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/kexec.h>
27 #include <linux/major.h>
28 #include <linux/root_dev.h>
29 #include <linux/kernel.h>
30 #include <linux/hrtimer.h>
31 #include <linux/tick.h>
33 #include <asm/processor.h>
34 #include <asm/machdep.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>
46 #include <asm/cache.h>
47 #include <asm/abs_addr.h>
48 #include <asm/iseries/hv_lp_config.h>
49 #include <asm/iseries/hv_call_event.h>
50 #include <asm/iseries/hv_call_xm.h>
51 #include <asm/iseries/it_lp_queue.h>
52 #include <asm/iseries/mf.h>
53 #include <asm/iseries/hv_lp_event.h>
54 #include <asm/iseries/lpar_map.h>
61 #include "vpd_areas.h"
62 #include "processor_vpd.h"
63 #include "it_lp_naca.h"
64 #include "main_store.h"
70 #define DBG(fmt...) udbg_printf(fmt)
75 /* Function Prototypes */
76 static unsigned long build_iSeries_Memory_Map(void);
77 static void iseries_shared_idle(void);
78 static void iseries_dedicated_idle(void);
82 unsigned long absStart;
84 unsigned long logicalStart;
85 unsigned long logicalEnd;
89 * Process the main store vpd to determine where the holes in memory are
90 * and return the number of physical blocks and fill in the array of
93 static unsigned long iSeries_process_Condor_mainstore_vpd(
94 struct MemoryBlock *mb_array, unsigned long max_entries)
96 unsigned long holeFirstChunk, holeSizeChunks;
97 unsigned long numMemoryBlocks = 1;
98 struct IoHriMainStoreSegment4 *msVpd =
99 (struct IoHriMainStoreSegment4 *)xMsVpd;
100 unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
101 unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
102 unsigned long holeSize = holeEnd - holeStart;
104 printk("Mainstore_VPD: Condor\n");
106 * Determine if absolute memory has any
107 * holes so that we can interpret the
108 * access map we get back from the hypervisor
111 mb_array[0].logicalStart = 0;
112 mb_array[0].logicalEnd = 0x100000000UL;
113 mb_array[0].absStart = 0;
114 mb_array[0].absEnd = 0x100000000UL;
118 holeStart = holeStart & 0x000fffffffffffffUL;
119 holeStart = addr_to_chunk(holeStart);
120 holeFirstChunk = holeStart;
121 holeSize = addr_to_chunk(holeSize);
122 holeSizeChunks = holeSize;
123 printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
124 holeFirstChunk, holeSizeChunks );
125 mb_array[0].logicalEnd = holeFirstChunk;
126 mb_array[0].absEnd = holeFirstChunk;
127 mb_array[1].logicalStart = holeFirstChunk;
128 mb_array[1].logicalEnd = 0x100000000UL - holeSizeChunks;
129 mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
130 mb_array[1].absEnd = 0x100000000UL;
132 return numMemoryBlocks;
135 #define MaxSegmentAreas 32
136 #define MaxSegmentAdrRangeBlocks 128
137 #define MaxAreaRangeBlocks 4
139 static unsigned long iSeries_process_Regatta_mainstore_vpd(
140 struct MemoryBlock *mb_array, unsigned long max_entries)
142 struct IoHriMainStoreSegment5 *msVpdP =
143 (struct IoHriMainStoreSegment5 *)xMsVpd;
144 unsigned long numSegmentBlocks = 0;
145 u32 existsBits = msVpdP->msAreaExists;
146 unsigned long area_num;
148 printk("Mainstore_VPD: Regatta\n");
150 for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
151 unsigned long numAreaBlocks;
152 struct IoHriMainStoreArea4 *currentArea;
154 if (existsBits & 0x80000000) {
155 unsigned long block_num;
157 currentArea = &msVpdP->msAreaArray[area_num];
158 numAreaBlocks = currentArea->numAdrRangeBlocks;
159 printk("ms_vpd: processing area %2ld blocks=%ld",
160 area_num, numAreaBlocks);
161 for (block_num = 0; block_num < numAreaBlocks;
163 /* Process an address range block */
164 struct MemoryBlock tempBlock;
168 (unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
170 (unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
171 tempBlock.logicalStart = 0;
172 tempBlock.logicalEnd = 0;
173 printk("\n block %ld absStart=%016lx absEnd=%016lx",
174 block_num, tempBlock.absStart,
177 for (i = 0; i < numSegmentBlocks; ++i) {
178 if (mb_array[i].absStart ==
182 if (i == numSegmentBlocks) {
183 if (numSegmentBlocks == max_entries)
184 panic("iSeries_process_mainstore_vpd: too many memory blocks");
185 mb_array[numSegmentBlocks] = tempBlock;
188 printk(" (duplicate)");
194 /* Now sort the blocks found into ascending sequence */
195 if (numSegmentBlocks > 1) {
198 for (m = 0; m < numSegmentBlocks - 1; ++m) {
199 for (n = numSegmentBlocks - 1; m < n; --n) {
200 if (mb_array[n].absStart <
201 mb_array[n-1].absStart) {
202 struct MemoryBlock tempBlock;
204 tempBlock = mb_array[n];
205 mb_array[n] = mb_array[n-1];
206 mb_array[n-1] = tempBlock;
212 * Assign "logical" addresses to each block. These
213 * addresses correspond to the hypervisor "bitmap" space.
214 * Convert all addresses into units of 256K chunks.
217 unsigned long i, nextBitmapAddress;
219 printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
220 nextBitmapAddress = 0;
221 for (i = 0; i < numSegmentBlocks; ++i) {
222 unsigned long length = mb_array[i].absEnd -
223 mb_array[i].absStart;
225 mb_array[i].logicalStart = nextBitmapAddress;
226 mb_array[i].logicalEnd = nextBitmapAddress + length;
227 nextBitmapAddress += length;
228 printk(" Bitmap range: %016lx - %016lx\n"
229 " Absolute range: %016lx - %016lx\n",
230 mb_array[i].logicalStart,
231 mb_array[i].logicalEnd,
232 mb_array[i].absStart, mb_array[i].absEnd);
233 mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
234 0x000fffffffffffffUL);
235 mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
236 0x000fffffffffffffUL);
237 mb_array[i].logicalStart =
238 addr_to_chunk(mb_array[i].logicalStart);
239 mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
243 return numSegmentBlocks;
246 static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
247 unsigned long max_entries)
250 unsigned long mem_blocks = 0;
252 if (cpu_has_feature(CPU_FTR_SLB))
253 mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
256 mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
259 printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
260 for (i = 0; i < mem_blocks; ++i) {
261 printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
262 " abs chunks %016lx - %016lx\n",
263 i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
264 mb_array[i].absStart, mb_array[i].absEnd);
269 static void __init iSeries_get_cmdline(void)
273 /* copy the command line parameter from the primary VSP */
274 HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
275 HvLpDma_Direction_RemoteToLocal);
280 if (!*p || *p == '\n')
287 static void __init iSeries_init_early(void)
289 DBG(" -> iSeries_init_early()\n");
291 /* Snapshot the timebase, for use in later recalibration */
292 iSeries_time_init_early();
295 * Initialize the DMA/TCE management
297 iommu_init_early_iSeries();
299 /* Initialize machine-dependency vectors */
304 /* Associate Lp Event Queue 0 with processor 0 */
305 HvCallEvent_setLpEventQueueInterruptProc(0, 0);
309 DBG(" <- iSeries_init_early()\n");
312 struct mschunks_map mschunks_map = {
313 /* XXX We don't use these, but Piranha might need them. */
314 .chunk_size = MSCHUNKS_CHUNK_SIZE,
315 .chunk_shift = MSCHUNKS_CHUNK_SHIFT,
316 .chunk_mask = MSCHUNKS_OFFSET_MASK,
318 EXPORT_SYMBOL(mschunks_map);
320 static void mschunks_alloc(unsigned long num_chunks)
322 klimit = _ALIGN(klimit, sizeof(u32));
323 mschunks_map.mapping = (u32 *)klimit;
324 klimit += num_chunks * sizeof(u32);
325 mschunks_map.num_chunks = num_chunks;
329 * The iSeries may have very large memories ( > 128 GB ) and a partition
330 * may get memory in "chunks" that may be anywhere in the 2**52 real
331 * address space. The chunks are 256K in size. To map this to the
332 * memory model Linux expects, the AS/400 specific code builds a
333 * translation table to translate what Linux thinks are "physical"
334 * addresses to the actual real addresses. This allows us to make
335 * it appear to Linux that we have contiguous memory starting at
336 * physical address zero while in fact this could be far from the truth.
337 * To avoid confusion, I'll let the words physical and/or real address
338 * apply to the Linux addresses while I'll use "absolute address" to
339 * refer to the actual hardware real address.
341 * build_iSeries_Memory_Map gets information from the Hypervisor and
342 * looks at the Main Store VPD to determine the absolute addresses
343 * of the memory that has been assigned to our partition and builds
344 * a table used to translate Linux's physical addresses to these
345 * absolute addresses. Absolute addresses are needed when
346 * communicating with the hypervisor (e.g. to build HPT entries)
348 * Returns the physical memory size
351 static unsigned long __init build_iSeries_Memory_Map(void)
353 u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
355 u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
356 u32 totalChunks,moreChunks;
357 u32 currChunk, thisChunk, absChunk;
361 struct MemoryBlock mb[32];
362 unsigned long numMemoryBlocks, curBlock;
364 /* Chunk size on iSeries is 256K bytes */
365 totalChunks = (u32)HvLpConfig_getMsChunks();
366 mschunks_alloc(totalChunks);
369 * Get absolute address of our load area
370 * and map it to physical address 0
371 * This guarantees that the loadarea ends up at physical 0
372 * otherwise, it might not be returned by PLIC as the first
376 loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
377 loadAreaSize = itLpNaca.xLoadAreaChunks;
380 * Only add the pages already mapped here.
381 * Otherwise we might add the hpt pages
382 * The rest of the pages of the load area
383 * aren't in the HPT yet and can still
384 * be assigned an arbitrary physical address
386 if ((loadAreaSize * 64) > HvPagesToMap)
387 loadAreaSize = HvPagesToMap / 64;
389 loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;
392 * TODO Do we need to do something if the HPT is in the 64MB load area?
393 * This would be required if the itLpNaca.xLoadAreaChunks includes
397 printk("Mapping load area - physical addr = 0000000000000000\n"
398 " absolute addr = %016lx\n",
399 chunk_to_addr(loadAreaFirstChunk));
400 printk("Load area size %dK\n", loadAreaSize * 256);
402 for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
403 mschunks_map.mapping[nextPhysChunk] =
404 loadAreaFirstChunk + nextPhysChunk;
407 * Get absolute address of our HPT and remember it so
408 * we won't map it to any physical address
410 hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
411 hptSizePages = (u32)HvCallHpt_getHptPages();
412 hptSizeChunks = hptSizePages >>
413 (MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
414 hptLastChunk = hptFirstChunk + hptSizeChunks - 1;
416 printk("HPT absolute addr = %016lx, size = %dK\n",
417 chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);
420 * Determine if absolute memory has any
421 * holes so that we can interpret the
422 * access map we get back from the hypervisor
425 numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);
428 * Process the main store access map from the hypervisor
429 * to build up our physical -> absolute translation table
434 moreChunks = totalChunks;
437 map = HvCallSm_get64BitsOfAccessMap(itLpNaca.xLpIndex,
439 thisChunk = currChunk;
441 chunkBit = map >> 63;
445 while (thisChunk >= mb[curBlock].logicalEnd) {
447 if (curBlock >= numMemoryBlocks)
448 panic("out of memory blocks");
450 if (thisChunk < mb[curBlock].logicalStart)
451 panic("memory block error");
453 absChunk = mb[curBlock].absStart +
454 (thisChunk - mb[curBlock].logicalStart);
455 if (((absChunk < hptFirstChunk) ||
456 (absChunk > hptLastChunk)) &&
457 ((absChunk < loadAreaFirstChunk) ||
458 (absChunk > loadAreaLastChunk))) {
459 mschunks_map.mapping[nextPhysChunk] =
471 * main store size (in chunks) is
472 * totalChunks - hptSizeChunks
473 * which should be equal to
476 return chunk_to_addr(nextPhysChunk);
482 static void __init iSeries_setup_arch(void)
484 if (get_lppaca()->shared_proc) {
485 ppc_md.idle_loop = iseries_shared_idle;
486 printk(KERN_DEBUG "Using shared processor idle loop\n");
488 ppc_md.idle_loop = iseries_dedicated_idle;
489 printk(KERN_DEBUG "Using dedicated idle loop\n");
492 /* Setup the Lp Event Queue */
493 setup_hvlpevent_queue();
495 printk("Max logical processors = %d\n",
496 itVpdAreas.xSlicMaxLogicalProcs);
497 printk("Max physical processors = %d\n",
498 itVpdAreas.xSlicMaxPhysicalProcs);
500 iSeries_pcibios_init();
503 static void iSeries_show_cpuinfo(struct seq_file *m)
505 seq_printf(m, "machine\t\t: 64-bit iSeries Logical Partition\n");
508 static void __init iSeries_progress(char * st, unsigned short code)
510 printk("Progress: [%04x] - %s\n", (unsigned)code, st);
511 mf_display_progress(code);
514 static void __init iSeries_fixup_klimit(void)
517 * Change klimit to take into account any ram disk
518 * that may be included
521 klimit = KERNELBASE + (u64)naca.xRamDisk +
522 (naca.xRamDiskSize * HW_PAGE_SIZE);
525 static int __init iSeries_src_init(void)
527 /* clear the progress line */
528 if (firmware_has_feature(FW_FEATURE_ISERIES))
529 ppc_md.progress(" ", 0xffff);
533 late_initcall(iSeries_src_init);
535 static inline void process_iSeries_events(void)
537 asm volatile ("li 0,0x5555; sc" : : : "r0", "r3");
540 static void yield_shared_processor(void)
544 HvCall_setEnabledInterrupts(HvCall_MaskIPI |
550 /* Compute future tb value when yield should expire */
551 HvCall_yieldProcessor(HvCall_YieldTimed, tb+tb_ticks_per_jiffy);
554 * The decrementer stops during the yield. Force a fake decrementer
555 * here and let the timer_interrupt code sort out the actual time.
557 get_lppaca()->int_dword.fields.decr_int = 1;
559 process_iSeries_events();
562 static void iseries_shared_idle(void)
565 tick_nohz_stop_sched_tick(1);
566 while (!need_resched() && !hvlpevent_is_pending()) {
568 ppc64_runlatch_off();
570 /* Recheck with irqs off */
571 if (!need_resched() && !hvlpevent_is_pending())
572 yield_shared_processor();
579 tick_nohz_restart_sched_tick();
581 if (hvlpevent_is_pending())
582 process_iSeries_events();
584 preempt_enable_no_resched();
590 static void iseries_dedicated_idle(void)
592 set_thread_flag(TIF_POLLING_NRFLAG);
595 tick_nohz_stop_sched_tick(1);
596 if (!need_resched()) {
597 while (!need_resched()) {
598 ppc64_runlatch_off();
601 if (hvlpevent_is_pending()) {
604 process_iSeries_events();
612 tick_nohz_restart_sched_tick();
613 preempt_enable_no_resched();
619 static void __iomem *iseries_ioremap(phys_addr_t address, unsigned long size,
620 unsigned long flags, void *caller)
622 return (void __iomem *)address;
625 static void iseries_iounmap(volatile void __iomem *token)
629 static int __init iseries_probe(void)
631 unsigned long root = of_get_flat_dt_root();
632 if (!of_flat_dt_is_compatible(root, "IBM,iSeries"))
636 /* iSeries does not support 16M pages */
637 cur_cpu_spec->cpu_features &= ~CPU_FTR_16M_PAGE;
643 static int iseries_kexec_prepare(struct kimage *image)
649 define_machine(iseries) {
651 .setup_arch = iSeries_setup_arch,
652 .show_cpuinfo = iSeries_show_cpuinfo,
653 .init_IRQ = iSeries_init_IRQ,
654 .get_irq = iSeries_get_irq,
655 .init_early = iSeries_init_early,
656 .pcibios_fixup = iSeries_pci_final_fixup,
657 .pcibios_fixup_resources= iSeries_pcibios_fixup_resources,
658 .restart = mf_reboot,
659 .power_off = mf_power_off,
660 .halt = mf_power_off,
661 .get_boot_time = iSeries_get_boot_time,
662 .set_rtc_time = iSeries_set_rtc_time,
663 .get_rtc_time = iSeries_get_rtc_time,
664 .calibrate_decr = generic_calibrate_decr,
665 .progress = iSeries_progress,
666 .probe = iseries_probe,
667 .ioremap = iseries_ioremap,
668 .iounmap = iseries_iounmap,
670 .machine_kexec_prepare = iseries_kexec_prepare,
672 /* XXX Implement enable_pmcs for iSeries */
675 void * __init iSeries_early_setup(void)
677 unsigned long phys_mem_size;
679 /* Identify CPU type. This is done again by the common code later
680 * on but calling this function multiple times is fine.
682 identify_cpu(0, mfspr(SPRN_PVR));
684 powerpc_firmware_features |= FW_FEATURE_ISERIES;
685 powerpc_firmware_features |= FW_FEATURE_LPAR;
687 iSeries_fixup_klimit();
690 * Initialize the table which translate Linux physical addresses to
691 * AS/400 absolute addresses
693 phys_mem_size = build_iSeries_Memory_Map();
695 iSeries_get_cmdline();
697 return (void *) __pa(build_flat_dt(phys_mem_size));
700 static void hvputc(char c)
705 HvCall_writeLogBuffer(&c, 1);
708 void __init udbg_init_iseries(void)