以，这也被叫做请求二进制全零的页
Dialogue: 0,0:53:04.98,0:53:08.26,English,,0,0,0,,Okay so it's brought in on demand and it's initialized to zero
Dialogue: 0,0:53:04.98,0:53:08.26,Chinese,,0,0,0,,它按需调入，并且初始化为 0
Dialogue: 0,0:53:10.12,0:53:16.18,English,,0,0,0,,Once now once an anonymous,once a file wants a page that's backed by an anonymous file
Dialogue: 0,0:53:10.12,0:53:16.18,Chinese,,0,0,0,,一旦一个被匿名文件初始化的页
Dialogue: 0,0:53:16.66,0:53:21.77,English,,0,0,0,,Is written to or dirtied then it's like any other page okay and
Dialogue: 0,0:53:16.66,0:53:21.77,Chinese,,0,0,0,,被修改过了，和其他页一样
Dialogue: 0,0:53:22.42,0:53:27.28,English,,0,0,0,,Then dirty pages are copied back and forth between memory and a special swap flow
Dialogue: 0,0:53:22.42,0:53:27.28,Chinese,,0,0,0,,它需要同步到文件中，只不过这里是同步到内核维护的专门的交换文件
Dialogue: 0,0:53:32.52,0:53:36.96,English,,0,0,0,,Now you remember in normally processes don't share anything with each other
Dialogue: 0,0:53:32.52,0:53:36.96,Chinese,,0,0,0,,前面说过，正常的进程不会和其他进程有任何共享的东西
Dialogue: 0,0:53:37.84,0:53:45.94,English,,0,0,0,,Okay but it's possible using this idea of mapping for processes to be mapped to the same objects
Dialogue: 0,0:53:37.84,0:53:45.94,Chinese,,0,0,0,,但是现在利用内存映射可以实现进程间共享对象，因为进程可以映射虚拟内存的区域到同一个对象
Dialogue: 0,0:53:46.74,0:53:54.50,English,,0,0,0,,So suppose that we have two processes with their own separate virtual address spaces
Dialogue: 0,0:53:46.74,0:53:54.50,Chinese,,0,0,0,,假设我们有两个进程，它们都有各自不同的虚拟地址空间
Dialogue: 0,0:53:55.20,0:53:59.60,English,,0,0,0,,And the pages in these two processes are being mapped to portions of physical memory
Dialogue: 0,0:53:55.20,0:53:59.60,Chinese,,0,0,0,,它们的虚拟页都被映射到物理内存的某个部分
Dialogue: 0,0:54:00.66,0:54:06.22,English,,0,0,0,,So let's say there's a an area,a segment in process one
Dialogue: 0,0:54:00.66,0:54:06.22,Chinese,,0,0,0,,我们假设有一个区域，是进程 1 的某个段
Dialogue: 0,0:54:06.86,0:54:11.74,English,,0,0,0,,That's been mapped to this object so to this portion of a file
Dialogue: 0,0:54:06.86,0:54:11.74,Chinese,,0,0,0,,被映射到这个对象，也就是一个文件的某个部分
Dialogue: 0,0:54:16.36,0:54:23.48,English,,0,0,0,,Okay now process 2 can also map to that same object
Dialogue: 0,0:54:16.36,0:54:23.48,Chinese,,0,0,0,,现在，进程 2 也把相同的对象映射到自己的虚拟内存空间
Dialogue: 0,0:54:26.14,0:54:29.50,English,,0,0,0,,And notice even at a completely different at virtual address
Dialogue: 0,0:54:26.14,0:54:29.50,Chinese,,0,0,0,,注意它们是不同的虚拟地址空间
Dialogue: 0,0:54:29.96,0:54:33.36,English,,0,0,0,,So there's no relationship between
Dialogue: 0,0:54:29.96,0:54:33.36,Chinese,,0,0,0,,所以，
Dialogue: 0,0:54:33.78,0:54:38.84,English,,0,0,0,,The region of this virtual address space that's mapped to the shared object in process 1
Dialogue: 0,0:54:33.78,0:54:38.84,Chinese,,0,0,0,,进程 1 映射到这个共享对象的区域
Dialogue: 0,0:54:40.04,0:54:44.92,English,,0,0,0,,And the region of the virtual address space in process 2 that's mapped to the same object
Dialogue: 0,0:54:40.04,0:54:44.92,Chinese,,0,0,0,,和进程 2 映射到共享对象的区域，是没有关系的
Dialogue: 0,0:54:46.62,0:54:50.70,English,,0,0,0,,And the kernel because these objects have unique names right their files
Dialogue: 0,0:54:46.62,0:54:50.70,Chinese,,0,0,0,,因为这个共享对象有一个唯一的名字，所以内核
Dialogue: 0,0:54:51.94,0:54:58.33,English,,0,0,0,,The kernel can check the other processes to see if any other processes are mapping to that object
Dialogue: 0,0:54:51.94,0:54:58.33,Chinese,,0,0,0,,可以检查有没有其他进程映射到那个对象
Dialogue: 0,0:54:58.62,0:55:03.58,English,,0,0,0,,And if so map this region of the virtual address space to those same physical pages
Dialogue: 0,0:54:58.62,0:55:03.58,Chinese,,0,0,0,,如果有，那么内核就把虚拟地址对应的区域映射到相同的物理地址
Dialogue: 0,0:55:04.46,0:55:06.02,English,,0,0,0,,Ok so now we have a situation
Dialogue: 0,0:55:04.46,0:55:06.02,Chinese,,0,0,0,,所以，现在我们有了一个办法
Dialogue: 0,0:55:06.68,0:55:11.86,English,,0,0,0,,Where each process is accessing some chunk of its virtual address space
Dialogue: 0,0:55:06.68,0:55:11.86,Chinese,,0,0,0,,使得每个独立虚拟地址空间的进程可以访问一些连续片（chunk）
Dialogue: 0,0:55:12.30,0:55:19.40,English,,0,0,0,,And those accesses are actually being done on the same region of physical memory
Dialogue: 0,0:55:12.30,0:55:19.40,Chinese,,0,0,0,,达到对同一个物理地址区域访问的目的
Dialogue: 0,0:55:20.62,0:55:25.32,English,,0,0,0,,So this might be useful you might imagine um you know say these processes are servers
Dialogue: 0,0:55:20.62,0:55:25.32,Chinese,,0,0,0,,这个方法是很有用的，想像一下，这些进程都是服务器
Dialogue: 0,0:55:26.26,0:55:30.62,English,,0,0,0,,You know say you've worked a bunch of copies of Apache
Dialogue: 0,0:55:26.26,0:55:30.62,Chinese,,0,0,0,,是多个 Apache 服务器的进程
Dialogue: 0,0:55:31.14,0:55:33.32,English,,0,0,0,,You might want to have some kind of shared cache
Dialogue: 0,0:55:31.14,0:55:33.32,Chinese,,0,0,0,,你可能需要共享一些缓存
Dialogue: 0,0:55:34.22,0:55:38.68,English,,0,0,0,,Maybe a shared memory cache between those, all those different copies
Dialogue: 0,0:55:34.22,0:55:38.68,Chinese,,0,0,0,,在这些 Apache 服务器之间共享内存缓存
Dialogue: 0,0:55:39.44,0:55:43.34,English,,0,0,0,,So this would be...in this case your cache would be a file
Dialogue: 0,0:55:39.44,0:55:43.34,Chinese,,0,0,0,,所以...在这个例子中，你的缓存是一个文件
Dialogue: 0,0:55:45.46,0:55:50.84,English,,0,0,0,,A file on disk and then as you access regions of that file
Dialogue: 0,0:55:45.46,0:55:50.84,Chinese,,0,0,0,,一个在磁盘上的文件，然后，当你访问文件的页表区域时
Dialogue: 0,0:55:51.94,0:55:54.02,English,,0,0,0,,Those pages get copied into physical memory
Dialogue: 0,0:55:51.94,0:55:54.02,Chinese,,0,0,0,,这些页表会拷贝到内存
Dialogue: 0,0:55:54.02,0:56:00.54,English,,0,0,0,,So that would allow you to have a shared cache in the memory across all those processes
Dialogue: 0,0:55:54.02,0:56:00.54,Chinese,,0,0,0,,所有的进程都可以共享这个缓存
Dialogue: 0,0:56:02.24,0:56:02.64,English,,0,0,0,,Now
Dialogue: 0,0:56:02.24,0:56:02.64,Chinese,,0,0,0,,好的
Dialogue: 0,0:56:04.36,0:56:13.12,English,,0,0,0,,Sometimes there's another objects can also be private meaning they shouldn't be shared ok across processes
Dialogue: 0,0:56:04.36,0:56:13.12,Chinese,,0,0,0,,有时候，我们也需要是私有对象，意味着他们不能在进程间共享
Dialogue: 0,0:56:15.90,0:56:24.14,English,,0,0,0,,And there's a special an especially interesting kind of private object called the private copy-on-write object
Dialogue: 0,0:56:15.90,0:56:24.14,Chinese,,0,0,0,,还有一种有趣的私有对象，叫做私有写时复制对象
Dialogue: 0,0:56:26.74,0:56:29.52,English,,0,0,0,,So the idea here
Dialogue: 0,0:56:26.74,0:56:29.52,Chinese,,0,0,0,,它的思想是
Dialogue: 0,0:56:30.50,0:56:35.20,English,,0,0,0,,And you'll see when... this will become very important when we look and see how fork actually works
Dialogue: 0,0:56:30.50,0:56:35.20,Chinese,,0,0,0,,后面我们会看到...
Dialogue: 0,0:56:36.08,0:56:38.80,English,,0,0,0,,The idea is that we have an object
Dialogue: 0,0:56:36.08,0:56:38.80,Chinese,,0,0,0,,它的思想是，我们有一个对象
Dialogue: 0,0:56:39.44,0:56:45.10,English,,0,0,0,,That's it we have two processes that are mapping to the same object
Dialogue: 0,0:56:39.44,0:56:45.10,Chinese,,0,0,0,,两个进程都将这个对象映射到他们的虚拟地址空间
Dialogue: 0,0:56:46.36,0:56:52.44,English,,0,0,0,,But  instead of this being a shared object it's what we call this private copy-on-write object
Dialogue: 0,0:56:46.36,0:56:52.44,Chinese,,0,0,0,,但是，这个对象不是共享对象，而是一个私有写时复制对象
Dialogue: 0,0:56:53.10,0:56:58.54,English,,0,0,0,,And just like before these two regions of the virtual address space
Dialogue: 0,0:56:53.10,0:56:58.54,Chinese,,0,0,0,,和上面讲的一样，两个虚拟地址空间的对应区域
Dialogue: 0,0:56:58.86,0:57:01.62,English,,0,0,0,,Map into the same region of physical memory
Dialogue: 0,0:56:58.86,0:57:01.62,Chinese,,0,0,0,,映射到了相同的物理区域
Dialogue: 0,0:57:02.20,0:57:06.10,English,,0,0,0,,But they're tagged using the and this is where those flag bits come in
Dialogue: 0,0:57:02.20,0:57:06.10,Chinese,,0,0,0,,他们使用标志位进行标记
Dialogue: 0,0:57:07.28,0:57:13.74,English,,0,0,0,,There they're...the pages in this area are flagged as private copy-on-write
Dialogue: 0,0:57:07.28,0:57:13.74,Chinese,,0,0,0,,他们...他们这个区域的页表标为私有写时复制
Dialogue: 0,0:57:16.44,0:57:20.22,English,,0,0,0,,Okay now what that means is normally if this was a shared object
Dialogue: 0,0:57:16.44,0:57:20.22,Chinese,,0,0,0,,这意味着，在正常的时候，它是一个共享对象
Dialogue: 0,0:57:24.08,0:57:31.04,English,,0,0,0,,Let's say this was a shared object and processed to wrote did a write into this region of the virtual address space
Dialogue: 0,0:57:24.08,0:57:31.04,Chinese,,0,0,0,,共享对象的含义是，如果一个进程对共享对象对应的虚拟地址空间进行了写操作
Dialogue: 0,0:57:31.84,0:57:37.02,English,,0,0,0,,If this was a shared object that write would also be reflected on the disk on the file on disk
Dialogue: 0,0:57:31.84,0:57:37.02,Chinese,,0,0,0,,那么这个写操作也会同步到磁盘上的文件
Dialogue: 0,0:57:41.54,0:57:50.74,English,,0,0,0,,But if it's flagged this instead of being shared if that area,this area is marked as flagged as private copy-on-write
Dialogue: 0,0:57:41.54,0:57:50.74,Chinese,,0,0,0,,但是如果这个对象被标记为私有写时复制而不是共享
Dialogue: 0,0:57:55.02,0:57:59.26,English,,0,0,0,,Then if process to does a write to a page in that area
Dialogue: 0,0:57:55.02,0:57:59.26,Chinese,,0,0,0,,那么，如果进程对这个区域进行了写操作
Dialogue: 0,0:58:01.18,0:58:07.02,English,,0,0,0,,Then instead of it reflecting that change in physical memory and on the the shared object
Dialogue: 0,0:58:01.18,0:58:07.02,Chinese,,0,0,0,,它并不会同步写操作到共享对象的物理内存
Dialogue: 0,0:58:08.40,0:58:12.52,English,,0,0,0,,It makes a copy of that of that page
Dialogue: 0,0:58:08.40,0:58:12.52,Chinese,,0,0,0,,而是将页面拷贝一份
Dialogue: 0,0:58:13.02,0:58:18.34,English,,0,0,0,,A separate copy of that page and maps it to some unused portion of the physical address space
Dialogue: 0,0:58:13.02,0:58:18.34,Chinese,,0,0,0,,一个独立的拷贝页，并且把它映射到一个没有使用的物理地址
Dialogue: 0,0:58:21.18,0:58:24.38,English,,0,0,0,,Okay so that's why,that's why we have the name copy on write
Dialogue: 0,0:58:21.18,0:58:24.38,Chinese,,0,0,0,,所以，这就是叫写时复制的原因
Dialogue: 0,0:58:25.24,0:58:30.88,English,,0,0,0,,So if we...but if we read,if we read values from this area
Dialogue: 0,0:58:25.24,0:58:30.88,Chinese,,0,0,0,,所以...如果我们从这个区域读数据
Dialogue: 0,0:58:31.50,0:58:36.32,English,,0,0,0,,Nothing happens it just it will just read from the this portion of the physical address space
Dialogue: 0,0:58:31.50,0:58:36.32,Chinese,,0,0,0,,那么和共享对象一样，它仅仅从对应的物理地址空间读数据
Dialogue: 0,0:58:36.94,0:58:40.06,English,,0,0,0,,Okay it's only when we do a write to some page
Dialogue: 0,0:58:36.94,0:58:40.06,Chinese,,0,0,0,,而当我们对
Dialogue: 0,0:58:41.82,0:58:44.66,English,,0,0,0,,In an area that's flagged as private copy-on-write
Dialogue: 0,0:58:41.82,0:58:44.66,Chinese,,0,0,0,,标记为私有写时复制的区域写的时候
Dialogue: 0,0:58:44.82,0:58:50.82,English,,0,0,0,,It's only when we do that write that the system first makes a copy of that page
Dialogue: 0,0:58:44.82,0:58:50.82,Chinese,,0,0,0,,第一次写的时候，系统会拷贝那个物理页
Dialogue: 0,0:58:52.16,0:58:53.38,English,,0,0,0,,And then does the write
Dialogue: 0,0:58:52.16,0:58:53.38,Chinese,,0,0,0,,然后在拷贝的物理页上写
Dialogue: 0,0:58:56.04,0:58:59.28,English,,0,0,0,,Okay now why on earth would why on earth would anybody want to do this
Dialogue: 0,0:58:56.04,0:58:59.28,Chinese,,0,0,0,,为什么有人要这样做呢
Dialogue: 0,0:58:59.28,0:59:03.14,English,,0,0,0,,Actually this notion of copy-on-write is is another one of these
Dialogue: 0,0:58:59.28,0:59:03.14,Chinese,,0,0,0,,实际上，写时复制是一个
Dialogue: 0,0:59:04.38,0:59:07.98,English,,0,0,0,,Kind of fundamental important system concepts that's used a lot
Dialogue: 0,0:59:04.38,0:59:07.98,Chinese,,0,0,0,,基本的系统概念，它在高效共享方面
Dialogue: 0,0:59:09.96,0:59:13.22,English,,0,0,0,,Whenever you want to share things efficiently yes
Dialogue: 0,0:59:09.96,0:59:13.22,Chinese,,0,0,0,,使用非常广泛，请说
Dialogue: 0,0:59:13.82,0:59:28.62,English,,0,0,0,,[student speaking]
Dialogue: 0,0:59:13.82,0:59:28.62,Chinese,,0,0,0,,[学生提问]
Dialogue: 0,0:59:28.70,0:59:29.52,English,,0,0,0,,If you
Dialogue: 0,0:59:28.70,0:59:29.52,Chinese,,0,0,0,,如果，你
Dialogue: 0,0:59:30.56,0:59:33.46,English,,0,0,0,,Well although the writes all go through
Dialogue: 0,0:59:30.56,0:59:33.46,Chinese,,0,0,0,,尽管所有的写操作都经过内存系统
Dialogue: 0,0:59:38.80,0:59:43.42,English,,0,0,0,,Yeah the memory system okay this is a really really good question
Dialogue: 0,0:59:38.80,0:59:43.42,Chinese,,0,0,0,,这是一个好问题
Dialogue: 0,0:59:44.16,0:59:47.52,English,,0,0,0,,And it's beyond our scope but the memory system takes care of those
Dialogue: 0,0:59:44.16,0:59:47.52,Chinese,,0,0,0,,但是超过了我们的讨论范围，操作系统负责维护写操作的顺序
Dialogue: 0,0:59:48.22,0:59:50.12,English,,0,0,0,,And it provides some guarantees
Dialogue: 0,0:59:48.22,0:59:50.12,Chinese,,0,0,0,,它提供了一些保证
Dialogue: 0,0:59:50.60,0:59:55.94,English,,0,0,0,,So it the memory system you're right it's receiving from multiple cores
Dialogue: 0,0:59:50.60,0:59:55.94,Chinese,,0,0,0,,你是对的，内存系统接收来自多核的
Dialogue: 0,0:59:55.94,1:00:01.42,English,,0,0,0,,It's receiving multiple instructions,multiple addresses
Dialogue: 0,0:59:55.94,1:00:01.42,Chinese,,0,0,0,,指令和地址
Dialogue: 0,1:00:01.92,1:00:06.92,English,,0,0,0,,And the memory system itself takes care of sort of providing some ordering on those
Dialogue: 0,1:00:01.92,1:00:06.92,Chinese,,0,0,0,,内存系统负责维护它们的顺序
Dialogue: 0,1:00:07.52,1:00:13.12,English,,0,0,0,,And it provides some sort of non-trivial guarantees on what you can assume the ordering is but
Dialogue: 0,1:00:07.52,1:00:13.12,Chinese,,0,0,0,,它提供了一些保护措施，所以你可以假设它们是有序的
Dialogue: 0,1:00:13.68,1:00:20.24,English,,0,0,0,,That's a pretty a pretty complicated topic that we'll get into
Dialogue: 0,1:00:13.68,1:00:20.24,Chinese,,0,0,0,,这是一个复杂的问题，
Dialogue: 0,1:00:23.08,1:00:25.82,English,,0,0,0,,It's called the consistency, it's called the consistency model
Dialogue: 0,1:00:23.08,1:00:25.82,Chinese,,0,0,0,,叫做连续性模型
Dialogue: 0,1:00:25.82,1:00:29.60,English,,0,0,0,,Right so every every processor provides some kind of consistency model
Dialogue: 0,1:00:25.82,1:00:29.60,Chinese,,0,0,0,,每一种处理器都提供了自己的连续性模型
Dialogue: 0,1:00:32.40,1:00:33.04,English,,0,0,0,,Any other questions
Dialogue: 0,1:00:32.40,1:00:33.04,Chinese,,0,0,0,,还有其他问题吗
Dialogue: 0,1:00:34.72,1:00:41.82,English,,0,0,0,,Okay so why is this copy-on-write technique useful and interesting
Dialogue: 0,1:00:34.72,1:00:41.82,Chinese,,0,0,0,,好的，为什么我们要有这个有用而有趣的写时复制技术呢
Dialogue: 0,1:00:42.88,1:00:46.02,English,,0,0,0,,Well it turns out I mean if you think about fork
Dialogue: 0,1:00:42.88,1:00:46.02,Chinese,,0,0,0,,嗯，你可以想一下 fork 函数
Dialogue: 0,1:00:47.32,1:00:49.76,English,,0,0,0,,It seems like it would be terribly expensive right you have
Dialogue: 0,1:00:47.32,1:00:49.76,Chinese,,0,0,0,,它看上去的开销是非常大的
Dialogue: 0,1:00:50.60,1:00:57.92,English,,0,0,0,,If you want to fork a process you know the naive way to do it would just be to copy
Dialogue: 0,1:00:50.60,1:00:57.92,Chinese,,0,0,0,,如果你想 fork 一个进程，最简单的方式就是拷贝
Dialogue: 0,1:00:57.92,1:01:01.14,English,,0,0,0,,You'd have you have to somehow get us exact copy of the address space
Dialogue: 0,1:00:57.92,1:01:01.14,Chinese,,0,0,0,,你需要拷贝完整的地址空间
Dialogue: 0,1:01:02.56,1:01:06.24,English,,0,0,0,,But a separate copy of the address space that's identical
Dialogue: 0,1:01:02.56,1:01:06.24,Chinese,,0,0,0,,一个独立但是一模一样的地址空间
Dialogue: 0,1:01:07.58,1:01:12.46,English,,0,0,0,,So if you approach this naively,you'd have to make copies of all the page tables
Dialogue: 0,1:01:07.58,1:01:12.46,Chinese,,0,0,0,,如果你使用这种简单的方式实现，你需要拷贝所有的页表
Dialogue: 0,1:01:13.14,1:01:18.84,English,,0,0,0,,All of the other data structures and you'd also have to copy all the memory
Dialogue: 0,1:01:13.14,1:01:18.84,Chinese,,0,0,0,,所有的用户数据结构和内存
Dialogue: 0,1:01:20.10,1:01:28.46,English,,0,0,0,,Right so if you're you know if you're forking a process that's created a huge number of virtual pages
Dialogue: 0,1:01:20.10,1:01:28.46,Chinese,,0,0,0,,对把，所以，如果你要 fork 的进程在它的地址空间内
Dialogue: 0,1:01:28.82,1:01:30.20,English,,0,0,0,,And it's address space
Dialogue: 0,1:01:28.82,1:01:30.20,Chinese,,0,0,0,,创建了很多虚拟页
Dialogue: 0,1:01:31.30,1:01:37.86,English,,0,0,0,,Those would all have to be copied to and mapped to different portions of the physical address of the memory
Dialogue: 0,1:01:31.30,1:01:37.86,Chinese,,0,0,0,,它们全部都需要拷贝，并且映射到不同的物理地址
Dialogue: 0,1:01:39.70,1:01:42.52,English,,0,0,0,,Okay so you'd have...it would be very expensive right
Dialogue: 0,1:01:39.70,1:01:42.52,Chinese,,0,0,0,,所以...开销会非常大
Dialogue: 0,1:01:42.52,1:01:45.86,English,,0,0,0,,Potentially almost unbounded depending on how much memory you are using
Dialogue: 0,1:01:42.52,1:01:45.86,Chinese,,0,0,0,,开销会随着你所使用的内存的大小的增长而增长，几乎是无限制的
Dialogue: 0,1:01:48.48,1:01:54.26,English,,0,0,0,,Right but fortunately this copy-on-write technique provides an efficient solution
Dialogue: 0,1:01:48.48,1:01:54.26,Chinese,,0,0,0,,但是幸运的是，写时复制技术提供了一个高效的解决方案
Dialogue: 0,1:01:54.88,1:01:59.32,English,,0,0,0,,And so the idea when we do fork
Dialogue: 0,1:01:54.88,1:01:59.32,Chinese,,0,0,0,,它的思想是，当我们执行 fork 的时候
Dialogue: 0,1:02:01.84,1:02:04.84,English,,0,0,0,,When we...when a process executes fork
Dialogue: 0,1:02:01.84,1:02:04.84,Chinese,,0,0,0,,我们...当我们执行 fork
Dialogue: 0,1:02:06.22,1:02:10.84,English,,0,0,0,,The kernel has to create exact copies of all those internal data structures
Dialogue: 0,1:02:06.22,1:02:10.84,Chinese,,0,0,0,,内核只拷贝所有的内核数据结构
Dialogue: 0,1:02:11.08,1:02:15.04,English,,0,0,0,,They mm_struck the area_struct and the page tables
Dialogue: 0,1:02:11.08,1:02:15.04,Chinese,,0,0,0,,mm_struct 和 area_struct 以及页表
Dialogue: 0,1:02:15.04,1:02:16.50,English,,0,0,0,,There's no way to get around that
Dialogue: 0,1:02:15.04,1:02:16.50,Chinese,,0,0,0,,这个操作是没有办法避免的
Dialogue: 0,1:02:17.58,1:02:21.14,English,,0,0,0,,Okay but those aren't those aren't really huge of
Dialogue: 0,1:02:17.58,1:02:21.14,Chinese,,0,0,0,,但是它们的大小并不大
Dialogue: 0,1:02:21.84,1:02:27.04,English,,0,0,0,,They're not as potentially huge as the the actual data that the program is is accessing
Dialogue: 0,1:02:21.84,1:02:27.04,Chinese,,0,0,0,,它们没有程序正在访问的实际数据那么大
Dialogue: 0,1:02:28.54,1:02:33.44,English,,0,0,0,,And then it flags each page in both processes as read-only
Dialogue: 0,1:02:28.54,1:02:33.44,Chinese,,0,0,0,,然后内核把两个进程中的每个页面都标记为只读
Dialogue: 0,1:02:35.94,1:02:38.50,English,,0,0,0,,Okay even though they're not no matter where they are any process
Dialogue: 0,1:02:35.94,1:02:38.50,Chinese,,0,0,0,,不管页面是哪一个进程
Dialogue: 0,1:02:39.30,1:02:41.82,English,,0,0,0,,Any page in that process it flags is read only
Dialogue: 0,1:02:39.30,1:02:41.82,Chinese,,0,0,0,,所有在进程中的页面都被标记为只读
Dialogue: 0,1:02:42.98,1:02:47.54,English,,0,0,0,,And then it flags each VM area_struct in both processes as private copy-on-write
Dialogue: 0,1:02:42.98,1:02:47.54,Chinese,,0,0,0,,然后把每一个 area_struct 都标记为私有写时复制
Dialogue: 0,1:02:50.34,1:02:55.46,English,,0,0,0,,So on return now each process has the identical address space
Dialogue: 0,1:02:50.34,1:02:55.46,Chinese,,0,0,0,,然后 fork 返回，在返回时，每个进程都有一样的地址空间
Dialogue: 0,1:02:55.50,1:02:58.40,English,,0,0,0,,Because it they have it has identical page tables
Dialogue: 0,1:02:55.50,1:02:58.40,Chinese,,0,0,0,,因为它们拥有相同的页表
Dialogue: 0,1:02:59.40,1:03:01.49,English,,0,0,0,,Okay we've made copies of the page tables
Dialogue: 0,1:02:59.40,1:03:01.49,Chinese,,0,0,0,,因为我们拷贝了这些页表
Dialogue: 0,1:03:02.08,1:03:04.54,English,,0,0,0,,And all of the other...all of the other structures
Dialogue: 0,1:03:02.08,1:03:04.54,Chinese,,0,0,0,,和其他所有的数据结构
Dialogue: 0,1:03:05.98,1:03:08.74,English,,0,0,0,,And if as those processes just read
Dialogue: 0,1:03:05.98,1:03:08.74,Chinese,,0,0,0,,如果两个进程仅仅是读的话
Dialogue: 0,1:03:10.18,1:03:12.36,English,,0,0,0,,Then they're sharing the same physical pages
Dialogue: 0,1:03:10.18,1:03:12.36,Chinese,,0,0,0,,它们共享相同的物理页
Dialogue: 0,1:03:13.66,1:03:16.82,English,,0,0,0,,Right it's only when a process does it write
Dialogue: 0,1:03:13.66,1:03:16.82,Chinese,,0,0,0,,当一个进程想要写的时候
Dialogue: 0,1:03:18.00,1:03:24.98,English,,0,0,0,,That the system will create a new page will create a new page using this copy-on-write mechanism
Dialogue: 0,1:03:18.00,1:03:24.98,Chinese,,0,0,0,,系统会使用写时复制的机制，创建一个新页面
Dialogue: 0,1:03:25.36,1:03:28.86,English,,0,0,0,,So it's only when a process write to a page will
Dialogue: 0,1:03:25.36,1:03:28.86,Chinese,,0,0,0,,因为当一个进程去写一个页面
Dialogue: 0,1:03:30.44,1:03:36.52,English,,0,0,0,,It tries to do a write the page is flagged as read-only okay in the PTE
Dialogue: 0,1:03:30.44,1:03:36.52,Chinese,,0,0,0,,而这个页面的 PTE 被标记为只读
Dialogue: 0,1:03:37.58,1:03:38.80,English,,0,0,0,,That triggers a fault
Dialogue: 0,1:03:37.58,1:03:38.80,Chinese,,0,0,0,,这会触发一个异常
Dialogue: 0,1:03:40.94,1:03:46.44,English,,0,0,0,,The kernel looks at that looks up the access that looks up the flags for that
Dialogue: 0,1:03:40.94,1:03:46.44,Chinese,,0,0,0,,内核查找访问地址的标志
Dialogue: 0,1:03:46.64,1:03:49.40,English,,0,0,0,,For that particular page sees that it's private copy-on-write
Dialogue: 0,1:03:46.64,1:03:49.40,Chinese,,0,0,0,,发现那个页面被标记为私有写时复制
Dialogue: 0,1:03:50.36,1:03:53.86,English,,0,0,0,,And so it makes a copy of the target page
Dialogue: 0,1:03:50.36,1:03:53.86,Chinese,,0,0,0,,所以它拷贝了那个页面
Dialogue: 0,1:03:54.96,1:03:58.96,English,,0,0,0,,And maps it to a new region of the physical address space
Dialogue: 0,1:03:54.96,1:03:58.96,Chinese,,0,0,0,,并把它映射到一个物理地址空间的新区域
Dialogue: 0,1:04:00.24,1:04:05.20,English,,0,0,0,,And then does the allows that...and then when the write executes again
Dialogue: 0,1:04:00.24,1:04:05.20,Chinese,,0,0,0,,它允许...当写再次执行的时候
Dialogue: 0,1:04:05.94,1:04:10.40,English,,0,0,0,,And then it restarts when that fault handler returns it restarts that instruction
Dialogue: 0,1:04:05.94,1:04:10.40,Chinese,,0,0,0,,也就是当异常处理返回的时候，重新执行当前指令
Dialogue: 0,1:04:10.84,1:04:13.44,English,,0,0,0,,And now the write is writing to the copy
Dialogue: 0,1:04:10.84,1:04:13.44,Chinese,,0,0,0,,写执行的对象就是拷贝的页面了
Dialogue: 0,1:04:14.70,1:04:19.64,English,,0,0,0,,Okay so what this does a very clever technique it defers the
Dialogue: 0,1:04:14.70,1:04:19.64,Chinese,,0,0,0,,所以，这是一个优秀的技巧，
Dialogue: 0,1:04:20.12,1:04:23.10,English,,0,0,0,,The copying till as long as till as late as possible
Dialogue: 0,1:04:20.12,1:04:23.10,Chinese,,0,0,0,,它尽可能的延迟了拷贝的发生
Dialogue: 0,1:04:23.48,1:04:26.60,English,,0,0,0,,And it only does copying when it's absolutely needed
Dialogue: 0,1:04:23.48,1:04:26.60,Chinese,,0,0,0,,当不得不需要拷贝的时候，拷贝才会执行
Dialogue: 0,1:04:27.54,1:04:32.02,English,,0,0,0,,Okay so in some sense it's the most it provides you with the most efficient way to
Dialogue: 0,1:04:27.54,1:04:32.02,Chinese,,0,0,0,,从某种意义上说，这是最有效率的办法
Dialogue: 0,1:04:33.38,1:04:37.34,English,,0,0,0,,A most space efficient way to represent those two
Dialogue: 0,1:04:33.38,1:04:37.34,Chinese,,0,0,0,,空间最优的方式来表示
Dialogue: 0,1:04:38.20,1:04:39.96,English,,0,0,0,,Those two different virtual address spaces
Dialogue: 0,1:04:38.20,1:04:39.96,Chinese,,0,0,0,,这两个不同的虚拟地址空间
Dialogue: 0,1:04:40.70,1:04:47.56,English,,0,0,0,,And so what's interesting is that regions of the virtual address space that read from never get copied
Dialogue: 0,1:04:40.70,1:04:47.56,Chinese,,0,0,0,,所以，有趣的是，对虚拟地址空间只进行读的区域永远不会被复制
Dialogue: 0,1:04:48.08,1:04:51.86,English,,0,0,0,,So they it's perfectly fine for those two processes to share
Dialogue: 0,1:04:48.08,1:04:51.86,Chinese,,0,0,0,,这是完美的，因为两个进程分享
Dialogue: 0,1:04:53.86,1:04:57.58,English,,0,0,0,,To share that data on physical memory because it's never written too right so
Dialogue: 0,1:04:53.86,1:04:57.58,Chinese,,0,0,0,,没有被修改的物理内存
Dialogue: 0,1:05:00.56,1:05:03.48,English,,0,0,0,,Okay so that's why fork isn't so bad right
Dialogue: 0,1:05:00.56,1:05:03.48,Chinese,,0,0,0,,这是为什么 fork 的效率并不低的原因
Dialogue: 0,1:05:07.04,1:05:11.88,English,,0,0,0,,Now execve now that we know about sharing
Dialogue: 0,1:05:07.04,1:05:11.88,Chinese,,0,0,0,,接下来说 execve 系统调用，我们知道了
Dialogue: 0,1:05:12.00,1:05:16.24,English,,0,0,0,,And now that we know about this mapping notion
Dialogue: 0,1:05:12.00,1:05:16.24,Chinese,,0,0,0,,内存映射的概念
Dialogue: 0,1:05:17.42,1:05:24.06,English,,0,0,0,,The execve function loads and runs a new program in the current process
Dialogue: 0,1:05:17.42,1:05:24.06,Chinese,,0,0,0,,execve 函数在当前进程中加载并执行一个新函数
Dialogue: 0,1:05:25.94,1:05:31.30,English,,0,0,0,,And so what it does is it frees all the area_struct for the current process right
Dialogue: 0,1:05:25.94,1:05:31.30,Chinese,,0,0,0,,它删除当前进程的所有 area_struct 结构
Dialogue: 0,1:05:31.30,1:05:33.42,English,,0,0,0,,So execve not creating a new process
Dialogue: 0,1:05:31.30,1:05:33.42,Chinese,,0,0,0,,所以，execve 并不创建进程
Dialogue: 0,1:05:34.02,1:05:40.52,English,,0,0,0,,It's running a new program in a new virtual address space within the current process
Dialogue: 0,1:05:34.02,1:05:40.52,Chinese,,0,0,0,,它只是在当前进程中，在一个新的虚拟地址空间上运行一个程序
Dialogue: 0,1:05:41.82,1:05:48.66,English,,0,0,0,,So it frees all the area_struct and page tables for the current process
Dialogue: 0,1:05:41.82,1:05:48.66,Chinese,,0,0,0,,所以，它删除了当前进程的所有 area_struct 和页表
Dialogue: 0,1:05:50.86,1:05:54.56,English,,0,0,0,,Ok and then it creates new area_structs and page tables for the new areas
Dialogue: 0,1:05:50.86,1:05:54.56,Chinese,,0,0,0,,然后为新区域创建了新的 area_struct 和页表
Dialogue: 0,1:05:56.18,1:06:02.16,English,,0,0,0,,And the programs so the program and initialize data
Dialogue: 0,1:05:56.18,1:06:02.16,Chinese,,0,0,0,,然后程序初始化数据
Dialogue: 0,1:06:03.02,1:06:09.04,English,,0,0,0,,Those areas are backed by the file in this case the executable binary
Dialogue: 0,1:06:03.02,1:06:09.04,Chinese,,0,0,0,,这些区域被映射为一个文件，在这个例子中，是一个可执行文件
Dialogue: 0,1:06:09.96,1:06:16.92,English,,0,0,0,,And their program is backed by the portion of the executable that contains code and data segment
Dialogue: 0,1:06:09.96,1:06:16.92,Chinese,,0,0,0,,程序的代码被映射为可执行文件的 .text 段，数据段
Dialogue: 0,1:06:17.22,1:06:22.92,English,,0,0,0,,Is backed by the portion of the executable file that contains initialize data
Dialogue: 0,1:06:17.22,1:06:22.92,Chinese,,0,0,0,,被映射为可执行文件的 .data 段
Dialogue: 0,1:06:24.02,1:06:25.98,English,,0,0,0,,So these two together are private
Dialogue: 0,1:06:24.02,1:06:25.98,Chinese,,0,0,0,,它们二者都是私有的
Dialogue: 0,1:06:26.86,1:06:28.14,English,,0,0,0,,What we call their private
Dialogue: 0,1:06:26.86,1:06:28.14,Chinese,,0,0,0,,它们是私有的
Dialogue: 0,1:06:29.08,1:06:34.68,English,,0,0,0,,Ok these this object it shouldn't being shared with any anything else
Dialogue: 0,1:06:29.08,1:06:34.68,Chinese,,0,0,0,,因为没有任何进程和它共享这两个部分
Dialogue: 0,1:06:35.28,1:06:40.40,English,,0,0,0,,And it's file backed because these this area is backed by portions of a file
Dialogue: 0,1:06:35.28,1:06:40.40,Chinese,,0,0,0,,并且，这部分区域被映射为文件的某个部分
Dialogue: 0,1:06:41.36,1:06:48.10,English,,0,0,0,,The uninitialized data which was specified in the .bss section of the of the binary
Dialogue: 0,1:06:41.36,1:06:48.10,Chinese,,0,0,0,,未初始化数据被文件中的 .bss 段定义
Dialogue: 0,1:06:49.32,1:06:52.58,English,,0,0,0,,That's defined as a private demand 0 area
Dialogue: 0,1:06:49.32,1:06:52.58,Chinese,,0,0,0,,被定义为私有的，并且是请求二进制零的
Dialogue: 0,1:06:55.04,1:07:02.76,English,,0,0,0,,Ok so these pages will all be remembered by definition BSS is uninitialized
Dialogue: 0,1:06:55.04,1:07:02.76,Chinese,,0,0,0,,因为 .bss 段是未初始化的
Dialogue: 0,1:07:03.82,1:07:06.34,English,,0,0,0,,Okay and so the system initializes these to zero
Dialogue: 0,1:07:03.82,1:07:06.34,Chinese,,0,0,0,,所以系统把它们全部初始化为 0
Dialogue: 0,1:07:09.68,1:07:15.32,English,,0,0,0,,So any pages then any pages in the heap are also private demands zero
Dialogue: 0,1:07:09.68,1:07:15.32,Chinese,,0,0,0,,堆区的页面页都是私有的和请求二进制 0 的
Dialogue: 0,1:07:17.52,1:07:19.70,English,,0,0,0,,Now the memory mapped region for shared libraries
Dialogue: 0,1:07:17.52,1:07:19.70,Chinese,,0,0,0,,然后把共享库区域映射到实际的物理内存
Dialogue: 0,1:07:19.70,1:07:24.84,English,,0,0,0,,Remember every process share is the same copy of libc and in memory
Dialogue: 0,1:07:19.70,1:07:24.84,Chinese,,0,0,0,,所有的进程都共享内存中的 libc 库
Dialogue: 0,1:07:26.26,1:07:30.70,English,,0,0,0,,So this is a region of the virtual address space that's shared
Dialogue: 0,1:07:26.26,1:07:30.70,Chinese,,0,0,0,,所以，这一部分虚拟地址空间被标记为共享的
Dialogue: 0,1:07:32.12,1:07:34.26,English,,0,0,0,,Okay because it's shared by multiple processes
Dialogue: 0,1:07:32.12,1:07:34.26,Chinese,,0,0,0,,因为它被多个进程共享
Dialogue: 0,1:07:34.82,1:07:38.68,English,,0,0,0,,And it's backed by files of
Dialogue: 0,1:07:34.82,1:07:38.68,Chinese,,0,0,0,,它被映射为一些文件
Dialogue: 0,1:07:39.52,1:07:43.30,English,,0,0,0,,And the de file in this case is the .so file the shared object file
Dialogue: 0,1:07:39.52,1:07:43.30,Chinese,,0,0,0,,在这个例子中，是一些 .so 文件，也就是共享目标文件
Dialogue: 0,1:07:44.04,1:07:46.62,English,,0,0,0,,And the code for this region is backed by text
Dialogue: 0,1:07:44.04,1:07:46.62,Chinese,,0,0,0,,这个区域的代码被映射为
Dialogue: 0,1:07:47.80,1:07:51.60,English,,0,0,0,,The portion of this object file that contains code and
Dialogue: 0,1:07:47.80,1:07:51.60,Chinese,,0,0,0,,目标文件中包含代码的 .text 段
Dialogue: 0,1:07:52.04,1:07:58.66,English,,0,0,0,,Data is initialized data is backed by a portion of this file that contains data
Dialogue: 0,1:07:52.04,1:07:58.66,Chinese,,0,0,0,,数据被映射为文件中包含数据的 .data 段
Dialogue: 0,1:07:59.38,1:08:04.82,English,,0,0,0,,Now there's...I haven't shown it here but there's other there's other portions of this
Dialogue: 0,1:07:59.38,1:08:04.82,Chinese,,0,0,0,,嗯...这里，我没有把共享库区域中的
Dialogue: 0,1:08:05.52,1:08:11.14,English,,0,0,0,,Of this region that need to be a private copy-on-write because different libc functions
Dialogue: 0,1:08:05.52,1:08:11.14,Chinese,,0,0,0,,私有写时复制的区域表示出来，因为 libc 有许多函数
Dialogue: 0,1:08:11.52,1:08:17.33,English,,0,0,0,,If a libc function has static variables can contain state
Dialogue: 0,1:08:11.52,1:08:17.33,Chinese,,0,0,0,,如果 libc 函数有维护状态的静态变量
Dialogue: 0,1:08:18.14,1:08:21.40,English,,0,0,0,,Like a random number generator typical random number generator
Dialogue: 0,1:08:18.14,1:08:21.40,Chinese,,0,0,0,,如一个随机数产生器，一个典型的随机数产生器
Dialogue: 0,1:08:23.38,1:08:25.54,English,,0,0,0,,Retains state across each invocation
Dialogue: 0,1:08:23.38,1:08:25.54,Chinese,,0,0,0,,在每次调用中保持状态
Dialogue: 0,1:08:26.36,1:08:29.32,English,,0,0,0,,Okay so that state would be different for each process
Dialogue: 0,1:08:26.36,1:08:29.32,Chinese,,0,0,0,,但是，不同调用此函数的状态是不同的
Dialogue: 0,1:08:29.34,1:08:36.60,English,,0,0,0,,So there needs to be a sort of private copy-on-write area for any data that's being written to
Dialogue: 0,1:08:29.34,1:08:36.60,Chinese,,0,0,0,,所以，对于要修改静态数据的函数，应该被标记为私有写时复制的
Dialogue: 0,1:08:40.86,1:08:43.36,English,,0,0,0,,And I suppose the system would adjust
Dialogue: 0,1:08:40.86,1:08:43.36,Chinese,,0,0,0,,我想，系统会进行调整
Dialogue: 0,1:08:44.96,1:08:48.56,English,,0,0,0,,adjust,if you made this entire region
Dialogue: 0,1:08:44.96,1:08:48.56,Chinese,,0,0,0,,如果你把这个区域弄成一个整体的话
Dialogue: 0,1:08:50.12,1:08:54.74,English,,0,0,0,,Now you'd have to have it you'd have to have second portions of this that would be private copy-on-write
Dialogue: 0,1:08:50.12,1:08:54.74,Chinese,,0,0,0,,现在你不得不有第二个区域，私有写时复制的区域
Dialogue: 0,1:08:55.66,1:08:57.82,English,,0,0,0,,And then the stack is private demand zero
Dialogue: 0,1:08:55.66,1:08:57.82,Chinese,,0,0,0,,栈是私有写时复制的
Dialogue: 0,1:08:59.66,1:09:03.28,English,,0,0,0,,Okay so what what execve does is it just sets
Dialogue: 0,1:08:59.66,1:09:03.28,Chinese,,0,0,0,,好了，所以，execve 做的仅仅是
Dialogue: 0,1:09:04.72,1:09:11.32,English,,0,0,0,,It creates new areas that are backed backed by a different by the object file that you want to execute
Dialogue: 0,1:09:04.72,1:09:11.32,Chinese,,0,0,0,,创建一个新区域，并把它映射到你想执行的对象文件中
Dialogue: 0,1:09:13.66,1:09:18.14,English,,0,0,0,,And it creates areas for BSS and stack that are backed by anonymous files
Dialogue: 0,1:09:13.66,1:09:18.14,Chinese,,0,0,0,,它创建 .bss 和栈，然后把它们映射到一个匿名文件
Dialogue: 0,1:09:18.14,1:09:25.82,English,,0,0,0,,And it creates a this memory-mapped region that's a shared object  which corresponds to libc
Dialogue: 0,1:09:18.14,1:09:25.82,Chinese,,0,0,0,,它创建一个内存映射区域，映射到 libc
Dialogue: 0,1:09:26.68,1:09:31.66,English,,0,0,0,,And then it sets the program counter %rip to the entry point in text
Dialogue: 0,1:09:26.68,1:09:31.66,Chinese,,0,0,0,,然后把程序计数器 %rip 设置为代码区域的入口点
Dialogue: 0,1:09:33.66,1:09:37.90,English,,0,0,0,,And then once as this program runs now notice we haven't loaded anything
Dialogue: 0,1:09:33.66,1:09:37.90,Chinese,,0,0,0,,然后当程序运行时，注意我们没有加载任何内容
Dialogue: 0,1:09:38.14,1:09:40.76,English,,0,0,0,,Right all we've done is set mappings
Dialogue: 0,1:09:38.14,1:09:40.76,Chinese,,0,0,0,,我们所做的，仅仅是设置内存映射
Dialogue: 0,1:09:42.40,1:09:44.70,English,,0,0,0,,We just created data structures in the kernel
Dialogue: 0,1:09:42.40,1:09:44.70,Chinese,,0,0,0,,在内核中创建数据结构
Dialogue: 0,1:09:45.48,1:09:50.18,English,,0,0,0,,And we've created mappings between portions of the address space and these objects
Dialogue: 0,1:09:45.48,1:09:50.18,Chinese,,0,0,0,,我们创建了地址空间和这些对象之间的映射
Dialogue: 0,1:09:50.98,1:09:53.80,English,,0,0,0,,But nothing is actually been copied into memory yet
Dialogue: 0,1:09:50.98,1:09:53.80,Chinese,,0,0,0,,但是现在还没有任何内容拷贝到内存
Dialogue: 0,1:09:55.02,1:10:00.78,English,,0,0,0,,Okay well this is all we've just created we've modified data structures in the kernel
Dialogue: 0,1:09:55.02,1:10:00.78,Chinese,,0,0,0,,我们仅仅做了映射，修改了内核中的数据结构
Dialogue: 0,1:10:02.24,1:10:08.92,English,,0,0,0,,Now but once the once the loader sets %rip to the entry point
Dialogue: 0,1:10:02.24,1:10:08.92,Chinese,,0,0,0,,但是当加载器把 %rip 设置为代码区域的入口点
Dialogue: 0,1:10:09.20,1:10:13.56,English,,0,0,0,,The first instruction in this text segment
Dialogue: 0,1:10:09.20,1:10:13.56,Chinese,,0,0,0,,当执行代码段的第一条指令的时候
Dialogue: 0,1:10:14.76,1:10:18.52,English,,0,0,0,,Then Linux will fault in all the code and data that's needed on demand
Dialogue: 0,1:10:14.76,1:10:18.52,Chinese,,0,0,0,,Linux 遇到的代码和数据没有时，会陷入异常
Dialogue: 0,1:10:19.78,1:10:27.56,English,,0,0,0,,So loading loading has is deferred until the loading of a page of code or data is deferred
Dialogue: 0,1:10:19.78,1:10:27.56,Chinese,,0,0,0,,所以加载不得不推迟，直到加载一个页面的代码或数据
Dialogue: 0,1:10:28.26,1:10:32.10,English,,0,0,0,,Until that code or data page is actually referenced and accessed
Dialogue: 0,1:10:28.26,1:10:32.10,Chinese,,0,0,0,,加载推迟到页面实际被引用或访问
Dialogue: 0,1:10:33.58,1:10:36.86,English,,0,0,0,,So it's very this is very clever and very interesting I think how
Dialogue: 0,1:10:33.58,1:10:36.86,Chinese,,0,0,0,,这个策略非常聪明且有趣
Dialogue: 0,1:10:37.74,1:10:42.16,English,,0,0,0,,How it's like yet another example of how useful and
Dialogue: 0,1:10:37.74,1:10:42.16,Chinese,,0,0,0,,我想，这是另一个例子，说明
Dialogue: 0,1:10:42.60,1:10:46.56,English,,0,0,0,,How tightly integrated virtual memory is with the operation of the system
Dialogue: 0,1:10:42.60,1:10:46.56,Chinese,,0,0,0,,虚拟内存与操作系统紧密结合
Dialogue: 0,1:10:47.12,1:10:49.16,English,,0,0,0,,Right you wouldn't have to do this
Dialogue: 0,1:10:47.12,1:10:49.16,Chinese,,0,0,0,,你不需要做加载
Dialogue: 0,1:10:49.66,1:10:55.16,English,,0,0,0,,You could just as part of the loading process you could just copy those take those object files
Dialogue: 0,1:10:49.66,1:10:55.16,Chinese,,0,0,0,,你也可以做加载的一部分，你可以把这些对象文件拷贝到内存
Dialogue: 0,1:10:56.56,1:11:02.06,English,,0,0,0,,You could just have a loop that would read these object files and just copy them into copy them into memory
Dialogue: 0,1:10:56.56,1:11:02.06,Chinese,,0,0,0,,你可以使用一个循环把这些对象文件拷贝到内存
Dialogue: 0,1:11:02.52,1:11:05.06,English,,0,0,0,,And create page table entries for of as as you did it
Dialogue: 0,1:11:02.52,1:11:05.06,Chinese,,0,0,0,,为它创建一个页表条目
Dialogue: 0,1:11:06.84,1:11:09.76,English,,0,0,0,,But even that would be efficient that would be inefficient too right
Dialogue: 0,1:11:06.84,1:11:09.76,Chinese,,0,0,0,,但这样做可能是高效的，也有可能是低效的
Dialogue: 0,1:11:09.76,1:11:15.54,English,,0,0,0,,What if you had a very large array that you'd initialized to some nonzero value
Dialogue: 0,1:11:09.76,1:11:15.54,Chinese,,0,0,0,,如果你有一个长的数组，并且你将他们初始化为一些非0值
Dialogue: 0,1:11:16.46,1:11:19.76,English,,0,0,0,,That would require you to copy that entire array in at load time
Dialogue: 0,1:11:16.46,1:11:19.76,Chinese,,0,0,0,,这就要求你在加载的时候拷贝整个数组到内存
Dialogue: 0,1:11:19.76,1:11:23.86,English,,0,0,0,,Even though you may be only accessing a small portion of that of that data structure
Dialogue: 0,1:11:19.76,1:11:23.86,Chinese,,0,0,0,,尽管你可能只访问这个数据结构中的一部分
Dialogue: 0,1:11:26.36,1:11:36.80,English,,0,0,0,,Now the kernel provides a function called mmap that allows you to do your memory mapping just like the kernel does
Dialogue: 0,1:11:26.36,1:11:36.80,Chinese,,0,0,0,,内核提供了一个叫做 mmap 的函数，它允许你像内核一样进行内存映射
Dialogue: 0,1:11:38.24,1:11:41.42,English,,0,0,0,,Okay and so it's basically a way it's a system called it that
Dialogue: 0,1:11:38.24,1:11:41.42,Chinese,,0,0,0,,它是一个基本的系统调用
Dialogue: 0,1:11:42.06,1:11:49.12,English,,0,0,0,,Allows you to request the kernel to map a region of virtual memory on your on its behalf
Dialogue: 0,1:11:42.06,1:11:49.12,Chinese,,0,0,0,,允许你像内核一样进行内存映射
Dialogue: 0,1:11:51.50,1:11:58.19,English,,0,0,0,,And so this mmap function takes a virtual pointer
Dialogue: 0,1:11:51.50,1:11:58.19,Chinese,,0,0,0,,mmap 有一个指针的参数
Dialogue: 0,1:11:59.04,1:12:01.66,English,,0,0,0,,Which is some pointer into the virtual address space
Dialogue: 0,1:11:59.04,1:12:01.66,Chinese,,0,0,0,,是一个指向虚拟地址空间的指针
Dialogue: 0,1:12:03.66,1:12:09.68,English,,0,0,0,,And it maps tries to map length bytes starting at this address
Dialogue: 0,1:12:03.66,1:12:09.68,Chinese,,0,0,0,,mmap 函数尝试将这个地址开始，长度为 length 字节的区域，映射到
Dialogue: 0,1:12:12.38,1:12:18.26,English,,0,0,0,,It tries to map that portion of the virtual address space to the object  to an offset
Dialogue: 0,1:12:12.38,1:12:18.26,Chinese,,0,0,0,,由这个文件描述符确定的
Dialogue: 0,1:12:20.30,1:12:23.26,English,,0,0,0,,To some objects specified by this file descriptor
Dialogue: 0,1:12:20.30,1:12:23.26,Chinese,,0,0,0,,某个对象的 offset 偏移位置
Dialogue: 0,1:12:25.12,1:12:27.86,English,,0,0,0,,Okay so remember we when you open a file you get back a file descriptor
Dialogue: 0,1:12:25.12,1:12:27.86,Chinese,,0,0,0,,当你打开一个文件，你可以得到一个文件描述符
Dialogue: 0,1:12:28.60,1:12:35.02,English,,0,0,0,,So it's mapping this region of the virtual address space starting at start for length bytes
Dialogue: 0,1:12:28.60,1:12:35.02,Chinese,,0,0,0,,mmap 函数把虚拟地址空间映射到
Dialogue: 0,1:12:36.70,1:12:42.00,English,,0,0,0,,To a region of the that the file denoted by this integer file descriptor
Dialogue: 0,1:12:36.70,1:12:42.00,Chinese,,0,0,0,,文件描述符确定的文件
Dialogue: 0,1:12:44.58,1:12:50.14,English,,0,0,0,,So it's mapping length bytes in that file starting at offset
Dialogue: 0,1:12:44.58,1:12:50.14,Chinese,,0,0,0,,mmap 把文件的 offset 开始的 length 个字节映射到虚拟地址空间
Dialogue: 0,1:12:51.38,1:12:58.64,English,,0,0,0,,Now and then there's...the user can specify different flags what kind of protection they want for this
Dialogue: 0,1:12:51.38,1:12:58.64,Chinese,,0,0,0,,嗯...用户可以指定不同的标志,指定这个页面的保护类型
Dialogue: 0,1:12:59.46,1:13:05.60,English,,0,0,0,,This whether its private or whether it's read or whether it can be it's read-only or read/write
Dialogue: 0,1:12:59.46,1:13:05.60,Chinese,,0,0,0,,是私有的，读的，只读的或是可读可写的
Dialogue: 0,1:13:06.32,1:13:11.82,English,,0,0,0,,And they can also specify by the type of of object
Dialogue: 0,1:13:06.32,1:13:11.82,Chinese,,0,0,0,,也可以指定文件对象的类型
Dialogue: 0,1:13:12.60,1:13:19.22,English,,0,0,0,,If you map and anonymous, if you use this flag and mmap then you get a demand 0 page
Dialogue: 0,1:13:12.60,1:13:19.22,Chinese,,0,0,0,,如果映射为匿名文件，你设置 flag，然后你可以得到一个请求二进制零页面
Dialogue: 0,1:13:19.98,1:13:21.18,English,,0,0,0,,Okay that's the anonymous file
Dialogue: 0,1:13:19.98,1:13:21.18,Chinese,,0,0,0,,这是映射为匿名文件
Dialogue: 0,1:13:22.88,1:13:26.16,English,,0,0,0,,In which case you don't need to specify this file descriptor
Dialogue: 0,1:13:22.88,1:13:26.16,Chinese,,0,0,0,,你不需要指定文件描述符参数
Dialogue: 0,1:13:27.46,1:13:29.98,English,,0,0,0,,And you can map that object as being private or shared
Dialogue: 0,1:13:27.46,1:13:29.98,Chinese,,0,0,0,,你可以指定该对象是私有的还是共享的
Dialogue: 0,1:13:31.34,1:13:32.70,English,,0,0,0,,Okay like we talked about before
Dialogue: 0,1:13:31.34,1:13:32.70,Chinese,,0,0,0,,就像我们前面说的那样
Dialogue: 0,1:13:34.38,1:13:41.96,English,,0,0,0,,Now the kernel will, it will return an pointer to the start of this mapped area
Dialogue: 0,1:13:34.38,1:13:41.96,Chinese,,0,0,0,,内核会返回一个指向映射区域的指针
Dialogue: 0,1:13:44.28,1:13:47.66,English,,0,0,0,,not necessarily,this start address is like a hint to the kernel
Dialogue: 0,1:13:44.28,1:13:47.66,Chinese,,0,0,0,,start 参数不是必须的，这个参数对于内核来说只是一个参考
Dialogue: 0,1:13:47.66,1:13:51.70,English,,0,0,0,,So if it can,it will try to map that area
Dialogue: 0,1:13:47.66,1:13:51.70,Chinese,,0,0,0,,如果该区域有效，它可以映射到这个虚拟地址空间区域
Dialogue: 0,1:13:51.70,1:13:57.02,English,,0,0,0,,But if that portion of the virtual address space is already contained in some already existing area
Dialogue: 0,1:13:51.70,1:13:57.02,Chinese,,0,0,0,,如果这个区域的部分虚拟地址空间已经被某个存在的区域包含了
Dialogue: 0,1:13:57.02,1:14:02.16,English,,0,0,0,,Then it will pick an unused portion of the virtual address space to to map
Dialogue: 0,1:13:57.02,1:14:02.16,Chinese,,0,0,0,,它会映射到虚拟地址空间中没有使用的部分
Dialogue: 0,1:14:04.22,1:14:07.86,English,,0,0,0,,Okay so what we're doing really is we're taking
Dialogue: 0,1:14:04.22,1:14:07.86,Chinese,,0,0,0,,我们要做的是将
Dialogue: 0,1:14:10.32,1:14:19.66,English,,0,0,0,,An offset into some file specified by that fd argument and length number of bytes
Dialogue: 0,1:14:10.32,1:14:19.66,Chinese,,0,0,0,,fd 参数指定的某个文件，offset 开始的长度 length 字节的区域
Dialogue: 0,1:14:21.82,1:14:27.66,English,,0,0,0,,And we're associating that or mapping that into us the same sized region of the virtual address space
Dialogue: 0,1:14:21.82,1:14:27.66,Chinese,,0,0,0,,映射到同样大小的虚拟地址空间中
Dialogue: 0,1:14:30.36,1:14:34.00,English,,0,0,0,,Okay so now if we do, if we read
Dialogue: 0,1:14:30.36,1:14:34.00,Chinese,,0,0,0,,好，如果我们读这个区域
Dialogue: 0,1:14:35.22,1:14:36.82,English,,0,0,0,,Now again nothing gets copied
Dialogue: 0,1:14:35.22,1:14:36.82,Chinese,,0,0,0,,这里我要强调的是，mmap 没有任何从文件拷贝到内存的动作
Dialogue: 0,1:14:38.70,1:14:39.82,English,,0,0,0,,Okay it's just a mapping
Dialogue: 0,1:14:38.70,1:14:39.82,Chinese,,0,0,0,,仅仅是内存映射
Dialogue: 0,1:14:40.88,1:14:48.64,English,,0,0,0,,So that if we were to read a portion of let's say we mapped an unused portion of the virtual address space
Dialogue: 0,1:14:40.88,1:14:48.64,Chinese,,0,0,0,,我们使用了 mmap 把未使用的虚拟地址空间的某个部分
Dialogue: 0,1:14:50.12,1:14:51.40,English,,0,0,0,,And then at using mmap
Dialogue: 0,1:14:50.12,1:14:51.40,Chinese,,0,0,0,,映射到一个文件
Dialogue: 0,1:14:51.46,1:14:56.26,English,,0,0,0,,And then after mmap we just started scanning through this portion of the virtual address space
Dialogue: 0,1:14:51.46,1:14:56.26,Chinese,,0,0,0,,mmap 之后，我们开始扫描这个虚拟地址空间的这个部分
Dialogue: 0,1:14:56.28,1:15:00.02,English,,0,0,0,,You know say it's an array, we've mmap a big array
Dialogue: 0,1:14:56.28,1:15:00.02,Chinese,,0,0,0,,它是一个数组，我们映射了一个很长的数组
Dialogue: 0,1:15:01.98,1:15:09.12,English,,0,0,0,,As we read each non-existing page will be swapped in automatically
Dialogue: 0,1:15:01.98,1:15:09.12,Chinese,,0,0,0,,当我们读到不存在的页面时，它会自动换入
Dialogue: 0,1:15:10.32,1:15:11.62,English,,0,0,0,,kernel will trigger page fault
Dialogue: 0,1:15:10.32,1:15:11.62,Chinese,,0,0,0,,内核会触发一个缺页异常
Dialogue: 0,1:15:12.72,1:15:21.42,English,,0,0,0,,And it will be swapped in and the value that it has will be determined by the the contents of this portion of the file
Dialogue: 0,1:15:12.72,1:15:21.42,Chinese,,0,0,0,,它会换入对应的页面，页面的初始值就是文件的某个部分
Dialogue: 0,1:15:26.48,1:15:32.50,English,,0,0,0,,Okay and then so one here's an example of how you can use mmap
Dialogue: 0,1:15:26.48,1:15:32.50,Chinese,,0,0,0,,这是一个关于使用 mmap 的例子
Dialogue: 0,1:15:33.80,1:15:38.92,English,,0,0,0,,To do and this is kind of an interesting example we're copying a file from standard into standard out
Dialogue: 0,1:15:33.80,1:15:38.92,Chinese,,0,0,0,,这是一个有趣的例子，我们从标准输入拷贝一个文件到标准输出
Dialogue: 0,1:15:39.48,1:15:43.74,English,,0,0,0,,Without ever transferring data to user space
Dialogue: 0,1:15:39.48,1:15:43.74,Chinese,,0,0,0,,并且没有把数据拷贝到用户空间
Dialogue: 0,1:15:43.74,1:15:48.82,English,,0,0,0,,So the normal sort of into naive way to do this would be to read from a file
Dialogue: 0,1:15:43.74,1:15:48.82,Chinese,,0,0,0,,要做到这一点，通常的方法是从一个文件中读取
Dialogue: 0,1:15:48.92,1:15:53.54,English,,0,0,0,,Read from standard in and then write to standard out so two system calls one read and one write
Dialogue: 0,1:15:48.92,1:15:53.54,Chinese,,0,0,0,,从标准输入中读，然后写到标准输出中去，需要用两个系统调用，read 和 write
Dialogue: 0,1:15:53.96,1:15:59.52,English,,0,0,0,,We can actually do this using 1 mmap and then a single right
Dialogue: 0,1:15:53.96,1:15:59.52,Chinese,,0,0,0,,但实际上，我们可以只使用一个系统调用 mmap 就可以做到这一点
Dialogue: 0,1:16:01.34,1:16:07.32,English,,0,0,0,,Okay so what we do is we open a file argv[1] is standard in
Dialogue: 0,1:16:01.34,1:16:07.32,Chinese,,0,0,0,,首先，我们标准输入中得到一个文件名，然后打开这个文件
Dialogue: 0,1:16:09.04,1:16:13.12,English,,0,0,0,,Or we I'm sorry we open a file that was passed in on the command line
Dialogue: 0,1:16:09.04,1:16:13.12,Chinese,,0,0,0,,抱歉，文件名的传递是通过命令行来传递
Dialogue: 0,1:16:13.64,1:16:19.88,English,,0,0,0,,We get its size and then we call this mmap copy function to say
Dialogue: 0,1:16:13.64,1:16:19.88,Chinese,,0,0,0,,然后我们得到文件的大小，再调用 mmap 函数
Dialogue: 0,1:16:20.54,1:16:25.92,English,,0,0,0,,Copy that the copy of the bytes the size bytes
Dialogue: 0,1:16:20.54,1:16:25.92,Chinese,,0,0,0,,拷贝文件的 size 字节
Dialogue: 0,1:16:26.84,1:16:29.72,English,,0,0,0,,From this file to standard out
Dialogue: 0,1:16:26.84,1:16:29.72,Chinese,,0,0,0,,到标准输出去
Dialogue: 0,1:16:30.42,1:16:35.96,English,,0,0,0,,So this mmap copy function does mmap using that file descriptor
Dialogue: 0,1:16:30.42,1:16:35.96,Chinese,,0,0,0,,mmapcopy 函数使用 mmap，传递一个文件描述符
Dialogue: 0,1:16:36.70,1:16:40.04,English,,0,0,0,,Sets that is read only private sets the size
Dialogue: 0,1:16:36.70,1:16:40.04,Chinese,,0,0,0,,设置标志位为私有的，设置大小
Dialogue: 0,1:16:43.70,1:16:46.18,English,,0,0,0,,And length means everything
Dialogue: 0,1:16:43.70,1:16:46.18,Chinese,,0,0,0,,长度和其他的参数
Dialogue: 0,1:16:48.32,1:16:50.34,English,,0,0,0,,And offset set zero
Dialogue: 0,1:16:48.32,1:16:50.34,Chinese,,0,0,0,,偏移设置为 0
Dialogue: 0,1:16:50.82,1:16:59.56,English,,0,0,0,,And then we do one writes this call to standard out passing at the pointer to the buffer that we want to copy
Dialogue: 0,1:16:50.82,1:16:59.56,Chinese,,0,0,0,,然后我们调用 write，把 buffer 指向的内容拷贝到标准输出
Dialogue: 0,1:17:00.96,1:17:02.56,English,,0,0,0,,And of size bytes
Dialogue: 0,1:17:00.96,1:17:02.56,Chinese,,0,0,0,,设置需要拷贝的大小
Dialogue: 0,1:17:03.30,1:17:03.80,English,,0,0,0,,So as this
Dialogue: 0,1:17:03.30,1:17:03.80,Chinese,,0,0,0,,所以
Dialogue: 0,1:17:05.10,1:17:05.54,English,,0,0,0,,As we
Dialogue: 0,1:17:05.10,1:17:05.54,Chinese,,0,0,0,,当我们
Dialogue: 0,1:17:06.20,1:17:06.86,English,,0,0,0,,so as we...
Dialogue: 0,1:17:06.20,1:17:06.86,Chinese,,0,0,0,,当我们
Dialogue: 0,1:17:07.92,1:17:09.06,English,,0,0,0,,So this will read
Dialogue: 0,1:17:07.92,1:17:09.06,Chinese,,0,0,0,,write 函数会读
Dialogue: 0,1:17:10.88,1:17:18.46,English,,0,0,0,,This will read bytes 1f the right call will read bytes in bufp one after the other and as that happens they'll be faulted in
Dialogue: 0,1:17:10.88,1:17:18.46,Chinese,,0,0,0,,write 函数会从 bufp 中一个字节一个字节地读取，执行的时候会出现异常
Dialogue: 0,1:17:19.48,1:17:21.92,English,,0,0,0,,And then it will write them
Dialogue: 0,1:17:19.48,1:17:21.92,Chinese,,0,0,0,,异常处理完后，write 函数可以
Dialogue: 0,1:17:22.96,1:17:29.02,English,,0,0,0,,I will write them to the descriptor indicated by the one which is standard out
Dialogue: 0,1:17:22.96,1:17:29.02,Chinese,,0,0,0,,把 bufp 写到文件描述符指示的文件中，这里是 1，也就是标准输出
Dialogue: 0,1:17:29.94,1:17:32.00,English,,0,0,0,,Okay so that's it for today
Dialogue: 0,1:17:29.94,1:17:32.00,Chinese,,0,0,0,,今天的课程结束了
Dialogue: 0,1:17:33.14,1:17:34.94,English,,0,0,0,,Hope you guys have a good weekend
Dialogue: 0,1:17:33.14,1:17:34.94,Chinese,,0,0,0,,祝你们有一个愉快的周末
Dialogue: 0,1:17:35.94,1:17:36.94,English,,0,0,0,,And see you next week
Dialogue: 0,1:17:35.94,1:17:36.94,Chinese,,0,0,0,,下周见