---
tags: 作業系統
---

# 作業系統 ch9 Main Memory

### 基本概念

- CPU register -> cache -> main memory, three type CPU can access directly
- stall: cycles, main memory too slow
- register access: CPU a clock time ( or less)
- memory unit: (address, read request)、(data, write request)
- program must be load into memory => then execute by CPU
- 

---

### address binding
- input queue
- 準備進入 Memory
- load into 0000 (depend on PC)

- addreess of diff stages
- source code ( symbolic )
- compiler ( relocation address )
- linker \ loader ( absoult address )

- bind in diff time
- compile time
- priori: absoult code
- other program load -> recompile memory address
- load time
- relocate, if memory location not known at compile time
- execution time
- binding delay
- hardware support (Base and Limit registers)
a. logical address
b. hardware protection

---
### address space
- address type
- logical (virtual) address: generate by CPU
- physical address: memory unit use

- address space
- address generate by program

- logical and physical address diff in load / run time address-binding scheme

---

### Memory management unit ( MMU )
- hardware
- map logical -> physical address
- base register (relocation register、4)
- user mode use logical address

---

### dynamic relocation use relocation register
- load routine when called
- don't need special support from OS

---

### Dynamic linking (shared lib)
- linking in executetion time
- stub (一段 code): find address of memory-resident lib routine, stub 將自已跟 routine 交換
- patch system lib
- versioning, 需要 OS support

---

### Dynamic loading
- user 自行規劃, 不由 OS 負責
- 古老方法

---

### swapping
- memory ->(swap) -> back store
- backing store
- fast disk
- DMA to memory image
- roll in, roll out (priority)
- swap time -> transfer time -> amount of memory swap
- ready queue(have memory image in disk)
- can load in diff original position? A: yes if dynamic binding

---

### context swap
- problem : pending I/O swap out case error
- can't swap out process which use I/O
- transfer I/O to kernel space (double buffering)

---

### swap in mobile
- no support swap, because flash memory
- both support paging
- ios
- 自願放棄 (voluntarily relinquish)
- throw out read-only data
- termination if fail to free

- Android
- termination if low free memory, write application state for fast restart

---

### contiguous allocation
- Main memory support user and OS process, each include a single section
- two partition
- kernal process(resident OS): low memory, interrupt vector
- user process: high memory

- relocation register
- base: smallest physical address
- limit register: logical address must be less than the limit register
- MMU maps logical address dynamic
- kernal code transient, kernal change size

---

### multiple partition allocation
- degree: number of partition
- partition with different size, continuous memory
- hole: available memory
- OS record some information
- a. allocated partition
- b. free partition(hole)
---

### dynamic allocation ways
- first fit
- best fit
- worst fit(largest)

---

### Fragmentation
- external fragmentation
- reduce by compaction : must be dynamic binding, done at execution time (shuffle memory content)
- internal fragmentation

---

### Segmentation
- user view of memory
- a segment = logic unit
- no internal fragmentation
- each process have segment table, map physical address
- base、limit
- segment table base register、segment table length register(check segment number < STLR)
- logic address < segment number, offset>
- protection: validation bit、privilege

---

### Page
- physical memory(frame)、logic memory(page)
- track all free frame
- page table
- each process have one page table
- save in main memory
- logical -> (page table) -> physical memory
- logic address <page number(page table index), offset>
- PTBR、PTLR
- twice memory access
- TLB(translate lookaside buffer): store address-space identifier(ASIDs), like cache
- 計算 Effective Access Time： hit:seach+acess*1, miss: search+acess*2
- associative memory(parallel search)

---

### Memory Protection
- protection bit(R、RW)
- valid-invalid bit( can use PTLR )
- illegel -> trap to kernal

---

### Share page
- share code reentrant
- use in interprocess communication
- private code
- each process have a copy

---

### structure of page table
- hierarchical paging
- 32 bit (12, 10, 10 => outer, inner page number, offset)
- multiple page table
- two-level (forward-mapped)=> outer、inner
- hashed page table
- \>32 bit
- useful for sparse address
- virtual page number
- inverted page table
- track physical address
- one page、frame have one entry in page table
- < PID, page number >