1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License.  See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2000, 05 by Ralf Baechle (ralf@linux-mips.org)
7 * Copyright (C) 2000 by Silicon Graphics, Inc.
8 * Copyright (C) 2004 by Christoph Hellwig
9 *
10 * On SGI IP27 the ARC memory configuration data is completly bogus but
11 * alternate easier to use mechanisms are available.
12 */
13#include <linux/init.h>
14#include <linux/kernel.h>
15#include <linux/mm.h>
16#include <linux/mmzone.h>
17#include <linux/module.h>
18#include <linux/nodemask.h>
19#include <linux/swap.h>
20#include <linux/bootmem.h>
21#include <linux/pfn.h>
22#include <linux/highmem.h>
23#include <asm/page.h>
24#include <asm/pgalloc.h>
25#include <asm/sections.h>
26
27#include <asm/sn/arch.h>
28#include <asm/sn/hub.h>
29#include <asm/sn/klconfig.h>
30#include <asm/sn/sn_private.h>
31
32
33#define SLOT_PFNSHIFT           (SLOT_SHIFT - PAGE_SHIFT)
34#define PFN_NASIDSHFT           (NASID_SHFT - PAGE_SHIFT)
35
36#define SLOT_IGNORED		0xffff
37
38static short __initdata slot_lastfilled_cache[MAX_COMPACT_NODES];
39static unsigned short __initdata slot_psize_cache[MAX_COMPACT_NODES][MAX_MEM_SLOTS];
40static struct bootmem_data __initdata plat_node_bdata[MAX_COMPACT_NODES];
41
42struct node_data *__node_data[MAX_COMPACT_NODES];
43
44EXPORT_SYMBOL(__node_data);
45
46static int fine_mode;
47
48static int is_fine_dirmode(void)
49{
50	return (((LOCAL_HUB_L(NI_STATUS_REV_ID) & NSRI_REGIONSIZE_MASK)
51	        >> NSRI_REGIONSIZE_SHFT) & REGIONSIZE_FINE);
52}
53
54static hubreg_t get_region(cnodeid_t cnode)
55{
56	if (fine_mode)
57		return COMPACT_TO_NASID_NODEID(cnode) >> NASID_TO_FINEREG_SHFT;
58	else
59		return COMPACT_TO_NASID_NODEID(cnode) >> NASID_TO_COARSEREG_SHFT;
60}
61
62static hubreg_t region_mask;
63
64static void gen_region_mask(hubreg_t *region_mask)
65{
66	cnodeid_t cnode;
67
68	(*region_mask) = 0;
69	for_each_online_node(cnode) {
70		(*region_mask) |= 1ULL << get_region(cnode);
71	}
72}
73
74#define	rou_rflag	rou_flags
75
76static int router_distance;
77
78static void router_recurse(klrou_t *router_a, klrou_t *router_b, int depth)
79{
80	klrou_t *router;
81	lboard_t *brd;
82	int	port;
83
84	if (router_a->rou_rflag == 1)
85		return;
86
87	if (depth >= router_distance)
88		return;
89
90	router_a->rou_rflag = 1;
91
92	for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
93		if (router_a->rou_port[port].port_nasid == INVALID_NASID)
94			continue;
95
96		brd = (lboard_t *)NODE_OFFSET_TO_K0(
97			router_a->rou_port[port].port_nasid,
98			router_a->rou_port[port].port_offset);
99
100		if (brd->brd_type == KLTYPE_ROUTER) {
101			router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
102			if (router == router_b) {
103				if (depth < router_distance)
104					router_distance = depth;
105			}
106			else
107				router_recurse(router, router_b, depth + 1);
108		}
109	}
110
111	router_a->rou_rflag = 0;
112}
113
114unsigned char __node_distances[MAX_COMPACT_NODES][MAX_COMPACT_NODES];
115
116static int __init compute_node_distance(nasid_t nasid_a, nasid_t nasid_b)
117{
118	klrou_t *router, *router_a = NULL, *router_b = NULL;
119	lboard_t *brd, *dest_brd;
120	cnodeid_t cnode;
121	nasid_t nasid;
122	int port;
123
124	/* Figure out which routers nodes in question are connected to */
125	for_each_online_node(cnode) {
126		nasid = COMPACT_TO_NASID_NODEID(cnode);
127
128		if (nasid == -1) continue;
129
130		brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
131					KLTYPE_ROUTER);
132
133		if (!brd)
134			continue;
135
136		do {
137			if (brd->brd_flags & DUPLICATE_BOARD)
138				continue;
139
140			router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
141			router->rou_rflag = 0;
142
143			for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
144				if (router->rou_port[port].port_nasid == INVALID_NASID)
145					continue;
146
147				dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
148					router->rou_port[port].port_nasid,
149					router->rou_port[port].port_offset);
150
151				if (dest_brd->brd_type == KLTYPE_IP27) {
152					if (dest_brd->brd_nasid == nasid_a)
153						router_a = router;
154					if (dest_brd->brd_nasid == nasid_b)
155						router_b = router;
156				}
157			}
158
159		} while ((brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)));
160	}
161
162	if (router_a == NULL) {
163		printk("node_distance: router_a NULL\n");
164		return -1;
165	}
166	if (router_b == NULL) {
167		printk("node_distance: router_b NULL\n");
168		return -1;
169	}
170
171	if (nasid_a == nasid_b)
172		return 0;
173
174	if (router_a == router_b)
175		return 1;
176
177	router_distance = 100;
178	router_recurse(router_a, router_b, 2);
179
180	return router_distance;
181}
182
183static void __init init_topology_matrix(void)
184{
185	nasid_t nasid, nasid2;
186	cnodeid_t row, col;
187
188	for (row = 0; row < MAX_COMPACT_NODES; row++)
189		for (col = 0; col < MAX_COMPACT_NODES; col++)
190			__node_distances[row][col] = -1;
191
192	for_each_online_node(row) {
193		nasid = COMPACT_TO_NASID_NODEID(row);
194		for_each_online_node(col) {
195			nasid2 = COMPACT_TO_NASID_NODEID(col);
196			__node_distances[row][col] =
197				compute_node_distance(nasid, nasid2);
198		}
199	}
200}
201
202static void __init dump_topology(void)
203{
204	nasid_t nasid;
205	cnodeid_t cnode;
206	lboard_t *brd, *dest_brd;
207	int port;
208	int router_num = 0;
209	klrou_t *router;
210	cnodeid_t row, col;
211
212	printk("************** Topology ********************\n");
213
214	printk("    ");
215	for_each_online_node(col)
216		printk("%02d ", col);
217	printk("\n");
218	for_each_online_node(row) {
219		printk("%02d  ", row);
220		for_each_online_node(col)
221			printk("%2d ", node_distance(row, col));
222		printk("\n");
223	}
224
225	for_each_online_node(cnode) {
226		nasid = COMPACT_TO_NASID_NODEID(cnode);
227
228		if (nasid == -1) continue;
229
230		brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
231					KLTYPE_ROUTER);
232
233		if (!brd)
234			continue;
235
236		do {
237			if (brd->brd_flags & DUPLICATE_BOARD)
238				continue;
239			printk("Router %d:", router_num);
240			router_num++;
241
242			router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
243
244			for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
245				if (router->rou_port[port].port_nasid == INVALID_NASID)
246					continue;
247
248				dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
249					router->rou_port[port].port_nasid,
250					router->rou_port[port].port_offset);
251
252				if (dest_brd->brd_type == KLTYPE_IP27)
253					printk(" %d", dest_brd->brd_nasid);
254				if (dest_brd->brd_type == KLTYPE_ROUTER)
255					printk(" r");
256			}
257			printk("\n");
258
259		} while ( (brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)) );
260	}
261}
262
263static pfn_t __init slot_getbasepfn(cnodeid_t cnode, int slot)
264{
265	nasid_t nasid = COMPACT_TO_NASID_NODEID(cnode);
266
267	return ((pfn_t)nasid << PFN_NASIDSHFT) | (slot << SLOT_PFNSHIFT);
268}
269
270/*
271 * Return the number of pages of memory provided by the given slot
272 * on the specified node.
273 */
274static pfn_t __init slot_getsize(cnodeid_t node, int slot)
275{
276	return (pfn_t) slot_psize_cache[node][slot];
277}
278
279/*
280 * Return highest slot filled
281 */
282static int __init node_getlastslot(cnodeid_t node)
283{
284	return (int) slot_lastfilled_cache[node];
285}
286
287/*
288 * Return the pfn of the last free page of memory on a node.
289 */
290static pfn_t __init node_getmaxclick(cnodeid_t node)
291{
292	pfn_t	slot_psize;
293	int	slot;
294
295	/*
296	 * Start at the top slot. When we find a slot with memory in it,
297	 * that's the winner.
298	 */
299	for (slot = (MAX_MEM_SLOTS - 1); slot >= 0; slot--) {
300		if ((slot_psize = slot_getsize(node, slot))) {
301			if (slot_psize == SLOT_IGNORED)
302				continue;
303			/* Return the basepfn + the slot size, minus 1. */
304			return slot_getbasepfn(node, slot) + slot_psize - 1;
305		}
306	}
307
308	/*
309	 * If there's no memory on the node, return 0. This is likely
310	 * to cause problems.
311	 */
312	return 0;
313}
314
315static pfn_t __init slot_psize_compute(cnodeid_t node, int slot)
316{
317	nasid_t nasid;
318	lboard_t *brd;
319	klmembnk_t *banks;
320	unsigned long size;
321
322	nasid = COMPACT_TO_NASID_NODEID(node);
323	/* Find the node board */
324	brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27);
325	if (!brd)
326		return 0;
327
328	/* Get the memory bank structure */
329	banks = (klmembnk_t *) find_first_component(brd, KLSTRUCT_MEMBNK);
330	if (!banks)
331		return 0;
332
333	/* Size in _Megabytes_ */
334	size = (unsigned long)banks->membnk_bnksz[slot/4];
335
336	/* hack for 128 dimm banks */
337	if (size <= 128) {
338		if (slot % 4 == 0) {
339			size <<= 20;		/* size in bytes */
340			return(size >> PAGE_SHIFT);
341		} else
342			return 0;
343	} else {
344		size /= 4;
345		size <<= 20;
346		return size >> PAGE_SHIFT;
347	}
348}
349
350static void __init mlreset(void)
351{
352	int i;
353
354	master_nasid = get_nasid();
355	fine_mode = is_fine_dirmode();
356
357	/*
358	 * Probe for all CPUs - this creates the cpumask and sets up the
359	 * mapping tables.  We need to do this as early as possible.
360	 */
361#ifdef CONFIG_SMP
362	cpu_node_probe();
363#endif
364
365	init_topology_matrix();
366	dump_topology();
367
368	gen_region_mask(&region_mask);
369
370	setup_replication_mask();
371
372	/*
373	 * Set all nodes' calias sizes to 8k
374	 */
375	for_each_online_node(i) {
376		nasid_t nasid;
377
378		nasid = COMPACT_TO_NASID_NODEID(i);
379
380		/*
381		 * Always have node 0 in the region mask, otherwise
382		 * CALIAS accesses get exceptions since the hub
383		 * thinks it is a node 0 address.
384		 */
385		REMOTE_HUB_S(nasid, PI_REGION_PRESENT, (region_mask | 1));
386#ifdef CONFIG_REPLICATE_EXHANDLERS
387		REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_8K);
388#else
389		REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_0);
390#endif
391
392#ifdef LATER
393		/*
394		 * Set up all hubs to have a big window pointing at
395		 * widget 0. Memory mode, widget 0, offset 0
396		 */
397		REMOTE_HUB_S(nasid, IIO_ITTE(SWIN0_BIGWIN),
398			((HUB_PIO_MAP_TO_MEM << IIO_ITTE_IOSP_SHIFT) |
399			(0 << IIO_ITTE_WIDGET_SHIFT)));
400#endif
401	}
402}
403
404static void __init szmem(void)
405{
406	pfn_t slot_psize, slot0sz = 0, nodebytes;	/* Hack to detect problem configs */
407	int slot, ignore;
408	cnodeid_t node;
409
410	num_physpages = 0;
411
412	for_each_online_node(node) {
413		ignore = nodebytes = 0;
414		for (slot = 0; slot < MAX_MEM_SLOTS; slot++) {
415			slot_psize = slot_psize_compute(node, slot);
416			if (slot == 0)
417				slot0sz = slot_psize;
418			/*
419			 * We need to refine the hack when we have replicated
420			 * kernel text.
421			 */
422			nodebytes += (1LL << SLOT_SHIFT);
423			if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) >
424						(slot0sz << PAGE_SHIFT))
425				ignore = 1;
426			if (ignore && slot_psize) {
427				printk("Ignoring slot %d onwards on node %d\n",
428								slot, node);
429				slot_psize_cache[node][slot] = SLOT_IGNORED;
430				slot = MAX_MEM_SLOTS;
431				continue;
432			}
433			num_physpages += slot_psize;
434			slot_psize_cache[node][slot] =
435					(unsigned short) slot_psize;
436			if (slot_psize)
437				slot_lastfilled_cache[node] = slot;
438		}
439	}
440}
441
442static void __init node_mem_init(cnodeid_t node)
443{
444	pfn_t slot_firstpfn = slot_getbasepfn(node, 0);
445	pfn_t slot_lastpfn = slot_firstpfn + slot_getsize(node, 0);
446	pfn_t slot_freepfn = node_getfirstfree(node);
447	struct pglist_data *pd;
448	unsigned long bootmap_size;
449
450	/*
451	 * Allocate the node data structures on the node first.
452	 */
453	__node_data[node] = __va(slot_freepfn << PAGE_SHIFT);
454
455	pd = NODE_DATA(node);
456	pd->bdata = &plat_node_bdata[node];
457
458	cpus_clear(hub_data(node)->h_cpus);
459
460	slot_freepfn += PFN_UP(sizeof(struct pglist_data) +
461			       sizeof(struct hub_data));
462
463  	bootmap_size = init_bootmem_node(NODE_DATA(node), slot_freepfn,
464					slot_firstpfn, slot_lastpfn);
465	free_bootmem_node(NODE_DATA(node), slot_firstpfn << PAGE_SHIFT,
466			(slot_lastpfn - slot_firstpfn) << PAGE_SHIFT);
467	reserve_bootmem_node(NODE_DATA(node), slot_firstpfn << PAGE_SHIFT,
468		((slot_freepfn - slot_firstpfn) << PAGE_SHIFT) + bootmap_size);
469}
470
471/*
472 * A node with nothing.  We use it to avoid any special casing in
473 * node_to_cpumask
474 */
475static struct node_data null_node = {
476	.hub = {
477		.h_cpus = CPU_MASK_NONE
478	}
479};
480
481/*
482 * Currently, the intranode memory hole support assumes that each slot
483 * contains at least 32 MBytes of memory. We assume all bootmem data
484 * fits on the first slot.
485 */
486void __init prom_meminit(void)
487{
488	cnodeid_t node;
489
490	mlreset();
491	szmem();
492
493	for (node = 0; node < MAX_COMPACT_NODES; node++) {
494		if (node_online(node)) {
495			node_mem_init(node);
496			continue;
497		}
498		__node_data[node] = &null_node;
499	}
500}
501
502void __init prom_free_prom_memory(void)
503{
504	/* We got nothing to free here ...  */
505}
506
507extern unsigned long setup_zero_pages(void);
508
509void __init paging_init(void)
510{
511	unsigned long zones_size[MAX_NR_ZONES] = {0, };
512	unsigned node;
513
514	pagetable_init();
515
516	for_each_online_node(node) {
517		pfn_t start_pfn = slot_getbasepfn(node, 0);
518		pfn_t end_pfn = node_getmaxclick(node) + 1;
519
520		zones_size[ZONE_NORMAL] = end_pfn - start_pfn;
521		free_area_init_node(node, NODE_DATA(node),
522				zones_size, start_pfn, NULL);
523
524		if (end_pfn > max_low_pfn)
525			max_low_pfn = end_pfn;
526	}
527}
528
529void __init mem_init(void)
530{
531	unsigned long codesize, datasize, initsize, tmp;
532	unsigned node;
533
534	high_memory = (void *) __va(num_physpages << PAGE_SHIFT);
535
536	for_each_online_node(node) {
537		unsigned slot, numslots;
538		struct page *end, *p;
539
540		/*
541		 * This will free up the bootmem, ie, slot 0 memory.
542		 */
543		totalram_pages += free_all_bootmem_node(NODE_DATA(node));
544
545		/*
546		 * We need to manually do the other slots.
547		 */
548		numslots = node_getlastslot(node);
549		for (slot = 1; slot <= numslots; slot++) {
550			p = nid_page_nr(node, slot_getbasepfn(node, slot) -
551					      slot_getbasepfn(node, 0));
552
553			/*
554			 * Free valid memory in current slot.
555			 */
556			for (end = p + slot_getsize(node, slot); p < end; p++) {
557				/* if (!page_is_ram(pgnr)) continue; */
558				/* commented out until page_is_ram works */
559				ClearPageReserved(p);
560				init_page_count(p);
561				__free_page(p);
562				totalram_pages++;
563			}
564		}
565	}
566
567	totalram_pages -= setup_zero_pages();	/* This comes from node 0 */
568
569	codesize =  (unsigned long) &_etext - (unsigned long) &_text;
570	datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
571	initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
572
573	tmp = nr_free_pages();
574	printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
575	       "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
576	       tmp << (PAGE_SHIFT-10),
577	       num_physpages << (PAGE_SHIFT-10),
578	       codesize >> 10,
579	       (num_physpages - tmp) << (PAGE_SHIFT-10),
580	       datasize >> 10,
581	       initsize >> 10,
582	       (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
583}
584