1Memory management for CRIS/MMU 2------------------------------ 3HISTORY: 4 5$Log: README.mm,v $ 6Revision 1.1.1.1 2007/08/03 18:51:41 rnuti 7Importing Linux MIPS Kernel 2.6.22 8 9Revision 1.1 2001/12/17 13:59:27 bjornw 10Initial revision 11 12Revision 1.1 2000/07/10 16:25:21 bjornw 13Initial revision 14 15Revision 1.4 2000/01/17 02:31:59 bjornw 16Added discussion of paging and VM. 17 18Revision 1.3 1999/12/03 16:43:23 hp 19Blurb about that the 3.5G-limitation is not a MMU limitation 20 21Revision 1.2 1999/12/03 16:04:21 hp 22Picky comment about not mapping the first page 23 24Revision 1.1 1999/12/03 15:41:30 bjornw 25First version of CRIS/MMU memory layout specification. 26 27 28 29 30 31------------------------------ 32 33See the ETRAX-NG HSDD for reference. 34 35We use the page-size of 8 kbytes, as opposed to the i386 page-size of 4 kbytes. 36 37The MMU can, apart from the normal mapping of pages, also do a top-level 38segmentation of the kernel memory space. We use this feature to avoid having 39to use page-tables to map the physical memory into the kernel's address 40space. We also use it to keep the user-mode virtual mapping in the same 41map during kernel-mode, so that the kernel easily can access the corresponding 42user-mode process' data. 43 44As a comparision, the Linux/i386 2.0 puts the kernel and physical RAM at 45address 0, overlapping with the user-mode virtual space, so that descriptor 46registers are needed for each memory access to specify which MMU space to 47map through. That changed in 2.2, putting the kernel/physical RAM at 480xc0000000, to co-exist with the user-mode mapping. We will do something 49quite similar, but with the additional complexity of having to map the 50internal chip I/O registers and the flash memory area (including SRAM 51and peripherial chip-selets). 52 53The kernel-mode segmentation map: 54 55 ------------------------ ------------------------ 56FFFFFFFF| | => cached | | 57 | kernel seg_f | flash | | 58F0000000|______________________| | | 59EFFFFFFF| | => uncached | | 60 | kernel seg_e | flash | | 61E0000000|______________________| | DRAM | 62DFFFFFFF| | paged to any | Un-cached | 63 | kernel seg_d | =======> | | 64D0000000|______________________| | | 65CFFFFFFF| | | | 66 | kernel seg_c |==\ | | 67C0000000|______________________| \ |______________________| 68BFFFFFFF| | uncached | | 69 | kernel seg_b |=====\=========>| Registers | 70B0000000|______________________| \c |______________________| 71AFFFFFFF| | \a | | 72 | | \c | FLASH/SRAM/Peripheral| 73 | | \h |______________________| 74 | | \e | | 75 | | \d | | 76 | kernel seg_0 - seg_a | \==>| DRAM | 77 | | | Cached | 78 | | paged to any | | 79 | | =======> |______________________| 80 | | | | 81 | | | Illegal | 82 | | |______________________| 83 | | | | 84 | | | FLASH/SRAM/Peripheral| 8500000000|______________________| |______________________| 86 87In user-mode it looks the same except that only the space 0-AFFFFFFF is 88available. Therefore, in this model, the virtual address space per process 89is limited to 0xb0000000 bytes (minus 8192 bytes, since the first page, 900..8191, is never mapped, in order to trap NULL references). 91 92It also means that the total physical RAM that can be mapped is 256 MB 93(kseg_c above). More RAM can be mapped by choosing a different segmentation 94and shrinking the user-mode memory space. 95 96The MMU can map all 4 GB in user mode, but doing that would mean that a 97few extra instructions would be needed for each access to user mode 98memory. 99 100The kernel needs access to both cached and uncached flash. Uncached is 101necessary because of the special write/erase sequences. Also, the 102peripherial chip-selects are decoded from that region. 103 104The kernel also needs its own virtual memory space. That is kseg_d. It 105is used by the vmalloc() kernel function to allocate virtual contiguous 106chunks of memory not possible using the normal kmalloc physical RAM 107allocator. 108 109The setting of the actual MMU control registers to use this layout would 110be something like this: 111 112R_MMU_KSEG = ( ( seg_f, seg ) | // Flash cached 113 ( seg_e, seg ) | // Flash uncached 114 ( seg_d, page ) | // kernel vmalloc area 115 ( seg_c, seg ) | // kernel linear segment 116 ( seg_b, seg ) | // kernel linear segment 117 ( seg_a, page ) | 118 ( seg_9, page ) | 119 ( seg_8, page ) | 120 ( seg_7, page ) | 121 ( seg_6, page ) | 122 ( seg_5, page ) | 123 ( seg_4, page ) | 124 ( seg_3, page ) | 125 ( seg_2, page ) | 126 ( seg_1, page ) | 127 ( seg_0, page ) ); 128 129R_MMU_KBASE_HI = ( ( base_f, 0x0 ) | // flash/sram/periph cached 130 ( base_e, 0x8 ) | // flash/sram/periph uncached 131 ( base_d, 0x0 ) | // don't care 132 ( base_c, 0x4 ) | // physical RAM cached area 133 ( base_b, 0xb ) | // uncached on-chip registers 134 ( base_a, 0x0 ) | // don't care 135 ( base_9, 0x0 ) | // don't care 136 ( base_8, 0x0 ) ); // don't care 137 138R_MMU_KBASE_LO = ( ( base_7, 0x0 ) | // don't care 139 ( base_6, 0x0 ) | // don't care 140 ( base_5, 0x0 ) | // don't care 141 ( base_4, 0x0 ) | // don't care 142 ( base_3, 0x0 ) | // don't care 143 ( base_2, 0x0 ) | // don't care 144 ( base_1, 0x0 ) | // don't care 145 ( base_0, 0x0 ) ); // don't care 146 147NOTE: while setting up the MMU, we run in a non-mapped mode in the DRAM (0x40 148segment) and need to setup the seg_4 to a unity mapping, so that we don't get 149a fault before we have had time to jump into the real kernel segment (0xc0). This 150is done in head.S temporarily, but fixed by the kernel later in paging_init. 151 152 153Paging - PTE's, PMD's and PGD's 154------------------------------- 155 156[ References: asm/pgtable.h, asm/page.h, asm/mmu.h ] 157 158The paging mechanism uses virtual addresses to split a process memory-space into 159pages, a page being the smallest unit that can be freely remapped in memory. On 160Linux/CRIS, a page is 8192 bytes (for technical reasons not equal to 4096 as in 161most other 32-bit architectures). It would be inefficient to let a virtual memory 162mapping be controlled by a long table of page mappings, so it is broken down into 163a 2-level structure with a Page Directory containing pointers to Page Tables which 164each have maps of up to 2048 pages (8192 / sizeof(void *)). Linux can actually 165handle 3-level structures as well, with a Page Middle Directory in between, but 166in many cases, this is folded into a two-level structure by excluding the Middle 167Directory. 168 169We'll take a look at how an address is translated while we discuss how it's handled 170in the Linux kernel. 171 172The example address is 0xd004000c; in binary this is: 173 17431 23 15 7 0 17511010000 00000100 00000000 00001100 176 177|______| |__________||____________| 178 PGD PTE page offset 179 180Given the top-level Page Directory, the offset in that directory is calculated 181using the upper 8 bits: 182 183static inline pgd_t * pgd_offset(struct mm_struct * mm, unsigned long address) 184{ 185 return mm->pgd + (address >> PGDIR_SHIFT); 186} 187 188PGDIR_SHIFT is the log2 of the amount of memory an entry in the PGD can map; in our 189case it is 24, corresponding to 16 MB. This means that each entry in the PGD 190corresponds to 16 MB of virtual memory. 191 192The pgd_t from our example will therefore be the 208'th (0xd0) entry in mm->pgd. 193 194Since the Middle Directory does not exist, it is a unity mapping: 195 196static inline pmd_t * pmd_offset(pgd_t * dir, unsigned long address) 197{ 198 return (pmd_t *) dir; 199} 200 201The Page Table provides the final lookup by using bits 13 to 23 as index: 202 203static inline pte_t * pte_offset(pmd_t * dir, unsigned long address) 204{ 205 return (pte_t *) pmd_page(*dir) + ((address >> PAGE_SHIFT) & 206 (PTRS_PER_PTE - 1)); 207} 208 209PAGE_SHIFT is the log2 of the size of a page; 13 in our case. PTRS_PER_PTE is 210the number of pointers that fit in a Page Table and is used to mask off the 211PGD-part of the address. 212 213The so-far unused bits 0 to 12 are used to index inside a page linearily. 214 215The VM system 216------------- 217 218The kernels own page-directory is the swapper_pg_dir, cleared in paging_init, 219and contains the kernels virtual mappings (the kernel itself is not paged - it 220is mapped linearily using kseg_c as described above). Architectures without 221kernel segments like the i386, need to setup swapper_pg_dir directly in head.S 222to map the kernel itself. swapper_pg_dir is pointed to by init_mm.pgd as the 223init-task's PGD. 224 225To see what support functions are used to setup a page-table, let's look at the 226kernel's internal paged memory system, vmalloc/vfree. 227 228void * vmalloc(unsigned long size) 229 230The vmalloc-system keeps a paged segment in kernel-space at 0xd0000000. What 231happens first is that a virtual address chunk is allocated to the request using 232get_vm_area(size). After that, physical RAM pages are allocated and put into 233the kernel's page-table using alloc_area_pages(addr, size). 234 235static int alloc_area_pages(unsigned long address, unsigned long size) 236 237First the PGD entry is found using init_mm.pgd. This is passed to 238alloc_area_pmd (remember the 3->2 folding). It uses pte_alloc_kernel to 239check if the PGD entry points anywhere - if not, a page table page is 240allocated and the PGD entry updated. Then the alloc_area_pte function is 241used just like alloc_area_pmd to check which page table entry is desired, 242and a physical page is allocated and the table entry updated. All of this 243is repeated at the top-level until the entire address range specified has 244been mapped. 245