Transcript Memory Management in Linux

W4118 Operating Systems
Instructor: Junfeng Yang
Logistics

Homework 5 due 3:09pm this Thursday
1
Last lecture: LRU page replacement

Least-Recently-Used: throw out a page that
has not been referenced for the longest time


With locality, approximate OPT
However, difficult to implement efficiently

Clock: efficient, approximate LRU
Clock extension: bias toward replacing clean pages

Problem with LRU



Doesn’t work well repeated scans with data set larger
than memory
Doesn’t consider frequency
• LFU
• LRU-k and LRU-2Q
2
Today

Linux Memory Management

Address Space Translation

Address Space Layout

Dynamic Memory Allocation

Virtual Memory
• How does Linux translate address?
• What does a process Address Space look like?
• What does the kernel Address Space look like?
• How to implement Copy-on-Write (COW)?
• How to allocate pages?
• How to allocate objects?
• How to replace pages?
• How to load page from disk?
3
Address Translation

Linux uses paging to translate logical
addresses to physical addresses. Linux does
not use segmentation


More portable since some RISC architectures don’t
support segmentation
Hierarchical paging is flexible enough
4
Recall: x86 segmentation and paging
hardware

CPU generates logical address (seg, offset)

Given to segmentation unit
• Which produces linear addresses

Linear address given to paging unit
• Which generates physical address in main memory
• Paging units form equivalent of MMU
5
Segmentation

Since x86 segmentation hardware cannot be
disabled, Linux just uses NULL mappings




Set segment base to 0x00000000, limit to
0xffffffff
segment offset == linear addresses
arch/i386/kernel/head.S
Linux defines four segments




User code (__USER_CS)
User data (__USER_DS)
Kernel code (__KERNEL_CS)
Kernel data (__KERNEL_DATA)
Segment Protection

Current Privilege level (CPL) specifies privileged
mode or user mode




Descriptor Privilege Level (DPL) specifies
protection


Stored in current code segment descriptor
User code segment: CPL = 3
Kernel code segment: CPL = 0
Only accessible if CPL <= DPL
Switch between user mode and kernel mode (e.g.
system call and return)

Hardware load the corresponding segment selector
(__USER_CS or __KERNEL_CS) into register cs
7
Paging


Linux uses 4-level hierarchical paging
A linear address is split into five parts, to seamlessly
handle a range of different addressing modes






Page Global Dir
Page Upper Dir
Page Middle Dir
Page Table
Page Offset
Example: 32-bit address space, 4KB page without
physical address extension




Page Global dir: 10 bits
Page Upper dir and Page Middle dir are not used
Page Table: 10 bits
Page Offset: 12 bits
8
Page Table Operations

Linux provides data structures and operations
to create, delete, read and write page
directories



include/asm-i386/pgtable.h
arch/i386/mm/hugetlbpage.c
Naming convention





pgd: Page Global Directory
pmd: Page Middle Directory
pud: Page Upper Directory
pte: Page Table Entry
Example: mk_pte(p, prot)
9
TLB Operations

x86 uses hardware TLB


OS does not manage TLB
Only operation: flush TLB entries



include/asm-i386/tlbflush.h
load cr3: flush all TLB entries
invlpg addr: flush a single TLB entry
• Much more efficient than flushing all TLB entries
10
Today

Linux Memory Management

Address Space Translation

Address Space Layout

Dynamic Memory Allocation

Virtual Memory
• How does Linux translate address?
• What does a process Address Space look like?
• What does the kernel Address Space look like?
• How to implement Copy-on-Write (COW)?
• How to allocate pages?
• How to allocate objects?
• How to replace pages?
• How to load page from disk?
11
Recall: Linux Process Address Space
0xFFFFFFFF
kernel mode
Kernel space
0xC0000000
User-mode stack-area
User space
user mode
Shared runtime-libraries
Task’s code and data
0
Kernel data
structures
Kernel space is also
mapped into user
space  from user
mode to kernel
mode, no need to
switch address
spaces
Kernel Address Space

Kernel Address Space [3G, 4G)

[0, 896 MB) Physical Memory Mapping
• Kernel uses this mapping to manage physical memory
• physical address = logical address – 3G

The rest 128 MB are used for
• Accessing High Memory
• Non-contiguous pages
• Fix-mapped Linear Addresses
Physical
Memory
Mapping
0xC0000000
Persistent
vmalloc … vmalloc
area
area
High
Memory
Mappings
Fix-mapped
Linear
addresses
0xFFFFFFFF
13
A Cool Trick: Copy-on-Write

In fork(), parent and child often share
significant amount of memory


Expensive to do copy all pages
COW Idea: exploit VA to PA indirection



Instead of copying all pages, share parent pages in
child address space
If either process writes to shared pages, only then
is the page copied
How to detect page write?
• Mark pages as read-only in both parent and child
address space
• On write, page fault occurs
14
Sharing Pages








copy_process() in kernel/fork.c
copy_mm()
dup_mmap() // copy page tables
copy_page_range() in mm/memory.c
copy_pud_range()
copy_pmd_range()
copy_pte_range()
copy_one_pte() // mark readonly
15
Copy Page on Page Fault






do_page_fault in arch/i386/mm/fault.c
cr2 stores faulting virtual address
handle_mm_fault in mm/memory.c
handle_pte_fault in mm/memory.c
if(write_access)
do_wp_page()
16
Today

Linux Memory Management

Address Space Translation

Address Space Layout

Dynamic Memory Allocation

Virtual Memory
• How does Linux translate address?
• What does a process Address Space look like?
• What does the kernel Address Space look like?
• How to implement Copy-on-Write (COW)?
• How to allocate pages?
• How to allocate objects?
• How to replace pages?
• How to load page from disk?
17
Dynamic Memory Allocation

How to allocate pages?



Data structures for page allocation
Buddy algorithm for page allocation
How to allocate objects?

Slab allocation
18
Page Descriptor

Keep track of the status of each page frame



struct page, include/linux/mm.h
All stored in mem_map array
Simple mapping between a page and its
descriptor



Nth page’s descriptor is mem_map[N]
virt_to_page
page_to_pfn
Memory Zone

Keep track of pages in different zones




struct zone, include/linux/mmzone.h
ZONE_DMA: <16MB
ZONE_NORMAL: 16MB-896MB
ZONE_HIGHMEM: >896MB
Page Allocator

Linux use a buddy allocator for page allocation




Data structure: free lists of blocks of pages of size
2^0, 2^1, …
Allocation restrictions: 2^n pages
Allocation of 2^n pages:



Search free lists (n, n+1, n+2, …) for appropriate size
• Recursively divide larger blocks until reach block of correct
size
• Insert “buddy” blocks into free lists
Free


Buddy Allocator: Fast, simple allocation for blocks that
are 2^n bytes [Knuth 1968]
Recursively coalesce block with buddy if buddy free
Example: allocate a 256-page block
mm/page_alloc.c
Advantages and Disadvantages of Buddy
Allocator

Advantages


Fast and simple compared to general dynamic
memory allocation
Avoid external fragmentation by keeping free
physical pages contiguous
• Can use paging, but three problems:
– DMA bypasses paging
– Modifying page table leads to TLB flush
– Cannot use “super page” to increase TLB coverage

Disadvantages

Internal fragmentation
• Allocation of block of k pages when k != 2^n
Slab Allocator

For objects smaller than a page
Implemented on top of page allocator
Each memory region is called a cache

Two types of slab allocator



Fixed-size slab allocator: cache contains objects of same
size
• for frequently allocated objects

General-purpose slab allocator: caches contain objects of
size 2^n
• for less frequently allocated objects
• For allocation of object with size k, round to nearest 2^n

mm/slab.c
Advantages and Disadvantages of slab
allocator

Advantages


Reduce internal fragmentation: many objects in one
page
Fast: no need to allocate and free page frames
• Allocation: no search of objects with the right size
for fixed-size allocator; simple search for generalpurpose allocator
• Free: no merge with adjacent free blocks

Disadvantages


Memory overhead for bookkeeping
Internal fragmentation for general-purpose slab
allocator
Today

Linux Memory Management

Address Space Translation

Address Space Layout

Dynamic Memory Allocation

Virtual Memory
• How does Linux translate address?
• What does a process Address Space look like?
• What does the kernel Address Space look like?
• How to implement Copy-on-Write (COW)?
• How to allocate pages?
• How to allocate objects?
• How to replace pages?
• How to load page from disk?
25
Data Structures for Page
Replacement

Two lists in struct zone



active_list: hot pages
inactive_list: cold pages
Two bits in struct page


PG_active: is page on active list?
PG_referenced: has page been referenced recently?
26
Functions for Page Replacement




lru_cache_add*(): add to inactive or active list
mark_page_accessed(): called twice to move a
page from inactive to active
page_referenced(): test if a page is
referenced
refill_inactive_zone(): move pages from active
to inactive
When to Replace Pages?








Usually when free_more_memory() in
fs/buffer.c called
try_to_free_pages in mm/vmscan.c
shrink_caches
shrink_zone
refill_inactive_zone
shrink_cache
shrink_list
pageout()
28
How to Load Page?








do_page_fault() in arch/i386/mm/fault.c
cr2 stores faulting virtual address
handle_mm_fault() in mm/memory.c
handle_pte_fault()
if(!pte_present(entry))
do_no_page() // demand paging
do_file_page() // file mapped page
do_swap_page() // swapped out page
29