[PATCH] powerpc: update iseries_veth device-tree information
[pandora-kernel.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 #include <linux/if_ether.h>     /* ETH_ALEN */
32
33 #include <asm/processor.h>
34 #include <asm/machdep.h>
35 #include <asm/page.h>
36 #include <asm/mmu.h>
37 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
39 #include <asm/cputable.h>
40 #include <asm/sections.h>
41 #include <asm/iommu.h>
42 #include <asm/firmware.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_types.h>
50 #include <asm/iseries/hv_lp_config.h>
51 #include <asm/iseries/hv_call_event.h>
52 #include <asm/iseries/hv_call_xm.h>
53 #include <asm/iseries/it_lp_queue.h>
54 #include <asm/iseries/mf.h>
55 #include <asm/iseries/it_exp_vpd_panel.h>
56 #include <asm/iseries/hv_lp_event.h>
57 #include <asm/iseries/lpar_map.h>
58 #include <asm/udbg.h>
59 #include <asm/irq.h>
60
61 #include "naca.h"
62 #include "setup.h"
63 #include "irq.h"
64 #include "vpd_areas.h"
65 #include "processor_vpd.h"
66 #include "main_store.h"
67 #include "call_sm.h"
68 #include "call_hpt.h"
69
70 #ifdef DEBUG
71 #define DBG(fmt...) udbg_printf(fmt)
72 #else
73 #define DBG(fmt...)
74 #endif
75
76 /* Function Prototypes */
77 static unsigned long build_iSeries_Memory_Map(void);
78 static void iseries_shared_idle(void);
79 static void iseries_dedicated_idle(void);
80 #ifdef CONFIG_PCI
81 extern void iSeries_pci_final_fixup(void);
82 #else
83 static void iSeries_pci_final_fixup(void) { }
84 #endif
85
86 extern int rd_size;             /* Defined in drivers/block/rd.c */
87 extern unsigned long embedded_sysmap_start;
88 extern unsigned long embedded_sysmap_end;
89
90 extern unsigned long iSeries_recal_tb;
91 extern unsigned long iSeries_recal_titan;
92
93 static unsigned long cmd_mem_limit;
94
95 struct MemoryBlock {
96         unsigned long absStart;
97         unsigned long absEnd;
98         unsigned long logicalStart;
99         unsigned long logicalEnd;
100 };
101
102 /*
103  * Process the main store vpd to determine where the holes in memory are
104  * and return the number of physical blocks and fill in the array of
105  * block data.
106  */
107 static unsigned long iSeries_process_Condor_mainstore_vpd(
108                 struct MemoryBlock *mb_array, unsigned long max_entries)
109 {
110         unsigned long holeFirstChunk, holeSizeChunks;
111         unsigned long numMemoryBlocks = 1;
112         struct IoHriMainStoreSegment4 *msVpd =
113                 (struct IoHriMainStoreSegment4 *)xMsVpd;
114         unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
115         unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
116         unsigned long holeSize = holeEnd - holeStart;
117
118         printk("Mainstore_VPD: Condor\n");
119         /*
120          * Determine if absolute memory has any
121          * holes so that we can interpret the
122          * access map we get back from the hypervisor
123          * correctly.
124          */
125         mb_array[0].logicalStart = 0;
126         mb_array[0].logicalEnd = 0x100000000;
127         mb_array[0].absStart = 0;
128         mb_array[0].absEnd = 0x100000000;
129
130         if (holeSize) {
131                 numMemoryBlocks = 2;
132                 holeStart = holeStart & 0x000fffffffffffff;
133                 holeStart = addr_to_chunk(holeStart);
134                 holeFirstChunk = holeStart;
135                 holeSize = addr_to_chunk(holeSize);
136                 holeSizeChunks = holeSize;
137                 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
138                                 holeFirstChunk, holeSizeChunks );
139                 mb_array[0].logicalEnd = holeFirstChunk;
140                 mb_array[0].absEnd = holeFirstChunk;
141                 mb_array[1].logicalStart = holeFirstChunk;
142                 mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks;
143                 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
144                 mb_array[1].absEnd = 0x100000000;
145         }
146         return numMemoryBlocks;
147 }
148
149 #define MaxSegmentAreas                 32
150 #define MaxSegmentAdrRangeBlocks        128
151 #define MaxAreaRangeBlocks              4
152
153 static unsigned long iSeries_process_Regatta_mainstore_vpd(
154                 struct MemoryBlock *mb_array, unsigned long max_entries)
155 {
156         struct IoHriMainStoreSegment5 *msVpdP =
157                 (struct IoHriMainStoreSegment5 *)xMsVpd;
158         unsigned long numSegmentBlocks = 0;
159         u32 existsBits = msVpdP->msAreaExists;
160         unsigned long area_num;
161
162         printk("Mainstore_VPD: Regatta\n");
163
164         for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
165                 unsigned long numAreaBlocks;
166                 struct IoHriMainStoreArea4 *currentArea;
167
168                 if (existsBits & 0x80000000) {
169                         unsigned long block_num;
170
171                         currentArea = &msVpdP->msAreaArray[area_num];
172                         numAreaBlocks = currentArea->numAdrRangeBlocks;
173                         printk("ms_vpd: processing area %2ld  blocks=%ld",
174                                         area_num, numAreaBlocks);
175                         for (block_num = 0; block_num < numAreaBlocks;
176                                         ++block_num ) {
177                                 /* Process an address range block */
178                                 struct MemoryBlock tempBlock;
179                                 unsigned long i;
180
181                                 tempBlock.absStart =
182                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
183                                 tempBlock.absEnd =
184                                         (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
185                                 tempBlock.logicalStart = 0;
186                                 tempBlock.logicalEnd   = 0;
187                                 printk("\n          block %ld absStart=%016lx absEnd=%016lx",
188                                                 block_num, tempBlock.absStart,
189                                                 tempBlock.absEnd);
190
191                                 for (i = 0; i < numSegmentBlocks; ++i) {
192                                         if (mb_array[i].absStart ==
193                                                         tempBlock.absStart)
194                                                 break;
195                                 }
196                                 if (i == numSegmentBlocks) {
197                                         if (numSegmentBlocks == max_entries)
198                                                 panic("iSeries_process_mainstore_vpd: too many memory blocks");
199                                         mb_array[numSegmentBlocks] = tempBlock;
200                                         ++numSegmentBlocks;
201                                 } else
202                                         printk(" (duplicate)");
203                         }
204                         printk("\n");
205                 }
206                 existsBits <<= 1;
207         }
208         /* Now sort the blocks found into ascending sequence */
209         if (numSegmentBlocks > 1) {
210                 unsigned long m, n;
211
212                 for (m = 0; m < numSegmentBlocks - 1; ++m) {
213                         for (n = numSegmentBlocks - 1; m < n; --n) {
214                                 if (mb_array[n].absStart <
215                                                 mb_array[n-1].absStart) {
216                                         struct MemoryBlock tempBlock;
217
218                                         tempBlock = mb_array[n];
219                                         mb_array[n] = mb_array[n-1];
220                                         mb_array[n-1] = tempBlock;
221                                 }
222                         }
223                 }
224         }
225         /*
226          * Assign "logical" addresses to each block.  These
227          * addresses correspond to the hypervisor "bitmap" space.
228          * Convert all addresses into units of 256K chunks.
229          */
230         {
231         unsigned long i, nextBitmapAddress;
232
233         printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
234         nextBitmapAddress = 0;
235         for (i = 0; i < numSegmentBlocks; ++i) {
236                 unsigned long length = mb_array[i].absEnd -
237                         mb_array[i].absStart;
238
239                 mb_array[i].logicalStart = nextBitmapAddress;
240                 mb_array[i].logicalEnd = nextBitmapAddress + length;
241                 nextBitmapAddress += length;
242                 printk("          Bitmap range: %016lx - %016lx\n"
243                                 "        Absolute range: %016lx - %016lx\n",
244                                 mb_array[i].logicalStart,
245                                 mb_array[i].logicalEnd,
246                                 mb_array[i].absStart, mb_array[i].absEnd);
247                 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
248                                 0x000fffffffffffff);
249                 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
250                                 0x000fffffffffffff);
251                 mb_array[i].logicalStart =
252                         addr_to_chunk(mb_array[i].logicalStart);
253                 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
254         }
255         }
256
257         return numSegmentBlocks;
258 }
259
260 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
261                 unsigned long max_entries)
262 {
263         unsigned long i;
264         unsigned long mem_blocks = 0;
265
266         if (cpu_has_feature(CPU_FTR_SLB))
267                 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
268                                 max_entries);
269         else
270                 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
271                                 max_entries);
272
273         printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
274         for (i = 0; i < mem_blocks; ++i) {
275                 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
276                        "                             abs chunks %016lx - %016lx\n",
277                         i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
278                         mb_array[i].absStart, mb_array[i].absEnd);
279         }
280         return mem_blocks;
281 }
282
283 static void __init iSeries_get_cmdline(void)
284 {
285         char *p, *q;
286
287         /* copy the command line parameter from the primary VSP  */
288         HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
289                         HvLpDma_Direction_RemoteToLocal);
290
291         p = cmd_line;
292         q = cmd_line + 255;
293         while(p < q) {
294                 if (!*p || *p == '\n')
295                         break;
296                 ++p;
297         }
298         *p = 0;
299 }
300
301 static void __init iSeries_init_early(void)
302 {
303         DBG(" -> iSeries_init_early()\n");
304
305         ppc64_interrupt_controller = IC_ISERIES;
306
307 #if defined(CONFIG_BLK_DEV_INITRD)
308         /*
309          * If the init RAM disk has been configured and there is
310          * a non-zero starting address for it, set it up
311          */
312         if (naca.xRamDisk) {
313                 initrd_start = (unsigned long)__va(naca.xRamDisk);
314                 initrd_end = initrd_start + naca.xRamDiskSize * HW_PAGE_SIZE;
315                 initrd_below_start_ok = 1;      // ramdisk in kernel space
316                 ROOT_DEV = Root_RAM0;
317                 if (((rd_size * 1024) / HW_PAGE_SIZE) < naca.xRamDiskSize)
318                         rd_size = (naca.xRamDiskSize * HW_PAGE_SIZE) / 1024;
319         } else
320 #endif /* CONFIG_BLK_DEV_INITRD */
321         {
322             /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
323         }
324
325         iSeries_recal_tb = get_tb();
326         iSeries_recal_titan = HvCallXm_loadTod();
327
328         /*
329          * Initialize the hash table management pointers
330          */
331         hpte_init_iSeries();
332
333         /*
334          * Initialize the DMA/TCE management
335          */
336         iommu_init_early_iSeries();
337
338         /* Initialize machine-dependency vectors */
339 #ifdef CONFIG_SMP
340         smp_init_iSeries();
341 #endif
342
343         /* Associate Lp Event Queue 0 with processor 0 */
344         HvCallEvent_setLpEventQueueInterruptProc(0, 0);
345
346         mf_init();
347
348         /* If we were passed an initrd, set the ROOT_DEV properly if the values
349          * look sensible. If not, clear initrd reference.
350          */
351 #ifdef CONFIG_BLK_DEV_INITRD
352         if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
353             initrd_end > initrd_start)
354                 ROOT_DEV = Root_RAM0;
355         else
356                 initrd_start = initrd_end = 0;
357 #endif /* CONFIG_BLK_DEV_INITRD */
358
359         DBG(" <- iSeries_init_early()\n");
360 }
361
362 struct mschunks_map mschunks_map = {
363         /* XXX We don't use these, but Piranha might need them. */
364         .chunk_size  = MSCHUNKS_CHUNK_SIZE,
365         .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
366         .chunk_mask  = MSCHUNKS_OFFSET_MASK,
367 };
368 EXPORT_SYMBOL(mschunks_map);
369
370 void mschunks_alloc(unsigned long num_chunks)
371 {
372         klimit = _ALIGN(klimit, sizeof(u32));
373         mschunks_map.mapping = (u32 *)klimit;
374         klimit += num_chunks * sizeof(u32);
375         mschunks_map.num_chunks = num_chunks;
376 }
377
378 /*
379  * The iSeries may have very large memories ( > 128 GB ) and a partition
380  * may get memory in "chunks" that may be anywhere in the 2**52 real
381  * address space.  The chunks are 256K in size.  To map this to the
382  * memory model Linux expects, the AS/400 specific code builds a
383  * translation table to translate what Linux thinks are "physical"
384  * addresses to the actual real addresses.  This allows us to make
385  * it appear to Linux that we have contiguous memory starting at
386  * physical address zero while in fact this could be far from the truth.
387  * To avoid confusion, I'll let the words physical and/or real address
388  * apply to the Linux addresses while I'll use "absolute address" to
389  * refer to the actual hardware real address.
390  *
391  * build_iSeries_Memory_Map gets information from the Hypervisor and
392  * looks at the Main Store VPD to determine the absolute addresses
393  * of the memory that has been assigned to our partition and builds
394  * a table used to translate Linux's physical addresses to these
395  * absolute addresses.  Absolute addresses are needed when
396  * communicating with the hypervisor (e.g. to build HPT entries)
397  *
398  * Returns the physical memory size
399  */
400
401 static unsigned long __init build_iSeries_Memory_Map(void)
402 {
403         u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
404         u32 nextPhysChunk;
405         u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
406         u32 totalChunks,moreChunks;
407         u32 currChunk, thisChunk, absChunk;
408         u32 currDword;
409         u32 chunkBit;
410         u64 map;
411         struct MemoryBlock mb[32];
412         unsigned long numMemoryBlocks, curBlock;
413
414         /* Chunk size on iSeries is 256K bytes */
415         totalChunks = (u32)HvLpConfig_getMsChunks();
416         mschunks_alloc(totalChunks);
417
418         /*
419          * Get absolute address of our load area
420          * and map it to physical address 0
421          * This guarantees that the loadarea ends up at physical 0
422          * otherwise, it might not be returned by PLIC as the first
423          * chunks
424          */
425
426         loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
427         loadAreaSize =  itLpNaca.xLoadAreaChunks;
428
429         /*
430          * Only add the pages already mapped here.
431          * Otherwise we might add the hpt pages
432          * The rest of the pages of the load area
433          * aren't in the HPT yet and can still
434          * be assigned an arbitrary physical address
435          */
436         if ((loadAreaSize * 64) > HvPagesToMap)
437                 loadAreaSize = HvPagesToMap / 64;
438
439         loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
440
441         /*
442          * TODO Do we need to do something if the HPT is in the 64MB load area?
443          * This would be required if the itLpNaca.xLoadAreaChunks includes
444          * the HPT size
445          */
446
447         printk("Mapping load area - physical addr = 0000000000000000\n"
448                 "                    absolute addr = %016lx\n",
449                 chunk_to_addr(loadAreaFirstChunk));
450         printk("Load area size %dK\n", loadAreaSize * 256);
451
452         for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
453                 mschunks_map.mapping[nextPhysChunk] =
454                         loadAreaFirstChunk + nextPhysChunk;
455
456         /*
457          * Get absolute address of our HPT and remember it so
458          * we won't map it to any physical address
459          */
460         hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
461         hptSizePages = (u32)HvCallHpt_getHptPages();
462         hptSizeChunks = hptSizePages >>
463                 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
464         hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
465
466         printk("HPT absolute addr = %016lx, size = %dK\n",
467                         chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
468
469         /*
470          * Determine if absolute memory has any
471          * holes so that we can interpret the
472          * access map we get back from the hypervisor
473          * correctly.
474          */
475         numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
476
477         /*
478          * Process the main store access map from the hypervisor
479          * to build up our physical -> absolute translation table
480          */
481         curBlock = 0;
482         currChunk = 0;
483         currDword = 0;
484         moreChunks = totalChunks;
485
486         while (moreChunks) {
487                 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
488                                 currDword);
489                 thisChunk = currChunk;
490                 while (map) {
491                         chunkBit = map >> 63;
492                         map <<= 1;
493                         if (chunkBit) {
494                                 --moreChunks;
495                                 while (thisChunk >= mb[curBlock].logicalEnd) {
496                                         ++curBlock;
497                                         if (curBlock >= numMemoryBlocks)
498                                                 panic("out of memory blocks");
499                                 }
500                                 if (thisChunk < mb[curBlock].logicalStart)
501                                         panic("memory block error");
502
503                                 absChunk = mb[curBlock].absStart +
504                                         (thisChunk - mb[curBlock].logicalStart);
505                                 if (((absChunk < hptFirstChunk) ||
506                                      (absChunk > hptLastChunk)) &&
507                                     ((absChunk < loadAreaFirstChunk) ||
508                                      (absChunk > loadAreaLastChunk))) {
509                                         mschunks_map.mapping[nextPhysChunk] =
510                                                 absChunk;
511                                         ++nextPhysChunk;
512                                 }
513                         }
514                         ++thisChunk;
515                 }
516                 ++currDword;
517                 currChunk += 64;
518         }
519
520         /*
521          * main store size (in chunks) is
522          *   totalChunks - hptSizeChunks
523          * which should be equal to
524          *   nextPhysChunk
525          */
526         return chunk_to_addr(nextPhysChunk);
527 }
528
529 /*
530  * Document me.
531  */
532 static void __init iSeries_setup_arch(void)
533 {
534         if (get_lppaca()->shared_proc) {
535                 ppc_md.idle_loop = iseries_shared_idle;
536                 printk(KERN_DEBUG "Using shared processor idle loop\n");
537         } else {
538                 ppc_md.idle_loop = iseries_dedicated_idle;
539                 printk(KERN_DEBUG "Using dedicated idle loop\n");
540         }
541
542         /* Setup the Lp Event Queue */
543         setup_hvlpevent_queue();
544
545         printk("Max  logical processors = %d\n",
546                         itVpdAreas.xSlicMaxLogicalProcs);
547         printk("Max physical processors = %d\n",
548                         itVpdAreas.xSlicMaxPhysicalProcs);
549 }
550
551 static void iSeries_show_cpuinfo(struct seq_file *m)
552 {
553         seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
554 }
555
556 static void __init iSeries_progress(char * st, unsigned short code)
557 {
558         printk("Progress: [%04x] - %s\n", (unsigned)code, st);
559         mf_display_progress(code);
560 }
561
562 static void __init iSeries_fixup_klimit(void)
563 {
564         /*
565          * Change klimit to take into account any ram disk
566          * that may be included
567          */
568         if (naca.xRamDisk)
569                 klimit = KERNELBASE + (u64)naca.xRamDisk +
570                         (naca.xRamDiskSize * HW_PAGE_SIZE);
571         else {
572                 /*
573                  * No ram disk was included - check and see if there
574                  * was an embedded system map.  Change klimit to take
575                  * into account any embedded system map
576                  */
577                 if (embedded_sysmap_end)
578                         klimit = KERNELBASE + ((embedded_sysmap_end + 4095) &
579                                         0xfffffffffffff000);
580         }
581 }
582
583 static int __init iSeries_src_init(void)
584 {
585         /* clear the progress line */
586         ppc_md.progress(" ", 0xffff);
587         return 0;
588 }
589
590 late_initcall(iSeries_src_init);
591
592 static inline void process_iSeries_events(void)
593 {
594         asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
595 }
596
597 static void yield_shared_processor(void)
598 {
599         unsigned long tb;
600
601         HvCall_setEnabledInterrupts(HvCall_MaskIPI |
602                                     HvCall_MaskLpEvent |
603                                     HvCall_MaskLpProd |
604                                     HvCall_MaskTimeout);
605
606         tb = get_tb();
607         /* Compute future tb value when yield should expire */
608         HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
609
610         /*
611          * The decrementer stops during the yield.  Force a fake decrementer
612          * here and let the timer_interrupt code sort out the actual time.
613          */
614         get_lppaca()->int_dword.fields.decr_int = 1;
615         ppc64_runlatch_on();
616         process_iSeries_events();
617 }
618
619 static void iseries_shared_idle(void)
620 {
621         while (1) {
622                 while (!need_resched() && !hvlpevent_is_pending()) {
623                         local_irq_disable();
624                         ppc64_runlatch_off();
625
626                         /* Recheck with irqs off */
627                         if (!need_resched() && !hvlpevent_is_pending())
628                                 yield_shared_processor();
629
630                         HMT_medium();
631                         local_irq_enable();
632                 }
633
634                 ppc64_runlatch_on();
635
636                 if (hvlpevent_is_pending())
637                         process_iSeries_events();
638
639                 preempt_enable_no_resched();
640                 schedule();
641                 preempt_disable();
642         }
643 }
644
645 static void iseries_dedicated_idle(void)
646 {
647         set_thread_flag(TIF_POLLING_NRFLAG);
648
649         while (1) {
650                 if (!need_resched()) {
651                         while (!need_resched()) {
652                                 ppc64_runlatch_off();
653                                 HMT_low();
654
655                                 if (hvlpevent_is_pending()) {
656                                         HMT_medium();
657                                         ppc64_runlatch_on();
658                                         process_iSeries_events();
659                                 }
660                         }
661
662                         HMT_medium();
663                 }
664
665                 ppc64_runlatch_on();
666                 preempt_enable_no_resched();
667                 schedule();
668                 preempt_disable();
669         }
670 }
671
672 #ifndef CONFIG_PCI
673 void __init iSeries_init_IRQ(void) { }
674 #endif
675
676 static int __init iseries_probe(void)
677 {
678         unsigned long root = of_get_flat_dt_root();
679         if (!of_flat_dt_is_compatible(root, "IBM,iSeries"))
680                 return 0;
681
682         powerpc_firmware_features |= FW_FEATURE_ISERIES;
683         powerpc_firmware_features |= FW_FEATURE_LPAR;
684
685         /*
686          * The Hypervisor only allows us up to 256 interrupt
687          * sources (the irq number is passed in a u8).
688          */
689         virt_irq_max = 255;
690
691         return 1;
692 }
693
694 define_machine(iseries) {
695         .name           = "iSeries",
696         .setup_arch     = iSeries_setup_arch,
697         .show_cpuinfo   = iSeries_show_cpuinfo,
698         .init_IRQ       = iSeries_init_IRQ,
699         .get_irq        = iSeries_get_irq,
700         .init_early     = iSeries_init_early,
701         .pcibios_fixup  = iSeries_pci_final_fixup,
702         .restart        = mf_reboot,
703         .power_off      = mf_power_off,
704         .halt           = mf_power_off,
705         .get_boot_time  = iSeries_get_boot_time,
706         .set_rtc_time   = iSeries_set_rtc_time,
707         .get_rtc_time   = iSeries_get_rtc_time,
708         .calibrate_decr = generic_calibrate_decr,
709         .progress       = iSeries_progress,
710         .probe          = iseries_probe,
711         /* XXX Implement enable_pmcs for iSeries */
712 };
713
714 struct blob {
715         unsigned char data[PAGE_SIZE * 2];
716         unsigned long next;
717 };
718
719 struct iseries_flat_dt {
720         struct boot_param_header header;
721         u64 reserve_map[2];
722         struct blob dt;
723         struct blob strings;
724 };
725
726 struct iseries_flat_dt iseries_dt;
727
728 void dt_init(struct iseries_flat_dt *dt)
729 {
730         dt->header.off_mem_rsvmap =
731                 offsetof(struct iseries_flat_dt, reserve_map);
732         dt->header.off_dt_struct = offsetof(struct iseries_flat_dt, dt);
733         dt->header.off_dt_strings = offsetof(struct iseries_flat_dt, strings);
734         dt->header.totalsize = sizeof(struct iseries_flat_dt);
735         dt->header.dt_strings_size = sizeof(struct blob);
736
737         /* There is no notion of hardware cpu id on iSeries */
738         dt->header.boot_cpuid_phys = smp_processor_id();
739
740         dt->dt.next = (unsigned long)&dt->dt.data;
741         dt->strings.next = (unsigned long)&dt->strings.data;
742
743         dt->header.magic = OF_DT_HEADER;
744         dt->header.version = 0x10;
745         dt->header.last_comp_version = 0x10;
746
747         dt->reserve_map[0] = 0;
748         dt->reserve_map[1] = 0;
749 }
750
751 void dt_check_blob(struct blob *b)
752 {
753         if (b->next >= (unsigned long)&b->next) {
754                 DBG("Ran out of space in flat device tree blob!\n");
755                 BUG();
756         }
757 }
758
759 void dt_push_u32(struct iseries_flat_dt *dt, u32 value)
760 {
761         *((u32*)dt->dt.next) = value;
762         dt->dt.next += sizeof(u32);
763
764         dt_check_blob(&dt->dt);
765 }
766
767 void dt_push_u64(struct iseries_flat_dt *dt, u64 value)
768 {
769         *((u64*)dt->dt.next) = value;
770         dt->dt.next += sizeof(u64);
771
772         dt_check_blob(&dt->dt);
773 }
774
775 unsigned long dt_push_bytes(struct blob *blob, char *data, int len)
776 {
777         unsigned long start = blob->next - (unsigned long)blob->data;
778
779         memcpy((char *)blob->next, data, len);
780         blob->next = _ALIGN(blob->next + len, 4);
781
782         dt_check_blob(blob);
783
784         return start;
785 }
786
787 void dt_start_node(struct iseries_flat_dt *dt, char *name)
788 {
789         dt_push_u32(dt, OF_DT_BEGIN_NODE);
790         dt_push_bytes(&dt->dt, name, strlen(name) + 1);
791 }
792
793 #define dt_end_node(dt) dt_push_u32(dt, OF_DT_END_NODE)
794
795 void dt_prop(struct iseries_flat_dt *dt, char *name, char *data, int len)
796 {
797         unsigned long offset;
798
799         dt_push_u32(dt, OF_DT_PROP);
800
801         /* Length of the data */
802         dt_push_u32(dt, len);
803
804         /* Put the property name in the string blob. */
805         offset = dt_push_bytes(&dt->strings, name, strlen(name) + 1);
806
807         /* The offset of the properties name in the string blob. */
808         dt_push_u32(dt, (u32)offset);
809
810         /* The actual data. */
811         dt_push_bytes(&dt->dt, data, len);
812 }
813
814 void dt_prop_str(struct iseries_flat_dt *dt, char *name, char *data)
815 {
816         dt_prop(dt, name, data, strlen(data) + 1); /* + 1 for NULL */
817 }
818
819 void dt_prop_u32(struct iseries_flat_dt *dt, char *name, u32 data)
820 {
821         dt_prop(dt, name, (char *)&data, sizeof(u32));
822 }
823
824 void dt_prop_u64(struct iseries_flat_dt *dt, char *name, u64 data)
825 {
826         dt_prop(dt, name, (char *)&data, sizeof(u64));
827 }
828
829 void dt_prop_u64_list(struct iseries_flat_dt *dt, char *name, u64 *data, int n)
830 {
831         dt_prop(dt, name, (char *)data, sizeof(u64) * n);
832 }
833
834 void dt_prop_u32_list(struct iseries_flat_dt *dt, char *name, u32 *data, int n)
835 {
836         dt_prop(dt, name, (char *)data, sizeof(u32) * n);
837 }
838
839 void dt_prop_empty(struct iseries_flat_dt *dt, char *name)
840 {
841         dt_prop(dt, name, NULL, 0);
842 }
843
844 void dt_cpus(struct iseries_flat_dt *dt)
845 {
846         unsigned char buf[32];
847         unsigned char *p;
848         unsigned int i, index;
849         struct IoHriProcessorVpd *d;
850         u32 pft_size[2];
851
852         /* yuck */
853         snprintf(buf, 32, "PowerPC,%s", cur_cpu_spec->cpu_name);
854         p = strchr(buf, ' ');
855         if (!p) p = buf + strlen(buf);
856
857         dt_start_node(dt, "cpus");
858         dt_prop_u32(dt, "#address-cells", 1);
859         dt_prop_u32(dt, "#size-cells", 0);
860
861         pft_size[0] = 0; /* NUMA CEC cookie, 0 for non NUMA  */
862         pft_size[1] = __ilog2(HvCallHpt_getHptPages() * HW_PAGE_SIZE);
863
864         for (i = 0; i < NR_CPUS; i++) {
865                 if (lppaca[i].dyn_proc_status >= 2)
866                         continue;
867
868                 snprintf(p, 32 - (p - buf), "@%d", i);
869                 dt_start_node(dt, buf);
870
871                 dt_prop_str(dt, "device_type", "cpu");
872
873                 index = lppaca[i].dyn_hv_phys_proc_index;
874                 d = &xIoHriProcessorVpd[index];
875
876                 dt_prop_u32(dt, "i-cache-size", d->xInstCacheSize * 1024);
877                 dt_prop_u32(dt, "i-cache-line-size", d->xInstCacheOperandSize);
878
879                 dt_prop_u32(dt, "d-cache-size", d->xDataL1CacheSizeKB * 1024);
880                 dt_prop_u32(dt, "d-cache-line-size", d->xDataCacheOperandSize);
881
882                 /* magic conversions to Hz copied from old code */
883                 dt_prop_u32(dt, "clock-frequency",
884                         ((1UL << 34) * 1000000) / d->xProcFreq);
885                 dt_prop_u32(dt, "timebase-frequency",
886                         ((1UL << 32) * 1000000) / d->xTimeBaseFreq);
887
888                 dt_prop_u32(dt, "reg", i);
889
890                 dt_prop_u32_list(dt, "ibm,pft-size", pft_size, 2);
891
892                 dt_end_node(dt);
893         }
894
895         dt_end_node(dt);
896 }
897
898 void dt_model(struct iseries_flat_dt *dt)
899 {
900         char buf[16] = "IBM,";
901
902         /* "IBM," + mfgId[2:3] + systemSerial[1:5] */
903         strne2a(buf + 4, xItExtVpdPanel.mfgID + 2, 2);
904         strne2a(buf + 6, xItExtVpdPanel.systemSerial + 1, 5);
905         buf[11] = '\0';
906         dt_prop_str(dt, "system-id", buf);
907
908         /* "IBM," + machineType[0:4] */
909         strne2a(buf + 4, xItExtVpdPanel.machineType, 4);
910         buf[8] = '\0';
911         dt_prop_str(dt, "model", buf);
912
913         dt_prop_str(dt, "compatible", "IBM,iSeries");
914 }
915
916 void dt_vdevices(struct iseries_flat_dt *dt)
917 {
918         u32 reg = 0;
919         HvLpIndexMap vlan_map;
920         int i;
921         char buf[32];
922
923         dt_start_node(dt, "vdevice");
924         dt_prop_u32(dt, "#address-cells", 1);
925         dt_prop_u32(dt, "#size-cells", 0);
926
927         snprintf(buf, sizeof(buf), "viocons@%08x", reg);
928         dt_start_node(dt, buf);
929         dt_prop_str(dt, "device_type", "serial");
930         dt_prop_str(dt, "compatible", "");
931         dt_prop_u32(dt, "reg", reg);
932         dt_end_node(dt);
933         reg++;
934
935         snprintf(buf, sizeof(buf), "v-scsi@%08x", reg);
936         dt_start_node(dt, buf);
937         dt_prop_str(dt, "device_type", "vscsi");
938         dt_prop_str(dt, "compatible", "IBM,v-scsi");
939         dt_prop_u32(dt, "reg", reg);
940         dt_end_node(dt);
941         reg++;
942
943         vlan_map = HvLpConfig_getVirtualLanIndexMap();
944         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
945                 unsigned char mac_addr[ETH_ALEN];
946
947                 if ((vlan_map & (0x8000 >> i)) == 0)
948                         continue;
949                 snprintf(buf, 32, "l-lan@%08x", reg + i);
950                 dt_start_node(dt, buf);
951                 dt_prop_str(dt, "device_type", "network");
952                 dt_prop_str(dt, "compatible", "IBM,iSeries-l-lan");
953                 dt_prop_u32(dt, "reg", reg + i);
954                 dt_prop_u32(dt, "linux,unit_address", i);
955
956                 mac_addr[0] = 0x02;
957                 mac_addr[1] = 0x01;
958                 mac_addr[2] = 0xff;
959                 mac_addr[3] = i;
960                 mac_addr[4] = 0xff;
961                 mac_addr[5] = HvLpConfig_getLpIndex_outline();
962                 dt_prop(dt, "local-mac-address", (char *)mac_addr, ETH_ALEN);
963                 dt_prop(dt, "mac-address", (char *)mac_addr, ETH_ALEN);
964                 dt_prop_u32(dt, "max-frame-size", 9000);
965                 dt_prop_u32(dt, "address-bits", 48);
966
967                 dt_end_node(dt);
968         }
969         reg += HVMAXARCHITECTEDVIRTUALLANS;
970
971         for (i = 0; i < HVMAXARCHITECTEDVIRTUALDISKS; i++) {
972                 snprintf(buf, 32, "viodasd@%08x", reg + i);
973                 dt_start_node(dt, buf);
974                 dt_prop_str(dt, "device_type", "viodasd");
975                 dt_prop_str(dt, "compatible", "");
976                 dt_prop_u32(dt, "reg", reg + i);
977                 dt_prop_u32(dt, "linux,unit_address", i);
978                 dt_end_node(dt);
979         }
980         reg += HVMAXARCHITECTEDVIRTUALDISKS;
981         for (i = 0; i < HVMAXARCHITECTEDVIRTUALCDROMS; i++) {
982                 snprintf(buf, 32, "viocd@%08x", reg + i);
983                 dt_start_node(dt, buf);
984                 dt_prop_str(dt, "device_type", "viocd");
985                 dt_prop_str(dt, "compatible", "");
986                 dt_prop_u32(dt, "reg", reg + i);
987                 dt_prop_u32(dt, "linux,unit_address", i);
988                 dt_end_node(dt);
989         }
990         reg += HVMAXARCHITECTEDVIRTUALCDROMS;
991         for (i = 0; i < HVMAXARCHITECTEDVIRTUALTAPES; i++) {
992                 snprintf(buf, 32, "viotape@%08x", reg + i);
993                 dt_start_node(dt, buf);
994                 dt_prop_str(dt, "device_type", "viotape");
995                 dt_prop_str(dt, "compatible", "");
996                 dt_prop_u32(dt, "reg", reg + i);
997                 dt_prop_u32(dt, "linux,unit_address", i);
998                 dt_end_node(dt);
999         }
1000
1001         dt_end_node(dt);
1002 }
1003
1004 void build_flat_dt(struct iseries_flat_dt *dt, unsigned long phys_mem_size)
1005 {
1006         u64 tmp[2];
1007
1008         dt_init(dt);
1009
1010         dt_start_node(dt, "");
1011
1012         dt_prop_u32(dt, "#address-cells", 2);
1013         dt_prop_u32(dt, "#size-cells", 2);
1014         dt_model(dt);
1015
1016         /* /memory */
1017         dt_start_node(dt, "memory@0");
1018         dt_prop_str(dt, "name", "memory");
1019         dt_prop_str(dt, "device_type", "memory");
1020         tmp[0] = 0;
1021         tmp[1] = phys_mem_size;
1022         dt_prop_u64_list(dt, "reg", tmp, 2);
1023         dt_end_node(dt);
1024
1025         /* /chosen */
1026         dt_start_node(dt, "chosen");
1027         dt_prop_str(dt, "bootargs", cmd_line);
1028         if (cmd_mem_limit)
1029                 dt_prop_u64(dt, "linux,memory-limit", cmd_mem_limit);
1030         dt_end_node(dt);
1031
1032         dt_cpus(dt);
1033
1034         dt_vdevices(dt);
1035
1036         dt_end_node(dt);
1037
1038         dt_push_u32(dt, OF_DT_END);
1039 }
1040
1041 void * __init iSeries_early_setup(void)
1042 {
1043         unsigned long phys_mem_size;
1044
1045         iSeries_fixup_klimit();
1046
1047         /*
1048          * Initialize the table which translate Linux physical addresses to
1049          * AS/400 absolute addresses
1050          */
1051         phys_mem_size = build_iSeries_Memory_Map();
1052
1053         iSeries_get_cmdline();
1054
1055         /* Save unparsed command line copy for /proc/cmdline */
1056         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
1057
1058         /* Parse early parameters, in particular mem=x */
1059         parse_early_param();
1060
1061         build_flat_dt(&iseries_dt, phys_mem_size);
1062
1063         return (void *) __pa(&iseries_dt);
1064 }
1065
1066 /*
1067  * On iSeries we just parse the mem=X option from the command line.
1068  * On pSeries it's a bit more complicated, see prom_init_mem()
1069  */
1070 static int __init early_parsemem(char *p)
1071 {
1072         if (p)
1073                 cmd_mem_limit = ALIGN(memparse(p, &p), PAGE_SIZE);
1074         return 0;
1075 }
1076 early_param("mem", early_parsemem);
1077
1078 static void hvputc(char c)
1079 {
1080         if (c == '\n')
1081                 hvputc('\r');
1082
1083         HvCall_writeLogBuffer(&c, 1);
1084 }
1085
1086 void __init udbg_init_iseries(void)
1087 {
1088         udbg_putc = hvputc;
1089 }