Initial rudimentary support for mapped kernel. The kernel text and data

are placed in the cksseg area, and such behavior is tuned thru a make
option (off by default). Basis for future kernel text replication and
partitioning work/study.

1 files changed, 368 insertions, 87 deletions

diff --git a/arch/mips64/kernel/head.S b/arch/mips64/kernel/head.S
index 0615209f9..b2c006886 100644
--- a/arch/mips64/kernel/head.S
+++ b/arch/mips64/kernel/head.S
@@ -1,107 +1,388 @@
-/* $Id: head.S,v 1.6 2000/03/27 21:05:04 ulfc Exp $
+/* $Id: ip27-memory.c,v 1.2 2000/01/27 01:05:24 ralf Exp $
  *
  * This file is subject to the terms and conditions of the GNU General Public
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Head.S contains the MIPS exception handler and startup code.
+ * Copyright (C) 2000 by Ralf Baechle
+ * Copyright (C) 2000 by Silicon Graphics, Inc.
  *
- * Copyright (C) 1994, 1995 Waldorf Electronics
- * Written by Ralf Baechle and Andreas Busse
- * Copyright (C) 1994, 1995, 1996, 1997, 1998, 1999 Ralf Baechle
- * Copyright (C) 1999 Silicon Graphics, Inc.
+ * On SGI IP27 the ARC memory configuration data is completly bogus but
+ * alternate easier to use mechanisms are available.
  */
-#define __ASSEMBLY__
-#include <linux/config.h>
 #include <linux/init.h>
-#include <asm/asm.h>
-#include <asm/regdef.h>
-#include <asm/processor.h>
-#include <asm/mipsregs.h>
-#include <asm/stackframe.h>
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/bootmem.h>
+#include <linux/swap.h>
+
+#include <asm/page.h>
+#include <asm/bootinfo.h>
+#include <asm/addrspace.h>
+#include <asm/pgtable.h>
+#include <asm/sn/types.h>
+#include <asm/sn/addrs.h>
+#include <asm/sn/klconfig.h>
+#include <asm/sn/arch.h>
+#include <asm/mmzone.h>
+
+typedef unsigned long pfn_t;			/* into <asm/sn/types.h> */
+#define KDM_TO_PHYS(x)	((x) & TO_PHYS_MASK)	/* into asm/addrspace.h */
+
+extern char _end;
+
+#define PFN_UP(x)	(((x) + PAGE_SIZE-1) >> PAGE_SHIFT)
+#define SLOT_IGNORED	0xffff
 
-	.text
+short slot_lastfilled_cache[MAX_COMPACT_NODES];
+unsigned short slot_psize_cache[MAX_COMPACT_NODES][MAX_MEM_SLOTS];
+static pfn_t numpages;
+static pfn_t pagenr = 0;
 
-EXPORT(stext)					# used for profiling
-EXPORT(_stext)
+plat_pg_data_t *plat_node_data[MAX_COMPACT_NODES];
+bootmem_data_t plat_node_bdata[MAX_COMPACT_NODES];
 
-	__INIT
+int numa_debug(void)
+{
+	printk("NUMA debug\n");
+	*(int *)0 = 0;
+	return(0);
+}
 
-NESTED(kernel_entry, 16, sp)			# kernel entry point
+/*
+ * Return pfn of first free page of memory on a node. PROM may allocate
+ * data structures on the first couple of pages of the first slot of each 
+ * node. If this is the case, getfirstfree(node) > getslotstart(node, 0).
+ */
+pfn_t node_getfirstfree(cnodeid_t cnode)
+{
+	unsigned long loadbase = CKSEG0;
+	nasid_t nasid = COMPACT_TO_NASID_NODEID(cnode);
 
-#ifdef CONFIG_ARC64
-	la	t0, 1f
-	jr	t0
-1:
+#ifdef CONFIG_MAPPED_KERNEL
+	loadbase = CKSSEG;
 #endif
+	if (cnode == 0)
+		return (KDM_TO_PHYS(PAGE_ALIGN((unsigned long)(&_end)) - 
+					(loadbase - K0BASE)) >> PAGE_SHIFT);
+	return (KDM_TO_PHYS(PAGE_ALIGN(SYMMON_STK_ADDR(nasid, 0))) >> PAGE_SHIFT);
+}
 
-	ori	sp, 0xf				# align stack on 16 byte.
-	xori	sp, 0xf
+/*
+ * Return the number of pages of memory provided by the given slot
+ * on the specified node.
+ */
+pfn_t slot_getsize(cnodeid_t node, int slot)
+{
+	return (pfn_t) slot_psize_cache[node][slot];
+}
 
-	/* Note that all firmware passed argument registers still
-	   have their values.  */
-	jal	prom_init			# initialize firmware
+/*
+ * Return highest slot filled
+ */
+int node_getlastslot(cnodeid_t node)
+{
+	return (int) slot_lastfilled_cache[node];
+}
 
-	CLI					# disable interrupts
+/*
+ * Return the pfn of the last free page of memory on a node.
+ */
+pfn_t node_getmaxclick(cnodeid_t node)
+{
+	pfn_t	slot_psize;
+	int	slot;
 
-	mfc0	t0, CP0_STATUS
 	/*
-	 * On IP27, I am seeing the TS bit set when the
-	 * kernel is loaded. Maybe because the kernel is
-	 * in ckseg0 and not xkphys? Clear it anyway ...
+	 * Start at the top slot. When we find a slot with memory in it,
+	 * that's the winner.
 	 */
-	li	t1, ~(ST0_TS|ST0_CU1|ST0_CU2|ST0_CU3)
-	and	t0, t1
-	or	t0, (ST0_CU0|ST0_KX|ST0_SX|ST0_FR) # Bogosity: cu0 indicates kernel
-	mtc0	t0, CP0_STATUS			   # thread in copy_thread.
-
-	la	$28, init_task_union		# init current pointer
-	daddiu	t0, $28, KERNEL_STACK_SIZE-32
-	sd	t0, kernelsp
-	dsubu	sp, t0, 4*SZREG			# init stack pointer
-	move	t0, $28
-#ifdef CONFIG_SMP
-	mtc0	t0, CP0_WATCHLO
-	dsrl32	t0, t0, 0
-	mtc0	t0, CP0_WATCHHI
-#endif
-	jal	start_kernel
-1:	b	1b				# just in case ...
-	END(kernel_entry)
-
-NESTED(bootstrap, 16, sp)
-#ifdef CONFIG_ARC64
-	la	t0, 1f
-	jr	t0
-1:
-#endif
-	CLI
-	mfc0	t0, CP0_STATUS
-	li	t1, ~(ST0_CU1|ST0_CU2|ST0_CU3)
-	and	t0, t1
-	or	t0, (ST0_CU0|ST0_KX|ST0_SX|ST0_FR) # Bogosity: cu0 indicates kernel
-	mtc0	t0, CP0_STATUS			   # thread in copy_thread.
-	jal	cboot
-	END(bootstrap)
-
-	__FINIT
-
-	.comm	kernelsp, 8, 8			# current stackpointer
-
-#define PAGE_SIZE	0x1000
-
-	.macro	page name, order=0
-	.globl	\name
-	.org	. + (PAGE_SIZE << \order)
-\name:	.size	\name, (PAGE_SIZE << \order)
-	.type	\name, @object
-	.endm
-
-	.align	12
-	.data
-
-	page	swapper_pg_dir, 1
-	page	invalid_pte_table, 1
-	page	invalid_pmd_table, 1
-	page	empty_bad_page_table, 1
-	page	empty_bad_page
+	for (slot = (node_getnumslots(node) - 1); slot >= 0; slot--) {
+		if ((slot_psize = slot_getsize(node, slot))) {
+			if (slot_psize == SLOT_IGNORED)
+				continue;
+			/* Return the basepfn + the slot size, minus 1. */
+			return slot_getbasepfn(node, slot) + slot_psize - 1;
+		}
+	}
+
+	/*
+	 * If there's no memory on the node, return 0. This is likely
+	 * to cause problems.
+	 */
+	return (pfn_t)0;
+}
+
+static pfn_t slot_psize_compute(cnodeid_t node, int slot)
+{
+	nasid_t nasid;
+	lboard_t *brd;
+	klmembnk_t *banks;
+	unsigned long size;
+
+	nasid = COMPACT_TO_NASID_NODEID(node);
+	/* Find the node board */
+	brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27);
+	if (!brd)
+		return 0;
+
+	/* Get the memory bank structure */
+	banks = (klmembnk_t *)find_first_component(brd, KLSTRUCT_MEMBNK);
+	if (!banks)
+		return 0;
+
+	/* Size in _Megabytes_ */
+	size = (unsigned long)banks->membnk_bnksz[slot/4];
+
+	/* hack for 128 dimm banks */
+	if (size <= 128) {
+		if (slot%4 == 0) {
+			size <<= 20;		/* size in bytes */
+			return(size >> PAGE_SHIFT);
+		} else {
+			return 0;
+		}
+	} else {
+		size /= 4;
+		size <<= 20;
+		return(size >> PAGE_SHIFT);
+	}
+}
+
+pfn_t szmem(pfn_t fpage, pfn_t maxpmem)
+{
+	cnodeid_t node;
+	int slot, numslots;
+	pfn_t num_pages = 0, slot_psize;
+	pfn_t slot0sz = 0, nodebytes;	/* Hack to detect problem configs */
+	int ignore;
+
+	for (node = 0; node < numnodes; node++) {
+		numslots = node_getnumslots(node);
+		ignore = nodebytes = 0;
+		for (slot = 0; slot < numslots; slot++) {
+			slot_psize = slot_psize_compute(node, slot);
+			if (slot == 0) slot0sz = slot_psize;
+			/* 
+			 * We need to refine the hack when we have replicated
+			 * kernel text.
+			 */
+			nodebytes += SLOT_SIZE;
+			if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) >
+						(slot0sz << PAGE_SHIFT))
+				ignore = 1;
+			if (ignore && slot_psize) {
+				printk("Ignoring slot %d onwards on node %d\n",
+								slot, node);
+				slot_psize_cache[node][slot] = SLOT_IGNORED;
+				slot = numslots;
+				continue;
+			}
+			num_pages += slot_psize;
+			slot_psize_cache[node][slot] = 
+					(unsigned short) slot_psize;
+			if (slot_psize)
+				slot_lastfilled_cache[node] = slot;
+		}
+	}
+	if (maxpmem)
+		return((maxpmem > num_pages) ? num_pages : maxpmem);
+	else
+		return num_pages;
+}
+
+/*
+ * HACK ALERT - Things do not work if this is not here. Maybe this is
+ * acting as a pseudocacheflush operation. The write pattern seems to
+ * be important, writing a 0 does not help.
+ */
+void setup_test(cnodeid_t node, pfn_t start, pfn_t end)
+{
+	unsigned long *ptr = __va(start << PAGE_SHIFT);
+	unsigned long size = 4 * 1024 * 1024;	/* 4M L2 caches */
+
+	while (size) {
+		size -= sizeof(unsigned long);
+		*ptr = (0xdeadbeefbabeb000UL|node);
+		/* *ptr = 0; */
+		ptr++;
+	}
+}
+
+/*
+ * Currently, the intranode memory hole support assumes that each slot
+ * contains at least 32 MBytes of memory. We assume all bootmem data
+ * fits on the first slot.
+ */
+void __init prom_meminit(void)
+{
+	extern void mlreset(void);
+	cnodeid_t node;
+	pfn_t slot_firstpfn, slot_lastpfn, slot_freepfn;
+	unsigned long bootmap_size;
+	int node_datasz;
+
+	node_datasz = PFN_UP(sizeof(plat_pg_data_t));
+	mlreset();
+	numpages = szmem(0, 0);
+	for (node = (numnodes - 1); node >= 0; node--) {
+		slot_firstpfn = slot_getbasepfn(node, 0);
+		slot_lastpfn = slot_firstpfn + slot_getsize(node, 0);
+		slot_freepfn = node_getfirstfree(node);
+		/* Foll line hack for non discontigmem; remove once discontigmem
+		 * becomes the default. */
+		max_low_pfn = (slot_lastpfn - slot_firstpfn);
+		if (node != 0) 
+			setup_test(node, slot_freepfn, slot_lastpfn);
+		/*
+		 * Allocate the node data structure on the node first.
+		 */
+		plat_node_data[node] = (plat_pg_data_t *)(__va(slot_freepfn \
+							<< PAGE_SHIFT));
+		NODE_DATA(node)->bdata = plat_node_bdata + node;
+		slot_freepfn += node_datasz;
+	  	bootmap_size = init_bootmem_node(node, slot_freepfn, 
+						slot_firstpfn, slot_lastpfn);
+		free_bootmem_node(node, slot_firstpfn << PAGE_SHIFT, 
+				(slot_lastpfn - slot_firstpfn) << PAGE_SHIFT);
+		reserve_bootmem_node(node, slot_firstpfn << PAGE_SHIFT,
+		  ((slot_freepfn - slot_firstpfn) << PAGE_SHIFT) + bootmap_size);
+	}
+	printk("Total memory probed : 0x%lx pages\n", numpages);
+}
+
+int __init page_is_ram(unsigned long pagenr)
+{
+        return 1;
+}
+
+void __init
+prom_free_prom_memory (void)
+{
+	/* We got nothing to free here ...  */
+}
+
+#ifdef CONFIG_DISCONTIGMEM
+
+void __init paging_init(void)
+{
+	cnodeid_t node;
+	unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
+
+	/* Initialize the entire pgd.  */
+	pgd_init((unsigned long)swapper_pg_dir);
+	pmd_init((unsigned long)invalid_pmd_table);
+	memset((void *)invalid_pte_table, 0, sizeof(pte_t) * 2 * PTRS_PER_PTE);
+
+	for (node = 0; node < numnodes; node++) {
+		pfn_t start_pfn = slot_getbasepfn(node, 0);
+		pfn_t end_pfn = node_getmaxclick(node);
+
+		zones_size[ZONE_DMA] = end_pfn + 1 - start_pfn;
+		free_area_init_node(node, NODE_DATA(node), zones_size, 
+						start_pfn << PAGE_SHIFT, 0);
+		if ((PLAT_NODE_DATA_STARTNR(node) + 
+					PLAT_NODE_DATA_SIZE(node)) > pagenr)
+			pagenr = PLAT_NODE_DATA_STARTNR(node) +
+					PLAT_NODE_DATA_SIZE(node);
+	}
+}
+
+void __init mem_init(void)
+{
+	extern char _stext, _etext, _fdata, _edata;
+	extern char __init_begin, __init_end;
+	extern unsigned long totalram_pages;
+	extern unsigned long setup_zero_pages(void);
+	cnodeid_t nid;
+	unsigned long tmp, ram;
+	unsigned long codesize, reservedpages, datasize, initsize;
+	int slot, numslots;
+	struct page *pg, *pslot;
+	pfn_t pgnr;
+
+	num_physpages = numpages;	/* memory already sized by szmem */
+	max_mapnr = pagenr;		/* already found during paging_init */
+	high_memory = (void *) __va(max_mapnr << PAGE_SHIFT);
+
+	for (nid = 0; nid < numnodes; nid++) {
+
+		/*
+		 * Hack till free_area_init_core() zeroes free_pages
+		 */
+		for (tmp = 0; tmp < MAX_NR_ZONES; tmp++)
+			PLAT_NODE_DATA(nid)->gendata.node_zones[tmp].free_pages=0;
+		/*
+	 	 * This will free up the bootmem, ie, slot 0 memory.
+	 	 */
+		totalram_pages += free_all_bootmem_node(nid);
+
+		/*
+		 * We need to manually do the other slots.
+		 */
+		pg = NODE_DATA(nid)->node_mem_map + slot_getsize(nid, 0);
+		pgnr = PLAT_NODE_DATA_STARTNR(nid) + slot_getsize(nid, 0);
+		numslots = node_getlastslot(nid);
+		for (slot = 1; slot <= numslots; slot++) {
+			pslot = NODE_DATA(nid)->node_mem_map + 
+			   slot_getbasepfn(nid, slot) - slot_getbasepfn(nid, 0);
+
+			/*
+			 * Mark holes in previous slot. May also want to
+			 * free up the pages that hold the memmap entries.
+			 */
+			while (pg < pslot) {
+				pg->flags |= (1<<PG_skip);
+				pg++; pgnr++;
+			}
+
+			/*
+			 * Free valid memory in current slot.
+			 */
+			pslot += slot_getsize(nid, slot);
+			while (pg < pslot) {
+				if (!page_is_ram(pgnr))
+					continue;
+				ClearPageReserved(pg);
+				atomic_set(&pg->count, 1);
+				__free_page(pg);
+				totalram_pages++;
+				pg++; pgnr++;
+			}
+		}
+	}
+
+	totalram_pages -= setup_zero_pages();	/* This comes from node 0 */
+
+	reservedpages = ram = 0;
+	for (nid = 0; nid < numnodes; nid++) {
+		for (tmp = PLAT_NODE_DATA_STARTNR(nid); tmp < 
+			(PLAT_NODE_DATA_STARTNR(nid) +
+			PLAT_NODE_DATA_SIZE(nid)); tmp++) {
+			/* Ignore holes */
+			if (PageSkip(mem_map+tmp))
+				continue;
+			if (page_is_ram(tmp)) {
+				ram++;
+				if (PageReserved(mem_map+tmp))
+					reservedpages++;
+			}
+		}
+	}
+
+	codesize =  (unsigned long) &_etext - (unsigned long) &_stext;
+	datasize =  (unsigned long) &_edata - (unsigned long) &_fdata;
+	initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
+
+	printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, "
+		"%ldk data, %ldk init)\n",
+		(unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
+		ram << (PAGE_SHIFT-10),
+		codesize >> 10,
+		reservedpages << (PAGE_SHIFT-10),
+		datasize >> 10,
+		initsize >> 10);
+}
+
+#endif /* CONFIG_DISCONTIGMEM */