1Memory management for CRIS/MMU 2------------------------------ 3HISTORY: 4 5$Log: README.mm,v $ 6Revision 1.1 2000/07/10 16:25:21 bjornw 7Initial revision 8 9Revision 1.4 2000/01/17 02:31:59 bjornw 10Added discussion of paging and VM. 11 12Revision 1.3 1999/12/03 16:43:23 hp 13Blurb about that the 3.5G-limitation is not a MMU limitation 14 15Revision 1.2 1999/12/03 16:04:21 hp 16Picky comment about not mapping the first page 17 18Revision 1.1 1999/12/03 15:41:30 bjornw 19First version of CRIS/MMU memory layout specification. 20 21 22 23 24 25------------------------------ 26 27See the ETRAX-NG HSDD for reference. 28 29We use the page-size of 8 kbytes, as opposed to the i386 page-size of 4 kbytes. 30 31The MMU can, apart from the normal mapping of pages, also do a top-level 32segmentation of the kernel memory space. We use this feature to avoid having 33to use page-tables to map the physical memory into the kernel's address 34space. We also use it to keep the user-mode virtual mapping in the same 35map during kernel-mode, so that the kernel easily can access the corresponding 36user-mode process' data. 37 38As a comparision, the Linux/i386 2.0 puts the kernel and physical RAM at 39address 0, overlapping with the user-mode virtual space, so that descriptor 40registers are needed for each memory access to specify which MMU space to 41map through. That changed in 2.2, putting the kernel/physical RAM at 420xc0000000, to co-exist with the user-mode mapping. We will do something 43quite similar, but with the additional complexity of having to map the 44internal chip I/O registers and the flash memory area (including SRAM 45and peripherial chip-selets). 46 47The kernel-mode segmentation map: 48 49 ------------------------ ------------------------ 50FFFFFFFF| | => cached | | 51 | kernel seg_f | flash | | 52F0000000|______________________| | | 53EFFFFFFF| | => uncached | | 54 | kernel seg_e | flash | | 55E0000000|______________________| | DRAM | 56DFFFFFFF| | paged to any | Un-cached | 57 | kernel seg_d | =======> | | 58D0000000|______________________| | | 59CFFFFFFF| | | | 60 | kernel seg_c |==\ | | 61C0000000|______________________| \ |______________________| 62BFFFFFFF| | uncached | | 63 | kernel seg_b |=====\=========>| Registers | 64B0000000|______________________| \c |______________________| 65AFFFFFFF| | \a | | 66 | | \c | FLASH/SRAM/Peripheral| 67 | | \h |______________________| 68 | | \e | | 69 | | \d | | 70 | kernel seg_0 - seg_a | \==>| DRAM | 71 | | | Cached | 72 | | paged to any | | 73 | | =======> |______________________| 74 | | | | 75 | | | Illegal | 76 | | |______________________| 77 | | | | 78 | | | FLASH/SRAM/Peripheral| 7900000000|______________________| |______________________| 80 81In user-mode it looks the same except that only the space 0-AFFFFFFF is 82available. Therefore, in this model, the virtual address space per process 83is limited to 0xb0000000 bytes (minus 8192 bytes, since the first page, 840..8191, is never mapped, in order to trap NULL references). 85 86It also means that the total physical RAM that can be mapped is 256 MB 87(kseg_c above). More RAM can be mapped by choosing a different segmentation 88and shrinking the user-mode memory space. 89 90The MMU can map all 4 GB in user mode, but doing that would mean that a 91few extra instructions would be needed for each access to user mode 92memory. 93 94The kernel needs access to both cached and uncached flash. Uncached is 95necessary because of the special write/erase sequences. Also, the 96peripherial chip-selects are decoded from that region. 97 98The kernel also needs its own virtual memory space. That is kseg_d. It 99is used by the vmalloc() kernel function to allocate virtual contiguous 100chunks of memory not possible using the normal kmalloc physical RAM 101allocator. 102 103The setting of the actual MMU control registers to use this layout would 104be something like this: 105 106R_MMU_KSEG = ( ( seg_f, seg ) | // Flash cached 107 ( seg_e, seg ) | // Flash uncached 108 ( seg_d, page ) | // kernel vmalloc area 109 ( seg_c, seg ) | // kernel linear segment 110 ( seg_b, seg ) | // kernel linear segment 111 ( seg_a, page ) | 112 ( seg_9, page ) | 113 ( seg_8, page ) | 114 ( seg_7, page ) | 115 ( seg_6, page ) | 116 ( seg_5, page ) | 117 ( seg_4, page ) | 118 ( seg_3, page ) | 119 ( seg_2, page ) | 120 ( seg_1, page ) | 121 ( seg_0, page ) ); 122 123R_MMU_KBASE_HI = ( ( base_f, 0x0 ) | // flash/sram/periph cached 124 ( base_e, 0x8 ) | // flash/sram/periph uncached 125 ( base_d, 0x0 ) | // don't care 126 ( base_c, 0x4 ) | // physical RAM cached area 127 ( base_b, 0xb ) | // uncached on-chip registers 128 ( base_a, 0x0 ) | // don't care 129 ( base_9, 0x0 ) | // don't care 130 ( base_8, 0x0 ) ); // don't care 131 132R_MMU_KBASE_LO = ( ( base_7, 0x0 ) | // don't care 133 ( base_6, 0x0 ) | // don't care 134 ( base_5, 0x0 ) | // don't care 135 ( base_4, 0x0 ) | // don't care 136 ( base_3, 0x0 ) | // don't care 137 ( base_2, 0x0 ) | // don't care 138 ( base_1, 0x0 ) | // don't care 139 ( base_0, 0x0 ) ); // don't care 140 141NOTE: while setting up the MMU, we run in a non-mapped mode in the DRAM (0x40 142segment) and need to setup the seg_4 to a unity mapping, so that we don't get 143a fault before we have had time to jump into the real kernel segment (0xc0). This 144is done in head.S temporarily, but fixed by the kernel later in paging_init. 145 146 147Paging - PTE's, PMD's and PGD's 148------------------------------- 149 150[ References: asm/pgtable.h, asm/page.h, asm/mmu.h ] 151 152The paging mechanism uses virtual addresses to split a process memory-space into 153pages, a page being the smallest unit that can be freely remapped in memory. On 154Linux/CRIS, a page is 8192 bytes (for technical reasons not equal to 4096 as in 155most other 32-bit architectures). It would be inefficient to let a virtual memory 156mapping be controlled by a long table of page mappings, so it is broken down into 157a 2-level structure with a Page Directory containing pointers to Page Tables which 158each have maps of up to 2048 pages (8192 / sizeof(void *)). Linux can actually 159handle 3-level structures as well, with a Page Middle Directory in between, but 160in many cases, this is folded into a two-level structure by excluding the Middle 161Directory. 162 163We'll take a look at how an address is translated while we discuss how it's handled 164in the Linux kernel. 165 166The example address is 0xd004000c; in binary this is: 167 16831 23 15 7 0 16911010000 00000100 00000000 00001100 170 171|______| |__________||____________| 172 PGD PTE page offset 173 174Given the top-level Page Directory, the offset in that directory is calculated 175using the upper 8 bits: 176 177extern inline pgd_t * pgd_offset(struct mm_struct * mm, unsigned long address) 178{ 179 return mm->pgd + (address >> PGDIR_SHIFT); 180} 181 182PGDIR_SHIFT is the log2 of the amount of memory an entry in the PGD can map; in our 183case it is 24, corresponding to 16 MB. This means that each entry in the PGD 184corresponds to 16 MB of virtual memory. 185 186The pgd_t from our example will therefore be the 208'th (0xd0) entry in mm->pgd. 187 188Since the Middle Directory does not exist, it is a unity mapping: 189 190extern inline pmd_t * pmd_offset(pgd_t * dir, unsigned long address) 191{ 192 return (pmd_t *) dir; 193} 194 195The Page Table provides the final lookup by using bits 13 to 23 as index: 196 197extern inline pte_t * pte_offset(pmd_t * dir, unsigned long address) 198{ 199 return (pte_t *) pmd_page(*dir) + ((address >> PAGE_SHIFT) & 200 (PTRS_PER_PTE - 1)); 201} 202 203PAGE_SHIFT is the log2 of the size of a page; 13 in our case. PTRS_PER_PTE is 204the number of pointers that fit in a Page Table and is used to mask off the 205PGD-part of the address. 206 207The so-far unused bits 0 to 12 are used to index inside a page linearily. 208 209The VM system 210------------- 211 212The kernels own page-directory is the swapper_pg_dir, cleared in paging_init, 213and contains the kernels virtual mappings (the kernel itself is not paged - it 214is mapped linearily using kseg_c as described above). Architectures without 215kernel segments like the i386, need to setup swapper_pg_dir directly in head.S 216to map the kernel itself. swapper_pg_dir is pointed to by init_mm.pgd as the 217init-task's PGD. 218 219To see what support functions are used to setup a page-table, let's look at the 220kernel's internal paged memory system, vmalloc/vfree. 221 222void * vmalloc(unsigned long size) 223 224The vmalloc-system keeps a paged segment in kernel-space at 0xd0000000. What 225happens first is that a virtual address chunk is allocated to the request using 226get_vm_area(size). After that, physical RAM pages are allocated and put into 227the kernel's page-table using alloc_area_pages(addr, size). 228 229static int alloc_area_pages(unsigned long address, unsigned long size) 230 231First the PGD entry is found using init_mm.pgd. This is passed to 232alloc_area_pmd (remember the 3->2 folding). It uses pte_alloc_kernel to 233check if the PGD entry points anywhere - if not, a page table page is 234allocated and the PGD entry updated. Then the alloc_area_pte function is 235used just like alloc_area_pmd to check which page table entry is desired, 236and a physical page is allocated and the table entry updated. All of this 237is repeated at the top-level until the entire address range specified has 238been mapped. 239 240 241 242