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/memblock.h>
16#include <linux/mm.h>
17#include <linux/mmzone.h>
18#include <linux/module.h>
19#include <linux/nodemask.h>
20#include <linux/swap.h>
21#include <linux/bootmem.h>
22#include <linux/pfn.h>
23#include <linux/highmem.h>
24#include <asm/page.h>
25#include <asm/pgalloc.h>
26#include <asm/sections.h>
27
28#include <asm/sn/arch.h>
29#include <asm/sn/hub.h>
30#include <asm/sn/klconfig.h>
31#include <asm/sn/sn_private.h>
32
33
34#define SLOT_PFNSHIFT		(SLOT_SHIFT - PAGE_SHIFT)
35#define PFN_NASIDSHFT		(NASID_SHFT - PAGE_SHIFT)
36
37struct node_data *__node_data[MAX_COMPACT_NODES];
38
39EXPORT_SYMBOL(__node_data);
40
41static int fine_mode;
42
43static int is_fine_dirmode(void)
44{
45	return ((LOCAL_HUB_L(NI_STATUS_REV_ID) & NSRI_REGIONSIZE_MASK) >> NSRI_REGIONSIZE_SHFT) & REGIONSIZE_FINE;
46}
47
48static hubreg_t get_region(cnodeid_t cnode)
49{
50	if (fine_mode)
51		return COMPACT_TO_NASID_NODEID(cnode) >> NASID_TO_FINEREG_SHFT;
52	else
53		return COMPACT_TO_NASID_NODEID(cnode) >> NASID_TO_COARSEREG_SHFT;
54}
55
56static hubreg_t region_mask;
57
58static void gen_region_mask(hubreg_t *region_mask)
59{
60	cnodeid_t cnode;
61
62	(*region_mask) = 0;
63	for_each_online_node(cnode) {
64		(*region_mask) |= 1ULL << get_region(cnode);
65	}
66}
67
68#define rou_rflag	rou_flags
69
70static int router_distance;
71
72static void router_recurse(klrou_t *router_a, klrou_t *router_b, int depth)
73{
74	klrou_t *router;
75	lboard_t *brd;
76	int	port;
77
78	if (router_a->rou_rflag == 1)
79		return;
80
81	if (depth >= router_distance)
82		return;
83
84	router_a->rou_rflag = 1;
85
86	for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
87		if (router_a->rou_port[port].port_nasid == INVALID_NASID)
88			continue;
89
90		brd = (lboard_t *)NODE_OFFSET_TO_K0(
91			router_a->rou_port[port].port_nasid,
92			router_a->rou_port[port].port_offset);
93
94		if (brd->brd_type == KLTYPE_ROUTER) {
95			router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
96			if (router == router_b) {
97				if (depth < router_distance)
98					router_distance = depth;
99			}
100			else
101				router_recurse(router, router_b, depth + 1);
102		}
103	}
104
105	router_a->rou_rflag = 0;
106}
107
108unsigned char __node_distances[MAX_COMPACT_NODES][MAX_COMPACT_NODES];
109EXPORT_SYMBOL(__node_distances);
110
111static int __init compute_node_distance(nasid_t nasid_a, nasid_t nasid_b)
112{
113	klrou_t *router, *router_a = NULL, *router_b = NULL;
114	lboard_t *brd, *dest_brd;
115	cnodeid_t cnode;
116	nasid_t nasid;
117	int port;
118
119	/* Figure out which routers nodes in question are connected to */
120	for_each_online_node(cnode) {
121		nasid = COMPACT_TO_NASID_NODEID(cnode);
122
123		if (nasid == -1) continue;
124
125		brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
126					KLTYPE_ROUTER);
127
128		if (!brd)
129			continue;
130
131		do {
132			if (brd->brd_flags & DUPLICATE_BOARD)
133				continue;
134
135			router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
136			router->rou_rflag = 0;
137
138			for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
139				if (router->rou_port[port].port_nasid == INVALID_NASID)
140					continue;
141
142				dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
143					router->rou_port[port].port_nasid,
144					router->rou_port[port].port_offset);
145
146				if (dest_brd->brd_type == KLTYPE_IP27) {
147					if (dest_brd->brd_nasid == nasid_a)
148						router_a = router;
149					if (dest_brd->brd_nasid == nasid_b)
150						router_b = router;
151				}
152			}
153
154		} while ((brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)));
155	}
156
157	if (router_a == NULL) {
158		printk("node_distance: router_a NULL\n");
159		return -1;
160	}
161	if (router_b == NULL) {
162		printk("node_distance: router_b NULL\n");
163		return -1;
164	}
165
166	if (nasid_a == nasid_b)
167		return 0;
168
169	if (router_a == router_b)
170		return 1;
171
172	router_distance = 100;
173	router_recurse(router_a, router_b, 2);
174
175	return router_distance;
176}
177
178static void __init init_topology_matrix(void)
179{
180	nasid_t nasid, nasid2;
181	cnodeid_t row, col;
182
183	for (row = 0; row < MAX_COMPACT_NODES; row++)
184		for (col = 0; col < MAX_COMPACT_NODES; col++)
185			__node_distances[row][col] = -1;
186
187	for_each_online_node(row) {
188		nasid = COMPACT_TO_NASID_NODEID(row);
189		for_each_online_node(col) {
190			nasid2 = COMPACT_TO_NASID_NODEID(col);
191			__node_distances[row][col] =
192				compute_node_distance(nasid, nasid2);
193		}
194	}
195}
196
197static void __init dump_topology(void)
198{
199	nasid_t nasid;
200	cnodeid_t cnode;
201	lboard_t *brd, *dest_brd;
202	int port;
203	int router_num = 0;
204	klrou_t *router;
205	cnodeid_t row, col;
206
207	printk("************** Topology ********************\n");
208
209	printk("    ");
210	for_each_online_node(col)
211		printk("%02d ", col);
212	printk("\n");
213	for_each_online_node(row) {
214		printk("%02d  ", row);
215		for_each_online_node(col)
216			printk("%2d ", node_distance(row, col));
217		printk("\n");
218	}
219
220	for_each_online_node(cnode) {
221		nasid = COMPACT_TO_NASID_NODEID(cnode);
222
223		if (nasid == -1) continue;
224
225		brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
226					KLTYPE_ROUTER);
227
228		if (!brd)
229			continue;
230
231		do {
232			if (brd->brd_flags & DUPLICATE_BOARD)
233				continue;
234			printk("Router %d:", router_num);
235			router_num++;
236
237			router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
238
239			for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
240				if (router->rou_port[port].port_nasid == INVALID_NASID)
241					continue;
242
243				dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
244					router->rou_port[port].port_nasid,
245					router->rou_port[port].port_offset);
246
247				if (dest_brd->brd_type == KLTYPE_IP27)
248					printk(" %d", dest_brd->brd_nasid);
249				if (dest_brd->brd_type == KLTYPE_ROUTER)
250					printk(" r");
251			}
252			printk("\n");
253
254		} while ( (brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)) );
255	}
256}
257
258static unsigned long __init slot_getbasepfn(cnodeid_t cnode, int slot)
259{
260	nasid_t nasid = COMPACT_TO_NASID_NODEID(cnode);
261
262	return ((unsigned long)nasid << PFN_NASIDSHFT) | (slot << SLOT_PFNSHIFT);
263}
264
265static unsigned long __init slot_psize_compute(cnodeid_t node, int slot)
266{
267	nasid_t nasid;
268	lboard_t *brd;
269	klmembnk_t *banks;
270	unsigned long size;
271
272	nasid = COMPACT_TO_NASID_NODEID(node);
273	/* Find the node board */
274	brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27);
275	if (!brd)
276		return 0;
277
278	/* Get the memory bank structure */
279	banks = (klmembnk_t *) find_first_component(brd, KLSTRUCT_MEMBNK);
280	if (!banks)
281		return 0;
282
283	/* Size in _Megabytes_ */
284	size = (unsigned long)banks->membnk_bnksz[slot/4];
285
286	/* hack for 128 dimm banks */
287	if (size <= 128) {
288		if (slot % 4 == 0) {
289			size <<= 20;		/* size in bytes */
290			return size >> PAGE_SHIFT;
291		} else
292			return 0;
293	} else {
294		size /= 4;
295		size <<= 20;
296		return size >> PAGE_SHIFT;
297	}
298}
299
300static void __init mlreset(void)
301{
302	int i;
303
304	master_nasid = get_nasid();
305	fine_mode = is_fine_dirmode();
306
307	/*
308	 * Probe for all CPUs - this creates the cpumask and sets up the
309	 * mapping tables.  We need to do this as early as possible.
310	 */
311#ifdef CONFIG_SMP
312	cpu_node_probe();
313#endif
314
315	init_topology_matrix();
316	dump_topology();
317
318	gen_region_mask(&region_mask);
319
320	setup_replication_mask();
321
322	/*
323	 * Set all nodes' calias sizes to 8k
324	 */
325	for_each_online_node(i) {
326		nasid_t nasid;
327
328		nasid = COMPACT_TO_NASID_NODEID(i);
329
330		/*
331		 * Always have node 0 in the region mask, otherwise
332		 * CALIAS accesses get exceptions since the hub
333		 * thinks it is a node 0 address.
334		 */
335		REMOTE_HUB_S(nasid, PI_REGION_PRESENT, (region_mask | 1));
336#ifdef CONFIG_REPLICATE_EXHANDLERS
337		REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_8K);
338#else
339		REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_0);
340#endif
341
342#ifdef LATER
343		/*
344		 * Set up all hubs to have a big window pointing at
345		 * widget 0. Memory mode, widget 0, offset 0
346		 */
347		REMOTE_HUB_S(nasid, IIO_ITTE(SWIN0_BIGWIN),
348			((HUB_PIO_MAP_TO_MEM << IIO_ITTE_IOSP_SHIFT) |
349			(0 << IIO_ITTE_WIDGET_SHIFT)));
350#endif
351	}
352}
353
354static void __init szmem(void)
355{
356	unsigned long slot_psize, slot0sz = 0, nodebytes;	/* Hack to detect problem configs */
357	int slot;
358	cnodeid_t node;
359
360	for_each_online_node(node) {
361		nodebytes = 0;
362		for (slot = 0; slot < MAX_MEM_SLOTS; slot++) {
363			slot_psize = slot_psize_compute(node, slot);
364			if (slot == 0)
365				slot0sz = slot_psize;
366			/*
367			 * We need to refine the hack when we have replicated
368			 * kernel text.
369			 */
370			nodebytes += (1LL << SLOT_SHIFT);
371
372			if (!slot_psize)
373				continue;
374
375			if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) >
376						(slot0sz << PAGE_SHIFT)) {
377				printk("Ignoring slot %d onwards on node %d\n",
378								slot, node);
379				slot = MAX_MEM_SLOTS;
380				continue;
381			}
382			memblock_add_node(PFN_PHYS(slot_getbasepfn(node, slot)),
383					  PFN_PHYS(slot_psize), node);
384		}
385	}
386}
387
388static void __init node_mem_init(cnodeid_t node)
389{
390	unsigned long slot_firstpfn = slot_getbasepfn(node, 0);
391	unsigned long slot_freepfn = node_getfirstfree(node);
392	unsigned long bootmap_size;
393	unsigned long start_pfn, end_pfn;
394
395	get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
396
397	/*
398	 * Allocate the node data structures on the node first.
399	 */
400	__node_data[node] = __va(slot_freepfn << PAGE_SHIFT);
401	memset(__node_data[node], 0, PAGE_SIZE);
402
403	NODE_DATA(node)->bdata = &bootmem_node_data[node];
404	NODE_DATA(node)->node_start_pfn = start_pfn;
405	NODE_DATA(node)->node_spanned_pages = end_pfn - start_pfn;
406
407	cpumask_clear(&hub_data(node)->h_cpus);
408
409	slot_freepfn += PFN_UP(sizeof(struct pglist_data) +
410			       sizeof(struct hub_data));
411
412	bootmap_size = init_bootmem_node(NODE_DATA(node), slot_freepfn,
413					start_pfn, end_pfn);
414	free_bootmem_with_active_regions(node, end_pfn);
415	reserve_bootmem_node(NODE_DATA(node), slot_firstpfn << PAGE_SHIFT,
416		((slot_freepfn - slot_firstpfn) << PAGE_SHIFT) + bootmap_size,
417		BOOTMEM_DEFAULT);
418	sparse_memory_present_with_active_regions(node);
419}
420
421/*
422 * A node with nothing.	 We use it to avoid any special casing in
423 * cpumask_of_node
424 */
425static struct node_data null_node = {
426	.hub = {
427		.h_cpus = CPU_MASK_NONE
428	}
429};
430
431/*
432 * Currently, the intranode memory hole support assumes that each slot
433 * contains at least 32 MBytes of memory. We assume all bootmem data
434 * fits on the first slot.
435 */
436void __init prom_meminit(void)
437{
438	cnodeid_t node;
439
440	mlreset();
441	szmem();
442
443	for (node = 0; node < MAX_COMPACT_NODES; node++) {
444		if (node_online(node)) {
445			node_mem_init(node);
446			continue;
447		}
448		__node_data[node] = &null_node;
449	}
450}
451
452void __init prom_free_prom_memory(void)
453{
454	/* We got nothing to free here ...  */
455}
456
457extern void setup_zero_pages(void);
458
459void __init paging_init(void)
460{
461	unsigned long zones_size[MAX_NR_ZONES] = {0, };
462	unsigned node;
463
464	pagetable_init();
465
466	for_each_online_node(node) {
467		unsigned long start_pfn, end_pfn;
468
469		get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
470
471		if (end_pfn > max_low_pfn)
472			max_low_pfn = end_pfn;
473	}
474	zones_size[ZONE_NORMAL] = max_low_pfn;
475	free_area_init_nodes(zones_size);
476}
477
478void __init mem_init(void)
479{
480	high_memory = (void *) __va(get_num_physpages() << PAGE_SHIFT);
481	free_all_bootmem();
482	setup_zero_pages();	/* This comes from node 0 */
483	mem_init_print_info(NULL);
484}
485