Merge branch 'master'
[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/systemcfg.h>
43 #include <asm/system.h>
44 #include <asm/time.h>
45 #include <asm/paca.h>
46 #include <asm/cache.h>
47 #include <asm/sections.h>
48 #include <asm/abs_addr.h>
49 #include <asm/iseries/hv_lp_config.h>
50 #include <asm/iseries/hv_call_event.h>
51 #include <asm/iseries/hv_call_xm.h>
52 #include <asm/iseries/it_lp_queue.h>
53 #include <asm/iseries/mf.h>
54 #include <asm/iseries/hv_lp_event.h>
55 #include <asm/iseries/lpar_map.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 extern void hvlog(char *fmt, ...);
67
68 #ifdef DEBUG
69 #define DBG(fmt...) hvlog(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 int mf_initialized;
95
96 static unsigned long cmd_mem_limit;
97
98 struct MemoryBlock {
99         unsigned long absStart;
100         unsigned long absEnd;
101         unsigned long logicalStart;
102         unsigned long logicalEnd;
103 };
104
105 /*
106  * Process the main store vpd to determine where the holes in memory are
107  * and return the number of physical blocks and fill in the array of
108  * block data.
109  */
110 static unsigned long iSeries_process_Condor_mainstore_vpd(
111                 struct MemoryBlock *mb_array, unsigned long max_entries)
112 {
113         unsigned long holeFirstChunk, holeSizeChunks;
114         unsigned long numMemoryBlocks = 1;
115         struct IoHriMainStoreSegment4 *msVpd =
116                 (struct IoHriMainStoreSegment4 *)xMsVpd;
117         unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
118         unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
119         unsigned long holeSize = holeEnd - holeStart;
120
121         printk("Mainstore_VPD: Condor\n");
122         /*
123          * Determine if absolute memory has any
124          * holes so that we can interpret the
125          * access map we get back from the hypervisor
126          * correctly.
127          */
128         mb_array[0].logicalStart = 0;
129         mb_array[0].logicalEnd = 0x100000000;
130         mb_array[0].absStart = 0;
131         mb_array[0].absEnd = 0x100000000;
132
133         if (holeSize) {
134                 numMemoryBlocks = 2;
135                 holeStart = holeStart & 0x000fffffffffffff;
136                 holeStart = addr_to_chunk(holeStart);
137                 holeFirstChunk = holeStart;
138                 holeSize = addr_to_chunk(holeSize);
139                 holeSizeChunks = holeSize;
140                 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
141                                 holeFirstChunk, holeSizeChunks );
142                 mb_array[0].logicalEnd = holeFirstChunk;
143                 mb_array[0].absEnd = holeFirstChunk;
144                 mb_array[1].logicalStart = holeFirstChunk;
145                 mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks;
146                 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
147                 mb_array[1].absEnd = 0x100000000;
148         }
149         return numMemoryBlocks;
150 }
151
152 #define MaxSegmentAreas                 32
153 #define MaxSegmentAdrRangeBlocks        128
154 #define MaxAreaRangeBlocks              4
155
156 static unsigned long iSeries_process_Regatta_mainstore_vpd(
157                 struct MemoryBlock *mb_array, unsigned long max_entries)
158 {
159         struct IoHriMainStoreSegment5 *msVpdP =
160                 (struct IoHriMainStoreSegment5 *)xMsVpd;
161         unsigned long numSegmentBlocks = 0;
162         u32 existsBits = msVpdP->msAreaExists;
163         unsigned long area_num;
164
165         printk("Mainstore_VPD: Regatta\n");
166
167         for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
168                 unsigned long numAreaBlocks;
169                 struct IoHriMainStoreArea4 *currentArea;
170
171                 if (existsBits & 0x80000000) {
172                         unsigned long block_num;
173
174                         currentArea = &msVpdP->msAreaArray[area_num];
175                         numAreaBlocks = currentArea->numAdrRangeBlocks;
176                         printk("ms_vpd: processing area %2ld  blocks=%ld",
177                                         area_num, numAreaBlocks);
178                         for (block_num = 0; block_num < numAreaBlocks;
179                                         ++block_num ) {
180                                 /* Process an address range block */
181                                 struct MemoryBlock tempBlock;
182                                 unsigned long i;
183
184                                 tempBlock.absStart =
185                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
186                                 tempBlock.absEnd =
187                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
188                                 tempBlock.logicalStart = 0;
189                                 tempBlock.logicalEnd   = 0;
190                                 printk("\n          block %ld absStart=%016lx absEnd=%016lx",
191                                                 block_num, tempBlock.absStart,
192                                                 tempBlock.absEnd);
193
194                                 for (i = 0; i < numSegmentBlocks; ++i) {
195                                         if (mb_array[i].absStart ==
196                                                         tempBlock.absStart)
197                                                 break;
198                                 }
199                                 if (i == numSegmentBlocks) {
200                                         if (numSegmentBlocks == max_entries)
201                                                 panic("iSeries_process_mainstore_vpd: too many memory blocks");
202                                         mb_array[numSegmentBlocks] = tempBlock;
203                                         ++numSegmentBlocks;
204                                 } else
205                                         printk(" (duplicate)");
206                         }
207                         printk("\n");
208                 }
209                 existsBits <<= 1;
210         }
211         /* Now sort the blocks found into ascending sequence */
212         if (numSegmentBlocks > 1) {
213                 unsigned long m, n;
214
215                 for (m = 0; m < numSegmentBlocks - 1; ++m) {
216                         for (n = numSegmentBlocks - 1; m < n; --n) {
217                                 if (mb_array[n].absStart <
218                                                 mb_array[n-1].absStart) {
219                                         struct MemoryBlock tempBlock;
220
221                                         tempBlock = mb_array[n];
222                                         mb_array[n] = mb_array[n-1];
223                                         mb_array[n-1] = tempBlock;
224                                 }
225                         }
226                 }
227         }
228         /*
229          * Assign "logical" addresses to each block.  These
230          * addresses correspond to the hypervisor "bitmap" space.
231          * Convert all addresses into units of 256K chunks.
232          */
233         {
234         unsigned long i, nextBitmapAddress;
235
236         printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
237         nextBitmapAddress = 0;
238         for (i = 0; i < numSegmentBlocks; ++i) {
239                 unsigned long length = mb_array[i].absEnd -
240                         mb_array[i].absStart;
241
242                 mb_array[i].logicalStart = nextBitmapAddress;
243                 mb_array[i].logicalEnd = nextBitmapAddress + length;
244                 nextBitmapAddress += length;
245                 printk("          Bitmap range: %016lx - %016lx\n"
246                                 "        Absolute range: %016lx - %016lx\n",
247                                 mb_array[i].logicalStart,
248                                 mb_array[i].logicalEnd,
249                                 mb_array[i].absStart, mb_array[i].absEnd);
250                 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
251                                 0x000fffffffffffff);
252                 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
253                                 0x000fffffffffffff);
254                 mb_array[i].logicalStart =
255                         addr_to_chunk(mb_array[i].logicalStart);
256                 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
257         }
258         }
259
260         return numSegmentBlocks;
261 }
262
263 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
264                 unsigned long max_entries)
265 {
266         unsigned long i;
267         unsigned long mem_blocks = 0;
268
269         if (cpu_has_feature(CPU_FTR_SLB))
270                 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
271                                 max_entries);
272         else
273                 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
274                                 max_entries);
275
276         printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
277         for (i = 0; i < mem_blocks; ++i) {
278                 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
279                        "                             abs chunks %016lx - %016lx\n",
280                         i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
281                         mb_array[i].absStart, mb_array[i].absEnd);
282         }
283         return mem_blocks;
284 }
285
286 static void __init iSeries_get_cmdline(void)
287 {
288         char *p, *q;
289
290         /* copy the command line parameter from the primary VSP  */
291         HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
292                         HvLpDma_Direction_RemoteToLocal);
293
294         p = cmd_line;
295         q = cmd_line + 255;
296         while(p < q) {
297                 if (!*p || *p == '\n')
298                         break;
299                 ++p;
300         }
301         *p = 0;
302 }
303
304 static void __init iSeries_init_early(void)
305 {
306         DBG(" -> iSeries_init_early()\n");
307
308         ppc64_firmware_features = FW_FEATURE_ISERIES;
309
310         ppc64_interrupt_controller = IC_ISERIES;
311
312 #if defined(CONFIG_BLK_DEV_INITRD)
313         /*
314          * If the init RAM disk has been configured and there is
315          * a non-zero starting address for it, set it up
316          */
317         if (naca.xRamDisk) {
318                 initrd_start = (unsigned long)__va(naca.xRamDisk);
319                 initrd_end = initrd_start + naca.xRamDiskSize * HW_PAGE_SIZE;
320                 initrd_below_start_ok = 1;      // ramdisk in kernel space
321                 ROOT_DEV = Root_RAM0;
322                 if (((rd_size * 1024) / HW_PAGE_SIZE) < naca.xRamDiskSize)
323                         rd_size = (naca.xRamDiskSize * HW_PAGE_SIZE) / 1024;
324         } else
325 #endif /* CONFIG_BLK_DEV_INITRD */
326         {
327             /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
328         }
329
330         iSeries_recal_tb = get_tb();
331         iSeries_recal_titan = HvCallXm_loadTod();
332
333         /*
334          * Initialize the hash table management pointers
335          */
336         hpte_init_iSeries();
337
338         /*
339          * Initialize the DMA/TCE management
340          */
341         iommu_init_early_iSeries();
342
343         /* Initialize machine-dependency vectors */
344 #ifdef CONFIG_SMP
345         smp_init_iSeries();
346 #endif
347         if (itLpNaca.xPirEnvironMode == 0)
348                 piranha_simulator = 1;
349
350         /* Associate Lp Event Queue 0 with processor 0 */
351         HvCallEvent_setLpEventQueueInterruptProc(0, 0);
352
353         mf_init();
354         mf_initialized = 1;
355         mb();
356
357         /* If we were passed an initrd, set the ROOT_DEV properly if the values
358          * look sensible. If not, clear initrd reference.
359          */
360 #ifdef CONFIG_BLK_DEV_INITRD
361         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
362             initrd_end > initrd_start)
363                 ROOT_DEV = Root_RAM0;
364         else
365                 initrd_start = initrd_end = 0;
366 #endif /* CONFIG_BLK_DEV_INITRD */
367
368         DBG(" <- iSeries_init_early()\n");
369 }
370
371 struct mschunks_map mschunks_map = {
372         /* XXX We don't use these, but Piranha might need them. */
373         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
374         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
375         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
376 };
377 EXPORT_SYMBOL(mschunks_map);
378
379 void mschunks_alloc(unsigned long num_chunks)
380 {
381         klimit = _ALIGN(klimit, sizeof(u32));
382         mschunks_map.mapping = (u32 *)klimit;
383         klimit += num_chunks * sizeof(u32);
384         mschunks_map.num_chunks = num_chunks;
385 }
386
387 /*
388  * The iSeries may have very large memories ( > 128 GB ) and a partition
389  * may get memory in "chunks" that may be anywhere in the 2**52 real
390  * address space.  The chunks are 256K in size.  To map this to the
391  * memory model Linux expects, the AS/400 specific code builds a
392  * translation table to translate what Linux thinks are "physical"
393  * addresses to the actual real addresses.  This allows us to make
394  * it appear to Linux that we have contiguous memory starting at
395  * physical address zero while in fact this could be far from the truth.
396  * To avoid confusion, I'll let the words physical and/or real address
397  * apply to the Linux addresses while I'll use "absolute address" to
398  * refer to the actual hardware real address.
399  *
400  * build_iSeries_Memory_Map gets information from the Hypervisor and
401  * looks at the Main Store VPD to determine the absolute addresses
402  * of the memory that has been assigned to our partition and builds
403  * a table used to translate Linux's physical addresses to these
404  * absolute addresses.  Absolute addresses are needed when
405  * communicating with the hypervisor (e.g. to build HPT entries)
406  *
407  * Returns the physical memory size
408  */
409
410 static unsigned long __init build_iSeries_Memory_Map(void)
411 {
412         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
413         u32 nextPhysChunk;
414         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
415         u32 totalChunks,moreChunks;
416         u32 currChunk, thisChunk, absChunk;
417         u32 currDword;
418         u32 chunkBit;
419         u64 map;
420         struct MemoryBlock mb[32];
421         unsigned long numMemoryBlocks, curBlock;
422
423         /* Chunk size on iSeries is 256K bytes */
424         totalChunks = (u32)HvLpConfig_getMsChunks();
425         mschunks_alloc(totalChunks);
426
427         /*
428          * Get absolute address of our load area
429          * and map it to physical address 0
430          * This guarantees that the loadarea ends up at physical 0
431          * otherwise, it might not be returned by PLIC as the first
432          * chunks
433          */
434
435         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
436         loadAreaSize =  itLpNaca.xLoadAreaChunks;
437
438         /*
439          * Only add the pages already mapped here.
440          * Otherwise we might add the hpt pages
441          * The rest of the pages of the load area
442          * aren't in the HPT yet and can still
443          * be assigned an arbitrary physical address
444          */
445         if ((loadAreaSize * 64) > HvPagesToMap)
446                 loadAreaSize = HvPagesToMap / 64;
447
448         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
449
450         /*
451          * TODO Do we need to do something if the HPT is in the 64MB load area?
452          * This would be required if the itLpNaca.xLoadAreaChunks includes
453          * the HPT size
454          */
455
456         printk("Mapping load area - physical addr = 0000000000000000\n"
457                 "                    absolute addr = %016lx\n",
458                 chunk_to_addr(loadAreaFirstChunk));
459         printk("Load area size %dK\n", loadAreaSize * 256);
460
461         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
462                 mschunks_map.mapping[nextPhysChunk] =
463                         loadAreaFirstChunk + nextPhysChunk;
464
465         /*
466          * Get absolute address of our HPT and remember it so
467          * we won't map it to any physical address
468          */
469         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
470         hptSizePages = (u32)HvCallHpt_getHptPages();
471         hptSizeChunks = hptSizePages >>
472                 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
473         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
474
475         printk("HPT absolute addr = %016lx, size = %dK\n",
476                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
477
478         ppc64_pft_size = __ilog2(hptSizePages * HW_PAGE_SIZE);
479
480         /*
481          * The actual hashed page table is in the hypervisor,
482          * we have no direct access
483          */
484         htab_address = NULL;
485
486         /*
487          * Determine if absolute memory has any
488          * holes so that we can interpret the
489          * access map we get back from the hypervisor
490          * correctly.
491          */
492         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
493
494         /*
495          * Process the main store access map from the hypervisor
496          * to build up our physical -> absolute translation table
497          */
498         curBlock = 0;
499         currChunk = 0;
500         currDword = 0;
501         moreChunks = totalChunks;
502
503         while (moreChunks) {
504                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
505                                 currDword);
506                 thisChunk = currChunk;
507                 while (map) {
508                         chunkBit = map >> 63;
509                         map <<= 1;
510                         if (chunkBit) {
511                                 --moreChunks;
512                                 while (thisChunk >= mb[curBlock].logicalEnd) {
513                                         ++curBlock;
514                                         if (curBlock >= numMemoryBlocks)
515                                                 panic("out of memory blocks");
516                                 }
517                                 if (thisChunk < mb[curBlock].logicalStart)
518                                         panic("memory block error");
519
520                                 absChunk = mb[curBlock].absStart +
521                                         (thisChunk - mb[curBlock].logicalStart);
522                                 if (((absChunk < hptFirstChunk) ||
523                                      (absChunk > hptLastChunk)) &&
524                                     ((absChunk < loadAreaFirstChunk) ||
525                                      (absChunk > loadAreaLastChunk))) {
526                                         mschunks_map.mapping[nextPhysChunk] =
527                                                 absChunk;
528                                         ++nextPhysChunk;
529                                 }
530                         }
531                         ++thisChunk;
532                 }
533                 ++currDword;
534                 currChunk += 64;
535         }
536
537         /*
538          * main store size (in chunks) is
539          *   totalChunks - hptSizeChunks
540          * which should be equal to
541          *   nextPhysChunk
542          */
543         return chunk_to_addr(nextPhysChunk);
544 }
545
546 /*
547  * Document me.
548  */
549 static void __init iSeries_setup_arch(void)
550 {
551         unsigned procIx = get_paca()->lppaca.dyn_hv_phys_proc_index;
552
553         if (get_paca()->lppaca.shared_proc) {
554                 ppc_md.idle_loop = iseries_shared_idle;
555                 printk(KERN_INFO "Using shared processor idle loop\n");
556         } else {
557                 ppc_md.idle_loop = iseries_dedicated_idle;
558                 printk(KERN_INFO "Using dedicated idle loop\n");
559         }
560
561         /* Setup the Lp Event Queue */
562         setup_hvlpevent_queue();
563
564         printk("Max  logical processors = %d\n",
565                         itVpdAreas.xSlicMaxLogicalProcs);
566         printk("Max physical processors = %d\n",
567                         itVpdAreas.xSlicMaxPhysicalProcs);
568
569         _systemcfg->processor = xIoHriProcessorVpd[procIx].xPVR;
570         printk("Processor version = %x\n", _systemcfg->processor);
571 }
572
573 static void iSeries_show_cpuinfo(struct seq_file *m)
574 {
575         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
576 }
577
578 /*
579  * Document me.
580  * and Implement me.
581  */
582 static int iSeries_get_irq(struct pt_regs *regs)
583 {
584         /* -2 means ignore this interrupt */
585         return -2;
586 }
587
588 /*
589  * Document me.
590  */
591 static void iSeries_restart(char *cmd)
592 {
593         mf_reboot();
594 }
595
596 /*
597  * Document me.
598  */
599 static void iSeries_power_off(void)
600 {
601         mf_power_off();
602 }
603
604 /*
605  * Document me.
606  */
607 static void iSeries_halt(void)
608 {
609         mf_power_off();
610 }
611
612 static void __init iSeries_progress(char * st, unsigned short code)
613 {
614         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
615         if (!piranha_simulator && mf_initialized) {
616                 if (code != 0xffff)
617                         mf_display_progress(code);
618                 else
619                         mf_clear_src();
620         }
621 }
622
623 static void __init iSeries_fixup_klimit(void)
624 {
625         /*
626          * Change klimit to take into account any ram disk
627          * that may be included
628          */
629         if (naca.xRamDisk)
630                 klimit = KERNELBASE + (u64)naca.xRamDisk +
631                         (naca.xRamDiskSize * HW_PAGE_SIZE);
632         else {
633                 /*
634                  * No ram disk was included - check and see if there
635                  * was an embedded system map.  Change klimit to take
636                  * into account any embedded system map
637                  */
638                 if (embedded_sysmap_end)
639                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
640                                         0xfffffffffffff000);
641         }
642 }
643
644 static int __init iSeries_src_init(void)
645 {
646         /* clear the progress line */
647         ppc_md.progress(" ", 0xffff);
648         return 0;
649 }
650
651 late_initcall(iSeries_src_init);
652
653 static inline void process_iSeries_events(void)
654 {
655         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
656 }
657
658 static void yield_shared_processor(void)
659 {
660         unsigned long tb;
661
662         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
663                                     HvCall_MaskLpEvent |
664                                     HvCall_MaskLpProd |
665                                     HvCall_MaskTimeout);
666
667         tb = get_tb();
668         /* Compute future tb value when yield should expire */
669         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
670
671         /*
672          * The decrementer stops during the yield.  Force a fake decrementer
673          * here and let the timer_interrupt code sort out the actual time.
674          */
675         get_paca()->lppaca.int_dword.fields.decr_int = 1;
676         process_iSeries_events();
677 }
678
679 static void iseries_shared_idle(void)
680 {
681         while (1) {
682                 while (!need_resched() && !hvlpevent_is_pending()) {
683                         local_irq_disable();
684                         ppc64_runlatch_off();
685
686                         /* Recheck with irqs off */
687                         if (!need_resched() && !hvlpevent_is_pending())
688                                 yield_shared_processor();
689
690                         HMT_medium();
691                         local_irq_enable();
692                 }
693
694                 ppc64_runlatch_on();
695
696                 if (hvlpevent_is_pending())
697                         process_iSeries_events();
698
699                 preempt_enable_no_resched();
700                 schedule();
701                 preempt_disable();
702         }
703 }
704
705 static void iseries_dedicated_idle(void)
706 {
707         set_thread_flag(TIF_POLLING_NRFLAG);
708
709         while (1) {
710                 if (!need_resched()) {
711                         while (!need_resched()) {
712                                 ppc64_runlatch_off();
713                                 HMT_low();
714
715                                 if (hvlpevent_is_pending()) {
716                                         HMT_medium();
717                                         ppc64_runlatch_on();
718                                         process_iSeries_events();
719                                 }
720                         }
721
722                         HMT_medium();
723                 }
724
725                 ppc64_runlatch_on();
726                 preempt_enable_no_resched();
727                 schedule();
728                 preempt_disable();
729         }
730 }
731
732 #ifndef CONFIG_PCI
733 void __init iSeries_init_IRQ(void) { }
734 #endif
735
736 static int __init iseries_probe(int platform)
737 {
738         return PLATFORM_ISERIES_LPAR == platform;
739 }
740
741 struct machdep_calls __initdata iseries_md = {
742         .setup_arch     = iSeries_setup_arch,
743         .show_cpuinfo   = iSeries_show_cpuinfo,
744         .init_IRQ       = iSeries_init_IRQ,
745         .get_irq        = iSeries_get_irq,
746         .init_early     = iSeries_init_early,
747         .pcibios_fixup  = iSeries_pci_final_fixup,
748         .restart        = iSeries_restart,
749         .power_off      = iSeries_power_off,
750         .halt           = iSeries_halt,
751         .get_boot_time  = iSeries_get_boot_time,
752         .set_rtc_time   = iSeries_set_rtc_time,
753         .get_rtc_time   = iSeries_get_rtc_time,
754         .calibrate_decr = generic_calibrate_decr,
755         .progress       = iSeries_progress,
756         .probe          = iseries_probe,
757         /* XXX Implement enable_pmcs for iSeries */
758 };
759
760 struct blob {
761         unsigned char data[PAGE_SIZE];
762         unsigned long next;
763 };
764
765 struct iseries_flat_dt {
766         struct boot_param_header header;
767         u64 reserve_map[2];
768         struct blob dt;
769         struct blob strings;
770 };
771
772 struct iseries_flat_dt iseries_dt;
773
774 void dt_init(struct iseries_flat_dt *dt)
775 {
776         dt->header.off_mem_rsvmap =
777                 offsetof(struct iseries_flat_dt, reserve_map);
778         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
779         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
780         dt->header.totalsize = sizeof(struct iseries_flat_dt);
781         dt->header.dt_strings_size = sizeof(struct blob);
782
783         /* There is no notion of hardware cpu id on iSeries */
784         dt->header.boot_cpuid_phys = smp_processor_id();
785
786         dt->dt.next = (unsigned long)&dt->dt.data;
787         dt->strings.next = (unsigned long)&dt->strings.data;
788
789         dt->header.magic = OF_DT_HEADER;
790         dt->header.version = 0x10;
791         dt->header.last_comp_version = 0x10;
792
793         dt->reserve_map[0] = 0;
794         dt->reserve_map[1] = 0;
795 }
796
797 void dt_check_blob(struct blob *b)
798 {
799         if (b->next >= (unsigned long)&b->next) {
800                 DBG("Ran out of space in flat device tree blob!\n");
801                 BUG();
802         }
803 }
804
805 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
806 {
807         *((u32*)dt->dt.next) = value;
808         dt->dt.next += sizeof(u32);
809
810         dt_check_blob(&dt->dt);
811 }
812
813 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
814 {
815         *((u64*)dt->dt.next) = value;
816         dt->dt.next += sizeof(u64);
817
818         dt_check_blob(&dt->dt);
819 }
820
821 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
822 {
823         unsigned long start = blob->next - (unsigned long)blob->data;
824
825         memcpy((char *)blob->next, data, len);
826         blob->next = _ALIGN(blob->next + len, 4);
827
828         dt_check_blob(blob);
829
830         return start;
831 }
832
833 void dt_start_node(struct iseries_flat_dt *dt, char *name)
834 {
835         dt_push_u32(dt, OF_DT_BEGIN_NODE);
836         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
837 }
838
839 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
840
841 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
842 {
843         unsigned long offset;
844
845         dt_push_u32(dt, OF_DT_PROP);
846
847         /* Length of the data */
848         dt_push_u32(dt, len);
849
850         /* Put the property name in the string blob. */
851         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
852
853         /* The offset of the properties name in the string blob. */
854         dt_push_u32(dt, (u32)offset);
855
856         /* The actual data. */
857         dt_push_bytes(&dt->dt, data, len);
858 }
859
860 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
861 {
862         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
863 }
864
865 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
866 {
867         dt_prop(dt, name, (char *)&data, sizeof(u32));
868 }
869
870 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
871 {
872         dt_prop(dt, name, (char *)&data, sizeof(u64));
873 }
874
875 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
876 {
877         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
878 }
879
880 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
881 {
882         dt_prop(dt, name, NULL, 0);
883 }
884
885 void dt_cpus(struct iseries_flat_dt *dt)
886 {
887         unsigned char buf[32];
888         unsigned char *p;
889         unsigned int i, index;
890         struct IoHriProcessorVpd *d;
891
892         /* yuck */
893         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
894         p = strchr(buf, ' ');
895         if (!p) p = buf + strlen(buf);
896
897         dt_start_node(dt, "cpus");
898         dt_prop_u32(dt, "#address-cells", 1);
899         dt_prop_u32(dt, "#size-cells", 0);
900
901         for (i = 0; i < NR_CPUS; i++) {
902                 if (paca[i].lppaca.dyn_proc_status >= 2)
903                         continue;
904
905                 snprintf(p, 32 - (p - buf), "@%d", i);
906                 dt_start_node(dt, buf);
907
908                 dt_prop_str(dt, "device_type", "cpu");
909
910                 index = paca[i].lppaca.dyn_hv_phys_proc_index;
911                 d = &xIoHriProcessorVpd[index];
912
913                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
914                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
915
916                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
917                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
918
919                 /* magic conversions to Hz copied from old code */
920                 dt_prop_u32(dt, "clock-frequency",
921                         ((1UL << 34) * 1000000) / d->xProcFreq);
922                 dt_prop_u32(dt, "timebase-frequency",
923                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
924
925                 dt_prop_u32(dt, "reg", i);
926
927                 dt_end_node(dt);
928         }
929
930         dt_end_node(dt);
931 }
932
933 void build_flat_dt(struct iseries_flat_dt *dt, unsigned long phys_mem_size)
934 {
935         u64 tmp[2];
936
937         dt_init(dt);
938
939         dt_start_node(dt, "");
940
941         dt_prop_u32(dt, "#address-cells", 2);
942         dt_prop_u32(dt, "#size-cells", 2);
943
944         /* /memory */
945         dt_start_node(dt, "memory@0");
946         dt_prop_str(dt, "name", "memory");
947         dt_prop_str(dt, "device_type", "memory");
948         tmp[0] = 0;
949         tmp[1] = phys_mem_size;
950         dt_prop_u64_list(dt, "reg", tmp, 2);
951         dt_end_node(dt);
952
953         /* /chosen */
954         dt_start_node(dt, "chosen");
955         dt_prop_u32(dt, "linux,platform", PLATFORM_ISERIES_LPAR);
956         if (cmd_mem_limit)
957                 dt_prop_u64(dt, "linux,memory-limit", cmd_mem_limit);
958         dt_end_node(dt);
959
960         dt_cpus(dt);
961
962         dt_end_node(dt);
963
964         dt_push_u32(dt, OF_DT_END);
965 }
966
967 void * __init iSeries_early_setup(void)
968 {
969         unsigned long phys_mem_size;
970
971         iSeries_fixup_klimit();
972
973         /*
974          * Initialize the table which translate Linux physical addresses to
975          * AS/400 absolute addresses
976          */
977         phys_mem_size = build_iSeries_Memory_Map();
978
979         iSeries_get_cmdline();
980
981         /* Save unparsed command line copy for /proc/cmdline */
982         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
983
984         /* Parse early parameters, in particular mem=x */
985         parse_early_param();
986
987         build_flat_dt(&iseries_dt, phys_mem_size);
988
989         return (void *) __pa(&iseries_dt);
990 }
991
992 /*
993  * On iSeries we just parse the mem=X option from the command line.
994  * On pSeries it's a bit more complicated, see prom_init_mem()
995  */
996 static int __init early_parsemem(char *p)
997 {
998         if (p)
999                 cmd_mem_limit = ALIGN(memparse(p, &p), PAGE_SIZE);
1000         return 0;
1001 }
1002 early_param("mem", early_parsemem);