2 * Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com>
3 * Copyright (c) 1999-2000 Grant Erickson <grant@lcse.umn.edu>
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
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.
19 #include <linux/init.h>
20 #include <linux/threads.h>
21 #include <linux/smp.h>
22 #include <linux/param.h>
23 #include <linux/string.h>
24 #include <linux/seq_file.h>
25 #include <linux/kdev_t.h>
26 #include <linux/major.h>
27 #include <linux/root_dev.h>
28 #include <linux/kernel.h>
29 #include <linux/hrtimer.h>
30 #include <linux/tick.h>
32 #include <asm/processor.h>
33 #include <asm/machdep.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>
45 #include <asm/cache.h>
46 #include <asm/abs_addr.h>
47 #include <asm/iseries/hv_lp_config.h>
48 #include <asm/iseries/hv_call_event.h>
49 #include <asm/iseries/hv_call_xm.h>
50 #include <asm/iseries/it_lp_queue.h>
51 #include <asm/iseries/mf.h>
52 #include <asm/iseries/hv_lp_event.h>
53 #include <asm/iseries/lpar_map.h>
60 #include "vpd_areas.h"
61 #include "processor_vpd.h"
62 #include "it_lp_naca.h"
63 #include "main_store.h"
69 #define DBG(fmt...) udbg_printf(fmt)
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);
81 unsigned long absStart;
83 unsigned long logicalStart;
84 unsigned long logicalEnd;
88 * Process the main store vpd to determine where the holes in memory are
89 * and return the number of physical blocks and fill in the array of
92 static unsigned long iSeries_process_Condor_mainstore_vpd(
93 struct MemoryBlock *mb_array, unsigned long max_entries)
95 unsigned long holeFirstChunk, holeSizeChunks;
96 unsigned long numMemoryBlocks = 1;
97 struct IoHriMainStoreSegment4 *msVpd =
98 (struct IoHriMainStoreSegment4 *)xMsVpd;
99 unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
100 unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
101 unsigned long holeSize = holeEnd - holeStart;
103 printk("Mainstore_VPD: Condor\n");
105 * Determine if absolute memory has any
106 * holes so that we can interpret the
107 * access map we get back from the hypervisor
110 mb_array[0].logicalStart = 0;
111 mb_array[0].logicalEnd = 0x100000000UL;
112 mb_array[0].absStart = 0;
113 mb_array[0].absEnd = 0x100000000UL;
117 holeStart = holeStart & 0x000fffffffffffffUL;
118 holeStart = addr_to_chunk(holeStart);
119 holeFirstChunk = holeStart;
120 holeSize = addr_to_chunk(holeSize);
121 holeSizeChunks = holeSize;
122 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
123 holeFirstChunk, holeSizeChunks );
124 mb_array[0].logicalEnd = holeFirstChunk;
125 mb_array[0].absEnd = holeFirstChunk;
126 mb_array[1].logicalStart = holeFirstChunk;
127 mb_array[1].logicalEnd = 0x100000000UL - holeSizeChunks;
128 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
129 mb_array[1].absEnd = 0x100000000UL;
131 return numMemoryBlocks;
134 #define MaxSegmentAreas 32
135 #define MaxSegmentAdrRangeBlocks 128
136 #define MaxAreaRangeBlocks 4
138 static unsigned long iSeries_process_Regatta_mainstore_vpd(
139 struct MemoryBlock *mb_array, unsigned long max_entries)
141 struct IoHriMainStoreSegment5 *msVpdP =
142 (struct IoHriMainStoreSegment5 *)xMsVpd;
143 unsigned long numSegmentBlocks = 0;
144 u32 existsBits = msVpdP->msAreaExists;
145 unsigned long area_num;
147 printk("Mainstore_VPD: Regatta\n");
149 for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
150 unsigned long numAreaBlocks;
151 struct IoHriMainStoreArea4 *currentArea;
153 if (existsBits & 0x80000000) {
154 unsigned long block_num;
156 currentArea = &msVpdP->msAreaArray[area_num];
157 numAreaBlocks = currentArea->numAdrRangeBlocks;
158 printk("ms_vpd: processing area %2ld blocks=%ld",
159 area_num, numAreaBlocks);
160 for (block_num = 0; block_num < numAreaBlocks;
162 /* Process an address range block */
163 struct MemoryBlock tempBlock;
167 (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
169 (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
170 tempBlock.logicalStart = 0;
171 tempBlock.logicalEnd = 0;
172 printk("\n block %ld absStart=%016lx absEnd=%016lx",
173 block_num, tempBlock.absStart,
176 for (i = 0; i < numSegmentBlocks; ++i) {
177 if (mb_array[i].absStart ==
181 if (i == numSegmentBlocks) {
182 if (numSegmentBlocks == max_entries)
183 panic("iSeries_process_mainstore_vpd: too many memory blocks");
184 mb_array[numSegmentBlocks] = tempBlock;
187 printk(" (duplicate)");
193 /* Now sort the blocks found into ascending sequence */
194 if (numSegmentBlocks > 1) {
197 for (m = 0; m < numSegmentBlocks - 1; ++m) {
198 for (n = numSegmentBlocks - 1; m < n; --n) {
199 if (mb_array[n].absStart <
200 mb_array[n-1].absStart) {
201 struct MemoryBlock tempBlock;
203 tempBlock = mb_array[n];
204 mb_array[n] = mb_array[n-1];
205 mb_array[n-1] = tempBlock;
211 * Assign "logical" addresses to each block. These
212 * addresses correspond to the hypervisor "bitmap" space.
213 * Convert all addresses into units of 256K chunks.
216 unsigned long i, nextBitmapAddress;
218 printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
219 nextBitmapAddress = 0;
220 for (i = 0; i < numSegmentBlocks; ++i) {
221 unsigned long length = mb_array[i].absEnd -
222 mb_array[i].absStart;
224 mb_array[i].logicalStart = nextBitmapAddress;
225 mb_array[i].logicalEnd = nextBitmapAddress + length;
226 nextBitmapAddress += length;
227 printk(" Bitmap range: %016lx - %016lx\n"
228 " Absolute range: %016lx - %016lx\n",
229 mb_array[i].logicalStart,
230 mb_array[i].logicalEnd,
231 mb_array[i].absStart, mb_array[i].absEnd);
232 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
233 0x000fffffffffffffUL);
234 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
235 0x000fffffffffffffUL);
236 mb_array[i].logicalStart =
237 addr_to_chunk(mb_array[i].logicalStart);
238 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
242 return numSegmentBlocks;
245 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
246 unsigned long max_entries)
249 unsigned long mem_blocks = 0;
251 if (cpu_has_feature(CPU_FTR_SLB))
252 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
255 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
258 printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
259 for (i = 0; i < mem_blocks; ++i) {
260 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
261 " abs chunks %016lx - %016lx\n",
262 i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
263 mb_array[i].absStart, mb_array[i].absEnd);
268 static void __init iSeries_get_cmdline(void)
272 /* copy the command line parameter from the primary VSP */
273 HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
274 HvLpDma_Direction_RemoteToLocal);
279 if (!*p || *p == '\n')
286 static void __init iSeries_init_early(void)
288 DBG(" -> iSeries_init_early()\n");
290 /* Snapshot the timebase, for use in later recalibration */
291 iSeries_time_init_early();
294 * Initialize the DMA/TCE management
296 iommu_init_early_iSeries();
298 /* Initialize machine-dependency vectors */
303 /* Associate Lp Event Queue 0 with processor 0 */
304 HvCallEvent_setLpEventQueueInterruptProc(0, 0);
308 DBG(" <- iSeries_init_early()\n");
311 struct mschunks_map mschunks_map = {
312 /* XXX We don't use these, but Piranha might need them. */
313 .chunk_size = MSCHUNKS_CHUNK_SIZE,
314 .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
315 .chunk_mask = MSCHUNKS_OFFSET_MASK,
317 EXPORT_SYMBOL(mschunks_map);
319 static void mschunks_alloc(unsigned long num_chunks)
321 klimit = _ALIGN(klimit, sizeof(u32));
322 mschunks_map.mapping = (u32 *)klimit;
323 klimit += num_chunks * sizeof(u32);
324 mschunks_map.num_chunks = num_chunks;
328 * The iSeries may have very large memories ( > 128 GB ) and a partition
329 * may get memory in "chunks" that may be anywhere in the 2**52 real
330 * address space. The chunks are 256K in size. To map this to the
331 * memory model Linux expects, the AS/400 specific code builds a
332 * translation table to translate what Linux thinks are "physical"
333 * addresses to the actual real addresses. This allows us to make
334 * it appear to Linux that we have contiguous memory starting at
335 * physical address zero while in fact this could be far from the truth.
336 * To avoid confusion, I'll let the words physical and/or real address
337 * apply to the Linux addresses while I'll use "absolute address" to
338 * refer to the actual hardware real address.
340 * build_iSeries_Memory_Map gets information from the Hypervisor and
341 * looks at the Main Store VPD to determine the absolute addresses
342 * of the memory that has been assigned to our partition and builds
343 * a table used to translate Linux's physical addresses to these
344 * absolute addresses. Absolute addresses are needed when
345 * communicating with the hypervisor (e.g. to build HPT entries)
347 * Returns the physical memory size
350 static unsigned long __init build_iSeries_Memory_Map(void)
352 u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
354 u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
355 u32 totalChunks,moreChunks;
356 u32 currChunk, thisChunk, absChunk;
360 struct MemoryBlock mb[32];
361 unsigned long numMemoryBlocks, curBlock;
363 /* Chunk size on iSeries is 256K bytes */
364 totalChunks = (u32)HvLpConfig_getMsChunks();
365 mschunks_alloc(totalChunks);
368 * Get absolute address of our load area
369 * and map it to physical address 0
370 * This guarantees that the loadarea ends up at physical 0
371 * otherwise, it might not be returned by PLIC as the first
375 loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
376 loadAreaSize = itLpNaca.xLoadAreaChunks;
379 * Only add the pages already mapped here.
380 * Otherwise we might add the hpt pages
381 * The rest of the pages of the load area
382 * aren't in the HPT yet and can still
383 * be assigned an arbitrary physical address
385 if ((loadAreaSize * 64) > HvPagesToMap)
386 loadAreaSize = HvPagesToMap / 64;
388 loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
391 * TODO Do we need to do something if the HPT is in the 64MB load area?
392 * This would be required if the itLpNaca.xLoadAreaChunks includes
396 printk("Mapping load area - physical addr = 0000000000000000\n"
397 " absolute addr = %016lx\n",
398 chunk_to_addr(loadAreaFirstChunk));
399 printk("Load area size %dK\n", loadAreaSize * 256);
401 for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
402 mschunks_map.mapping[nextPhysChunk] =
403 loadAreaFirstChunk + nextPhysChunk;
406 * Get absolute address of our HPT and remember it so
407 * we won't map it to any physical address
409 hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
410 hptSizePages = (u32)HvCallHpt_getHptPages();
411 hptSizeChunks = hptSizePages >>
412 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
413 hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
415 printk("HPT absolute addr = %016lx, size = %dK\n",
416 chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
419 * Determine if absolute memory has any
420 * holes so that we can interpret the
421 * access map we get back from the hypervisor
424 numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
427 * Process the main store access map from the hypervisor
428 * to build up our physical -> absolute translation table
433 moreChunks = totalChunks;
436 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
438 thisChunk = currChunk;
440 chunkBit = map >> 63;
444 while (thisChunk >= mb[curBlock].logicalEnd) {
446 if (curBlock >= numMemoryBlocks)
447 panic("out of memory blocks");
449 if (thisChunk < mb[curBlock].logicalStart)
450 panic("memory block error");
452 absChunk = mb[curBlock].absStart +
453 (thisChunk - mb[curBlock].logicalStart);
454 if (((absChunk < hptFirstChunk) ||
455 (absChunk > hptLastChunk)) &&
456 ((absChunk < loadAreaFirstChunk) ||
457 (absChunk > loadAreaLastChunk))) {
458 mschunks_map.mapping[nextPhysChunk] =
470 * main store size (in chunks) is
471 * totalChunks - hptSizeChunks
472 * which should be equal to
475 return chunk_to_addr(nextPhysChunk);
481 static void __init iSeries_setup_arch(void)
483 if (get_lppaca()->shared_proc) {
484 ppc_md.idle_loop = iseries_shared_idle;
485 printk(KERN_DEBUG "Using shared processor idle loop\n");
487 ppc_md.idle_loop = iseries_dedicated_idle;
488 printk(KERN_DEBUG "Using dedicated idle loop\n");
491 /* Setup the Lp Event Queue */
492 setup_hvlpevent_queue();
494 printk("Max logical processors = %d\n",
495 itVpdAreas.xSlicMaxLogicalProcs);
496 printk("Max physical processors = %d\n",
497 itVpdAreas.xSlicMaxPhysicalProcs);
499 iSeries_pcibios_init();
502 static void iSeries_show_cpuinfo(struct seq_file *m)
504 seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
507 static void __init iSeries_progress(char * st, unsigned short code)
509 printk("Progress: [%04x] - %s\n", (unsigned)code, st);
510 mf_display_progress(code);
513 static void __init iSeries_fixup_klimit(void)
516 * Change klimit to take into account any ram disk
517 * that may be included
520 klimit = KERNELBASE + (u64)naca.xRamDisk +
521 (naca.xRamDiskSize * HW_PAGE_SIZE);
524 static int __init iSeries_src_init(void)
526 /* clear the progress line */
527 if (firmware_has_feature(FW_FEATURE_ISERIES))
528 ppc_md.progress(" ", 0xffff);
532 late_initcall(iSeries_src_init);
534 static inline void process_iSeries_events(void)
536 asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
539 static void yield_shared_processor(void)
543 HvCall_setEnabledInterrupts(HvCall_MaskIPI |
549 /* Compute future tb value when yield should expire */
550 HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
553 * The decrementer stops during the yield. Force a fake decrementer
554 * here and let the timer_interrupt code sort out the actual time.
556 get_lppaca()->int_dword.fields.decr_int = 1;
558 process_iSeries_events();
561 static void iseries_shared_idle(void)
564 tick_nohz_stop_sched_tick(1);
565 while (!need_resched() && !hvlpevent_is_pending()) {
567 ppc64_runlatch_off();
569 /* Recheck with irqs off */
570 if (!need_resched() && !hvlpevent_is_pending())
571 yield_shared_processor();
578 tick_nohz_restart_sched_tick();
580 if (hvlpevent_is_pending())
581 process_iSeries_events();
583 preempt_enable_no_resched();
589 static void iseries_dedicated_idle(void)
591 set_thread_flag(TIF_POLLING_NRFLAG);
594 tick_nohz_stop_sched_tick(1);
595 if (!need_resched()) {
596 while (!need_resched()) {
597 ppc64_runlatch_off();
600 if (hvlpevent_is_pending()) {
603 process_iSeries_events();
611 tick_nohz_restart_sched_tick();
612 preempt_enable_no_resched();
618 static void __iomem *iseries_ioremap(phys_addr_t address, unsigned long size,
621 return (void __iomem *)address;
624 static void iseries_iounmap(volatile void __iomem *token)
628 static int __init iseries_probe(void)
630 unsigned long root = of_get_flat_dt_root();
631 if (!of_flat_dt_is_compatible(root, "IBM,iSeries"))
635 /* iSeries does not support 16M pages */
636 cur_cpu_spec->cpu_features &= ~CPU_FTR_16M_PAGE;
641 define_machine(iseries) {
643 .setup_arch = iSeries_setup_arch,
644 .show_cpuinfo = iSeries_show_cpuinfo,
645 .init_IRQ = iSeries_init_IRQ,
646 .get_irq = iSeries_get_irq,
647 .init_early = iSeries_init_early,
648 .pcibios_fixup = iSeries_pci_final_fixup,
649 .pcibios_fixup_resources= iSeries_pcibios_fixup_resources,
650 .restart = mf_reboot,
651 .power_off = mf_power_off,
652 .halt = mf_power_off,
653 .get_boot_time = iSeries_get_boot_time,
654 .set_rtc_time = iSeries_set_rtc_time,
655 .get_rtc_time = iSeries_get_rtc_time,
656 .calibrate_decr = generic_calibrate_decr,
657 .progress = iSeries_progress,
658 .probe = iseries_probe,
659 .ioremap = iseries_ioremap,
660 .iounmap = iseries_iounmap,
661 /* XXX Implement enable_pmcs for iSeries */
664 void * __init iSeries_early_setup(void)
666 unsigned long phys_mem_size;
668 /* Identify CPU type. This is done again by the common code later
669 * on but calling this function multiple times is fine.
671 identify_cpu(0, mfspr(SPRN_PVR));
673 powerpc_firmware_features |= FW_FEATURE_ISERIES;
674 powerpc_firmware_features |= FW_FEATURE_LPAR;
676 iSeries_fixup_klimit();
679 * Initialize the table which translate Linux physical addresses to
680 * AS/400 absolute addresses
682 phys_mem_size = build_iSeries_Memory_Map();
684 iSeries_get_cmdline();
686 return (void *) __pa(build_flat_dt(phys_mem_size));
689 static void hvputc(char c)
694 HvCall_writeLogBuffer(&c, 1);
697 void __init udbg_init_iseries(void)