Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com> |
| 3 | * Copyright (c) 1999-2000 Grant Erickson <grant@lcse.umn.edu> |
| 4 | * |
| 5 | * Module name: iSeries_setup.c |
| 6 | * |
| 7 | * Description: |
| 8 | * Architecture- / platform-specific boot-time initialization code for |
| 9 | * the IBM iSeries LPAR. Adapted from original code by Grant Erickson and |
| 10 | * code by Gary Thomas, Cort Dougan <cort@fsmlabs.com>, and Dan Malek |
| 11 | * <dan@net4x.com>. |
| 12 | * |
| 13 | * This program is free software; you can redistribute it and/or |
| 14 | * modify it under the terms of the GNU General Public License |
| 15 | * as published by the Free Software Foundation; either version |
| 16 | * 2 of the License, or (at your option) any later version. |
| 17 | */ |
| 18 | |
| 19 | #undef DEBUG |
| 20 | |
| 21 | #include <linux/config.h> |
| 22 | #include <linux/init.h> |
| 23 | #include <linux/threads.h> |
| 24 | #include <linux/smp.h> |
| 25 | #include <linux/param.h> |
| 26 | #include <linux/string.h> |
| 27 | #include <linux/bootmem.h> |
| 28 | #include <linux/initrd.h> |
| 29 | #include <linux/seq_file.h> |
| 30 | #include <linux/kdev_t.h> |
| 31 | #include <linux/major.h> |
| 32 | #include <linux/root_dev.h> |
| 33 | |
| 34 | #include <asm/processor.h> |
| 35 | #include <asm/machdep.h> |
| 36 | #include <asm/page.h> |
| 37 | #include <asm/mmu.h> |
| 38 | #include <asm/pgtable.h> |
| 39 | #include <asm/mmu_context.h> |
| 40 | #include <asm/cputable.h> |
| 41 | #include <asm/sections.h> |
| 42 | #include <asm/iommu.h> |
| 43 | |
| 44 | #include <asm/time.h> |
| 45 | #include "iSeries_setup.h" |
| 46 | #include <asm/naca.h> |
| 47 | #include <asm/paca.h> |
| 48 | #include <asm/cache.h> |
| 49 | #include <asm/sections.h> |
| 50 | #include <asm/iSeries/LparData.h> |
| 51 | #include <asm/iSeries/HvCallHpt.h> |
| 52 | #include <asm/iSeries/HvLpConfig.h> |
| 53 | #include <asm/iSeries/HvCallEvent.h> |
| 54 | #include <asm/iSeries/HvCallSm.h> |
| 55 | #include <asm/iSeries/HvCallXm.h> |
| 56 | #include <asm/iSeries/ItLpQueue.h> |
| 57 | #include <asm/iSeries/IoHriMainStore.h> |
| 58 | #include <asm/iSeries/iSeries_proc.h> |
| 59 | #include <asm/iSeries/mf.h> |
| 60 | #include <asm/iSeries/HvLpEvent.h> |
| 61 | #include <asm/iSeries/iSeries_irq.h> |
| 62 | |
| 63 | extern void hvlog(char *fmt, ...); |
| 64 | |
| 65 | #ifdef DEBUG |
| 66 | #define DBG(fmt...) hvlog(fmt) |
| 67 | #else |
| 68 | #define DBG(fmt...) |
| 69 | #endif |
| 70 | |
| 71 | /* Function Prototypes */ |
| 72 | extern void ppcdbg_initialize(void); |
| 73 | |
| 74 | static void build_iSeries_Memory_Map(void); |
| 75 | static void setup_iSeries_cache_sizes(void); |
| 76 | static void iSeries_bolt_kernel(unsigned long saddr, unsigned long eaddr); |
| 77 | extern void iSeries_pci_final_fixup(void); |
| 78 | |
| 79 | /* Global Variables */ |
| 80 | static unsigned long procFreqHz; |
| 81 | static unsigned long procFreqMhz; |
| 82 | static unsigned long procFreqMhzHundreths; |
| 83 | |
| 84 | static unsigned long tbFreqHz; |
| 85 | static unsigned long tbFreqMhz; |
| 86 | static unsigned long tbFreqMhzHundreths; |
| 87 | |
| 88 | int piranha_simulator; |
| 89 | |
| 90 | extern int rd_size; /* Defined in drivers/block/rd.c */ |
| 91 | extern unsigned long klimit; |
| 92 | extern unsigned long embedded_sysmap_start; |
| 93 | extern unsigned long embedded_sysmap_end; |
| 94 | |
| 95 | extern unsigned long iSeries_recal_tb; |
| 96 | extern unsigned long iSeries_recal_titan; |
| 97 | |
| 98 | static int mf_initialized; |
| 99 | |
| 100 | struct MemoryBlock { |
| 101 | unsigned long absStart; |
| 102 | unsigned long absEnd; |
| 103 | unsigned long logicalStart; |
| 104 | unsigned long logicalEnd; |
| 105 | }; |
| 106 | |
| 107 | /* |
| 108 | * Process the main store vpd to determine where the holes in memory are |
| 109 | * and return the number of physical blocks and fill in the array of |
| 110 | * block data. |
| 111 | */ |
| 112 | static unsigned long iSeries_process_Condor_mainstore_vpd( |
| 113 | struct MemoryBlock *mb_array, unsigned long max_entries) |
| 114 | { |
| 115 | unsigned long holeFirstChunk, holeSizeChunks; |
| 116 | unsigned long numMemoryBlocks = 1; |
| 117 | struct IoHriMainStoreSegment4 *msVpd = |
| 118 | (struct IoHriMainStoreSegment4 *)xMsVpd; |
| 119 | unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr; |
| 120 | unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr; |
| 121 | unsigned long holeSize = holeEnd - holeStart; |
| 122 | |
| 123 | printk("Mainstore_VPD: Condor\n"); |
| 124 | /* |
| 125 | * Determine if absolute memory has any |
| 126 | * holes so that we can interpret the |
| 127 | * access map we get back from the hypervisor |
| 128 | * correctly. |
| 129 | */ |
| 130 | mb_array[0].logicalStart = 0; |
| 131 | mb_array[0].logicalEnd = 0x100000000; |
| 132 | mb_array[0].absStart = 0; |
| 133 | mb_array[0].absEnd = 0x100000000; |
| 134 | |
| 135 | if (holeSize) { |
| 136 | numMemoryBlocks = 2; |
| 137 | holeStart = holeStart & 0x000fffffffffffff; |
| 138 | holeStart = addr_to_chunk(holeStart); |
| 139 | holeFirstChunk = holeStart; |
| 140 | holeSize = addr_to_chunk(holeSize); |
| 141 | holeSizeChunks = holeSize; |
| 142 | printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n", |
| 143 | holeFirstChunk, holeSizeChunks ); |
| 144 | mb_array[0].logicalEnd = holeFirstChunk; |
| 145 | mb_array[0].absEnd = holeFirstChunk; |
| 146 | mb_array[1].logicalStart = holeFirstChunk; |
| 147 | mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks; |
| 148 | mb_array[1].absStart = holeFirstChunk + holeSizeChunks; |
| 149 | mb_array[1].absEnd = 0x100000000; |
| 150 | } |
| 151 | return numMemoryBlocks; |
| 152 | } |
| 153 | |
| 154 | #define MaxSegmentAreas 32 |
| 155 | #define MaxSegmentAdrRangeBlocks 128 |
| 156 | #define MaxAreaRangeBlocks 4 |
| 157 | |
| 158 | static unsigned long iSeries_process_Regatta_mainstore_vpd( |
| 159 | struct MemoryBlock *mb_array, unsigned long max_entries) |
| 160 | { |
| 161 | struct IoHriMainStoreSegment5 *msVpdP = |
| 162 | (struct IoHriMainStoreSegment5 *)xMsVpd; |
| 163 | unsigned long numSegmentBlocks = 0; |
| 164 | u32 existsBits = msVpdP->msAreaExists; |
| 165 | unsigned long area_num; |
| 166 | |
| 167 | printk("Mainstore_VPD: Regatta\n"); |
| 168 | |
| 169 | for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) { |
| 170 | unsigned long numAreaBlocks; |
| 171 | struct IoHriMainStoreArea4 *currentArea; |
| 172 | |
| 173 | if (existsBits & 0x80000000) { |
| 174 | unsigned long block_num; |
| 175 | |
| 176 | currentArea = &msVpdP->msAreaArray[area_num]; |
| 177 | numAreaBlocks = currentArea->numAdrRangeBlocks; |
| 178 | printk("ms_vpd: processing area %2ld blocks=%ld", |
| 179 | area_num, numAreaBlocks); |
| 180 | for (block_num = 0; block_num < numAreaBlocks; |
| 181 | ++block_num ) { |
| 182 | /* Process an address range block */ |
| 183 | struct MemoryBlock tempBlock; |
| 184 | unsigned long i; |
| 185 | |
| 186 | tempBlock.absStart = |
| 187 | (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart; |
| 188 | tempBlock.absEnd = |
| 189 | (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd; |
| 190 | tempBlock.logicalStart = 0; |
| 191 | tempBlock.logicalEnd = 0; |
| 192 | printk("\n block %ld absStart=%016lx absEnd=%016lx", |
| 193 | block_num, tempBlock.absStart, |
| 194 | tempBlock.absEnd); |
| 195 | |
| 196 | for (i = 0; i < numSegmentBlocks; ++i) { |
| 197 | if (mb_array[i].absStart == |
| 198 | tempBlock.absStart) |
| 199 | break; |
| 200 | } |
| 201 | if (i == numSegmentBlocks) { |
| 202 | if (numSegmentBlocks == max_entries) |
| 203 | panic("iSeries_process_mainstore_vpd: too many memory blocks"); |
| 204 | mb_array[numSegmentBlocks] = tempBlock; |
| 205 | ++numSegmentBlocks; |
| 206 | } else |
| 207 | printk(" (duplicate)"); |
| 208 | } |
| 209 | printk("\n"); |
| 210 | } |
| 211 | existsBits <<= 1; |
| 212 | } |
| 213 | /* Now sort the blocks found into ascending sequence */ |
| 214 | if (numSegmentBlocks > 1) { |
| 215 | unsigned long m, n; |
| 216 | |
| 217 | for (m = 0; m < numSegmentBlocks - 1; ++m) { |
| 218 | for (n = numSegmentBlocks - 1; m < n; --n) { |
| 219 | if (mb_array[n].absStart < |
| 220 | mb_array[n-1].absStart) { |
| 221 | struct MemoryBlock tempBlock; |
| 222 | |
| 223 | tempBlock = mb_array[n]; |
| 224 | mb_array[n] = mb_array[n-1]; |
| 225 | mb_array[n-1] = tempBlock; |
| 226 | } |
| 227 | } |
| 228 | } |
| 229 | } |
| 230 | /* |
| 231 | * Assign "logical" addresses to each block. These |
| 232 | * addresses correspond to the hypervisor "bitmap" space. |
| 233 | * Convert all addresses into units of 256K chunks. |
| 234 | */ |
| 235 | { |
| 236 | unsigned long i, nextBitmapAddress; |
| 237 | |
| 238 | printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks); |
| 239 | nextBitmapAddress = 0; |
| 240 | for (i = 0; i < numSegmentBlocks; ++i) { |
| 241 | unsigned long length = mb_array[i].absEnd - |
| 242 | mb_array[i].absStart; |
| 243 | |
| 244 | mb_array[i].logicalStart = nextBitmapAddress; |
| 245 | mb_array[i].logicalEnd = nextBitmapAddress + length; |
| 246 | nextBitmapAddress += length; |
| 247 | printk(" Bitmap range: %016lx - %016lx\n" |
| 248 | " Absolute range: %016lx - %016lx\n", |
| 249 | mb_array[i].logicalStart, |
| 250 | mb_array[i].logicalEnd, |
| 251 | mb_array[i].absStart, mb_array[i].absEnd); |
| 252 | mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart & |
| 253 | 0x000fffffffffffff); |
| 254 | mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd & |
| 255 | 0x000fffffffffffff); |
| 256 | mb_array[i].logicalStart = |
| 257 | addr_to_chunk(mb_array[i].logicalStart); |
| 258 | mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd); |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | return numSegmentBlocks; |
| 263 | } |
| 264 | |
| 265 | static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array, |
| 266 | unsigned long max_entries) |
| 267 | { |
| 268 | unsigned long i; |
| 269 | unsigned long mem_blocks = 0; |
| 270 | |
| 271 | if (cpu_has_feature(CPU_FTR_SLB)) |
| 272 | mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array, |
| 273 | max_entries); |
| 274 | else |
| 275 | mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array, |
| 276 | max_entries); |
| 277 | |
| 278 | printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks); |
| 279 | for (i = 0; i < mem_blocks; ++i) { |
| 280 | printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n" |
| 281 | " abs chunks %016lx - %016lx\n", |
| 282 | i, mb_array[i].logicalStart, mb_array[i].logicalEnd, |
| 283 | mb_array[i].absStart, mb_array[i].absEnd); |
| 284 | } |
| 285 | return mem_blocks; |
| 286 | } |
| 287 | |
| 288 | static void __init iSeries_get_cmdline(void) |
| 289 | { |
| 290 | char *p, *q; |
| 291 | |
| 292 | /* copy the command line parameter from the primary VSP */ |
| 293 | HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256, |
| 294 | HvLpDma_Direction_RemoteToLocal); |
| 295 | |
| 296 | p = cmd_line; |
| 297 | q = cmd_line + 255; |
| 298 | while(p < q) { |
| 299 | if (!*p || *p == '\n') |
| 300 | break; |
| 301 | ++p; |
| 302 | } |
| 303 | *p = 0; |
| 304 | } |
| 305 | |
| 306 | static void __init iSeries_init_early(void) |
| 307 | { |
| 308 | extern unsigned long memory_limit; |
| 309 | |
| 310 | DBG(" -> iSeries_init_early()\n"); |
| 311 | |
| 312 | ppcdbg_initialize(); |
| 313 | |
| 314 | #if defined(CONFIG_BLK_DEV_INITRD) |
| 315 | /* |
| 316 | * If the init RAM disk has been configured and there is |
| 317 | * a non-zero starting address for it, set it up |
| 318 | */ |
| 319 | if (naca.xRamDisk) { |
| 320 | initrd_start = (unsigned long)__va(naca.xRamDisk); |
| 321 | initrd_end = initrd_start + naca.xRamDiskSize * PAGE_SIZE; |
| 322 | initrd_below_start_ok = 1; // ramdisk in kernel space |
| 323 | ROOT_DEV = Root_RAM0; |
| 324 | if (((rd_size * 1024) / PAGE_SIZE) < naca.xRamDiskSize) |
| 325 | rd_size = (naca.xRamDiskSize * PAGE_SIZE) / 1024; |
| 326 | } else |
| 327 | #endif /* CONFIG_BLK_DEV_INITRD */ |
| 328 | { |
| 329 | /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */ |
| 330 | } |
| 331 | |
| 332 | iSeries_recal_tb = get_tb(); |
| 333 | iSeries_recal_titan = HvCallXm_loadTod(); |
| 334 | |
| 335 | /* |
| 336 | * Cache sizes must be initialized before hpte_init_iSeries is called |
| 337 | * as the later need them for flush_icache_range() |
| 338 | */ |
| 339 | setup_iSeries_cache_sizes(); |
| 340 | |
| 341 | /* |
| 342 | * Initialize the hash table management pointers |
| 343 | */ |
| 344 | hpte_init_iSeries(); |
| 345 | |
| 346 | /* |
| 347 | * Initialize the DMA/TCE management |
| 348 | */ |
| 349 | iommu_init_early_iSeries(); |
| 350 | |
| 351 | /* |
| 352 | * Initialize the table which translate Linux physical addresses to |
| 353 | * AS/400 absolute addresses |
| 354 | */ |
| 355 | build_iSeries_Memory_Map(); |
| 356 | |
| 357 | iSeries_get_cmdline(); |
| 358 | |
| 359 | /* Save unparsed command line copy for /proc/cmdline */ |
| 360 | strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE); |
| 361 | |
| 362 | /* Parse early parameters, in particular mem=x */ |
| 363 | parse_early_param(); |
| 364 | |
| 365 | if (memory_limit) { |
| 366 | if (memory_limit < systemcfg->physicalMemorySize) |
| 367 | systemcfg->physicalMemorySize = memory_limit; |
| 368 | else { |
| 369 | printk("Ignoring mem=%lu >= ram_top.\n", memory_limit); |
| 370 | memory_limit = 0; |
| 371 | } |
| 372 | } |
| 373 | |
| 374 | /* Bolt kernel mappings for all of memory (or just a bit if we've got a limit) */ |
| 375 | iSeries_bolt_kernel(0, systemcfg->physicalMemorySize); |
| 376 | |
| 377 | lmb_init(); |
| 378 | lmb_add(0, systemcfg->physicalMemorySize); |
| 379 | lmb_analyze(); |
| 380 | lmb_reserve(0, __pa(klimit)); |
| 381 | |
| 382 | /* Initialize machine-dependency vectors */ |
| 383 | #ifdef CONFIG_SMP |
| 384 | smp_init_iSeries(); |
| 385 | #endif |
| 386 | if (itLpNaca.xPirEnvironMode == 0) |
| 387 | piranha_simulator = 1; |
| 388 | |
| 389 | /* Associate Lp Event Queue 0 with processor 0 */ |
| 390 | HvCallEvent_setLpEventQueueInterruptProc(0, 0); |
| 391 | |
| 392 | mf_init(); |
| 393 | mf_initialized = 1; |
| 394 | mb(); |
| 395 | |
| 396 | /* If we were passed an initrd, set the ROOT_DEV properly if the values |
| 397 | * look sensible. If not, clear initrd reference. |
| 398 | */ |
| 399 | #ifdef CONFIG_BLK_DEV_INITRD |
| 400 | if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE && |
| 401 | initrd_end > initrd_start) |
| 402 | ROOT_DEV = Root_RAM0; |
| 403 | else |
| 404 | initrd_start = initrd_end = 0; |
| 405 | #endif /* CONFIG_BLK_DEV_INITRD */ |
| 406 | |
| 407 | DBG(" <- iSeries_init_early()\n"); |
| 408 | } |
| 409 | |
| 410 | /* |
| 411 | * The iSeries may have very large memories ( > 128 GB ) and a partition |
| 412 | * may get memory in "chunks" that may be anywhere in the 2**52 real |
| 413 | * address space. The chunks are 256K in size. To map this to the |
| 414 | * memory model Linux expects, the AS/400 specific code builds a |
| 415 | * translation table to translate what Linux thinks are "physical" |
| 416 | * addresses to the actual real addresses. This allows us to make |
| 417 | * it appear to Linux that we have contiguous memory starting at |
| 418 | * physical address zero while in fact this could be far from the truth. |
| 419 | * To avoid confusion, I'll let the words physical and/or real address |
| 420 | * apply to the Linux addresses while I'll use "absolute address" to |
| 421 | * refer to the actual hardware real address. |
| 422 | * |
| 423 | * build_iSeries_Memory_Map gets information from the Hypervisor and |
| 424 | * looks at the Main Store VPD to determine the absolute addresses |
| 425 | * of the memory that has been assigned to our partition and builds |
| 426 | * a table used to translate Linux's physical addresses to these |
| 427 | * absolute addresses. Absolute addresses are needed when |
| 428 | * communicating with the hypervisor (e.g. to build HPT entries) |
| 429 | */ |
| 430 | |
| 431 | static void __init build_iSeries_Memory_Map(void) |
| 432 | { |
| 433 | u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize; |
| 434 | u32 nextPhysChunk; |
| 435 | u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages; |
| 436 | u32 num_ptegs; |
| 437 | u32 totalChunks,moreChunks; |
| 438 | u32 currChunk, thisChunk, absChunk; |
| 439 | u32 currDword; |
| 440 | u32 chunkBit; |
| 441 | u64 map; |
| 442 | struct MemoryBlock mb[32]; |
| 443 | unsigned long numMemoryBlocks, curBlock; |
| 444 | |
| 445 | /* Chunk size on iSeries is 256K bytes */ |
| 446 | totalChunks = (u32)HvLpConfig_getMsChunks(); |
| 447 | klimit = msChunks_alloc(klimit, totalChunks, 1UL << 18); |
| 448 | |
| 449 | /* |
| 450 | * Get absolute address of our load area |
| 451 | * and map it to physical address 0 |
| 452 | * This guarantees that the loadarea ends up at physical 0 |
| 453 | * otherwise, it might not be returned by PLIC as the first |
| 454 | * chunks |
| 455 | */ |
| 456 | |
| 457 | loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr); |
| 458 | loadAreaSize = itLpNaca.xLoadAreaChunks; |
| 459 | |
| 460 | /* |
| 461 | * Only add the pages already mapped here. |
| 462 | * Otherwise we might add the hpt pages |
| 463 | * The rest of the pages of the load area |
| 464 | * aren't in the HPT yet and can still |
| 465 | * be assigned an arbitrary physical address |
| 466 | */ |
| 467 | if ((loadAreaSize * 64) > HvPagesToMap) |
| 468 | loadAreaSize = HvPagesToMap / 64; |
| 469 | |
| 470 | loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1; |
| 471 | |
| 472 | /* |
| 473 | * TODO Do we need to do something if the HPT is in the 64MB load area? |
| 474 | * This would be required if the itLpNaca.xLoadAreaChunks includes |
| 475 | * the HPT size |
| 476 | */ |
| 477 | |
| 478 | printk("Mapping load area - physical addr = 0000000000000000\n" |
| 479 | " absolute addr = %016lx\n", |
| 480 | chunk_to_addr(loadAreaFirstChunk)); |
| 481 | printk("Load area size %dK\n", loadAreaSize * 256); |
| 482 | |
| 483 | for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk) |
| 484 | msChunks.abs[nextPhysChunk] = |
| 485 | loadAreaFirstChunk + nextPhysChunk; |
| 486 | |
| 487 | /* |
| 488 | * Get absolute address of our HPT and remember it so |
| 489 | * we won't map it to any physical address |
| 490 | */ |
| 491 | hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress()); |
| 492 | hptSizePages = (u32)HvCallHpt_getHptPages(); |
| 493 | hptSizeChunks = hptSizePages >> (msChunks.chunk_shift - PAGE_SHIFT); |
| 494 | hptLastChunk = hptFirstChunk + hptSizeChunks - 1; |
| 495 | |
| 496 | printk("HPT absolute addr = %016lx, size = %dK\n", |
| 497 | chunk_to_addr(hptFirstChunk), hptSizeChunks * 256); |
| 498 | |
| 499 | /* Fill in the hashed page table hash mask */ |
| 500 | num_ptegs = hptSizePages * |
| 501 | (PAGE_SIZE / (sizeof(HPTE) * HPTES_PER_GROUP)); |
| 502 | htab_hash_mask = num_ptegs - 1; |
| 503 | |
| 504 | /* |
| 505 | * The actual hashed page table is in the hypervisor, |
| 506 | * we have no direct access |
| 507 | */ |
| 508 | htab_address = NULL; |
| 509 | |
| 510 | /* |
| 511 | * Determine if absolute memory has any |
| 512 | * holes so that we can interpret the |
| 513 | * access map we get back from the hypervisor |
| 514 | * correctly. |
| 515 | */ |
| 516 | numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32); |
| 517 | |
| 518 | /* |
| 519 | * Process the main store access map from the hypervisor |
| 520 | * to build up our physical -> absolute translation table |
| 521 | */ |
| 522 | curBlock = 0; |
| 523 | currChunk = 0; |
| 524 | currDword = 0; |
| 525 | moreChunks = totalChunks; |
| 526 | |
| 527 | while (moreChunks) { |
| 528 | map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex, |
| 529 | currDword); |
| 530 | thisChunk = currChunk; |
| 531 | while (map) { |
| 532 | chunkBit = map >> 63; |
| 533 | map <<= 1; |
| 534 | if (chunkBit) { |
| 535 | --moreChunks; |
| 536 | while (thisChunk >= mb[curBlock].logicalEnd) { |
| 537 | ++curBlock; |
| 538 | if (curBlock >= numMemoryBlocks) |
| 539 | panic("out of memory blocks"); |
| 540 | } |
| 541 | if (thisChunk < mb[curBlock].logicalStart) |
| 542 | panic("memory block error"); |
| 543 | |
| 544 | absChunk = mb[curBlock].absStart + |
| 545 | (thisChunk - mb[curBlock].logicalStart); |
| 546 | if (((absChunk < hptFirstChunk) || |
| 547 | (absChunk > hptLastChunk)) && |
| 548 | ((absChunk < loadAreaFirstChunk) || |
| 549 | (absChunk > loadAreaLastChunk))) { |
| 550 | msChunks.abs[nextPhysChunk] = absChunk; |
| 551 | ++nextPhysChunk; |
| 552 | } |
| 553 | } |
| 554 | ++thisChunk; |
| 555 | } |
| 556 | ++currDword; |
| 557 | currChunk += 64; |
| 558 | } |
| 559 | |
| 560 | /* |
| 561 | * main store size (in chunks) is |
| 562 | * totalChunks - hptSizeChunks |
| 563 | * which should be equal to |
| 564 | * nextPhysChunk |
| 565 | */ |
| 566 | systemcfg->physicalMemorySize = chunk_to_addr(nextPhysChunk); |
| 567 | } |
| 568 | |
| 569 | /* |
| 570 | * Set up the variables that describe the cache line sizes |
| 571 | * for this machine. |
| 572 | */ |
| 573 | static void __init setup_iSeries_cache_sizes(void) |
| 574 | { |
| 575 | unsigned int i, n; |
| 576 | unsigned int procIx = get_paca()->lppaca.dyn_hv_phys_proc_index; |
| 577 | |
| 578 | systemcfg->icache_size = |
| 579 | ppc64_caches.isize = xIoHriProcessorVpd[procIx].xInstCacheSize * 1024; |
| 580 | systemcfg->icache_line_size = |
| 581 | ppc64_caches.iline_size = |
| 582 | xIoHriProcessorVpd[procIx].xInstCacheOperandSize; |
| 583 | systemcfg->dcache_size = |
| 584 | ppc64_caches.dsize = |
| 585 | xIoHriProcessorVpd[procIx].xDataL1CacheSizeKB * 1024; |
| 586 | systemcfg->dcache_line_size = |
| 587 | ppc64_caches.dline_size = |
| 588 | xIoHriProcessorVpd[procIx].xDataCacheOperandSize; |
| 589 | ppc64_caches.ilines_per_page = PAGE_SIZE / ppc64_caches.iline_size; |
| 590 | ppc64_caches.dlines_per_page = PAGE_SIZE / ppc64_caches.dline_size; |
| 591 | |
| 592 | i = ppc64_caches.iline_size; |
| 593 | n = 0; |
| 594 | while ((i = (i / 2))) |
| 595 | ++n; |
| 596 | ppc64_caches.log_iline_size = n; |
| 597 | |
| 598 | i = ppc64_caches.dline_size; |
| 599 | n = 0; |
| 600 | while ((i = (i / 2))) |
| 601 | ++n; |
| 602 | ppc64_caches.log_dline_size = n; |
| 603 | |
| 604 | printk("D-cache line size = %d\n", |
| 605 | (unsigned int)ppc64_caches.dline_size); |
| 606 | printk("I-cache line size = %d\n", |
| 607 | (unsigned int)ppc64_caches.iline_size); |
| 608 | } |
| 609 | |
| 610 | /* |
| 611 | * Create a pte. Used during initialization only. |
| 612 | */ |
| 613 | static void iSeries_make_pte(unsigned long va, unsigned long pa, |
| 614 | int mode) |
| 615 | { |
| 616 | HPTE local_hpte, rhpte; |
| 617 | unsigned long hash, vpn; |
| 618 | long slot; |
| 619 | |
| 620 | vpn = va >> PAGE_SHIFT; |
| 621 | hash = hpt_hash(vpn, 0); |
| 622 | |
| 623 | local_hpte.dw1.dword1 = pa | mode; |
| 624 | local_hpte.dw0.dword0 = 0; |
| 625 | local_hpte.dw0.dw0.avpn = va >> 23; |
| 626 | local_hpte.dw0.dw0.bolted = 1; /* bolted */ |
| 627 | local_hpte.dw0.dw0.v = 1; |
| 628 | |
| 629 | slot = HvCallHpt_findValid(&rhpte, vpn); |
| 630 | if (slot < 0) { |
| 631 | /* Must find space in primary group */ |
| 632 | panic("hash_page: hpte already exists\n"); |
| 633 | } |
| 634 | HvCallHpt_addValidate(slot, 0, (HPTE *)&local_hpte ); |
| 635 | } |
| 636 | |
| 637 | /* |
| 638 | * Bolt the kernel addr space into the HPT |
| 639 | */ |
| 640 | static void __init iSeries_bolt_kernel(unsigned long saddr, unsigned long eaddr) |
| 641 | { |
| 642 | unsigned long pa; |
| 643 | unsigned long mode_rw = _PAGE_ACCESSED | _PAGE_COHERENT | PP_RWXX; |
| 644 | HPTE hpte; |
| 645 | |
| 646 | for (pa = saddr; pa < eaddr ;pa += PAGE_SIZE) { |
| 647 | unsigned long ea = (unsigned long)__va(pa); |
| 648 | unsigned long vsid = get_kernel_vsid(ea); |
| 649 | unsigned long va = (vsid << 28) | (pa & 0xfffffff); |
| 650 | unsigned long vpn = va >> PAGE_SHIFT; |
| 651 | unsigned long slot = HvCallHpt_findValid(&hpte, vpn); |
| 652 | |
| 653 | /* Make non-kernel text non-executable */ |
| 654 | if (!in_kernel_text(ea)) |
| 655 | mode_rw |= HW_NO_EXEC; |
| 656 | |
| 657 | if (hpte.dw0.dw0.v) { |
| 658 | /* HPTE exists, so just bolt it */ |
| 659 | HvCallHpt_setSwBits(slot, 0x10, 0); |
| 660 | /* And make sure the pp bits are correct */ |
| 661 | HvCallHpt_setPp(slot, PP_RWXX); |
| 662 | } else |
| 663 | /* No HPTE exists, so create a new bolted one */ |
| 664 | iSeries_make_pte(va, phys_to_abs(pa), mode_rw); |
| 665 | } |
| 666 | } |
| 667 | |
| 668 | extern unsigned long ppc_proc_freq; |
| 669 | extern unsigned long ppc_tb_freq; |
| 670 | |
| 671 | /* |
| 672 | * Document me. |
| 673 | */ |
| 674 | static void __init iSeries_setup_arch(void) |
| 675 | { |
| 676 | void *eventStack; |
| 677 | unsigned procIx = get_paca()->lppaca.dyn_hv_phys_proc_index; |
| 678 | |
| 679 | /* Add an eye catcher and the systemcfg layout version number */ |
| 680 | strcpy(systemcfg->eye_catcher, "SYSTEMCFG:PPC64"); |
| 681 | systemcfg->version.major = SYSTEMCFG_MAJOR; |
| 682 | systemcfg->version.minor = SYSTEMCFG_MINOR; |
| 683 | |
| 684 | /* Setup the Lp Event Queue */ |
| 685 | |
| 686 | /* Allocate a page for the Event Stack |
| 687 | * The hypervisor wants the absolute real address, so |
| 688 | * we subtract out the KERNELBASE and add in the |
| 689 | * absolute real address of the kernel load area |
| 690 | */ |
| 691 | eventStack = alloc_bootmem_pages(LpEventStackSize); |
| 692 | memset(eventStack, 0, LpEventStackSize); |
| 693 | |
| 694 | /* Invoke the hypervisor to initialize the event stack */ |
| 695 | HvCallEvent_setLpEventStack(0, eventStack, LpEventStackSize); |
| 696 | |
| 697 | /* Initialize fields in our Lp Event Queue */ |
| 698 | xItLpQueue.xSlicEventStackPtr = (char *)eventStack; |
| 699 | xItLpQueue.xSlicCurEventPtr = (char *)eventStack; |
| 700 | xItLpQueue.xSlicLastValidEventPtr = (char *)eventStack + |
| 701 | (LpEventStackSize - LpEventMaxSize); |
| 702 | xItLpQueue.xIndex = 0; |
| 703 | |
| 704 | /* Compute processor frequency */ |
| 705 | procFreqHz = ((1UL << 34) * 1000000) / |
| 706 | xIoHriProcessorVpd[procIx].xProcFreq; |
| 707 | procFreqMhz = procFreqHz / 1000000; |
| 708 | procFreqMhzHundreths = (procFreqHz / 10000) - (procFreqMhz * 100); |
| 709 | ppc_proc_freq = procFreqHz; |
| 710 | |
| 711 | /* Compute time base frequency */ |
| 712 | tbFreqHz = ((1UL << 32) * 1000000) / |
| 713 | xIoHriProcessorVpd[procIx].xTimeBaseFreq; |
| 714 | tbFreqMhz = tbFreqHz / 1000000; |
| 715 | tbFreqMhzHundreths = (tbFreqHz / 10000) - (tbFreqMhz * 100); |
| 716 | ppc_tb_freq = tbFreqHz; |
| 717 | |
| 718 | printk("Max logical processors = %d\n", |
| 719 | itVpdAreas.xSlicMaxLogicalProcs); |
| 720 | printk("Max physical processors = %d\n", |
| 721 | itVpdAreas.xSlicMaxPhysicalProcs); |
| 722 | printk("Processor frequency = %lu.%02lu\n", procFreqMhz, |
| 723 | procFreqMhzHundreths); |
| 724 | printk("Time base frequency = %lu.%02lu\n", tbFreqMhz, |
| 725 | tbFreqMhzHundreths); |
| 726 | systemcfg->processor = xIoHriProcessorVpd[procIx].xPVR; |
| 727 | printk("Processor version = %x\n", systemcfg->processor); |
| 728 | } |
| 729 | |
| 730 | static void iSeries_get_cpuinfo(struct seq_file *m) |
| 731 | { |
| 732 | seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n"); |
| 733 | } |
| 734 | |
| 735 | /* |
| 736 | * Document me. |
| 737 | * and Implement me. |
| 738 | */ |
| 739 | static int iSeries_get_irq(struct pt_regs *regs) |
| 740 | { |
| 741 | /* -2 means ignore this interrupt */ |
| 742 | return -2; |
| 743 | } |
| 744 | |
| 745 | /* |
| 746 | * Document me. |
| 747 | */ |
| 748 | static void iSeries_restart(char *cmd) |
| 749 | { |
| 750 | mf_reboot(); |
| 751 | } |
| 752 | |
| 753 | /* |
| 754 | * Document me. |
| 755 | */ |
| 756 | static void iSeries_power_off(void) |
| 757 | { |
| 758 | mf_power_off(); |
| 759 | } |
| 760 | |
| 761 | /* |
| 762 | * Document me. |
| 763 | */ |
| 764 | static void iSeries_halt(void) |
| 765 | { |
| 766 | mf_power_off(); |
| 767 | } |
| 768 | |
| 769 | extern void setup_default_decr(void); |
| 770 | |
| 771 | /* |
| 772 | * void __init iSeries_calibrate_decr() |
| 773 | * |
| 774 | * Description: |
| 775 | * This routine retrieves the internal processor frequency from the VPD, |
| 776 | * and sets up the kernel timer decrementer based on that value. |
| 777 | * |
| 778 | */ |
| 779 | static void __init iSeries_calibrate_decr(void) |
| 780 | { |
| 781 | unsigned long cyclesPerUsec; |
| 782 | struct div_result divres; |
| 783 | |
| 784 | /* Compute decrementer (and TB) frequency in cycles/sec */ |
| 785 | cyclesPerUsec = ppc_tb_freq / 1000000; |
| 786 | |
| 787 | /* |
| 788 | * Set the amount to refresh the decrementer by. This |
| 789 | * is the number of decrementer ticks it takes for |
| 790 | * 1/HZ seconds. |
| 791 | */ |
| 792 | tb_ticks_per_jiffy = ppc_tb_freq / HZ; |
| 793 | |
| 794 | #if 0 |
| 795 | /* TEST CODE FOR ADJTIME */ |
| 796 | tb_ticks_per_jiffy += tb_ticks_per_jiffy / 5000; |
| 797 | /* END OF TEST CODE */ |
| 798 | #endif |
| 799 | |
| 800 | /* |
| 801 | * tb_ticks_per_sec = freq; would give better accuracy |
| 802 | * but tb_ticks_per_sec = tb_ticks_per_jiffy*HZ; assures |
| 803 | * that jiffies (and xtime) will match the time returned |
| 804 | * by do_gettimeofday. |
| 805 | */ |
| 806 | tb_ticks_per_sec = tb_ticks_per_jiffy * HZ; |
| 807 | tb_ticks_per_usec = cyclesPerUsec; |
| 808 | tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000); |
| 809 | div128_by_32(1024 * 1024, 0, tb_ticks_per_sec, &divres); |
| 810 | tb_to_xs = divres.result_low; |
| 811 | setup_default_decr(); |
| 812 | } |
| 813 | |
| 814 | static void __init iSeries_progress(char * st, unsigned short code) |
| 815 | { |
| 816 | printk("Progress: [%04x] - %s\n", (unsigned)code, st); |
| 817 | if (!piranha_simulator && mf_initialized) { |
| 818 | if (code != 0xffff) |
| 819 | mf_display_progress(code); |
| 820 | else |
| 821 | mf_clear_src(); |
| 822 | } |
| 823 | } |
| 824 | |
| 825 | static void __init iSeries_fixup_klimit(void) |
| 826 | { |
| 827 | /* |
| 828 | * Change klimit to take into account any ram disk |
| 829 | * that may be included |
| 830 | */ |
| 831 | if (naca.xRamDisk) |
| 832 | klimit = KERNELBASE + (u64)naca.xRamDisk + |
| 833 | (naca.xRamDiskSize * PAGE_SIZE); |
| 834 | else { |
| 835 | /* |
| 836 | * No ram disk was included - check and see if there |
| 837 | * was an embedded system map. Change klimit to take |
| 838 | * into account any embedded system map |
| 839 | */ |
| 840 | if (embedded_sysmap_end) |
| 841 | klimit = KERNELBASE + ((embedded_sysmap_end + 4095) & |
| 842 | 0xfffffffffffff000); |
| 843 | } |
| 844 | } |
| 845 | |
| 846 | static int __init iSeries_src_init(void) |
| 847 | { |
| 848 | /* clear the progress line */ |
| 849 | ppc_md.progress(" ", 0xffff); |
| 850 | return 0; |
| 851 | } |
| 852 | |
| 853 | late_initcall(iSeries_src_init); |
| 854 | |
| 855 | void __init iSeries_early_setup(void) |
| 856 | { |
| 857 | iSeries_fixup_klimit(); |
| 858 | |
| 859 | ppc_md.setup_arch = iSeries_setup_arch; |
| 860 | ppc_md.get_cpuinfo = iSeries_get_cpuinfo; |
| 861 | ppc_md.init_IRQ = iSeries_init_IRQ; |
| 862 | ppc_md.get_irq = iSeries_get_irq; |
| 863 | ppc_md.init_early = iSeries_init_early, |
| 864 | |
| 865 | ppc_md.pcibios_fixup = iSeries_pci_final_fixup; |
| 866 | |
| 867 | ppc_md.restart = iSeries_restart; |
| 868 | ppc_md.power_off = iSeries_power_off; |
| 869 | ppc_md.halt = iSeries_halt; |
| 870 | |
| 871 | ppc_md.get_boot_time = iSeries_get_boot_time; |
| 872 | ppc_md.set_rtc_time = iSeries_set_rtc_time; |
| 873 | ppc_md.get_rtc_time = iSeries_get_rtc_time; |
| 874 | ppc_md.calibrate_decr = iSeries_calibrate_decr; |
| 875 | ppc_md.progress = iSeries_progress; |
| 876 | } |
| 877 | |