setup.c

linux-2.6.15.6

/*
 *    Copyright (c) 2000 Mike Corrigan <[e-mail]>
 *    Copyright (c) 1999-2000 Grant Erickson <[e-mail]>
 *
 *    Description:
 *      Architecture- / platform-specific boot-time initialization code for
 *      the IBM iSeries LPAR.  Adapted from original code by Grant Erickson and
 *      code by Gary Thomas, Cort Dougan <[e-mail]>, and Dan Malek
 *      <[e-mail]>.
 *
 *      This program is free software; you can redistribute it and/or
 *      modify it under the terms of the GNU General Public License
 *      as published by the Free Software Foundation; either version
 *      2 of the License, or (at your option) any later version.
 */

#undef DEBUG

#include <linux/config.h>
#include <linux/init.h>
#include <linux/threads.h>
#include <linux/smp.h>
#include <linux/param.h>
#include <linux/string.h>
#include <linux/initrd.h>
#include <linux/seq_file.h>
#include <linux/kdev_t.h>
#include <linux/major.h>
#include <linux/root_dev.h>
#include <linux/kernel.h>

#include <asm/processor.h>
#include <asm/machdep.h>
#include <asm/page.h>
#include <asm/mmu.h>
#include <asm/pgtable.h>
#include <asm/mmu_context.h>
#include <asm/cputable.h>
#include <asm/sections.h>
#include <asm/iommu.h>
#include <asm/firmware.h>
#include <asm/system.h>
#include <asm/time.h>
#include <asm/paca.h>
#include <asm/cache.h>
#include <asm/sections.h>
#include <asm/abs_addr.h>
#include <asm/iseries/hv_lp_config.h>
#include <asm/iseries/hv_call_event.h>
#include <asm/iseries/hv_call_xm.h>
#include <asm/iseries/it_lp_queue.h>
#include <asm/iseries/mf.h>
#include <asm/iseries/hv_lp_event.h>
#include <asm/iseries/lpar_map.h>

#include "naca.h"
#include "setup.h"
#include "irq.h"
#include "vpd_areas.h"
#include "processor_vpd.h"
#include "main_store.h"
#include "call_sm.h"
#include "call_hpt.h"

extern void hvlog(char *fmt, ...);

#ifdef DEBUG
#define DBG(fmt...) hvlog(fmt)
#else
#define DBG(fmt...)
#endif

/* Function Prototypes */
static unsigned long build_iSeries_Memory_Map(void);
static void iseries_shared_idle(void);
static void iseries_dedicated_idle(void);
#ifdef CONFIG_PCI
extern void iSeries_pci_final_fixup(void);
#else
static void iSeries_pci_final_fixup(void) { }
#endif

/* Global Variables */
int piranha_simulator;

extern int rd_size;		/* Defined in drivers/block/rd.c */
extern unsigned long embedded_sysmap_start;
extern unsigned long embedded_sysmap_end;

extern unsigned long iSeries_recal_tb;
extern unsigned long iSeries_recal_titan;

static int mf_initialized;

static unsigned long cmd_mem_limit;

struct MemoryBlock {
	unsigned long absStart;
	unsigned long absEnd;
	unsigned long logicalStart;
	unsigned long logicalEnd;
};

/*
 * Process the main store vpd to determine where the holes in memory are
 * and return the number of physical blocks and fill in the array of
 * block data.
 */
static unsigned long iSeries_process_Condor_mainstore_vpd(
		struct MemoryBlock *mb_array, unsigned long max_entries)
{
	unsigned long holeFirstChunk, holeSizeChunks;
	unsigned long numMemoryBlocks = 1;
	struct IoHriMainStoreSegment4 *msVpd =
		(struct IoHriMainStoreSegment4 *)xMsVpd;
	unsigned long holeStart = msVpd->nonInterleavedBlocksStartAdr;
	unsigned long holeEnd = msVpd->nonInterleavedBlocksEndAdr;
	unsigned long holeSize = holeEnd - holeStart;

	printk("Mainstore_VPD: Condor\n");
	/*
	 * Determine if absolute memory has any
	 * holes so that we can interpret the
	 * access map we get back from the hypervisor
	 * correctly.
	 */
	mb_array[0].logicalStart = 0;
	mb_array[0].logicalEnd = 0x100000000;
	mb_array[0].absStart = 0;
	mb_array[0].absEnd = 0x100000000;

	if (holeSize) {
		numMemoryBlocks = 2;
		holeStart = holeStart & 0x000fffffffffffff;
		holeStart = addr_to_chunk(holeStart);
		holeFirstChunk = holeStart;
		holeSize = addr_to_chunk(holeSize);
		holeSizeChunks = holeSize;
		printk( "Main store hole: start chunk = %0lx, size = %0lx chunks\n",
				holeFirstChunk, holeSizeChunks );
		mb_array[0].logicalEnd = holeFirstChunk;
		mb_array[0].absEnd = holeFirstChunk;
		mb_array[1].logicalStart = holeFirstChunk;
		mb_array[1].logicalEnd = 0x100000000 - holeSizeChunks;
		mb_array[1].absStart = holeFirstChunk + holeSizeChunks;
		mb_array[1].absEnd = 0x100000000;
	}
	return numMemoryBlocks;
}

#define MaxSegmentAreas			32
#define MaxSegmentAdrRangeBlocks	128
#define MaxAreaRangeBlocks		4

static unsigned long iSeries_process_Regatta_mainstore_vpd(
		struct MemoryBlock *mb_array, unsigned long max_entries)
{
	struct IoHriMainStoreSegment5 *msVpdP =
		(struct IoHriMainStoreSegment5 *)xMsVpd;
	unsigned long numSegmentBlocks = 0;
	u32 existsBits = msVpdP->msAreaExists;
	unsigned long area_num;

	printk("Mainstore_VPD: Regatta\n");

	for (area_num = 0; area_num < MaxSegmentAreas; ++area_num ) {
		unsigned long numAreaBlocks;
		struct IoHriMainStoreArea4 *currentArea;

		if (existsBits & 0x80000000) {
			unsigned long block_num;

			currentArea = &msVpdP->msAreaArray[area_num];
			numAreaBlocks = currentArea->numAdrRangeBlocks;
			printk("ms_vpd: processing area %2ld  blocks=%ld",
					area_num, numAreaBlocks);
			for (block_num = 0; block_num < numAreaBlocks;
					++block_num ) {
				/* Process an address range block */
				struct MemoryBlock tempBlock;
				unsigned long i;

				tempBlock.absStart =
					(unsigned long)currentArea->xAdrRangeBlock[block_num].blockStart;
				tempBlock.absEnd =
					(unsigned long)currentArea->xAdrRangeBlock[block_num].blockEnd;
				tempBlock.logicalStart = 0;
				tempBlock.logicalEnd   = 0;
				printk("\n          block %ld absStart=%016lx absEnd=%016lx",
						block_num, tempBlock.absStart,
						tempBlock.absEnd);

				for (i = 0; i < numSegmentBlocks; ++i) {
					if (mb_array[i].absStart ==
							tempBlock.absStart)
						break;
				}
				if (i == numSegmentBlocks) {
					if (numSegmentBlocks == max_entries)
						panic("iSeries_process_mainstore_vpd: too many memory blocks");
					mb_array[numSegmentBlocks] = tempBlock;
					++numSegmentBlocks;
				} else
					printk(" (duplicate)");
			}
			printk("\n");
		}
		existsBits <<= 1;
	}
	/* Now sort the blocks found into ascending sequence */
	if (numSegmentBlocks > 1) {
		unsigned long m, n;

		for (m = 0; m < numSegmentBlocks - 1; ++m) {
			for (n = numSegmentBlocks - 1; m < n; --n) {
				if (mb_array[n].absStart <
						mb_array[n-1].absStart) {
					struct MemoryBlock tempBlock;

					tempBlock = mb_array[n];
					mb_array[n] = mb_array[n-1];
					mb_array[n-1] = tempBlock;
				}
			}
		}
	}
	/*
	 * Assign "logical" addresses to each block.  These
	 * addresses correspond to the hypervisor "bitmap" space.
	 * Convert all addresses into units of 256K chunks.
	 */
	{
	unsigned long i, nextBitmapAddress;

	printk("ms_vpd: %ld sorted memory blocks\n", numSegmentBlocks);
	nextBitmapAddress = 0;
	for (i = 0; i < numSegmentBlocks; ++i) {
		unsigned long length = mb_array[i].absEnd -
			mb_array[i].absStart;

		mb_array[i].logicalStart = nextBitmapAddress;
		mb_array[i].logicalEnd = nextBitmapAddress + length;
		nextBitmapAddress += length;
		printk("          Bitmap range: %016lx - %016lx\n"
				"        Absolute range: %016lx - %016lx\n",
				mb_array[i].logicalStart,
				mb_array[i].logicalEnd,
				mb_array[i].absStart, mb_array[i].absEnd);
		mb_array[i].absStart = addr_to_chunk(mb_array[i].absStart &
				0x000fffffffffffff);
		mb_array[i].absEnd = addr_to_chunk(mb_array[i].absEnd &
				0x000fffffffffffff);
		mb_array[i].logicalStart =
			addr_to_chunk(mb_array[i].logicalStart);
		mb_array[i].logicalEnd = addr_to_chunk(mb_array[i].logicalEnd);
	}
	}

	return numSegmentBlocks;
}

static unsigned long iSeries_process_mainstore_vpd(struct MemoryBlock *mb_array,
		unsigned long max_entries)
{
	unsigned long i;
	unsigned long mem_blocks = 0;

	if (cpu_has_feature(CPU_FTR_SLB))
		mem_blocks = iSeries_process_Regatta_mainstore_vpd(mb_array,
				max_entries);
	else
		mem_blocks = iSeries_process_Condor_mainstore_vpd(mb_array,
				max_entries);

	printk("Mainstore_VPD: numMemoryBlocks = %ld \n", mem_blocks);
	for (i = 0; i < mem_blocks; ++i) {
		printk("Mainstore_VPD: block %3ld logical chunks %016lx - %016lx\n"
		       "                             abs chunks %016lx - %016lx\n",
			i, mb_array[i].logicalStart, mb_array[i].logicalEnd,
			mb_array[i].absStart, mb_array[i].absEnd);
	}
	return mem_blocks;
}

static void __init iSeries_get_cmdline(void)
{
	char *p, *q;

	/* copy the command line parameter from the primary VSP  */
	HvCallEvent_dmaToSp(cmd_line, 2 * 64* 1024, 256,
			HvLpDma_Direction_RemoteToLocal);

	p = cmd_line;
	q = cmd_line + 255;
	while(p < q) {
		if (!*p || *p == '\n')
			break;
		++p;
	}
	*p = 0;
}

static void __init iSeries_init_early(void)
{
	DBG(" -> iSeries_init_early()\n");

	ppc64_firmware_features = FW_FEATURE_ISERIES;

	ppc64_interrupt_controller = IC_ISERIES;

#if defined(CONFIG_BLK_DEV_INITRD)
	/*
	 * If the init RAM disk has been configured and there is
	 * a non-zero starting address for it, set it up
	 */
	if (naca.xRamDisk) {
		initrd_start = (unsigned long)__va(naca.xRamDisk);
		initrd_end = initrd_start + naca.xRamDiskSize * HW_PAGE_SIZE;
		initrd_below_start_ok = 1;	// ramdisk in kernel space
		ROOT_DEV = Root_RAM0;
		if (((rd_size * 1024) / HW_PAGE_SIZE) < naca.xRamDiskSize)
			rd_size = (naca.xRamDiskSize * HW_PAGE_SIZE) / 1024;
	} else
#endif /* CONFIG_BLK_DEV_INITRD */
	{
	    /* ROOT_DEV = MKDEV(VIODASD_MAJOR, 1); */
	}

	iSeries_recal_tb = get_tb();
	iSeries_recal_titan = HvCallXm_loadTod();

	/*
	 * Initialize the hash table management pointers
	 */
	hpte_init_iSeries();

	/*
	 * Initialize the DMA/TCE management
	 */
	iommu_init_early_iSeries();

	/* Initialize machine-dependency vectors */
#ifdef CONFIG_SMP
	smp_init_iSeries();
#endif
	if (itLpNaca.xPirEnvironMode == 0)
		piranha_simulator = 1;

	/* Associate Lp Event Queue 0 with processor 0 */
	HvCallEvent_setLpEventQueueInterruptProc(0, 0);

	mf_init();
	mf_initialized = 1;
	mb();

	/* If we were passed an initrd, set the ROOT_DEV properly if the values
	 * look sensible. If not, clear initrd reference.
	 */
#ifdef CONFIG_BLK_DEV_INITRD
	if (initrd_start >= KERNELBASE && initrd_end >= KERNELBASE &&
	    initrd_end > initrd_start)
		ROOT_DEV = Root_RAM0;
	else
		initrd_start = initrd_end = 0;
#endif /* CONFIG_BLK_DEV_INITRD */

	DBG(" <- iSeries_init_early()\n");
}

struct mschunks_map mschunks_map = {
	/* XXX We don't use these, but Piranha might need them. */
	.chunk_size  = MSCHUNKS_CHUNK_SIZE,
	.chunk_shift = MSCHUNKS_CHUNK_SHIFT,
	.chunk_mask  = MSCHUNKS_OFFSET_MASK,
};
EXPORT_SYMBOL(mschunks_map);

void mschunks_alloc(unsigned long num_chunks)
{
	klimit = _ALIGN(klimit, sizeof(u32));
	mschunks_map.mapping = (u32 *)klimit;
	klimit += num_chunks * sizeof(u32);
	mschunks_map.num_chunks = num_chunks;
}

/*
 * The iSeries may have very large memories ( > 128 GB ) and a partition
 * may get memory in "chunks" that may be anywhere in the 2**52 real
 * address space.  The chunks are 256K in size.  To map this to the
 * memory model Linux expects, the AS/400 specific code builds a
 * translation table to translate what Linux thinks are "physical"
 * addresses to the actual real addresses.  This allows us to make
 * it appear to Linux that we have contiguous memory starting at
 * physical address zero while in fact this could be far from the truth.
 * To avoid confusion, I'll let the words physical and/or real address
 * apply to the Linux addresses while I'll use "absolute address" to
 * refer to the actual hardware real address.
 *
 * build_iSeries_Memory_Map gets information from the Hypervisor and
 * looks at the Main Store VPD to determine the absolute addresses
 * of the memory that has been assigned to our partition and builds
 * a table used to translate Linux's physical addresses to these
 * absolute addresses.  Absolute addresses are needed when
 * communicating with the hypervisor (e.g. to build HPT entries)
 *
 * Returns the physical memory size
 */

static unsigned long __init build_iSeries_Memory_Map(void)
{
	u32 loadAreaFirstChunk, loadAreaLastChunk, loadAreaSize;
	u32 nextPhysChunk;
	u32 hptFirstChunk, hptLastChunk, hptSizeChunks, hptSizePages;
	u32 totalChunks,moreChunks;
	u32 currChunk, thisChunk, absChunk;
	u32 currDword;
	u32 chunkBit;
	u64 map;
	struct MemoryBlock mb[32];
	unsigned long numMemoryBlocks, curBlock;

	/* Chunk size on iSeries is 256K bytes */
	totalChunks = (u32)HvLpConfig_getMsChunks();
	mschunks_alloc(totalChunks);

	/*
	 * Get absolute address of our load area
	 * and map it to physical address 0
	 * This guarantees that the loadarea ends up at physical 0
	 * otherwise, it might not be returned by PLIC as the first
	 * chunks
	 */

	loadAreaFirstChunk = (u32)addr_to_chunk(itLpNaca.xLoadAreaAddr);
	loadAreaSize =  itLpNaca.xLoadAreaChunks;

	/*
	 * Only add the pages already mapped here.
	 * Otherwise we might add the hpt pages
	 * The rest of the pages of the load area
	 * aren't in the HPT yet and can still
	 * be assigned an arbitrary physical address
	 */
	if ((loadAreaSize * 64) > HvPagesToMap)
		loadAreaSize = HvPagesToMap / 64;

	loadAreaLastChunk = loadAreaFirstChunk + loadAreaSize - 1;

	/*
	 * TODO Do we need to do something if the HPT is in the 64MB load area?
	 * This would be required if the itLpNaca.xLoadAreaChunks includes
	 * the HPT size
	 */

	printk("Mapping load area - physical addr = 0000000000000000\n"
		"                    absolute addr = %016lx\n",
		chunk_to_addr(loadAreaFirstChunk));
	printk("Load area size %dK\n", loadAreaSize * 256);

	for (nextPhysChunk = 0; nextPhysChunk < loadAreaSize; ++nextPhysChunk)
		mschunks_map.mapping[nextPhysChunk] =
			loadAreaFirstChunk + nextPhysChunk;

	/*
	 * Get absolute address of our HPT and remember it so
	 * we won't map it to any physical address
	 */
	hptFirstChunk = (u32)addr_to_chunk(HvCallHpt_getHptAddress());
	hptSizePages = (u32)HvCallHpt_getHptPages();
	hptSizeChunks = hptSizePages >>
		(MSCHUNKS_CHUNK_SHIFT - HW_PAGE_SHIFT);
	hptLastChunk = hptFirstChunk + hptSizeChunks - 1;

	printk("HPT absolute addr = %016lx, size = %dK\n",
			chunk_to_addr(hptFirstChunk), hptSizeChunks * 256);

	ppc64_pft_size = __ilog2(hptSizePages * HW_PAGE_SIZE);

	/*
	 * The actual hashed page table is in the hypervisor,
	 * we have no direct access
	 */
	htab_address = NULL;

	/*
	 * Determine if absolute memory has any
	 * holes so that we can interpret the
	 * access map we get back from the hypervisor
	 * correctly.
	 */
	numMemoryBlocks = iSeries_process_mainstore_vpd(mb, 32);

	/*
	 * Process the main store access map from the hypervisor
	 * to build up our physical -> absolute translation table
	 */
	curBlock = 0;
	currChunk = 0;
	currDword = 0;