powerpc: Move LparData.c to powerpc platforms
[linux-2.6.git] / arch / ppc64 / kernel / iSeries_setup.c
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/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>
42 #include <asm/firmware.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/abs_addr.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/mf.h>
59 #include <asm/iSeries/HvLpEvent.h>
60 #include <asm/iSeries/iSeries_irq.h>
61 #include <asm/iSeries/IoHriProcessorVpd.h>
62 #include <asm/iSeries/ItVpdAreas.h>
63 #include <asm/iSeries/LparMap.h>
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 int iseries_shared_idle(void);
78 static int iseries_dedicated_idle(void);
79 #ifdef CONFIG_PCI
80 extern void iSeries_pci_final_fixup(void);
81 #else
82 static void iSeries_pci_final_fixup(void) { }
83 #endif
84
85 /* Global Variables */
86 int piranha_simulator;
87
88 extern int rd_size;             /* Defined in drivers/block/rd.c */
89 extern unsigned long klimit;
90 extern unsigned long embedded_sysmap_start;
91 extern unsigned long embedded_sysmap_end;
92
93 extern unsigned long iSeries_recal_tb;
94 extern unsigned long iSeries_recal_titan;
95
96 static int mf_initialized;
97
98 struct MemoryBlock {
99         unsigned long absStart;
100         unsigned long absEnd;
101         unsigned long logicalStart;
102         unsigned long logicalEnd;
103 };
104
105 /*
106  * Process the main store vpd to determine where the holes in memory are
107  * and return the number of physical blocks and fill in the array of
108  * block data.
109  */
110 static unsigned long iSeries_process_Condor_mainstore_vpd(
111                 struct MemoryBlock *mb_array, unsigned long max_entries)
112 {
113         unsigned long holeFirstChunk, holeSizeChunks;
114         unsigned long numMemoryBlocks = 1;
115         struct IoHriMainStoreSegment4 *msVpd =
116                 (struct IoHriMainStoreSegment4 *)xMsVpd;
117         unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
118         unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
119         unsigned long holeSize = holeEnd - holeStart;
120
121         printk("Mainstore_VPD: Condor\n");
122         /*
123          * Determine if absolute memory has any
124          * holes so that we can interpret the
125          * access map we get back from the hypervisor
126          * correctly.
127          */
128         mb_array[0].logicalStart = 0;
129         mb_array[0].logicalEnd = 0x100000000;
130         mb_array[0].absStart = 0;
131         mb_array[0].absEnd = 0x100000000;
132
133         if (holeSize) {
134                 numMemoryBlocks = 2;
135                 holeStart = holeStart & 0x000fffffffffffff;
136                 holeStart = addr_to_chunk(holeStart);
137                 holeFirstChunk = holeStart;
138                 holeSize = addr_to_chunk(holeSize);
139                 holeSizeChunks = holeSize;
140                 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
141                                 holeFirstChunk, holeSizeChunks );
142                 mb_array[0].logicalEnd = holeFirstChunk;
143                 mb_array[0].absEnd = holeFirstChunk;
144                 mb_array[1].logicalStart = holeFirstChunk;
145                 mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks;
146                 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
147                 mb_array[1].absEnd = 0x100000000;
148         }
149         return numMemoryBlocks;
150 }
151
152 #define MaxSegmentAreas                 32
153 #define MaxSegmentAdrRangeBlocks        128
154 #define MaxAreaRangeBlocks              4
155
156 static unsigned long iSeries_process_Regatta_mainstore_vpd(
157                 struct MemoryBlock *mb_array, unsigned long max_entries)
158 {
159         struct IoHriMainStoreSegment5 *msVpdP =
160                 (struct IoHriMainStoreSegment5 *)xMsVpd;
161         unsigned long numSegmentBlocks = 0;
162         u32 existsBits = msVpdP->msAreaExists;
163         unsigned long area_num;
164
165         printk("Mainstore_VPD: Regatta\n");
166
167         for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
168                 unsigned long numAreaBlocks;
169                 struct IoHriMainStoreArea4 *currentArea;
170
171                 if (existsBits & 0x80000000) {
172                         unsigned long block_num;
173
174                         currentArea = &msVpdP->msAreaArray[area_num];
175                         numAreaBlocks = currentArea->numAdrRangeBlocks;
176                         printk("ms_vpd: processing area %2ld  blocks=%ld",
177                                         area_num, numAreaBlocks);
178                         for (block_num = 0; block_num < numAreaBlocks;
179                                         ++block_num ) {
180                                 /* Process an address range block */
181                                 struct MemoryBlock tempBlock;
182                                 unsigned long i;
183
184                                 tempBlock.absStart =
185                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
186                                 tempBlock.absEnd =
187                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
188                                 tempBlock.logicalStart = 0;
189                                 tempBlock.logicalEnd   = 0;
190                                 printk("\n          block %ld absStart=%016lx absEnd=%016lx",
191                                                 block_num, tempBlock.absStart,
192                                                 tempBlock.absEnd);
193
194                                 for (i = 0; i < numSegmentBlocks; ++i) {
195                                         if (mb_array[i].absStart ==
196                                                         tempBlock.absStart)
197                                                 break;
198                                 }
199                                 if (i == numSegmentBlocks) {
200                                         if (numSegmentBlocks == max_entries)
201                                                 panic("iSeries_process_mainstore_vpd: too many memory blocks");
202                                         mb_array[numSegmentBlocks] = tempBlock;
203                                         ++numSegmentBlocks;
204                                 } else
205                                         printk(" (duplicate)");
206                         }
207                         printk("\n");
208                 }
209                 existsBits <<= 1;
210         }
211         /* Now sort the blocks found into ascending sequence */
212         if (numSegmentBlocks > 1) {
213                 unsigned long m, n;
214
215                 for (m = 0; m < numSegmentBlocks - 1; ++m) {
216                         for (n = numSegmentBlocks - 1; m < n; --n) {
217                                 if (mb_array[n].absStart <
218                                                 mb_array[n-1].absStart) {
219                                         struct MemoryBlock tempBlock;
220
221                                         tempBlock = mb_array[n];
222                                         mb_array[n] = mb_array[n-1];
223                                         mb_array[n-1] = tempBlock;
224                                 }
225                         }
226                 }
227         }
228         /*
229          * Assign "logical" addresses to each block.  These
230          * addresses correspond to the hypervisor "bitmap" space.
231          * Convert all addresses into units of 256K chunks.
232          */
233         {
234         unsigned long i, nextBitmapAddress;
235
236         printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
237         nextBitmapAddress = 0;
238         for (i = 0; i < numSegmentBlocks; ++i) {
239                 unsigned long length = mb_array[i].absEnd -
240                         mb_array[i].absStart;
241
242                 mb_array[i].logicalStart = nextBitmapAddress;
243                 mb_array[i].logicalEnd = nextBitmapAddress + length;
244                 nextBitmapAddress += length;
245                 printk("          Bitmap range: %016lx - %016lx\n"
246                                 "        Absolute range: %016lx - %016lx\n",
247                                 mb_array[i].logicalStart,
248                                 mb_array[i].logicalEnd,
249                                 mb_array[i].absStart, mb_array[i].absEnd);
250                 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
251                                 0x000fffffffffffff);
252                 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
253                                 0x000fffffffffffff);
254                 mb_array[i].logicalStart =
255                         addr_to_chunk(mb_array[i].logicalStart);
256                 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
257         }
258         }
259
260         return numSegmentBlocks;
261 }
262
263 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
264                 unsigned long max_entries)
265 {
266         unsigned long i;
267         unsigned long mem_blocks = 0;
268
269         if (cpu_has_feature(CPU_FTR_SLB))
270                 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
271                                 max_entries);
272         else
273                 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
274                                 max_entries);
275
276         printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
277         for (i = 0; i < mem_blocks; ++i) {
278                 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
279                        "                             abs chunks %016lx - %016lx\n",
280                         i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
281                         mb_array[i].absStart, mb_array[i].absEnd);
282         }
283         return mem_blocks;
284 }
285
286 static void __init iSeries_get_cmdline(void)
287 {
288         char *p, *q;
289
290         /* copy the command line parameter from the primary VSP  */
291         HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
292                         HvLpDma_Direction_RemoteToLocal);
293
294         p = cmd_line;
295         q = cmd_line + 255;
296         while(p < q) {
297                 if (!*p || *p == '\n')
298                         break;
299                 ++p;
300         }
301         *p = 0;
302 }
303
304 static void __init iSeries_init_early(void)
305 {
306         extern unsigned long memory_limit;
307
308         DBG(" -> iSeries_init_early()\n");
309
310         ppc64_firmware_features = FW_FEATURE_ISERIES;
311
312         ppcdbg_initialize();
313
314         ppc64_interrupt_controller = IC_ISERIES;
315
316 #if defined(CONFIG_BLK_DEV_INITRD)
317         /*
318          * If the init RAM disk has been configured and there is
319          * a non-zero starting address for it, set it up
320          */
321         if (naca.xRamDisk) {
322                 initrd_start = (unsigned long)__va(naca.xRamDisk);
323                 initrd_end = initrd_start + naca.xRamDiskSize * PAGE_SIZE;
324                 initrd_below_start_ok = 1;      // ramdisk in kernel space
325                 ROOT_DEV = Root_RAM0;
326                 if (((rd_size * 1024) / PAGE_SIZE) < naca.xRamDiskSize)
327                         rd_size = (naca.xRamDiskSize * PAGE_SIZE) / 1024;
328         } else
329 #endif /* CONFIG_BLK_DEV_INITRD */
330         {
331             /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
332         }
333
334         iSeries_recal_tb = get_tb();
335         iSeries_recal_titan = HvCallXm_loadTod();
336
337         /*
338          * Initialize the hash table management pointers
339          */
340         hpte_init_iSeries();
341
342         /*
343          * Initialize the DMA/TCE management
344          */
345         iommu_init_early_iSeries();
346
347         iSeries_get_cmdline();
348
349         /* Save unparsed command line copy for /proc/cmdline */
350         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
351
352         /* Parse early parameters, in particular mem=x */
353         parse_early_param();
354
355         if (memory_limit) {
356                 if (memory_limit < systemcfg->physicalMemorySize)
357                         systemcfg->physicalMemorySize = memory_limit;
358                 else {
359                         printk("Ignoring mem=%lu >= ram_top.\n", memory_limit);
360                         memory_limit = 0;
361                 }
362         }
363
364         /* Initialize machine-dependency vectors */
365 #ifdef CONFIG_SMP
366         smp_init_iSeries();
367 #endif
368         if (itLpNaca.xPirEnvironMode == 0)
369                 piranha_simulator = 1;
370
371         /* Associate Lp Event Queue 0 with processor 0 */
372         HvCallEvent_setLpEventQueueInterruptProc(0, 0);
373
374         mf_init();
375         mf_initialized = 1;
376         mb();
377
378         /* If we were passed an initrd, set the ROOT_DEV properly if the values
379          * look sensible. If not, clear initrd reference.
380          */
381 #ifdef CONFIG_BLK_DEV_INITRD
382         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
383             initrd_end > initrd_start)
384                 ROOT_DEV = Root_RAM0;
385         else
386                 initrd_start = initrd_end = 0;
387 #endif /* CONFIG_BLK_DEV_INITRD */
388
389         DBG(" <- iSeries_init_early()\n");
390 }
391
392 struct mschunks_map mschunks_map = {
393         /* XXX We don't use these, but Piranha might need them. */
394         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
395         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
396         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
397 };
398 EXPORT_SYMBOL(mschunks_map);
399
400 void mschunks_alloc(unsigned long num_chunks)
401 {
402         klimit = _ALIGN(klimit, sizeof(u32));
403         mschunks_map.mapping = (u32 *)klimit;
404         klimit += num_chunks * sizeof(u32);
405         mschunks_map.num_chunks = num_chunks;
406 }
407
408 /*
409  * The iSeries may have very large memories ( > 128 GB ) and a partition
410  * may get memory in "chunks" that may be anywhere in the 2**52 real
411  * address space.  The chunks are 256K in size.  To map this to the
412  * memory model Linux expects, the AS/400 specific code builds a
413  * translation table to translate what Linux thinks are "physical"
414  * addresses to the actual real addresses.  This allows us to make
415  * it appear to Linux that we have contiguous memory starting at
416  * physical address zero while in fact this could be far from the truth.
417  * To avoid confusion, I'll let the words physical and/or real address
418  * apply to the Linux addresses while I'll use "absolute address" to
419  * refer to the actual hardware real address.
420  *
421  * build_iSeries_Memory_Map gets information from the Hypervisor and
422  * looks at the Main Store VPD to determine the absolute addresses
423  * of the memory that has been assigned to our partition and builds
424  * a table used to translate Linux's physical addresses to these
425  * absolute addresses.  Absolute addresses are needed when
426  * communicating with the hypervisor (e.g. to build HPT entries)
427  */
428
429 static void __init build_iSeries_Memory_Map(void)
430 {
431         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
432         u32 nextPhysChunk;
433         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
434         u32 num_ptegs;
435         u32 totalChunks,moreChunks;
436         u32 currChunk, thisChunk, absChunk;
437         u32 currDword;
438         u32 chunkBit;
439         u64 map;
440         struct MemoryBlock mb[32];
441         unsigned long numMemoryBlocks, curBlock;
442
443         /* Chunk size on iSeries is 256K bytes */
444         totalChunks = (u32)HvLpConfig_getMsChunks();
445         mschunks_alloc(totalChunks);
446
447         /*
448          * Get absolute address of our load area
449          * and map it to physical address 0
450          * This guarantees that the loadarea ends up at physical 0
451          * otherwise, it might not be returned by PLIC as the first
452          * chunks
453          */
454
455         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
456         loadAreaSize =  itLpNaca.xLoadAreaChunks;
457
458         /*
459          * Only add the pages already mapped here.
460          * Otherwise we might add the hpt pages
461          * The rest of the pages of the load area
462          * aren't in the HPT yet and can still
463          * be assigned an arbitrary physical address
464          */
465         if ((loadAreaSize * 64) > HvPagesToMap)
466                 loadAreaSize = HvPagesToMap / 64;
467
468         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
469
470         /*
471          * TODO Do we need to do something if the HPT is in the 64MB load area?
472          * This would be required if the itLpNaca.xLoadAreaChunks includes
473          * the HPT size
474          */
475
476         printk("Mapping load area - physical addr = 0000000000000000\n"
477                 "                    absolute addr = %016lx\n",
478                 chunk_to_addr(loadAreaFirstChunk));
479         printk("Load area size %dK\n", loadAreaSize * 256);
480
481         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
482                 mschunks_map.mapping[nextPhysChunk] =
483                         loadAreaFirstChunk + nextPhysChunk;
484
485         /*
486          * Get absolute address of our HPT and remember it so
487          * we won't map it to any physical address
488          */
489         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
490         hptSizePages = (u32)HvCallHpt_getHptPages();
491         hptSizeChunks = hptSizePages >> (MSCHUNKS_CHUNK_SHIFT - PAGE_SHIFT);
492         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
493
494         printk("HPT absolute addr = %016lx, size = %dK\n",
495                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
496
497         /* Fill in the hashed page table hash mask */
498         num_ptegs = hptSizePages *
499                 (PAGE_SIZE / (sizeof(hpte_t) * HPTES_PER_GROUP));
500         htab_hash_mask = num_ptegs - 1;
501
502         /*
503          * The actual hashed page table is in the hypervisor,
504          * we have no direct access
505          */
506         htab_address = NULL;
507
508         /*
509          * Determine if absolute memory has any
510          * holes so that we can interpret the
511          * access map we get back from the hypervisor
512          * correctly.
513          */
514         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
515
516         /*
517          * Process the main store access map from the hypervisor
518          * to build up our physical -> absolute translation table
519          */
520         curBlock = 0;
521         currChunk = 0;
522         currDword = 0;
523         moreChunks = totalChunks;
524
525         while (moreChunks) {
526                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
527                                 currDword);
528                 thisChunk = currChunk;
529                 while (map) {
530                         chunkBit = map >> 63;
531                         map <<= 1;
532                         if (chunkBit) {
533                                 --moreChunks;
534                                 while (thisChunk >= mb[curBlock].logicalEnd) {
535                                         ++curBlock;
536                                         if (curBlock >= numMemoryBlocks)
537                                                 panic("out of memory blocks");
538                                 }
539                                 if (thisChunk < mb[curBlock].logicalStart)
540                                         panic("memory block error");
541
542                                 absChunk = mb[curBlock].absStart +
543                                         (thisChunk - mb[curBlock].logicalStart);
544                                 if (((absChunk < hptFirstChunk) ||
545                                      (absChunk > hptLastChunk)) &&
546                                     ((absChunk < loadAreaFirstChunk) ||
547                                      (absChunk > loadAreaLastChunk))) {
548                                         mschunks_map.mapping[nextPhysChunk] =
549                                                 absChunk;
550                                         ++nextPhysChunk;
551                                 }
552                         }
553                         ++thisChunk;
554                 }
555                 ++currDword;
556                 currChunk += 64;
557         }
558
559         /*
560          * main store size (in chunks) is
561          *   totalChunks - hptSizeChunks
562          * which should be equal to
563          *   nextPhysChunk
564          */
565         systemcfg->physicalMemorySize = chunk_to_addr(nextPhysChunk);
566 }
567
568 /*
569  * Document me.
570  */
571 static void __init iSeries_setup_arch(void)
572 {
573         unsigned procIx = get_paca()->lppaca.dyn_hv_phys_proc_index;
574
575         if (get_paca()->lppaca.shared_proc) {
576                 ppc_md.idle_loop = iseries_shared_idle;
577                 printk(KERN_INFO "Using shared processor idle loop\n");
578         } else {
579                 ppc_md.idle_loop = iseries_dedicated_idle;
580                 printk(KERN_INFO "Using dedicated idle loop\n");
581         }
582
583         /* Setup the Lp Event Queue */
584         setup_hvlpevent_queue();
585
586         printk("Max  logical processors = %d\n",
587                         itVpdAreas.xSlicMaxLogicalProcs);
588         printk("Max physical processors = %d\n",
589                         itVpdAreas.xSlicMaxPhysicalProcs);
590
591         systemcfg->processor = xIoHriProcessorVpd[procIx].xPVR;
592         printk("Processor version = %x\n", systemcfg->processor);
593 }
594
595 static void iSeries_get_cpuinfo(struct seq_file *m)
596 {
597         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
598 }
599
600 /*
601  * Document me.
602  * and Implement me.
603  */
604 static int iSeries_get_irq(struct pt_regs *regs)
605 {
606         /* -2 means ignore this interrupt */
607         return -2;
608 }
609
610 /*
611  * Document me.
612  */
613 static void iSeries_restart(char *cmd)
614 {
615         mf_reboot();
616 }
617
618 /*
619  * Document me.
620  */
621 static void iSeries_power_off(void)
622 {
623         mf_power_off();
624 }
625
626 /*
627  * Document me.
628  */
629 static void iSeries_halt(void)
630 {
631         mf_power_off();
632 }
633
634 static void __init iSeries_progress(char * st, unsigned short code)
635 {
636         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
637         if (!piranha_simulator && mf_initialized) {
638                 if (code != 0xffff)
639                         mf_display_progress(code);
640                 else
641                         mf_clear_src();
642         }
643 }
644
645 static void __init iSeries_fixup_klimit(void)
646 {
647         /*
648          * Change klimit to take into account any ram disk
649          * that may be included
650          */
651         if (naca.xRamDisk)
652                 klimit = KERNELBASE + (u64)naca.xRamDisk +
653                         (naca.xRamDiskSize * PAGE_SIZE);
654         else {
655                 /*
656                  * No ram disk was included - check and see if there
657                  * was an embedded system map.  Change klimit to take
658                  * into account any embedded system map
659                  */
660                 if (embedded_sysmap_end)
661                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
662                                         0xfffffffffffff000);
663         }
664 }
665
666 static int __init iSeries_src_init(void)
667 {
668         /* clear the progress line */
669         ppc_md.progress(" ", 0xffff);
670         return 0;
671 }
672
673 late_initcall(iSeries_src_init);
674
675 static inline void process_iSeries_events(void)
676 {
677         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
678 }
679
680 static void yield_shared_processor(void)
681 {
682         unsigned long tb;
683
684         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
685                                     HvCall_MaskLpEvent |
686                                     HvCall_MaskLpProd |
687                                     HvCall_MaskTimeout);
688
689         tb = get_tb();
690         /* Compute future tb value when yield should expire */
691         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
692
693         /*
694          * The decrementer stops during the yield.  Force a fake decrementer
695          * here and let the timer_interrupt code sort out the actual time.
696          */
697         get_paca()->lppaca.int_dword.fields.decr_int = 1;
698         process_iSeries_events();
699 }
700
701 static int iseries_shared_idle(void)
702 {
703         while (1) {
704                 while (!need_resched() && !hvlpevent_is_pending()) {
705                         local_irq_disable();
706                         ppc64_runlatch_off();
707
708                         /* Recheck with irqs off */
709                         if (!need_resched() && !hvlpevent_is_pending())
710                                 yield_shared_processor();
711
712                         HMT_medium();
713                         local_irq_enable();
714                 }
715
716                 ppc64_runlatch_on();
717
718                 if (hvlpevent_is_pending())
719                         process_iSeries_events();
720
721                 schedule();
722         }
723
724         return 0;
725 }
726
727 static int iseries_dedicated_idle(void)
728 {
729         long oldval;
730
731         while (1) {
732                 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
733
734                 if (!oldval) {
735                         set_thread_flag(TIF_POLLING_NRFLAG);
736
737                         while (!need_resched()) {
738                                 ppc64_runlatch_off();
739                                 HMT_low();
740
741                                 if (hvlpevent_is_pending()) {
742                                         HMT_medium();
743                                         ppc64_runlatch_on();
744                                         process_iSeries_events();
745                                 }
746                         }
747
748                         HMT_medium();
749                         clear_thread_flag(TIF_POLLING_NRFLAG);
750                 } else {
751                         set_need_resched();
752                 }
753
754                 ppc64_runlatch_on();
755                 schedule();
756         }
757
758         return 0;
759 }
760
761 #ifndef CONFIG_PCI
762 void __init iSeries_init_IRQ(void) { }
763 #endif
764
765 static int __init iseries_probe(int platform)
766 {
767         return PLATFORM_ISERIES_LPAR == platform;
768 }
769
770 struct machdep_calls __initdata iseries_md = {
771         .setup_arch     = iSeries_setup_arch,
772         .get_cpuinfo    = iSeries_get_cpuinfo,
773         .init_IRQ       = iSeries_init_IRQ,
774         .get_irq        = iSeries_get_irq,
775         .init_early     = iSeries_init_early,
776         .pcibios_fixup  = iSeries_pci_final_fixup,
777         .restart        = iSeries_restart,
778         .power_off      = iSeries_power_off,
779         .halt           = iSeries_halt,
780         .get_boot_time  = iSeries_get_boot_time,
781         .set_rtc_time   = iSeries_set_rtc_time,
782         .get_rtc_time   = iSeries_get_rtc_time,
783         .calibrate_decr = generic_calibrate_decr,
784         .progress       = iSeries_progress,
785         .probe          = iseries_probe,
786         /* XXX Implement enable_pmcs for iSeries */
787 };
788
789 struct blob {
790         unsigned char data[PAGE_SIZE];
791         unsigned long next;
792 };
793
794 struct iseries_flat_dt {
795         struct boot_param_header header;
796         u64 reserve_map[2];
797         struct blob dt;
798         struct blob strings;
799 };
800
801 struct iseries_flat_dt iseries_dt;
802
803 void dt_init(struct iseries_flat_dt *dt)
804 {
805         dt->header.off_mem_rsvmap =
806                 offsetof(struct iseries_flat_dt, reserve_map);
807         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
808         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
809         dt->header.totalsize = sizeof(struct iseries_flat_dt);
810         dt->header.dt_strings_size = sizeof(struct blob);
811
812         /* There is no notion of hardware cpu id on iSeries */
813         dt->header.boot_cpuid_phys = smp_processor_id();
814
815         dt->dt.next = (unsigned long)&dt->dt.data;
816         dt->strings.next = (unsigned long)&dt->strings.data;
817
818         dt->header.magic = OF_DT_HEADER;
819         dt->header.version = 0x10;
820         dt->header.last_comp_version = 0x10;
821
822         dt->reserve_map[0] = 0;
823         dt->reserve_map[1] = 0;
824 }
825
826 void dt_check_blob(struct blob *b)
827 {
828         if (b->next >= (unsigned long)&b->next) {
829                 DBG("Ran out of space in flat device tree blob!\n");
830                 BUG();
831         }
832 }
833
834 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
835 {
836         *((u32*)dt->dt.next) = value;
837         dt->dt.next += sizeof(u32);
838
839         dt_check_blob(&dt->dt);
840 }
841
842 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
843 {
844         *((u64*)dt->dt.next) = value;
845         dt->dt.next += sizeof(u64);
846
847         dt_check_blob(&dt->dt);
848 }
849
850 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
851 {
852         unsigned long start = blob->next - (unsigned long)blob->data;
853
854         memcpy((char *)blob->next, data, len);
855         blob->next = _ALIGN(blob->next + len, 4);
856
857         dt_check_blob(blob);
858
859         return start;
860 }
861
862 void dt_start_node(struct iseries_flat_dt *dt, char *name)
863 {
864         dt_push_u32(dt, OF_DT_BEGIN_NODE);
865         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
866 }
867
868 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
869
870 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
871 {
872         unsigned long offset;
873
874         dt_push_u32(dt, OF_DT_PROP);
875
876         /* Length of the data */
877         dt_push_u32(dt, len);
878
879         /* Put the property name in the string blob. */
880         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
881
882         /* The offset of the properties name in the string blob. */
883         dt_push_u32(dt, (u32)offset);
884
885         /* The actual data. */
886         dt_push_bytes(&dt->dt, data, len);
887 }
888
889 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
890 {
891         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
892 }
893
894 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
895 {
896         dt_prop(dt, name, (char *)&data, sizeof(u32));
897 }
898
899 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
900 {
901         dt_prop(dt, name, (char *)&data, sizeof(u64));
902 }
903
904 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
905 {
906         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
907 }
908
909 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
910 {
911         dt_prop(dt, name, NULL, 0);
912 }
913
914 void dt_cpus(struct iseries_flat_dt *dt)
915 {
916         unsigned char buf[32];
917         unsigned char *p;
918         unsigned int i, index;
919         struct IoHriProcessorVpd *d;
920
921         /* yuck */
922         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
923         p = strchr(buf, ' ');
924         if (!p) p = buf + strlen(buf);
925
926         dt_start_node(dt, "cpus");
927         dt_prop_u32(dt, "#address-cells", 1);
928         dt_prop_u32(dt, "#size-cells", 0);
929
930         for (i = 0; i < NR_CPUS; i++) {
931                 if (paca[i].lppaca.dyn_proc_status >= 2)
932                         continue;
933
934                 snprintf(p, 32 - (p - buf), "@%d", i);
935                 dt_start_node(dt, buf);
936
937                 dt_prop_str(dt, "device_type", "cpu");
938
939                 index = paca[i].lppaca.dyn_hv_phys_proc_index;
940                 d = &xIoHriProcessorVpd[index];
941
942                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
943                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
944
945                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
946                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
947
948                 /* magic conversions to Hz copied from old code */
949                 dt_prop_u32(dt, "clock-frequency",
950                         ((1UL << 34) * 1000000) / d->xProcFreq);
951                 dt_prop_u32(dt, "timebase-frequency",
952                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
953
954                 dt_prop_u32(dt, "reg", i);
955
956                 dt_end_node(dt);
957         }
958
959         dt_end_node(dt);
960 }
961
962 void build_flat_dt(struct iseries_flat_dt *dt)
963 {
964         u64 tmp[2];
965
966         dt_init(dt);
967
968         dt_start_node(dt, "");
969
970         dt_prop_u32(dt, "#address-cells", 2);
971         dt_prop_u32(dt, "#size-cells", 2);
972
973         /* /memory */
974         dt_start_node(dt, "memory@0");
975         dt_prop_str(dt, "name", "memory");
976         dt_prop_str(dt, "device_type", "memory");
977         tmp[0] = 0;
978         tmp[1] = systemcfg->physicalMemorySize;
979         dt_prop_u64_list(dt, "reg", tmp, 2);
980         dt_end_node(dt);
981
982         /* /chosen */
983         dt_start_node(dt, "chosen");
984         dt_prop_u32(dt, "linux,platform", PLATFORM_ISERIES_LPAR);
985         dt_end_node(dt);
986
987         dt_cpus(dt);
988
989         dt_end_node(dt);
990
991         dt_push_u32(dt, OF_DT_END);
992 }
993
994 void * __init iSeries_early_setup(void)
995 {
996         iSeries_fixup_klimit();
997
998         /*
999          * Initialize the table which translate Linux physical addresses to
1000          * AS/400 absolute addresses
1001          */
1002         build_iSeries_Memory_Map();
1003
1004         build_flat_dt(&iseries_dt);
1005
1006         return (void *) __pa(&iseries_dt);
1007 }