powerpc: move iSeries_setup.[ch] and mf.c into platforms/iseries
[linux-2.6.git] / arch / powerpc / platforms / 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  *    Description:
6  *      Architecture- / platform-specific boot-time initialization code for
7  *      the IBM iSeries LPAR.  Adapted from original code by Grant Erickson and
8  *      code by Gary Thomas, Cort Dougan <cort@fsmlabs.com>, and Dan Malek
9  *      <dan@net4x.com>.
10  *
11  *      This program is free software; you can redistribute it and/or
12  *      modify it under the terms of the GNU General Public License
13  *      as published by the Free Software Foundation; either version
14  *      2 of the License, or (at your option) any later version.
15  */
16
17 #undef DEBUG
18
19 #include <linux/config.h>
20 #include <linux/init.h>
21 #include <linux/threads.h>
22 #include <linux/smp.h>
23 #include <linux/param.h>
24 #include <linux/string.h>
25 #include <linux/initrd.h>
26 #include <linux/seq_file.h>
27 #include <linux/kdev_t.h>
28 #include <linux/major.h>
29 #include <linux/root_dev.h>
30
31 #include <asm/processor.h>
32 #include <asm/machdep.h>
33 #include <asm/page.h>
34 #include <asm/mmu.h>
35 #include <asm/pgtable.h>
36 #include <asm/mmu_context.h>
37 #include <asm/cputable.h>
38 #include <asm/sections.h>
39 #include <asm/iommu.h>
40 #include <asm/firmware.h>
41
42 #include <asm/time.h>
43 #include <asm/naca.h>
44 #include <asm/paca.h>
45 #include <asm/cache.h>
46 #include <asm/sections.h>
47 #include <asm/abs_addr.h>
48 #include <asm/iSeries/HvCallHpt.h>
49 #include <asm/iSeries/HvLpConfig.h>
50 #include <asm/iSeries/HvCallEvent.h>
51 #include <asm/iSeries/HvCallSm.h>
52 #include <asm/iSeries/HvCallXm.h>
53 #include <asm/iSeries/ItLpQueue.h>
54 #include <asm/iSeries/IoHriMainStore.h>
55 #include <asm/iSeries/mf.h>
56 #include <asm/iSeries/HvLpEvent.h>
57 #include <asm/iSeries/iSeries_irq.h>
58 #include <asm/iSeries/IoHriProcessorVpd.h>
59 #include <asm/iSeries/ItVpdAreas.h>
60 #include <asm/iSeries/LparMap.h>
61
62 #include "setup.h"
63
64 extern void hvlog(char *fmt, ...);
65
66 #ifdef DEBUG
67 #define DBG(fmt...) hvlog(fmt)
68 #else
69 #define DBG(fmt...)
70 #endif
71
72 /* Function Prototypes */
73 extern void ppcdbg_initialize(void);
74
75 static void build_iSeries_Memory_Map(void);
76 static int iseries_shared_idle(void);
77 static int iseries_dedicated_idle(void);
78 #ifdef CONFIG_PCI
79 extern void iSeries_pci_final_fixup(void);
80 #else
81 static void iSeries_pci_final_fixup(void) { }
82 #endif
83
84 /* Global Variables */
85 int piranha_simulator;
86
87 extern int rd_size;             /* Defined in drivers/block/rd.c */
88 extern unsigned long klimit;
89 extern unsigned long embedded_sysmap_start;
90 extern unsigned long embedded_sysmap_end;
91
92 extern unsigned long iSeries_recal_tb;
93 extern unsigned long iSeries_recal_titan;
94
95 static int mf_initialized;
96
97 struct MemoryBlock {
98         unsigned long absStart;
99         unsigned long absEnd;
100         unsigned long logicalStart;
101         unsigned long logicalEnd;
102 };
103
104 /*
105  * Process the main store vpd to determine where the holes in memory are
106  * and return the number of physical blocks and fill in the array of
107  * block data.
108  */
109 static unsigned long iSeries_process_Condor_mainstore_vpd(
110                 struct MemoryBlock *mb_array, unsigned long max_entries)
111 {
112         unsigned long holeFirstChunk, holeSizeChunks;
113         unsigned long numMemoryBlocks = 1;
114         struct IoHriMainStoreSegment4 *msVpd =
115                 (struct IoHriMainStoreSegment4 *)xMsVpd;
116         unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
117         unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
118         unsigned long holeSize = holeEnd - holeStart;
119
120         printk("Mainstore_VPD: Condor\n");
121         /*
122          * Determine if absolute memory has any
123          * holes so that we can interpret the
124          * access map we get back from the hypervisor
125          * correctly.
126          */
127         mb_array[0].logicalStart = 0;
128         mb_array[0].logicalEnd = 0x100000000;
129         mb_array[0].absStart = 0;
130         mb_array[0].absEnd = 0x100000000;
131
132         if (holeSize) {
133                 numMemoryBlocks = 2;
134                 holeStart = holeStart & 0x000fffffffffffff;
135                 holeStart = addr_to_chunk(holeStart);
136                 holeFirstChunk = holeStart;
137                 holeSize = addr_to_chunk(holeSize);
138                 holeSizeChunks = holeSize;
139                 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
140                                 holeFirstChunk, holeSizeChunks );
141                 mb_array[0].logicalEnd = holeFirstChunk;
142                 mb_array[0].absEnd = holeFirstChunk;
143                 mb_array[1].logicalStart = holeFirstChunk;
144                 mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks;
145                 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
146                 mb_array[1].absEnd = 0x100000000;
147         }
148         return numMemoryBlocks;
149 }
150
151 #define MaxSegmentAreas                 32
152 #define MaxSegmentAdrRangeBlocks        128
153 #define MaxAreaRangeBlocks              4
154
155 static unsigned long iSeries_process_Regatta_mainstore_vpd(
156                 struct MemoryBlock *mb_array, unsigned long max_entries)
157 {
158         struct IoHriMainStoreSegment5 *msVpdP =
159                 (struct IoHriMainStoreSegment5 *)xMsVpd;
160         unsigned long numSegmentBlocks = 0;
161         u32 existsBits = msVpdP->msAreaExists;
162         unsigned long area_num;
163
164         printk("Mainstore_VPD: Regatta\n");
165
166         for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
167                 unsigned long numAreaBlocks;
168                 struct IoHriMainStoreArea4 *currentArea;
169
170                 if (existsBits & 0x80000000) {
171                         unsigned long block_num;
172
173                         currentArea = &msVpdP->msAreaArray[area_num];
174                         numAreaBlocks = currentArea->numAdrRangeBlocks;
175                         printk("ms_vpd: processing area %2ld  blocks=%ld",
176                                         area_num, numAreaBlocks);
177                         for (block_num = 0; block_num < numAreaBlocks;
178                                         ++block_num ) {
179                                 /* Process an address range block */
180                                 struct MemoryBlock tempBlock;
181                                 unsigned long i;
182
183                                 tempBlock.absStart =
184                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
185                                 tempBlock.absEnd =
186                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
187                                 tempBlock.logicalStart = 0;
188                                 tempBlock.logicalEnd   = 0;
189                                 printk("\n          block %ld absStart=%016lx absEnd=%016lx",
190                                                 block_num, tempBlock.absStart,
191                                                 tempBlock.absEnd);
192
193                                 for (i = 0; i < numSegmentBlocks; ++i) {
194                                         if (mb_array[i].absStart ==
195                                                         tempBlock.absStart)
196                                                 break;
197                                 }
198                                 if (i == numSegmentBlocks) {
199                                         if (numSegmentBlocks == max_entries)
200                                                 panic("iSeries_process_mainstore_vpd: too many memory blocks");
201                                         mb_array[numSegmentBlocks] = tempBlock;
202                                         ++numSegmentBlocks;
203                                 } else
204                                         printk(" (duplicate)");
205                         }
206                         printk("\n");
207                 }
208                 existsBits <<= 1;
209         }
210         /* Now sort the blocks found into ascending sequence */
211         if (numSegmentBlocks > 1) {
212                 unsigned long m, n;
213
214                 for (m = 0; m < numSegmentBlocks - 1; ++m) {
215                         for (n = numSegmentBlocks - 1; m < n; --n) {
216                                 if (mb_array[n].absStart <
217                                                 mb_array[n-1].absStart) {
218                                         struct MemoryBlock tempBlock;
219
220                                         tempBlock = mb_array[n];
221                                         mb_array[n] = mb_array[n-1];
222                                         mb_array[n-1] = tempBlock;
223                                 }
224                         }
225                 }
226         }
227         /*
228          * Assign "logical" addresses to each block.  These
229          * addresses correspond to the hypervisor "bitmap" space.
230          * Convert all addresses into units of 256K chunks.
231          */
232         {
233         unsigned long i, nextBitmapAddress;
234
235         printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
236         nextBitmapAddress = 0;
237         for (i = 0; i < numSegmentBlocks; ++i) {
238                 unsigned long length = mb_array[i].absEnd -
239                         mb_array[i].absStart;
240
241                 mb_array[i].logicalStart = nextBitmapAddress;
242                 mb_array[i].logicalEnd = nextBitmapAddress + length;
243                 nextBitmapAddress += length;
244                 printk("          Bitmap range: %016lx - %016lx\n"
245                                 "        Absolute range: %016lx - %016lx\n",
246                                 mb_array[i].logicalStart,
247                                 mb_array[i].logicalEnd,
248                                 mb_array[i].absStart, mb_array[i].absEnd);
249                 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
250                                 0x000fffffffffffff);
251                 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
252                                 0x000fffffffffffff);
253                 mb_array[i].logicalStart =
254                         addr_to_chunk(mb_array[i].logicalStart);
255                 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
256         }
257         }
258
259         return numSegmentBlocks;
260 }
261
262 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
263                 unsigned long max_entries)
264 {
265         unsigned long i;
266         unsigned long mem_blocks = 0;
267
268         if (cpu_has_feature(CPU_FTR_SLB))
269                 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
270                                 max_entries);
271         else
272                 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
273                                 max_entries);
274
275         printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
276         for (i = 0; i < mem_blocks; ++i) {
277                 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
278                        "                             abs chunks %016lx - %016lx\n",
279                         i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
280                         mb_array[i].absStart, mb_array[i].absEnd);
281         }
282         return mem_blocks;
283 }
284
285 static void __init iSeries_get_cmdline(void)
286 {
287         char *p, *q;
288
289         /* copy the command line parameter from the primary VSP  */
290         HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
291                         HvLpDma_Direction_RemoteToLocal);
292
293         p = cmd_line;
294         q = cmd_line + 255;
295         while(p < q) {
296                 if (!*p || *p == '\n')
297                         break;
298                 ++p;
299         }
300         *p = 0;
301 }
302
303 static void __init iSeries_init_early(void)
304 {
305         extern unsigned long memory_limit;
306
307         DBG(" -> iSeries_init_early()\n");
308
309         ppc64_firmware_features = FW_FEATURE_ISERIES;
310
311         ppcdbg_initialize();
312
313         ppc64_interrupt_controller = IC_ISERIES;
314
315 #if defined(CONFIG_BLK_DEV_INITRD)
316         /*
317          * If the init RAM disk has been configured and there is
318          * a non-zero starting address for it, set it up
319          */
320         if (naca.xRamDisk) {
321                 initrd_start = (unsigned long)__va(naca.xRamDisk);
322                 initrd_end = initrd_start + naca.xRamDiskSize * PAGE_SIZE;
323                 initrd_below_start_ok = 1;      // ramdisk in kernel space
324                 ROOT_DEV = Root_RAM0;
325                 if (((rd_size * 1024) / PAGE_SIZE) < naca.xRamDiskSize)
326                         rd_size = (naca.xRamDiskSize * PAGE_SIZE) / 1024;
327         } else
328 #endif /* CONFIG_BLK_DEV_INITRD */
329         {
330             /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
331         }
332
333         iSeries_recal_tb = get_tb();
334         iSeries_recal_titan = HvCallXm_loadTod();
335
336         /*
337          * Initialize the hash table management pointers
338          */
339         hpte_init_iSeries();
340
341         /*
342          * Initialize the DMA/TCE management
343          */
344         iommu_init_early_iSeries();
345
346         iSeries_get_cmdline();
347
348         /* Save unparsed command line copy for /proc/cmdline */
349         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
350
351         /* Parse early parameters, in particular mem=x */
352         parse_early_param();
353
354         if (memory_limit) {
355                 if (memory_limit < systemcfg->physicalMemorySize)
356                         systemcfg->physicalMemorySize = memory_limit;
357                 else {
358                         printk("Ignoring mem=%lu >= ram_top.\n", memory_limit);
359                         memory_limit = 0;
360                 }
361         }
362
363         /* Initialize machine-dependency vectors */
364 #ifdef CONFIG_SMP
365         smp_init_iSeries();
366 #endif
367         if (itLpNaca.xPirEnvironMode == 0)
368                 piranha_simulator = 1;
369
370         /* Associate Lp Event Queue 0 with processor 0 */
371         HvCallEvent_setLpEventQueueInterruptProc(0, 0);
372
373         mf_init();
374         mf_initialized = 1;
375         mb();
376
377         /* If we were passed an initrd, set the ROOT_DEV properly if the values
378          * look sensible. If not, clear initrd reference.
379          */
380 #ifdef CONFIG_BLK_DEV_INITRD
381         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
382             initrd_end > initrd_start)
383                 ROOT_DEV = Root_RAM0;
384         else
385                 initrd_start = initrd_end = 0;
386 #endif /* CONFIG_BLK_DEV_INITRD */
387
388         DBG(" <- iSeries_init_early()\n");
389 }
390
391 struct mschunks_map mschunks_map = {
392         /* XXX We don't use these, but Piranha might need them. */
393         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
394         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
395         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
396 };
397 EXPORT_SYMBOL(mschunks_map);
398
399 void mschunks_alloc(unsigned long num_chunks)
400 {
401         klimit = _ALIGN(klimit, sizeof(u32));
402         mschunks_map.mapping = (u32 *)klimit;
403         klimit += num_chunks * sizeof(u32);
404         mschunks_map.num_chunks = num_chunks;
405 }
406
407 /*
408  * The iSeries may have very large memories ( > 128 GB ) and a partition
409  * may get memory in "chunks" that may be anywhere in the 2**52 real
410  * address space.  The chunks are 256K in size.  To map this to the
411  * memory model Linux expects, the AS/400 specific code builds a
412  * translation table to translate what Linux thinks are "physical"
413  * addresses to the actual real addresses.  This allows us to make
414  * it appear to Linux that we have contiguous memory starting at
415  * physical address zero while in fact this could be far from the truth.
416  * To avoid confusion, I'll let the words physical and/or real address
417  * apply to the Linux addresses while I'll use "absolute address" to
418  * refer to the actual hardware real address.
419  *
420  * build_iSeries_Memory_Map gets information from the Hypervisor and
421  * looks at the Main Store VPD to determine the absolute addresses
422  * of the memory that has been assigned to our partition and builds
423  * a table used to translate Linux's physical addresses to these
424  * absolute addresses.  Absolute addresses are needed when
425  * communicating with the hypervisor (e.g. to build HPT entries)
426  */
427
428 static void __init build_iSeries_Memory_Map(void)
429 {
430         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
431         u32 nextPhysChunk;
432         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
433         u32 num_ptegs;
434         u32 totalChunks,moreChunks;
435         u32 currChunk, thisChunk, absChunk;
436         u32 currDword;
437         u32 chunkBit;
438         u64 map;
439         struct MemoryBlock mb[32];
440         unsigned long numMemoryBlocks, curBlock;
441
442         /* Chunk size on iSeries is 256K bytes */
443         totalChunks = (u32)HvLpConfig_getMsChunks();
444         mschunks_alloc(totalChunks);
445
446         /*
447          * Get absolute address of our load area
448          * and map it to physical address 0
449          * This guarantees that the loadarea ends up at physical 0
450          * otherwise, it might not be returned by PLIC as the first
451          * chunks
452          */
453
454         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
455         loadAreaSize =  itLpNaca.xLoadAreaChunks;
456
457         /*
458          * Only add the pages already mapped here.
459          * Otherwise we might add the hpt pages
460          * The rest of the pages of the load area
461          * aren't in the HPT yet and can still
462          * be assigned an arbitrary physical address
463          */
464         if ((loadAreaSize * 64) > HvPagesToMap)
465                 loadAreaSize = HvPagesToMap / 64;
466
467         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
468
469         /*
470          * TODO Do we need to do something if the HPT is in the 64MB load area?
471          * This would be required if the itLpNaca.xLoadAreaChunks includes
472          * the HPT size
473          */
474
475         printk("Mapping load area - physical addr = 0000000000000000\n"
476                 "                    absolute addr = %016lx\n",
477                 chunk_to_addr(loadAreaFirstChunk));
478         printk("Load area size %dK\n", loadAreaSize * 256);
479
480         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
481                 mschunks_map.mapping[nextPhysChunk] =
482                         loadAreaFirstChunk + nextPhysChunk;
483
484         /*
485          * Get absolute address of our HPT and remember it so
486          * we won't map it to any physical address
487          */
488         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
489         hptSizePages = (u32)HvCallHpt_getHptPages();
490         hptSizeChunks = hptSizePages >> (MSCHUNKS_CHUNK_SHIFT - PAGE_SHIFT);
491         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
492
493         printk("HPT absolute addr = %016lx, size = %dK\n",
494                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
495
496         /* Fill in the hashed page table hash mask */
497         num_ptegs = hptSizePages *
498                 (PAGE_SIZE / (sizeof(hpte_t) * HPTES_PER_GROUP));
499         htab_hash_mask = num_ptegs - 1;
500
501         /*
502          * The actual hashed page table is in the hypervisor,
503          * we have no direct access
504          */
505         htab_address = NULL;
506
507         /*
508          * Determine if absolute memory has any
509          * holes so that we can interpret the
510          * access map we get back from the hypervisor
511          * correctly.
512          */
513         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
514
515         /*
516          * Process the main store access map from the hypervisor
517          * to build up our physical -> absolute translation table
518          */
519         curBlock = 0;
520         currChunk = 0;
521         currDword = 0;
522         moreChunks = totalChunks;
523
524         while (moreChunks) {
525                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
526                                 currDword);
527                 thisChunk = currChunk;
528                 while (map) {
529                         chunkBit = map >> 63;
530                         map <<= 1;
531                         if (chunkBit) {
532                                 --moreChunks;
533                                 while (thisChunk >= mb[curBlock].logicalEnd) {
534                                         ++curBlock;
535                                         if (curBlock >= numMemoryBlocks)
536                                                 panic("out of memory blocks");
537                                 }
538                                 if (thisChunk < mb[curBlock].logicalStart)
539                                         panic("memory block error");
540
541                                 absChunk = mb[curBlock].absStart +
542                                         (thisChunk - mb[curBlock].logicalStart);
543                                 if (((absChunk < hptFirstChunk) ||
544                                      (absChunk > hptLastChunk)) &&
545                                     ((absChunk < loadAreaFirstChunk) ||
546                                      (absChunk > loadAreaLastChunk))) {
547                                         mschunks_map.mapping[nextPhysChunk] =
548                                                 absChunk;
549                                         ++nextPhysChunk;
550                                 }
551                         }
552                         ++thisChunk;
553                 }
554                 ++currDword;
555                 currChunk += 64;
556         }
557
558         /*
559          * main store size (in chunks) is
560          *   totalChunks - hptSizeChunks
561          * which should be equal to
562          *   nextPhysChunk
563          */
564         systemcfg->physicalMemorySize = chunk_to_addr(nextPhysChunk);
565 }
566
567 /*
568  * Document me.
569  */
570 static void __init iSeries_setup_arch(void)
571 {
572         unsigned procIx = get_paca()->lppaca.dyn_hv_phys_proc_index;
573
574         if (get_paca()->lppaca.shared_proc) {
575                 ppc_md.idle_loop = iseries_shared_idle;
576                 printk(KERN_INFO "Using shared processor idle loop\n");
577         } else {
578                 ppc_md.idle_loop = iseries_dedicated_idle;
579                 printk(KERN_INFO "Using dedicated idle loop\n");
580         }
581
582         /* Setup the Lp Event Queue */
583         setup_hvlpevent_queue();
584
585         printk("Max  logical processors = %d\n",
586                         itVpdAreas.xSlicMaxLogicalProcs);
587         printk("Max physical processors = %d\n",
588                         itVpdAreas.xSlicMaxPhysicalProcs);
589
590         systemcfg->processor = xIoHriProcessorVpd[procIx].xPVR;
591         printk("Processor version = %x\n", systemcfg->processor);
592 }
593
594 static void iSeries_get_cpuinfo(struct seq_file *m)
595 {
596         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
597 }
598
599 /*
600  * Document me.
601  * and Implement me.
602  */
603 static int iSeries_get_irq(struct pt_regs *regs)
604 {
605         /* -2 means ignore this interrupt */
606         return -2;
607 }
608
609 /*
610  * Document me.
611  */
612 static void iSeries_restart(char *cmd)
613 {
614         mf_reboot();
615 }
616
617 /*
618  * Document me.
619  */
620 static void iSeries_power_off(void)
621 {
622         mf_power_off();
623 }
624
625 /*
626  * Document me.
627  */
628 static void iSeries_halt(void)
629 {
630         mf_power_off();
631 }
632
633 static void __init iSeries_progress(char * st, unsigned short code)
634 {
635         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
636         if (!piranha_simulator && mf_initialized) {
637                 if (code != 0xffff)
638                         mf_display_progress(code);
639                 else
640                         mf_clear_src();
641         }
642 }
643
644 static void __init iSeries_fixup_klimit(void)
645 {
646         /*
647          * Change klimit to take into account any ram disk
648          * that may be included
649          */
650         if (naca.xRamDisk)
651                 klimit = KERNELBASE + (u64)naca.xRamDisk +
652                         (naca.xRamDiskSize * PAGE_SIZE);
653         else {
654                 /*
655                  * No ram disk was included - check and see if there
656                  * was an embedded system map.  Change klimit to take
657                  * into account any embedded system map
658                  */
659                 if (embedded_sysmap_end)
660                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
661                                         0xfffffffffffff000);
662         }
663 }
664
665 static int __init iSeries_src_init(void)
666 {
667         /* clear the progress line */
668         ppc_md.progress(" ", 0xffff);
669         return 0;
670 }
671
672 late_initcall(iSeries_src_init);
673
674 static inline void process_iSeries_events(void)
675 {
676         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
677 }
678
679 static void yield_shared_processor(void)
680 {
681         unsigned long tb;
682
683         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
684                                     HvCall_MaskLpEvent |
685                                     HvCall_MaskLpProd |
686                                     HvCall_MaskTimeout);
687
688         tb = get_tb();
689         /* Compute future tb value when yield should expire */
690         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
691
692         /*
693          * The decrementer stops during the yield.  Force a fake decrementer
694          * here and let the timer_interrupt code sort out the actual time.
695          */
696         get_paca()->lppaca.int_dword.fields.decr_int = 1;
697         process_iSeries_events();
698 }
699
700 static int iseries_shared_idle(void)
701 {
702         while (1) {
703                 while (!need_resched() && !hvlpevent_is_pending()) {
704                         local_irq_disable();
705                         ppc64_runlatch_off();
706
707                         /* Recheck with irqs off */
708                         if (!need_resched() && !hvlpevent_is_pending())
709                                 yield_shared_processor();
710
711                         HMT_medium();
712                         local_irq_enable();
713                 }
714
715                 ppc64_runlatch_on();
716
717                 if (hvlpevent_is_pending())
718                         process_iSeries_events();
719
720                 schedule();
721         }
722
723         return 0;
724 }
725
726 static int iseries_dedicated_idle(void)
727 {
728         long oldval;
729
730         while (1) {
731                 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
732
733                 if (!oldval) {
734                         set_thread_flag(TIF_POLLING_NRFLAG);
735
736                         while (!need_resched()) {
737                                 ppc64_runlatch_off();
738                                 HMT_low();
739
740                                 if (hvlpevent_is_pending()) {
741                                         HMT_medium();
742                                         ppc64_runlatch_on();
743                                         process_iSeries_events();
744                                 }
745                         }
746
747                         HMT_medium();
748                         clear_thread_flag(TIF_POLLING_NRFLAG);
749                 } else {
750                         set_need_resched();
751                 }
752
753                 ppc64_runlatch_on();
754                 schedule();
755         }
756
757         return 0;
758 }
759
760 #ifndef CONFIG_PCI
761 void __init iSeries_init_IRQ(void) { }
762 #endif
763
764 static int __init iseries_probe(int platform)
765 {
766         return PLATFORM_ISERIES_LPAR == platform;
767 }
768
769 struct machdep_calls __initdata iseries_md = {
770         .setup_arch     = iSeries_setup_arch,
771         .get_cpuinfo    = iSeries_get_cpuinfo,
772         .init_IRQ       = iSeries_init_IRQ,
773         .get_irq        = iSeries_get_irq,
774         .init_early     = iSeries_init_early,
775         .pcibios_fixup  = iSeries_pci_final_fixup,
776         .restart        = iSeries_restart,
777         .power_off      = iSeries_power_off,
778         .halt           = iSeries_halt,
779         .get_boot_time  = iSeries_get_boot_time,
780         .set_rtc_time   = iSeries_set_rtc_time,
781         .get_rtc_time   = iSeries_get_rtc_time,
782         .calibrate_decr = generic_calibrate_decr,
783         .progress       = iSeries_progress,
784         .probe          = iseries_probe,
785         /* XXX Implement enable_pmcs for iSeries */
786 };
787
788 struct blob {
789         unsigned char data[PAGE_SIZE];
790         unsigned long next;
791 };
792
793 struct iseries_flat_dt {
794         struct boot_param_header header;
795         u64 reserve_map[2];
796         struct blob dt;
797         struct blob strings;
798 };
799
800 struct iseries_flat_dt iseries_dt;
801
802 void dt_init(struct iseries_flat_dt *dt)
803 {
804         dt->header.off_mem_rsvmap =
805                 offsetof(struct iseries_flat_dt, reserve_map);
806         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
807         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
808         dt->header.totalsize = sizeof(struct iseries_flat_dt);
809         dt->header.dt_strings_size = sizeof(struct blob);
810
811         /* There is no notion of hardware cpu id on iSeries */
812         dt->header.boot_cpuid_phys = smp_processor_id();
813
814         dt->dt.next = (unsigned long)&dt->dt.data;
815         dt->strings.next = (unsigned long)&dt->strings.data;
816
817         dt->header.magic = OF_DT_HEADER;
818         dt->header.version = 0x10;
819         dt->header.last_comp_version = 0x10;
820
821         dt->reserve_map[0] = 0;
822         dt->reserve_map[1] = 0;
823 }
824
825 void dt_check_blob(struct blob *b)
826 {
827         if (b->next >= (unsigned long)&b->next) {
828                 DBG("Ran out of space in flat device tree blob!\n");
829                 BUG();
830         }
831 }
832
833 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
834 {
835         *((u32*)dt->dt.next) = value;
836         dt->dt.next += sizeof(u32);
837
838         dt_check_blob(&dt->dt);
839 }
840
841 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
842 {
843         *((u64*)dt->dt.next) = value;
844         dt->dt.next += sizeof(u64);
845
846         dt_check_blob(&dt->dt);
847 }
848
849 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
850 {
851         unsigned long start = blob->next - (unsigned long)blob->data;
852
853         memcpy((char *)blob->next, data, len);
854         blob->next = _ALIGN(blob->next + len, 4);
855
856         dt_check_blob(blob);
857
858         return start;
859 }
860
861 void dt_start_node(struct iseries_flat_dt *dt, char *name)
862 {
863         dt_push_u32(dt, OF_DT_BEGIN_NODE);
864         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
865 }
866
867 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
868
869 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
870 {
871         unsigned long offset;
872
873         dt_push_u32(dt, OF_DT_PROP);
874
875         /* Length of the data */
876         dt_push_u32(dt, len);
877
878         /* Put the property name in the string blob. */
879         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
880
881         /* The offset of the properties name in the string blob. */
882         dt_push_u32(dt, (u32)offset);
883
884         /* The actual data. */
885         dt_push_bytes(&dt->dt, data, len);
886 }
887
888 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
889 {
890         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
891 }
892
893 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
894 {
895         dt_prop(dt, name, (char *)&data, sizeof(u32));
896 }
897
898 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
899 {
900         dt_prop(dt, name, (char *)&data, sizeof(u64));
901 }
902
903 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
904 {
905         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
906 }
907
908 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
909 {
910         dt_prop(dt, name, NULL, 0);
911 }
912
913 void dt_cpus(struct iseries_flat_dt *dt)
914 {
915         unsigned char buf[32];
916         unsigned char *p;
917         unsigned int i, index;
918         struct IoHriProcessorVpd *d;
919
920         /* yuck */
921         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
922         p = strchr(buf, ' ');
923         if (!p) p = buf + strlen(buf);
924
925         dt_start_node(dt, "cpus");
926         dt_prop_u32(dt, "#address-cells", 1);
927         dt_prop_u32(dt, "#size-cells", 0);
928
929         for (i = 0; i < NR_CPUS; i++) {
930                 if (paca[i].lppaca.dyn_proc_status >= 2)
931                         continue;
932
933                 snprintf(p, 32 - (p - buf), "@%d", i);
934                 dt_start_node(dt, buf);
935
936                 dt_prop_str(dt, "device_type", "cpu");
937
938                 index = paca[i].lppaca.dyn_hv_phys_proc_index;
939                 d = &xIoHriProcessorVpd[index];
940
941                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
942                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
943
944                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
945                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
946
947                 /* magic conversions to Hz copied from old code */
948                 dt_prop_u32(dt, "clock-frequency",
949                         ((1UL << 34) * 1000000) / d->xProcFreq);
950                 dt_prop_u32(dt, "timebase-frequency",
951                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
952
953                 dt_prop_u32(dt, "reg", i);
954
955                 dt_end_node(dt);
956         }
957
958         dt_end_node(dt);
959 }
960
961 void build_flat_dt(struct iseries_flat_dt *dt)
962 {
963         u64 tmp[2];
964
965         dt_init(dt);
966
967         dt_start_node(dt, "");
968
969         dt_prop_u32(dt, "#address-cells", 2);
970         dt_prop_u32(dt, "#size-cells", 2);
971
972         /* /memory */
973         dt_start_node(dt, "memory@0");
974         dt_prop_str(dt, "name", "memory");
975         dt_prop_str(dt, "device_type", "memory");
976         tmp[0] = 0;
977         tmp[1] = systemcfg->physicalMemorySize;
978         dt_prop_u64_list(dt, "reg", tmp, 2);
979         dt_end_node(dt);
980
981         /* /chosen */
982         dt_start_node(dt, "chosen");
983         dt_prop_u32(dt, "linux,platform", PLATFORM_ISERIES_LPAR);
984         dt_end_node(dt);
985
986         dt_cpus(dt);
987
988         dt_end_node(dt);
989
990         dt_push_u32(dt, OF_DT_END);
991 }
992
993 void * __init iSeries_early_setup(void)
994 {
995         iSeries_fixup_klimit();
996
997         /*
998          * Initialize the table which translate Linux physical addresses to
999          * AS/400 absolute addresses
1000          */
1001         build_iSeries_Memory_Map();
1002
1003         build_flat_dt(&iseries_dt);
1004
1005         return (void *) __pa(&iseries_dt);
1006 }