[CS61C FA20] Lecture 29.3 - Virtual Memory I: Memory Manager — Transcript

Full transcript

1. 0:00[Music]
2. 0:08hello
3. 0:09welcome back to our discussion on
4. 0:11operating system support in which
5. 0:12virtual memory
6. 0:14so we have seen how does the
7. 0:17real physical memory work and one thing
8. 0:19that you would like to take away from
9. 0:20there is that
10. 0:22that every uh location in that memory
11. 0:25every word or a block
12. 0:27has its own physical address that is the
13. 0:30address that corresponds to the library
14. 0:31or congress call
15. 0:32for the books or a street address
16. 0:36that the postal service uses to deliver
17. 0:38mail
18. 0:40let's see how does this memory and
19. 0:43addressing of the memory
20. 0:44work when there are multiple processes
21. 0:47that are trying to use it so
22. 0:51in the simplest case when we are running
23. 0:54so-called a bare metal system
24. 0:56where there is no other support
25. 0:59addresses
26. 1:00issued by loads and stores that
27. 1:03the program has on real physical
28. 1:05addresses
29. 1:06so you know this process that is running
30. 1:08on on a
31. 1:09processor is going to be addressing
32. 1:13real physical locations in the memory
33. 1:18now if we have multiple processes
34. 1:23then each of these processes can issue
35. 1:25any address
36. 1:26and can therefore address any part of
37. 1:29the memory
38. 1:30uh even those parts that it
39. 1:33does not necessarily own that you know
40. 1:35where another process believes
41. 1:38um that that where another process
42. 1:40believes that it has ownership up
43. 1:43so that's kind of a problem because
44. 1:44multiple processes can
45. 1:46run over each other in the memory
46. 1:50um even more dangerously um
47. 1:53some processes when there is an
48. 1:54operating system in this model
49. 1:57would be able to run over the operating
50. 1:59systems data structures
51. 2:02in order to avoid that we need to have a
52. 2:05translation mechanism
53. 2:07where there would be
54. 2:10each process would be
55. 2:13using a virtual outer space and then
56. 2:18those addresses would be somehow
57. 2:20decoupled in the physical
58. 2:22address space so before accessing a
59. 2:25particular physical
60. 2:26location in memory we would check if a
61. 2:29particular process
62. 2:30has access rights to that particular
63. 2:34spot
64. 2:35so when we look how things look with the
65. 2:38virtual memory
66. 2:40we are going to have hundreds of
67. 2:41processes that are managed by the
68. 2:43operating system
69. 2:45each line here corresponds to a
70. 2:48different process and each process lives
71. 2:51under this illusion that it
72. 2:52got at the time according to itself
73. 2:56and the entire memory to itself
74. 2:59and the next line here is another
75. 3:01process that also
76. 3:02lives under the illusion that it got the
77. 3:04process to itself
78. 3:06and all of the memory and the next one
79. 3:08and the next one
80. 3:10hundreds of processes can be there
81. 3:14and they're all multiplexed onto the
82. 3:16core
83. 3:17by the operating system um each one of
84. 3:20them runs for
85. 3:21some time on the core and then
86. 3:25through the mechanism of a contact
87. 3:26switch another process
88. 3:28takes ownership of the core but what are
89. 3:31we going to do with the memory there is
90. 3:33only one memory
91. 3:34and we cannot just save its contents
92. 3:37it's only because we have all of the
93. 3:39memory
94. 3:40and you know in the content switch run
95. 3:43you know put the data from another
96. 3:45process on it
97. 3:48so what we do we run
98. 3:51we we have to run this translation
99. 3:55process we have seen
100. 3:57a conceptual picture of that we are
101. 3:58going to get into a more
102. 4:00more details of its operation now so
103. 4:03each process
104. 4:07runs in its own world um it sees
105. 4:11this memory here the entire memory from
106. 4:13zero zero zero
107. 4:14zero zero to at the fff
108. 4:18where it can store the code the static
109. 4:20data the heap
110. 4:21and the stack
111. 4:25but then then
112. 4:28in some way those virtual addresses
113. 4:31get translated to physical addresses in
114. 4:33a different way for every single process
115. 4:37that concept that the translation is
116. 4:40owned by
117. 4:41something it is called the memory
118. 4:42manager
119. 4:44memory manager is the thing that maps
120. 4:47virtual addresses
121. 4:48to physical addresses
122. 4:51conceptually what it does it essentially
123. 4:55maps
124. 4:57each of these processes to a part of a
125. 5:00memory
126. 5:00so each of the processes um
127. 5:04although it thinks it is running from
128. 5:05the zero address to the
129. 5:07top of the address actually gets to use
130. 5:10only a part of the memory
131. 5:13these parts of the memory are not
132. 5:15necessarily contiguous
133. 5:17as drawn here in practice they're all
134. 5:21going to be more like interleaved
135. 5:22they're going to be all over the place
136. 5:25and
137. 5:25mixed with each other
138. 5:29furthermore conceptually what does the
139. 5:31memory manager need to do first it needs
140. 5:33to provide the translation
141. 5:34of virtual to physical address
142. 5:37spaces and then very importantly needs
143. 5:41to provide the protection
144. 5:42so the memory needs to be isolated
145. 5:44between the processes
146. 5:45such that each of these processes gets
147. 5:48its own
148. 5:49private memory consequence of that is
149. 5:52that errors in one program
150. 5:54you know coding errors that can happen
151. 5:57accidentally in one
152. 5:58program do not corrupt the memory of
153. 6:00another program
154. 6:01and very importantly prevents a user
155. 6:04program
156. 6:05from messing up with os's memory and
157. 6:09crashing the system
158. 6:12also since we have so many processes we
159. 6:15may run out of dram
160. 6:19but the memory manager is actually going
161. 6:22to prevent that
162. 6:22by expanding the memory system onto the
163. 6:25disk
164. 6:26so it will provide the solution that our
165. 6:29dram is much bigger
166. 6:32by swapping parts of the memory
167. 6:35of dram contents onto the disk now disk
168. 6:38is a lot
169. 6:39slower than our dram
170. 6:42so this should happen infrequently and
171. 6:44what we are going to see here
172. 6:45our dram is essentially going to be used
173. 6:48as a cache some sort of a cache for the
174. 6:52slower disk how does that all happen
175. 6:55you're going to see after a quick break