Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
README.mm
Go to the documentation of this file.
1 Memory management for CRIS/MMU
2 ------------------------------
3 HISTORY:
4 
5 $Log: README.mm,v $
6 Revision 1.1 2001/12/17 13:59:27 bjornw
7 Initial revision
8 
9 Revision 1.1 2000/07/10 16:25:21 bjornw
10 Initial revision
11 
12 Revision 1.4 2000/01/17 02:31:59 bjornw
13 Added discussion of paging and VM.
14 
15 Revision 1.3 1999/12/03 16:43:23 hp
16 Blurb about that the 3.5G-limitation is not a MMU limitation
17 
18 Revision 1.2 1999/12/03 16:04:21 hp
19 Picky comment about not mapping the first page
20 
21 Revision 1.1 1999/12/03 15:41:30 bjornw
22 First version of CRIS/MMU memory layout specification.
23 
24 
25 
26 
27 
28 ------------------------------
29 
30 See the ETRAX-NG HSDD for reference.
31 
32 We use the page-size of 8 kbytes, as opposed to the i386 page-size of 4 kbytes.
33 
34 The MMU can, apart from the normal mapping of pages, also do a top-level
35 segmentation of the kernel memory space. We use this feature to avoid having
36 to use page-tables to map the physical memory into the kernel's address
37 space. We also use it to keep the user-mode virtual mapping in the same
38 map during kernel-mode, so that the kernel easily can access the corresponding
39 user-mode process' data.
40 
41 As a comparison, the Linux/i386 2.0 puts the kernel and physical RAM at
42 address 0, overlapping with the user-mode virtual space, so that descriptor
43 registers are needed for each memory access to specify which MMU space to
44 map through. That changed in 2.2, putting the kernel/physical RAM at
45 0xc0000000, to co-exist with the user-mode mapping. We will do something
46 quite similar, but with the additional complexity of having to map the
47 internal chip I/O registers and the flash memory area (including SRAM
48 and peripherial chip-selets).
49 
50 The kernel-mode segmentation map:
51 
52  ------------------------ ------------------------
53 FFFFFFFF| | => cached | |
54  | kernel seg_f | flash | |
55 F0000000|______________________| | |
56 EFFFFFFF| | => uncached | |
57  | kernel seg_e | flash | |
58 E0000000|______________________| | DRAM |
59 DFFFFFFF| | paged to any | Un-cached |
60  | kernel seg_d | =======> | |
61 D0000000|______________________| | |
62 CFFFFFFF| | | |
63  | kernel seg_c |==\ | |
64 C0000000|______________________| \ |______________________|
65 BFFFFFFF| | uncached | |
66  | kernel seg_b |=====\=========>| Registers |
67 B0000000|______________________| \c |______________________|
68 AFFFFFFF| | \a | |
69  | | \c | FLASH/SRAM/Peripheral|
70  | | \h |______________________|
71  | | \e | |
72  | | \d | |
73  | kernel seg_0 - seg_a | \==>| DRAM |
74  | | | Cached |
75  | | paged to any | |
76  | | =======> |______________________|
77  | | | |
78  | | | Illegal |
79  | | |______________________|
80  | | | |
81  | | | FLASH/SRAM/Peripheral|
82 00000000|______________________| |______________________|
83 
84 In user-mode it looks the same except that only the space 0-AFFFFFFF is
85 available. Therefore, in this model, the virtual address space per process
86 is limited to 0xb0000000 bytes (minus 8192 bytes, since the first page,
87 0..8191, is never mapped, in order to trap NULL references).
88 
89 It also means that the total physical RAM that can be mapped is 256 MB
90 (kseg_c above). More RAM can be mapped by choosing a different segmentation
91 and shrinking the user-mode memory space.
92 
93 The MMU can map all 4 GB in user mode, but doing that would mean that a
94 few extra instructions would be needed for each access to user mode
95 memory.
96 
97 The kernel needs access to both cached and uncached flash. Uncached is
98 necessary because of the special write/erase sequences. Also, the
99 peripherial chip-selects are decoded from that region.
100 
101 The kernel also needs its own virtual memory space. That is kseg_d. It
102 is used by the vmalloc() kernel function to allocate virtual contiguous
103 chunks of memory not possible using the normal kmalloc physical RAM
104 allocator.
105 
106 The setting of the actual MMU control registers to use this layout would
107 be something like this:
108 
109 R_MMU_KSEG = ( ( seg_f, seg ) | // Flash cached
110  ( seg_e, seg ) | // Flash uncached
111  ( seg_d, page ) | // kernel vmalloc area
112  ( seg_c, seg ) | // kernel linear segment
113  ( seg_b, seg ) | // kernel linear segment
114  ( seg_a, page ) |
115  ( seg_9, page ) |
116  ( seg_8, page ) |
117  ( seg_7, page ) |
118  ( seg_6, page ) |
119  ( seg_5, page ) |
120  ( seg_4, page ) |
121  ( seg_3, page ) |
122  ( seg_2, page ) |
123  ( seg_1, page ) |
124  ( seg_0, page ) );
125 
126 R_MMU_KBASE_HI = ( ( base_f, 0x0 ) | // flash/sram/periph cached
127  ( base_e, 0x8 ) | // flash/sram/periph uncached
128  ( base_d, 0x0 ) | // don't care
129  ( base_c, 0x4 ) | // physical RAM cached area
130  ( base_b, 0xb ) | // uncached on-chip registers
131  ( base_a, 0x0 ) | // don't care
132  ( base_9, 0x0 ) | // don't care
133  ( base_8, 0x0 ) ); // don't care
134 
135 R_MMU_KBASE_LO = ( ( base_7, 0x0 ) | // don't care
136  ( base_6, 0x0 ) | // don't care
137  ( base_5, 0x0 ) | // don't care
138  ( base_4, 0x0 ) | // don't care
139  ( base_3, 0x0 ) | // don't care
140  ( base_2, 0x0 ) | // don't care
141  ( base_1, 0x0 ) | // don't care
142  ( base_0, 0x0 ) ); // don't care
143 
144 NOTE: while setting up the MMU, we run in a non-mapped mode in the DRAM (0x40
145 segment) and need to setup the seg_4 to a unity mapping, so that we don't get
146 a fault before we have had time to jump into the real kernel segment (0xc0). This
147 is done in head.S temporarily, but fixed by the kernel later in paging_init.
148 
149 
150 Paging - PTE's, PMD's and PGD's
151 -------------------------------
152 
153 [ References: asm/pgtable.h, asm/page.h, asm/mmu.h ]
154 
155 The paging mechanism uses virtual addresses to split a process memory-space into
156 pages, a page being the smallest unit that can be freely remapped in memory. On
157 Linux/CRIS, a page is 8192 bytes (for technical reasons not equal to 4096 as in
158 most other 32-bit architectures). It would be inefficient to let a virtual memory
159 mapping be controlled by a long table of page mappings, so it is broken down into
160 a 2-level structure with a Page Directory containing pointers to Page Tables which
161 each have maps of up to 2048 pages (8192 / sizeof(void *)). Linux can actually
162 handle 3-level structures as well, with a Page Middle Directory in between, but
163 in many cases, this is folded into a two-level structure by excluding the Middle
164 Directory.
165 
166 We'll take a look at how an address is translated while we discuss how it's handled
167 in the Linux kernel.
168 
169 The example address is 0xd004000c; in binary this is:
170 
171 31 23 15 7 0
172 11010000 00000100 00000000 00001100
173 
174 |______| |__________||____________|
175  PGD PTE page offset
176 
177 Given the top-level Page Directory, the offset in that directory is calculated
178 using the upper 8 bits:
179 
180 static inline pgd_t * pgd_offset(struct mm_struct * mm, unsigned long address)
181 {
182  return mm->pgd + (address >> PGDIR_SHIFT);
183 }
184 
185 PGDIR_SHIFT is the log2 of the amount of memory an entry in the PGD can map; in our
186 case it is 24, corresponding to 16 MB. This means that each entry in the PGD
187 corresponds to 16 MB of virtual memory.
188 
189 The pgd_t from our example will therefore be the 208'th (0xd0) entry in mm->pgd.
190 
191 Since the Middle Directory does not exist, it is a unity mapping:
192 
193 static inline pmd_t * pmd_offset(pgd_t * dir, unsigned long address)
194 {
195  return (pmd_t *) dir;
196 }
197 
198 The Page Table provides the final lookup by using bits 13 to 23 as index:
199 
200 static inline pte_t * pte_offset(pmd_t * dir, unsigned long address)
201 {
202  return (pte_t *) pmd_page(*dir) + ((address >> PAGE_SHIFT) &
203  (PTRS_PER_PTE - 1));
204 }
205 
206 PAGE_SHIFT is the log2 of the size of a page; 13 in our case. PTRS_PER_PTE is
207 the number of pointers that fit in a Page Table and is used to mask off the
208 PGD-part of the address.
209 
210 The so-far unused bits 0 to 12 are used to index inside a page linearily.
211 
212 The VM system
213 -------------
214 
215 The kernels own page-directory is the swapper_pg_dir, cleared in paging_init,
216 and contains the kernels virtual mappings (the kernel itself is not paged - it
217 is mapped linearily using kseg_c as described above). Architectures without
218 kernel segments like the i386, need to setup swapper_pg_dir directly in head.S
219 to map the kernel itself. swapper_pg_dir is pointed to by init_mm.pgd as the
220 init-task's PGD.
221 
222 To see what support functions are used to setup a page-table, let's look at the
223 kernel's internal paged memory system, vmalloc/vfree.
224 
225 void * vmalloc(unsigned long size)
226 
227 The vmalloc-system keeps a paged segment in kernel-space at 0xd0000000. What
228 happens first is that a virtual address chunk is allocated to the request using
229 get_vm_area(size). After that, physical RAM pages are allocated and put into
230 the kernel's page-table using alloc_area_pages(addr, size).
231 
232 static int alloc_area_pages(unsigned long address, unsigned long size)
233 
234 First the PGD entry is found using init_mm.pgd. This is passed to
235 alloc_area_pmd (remember the 3->2 folding). It uses pte_alloc_kernel to
236 check if the PGD entry points anywhere - if not, a page table page is
237 allocated and the PGD entry updated. Then the alloc_area_pte function is
238 used just like alloc_area_pmd to check which page table entry is desired,
239 and a physical page is allocated and the table entry updated. All of this
240 is repeated at the top-level until the entire address range specified has
241 been mapped.
242 
243 
244 
Generated on Thu Jan 10 2013 13:05:13 for Linux Kernel by   doxygen 1.8.2