Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
1da177e4 LT |
27 | #include <linux/initrd.h> |
28 | #include <linux/seq_file.h> | |
29 | #include <linux/kdev_t.h> | |
30 | #include <linux/major.h> | |
31 | #include <linux/root_dev.h> | |
32 | ||
33 | #include <asm/processor.h> | |
34 | #include <asm/machdep.h> | |
35 | #include <asm/page.h> | |
36 | #include <asm/mmu.h> | |
37 | #include <asm/pgtable.h> | |
38 | #include <asm/mmu_context.h> | |
39 | #include <asm/cputable.h> | |
40 | #include <asm/sections.h> | |
41 | #include <asm/iommu.h> | |
aed31351 | 42 | #include <asm/firmware.h> |
1da177e4 LT |
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> | |
0bc0ffd5 | 50 | #include <asm/abs_addr.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
58 | #include <asm/iSeries/mf.h> |
59 | #include <asm/iSeries/HvLpEvent.h> | |
60 | #include <asm/iSeries/iSeries_irq.h> | |
0bc0ffd5 SR |
61 | #include <asm/iSeries/IoHriProcessorVpd.h> |
62 | #include <asm/iSeries/ItVpdAreas.h> | |
63 | #include <asm/iSeries/LparMap.h> | |
1da177e4 LT |
64 | |
65 | extern void hvlog(char *fmt, ...); | |
66 | ||
67 | #ifdef DEBUG | |
68 | #define DBG(fmt...) hvlog(fmt) | |
69 | #else | |
70 | #define DBG(fmt...) | |
71 | #endif | |
72 | ||
73 | /* Function Prototypes */ | |
74 | extern void ppcdbg_initialize(void); | |
75 | ||
76 | static void build_iSeries_Memory_Map(void); | |
77 | static void setup_iSeries_cache_sizes(void); | |
78 | static void iSeries_bolt_kernel(unsigned long saddr, unsigned long eaddr); | |
145d01e4 | 79 | #ifdef CONFIG_PCI |
1da177e4 | 80 | extern void iSeries_pci_final_fixup(void); |
145d01e4 SR |
81 | #else |
82 | static void iSeries_pci_final_fixup(void) { } | |
83 | #endif | |
1da177e4 LT |
84 | |
85 | /* Global Variables */ | |
86 | static unsigned long procFreqHz; | |
87 | static unsigned long procFreqMhz; | |
88 | static unsigned long procFreqMhzHundreths; | |
89 | ||
90 | static unsigned long tbFreqHz; | |
91 | static unsigned long tbFreqMhz; | |
92 | static unsigned long tbFreqMhzHundreths; | |
93 | ||
94 | int piranha_simulator; | |
95 | ||
96 | extern int rd_size; /* Defined in drivers/block/rd.c */ | |
97 | extern unsigned long klimit; | |
98 | extern unsigned long embedded_sysmap_start; | |
99 | extern unsigned long embedded_sysmap_end; | |
100 | ||
101 | extern unsigned long iSeries_recal_tb; | |
102 | extern unsigned long iSeries_recal_titan; | |
103 | ||
104 | static int mf_initialized; | |
105 | ||
106 | struct MemoryBlock { | |
107 | unsigned long absStart; | |
108 | unsigned long absEnd; | |
109 | unsigned long logicalStart; | |
110 | unsigned long logicalEnd; | |
111 | }; | |
112 | ||
113 | /* | |
114 | * Process the main store vpd to determine where the holes in memory are | |
115 | * and return the number of physical blocks and fill in the array of | |
116 | * block data. | |
117 | */ | |
118 | static unsigned long iSeries_process_Condor_mainstore_vpd( | |
119 | struct MemoryBlock *mb_array, unsigned long max_entries) | |
120 | { | |
121 | unsigned long holeFirstChunk, holeSizeChunks; | |
122 | unsigned long numMemoryBlocks = 1; | |
123 | struct IoHriMainStoreSegment4 *msVpd = | |
124 | (struct IoHriMainStoreSegment4 *)xMsVpd; | |
125 | unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr; | |
126 | unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr; | |
127 | unsigned long holeSize = holeEnd - holeStart; | |
128 | ||
129 | printk("Mainstore_VPD: Condor\n"); | |
130 | /* | |
131 | * Determine if absolute memory has any | |
132 | * holes so that we can interpret the | |
133 | * access map we get back from the hypervisor | |
134 | * correctly. | |
135 | */ | |
136 | mb_array[0].logicalStart = 0; | |
137 | mb_array[0].logicalEnd = 0x100000000; | |
138 | mb_array[0].absStart = 0; | |
139 | mb_array[0].absEnd = 0x100000000; | |
140 | ||
141 | if (holeSize) { | |
142 | numMemoryBlocks = 2; | |
143 | holeStart = holeStart & 0x000fffffffffffff; | |
144 | holeStart = addr_to_chunk(holeStart); | |
145 | holeFirstChunk = holeStart; | |
146 | holeSize = addr_to_chunk(holeSize); | |
147 | holeSizeChunks = holeSize; | |
148 | printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n", | |
149 | holeFirstChunk, holeSizeChunks ); | |
150 | mb_array[0].logicalEnd = holeFirstChunk; | |
151 | mb_array[0].absEnd = holeFirstChunk; | |
152 | mb_array[1].logicalStart = holeFirstChunk; | |
153 | mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks; | |
154 | mb_array[1].absStart = holeFirstChunk + holeSizeChunks; | |
155 | mb_array[1].absEnd = 0x100000000; | |
156 | } | |
157 | return numMemoryBlocks; | |
158 | } | |
159 | ||
160 | #define MaxSegmentAreas 32 | |
161 | #define MaxSegmentAdrRangeBlocks 128 | |
162 | #define MaxAreaRangeBlocks 4 | |
163 | ||
164 | static unsigned long iSeries_process_Regatta_mainstore_vpd( | |
165 | struct MemoryBlock *mb_array, unsigned long max_entries) | |
166 | { | |
167 | struct IoHriMainStoreSegment5 *msVpdP = | |
168 | (struct IoHriMainStoreSegment5 *)xMsVpd; | |
169 | unsigned long numSegmentBlocks = 0; | |
170 | u32 existsBits = msVpdP->msAreaExists; | |
171 | unsigned long area_num; | |
172 | ||
173 | printk("Mainstore_VPD: Regatta\n"); | |
174 | ||
175 | for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) { | |
176 | unsigned long numAreaBlocks; | |
177 | struct IoHriMainStoreArea4 *currentArea; | |
178 | ||
179 | if (existsBits & 0x80000000) { | |
180 | unsigned long block_num; | |
181 | ||
182 | currentArea = &msVpdP->msAreaArray[area_num]; | |
183 | numAreaBlocks = currentArea->numAdrRangeBlocks; | |
184 | printk("ms_vpd: processing area %2ld blocks=%ld", | |
185 | area_num, numAreaBlocks); | |
186 | for (block_num = 0; block_num < numAreaBlocks; | |
187 | ++block_num ) { | |
188 | /* Process an address range block */ | |
189 | struct MemoryBlock tempBlock; | |
190 | unsigned long i; | |
191 | ||
192 | tempBlock.absStart = | |
193 | (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart; | |
194 | tempBlock.absEnd = | |
195 | (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd; | |
196 | tempBlock.logicalStart = 0; | |
197 | tempBlock.logicalEnd = 0; | |
198 | printk("\n block %ld absStart=%016lx absEnd=%016lx", | |
199 | block_num, tempBlock.absStart, | |
200 | tempBlock.absEnd); | |
201 | ||
202 | for (i = 0; i < numSegmentBlocks; ++i) { | |
203 | if (mb_array[i].absStart == | |
204 | tempBlock.absStart) | |
205 | break; | |
206 | } | |
207 | if (i == numSegmentBlocks) { | |
208 | if (numSegmentBlocks == max_entries) | |
209 | panic("iSeries_process_mainstore_vpd: too many memory blocks"); | |
210 | mb_array[numSegmentBlocks] = tempBlock; | |
211 | ++numSegmentBlocks; | |
212 | } else | |
213 | printk(" (duplicate)"); | |
214 | } | |
215 | printk("\n"); | |
216 | } | |
217 | existsBits <<= 1; | |
218 | } | |
219 | /* Now sort the blocks found into ascending sequence */ | |
220 | if (numSegmentBlocks > 1) { | |
221 | unsigned long m, n; | |
222 | ||
223 | for (m = 0; m < numSegmentBlocks - 1; ++m) { | |
224 | for (n = numSegmentBlocks - 1; m < n; --n) { | |
225 | if (mb_array[n].absStart < | |
226 | mb_array[n-1].absStart) { | |
227 | struct MemoryBlock tempBlock; | |
228 | ||
229 | tempBlock = mb_array[n]; | |
230 | mb_array[n] = mb_array[n-1]; | |
231 | mb_array[n-1] = tempBlock; | |
232 | } | |
233 | } | |
234 | } | |
235 | } | |
236 | /* | |
237 | * Assign "logical" addresses to each block. These | |
238 | * addresses correspond to the hypervisor "bitmap" space. | |
239 | * Convert all addresses into units of 256K chunks. | |
240 | */ | |
241 | { | |
242 | unsigned long i, nextBitmapAddress; | |
243 | ||
244 | printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks); | |
245 | nextBitmapAddress = 0; | |
246 | for (i = 0; i < numSegmentBlocks; ++i) { | |
247 | unsigned long length = mb_array[i].absEnd - | |
248 | mb_array[i].absStart; | |
249 | ||
250 | mb_array[i].logicalStart = nextBitmapAddress; | |
251 | mb_array[i].logicalEnd = nextBitmapAddress + length; | |
252 | nextBitmapAddress += length; | |
253 | printk(" Bitmap range: %016lx - %016lx\n" | |
254 | " Absolute range: %016lx - %016lx\n", | |
255 | mb_array[i].logicalStart, | |
256 | mb_array[i].logicalEnd, | |
257 | mb_array[i].absStart, mb_array[i].absEnd); | |
258 | mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart & | |
259 | 0x000fffffffffffff); | |
260 | mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd & | |
261 | 0x000fffffffffffff); | |
262 | mb_array[i].logicalStart = | |
263 | addr_to_chunk(mb_array[i].logicalStart); | |
264 | mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd); | |
265 | } | |
266 | } | |
267 | ||
268 | return numSegmentBlocks; | |
269 | } | |
270 | ||
271 | static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array, | |
272 | unsigned long max_entries) | |
273 | { | |
274 | unsigned long i; | |
275 | unsigned long mem_blocks = 0; | |
276 | ||
277 | if (cpu_has_feature(CPU_FTR_SLB)) | |
278 | mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array, | |
279 | max_entries); | |
280 | else | |
281 | mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array, | |
282 | max_entries); | |
283 | ||
284 | printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks); | |
285 | for (i = 0; i < mem_blocks; ++i) { | |
286 | printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n" | |
287 | " abs chunks %016lx - %016lx\n", | |
288 | i, mb_array[i].logicalStart, mb_array[i].logicalEnd, | |
289 | mb_array[i].absStart, mb_array[i].absEnd); | |
290 | } | |
291 | return mem_blocks; | |
292 | } | |
293 | ||
294 | static void __init iSeries_get_cmdline(void) | |
295 | { | |
296 | char *p, *q; | |
297 | ||
298 | /* copy the command line parameter from the primary VSP */ | |
299 | HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256, | |
300 | HvLpDma_Direction_RemoteToLocal); | |
301 | ||
302 | p = cmd_line; | |
303 | q = cmd_line + 255; | |
304 | while(p < q) { | |
305 | if (!*p || *p == '\n') | |
306 | break; | |
307 | ++p; | |
308 | } | |
309 | *p = 0; | |
310 | } | |
311 | ||
312 | static void __init iSeries_init_early(void) | |
313 | { | |
314 | extern unsigned long memory_limit; | |
315 | ||
316 | DBG(" -> iSeries_init_early()\n"); | |
317 | ||
aed31351 SR |
318 | ppc64_firmware_features = FW_FEATURE_ISERIES; |
319 | ||
1da177e4 LT |
320 | ppcdbg_initialize(); |
321 | ||
322 | #if defined(CONFIG_BLK_DEV_INITRD) | |
323 | /* | |
324 | * If the init RAM disk has been configured and there is | |
325 | * a non-zero starting address for it, set it up | |
326 | */ | |
327 | if (naca.xRamDisk) { | |
328 | initrd_start = (unsigned long)__va(naca.xRamDisk); | |
329 | initrd_end = initrd_start + naca.xRamDiskSize * PAGE_SIZE; | |
330 | initrd_below_start_ok = 1; // ramdisk in kernel space | |
331 | ROOT_DEV = Root_RAM0; | |
332 | if (((rd_size * 1024) / PAGE_SIZE) < naca.xRamDiskSize) | |
333 | rd_size = (naca.xRamDiskSize * PAGE_SIZE) / 1024; | |
334 | } else | |
335 | #endif /* CONFIG_BLK_DEV_INITRD */ | |
336 | { | |
337 | /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */ | |
338 | } | |
339 | ||
340 | iSeries_recal_tb = get_tb(); | |
341 | iSeries_recal_titan = HvCallXm_loadTod(); | |
342 | ||
343 | /* | |
344 | * Cache sizes must be initialized before hpte_init_iSeries is called | |
345 | * as the later need them for flush_icache_range() | |
346 | */ | |
347 | setup_iSeries_cache_sizes(); | |
348 | ||
349 | /* | |
350 | * Initialize the hash table management pointers | |
351 | */ | |
352 | hpte_init_iSeries(); | |
353 | ||
354 | /* | |
355 | * Initialize the DMA/TCE management | |
356 | */ | |
357 | iommu_init_early_iSeries(); | |
358 | ||
359 | /* | |
360 | * Initialize the table which translate Linux physical addresses to | |
361 | * AS/400 absolute addresses | |
362 | */ | |
363 | build_iSeries_Memory_Map(); | |
364 | ||
365 | iSeries_get_cmdline(); | |
366 | ||
367 | /* Save unparsed command line copy for /proc/cmdline */ | |
368 | strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE); | |
369 | ||
370 | /* Parse early parameters, in particular mem=x */ | |
371 | parse_early_param(); | |
372 | ||
373 | if (memory_limit) { | |
374 | if (memory_limit < systemcfg->physicalMemorySize) | |
375 | systemcfg->physicalMemorySize = memory_limit; | |
376 | else { | |
377 | printk("Ignoring mem=%lu >= ram_top.\n", memory_limit); | |
378 | memory_limit = 0; | |
379 | } | |
380 | } | |
381 | ||
382 | /* Bolt kernel mappings for all of memory (or just a bit if we've got a limit) */ | |
383 | iSeries_bolt_kernel(0, systemcfg->physicalMemorySize); | |
384 | ||
385 | lmb_init(); | |
386 | lmb_add(0, systemcfg->physicalMemorySize); | |
387 | lmb_analyze(); | |
388 | lmb_reserve(0, __pa(klimit)); | |
389 | ||
390 | /* Initialize machine-dependency vectors */ | |
391 | #ifdef CONFIG_SMP | |
392 | smp_init_iSeries(); | |
393 | #endif | |
394 | if (itLpNaca.xPirEnvironMode == 0) | |
395 | piranha_simulator = 1; | |
396 | ||
397 | /* Associate Lp Event Queue 0 with processor 0 */ | |
398 | HvCallEvent_setLpEventQueueInterruptProc(0, 0); | |
399 | ||
400 | mf_init(); | |
401 | mf_initialized = 1; | |
402 | mb(); | |
403 | ||
404 | /* If we were passed an initrd, set the ROOT_DEV properly if the values | |
405 | * look sensible. If not, clear initrd reference. | |
406 | */ | |
407 | #ifdef CONFIG_BLK_DEV_INITRD | |
408 | if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE && | |
409 | initrd_end > initrd_start) | |
410 | ROOT_DEV = Root_RAM0; | |
411 | else | |
412 | initrd_start = initrd_end = 0; | |
413 | #endif /* CONFIG_BLK_DEV_INITRD */ | |
414 | ||
415 | DBG(" <- iSeries_init_early()\n"); | |
416 | } | |
417 | ||
418 | /* | |
419 | * The iSeries may have very large memories ( > 128 GB ) and a partition | |
420 | * may get memory in "chunks" that may be anywhere in the 2**52 real | |
421 | * address space. The chunks are 256K in size. To map this to the | |
422 | * memory model Linux expects, the AS/400 specific code builds a | |
423 | * translation table to translate what Linux thinks are "physical" | |
424 | * addresses to the actual real addresses. This allows us to make | |
425 | * it appear to Linux that we have contiguous memory starting at | |
426 | * physical address zero while in fact this could be far from the truth. | |
427 | * To avoid confusion, I'll let the words physical and/or real address | |
428 | * apply to the Linux addresses while I'll use "absolute address" to | |
429 | * refer to the actual hardware real address. | |
430 | * | |
431 | * build_iSeries_Memory_Map gets information from the Hypervisor and | |
432 | * looks at the Main Store VPD to determine the absolute addresses | |
433 | * of the memory that has been assigned to our partition and builds | |
434 | * a table used to translate Linux's physical addresses to these | |
435 | * absolute addresses. Absolute addresses are needed when | |
436 | * communicating with the hypervisor (e.g. to build HPT entries) | |
437 | */ | |
438 | ||
439 | static void __init build_iSeries_Memory_Map(void) | |
440 | { | |
441 | u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize; | |
442 | u32 nextPhysChunk; | |
443 | u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages; | |
444 | u32 num_ptegs; | |
445 | u32 totalChunks,moreChunks; | |
446 | u32 currChunk, thisChunk, absChunk; | |
447 | u32 currDword; | |
448 | u32 chunkBit; | |
449 | u64 map; | |
450 | struct MemoryBlock mb[32]; | |
451 | unsigned long numMemoryBlocks, curBlock; | |
452 | ||
453 | /* Chunk size on iSeries is 256K bytes */ | |
454 | totalChunks = (u32)HvLpConfig_getMsChunks(); | |
455 | klimit = msChunks_alloc(klimit, totalChunks, 1UL << 18); | |
456 | ||
457 | /* | |
458 | * Get absolute address of our load area | |
459 | * and map it to physical address 0 | |
460 | * This guarantees that the loadarea ends up at physical 0 | |
461 | * otherwise, it might not be returned by PLIC as the first | |
462 | * chunks | |
463 | */ | |
464 | ||
465 | loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr); | |
466 | loadAreaSize = itLpNaca.xLoadAreaChunks; | |
467 | ||
468 | /* | |
469 | * Only add the pages already mapped here. | |
470 | * Otherwise we might add the hpt pages | |
471 | * The rest of the pages of the load area | |
472 | * aren't in the HPT yet and can still | |
473 | * be assigned an arbitrary physical address | |
474 | */ | |
475 | if ((loadAreaSize * 64) > HvPagesToMap) | |
476 | loadAreaSize = HvPagesToMap / 64; | |
477 | ||
478 | loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1; | |
479 | ||
480 | /* | |
481 | * TODO Do we need to do something if the HPT is in the 64MB load area? | |
482 | * This would be required if the itLpNaca.xLoadAreaChunks includes | |
483 | * the HPT size | |
484 | */ | |
485 | ||
486 | printk("Mapping load area - physical addr = 0000000000000000\n" | |
487 | " absolute addr = %016lx\n", | |
488 | chunk_to_addr(loadAreaFirstChunk)); | |
489 | printk("Load area size %dK\n", loadAreaSize * 256); | |
490 | ||
491 | for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk) | |
492 | msChunks.abs[nextPhysChunk] = | |
493 | loadAreaFirstChunk + nextPhysChunk; | |
494 | ||
495 | /* | |
496 | * Get absolute address of our HPT and remember it so | |
497 | * we won't map it to any physical address | |
498 | */ | |
499 | hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress()); | |
500 | hptSizePages = (u32)HvCallHpt_getHptPages(); | |
501 | hptSizeChunks = hptSizePages >> (msChunks.chunk_shift - PAGE_SHIFT); | |
502 | hptLastChunk = hptFirstChunk + hptSizeChunks - 1; | |
503 | ||
504 | printk("HPT absolute addr = %016lx, size = %dK\n", | |
505 | chunk_to_addr(hptFirstChunk), hptSizeChunks * 256); | |
506 | ||
507 | /* Fill in the hashed page table hash mask */ | |
508 | num_ptegs = hptSizePages * | |
96e28449 | 509 | (PAGE_SIZE / (sizeof(hpte_t) * HPTES_PER_GROUP)); |
1da177e4 LT |
510 | htab_hash_mask = num_ptegs - 1; |
511 | ||
512 | /* | |
513 | * The actual hashed page table is in the hypervisor, | |
514 | * we have no direct access | |
515 | */ | |
516 | htab_address = NULL; | |
517 | ||
518 | /* | |
519 | * Determine if absolute memory has any | |
520 | * holes so that we can interpret the | |
521 | * access map we get back from the hypervisor | |
522 | * correctly. | |
523 | */ | |
524 | numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32); | |
525 | ||
526 | /* | |
527 | * Process the main store access map from the hypervisor | |
528 | * to build up our physical -> absolute translation table | |
529 | */ | |
530 | curBlock = 0; | |
531 | currChunk = 0; | |
532 | currDword = 0; | |
533 | moreChunks = totalChunks; | |
534 | ||
535 | while (moreChunks) { | |
536 | map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex, | |
537 | currDword); | |
538 | thisChunk = currChunk; | |
539 | while (map) { | |
540 | chunkBit = map >> 63; | |
541 | map <<= 1; | |
542 | if (chunkBit) { | |
543 | --moreChunks; | |
544 | while (thisChunk >= mb[curBlock].logicalEnd) { | |
545 | ++curBlock; | |
546 | if (curBlock >= numMemoryBlocks) | |
547 | panic("out of memory blocks"); | |
548 | } | |
549 | if (thisChunk < mb[curBlock].logicalStart) | |
550 | panic("memory block error"); | |
551 | ||
552 | absChunk = mb[curBlock].absStart + | |
553 | (thisChunk - mb[curBlock].logicalStart); | |
554 | if (((absChunk < hptFirstChunk) || | |
555 | (absChunk > hptLastChunk)) && | |
556 | ((absChunk < loadAreaFirstChunk) || | |
557 | (absChunk > loadAreaLastChunk))) { | |
558 | msChunks.abs[nextPhysChunk] = absChunk; | |
559 | ++nextPhysChunk; | |
560 | } | |
561 | } | |
562 | ++thisChunk; | |
563 | } | |
564 | ++currDword; | |
565 | currChunk += 64; | |
566 | } | |
567 | ||
568 | /* | |
569 | * main store size (in chunks) is | |
570 | * totalChunks - hptSizeChunks | |
571 | * which should be equal to | |
572 | * nextPhysChunk | |
573 | */ | |
574 | systemcfg->physicalMemorySize = chunk_to_addr(nextPhysChunk); | |
575 | } | |
576 | ||
577 | /* | |
578 | * Set up the variables that describe the cache line sizes | |
579 | * for this machine. | |
580 | */ | |
581 | static void __init setup_iSeries_cache_sizes(void) | |
582 | { | |
583 | unsigned int i, n; | |
584 | unsigned int procIx = get_paca()->lppaca.dyn_hv_phys_proc_index; | |
585 | ||
586 | systemcfg->icache_size = | |
587 | ppc64_caches.isize = xIoHriProcessorVpd[procIx].xInstCacheSize * 1024; | |
588 | systemcfg->icache_line_size = | |
589 | ppc64_caches.iline_size = | |
590 | xIoHriProcessorVpd[procIx].xInstCacheOperandSize; | |
591 | systemcfg->dcache_size = | |
592 | ppc64_caches.dsize = | |
593 | xIoHriProcessorVpd[procIx].xDataL1CacheSizeKB * 1024; | |
594 | systemcfg->dcache_line_size = | |
595 | ppc64_caches.dline_size = | |
596 | xIoHriProcessorVpd[procIx].xDataCacheOperandSize; | |
597 | ppc64_caches.ilines_per_page = PAGE_SIZE / ppc64_caches.iline_size; | |
598 | ppc64_caches.dlines_per_page = PAGE_SIZE / ppc64_caches.dline_size; | |
599 | ||
600 | i = ppc64_caches.iline_size; | |
601 | n = 0; | |
602 | while ((i = (i / 2))) | |
603 | ++n; | |
604 | ppc64_caches.log_iline_size = n; | |
605 | ||
606 | i = ppc64_caches.dline_size; | |
607 | n = 0; | |
608 | while ((i = (i / 2))) | |
609 | ++n; | |
610 | ppc64_caches.log_dline_size = n; | |
611 | ||
612 | printk("D-cache line size = %d\n", | |
613 | (unsigned int)ppc64_caches.dline_size); | |
614 | printk("I-cache line size = %d\n", | |
615 | (unsigned int)ppc64_caches.iline_size); | |
616 | } | |
617 | ||
618 | /* | |
619 | * Create a pte. Used during initialization only. | |
620 | */ | |
621 | static void iSeries_make_pte(unsigned long va, unsigned long pa, | |
622 | int mode) | |
623 | { | |
96e28449 | 624 | hpte_t local_hpte, rhpte; |
1da177e4 LT |
625 | unsigned long hash, vpn; |
626 | long slot; | |
627 | ||
628 | vpn = va >> PAGE_SHIFT; | |
629 | hash = hpt_hash(vpn, 0); | |
630 | ||
96e28449 DG |
631 | local_hpte.r = pa | mode; |
632 | local_hpte.v = ((va >> 23) << HPTE_V_AVPN_SHIFT) | |
633 | | HPTE_V_BOLTED | HPTE_V_VALID; | |
1da177e4 LT |
634 | |
635 | slot = HvCallHpt_findValid(&rhpte, vpn); | |
636 | if (slot < 0) { | |
637 | /* Must find space in primary group */ | |
638 | panic("hash_page: hpte already exists\n"); | |
639 | } | |
96e28449 | 640 | HvCallHpt_addValidate(slot, 0, &local_hpte); |
1da177e4 LT |
641 | } |
642 | ||
643 | /* | |
644 | * Bolt the kernel addr space into the HPT | |
645 | */ | |
646 | static void __init iSeries_bolt_kernel(unsigned long saddr, unsigned long eaddr) | |
647 | { | |
648 | unsigned long pa; | |
649 | unsigned long mode_rw = _PAGE_ACCESSED | _PAGE_COHERENT | PP_RWXX; | |
96e28449 | 650 | hpte_t hpte; |
1da177e4 LT |
651 | |
652 | for (pa = saddr; pa < eaddr ;pa += PAGE_SIZE) { | |
653 | unsigned long ea = (unsigned long)__va(pa); | |
654 | unsigned long vsid = get_kernel_vsid(ea); | |
655 | unsigned long va = (vsid << 28) | (pa & 0xfffffff); | |
656 | unsigned long vpn = va >> PAGE_SHIFT; | |
657 | unsigned long slot = HvCallHpt_findValid(&hpte, vpn); | |
658 | ||
659 | /* Make non-kernel text non-executable */ | |
660 | if (!in_kernel_text(ea)) | |
661 | mode_rw |= HW_NO_EXEC; | |
662 | ||
96e28449 | 663 | if (hpte.v & HPTE_V_VALID) { |
1da177e4 LT |
664 | /* HPTE exists, so just bolt it */ |
665 | HvCallHpt_setSwBits(slot, 0x10, 0); | |
666 | /* And make sure the pp bits are correct */ | |
667 | HvCallHpt_setPp(slot, PP_RWXX); | |
668 | } else | |
669 | /* No HPTE exists, so create a new bolted one */ | |
670 | iSeries_make_pte(va, phys_to_abs(pa), mode_rw); | |
671 | } | |
672 | } | |
673 | ||
1da177e4 LT |
674 | /* |
675 | * Document me. | |
676 | */ | |
677 | static void __init iSeries_setup_arch(void) | |
678 | { | |
1da177e4 LT |
679 | unsigned procIx = get_paca()->lppaca.dyn_hv_phys_proc_index; |
680 | ||
681 | /* Add an eye catcher and the systemcfg layout version number */ | |
682 | strcpy(systemcfg->eye_catcher, "SYSTEMCFG:PPC64"); | |
683 | systemcfg->version.major = SYSTEMCFG_MAJOR; | |
684 | systemcfg->version.minor = SYSTEMCFG_MINOR; | |
685 | ||
686 | /* Setup the Lp Event Queue */ | |
512d31d6 | 687 | setup_hvlpevent_queue(); |
1da177e4 LT |
688 | |
689 | /* Compute processor frequency */ | |
690 | procFreqHz = ((1UL << 34) * 1000000) / | |
691 | xIoHriProcessorVpd[procIx].xProcFreq; | |
692 | procFreqMhz = procFreqHz / 1000000; | |
693 | procFreqMhzHundreths = (procFreqHz / 10000) - (procFreqMhz * 100); | |
694 | ppc_proc_freq = procFreqHz; | |
695 | ||
696 | /* Compute time base frequency */ | |
697 | tbFreqHz = ((1UL << 32) * 1000000) / | |
698 | xIoHriProcessorVpd[procIx].xTimeBaseFreq; | |
699 | tbFreqMhz = tbFreqHz / 1000000; | |
700 | tbFreqMhzHundreths = (tbFreqHz / 10000) - (tbFreqMhz * 100); | |
701 | ppc_tb_freq = tbFreqHz; | |
702 | ||
703 | printk("Max logical processors = %d\n", | |
704 | itVpdAreas.xSlicMaxLogicalProcs); | |
705 | printk("Max physical processors = %d\n", | |
706 | itVpdAreas.xSlicMaxPhysicalProcs); | |
707 | printk("Processor frequency = %lu.%02lu\n", procFreqMhz, | |
708 | procFreqMhzHundreths); | |
709 | printk("Time base frequency = %lu.%02lu\n", tbFreqMhz, | |
710 | tbFreqMhzHundreths); | |
711 | systemcfg->processor = xIoHriProcessorVpd[procIx].xPVR; | |
712 | printk("Processor version = %x\n", systemcfg->processor); | |
713 | } | |
714 | ||
715 | static void iSeries_get_cpuinfo(struct seq_file *m) | |
716 | { | |
717 | seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n"); | |
718 | } | |
719 | ||
720 | /* | |
721 | * Document me. | |
722 | * and Implement me. | |
723 | */ | |
724 | static int iSeries_get_irq(struct pt_regs *regs) | |
725 | { | |
726 | /* -2 means ignore this interrupt */ | |
727 | return -2; | |
728 | } | |
729 | ||
730 | /* | |
731 | * Document me. | |
732 | */ | |
733 | static void iSeries_restart(char *cmd) | |
734 | { | |
735 | mf_reboot(); | |
736 | } | |
737 | ||
738 | /* | |
739 | * Document me. | |
740 | */ | |
741 | static void iSeries_power_off(void) | |
742 | { | |
743 | mf_power_off(); | |
744 | } | |
745 | ||
746 | /* | |
747 | * Document me. | |
748 | */ | |
749 | static void iSeries_halt(void) | |
750 | { | |
751 | mf_power_off(); | |
752 | } | |
753 | ||
1da177e4 LT |
754 | /* |
755 | * void __init iSeries_calibrate_decr() | |
756 | * | |
757 | * Description: | |
758 | * This routine retrieves the internal processor frequency from the VPD, | |
759 | * and sets up the kernel timer decrementer based on that value. | |
760 | * | |
761 | */ | |
762 | static void __init iSeries_calibrate_decr(void) | |
763 | { | |
764 | unsigned long cyclesPerUsec; | |
765 | struct div_result divres; | |
766 | ||
767 | /* Compute decrementer (and TB) frequency in cycles/sec */ | |
768 | cyclesPerUsec = ppc_tb_freq / 1000000; | |
769 | ||
770 | /* | |
771 | * Set the amount to refresh the decrementer by. This | |
772 | * is the number of decrementer ticks it takes for | |
773 | * 1/HZ seconds. | |
774 | */ | |
775 | tb_ticks_per_jiffy = ppc_tb_freq / HZ; | |
776 | ||
777 | #if 0 | |
778 | /* TEST CODE FOR ADJTIME */ | |
779 | tb_ticks_per_jiffy += tb_ticks_per_jiffy / 5000; | |
780 | /* END OF TEST CODE */ | |
781 | #endif | |
782 | ||
783 | /* | |
784 | * tb_ticks_per_sec = freq; would give better accuracy | |
785 | * but tb_ticks_per_sec = tb_ticks_per_jiffy*HZ; assures | |
786 | * that jiffies (and xtime) will match the time returned | |
787 | * by do_gettimeofday. | |
788 | */ | |
789 | tb_ticks_per_sec = tb_ticks_per_jiffy * HZ; | |
790 | tb_ticks_per_usec = cyclesPerUsec; | |
791 | tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000); | |
792 | div128_by_32(1024 * 1024, 0, tb_ticks_per_sec, &divres); | |
793 | tb_to_xs = divres.result_low; | |
794 | setup_default_decr(); | |
795 | } | |
796 | ||
797 | static void __init iSeries_progress(char * st, unsigned short code) | |
798 | { | |
799 | printk("Progress: [%04x] - %s\n", (unsigned)code, st); | |
800 | if (!piranha_simulator && mf_initialized) { | |
801 | if (code != 0xffff) | |
802 | mf_display_progress(code); | |
803 | else | |
804 | mf_clear_src(); | |
805 | } | |
806 | } | |
807 | ||
808 | static void __init iSeries_fixup_klimit(void) | |
809 | { | |
810 | /* | |
811 | * Change klimit to take into account any ram disk | |
812 | * that may be included | |
813 | */ | |
814 | if (naca.xRamDisk) | |
815 | klimit = KERNELBASE + (u64)naca.xRamDisk + | |
816 | (naca.xRamDiskSize * PAGE_SIZE); | |
817 | else { | |
818 | /* | |
819 | * No ram disk was included - check and see if there | |
820 | * was an embedded system map. Change klimit to take | |
821 | * into account any embedded system map | |
822 | */ | |
823 | if (embedded_sysmap_end) | |
824 | klimit = KERNELBASE + ((embedded_sysmap_end + 4095) & | |
825 | 0xfffffffffffff000); | |
826 | } | |
827 | } | |
828 | ||
829 | static int __init iSeries_src_init(void) | |
830 | { | |
831 | /* clear the progress line */ | |
832 | ppc_md.progress(" ", 0xffff); | |
833 | return 0; | |
834 | } | |
835 | ||
836 | late_initcall(iSeries_src_init); | |
837 | ||
d200903e ME |
838 | static inline void process_iSeries_events(void) |
839 | { | |
840 | asm volatile ("li 0,0x5555; sc" : : : "r0", "r3"); | |
841 | } | |
842 | ||
843 | static void yield_shared_processor(void) | |
844 | { | |
845 | unsigned long tb; | |
d200903e ME |
846 | |
847 | HvCall_setEnabledInterrupts(HvCall_MaskIPI | | |
848 | HvCall_MaskLpEvent | | |
849 | HvCall_MaskLpProd | | |
850 | HvCall_MaskTimeout); | |
851 | ||
852 | tb = get_tb(); | |
853 | /* Compute future tb value when yield should expire */ | |
854 | HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy); | |
855 | ||
d200903e ME |
856 | /* |
857 | * The decrementer stops during the yield. Force a fake decrementer | |
858 | * here and let the timer_interrupt code sort out the actual time. | |
859 | */ | |
860 | get_paca()->lppaca.int_dword.fields.decr_int = 1; | |
861 | process_iSeries_events(); | |
862 | } | |
863 | ||
3c57bb9f | 864 | static int iseries_shared_idle(void) |
d200903e | 865 | { |
3c57bb9f AB |
866 | while (1) { |
867 | while (!need_resched() && !hvlpevent_is_pending()) { | |
868 | local_irq_disable(); | |
869 | ppc64_runlatch_off(); | |
870 | ||
871 | /* Recheck with irqs off */ | |
872 | if (!need_resched() && !hvlpevent_is_pending()) | |
873 | yield_shared_processor(); | |
d200903e | 874 | |
3c57bb9f AB |
875 | HMT_medium(); |
876 | local_irq_enable(); | |
877 | } | |
878 | ||
879 | ppc64_runlatch_on(); | |
d200903e | 880 | |
3c57bb9f AB |
881 | if (hvlpevent_is_pending()) |
882 | process_iSeries_events(); | |
883 | ||
884 | schedule(); | |
885 | } | |
886 | ||
887 | return 0; | |
888 | } | |
889 | ||
890 | static int iseries_dedicated_idle(void) | |
891 | { | |
3c57bb9f | 892 | long oldval; |
d200903e ME |
893 | |
894 | while (1) { | |
3c57bb9f | 895 | oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED); |
d200903e | 896 | |
3c57bb9f AB |
897 | if (!oldval) { |
898 | set_thread_flag(TIF_POLLING_NRFLAG); | |
d200903e | 899 | |
3c57bb9f AB |
900 | while (!need_resched()) { |
901 | ppc64_runlatch_off(); | |
902 | HMT_low(); | |
903 | ||
904 | if (hvlpevent_is_pending()) { | |
d200903e | 905 | HMT_medium(); |
3c57bb9f AB |
906 | ppc64_runlatch_on(); |
907 | process_iSeries_events(); | |
d200903e | 908 | } |
d200903e | 909 | } |
3c57bb9f AB |
910 | |
911 | HMT_medium(); | |
912 | clear_thread_flag(TIF_POLLING_NRFLAG); | |
913 | } else { | |
914 | set_need_resched(); | |
d200903e ME |
915 | } |
916 | ||
917 | ppc64_runlatch_on(); | |
918 | schedule(); | |
d200903e ME |
919 | } |
920 | ||
921 | return 0; | |
922 | } | |
923 | ||
145d01e4 SR |
924 | #ifndef CONFIG_PCI |
925 | void __init iSeries_init_IRQ(void) { } | |
926 | #endif | |
927 | ||
1da177e4 LT |
928 | void __init iSeries_early_setup(void) |
929 | { | |
930 | iSeries_fixup_klimit(); | |
931 | ||
932 | ppc_md.setup_arch = iSeries_setup_arch; | |
933 | ppc_md.get_cpuinfo = iSeries_get_cpuinfo; | |
934 | ppc_md.init_IRQ = iSeries_init_IRQ; | |
935 | ppc_md.get_irq = iSeries_get_irq; | |
936 | ppc_md.init_early = iSeries_init_early, | |
937 | ||
938 | ppc_md.pcibios_fixup = iSeries_pci_final_fixup; | |
939 | ||
940 | ppc_md.restart = iSeries_restart; | |
941 | ppc_md.power_off = iSeries_power_off; | |
942 | ppc_md.halt = iSeries_halt; | |
943 | ||
944 | ppc_md.get_boot_time = iSeries_get_boot_time; | |
945 | ppc_md.set_rtc_time = iSeries_set_rtc_time; | |
946 | ppc_md.get_rtc_time = iSeries_get_rtc_time; | |
947 | ppc_md.calibrate_decr = iSeries_calibrate_decr; | |
948 | ppc_md.progress = iSeries_progress; | |
3c57bb9f | 949 | |
b6bff397 | 950 | if (get_paca()->lppaca.shared_proc) { |
3c57bb9f | 951 | ppc_md.idle_loop = iseries_shared_idle; |
b6bff397 ME |
952 | printk(KERN_INFO "Using shared processor idle loop\n"); |
953 | } else { | |
3c57bb9f | 954 | ppc_md.idle_loop = iseries_dedicated_idle; |
b6bff397 ME |
955 | printk(KERN_INFO "Using dedicated idle loop\n"); |
956 | } | |
1da177e4 LT |
957 | } |
958 |