1 | Memory management
|
---|
2 | =================
|
---|
3 |
|
---|
4 | 1. Virtual Address Translation
|
---|
5 |
|
---|
6 | 1.1 Hierarchical 4-level per address space page tables
|
---|
7 |
|
---|
8 | SPARTAN kernel deploys generic interface for 4-level page tables
|
---|
9 | for these architectures: amd64, ia32, mips32 and ppc32. In this
|
---|
10 | setting, page tables are hierarchical and are not shared by
|
---|
11 | address spaces (i.e. one set of page tables per address space).
|
---|
12 |
|
---|
13 |
|
---|
14 | VADDR
|
---|
15 | +-----------------------------------------------------------------------------+
|
---|
16 | | PTL0_INDEX | PTL1_INDEX | PTL2_INDEX | PTL3_INDEX | OFFSET |
|
---|
17 | +-----------------------------------------------------------------------------+
|
---|
18 |
|
---|
19 |
|
---|
20 | PTL0 PTL1 PTL2 PTL3
|
---|
21 | +--------+ +--------+ +--------+ +--------+
|
---|
22 | | | | | | PTL3 | -----\ | |
|
---|
23 | | | | | +--------+ | | |
|
---|
24 | | | +--------+ | | | | |
|
---|
25 | | | | PTL2 | -----\ | | | | |
|
---|
26 | | | +--------+ | | | | | |
|
---|
27 | | | | | | | | | +--------+
|
---|
28 | +--------+ | | | | | | | FRAME |
|
---|
29 | | PTL1 | -----\ | | | | | | +--------+
|
---|
30 | +--------+ | | | | | | | | |
|
---|
31 | | | | | | | | | | | |
|
---|
32 | | | | | | | | | | | |
|
---|
33 | +--------+ \----> +--------+ \----> +--------+ \----> +--------+
|
---|
34 | ^
|
---|
35 | |
|
---|
36 | |
|
---|
37 | +--------+
|
---|
38 | | PTL0 |
|
---|
39 | +--------+
|
---|
40 |
|
---|
41 |
|
---|
42 | PTL0 Page Table Level 0 (Page Directory)
|
---|
43 | PTL1 Page Table Level 1
|
---|
44 | PTL2 Page Table Level 2
|
---|
45 | PTL3 Page Table Level 3
|
---|
46 |
|
---|
47 | PTL0_INDEX Index into PTL0
|
---|
48 | PTL1_INDEX Index into PTL1
|
---|
49 | PTL2_INDEX Index into PTL2
|
---|
50 | PTL3_INDEX Index into PTL3
|
---|
51 |
|
---|
52 | VADDR Virtual address for which mapping is looked up
|
---|
53 | FRAME Physical address of memory frame to which VADDR is mapped
|
---|
54 |
|
---|
55 |
|
---|
56 | On architectures whose hardware has fewer levels, PTL2 and, if need be, PTL1 are
|
---|
57 | left out. TLB-only architectures are to define custom format for software page
|
---|
58 | tables.
|
---|
59 |
|
---|
60 |
|
---|
61 |
|
---|
62 | 1.2 Single global page hash table
|
---|
63 |
|
---|
64 | Generic page hash table interface is deployed on 64-bit architectures without
|
---|
65 | implied hardware support for hierarchical page tables, i.e. ia64 and sparc64.
|
---|
66 | There is only one global page hash table in the system shared by all address
|
---|
67 | spaces.
|
---|
68 |
|
---|
69 | 2.1 General allocator
|
---|
70 |
|
---|
71 | 'malloc' function accepts flags as a second argument. The flags are directly
|
---|
72 | passed to the underlying frame_alloc function.
|
---|
73 |
|
---|
74 | 1) If the flags parameter contains FRAME_ATOMIC, the allocator will not sleep.
|
---|
75 | The allocator CAN return NULL, when memory is not directly available.
|
---|
76 | The caller MUST check if NULL was not returned
|
---|
77 |
|
---|
78 | 2) If the flags parameter does not contain FRAME_ATOMIC, the allocator
|
---|
79 | will never return NULL, but it CAN sleep indefinitely. The caller
|
---|
80 | does not have to check the return value.
|
---|
81 |
|
---|
82 | 3) The maximum size that can be allocated using malloc is 128K
|
---|
83 |
|
---|
84 | Rules 1) and 2) apply to slab_alloc as well. Using SLAB allocator
|
---|
85 | to allocate too large values is not recommended.
|
---|
86 |
|
---|