[PATCH] powerpc: Replace platform_is_lpar() with a firmware feature
[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_interrupt_controller = IC_ISERIES;
307
308 #if defined(CONFIG_BLK_DEV_INITRD)
309         /*
310          * If the init RAM disk has been configured and there is
311          * a non-zero starting address for it, set it up
312          */
313         if (naca.xRamDisk) {
314                 initrd_start = (unsigned long)__va(naca.xRamDisk);
315                 initrd_end = initrd_start + naca.xRamDiskSize * HW_PAGE_SIZE;
316                 initrd_below_start_ok = 1;      // ramdisk in kernel space
317                 ROOT_DEV = Root_RAM0;
318                 if (((rd_size * 1024) / HW_PAGE_SIZE) < naca.xRamDiskSize)
319                         rd_size = (naca.xRamDiskSize * HW_PAGE_SIZE) / 1024;
320         } else
321 #endif /* CONFIG_BLK_DEV_INITRD */
322         {
323             /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
324         }
325
326         iSeries_recal_tb = get_tb();
327         iSeries_recal_titan = HvCallXm_loadTod();
328
329         /*
330          * Initialize the hash table management pointers
331          */
332         hpte_init_iSeries();
333
334         /*
335          * Initialize the DMA/TCE management
336          */
337         iommu_init_early_iSeries();
338
339         /* Initialize machine-dependency vectors */
340 #ifdef CONFIG_SMP
341         smp_init_iSeries();
342 #endif
343         if (itLpNaca.xPirEnvironMode == 0)
344                 piranha_simulator = 1;
345
346         /* Associate Lp Event Queue 0 with processor 0 */
347         HvCallEvent_setLpEventQueueInterruptProc(0, 0);
348
349         mf_init();
350         mf_initialized = 1;
351         mb();
352
353         /* If we were passed an initrd, set the ROOT_DEV properly if the values
354          * look sensible. If not, clear initrd reference.
355          */
356 #ifdef CONFIG_BLK_DEV_INITRD
357         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
358             initrd_end > initrd_start)
359                 ROOT_DEV = Root_RAM0;
360         else
361                 initrd_start = initrd_end = 0;
362 #endif /* CONFIG_BLK_DEV_INITRD */
363
364         DBG(" <- iSeries_init_early()\n");
365 }
366
367 struct mschunks_map mschunks_map = {
368         /* XXX We don't use these, but Piranha might need them. */
369         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
370         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
371         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
372 };
373 EXPORT_SYMBOL(mschunks_map);
374
375 void mschunks_alloc(unsigned long num_chunks)
376 {
377         klimit = _ALIGN(klimit, sizeof(u32));
378         mschunks_map.mapping = (u32 *)klimit;
379         klimit += num_chunks * sizeof(u32);
380         mschunks_map.num_chunks = num_chunks;
381 }
382
383 /*
384  * The iSeries may have very large memories ( > 128 GB ) and a partition
385  * may get memory in "chunks" that may be anywhere in the 2**52 real
386  * address space.  The chunks are 256K in size.  To map this to the
387  * memory model Linux expects, the AS/400 specific code builds a
388  * translation table to translate what Linux thinks are "physical"
389  * addresses to the actual real addresses.  This allows us to make
390  * it appear to Linux that we have contiguous memory starting at
391  * physical address zero while in fact this could be far from the truth.
392  * To avoid confusion, I'll let the words physical and/or real address
393  * apply to the Linux addresses while I'll use "absolute address" to
394  * refer to the actual hardware real address.
395  *
396  * build_iSeries_Memory_Map gets information from the Hypervisor and
397  * looks at the Main Store VPD to determine the absolute addresses
398  * of the memory that has been assigned to our partition and builds
399  * a table used to translate Linux's physical addresses to these
400  * absolute addresses.  Absolute addresses are needed when
401  * communicating with the hypervisor (e.g. to build HPT entries)
402  *
403  * Returns the physical memory size
404  */
405
406 static unsigned long __init build_iSeries_Memory_Map(void)
407 {
408         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
409         u32 nextPhysChunk;
410         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
411         u32 totalChunks,moreChunks;
412         u32 currChunk, thisChunk, absChunk;
413         u32 currDword;
414         u32 chunkBit;
415         u64 map;
416         struct MemoryBlock mb[32];
417         unsigned long numMemoryBlocks, curBlock;
418
419         /* Chunk size on iSeries is 256K bytes */
420         totalChunks = (u32)HvLpConfig_getMsChunks();
421         mschunks_alloc(totalChunks);
422
423         /*
424          * Get absolute address of our load area
425          * and map it to physical address 0
426          * This guarantees that the loadarea ends up at physical 0
427          * otherwise, it might not be returned by PLIC as the first
428          * chunks
429          */
430
431         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
432         loadAreaSize =  itLpNaca.xLoadAreaChunks;
433
434         /*
435          * Only add the pages already mapped here.
436          * Otherwise we might add the hpt pages
437          * The rest of the pages of the load area
438          * aren't in the HPT yet and can still
439          * be assigned an arbitrary physical address
440          */
441         if ((loadAreaSize * 64) > HvPagesToMap)
442                 loadAreaSize = HvPagesToMap / 64;
443
444         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
445
446         /*
447          * TODO Do we need to do something if the HPT is in the 64MB load area?
448          * This would be required if the itLpNaca.xLoadAreaChunks includes
449          * the HPT size
450          */
451
452         printk("Mapping load area - physical addr = 0000000000000000\n"
453                 "                    absolute addr = %016lx\n",
454                 chunk_to_addr(loadAreaFirstChunk));
455         printk("Load area size %dK\n", loadAreaSize * 256);
456
457         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
458                 mschunks_map.mapping[nextPhysChunk] =
459                         loadAreaFirstChunk + nextPhysChunk;
460
461         /*
462          * Get absolute address of our HPT and remember it so
463          * we won't map it to any physical address
464          */
465         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
466         hptSizePages = (u32)HvCallHpt_getHptPages();
467         hptSizeChunks = hptSizePages >>
468                 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
469         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
470
471         printk("HPT absolute addr = %016lx, size = %dK\n",
472                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
473
474         /*
475          * Determine if absolute memory has any
476          * holes so that we can interpret the
477          * access map we get back from the hypervisor
478          * correctly.
479          */
480         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
481
482         /*
483          * Process the main store access map from the hypervisor
484          * to build up our physical -> absolute translation table
485          */
486         curBlock = 0;
487         currChunk = 0;
488         currDword = 0;
489         moreChunks = totalChunks;
490
491         while (moreChunks) {
492                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
493                                 currDword);
494                 thisChunk = currChunk;
495                 while (map) {
496                         chunkBit = map >> 63;
497                         map <<= 1;
498                         if (chunkBit) {
499                                 --moreChunks;
500                                 while (thisChunk >= mb[curBlock].logicalEnd) {
501                                         ++curBlock;
502                                         if (curBlock >= numMemoryBlocks)
503                                                 panic("out of memory blocks");
504                                 }
505                                 if (thisChunk < mb[curBlock].logicalStart)
506                                         panic("memory block error");
507
508                                 absChunk = mb[curBlock].absStart +
509                                         (thisChunk - mb[curBlock].logicalStart);
510                                 if (((absChunk < hptFirstChunk) ||
511                                      (absChunk > hptLastChunk)) &&
512                                     ((absChunk < loadAreaFirstChunk) ||
513                                      (absChunk > loadAreaLastChunk))) {
514                                         mschunks_map.mapping[nextPhysChunk] =
515                                                 absChunk;
516                                         ++nextPhysChunk;
517                                 }
518                         }
519                         ++thisChunk;
520                 }
521                 ++currDword;
522                 currChunk += 64;
523         }
524
525         /*
526          * main store size (in chunks) is
527          *   totalChunks - hptSizeChunks
528          * which should be equal to
529          *   nextPhysChunk
530          */
531         return chunk_to_addr(nextPhysChunk);
532 }
533
534 /*
535  * Document me.
536  */
537 static void __init iSeries_setup_arch(void)
538 {
539         if (get_lppaca()->shared_proc) {
540                 ppc_md.idle_loop = iseries_shared_idle;
541                 printk(KERN_INFO "Using shared processor idle loop\n");
542         } else {
543                 ppc_md.idle_loop = iseries_dedicated_idle;
544                 printk(KERN_INFO "Using dedicated idle loop\n");
545         }
546
547         /* Setup the Lp Event Queue */
548         setup_hvlpevent_queue();
549
550         printk("Max  logical processors = %d\n",
551                         itVpdAreas.xSlicMaxLogicalProcs);
552         printk("Max physical processors = %d\n",
553                         itVpdAreas.xSlicMaxPhysicalProcs);
554 }
555
556 static void iSeries_show_cpuinfo(struct seq_file *m)
557 {
558         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
559 }
560
561 /*
562  * Document me.
563  */
564 static void iSeries_restart(char *cmd)
565 {
566         mf_reboot();
567 }
568
569 /*
570  * Document me.
571  */
572 static void iSeries_power_off(void)
573 {
574         mf_power_off();
575 }
576
577 /*
578  * Document me.
579  */
580 static void iSeries_halt(void)
581 {
582         mf_power_off();
583 }
584
585 static void __init iSeries_progress(char * st, unsigned short code)
586 {
587         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
588         if (!piranha_simulator && mf_initialized) {
589                 if (code != 0xffff)
590                         mf_display_progress(code);
591                 else
592                         mf_clear_src();
593         }
594 }
595
596 static void __init iSeries_fixup_klimit(void)
597 {
598         /*
599          * Change klimit to take into account any ram disk
600          * that may be included
601          */
602         if (naca.xRamDisk)
603                 klimit = KERNELBASE + (u64)naca.xRamDisk +
604                         (naca.xRamDiskSize * HW_PAGE_SIZE);
605         else {
606                 /*
607                  * No ram disk was included - check and see if there
608                  * was an embedded system map.  Change klimit to take
609                  * into account any embedded system map
610                  */
611                 if (embedded_sysmap_end)
612                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
613                                         0xfffffffffffff000);
614         }
615 }
616
617 static int __init iSeries_src_init(void)
618 {
619         /* clear the progress line */
620         ppc_md.progress(" ", 0xffff);
621         return 0;
622 }
623
624 late_initcall(iSeries_src_init);
625
626 static inline void process_iSeries_events(void)
627 {
628         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
629 }
630
631 static void yield_shared_processor(void)
632 {
633         unsigned long tb;
634
635         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
636                                     HvCall_MaskLpEvent |
637                                     HvCall_MaskLpProd |
638                                     HvCall_MaskTimeout);
639
640         tb = get_tb();
641         /* Compute future tb value when yield should expire */
642         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
643
644         /*
645          * The decrementer stops during the yield.  Force a fake decrementer
646          * here and let the timer_interrupt code sort out the actual time.
647          */
648         get_lppaca()->int_dword.fields.decr_int = 1;
649         ppc64_runlatch_on();
650         process_iSeries_events();
651 }
652
653 static void iseries_shared_idle(void)
654 {
655         while (1) {
656                 while (!need_resched() && !hvlpevent_is_pending()) {
657                         local_irq_disable();
658                         ppc64_runlatch_off();
659
660                         /* Recheck with irqs off */
661                         if (!need_resched() && !hvlpevent_is_pending())
662                                 yield_shared_processor();
663
664                         HMT_medium();
665                         local_irq_enable();
666                 }
667
668                 ppc64_runlatch_on();
669
670                 if (hvlpevent_is_pending())
671                         process_iSeries_events();
672
673                 preempt_enable_no_resched();
674                 schedule();
675                 preempt_disable();
676         }
677 }
678
679 static void iseries_dedicated_idle(void)
680 {
681         set_thread_flag(TIF_POLLING_NRFLAG);
682
683         while (1) {
684                 if (!need_resched()) {
685                         while (!need_resched()) {
686                                 ppc64_runlatch_off();
687                                 HMT_low();
688
689                                 if (hvlpevent_is_pending()) {
690                                         HMT_medium();
691                                         ppc64_runlatch_on();
692                                         process_iSeries_events();
693                                 }
694                         }
695
696                         HMT_medium();
697                 }
698
699                 ppc64_runlatch_on();
700                 preempt_enable_no_resched();
701                 schedule();
702                 preempt_disable();
703         }
704 }
705
706 #ifndef CONFIG_PCI
707 void __init iSeries_init_IRQ(void) { }
708 #endif
709
710 static int __init iseries_probe(int platform)
711 {
712         if (PLATFORM_ISERIES_LPAR != platform)
713                 return 0;
714
715         ppc64_firmware_features |= FW_FEATURE_ISERIES;
716         ppc64_firmware_features |= FW_FEATURE_LPAR;
717
718         return 1;
719 }
720
721 struct machdep_calls __initdata iseries_md = {
722         .setup_arch     = iSeries_setup_arch,
723         .show_cpuinfo   = iSeries_show_cpuinfo,
724         .init_IRQ       = iSeries_init_IRQ,
725         .get_irq        = iSeries_get_irq,
726         .init_early     = iSeries_init_early,
727         .pcibios_fixup  = iSeries_pci_final_fixup,
728         .restart        = iSeries_restart,
729         .power_off      = iSeries_power_off,
730         .halt           = iSeries_halt,
731         .get_boot_time  = iSeries_get_boot_time,
732         .set_rtc_time   = iSeries_set_rtc_time,
733         .get_rtc_time   = iSeries_get_rtc_time,
734         .calibrate_decr = generic_calibrate_decr,
735         .progress       = iSeries_progress,
736         .probe          = iseries_probe,
737         /* XXX Implement enable_pmcs for iSeries */
738 };
739
740 struct blob {
741         unsigned char data[PAGE_SIZE];
742         unsigned long next;
743 };
744
745 struct iseries_flat_dt {
746         struct boot_param_header header;
747         u64 reserve_map[2];
748         struct blob dt;
749         struct blob strings;
750 };
751
752 struct iseries_flat_dt iseries_dt;
753
754 void dt_init(struct iseries_flat_dt *dt)
755 {
756         dt->header.off_mem_rsvmap =
757                 offsetof(struct iseries_flat_dt, reserve_map);
758         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
759         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
760         dt->header.totalsize = sizeof(struct iseries_flat_dt);
761         dt->header.dt_strings_size = sizeof(struct blob);
762
763         /* There is no notion of hardware cpu id on iSeries */
764         dt->header.boot_cpuid_phys = smp_processor_id();
765
766         dt->dt.next = (unsigned long)&dt->dt.data;
767         dt->strings.next = (unsigned long)&dt->strings.data;
768
769         dt->header.magic = OF_DT_HEADER;
770         dt->header.version = 0x10;
771         dt->header.last_comp_version = 0x10;
772
773         dt->reserve_map[0] = 0;
774         dt->reserve_map[1] = 0;
775 }
776
777 void dt_check_blob(struct blob *b)
778 {
779         if (b->next >= (unsigned long)&b->next) {
780                 DBG("Ran out of space in flat device tree blob!\n");
781                 BUG();
782         }
783 }
784
785 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
786 {
787         *((u32*)dt->dt.next) = value;
788         dt->dt.next += sizeof(u32);
789
790         dt_check_blob(&dt->dt);
791 }
792
793 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
794 {
795         *((u64*)dt->dt.next) = value;
796         dt->dt.next += sizeof(u64);
797
798         dt_check_blob(&dt->dt);
799 }
800
801 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
802 {
803         unsigned long start = blob->next - (unsigned long)blob->data;
804
805         memcpy((char *)blob->next, data, len);
806         blob->next = _ALIGN(blob->next + len, 4);
807
808         dt_check_blob(blob);
809
810         return start;
811 }
812
813 void dt_start_node(struct iseries_flat_dt *dt, char *name)
814 {
815         dt_push_u32(dt, OF_DT_BEGIN_NODE);
816         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
817 }
818
819 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
820
821 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
822 {
823         unsigned long offset;
824
825         dt_push_u32(dt, OF_DT_PROP);
826
827         /* Length of the data */
828         dt_push_u32(dt, len);
829
830         /* Put the property name in the string blob. */
831         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
832
833         /* The offset of the properties name in the string blob. */
834         dt_push_u32(dt, (u32)offset);
835
836         /* The actual data. */
837         dt_push_bytes(&dt->dt, data, len);
838 }
839
840 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
841 {
842         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
843 }
844
845 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
846 {
847         dt_prop(dt, name, (char *)&data, sizeof(u32));
848 }
849
850 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
851 {
852         dt_prop(dt, name, (char *)&data, sizeof(u64));
853 }
854
855 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
856 {
857         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
858 }
859
860 void dt_prop_u32_list(struct iseries_flat_dt *dt, char *name, u32 *data, int n)
861 {
862         dt_prop(dt, name, (char *)data, sizeof(u32) * n);
863 }
864
865 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
866 {
867         dt_prop(dt, name, NULL, 0);
868 }
869
870 void dt_cpus(struct iseries_flat_dt *dt)
871 {
872         unsigned char buf[32];
873         unsigned char *p;
874         unsigned int i, index;
875         struct IoHriProcessorVpd *d;
876         u32 pft_size[2];
877
878         /* yuck */
879         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
880         p = strchr(buf, ' ');
881         if (!p) p = buf + strlen(buf);
882
883         dt_start_node(dt, "cpus");
884         dt_prop_u32(dt, "#address-cells", 1);
885         dt_prop_u32(dt, "#size-cells", 0);
886
887         pft_size[0] = 0; /* NUMA CEC cookie, 0 for non NUMA  */
888         pft_size[1] = __ilog2(HvCallHpt_getHptPages() * HW_PAGE_SIZE);
889
890         for (i = 0; i < NR_CPUS; i++) {
891                 if (lppaca[i].dyn_proc_status >= 2)
892                         continue;
893
894                 snprintf(p, 32 - (p - buf), "@%d", i);
895                 dt_start_node(dt, buf);
896
897                 dt_prop_str(dt, "device_type", "cpu");
898
899                 index = lppaca[i].dyn_hv_phys_proc_index;
900                 d = &xIoHriProcessorVpd[index];
901
902                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
903                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
904
905                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
906                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
907
908                 /* magic conversions to Hz copied from old code */
909                 dt_prop_u32(dt, "clock-frequency",
910                         ((1UL << 34) * 1000000) / d->xProcFreq);
911                 dt_prop_u32(dt, "timebase-frequency",
912                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
913
914                 dt_prop_u32(dt, "reg", i);
915
916                 dt_prop_u32_list(dt, "ibm,pft-size", pft_size, 2);
917
918                 dt_end_node(dt);
919         }
920
921         dt_end_node(dt);
922 }
923
924 void build_flat_dt(struct iseries_flat_dt *dt, unsigned long phys_mem_size)
925 {
926         u64 tmp[2];
927
928         dt_init(dt);
929
930         dt_start_node(dt, "");
931
932         dt_prop_u32(dt, "#address-cells", 2);
933         dt_prop_u32(dt, "#size-cells", 2);
934
935         /* /memory */
936         dt_start_node(dt, "memory@0");
937         dt_prop_str(dt, "name", "memory");
938         dt_prop_str(dt, "device_type", "memory");
939         tmp[0] = 0;
940         tmp[1] = phys_mem_size;
941         dt_prop_u64_list(dt, "reg", tmp, 2);
942         dt_end_node(dt);
943
944         /* /chosen */
945         dt_start_node(dt, "chosen");
946         dt_prop_u32(dt, "linux,platform", PLATFORM_ISERIES_LPAR);
947         if (cmd_mem_limit)
948                 dt_prop_u64(dt, "linux,memory-limit", cmd_mem_limit);
949         dt_end_node(dt);
950
951         dt_cpus(dt);
952
953         dt_end_node(dt);
954
955         dt_push_u32(dt, OF_DT_END);
956 }
957
958 void * __init iSeries_early_setup(void)
959 {
960         unsigned long phys_mem_size;
961
962         iSeries_fixup_klimit();
963
964         /*
965          * Initialize the table which translate Linux physical addresses to
966          * AS/400 absolute addresses
967          */
968         phys_mem_size = build_iSeries_Memory_Map();
969
970         iSeries_get_cmdline();
971
972         /* Save unparsed command line copy for /proc/cmdline */
973         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
974
975         /* Parse early parameters, in particular mem=x */
976         parse_early_param();
977
978         build_flat_dt(&iseries_dt, phys_mem_size);
979
980         return (void *) __pa(&iseries_dt);
981 }
982
983 /*
984  * On iSeries we just parse the mem=X option from the command line.
985  * On pSeries it's a bit more complicated, see prom_init_mem()
986  */
987 static int __init early_parsemem(char *p)
988 {
989         if (p)
990                 cmd_mem_limit = ALIGN(memparse(p, &p), PAGE_SIZE);
991         return 0;
992 }
993 early_param("mem", early_parsemem);
994
995 static void hvputc(char c)
996 {
997         if (c == '\n')
998                 hvputc('\r');
999
1000         HvCall_writeLogBuffer(&c, 1);
1001 }
1002
1003 void __init udbg_init_iseries(void)
1004 {
1005         udbg_putc = hvputc;
1006 }