Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / 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/it_exp_vpd_panel.h>
54 #include <asm/iseries/hv_lp_event.h>
55 #include <asm/iseries/lpar_map.h>
56 #include <asm/udbg.h>
57 #include <asm/irq.h>
58
59 #include "naca.h"
60 #include "setup.h"
61 #include "irq.h"
62 #include "vpd_areas.h"
63 #include "processor_vpd.h"
64 #include "main_store.h"
65 #include "call_sm.h"
66 #include "call_hpt.h"
67
68 #ifdef DEBUG
69 #define DBG(fmt...) udbg_printf(fmt)
70 #else
71 #define DBG(fmt...)
72 #endif
73
74 /* Function Prototypes */
75 static unsigned long build_iSeries_Memory_Map(void);
76 static void iseries_shared_idle(void);
77 static void 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 embedded_sysmap_start;
89 extern unsigned long embedded_sysmap_end;
90
91 extern unsigned long iSeries_recal_tb;
92 extern unsigned long iSeries_recal_titan;
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
351         /* If we were passed an initrd, set the ROOT_DEV properly if the values
352          * look sensible. If not, clear initrd reference.
353          */
354 #ifdef CONFIG_BLK_DEV_INITRD
355         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
356             initrd_end > initrd_start)
357                 ROOT_DEV = Root_RAM0;
358         else
359                 initrd_start = initrd_end = 0;
360 #endif /* CONFIG_BLK_DEV_INITRD */
361
362         DBG(" <- iSeries_init_early()\n");
363 }
364
365 struct mschunks_map mschunks_map = {
366         /* XXX We don't use these, but Piranha might need them. */
367         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
368         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
369         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
370 };
371 EXPORT_SYMBOL(mschunks_map);
372
373 void mschunks_alloc(unsigned long num_chunks)
374 {
375         klimit = _ALIGN(klimit, sizeof(u32));
376         mschunks_map.mapping = (u32 *)klimit;
377         klimit += num_chunks * sizeof(u32);
378         mschunks_map.num_chunks = num_chunks;
379 }
380
381 /*
382  * The iSeries may have very large memories ( > 128 GB ) and a partition
383  * may get memory in "chunks" that may be anywhere in the 2**52 real
384  * address space.  The chunks are 256K in size.  To map this to the
385  * memory model Linux expects, the AS/400 specific code builds a
386  * translation table to translate what Linux thinks are "physical"
387  * addresses to the actual real addresses.  This allows us to make
388  * it appear to Linux that we have contiguous memory starting at
389  * physical address zero while in fact this could be far from the truth.
390  * To avoid confusion, I'll let the words physical and/or real address
391  * apply to the Linux addresses while I'll use "absolute address" to
392  * refer to the actual hardware real address.
393  *
394  * build_iSeries_Memory_Map gets information from the Hypervisor and
395  * looks at the Main Store VPD to determine the absolute addresses
396  * of the memory that has been assigned to our partition and builds
397  * a table used to translate Linux's physical addresses to these
398  * absolute addresses.  Absolute addresses are needed when
399  * communicating with the hypervisor (e.g. to build HPT entries)
400  *
401  * Returns the physical memory size
402  */
403
404 static unsigned long __init build_iSeries_Memory_Map(void)
405 {
406         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
407         u32 nextPhysChunk;
408         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
409         u32 totalChunks,moreChunks;
410         u32 currChunk, thisChunk, absChunk;
411         u32 currDword;
412         u32 chunkBit;
413         u64 map;
414         struct MemoryBlock mb[32];
415         unsigned long numMemoryBlocks, curBlock;
416
417         /* Chunk size on iSeries is 256K bytes */
418         totalChunks = (u32)HvLpConfig_getMsChunks();
419         mschunks_alloc(totalChunks);
420
421         /*
422          * Get absolute address of our load area
423          * and map it to physical address 0
424          * This guarantees that the loadarea ends up at physical 0
425          * otherwise, it might not be returned by PLIC as the first
426          * chunks
427          */
428
429         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
430         loadAreaSize =  itLpNaca.xLoadAreaChunks;
431
432         /*
433          * Only add the pages already mapped here.
434          * Otherwise we might add the hpt pages
435          * The rest of the pages of the load area
436          * aren't in the HPT yet and can still
437          * be assigned an arbitrary physical address
438          */
439         if ((loadAreaSize * 64) > HvPagesToMap)
440                 loadAreaSize = HvPagesToMap / 64;
441
442         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
443
444         /*
445          * TODO Do we need to do something if the HPT is in the 64MB load area?
446          * This would be required if the itLpNaca.xLoadAreaChunks includes
447          * the HPT size
448          */
449
450         printk("Mapping load area - physical addr = 0000000000000000\n"
451                 "                    absolute addr = %016lx\n",
452                 chunk_to_addr(loadAreaFirstChunk));
453         printk("Load area size %dK\n", loadAreaSize * 256);
454
455         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
456                 mschunks_map.mapping[nextPhysChunk] =
457                         loadAreaFirstChunk + nextPhysChunk;
458
459         /*
460          * Get absolute address of our HPT and remember it so
461          * we won't map it to any physical address
462          */
463         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
464         hptSizePages = (u32)HvCallHpt_getHptPages();
465         hptSizeChunks = hptSizePages >>
466                 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
467         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
468
469         printk("HPT absolute addr = %016lx, size = %dK\n",
470                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
471
472         /*
473          * Determine if absolute memory has any
474          * holes so that we can interpret the
475          * access map we get back from the hypervisor
476          * correctly.
477          */
478         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
479
480         /*
481          * Process the main store access map from the hypervisor
482          * to build up our physical -> absolute translation table
483          */
484         curBlock = 0;
485         currChunk = 0;
486         currDword = 0;
487         moreChunks = totalChunks;
488
489         while (moreChunks) {
490                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
491                                 currDword);
492                 thisChunk = currChunk;
493                 while (map) {
494                         chunkBit = map >> 63;
495                         map <<= 1;
496                         if (chunkBit) {
497                                 --moreChunks;
498                                 while (thisChunk >= mb[curBlock].logicalEnd) {
499                                         ++curBlock;
500                                         if (curBlock >= numMemoryBlocks)
501                                                 panic("out of memory blocks");
502                                 }
503                                 if (thisChunk < mb[curBlock].logicalStart)
504                                         panic("memory block error");
505
506                                 absChunk = mb[curBlock].absStart +
507                                         (thisChunk - mb[curBlock].logicalStart);
508                                 if (((absChunk < hptFirstChunk) ||
509                                      (absChunk > hptLastChunk)) &&
510                                     ((absChunk < loadAreaFirstChunk) ||
511                                      (absChunk > loadAreaLastChunk))) {
512                                         mschunks_map.mapping[nextPhysChunk] =
513                                                 absChunk;
514                                         ++nextPhysChunk;
515                                 }
516                         }
517                         ++thisChunk;
518                 }
519                 ++currDword;
520                 currChunk += 64;
521         }
522
523         /*
524          * main store size (in chunks) is
525          *   totalChunks - hptSizeChunks
526          * which should be equal to
527          *   nextPhysChunk
528          */
529         return chunk_to_addr(nextPhysChunk);
530 }
531
532 /*
533  * Document me.
534  */
535 static void __init iSeries_setup_arch(void)
536 {
537         if (get_lppaca()->shared_proc) {
538                 ppc_md.idle_loop = iseries_shared_idle;
539                 printk(KERN_INFO "Using shared processor idle loop\n");
540         } else {
541                 ppc_md.idle_loop = iseries_dedicated_idle;
542                 printk(KERN_INFO "Using dedicated idle loop\n");
543         }
544
545         /* Setup the Lp Event Queue */
546         setup_hvlpevent_queue();
547
548         printk("Max  logical processors = %d\n",
549                         itVpdAreas.xSlicMaxLogicalProcs);
550         printk("Max physical processors = %d\n",
551                         itVpdAreas.xSlicMaxPhysicalProcs);
552 }
553
554 static void iSeries_show_cpuinfo(struct seq_file *m)
555 {
556         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
557 }
558
559 static void __init iSeries_progress(char * st, unsigned short code)
560 {
561         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
562         mf_display_progress(code);
563 }
564
565 static void __init iSeries_fixup_klimit(void)
566 {
567         /*
568          * Change klimit to take into account any ram disk
569          * that may be included
570          */
571         if (naca.xRamDisk)
572                 klimit = KERNELBASE + (u64)naca.xRamDisk +
573                         (naca.xRamDiskSize * HW_PAGE_SIZE);
574         else {
575                 /*
576                  * No ram disk was included - check and see if there
577                  * was an embedded system map.  Change klimit to take
578                  * into account any embedded system map
579                  */
580                 if (embedded_sysmap_end)
581                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
582                                         0xfffffffffffff000);
583         }
584 }
585
586 static int __init iSeries_src_init(void)
587 {
588         /* clear the progress line */
589         ppc_md.progress(" ", 0xffff);
590         return 0;
591 }
592
593 late_initcall(iSeries_src_init);
594
595 static inline void process_iSeries_events(void)
596 {
597         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
598 }
599
600 static void yield_shared_processor(void)
601 {
602         unsigned long tb;
603
604         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
605                                     HvCall_MaskLpEvent |
606                                     HvCall_MaskLpProd |
607                                     HvCall_MaskTimeout);
608
609         tb = get_tb();
610         /* Compute future tb value when yield should expire */
611         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
612
613         /*
614          * The decrementer stops during the yield.  Force a fake decrementer
615          * here and let the timer_interrupt code sort out the actual time.
616          */
617         get_lppaca()->int_dword.fields.decr_int = 1;
618         ppc64_runlatch_on();
619         process_iSeries_events();
620 }
621
622 static void iseries_shared_idle(void)
623 {
624         while (1) {
625                 while (!need_resched() && !hvlpevent_is_pending()) {
626                         local_irq_disable();
627                         ppc64_runlatch_off();
628
629                         /* Recheck with irqs off */
630                         if (!need_resched() && !hvlpevent_is_pending())
631                                 yield_shared_processor();
632
633                         HMT_medium();
634                         local_irq_enable();
635                 }
636
637                 ppc64_runlatch_on();
638
639                 if (hvlpevent_is_pending())
640                         process_iSeries_events();
641
642                 preempt_enable_no_resched();
643                 schedule();
644                 preempt_disable();
645         }
646 }
647
648 static void iseries_dedicated_idle(void)
649 {
650         set_thread_flag(TIF_POLLING_NRFLAG);
651
652         while (1) {
653                 if (!need_resched()) {
654                         while (!need_resched()) {
655                                 ppc64_runlatch_off();
656                                 HMT_low();
657
658                                 if (hvlpevent_is_pending()) {
659                                         HMT_medium();
660                                         ppc64_runlatch_on();
661                                         process_iSeries_events();
662                                 }
663                         }
664
665                         HMT_medium();
666                 }
667
668                 ppc64_runlatch_on();
669                 preempt_enable_no_resched();
670                 schedule();
671                 preempt_disable();
672         }
673 }
674
675 #ifndef CONFIG_PCI
676 void __init iSeries_init_IRQ(void) { }
677 #endif
678
679 static int __init iseries_probe(void)
680 {
681         unsigned long root = of_get_flat_dt_root();
682         if (!of_flat_dt_is_compatible(root, "IBM,iSeries"))
683                 return 0;
684
685         powerpc_firmware_features |= FW_FEATURE_ISERIES;
686         powerpc_firmware_features |= FW_FEATURE_LPAR;
687
688         /*
689          * The Hypervisor only allows us up to 256 interrupt
690          * sources (the irq number is passed in a u8).
691          */
692         virt_irq_max = 255;
693
694         return 1;
695 }
696
697 define_machine(iseries) {
698         .name           = "iSeries",
699         .setup_arch     = iSeries_setup_arch,
700         .show_cpuinfo   = iSeries_show_cpuinfo,
701         .init_IRQ       = iSeries_init_IRQ,
702         .get_irq        = iSeries_get_irq,
703         .init_early     = iSeries_init_early,
704         .pcibios_fixup  = iSeries_pci_final_fixup,
705         .restart        = mf_reboot,
706         .power_off      = mf_power_off,
707         .halt           = mf_power_off,
708         .get_boot_time  = iSeries_get_boot_time,
709         .set_rtc_time   = iSeries_set_rtc_time,
710         .get_rtc_time   = iSeries_get_rtc_time,
711         .calibrate_decr = generic_calibrate_decr,
712         .progress       = iSeries_progress,
713         .probe          = iseries_probe,
714         /* XXX Implement enable_pmcs for iSeries */
715 };
716
717 struct blob {
718         unsigned char data[PAGE_SIZE];
719         unsigned long next;
720 };
721
722 struct iseries_flat_dt {
723         struct boot_param_header header;
724         u64 reserve_map[2];
725         struct blob dt;
726         struct blob strings;
727 };
728
729 struct iseries_flat_dt iseries_dt;
730
731 void dt_init(struct iseries_flat_dt *dt)
732 {
733         dt->header.off_mem_rsvmap =
734                 offsetof(struct iseries_flat_dt, reserve_map);
735         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
736         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
737         dt->header.totalsize = sizeof(struct iseries_flat_dt);
738         dt->header.dt_strings_size = sizeof(struct blob);
739
740         /* There is no notion of hardware cpu id on iSeries */
741         dt->header.boot_cpuid_phys = smp_processor_id();
742
743         dt->dt.next = (unsigned long)&dt->dt.data;
744         dt->strings.next = (unsigned long)&dt->strings.data;
745
746         dt->header.magic = OF_DT_HEADER;
747         dt->header.version = 0x10;
748         dt->header.last_comp_version = 0x10;
749
750         dt->reserve_map[0] = 0;
751         dt->reserve_map[1] = 0;
752 }
753
754 void dt_check_blob(struct blob *b)
755 {
756         if (b->next >= (unsigned long)&b->next) {
757                 DBG("Ran out of space in flat device tree blob!\n");
758                 BUG();
759         }
760 }
761
762 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
763 {
764         *((u32*)dt->dt.next) = value;
765         dt->dt.next += sizeof(u32);
766
767         dt_check_blob(&dt->dt);
768 }
769
770 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
771 {
772         *((u64*)dt->dt.next) = value;
773         dt->dt.next += sizeof(u64);
774
775         dt_check_blob(&dt->dt);
776 }
777
778 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
779 {
780         unsigned long start = blob->next - (unsigned long)blob->data;
781
782         memcpy((char *)blob->next, data, len);
783         blob->next = _ALIGN(blob->next + len, 4);
784
785         dt_check_blob(blob);
786
787         return start;
788 }
789
790 void dt_start_node(struct iseries_flat_dt *dt, char *name)
791 {
792         dt_push_u32(dt, OF_DT_BEGIN_NODE);
793         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
794 }
795
796 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
797
798 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
799 {
800         unsigned long offset;
801
802         dt_push_u32(dt, OF_DT_PROP);
803
804         /* Length of the data */
805         dt_push_u32(dt, len);
806
807         /* Put the property name in the string blob. */
808         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
809
810         /* The offset of the properties name in the string blob. */
811         dt_push_u32(dt, (u32)offset);
812
813         /* The actual data. */
814         dt_push_bytes(&dt->dt, data, len);
815 }
816
817 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
818 {
819         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
820 }
821
822 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
823 {
824         dt_prop(dt, name, (char *)&data, sizeof(u32));
825 }
826
827 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
828 {
829         dt_prop(dt, name, (char *)&data, sizeof(u64));
830 }
831
832 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
833 {
834         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
835 }
836
837 void dt_prop_u32_list(struct iseries_flat_dt *dt, char *name, u32 *data, int n)
838 {
839         dt_prop(dt, name, (char *)data, sizeof(u32) * n);
840 }
841
842 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
843 {
844         dt_prop(dt, name, NULL, 0);
845 }
846
847 void dt_cpus(struct iseries_flat_dt *dt)
848 {
849         unsigned char buf[32];
850         unsigned char *p;
851         unsigned int i, index;
852         struct IoHriProcessorVpd *d;
853         u32 pft_size[2];
854
855         /* yuck */
856         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
857         p = strchr(buf, ' ');
858         if (!p) p = buf + strlen(buf);
859
860         dt_start_node(dt, "cpus");
861         dt_prop_u32(dt, "#address-cells", 1);
862         dt_prop_u32(dt, "#size-cells", 0);
863
864         pft_size[0] = 0; /* NUMA CEC cookie, 0 for non NUMA  */
865         pft_size[1] = __ilog2(HvCallHpt_getHptPages() * HW_PAGE_SIZE);
866
867         for (i = 0; i < NR_CPUS; i++) {
868                 if (lppaca[i].dyn_proc_status >= 2)
869                         continue;
870
871                 snprintf(p, 32 - (p - buf), "@%d", i);
872                 dt_start_node(dt, buf);
873
874                 dt_prop_str(dt, "device_type", "cpu");
875
876                 index = lppaca[i].dyn_hv_phys_proc_index;
877                 d = &xIoHriProcessorVpd[index];
878
879                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
880                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
881
882                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
883                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
884
885                 /* magic conversions to Hz copied from old code */
886                 dt_prop_u32(dt, "clock-frequency",
887                         ((1UL << 34) * 1000000) / d->xProcFreq);
888                 dt_prop_u32(dt, "timebase-frequency",
889                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
890
891                 dt_prop_u32(dt, "reg", i);
892
893                 dt_prop_u32_list(dt, "ibm,pft-size", pft_size, 2);
894
895                 dt_end_node(dt);
896         }
897
898         dt_end_node(dt);
899 }
900
901 void dt_model(struct iseries_flat_dt *dt)
902 {
903         char buf[16] = "IBM,";
904
905         /* "IBM," + mfgId[2:3] + systemSerial[1:5] */
906         strne2a(buf + 4, xItExtVpdPanel.mfgID + 2, 2);
907         strne2a(buf + 6, xItExtVpdPanel.systemSerial + 1, 5);
908         buf[11] = '\0';
909         dt_prop_str(dt, "system-id", buf);
910
911         /* "IBM," + machineType[0:4] */
912         strne2a(buf + 4, xItExtVpdPanel.machineType, 4);
913         buf[8] = '\0';
914         dt_prop_str(dt, "model", buf);
915
916         dt_prop_str(dt, "compatible", "IBM,iSeries");
917 }
918
919 void build_flat_dt(struct iseries_flat_dt *dt, unsigned long phys_mem_size)
920 {
921         u64 tmp[2];
922
923         dt_init(dt);
924
925         dt_start_node(dt, "");
926
927         dt_prop_u32(dt, "#address-cells", 2);
928         dt_prop_u32(dt, "#size-cells", 2);
929         dt_model(dt);
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_str(dt, "bootargs", cmd_line);
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 }