3ecc4a652d82e55425f6d3c3b8441266e389e316
[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 #include <linux/kernel.h>
31
32 #include <asm/processor.h>
33 #include <asm/machdep.h>
34 #include <asm/page.h>
35 #include <asm/mmu.h>
36 #include <asm/pgtable.h>
37 #include <asm/mmu_context.h>
38 #include <asm/cputable.h>
39 #include <asm/sections.h>
40 #include <asm/iommu.h>
41 #include <asm/firmware.h>
42 #include <asm/system.h>
43 #include <asm/time.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/hv_lp_config.h>
49 #include <asm/iseries/hv_call_event.h>
50 #include <asm/iseries/hv_call_xm.h>
51 #include <asm/iseries/it_lp_queue.h>
52 #include <asm/iseries/mf.h>
53 #include <asm/iseries/hv_lp_event.h>
54 #include <asm/iseries/lpar_map.h>
55 #include <asm/udbg.h>
56
57 #include "naca.h"
58 #include "setup.h"
59 #include "irq.h"
60 #include "vpd_areas.h"
61 #include "processor_vpd.h"
62 #include "main_store.h"
63 #include "call_sm.h"
64 #include "call_hpt.h"
65
66 #ifdef DEBUG
67 #define DBG(fmt...) udbg_printf(fmt)
68 #else
69 #define DBG(fmt...)
70 #endif
71
72 /* Function Prototypes */
73 static unsigned long build_iSeries_Memory_Map(void);
74 static void iseries_shared_idle(void);
75 static void iseries_dedicated_idle(void);
76 #ifdef CONFIG_PCI
77 extern void iSeries_pci_final_fixup(void);
78 #else
79 static void iSeries_pci_final_fixup(void) { }
80 #endif
81
82 /* Global Variables */
83 int piranha_simulator;
84
85 extern int rd_size;             /* Defined in drivers/block/rd.c */
86 extern unsigned long embedded_sysmap_start;
87 extern unsigned long embedded_sysmap_end;
88
89 extern unsigned long iSeries_recal_tb;
90 extern unsigned long iSeries_recal_titan;
91
92 static int mf_initialized;
93
94 static unsigned long cmd_mem_limit;
95
96 struct MemoryBlock {
97         unsigned long absStart;
98         unsigned long absEnd;
99         unsigned long logicalStart;
100         unsigned long logicalEnd;
101 };
102
103 /*
104  * Process the main store vpd to determine where the holes in memory are
105  * and return the number of physical blocks and fill in the array of
106  * block data.
107  */
108 static unsigned long iSeries_process_Condor_mainstore_vpd(
109                 struct MemoryBlock *mb_array, unsigned long max_entries)
110 {
111         unsigned long holeFirstChunk, holeSizeChunks;
112         unsigned long numMemoryBlocks = 1;
113         struct IoHriMainStoreSegment4 *msVpd =
114                 (struct IoHriMainStoreSegment4 *)xMsVpd;
115         unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
116         unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
117         unsigned long holeSize = holeEnd - holeStart;
118
119         printk("Mainstore_VPD: Condor\n");
120         /*
121          * Determine if absolute memory has any
122          * holes so that we can interpret the
123          * access map we get back from the hypervisor
124          * correctly.
125          */
126         mb_array[0].logicalStart = 0;
127         mb_array[0].logicalEnd = 0x100000000;
128         mb_array[0].absStart = 0;
129         mb_array[0].absEnd = 0x100000000;
130
131         if (holeSize) {
132                 numMemoryBlocks = 2;
133                 holeStart = holeStart & 0x000fffffffffffff;
134                 holeStart = addr_to_chunk(holeStart);
135                 holeFirstChunk = holeStart;
136                 holeSize = addr_to_chunk(holeSize);
137                 holeSizeChunks = holeSize;
138                 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
139                                 holeFirstChunk, holeSizeChunks );
140                 mb_array[0].logicalEnd = holeFirstChunk;
141                 mb_array[0].absEnd = holeFirstChunk;
142                 mb_array[1].logicalStart = holeFirstChunk;
143                 mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks;
144                 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
145                 mb_array[1].absEnd = 0x100000000;
146         }
147         return numMemoryBlocks;
148 }
149
150 #define MaxSegmentAreas                 32
151 #define MaxSegmentAdrRangeBlocks        128
152 #define MaxAreaRangeBlocks              4
153
154 static unsigned long iSeries_process_Regatta_mainstore_vpd(
155                 struct MemoryBlock *mb_array, unsigned long max_entries)
156 {
157         struct IoHriMainStoreSegment5 *msVpdP =
158                 (struct IoHriMainStoreSegment5 *)xMsVpd;
159         unsigned long numSegmentBlocks = 0;
160         u32 existsBits = msVpdP->msAreaExists;
161         unsigned long area_num;
162
163         printk("Mainstore_VPD: Regatta\n");
164
165         for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
166                 unsigned long numAreaBlocks;
167                 struct IoHriMainStoreArea4 *currentArea;
168
169                 if (existsBits & 0x80000000) {
170                         unsigned long block_num;
171
172                         currentArea = &msVpdP->msAreaArray[area_num];
173                         numAreaBlocks = currentArea->numAdrRangeBlocks;
174                         printk("ms_vpd: processing area %2ld  blocks=%ld",
175                                         area_num, numAreaBlocks);
176                         for (block_num = 0; block_num < numAreaBlocks;
177                                         ++block_num ) {
178                                 /* Process an address range block */
179                                 struct MemoryBlock tempBlock;
180                                 unsigned long i;
181
182                                 tempBlock.absStart =
183                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
184                                 tempBlock.absEnd =
185                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
186                                 tempBlock.logicalStart = 0;
187                                 tempBlock.logicalEnd   = 0;
188                                 printk("\n          block %ld absStart=%016lx absEnd=%016lx",
189                                                 block_num, tempBlock.absStart,
190                                                 tempBlock.absEnd);
191
192                                 for (i = 0; i < numSegmentBlocks; ++i) {
193                                         if (mb_array[i].absStart ==
194                                                         tempBlock.absStart)
195                                                 break;
196                                 }
197                                 if (i == numSegmentBlocks) {
198                                         if (numSegmentBlocks == max_entries)
199                                                 panic("iSeries_process_mainstore_vpd: too many memory blocks");
200                                         mb_array[numSegmentBlocks] = tempBlock;
201                                         ++numSegmentBlocks;
202                                 } else
203                                         printk(" (duplicate)");
204                         }
205                         printk("\n");
206                 }
207                 existsBits <<= 1;
208         }
209         /* Now sort the blocks found into ascending sequence */
210         if (numSegmentBlocks > 1) {
211                 unsigned long m, n;
212
213                 for (m = 0; m < numSegmentBlocks - 1; ++m) {
214                         for (n = numSegmentBlocks - 1; m < n; --n) {
215                                 if (mb_array[n].absStart <
216                                                 mb_array[n-1].absStart) {
217                                         struct MemoryBlock tempBlock;
218
219                                         tempBlock = mb_array[n];
220                                         mb_array[n] = mb_array[n-1];
221                                         mb_array[n-1] = tempBlock;
222                                 }
223                         }
224                 }
225         }
226         /*
227          * Assign "logical" addresses to each block.  These
228          * addresses correspond to the hypervisor "bitmap" space.
229          * Convert all addresses into units of 256K chunks.
230          */
231         {
232         unsigned long i, nextBitmapAddress;
233
234         printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
235         nextBitmapAddress = 0;
236         for (i = 0; i < numSegmentBlocks; ++i) {
237                 unsigned long length = mb_array[i].absEnd -
238                         mb_array[i].absStart;
239
240                 mb_array[i].logicalStart = nextBitmapAddress;
241                 mb_array[i].logicalEnd = nextBitmapAddress + length;
242                 nextBitmapAddress += length;
243                 printk("          Bitmap range: %016lx - %016lx\n"
244                                 "        Absolute range: %016lx - %016lx\n",
245                                 mb_array[i].logicalStart,
246                                 mb_array[i].logicalEnd,
247                                 mb_array[i].absStart, mb_array[i].absEnd);
248                 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
249                                 0x000fffffffffffff);
250                 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
251                                 0x000fffffffffffff);
252                 mb_array[i].logicalStart =
253                         addr_to_chunk(mb_array[i].logicalStart);
254                 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
255         }
256         }
257
258         return numSegmentBlocks;
259 }
260
261 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
262                 unsigned long max_entries)
263 {
264         unsigned long i;
265         unsigned long mem_blocks = 0;
266
267         if (cpu_has_feature(CPU_FTR_SLB))
268                 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
269                                 max_entries);
270         else
271                 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
272                                 max_entries);
273
274         printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
275         for (i = 0; i < mem_blocks; ++i) {
276                 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
277                        "                             abs chunks %016lx - %016lx\n",
278                         i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
279                         mb_array[i].absStart, mb_array[i].absEnd);
280         }
281         return mem_blocks;
282 }
283
284 static void __init iSeries_get_cmdline(void)
285 {
286         char *p, *q;
287
288         /* copy the command line parameter from the primary VSP  */
289         HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
290                         HvLpDma_Direction_RemoteToLocal);
291
292         p = cmd_line;
293         q = cmd_line + 255;
294         while(p < q) {
295                 if (!*p || *p == '\n')
296                         break;
297                 ++p;
298         }
299         *p = 0;
300 }
301
302 static void __init iSeries_init_early(void)
303 {
304         DBG(" -> iSeries_init_early()\n");
305
306         ppc64_firmware_features = FW_FEATURE_ISERIES;
307
308         ppc64_interrupt_controller = IC_ISERIES;
309
310 #if defined(CONFIG_BLK_DEV_INITRD)
311         /*
312          * If the init RAM disk has been configured and there is
313          * a non-zero starting address for it, set it up
314          */
315         if (naca.xRamDisk) {
316                 initrd_start = (unsigned long)__va(naca.xRamDisk);
317                 initrd_end = initrd_start + naca.xRamDiskSize * HW_PAGE_SIZE;
318                 initrd_below_start_ok = 1;      // ramdisk in kernel space
319                 ROOT_DEV = Root_RAM0;
320                 if (((rd_size * 1024) / HW_PAGE_SIZE) < naca.xRamDiskSize)
321                         rd_size = (naca.xRamDiskSize * HW_PAGE_SIZE) / 1024;
322         } else
323 #endif /* CONFIG_BLK_DEV_INITRD */
324         {
325             /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
326         }
327
328         iSeries_recal_tb = get_tb();
329         iSeries_recal_titan = HvCallXm_loadTod();
330
331         /*
332          * Initialize the hash table management pointers
333          */
334         hpte_init_iSeries();
335
336         /*
337          * Initialize the DMA/TCE management
338          */
339         iommu_init_early_iSeries();
340
341         /* Initialize machine-dependency vectors */
342 #ifdef CONFIG_SMP
343         smp_init_iSeries();
344 #endif
345         if (itLpNaca.xPirEnvironMode == 0)
346                 piranha_simulator = 1;
347
348         /* Associate Lp Event Queue 0 with processor 0 */
349         HvCallEvent_setLpEventQueueInterruptProc(0, 0);
350
351         mf_init();
352         mf_initialized = 1;
353         mb();
354
355         /* If we were passed an initrd, set the ROOT_DEV properly if the values
356          * look sensible. If not, clear initrd reference.
357          */
358 #ifdef CONFIG_BLK_DEV_INITRD
359         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
360             initrd_end > initrd_start)
361                 ROOT_DEV = Root_RAM0;
362         else
363                 initrd_start = initrd_end = 0;
364 #endif /* CONFIG_BLK_DEV_INITRD */
365
366         DBG(" <- iSeries_init_early()\n");
367 }
368
369 struct mschunks_map mschunks_map = {
370         /* XXX We don't use these, but Piranha might need them. */
371         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
372         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
373         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
374 };
375 EXPORT_SYMBOL(mschunks_map);
376
377 void mschunks_alloc(unsigned long num_chunks)
378 {
379         klimit = _ALIGN(klimit, sizeof(u32));
380         mschunks_map.mapping = (u32 *)klimit;
381         klimit += num_chunks * sizeof(u32);
382         mschunks_map.num_chunks = num_chunks;
383 }
384
385 /*
386  * The iSeries may have very large memories ( > 128 GB ) and a partition
387  * may get memory in "chunks" that may be anywhere in the 2**52 real
388  * address space.  The chunks are 256K in size.  To map this to the
389  * memory model Linux expects, the AS/400 specific code builds a
390  * translation table to translate what Linux thinks are "physical"
391  * addresses to the actual real addresses.  This allows us to make
392  * it appear to Linux that we have contiguous memory starting at
393  * physical address zero while in fact this could be far from the truth.
394  * To avoid confusion, I'll let the words physical and/or real address
395  * apply to the Linux addresses while I'll use "absolute address" to
396  * refer to the actual hardware real address.
397  *
398  * build_iSeries_Memory_Map gets information from the Hypervisor and
399  * looks at the Main Store VPD to determine the absolute addresses
400  * of the memory that has been assigned to our partition and builds
401  * a table used to translate Linux's physical addresses to these
402  * absolute addresses.  Absolute addresses are needed when
403  * communicating with the hypervisor (e.g. to build HPT entries)
404  *
405  * Returns the physical memory size
406  */
407
408 static unsigned long __init build_iSeries_Memory_Map(void)
409 {
410         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
411         u32 nextPhysChunk;
412         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
413         u32 totalChunks,moreChunks;
414         u32 currChunk, thisChunk, absChunk;
415         u32 currDword;
416         u32 chunkBit;
417         u64 map;
418         struct MemoryBlock mb[32];
419         unsigned long numMemoryBlocks, curBlock;
420
421         /* Chunk size on iSeries is 256K bytes */
422         totalChunks = (u32)HvLpConfig_getMsChunks();
423         mschunks_alloc(totalChunks);
424
425         /*
426          * Get absolute address of our load area
427          * and map it to physical address 0
428          * This guarantees that the loadarea ends up at physical 0
429          * otherwise, it might not be returned by PLIC as the first
430          * chunks
431          */
432
433         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
434         loadAreaSize =  itLpNaca.xLoadAreaChunks;
435
436         /*
437          * Only add the pages already mapped here.
438          * Otherwise we might add the hpt pages
439          * The rest of the pages of the load area
440          * aren't in the HPT yet and can still
441          * be assigned an arbitrary physical address
442          */
443         if ((loadAreaSize * 64) > HvPagesToMap)
444                 loadAreaSize = HvPagesToMap / 64;
445
446         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
447
448         /*
449          * TODO Do we need to do something if the HPT is in the 64MB load area?
450          * This would be required if the itLpNaca.xLoadAreaChunks includes
451          * the HPT size
452          */
453
454         printk("Mapping load area - physical addr = 0000000000000000\n"
455                 "                    absolute addr = %016lx\n",
456                 chunk_to_addr(loadAreaFirstChunk));
457         printk("Load area size %dK\n", loadAreaSize * 256);
458
459         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
460                 mschunks_map.mapping[nextPhysChunk] =
461                         loadAreaFirstChunk + nextPhysChunk;
462
463         /*
464          * Get absolute address of our HPT and remember it so
465          * we won't map it to any physical address
466          */
467         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
468         hptSizePages = (u32)HvCallHpt_getHptPages();
469         hptSizeChunks = hptSizePages >>
470                 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
471         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
472
473         printk("HPT absolute addr = %016lx, size = %dK\n",
474                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
475
476         /*
477          * Determine if absolute memory has any
478          * holes so that we can interpret the
479          * access map we get back from the hypervisor
480          * correctly.
481          */
482         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
483
484         /*
485          * Process the main store access map from the hypervisor
486          * to build up our physical -> absolute translation table
487          */
488         curBlock = 0;
489         currChunk = 0;
490         currDword = 0;
491         moreChunks = totalChunks;
492
493         while (moreChunks) {
494                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
495                                 currDword);
496                 thisChunk = currChunk;
497                 while (map) {
498                         chunkBit = map >> 63;
499                         map <<= 1;
500                         if (chunkBit) {
501                                 --moreChunks;
502                                 while (thisChunk >= mb[curBlock].logicalEnd) {
503                                         ++curBlock;
504                                         if (curBlock >= numMemoryBlocks)
505                                                 panic("out of memory blocks");
506                                 }
507                                 if (thisChunk < mb[curBlock].logicalStart)
508                                         panic("memory block error");
509
510                                 absChunk = mb[curBlock].absStart +
511                                         (thisChunk - mb[curBlock].logicalStart);
512                                 if (((absChunk < hptFirstChunk) ||
513                                      (absChunk > hptLastChunk)) &&
514                                     ((absChunk < loadAreaFirstChunk) ||
515                                      (absChunk > loadAreaLastChunk))) {
516                                         mschunks_map.mapping[nextPhysChunk] =
517                                                 absChunk;
518                                         ++nextPhysChunk;
519                                 }
520                         }
521                         ++thisChunk;
522                 }
523                 ++currDword;
524                 currChunk += 64;
525         }
526
527         /*
528          * main store size (in chunks) is
529          *   totalChunks - hptSizeChunks
530          * which should be equal to
531          *   nextPhysChunk
532          */
533         return chunk_to_addr(nextPhysChunk);
534 }
535
536 /*
537  * Document me.
538  */
539 static void __init iSeries_setup_arch(void)
540 {
541         if (get_lppaca()->shared_proc) {
542                 ppc_md.idle_loop = iseries_shared_idle;
543                 printk(KERN_INFO "Using shared processor idle loop\n");
544         } else {
545                 ppc_md.idle_loop = iseries_dedicated_idle;
546                 printk(KERN_INFO "Using dedicated idle loop\n");
547         }
548
549         /* Setup the Lp Event Queue */
550         setup_hvlpevent_queue();
551
552         printk("Max  logical processors = %d\n",
553                         itVpdAreas.xSlicMaxLogicalProcs);
554         printk("Max physical processors = %d\n",
555                         itVpdAreas.xSlicMaxPhysicalProcs);
556 }
557
558 static void iSeries_show_cpuinfo(struct seq_file *m)
559 {
560         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
561 }
562
563 /*
564  * Document me.
565  */
566 static void iSeries_restart(char *cmd)
567 {
568         mf_reboot();
569 }
570
571 /*
572  * Document me.
573  */
574 static void iSeries_power_off(void)
575 {
576         mf_power_off();
577 }
578
579 /*
580  * Document me.
581  */
582 static void iSeries_halt(void)
583 {
584         mf_power_off();
585 }
586
587 static void __init iSeries_progress(char * st, unsigned short code)
588 {
589         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
590         if (!piranha_simulator && mf_initialized) {
591                 if (code != 0xffff)
592                         mf_display_progress(code);
593                 else
594                         mf_clear_src();
595         }
596 }
597
598 static void __init iSeries_fixup_klimit(void)
599 {
600         /*
601          * Change klimit to take into account any ram disk
602          * that may be included
603          */
604         if (naca.xRamDisk)
605                 klimit = KERNELBASE + (u64)naca.xRamDisk +
606                         (naca.xRamDiskSize * HW_PAGE_SIZE);
607         else {
608                 /*
609                  * No ram disk was included - check and see if there
610                  * was an embedded system map.  Change klimit to take
611                  * into account any embedded system map
612                  */
613                 if (embedded_sysmap_end)
614                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
615                                         0xfffffffffffff000);
616         }
617 }
618
619 static int __init iSeries_src_init(void)
620 {
621         /* clear the progress line */
622         ppc_md.progress(" ", 0xffff);
623         return 0;
624 }
625
626 late_initcall(iSeries_src_init);
627
628 static inline void process_iSeries_events(void)
629 {
630         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
631 }
632
633 static void yield_shared_processor(void)
634 {
635         unsigned long tb;
636
637         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
638                                     HvCall_MaskLpEvent |
639                                     HvCall_MaskLpProd |
640                                     HvCall_MaskTimeout);
641
642         tb = get_tb();
643         /* Compute future tb value when yield should expire */
644         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
645
646         /*
647          * The decrementer stops during the yield.  Force a fake decrementer
648          * here and let the timer_interrupt code sort out the actual time.
649          */
650         get_lppaca()->int_dword.fields.decr_int = 1;
651         ppc64_runlatch_on();
652         process_iSeries_events();
653 }
654
655 static void iseries_shared_idle(void)
656 {
657         while (1) {
658                 while (!need_resched() && !hvlpevent_is_pending()) {
659                         local_irq_disable();
660                         ppc64_runlatch_off();
661
662                         /* Recheck with irqs off */
663                         if (!need_resched() && !hvlpevent_is_pending())
664                                 yield_shared_processor();
665
666                         HMT_medium();
667                         local_irq_enable();
668                 }
669
670                 ppc64_runlatch_on();
671
672                 if (hvlpevent_is_pending())
673                         process_iSeries_events();
674
675                 preempt_enable_no_resched();
676                 schedule();
677                 preempt_disable();
678         }
679 }
680
681 static void iseries_dedicated_idle(void)
682 {
683         set_thread_flag(TIF_POLLING_NRFLAG);
684
685         while (1) {
686                 if (!need_resched()) {
687                         while (!need_resched()) {
688                                 ppc64_runlatch_off();
689                                 HMT_low();
690
691                                 if (hvlpevent_is_pending()) {
692                                         HMT_medium();
693                                         ppc64_runlatch_on();
694                                         process_iSeries_events();
695                                 }
696                         }
697
698                         HMT_medium();
699                 }
700
701                 ppc64_runlatch_on();
702                 preempt_enable_no_resched();
703                 schedule();
704                 preempt_disable();
705         }
706 }
707
708 #ifndef CONFIG_PCI
709 void __init iSeries_init_IRQ(void) { }
710 #endif
711
712 static int __init iseries_probe(int platform)
713 {
714         return PLATFORM_ISERIES_LPAR == platform;
715 }
716
717 struct machdep_calls __initdata iseries_md = {
718         .setup_arch     = iSeries_setup_arch,
719         .show_cpuinfo   = iSeries_show_cpuinfo,
720         .init_IRQ       = iSeries_init_IRQ,
721         .get_irq        = iSeries_get_irq,
722         .init_early     = iSeries_init_early,
723         .pcibios_fixup  = iSeries_pci_final_fixup,
724         .restart        = iSeries_restart,
725         .power_off      = iSeries_power_off,
726         .halt           = iSeries_halt,
727         .get_boot_time  = iSeries_get_boot_time,
728         .set_rtc_time   = iSeries_set_rtc_time,
729         .get_rtc_time   = iSeries_get_rtc_time,
730         .calibrate_decr = generic_calibrate_decr,
731         .progress       = iSeries_progress,
732         .probe          = iseries_probe,
733         /* XXX Implement enable_pmcs for iSeries */
734 };
735
736 struct blob {
737         unsigned char data[PAGE_SIZE];
738         unsigned long next;
739 };
740
741 struct iseries_flat_dt {
742         struct boot_param_header header;
743         u64 reserve_map[2];
744         struct blob dt;
745         struct blob strings;
746 };
747
748 struct iseries_flat_dt iseries_dt;
749
750 void dt_init(struct iseries_flat_dt *dt)
751 {
752         dt->header.off_mem_rsvmap =
753                 offsetof(struct iseries_flat_dt, reserve_map);
754         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
755         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
756         dt->header.totalsize = sizeof(struct iseries_flat_dt);
757         dt->header.dt_strings_size = sizeof(struct blob);
758
759         /* There is no notion of hardware cpu id on iSeries */
760         dt->header.boot_cpuid_phys = smp_processor_id();
761
762         dt->dt.next = (unsigned long)&dt->dt.data;
763         dt->strings.next = (unsigned long)&dt->strings.data;
764
765         dt->header.magic = OF_DT_HEADER;
766         dt->header.version = 0x10;
767         dt->header.last_comp_version = 0x10;
768
769         dt->reserve_map[0] = 0;
770         dt->reserve_map[1] = 0;
771 }
772
773 void dt_check_blob(struct blob *b)
774 {
775         if (b->next >= (unsigned long)&b->next) {
776                 DBG("Ran out of space in flat device tree blob!\n");
777                 BUG();
778         }
779 }
780
781 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
782 {
783         *((u32*)dt->dt.next) = value;
784         dt->dt.next += sizeof(u32);
785
786         dt_check_blob(&dt->dt);
787 }
788
789 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
790 {
791         *((u64*)dt->dt.next) = value;
792         dt->dt.next += sizeof(u64);
793
794         dt_check_blob(&dt->dt);
795 }
796
797 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
798 {
799         unsigned long start = blob->next - (unsigned long)blob->data;
800
801         memcpy((char *)blob->next, data, len);
802         blob->next = _ALIGN(blob->next + len, 4);
803
804         dt_check_blob(blob);
805
806         return start;
807 }
808
809 void dt_start_node(struct iseries_flat_dt *dt, char *name)
810 {
811         dt_push_u32(dt, OF_DT_BEGIN_NODE);
812         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
813 }
814
815 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
816
817 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
818 {
819         unsigned long offset;
820
821         dt_push_u32(dt, OF_DT_PROP);
822
823         /* Length of the data */
824         dt_push_u32(dt, len);
825
826         /* Put the property name in the string blob. */
827         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
828
829         /* The offset of the properties name in the string blob. */
830         dt_push_u32(dt, (u32)offset);
831
832         /* The actual data. */
833         dt_push_bytes(&dt->dt, data, len);
834 }
835
836 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
837 {
838         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
839 }
840
841 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
842 {
843         dt_prop(dt, name, (char *)&data, sizeof(u32));
844 }
845
846 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
847 {
848         dt_prop(dt, name, (char *)&data, sizeof(u64));
849 }
850
851 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
852 {
853         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
854 }
855
856 void dt_prop_u32_list(struct iseries_flat_dt *dt, char *name, u32 *data, int n)
857 {
858         dt_prop(dt, name, (char *)data, sizeof(u32) * n);
859 }
860
861 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
862 {
863         dt_prop(dt, name, NULL, 0);
864 }
865
866 void dt_cpus(struct iseries_flat_dt *dt)
867 {
868         unsigned char buf[32];
869         unsigned char *p;
870         unsigned int i, index;
871         struct IoHriProcessorVpd *d;
872         u32 pft_size[2];
873
874         /* yuck */
875         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
876         p = strchr(buf, ' ');
877         if (!p) p = buf + strlen(buf);
878
879         dt_start_node(dt, "cpus");
880         dt_prop_u32(dt, "#address-cells", 1);
881         dt_prop_u32(dt, "#size-cells", 0);
882
883         pft_size[0] = 0; /* NUMA CEC cookie, 0 for non NUMA  */
884         pft_size[1] = __ilog2(HvCallHpt_getHptPages() * HW_PAGE_SIZE);
885
886         for (i = 0; i < NR_CPUS; i++) {
887                 if (lppaca[i].dyn_proc_status >= 2)
888                         continue;
889
890                 snprintf(p, 32 - (p - buf), "@%d", i);
891                 dt_start_node(dt, buf);
892
893                 dt_prop_str(dt, "device_type", "cpu");
894
895                 index = lppaca[i].dyn_hv_phys_proc_index;
896                 d = &xIoHriProcessorVpd[index];
897
898                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
899                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
900
901                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
902                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
903
904                 /* magic conversions to Hz copied from old code */
905                 dt_prop_u32(dt, "clock-frequency",
906                         ((1UL << 34) * 1000000) / d->xProcFreq);
907                 dt_prop_u32(dt, "timebase-frequency",
908                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
909
910                 dt_prop_u32(dt, "reg", i);
911
912                 dt_prop_u32_list(dt, "ibm,pft-size", pft_size, 2);
913
914                 dt_end_node(dt);
915         }
916
917         dt_end_node(dt);
918 }
919
920 void build_flat_dt(struct iseries_flat_dt *dt, unsigned long phys_mem_size)
921 {
922         u64 tmp[2];
923
924         dt_init(dt);
925
926         dt_start_node(dt, "");
927
928         dt_prop_u32(dt, "#address-cells", 2);
929         dt_prop_u32(dt, "#size-cells", 2);
930
931         /* /memory */
932         dt_start_node(dt, "memory@0");
933         dt_prop_str(dt, "name", "memory");
934         dt_prop_str(dt, "device_type", "memory");
935         tmp[0] = 0;
936         tmp[1] = phys_mem_size;
937         dt_prop_u64_list(dt, "reg", tmp, 2);
938         dt_end_node(dt);
939
940         /* /chosen */
941         dt_start_node(dt, "chosen");
942         dt_prop_u32(dt, "linux,platform", PLATFORM_ISERIES_LPAR);
943         if (cmd_mem_limit)
944                 dt_prop_u64(dt, "linux,memory-limit", cmd_mem_limit);
945         dt_end_node(dt);
946
947         dt_cpus(dt);
948
949         dt_end_node(dt);
950
951         dt_push_u32(dt, OF_DT_END);
952 }
953
954 void * __init iSeries_early_setup(void)
955 {
956         unsigned long phys_mem_size;
957
958         iSeries_fixup_klimit();
959
960         /*
961          * Initialize the table which translate Linux physical addresses to
962          * AS/400 absolute addresses
963          */
964         phys_mem_size = build_iSeries_Memory_Map();
965
966         iSeries_get_cmdline();
967
968         /* Save unparsed command line copy for /proc/cmdline */
969         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
970
971         /* Parse early parameters, in particular mem=x */
972         parse_early_param();
973
974         build_flat_dt(&iseries_dt, phys_mem_size);
975
976         return (void *) __pa(&iseries_dt);
977 }
978
979 /*
980  * On iSeries we just parse the mem=X option from the command line.
981  * On pSeries it's a bit more complicated, see prom_init_mem()
982  */
983 static int __init early_parsemem(char *p)
984 {
985         if (p)
986                 cmd_mem_limit = ALIGN(memparse(p, &p), PAGE_SIZE);
987         return 0;
988 }
989 early_param("mem", early_parsemem);
990
991 static void hvputc(char c)
992 {
993         if (c == '\n')
994                 hvputc('\r');
995
996         HvCall_writeLogBuffer(&c, 1);
997 }
998
999 void __init udbg_init_iseries(void)
1000 {
1001         udbg_putc = hvputc;
1002 }