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 struct node_data *__node_data[MAX_COMPACT_NODES];
37
38 EXPORT_SYMBOL(__node_data);
39
40 static int fine_mode;
41
is_fine_dirmode(void)42 static int is_fine_dirmode(void)
43 {
44 return (((LOCAL_HUB_L(NI_STATUS_REV_ID) & NSRI_REGIONSIZE_MASK)
45 >> NSRI_REGIONSIZE_SHFT) & REGIONSIZE_FINE);
46 }
47
get_region(cnodeid_t cnode)48 static 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
56 static hubreg_t region_mask;
57
gen_region_mask(hubreg_t * region_mask)58 static 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
70 static int router_distance;
71
router_recurse(klrou_t * router_a,klrou_t * router_b,int depth)72 static 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
108 unsigned char __node_distances[MAX_COMPACT_NODES][MAX_COMPACT_NODES];
109
compute_node_distance(nasid_t nasid_a,nasid_t nasid_b)110 static int __init compute_node_distance(nasid_t nasid_a, nasid_t nasid_b)
111 {
112 klrou_t *router, *router_a = NULL, *router_b = NULL;
113 lboard_t *brd, *dest_brd;
114 cnodeid_t cnode;
115 nasid_t nasid;
116 int port;
117
118 /* Figure out which routers nodes in question are connected to */
119 for_each_online_node(cnode) {
120 nasid = COMPACT_TO_NASID_NODEID(cnode);
121
122 if (nasid == -1) continue;
123
124 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
125 KLTYPE_ROUTER);
126
127 if (!brd)
128 continue;
129
130 do {
131 if (brd->brd_flags & DUPLICATE_BOARD)
132 continue;
133
134 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
135 router->rou_rflag = 0;
136
137 for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
138 if (router->rou_port[port].port_nasid == INVALID_NASID)
139 continue;
140
141 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
142 router->rou_port[port].port_nasid,
143 router->rou_port[port].port_offset);
144
145 if (dest_brd->brd_type == KLTYPE_IP27) {
146 if (dest_brd->brd_nasid == nasid_a)
147 router_a = router;
148 if (dest_brd->brd_nasid == nasid_b)
149 router_b = router;
150 }
151 }
152
153 } while ((brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)));
154 }
155
156 if (router_a == NULL) {
157 printk("node_distance: router_a NULL\n");
158 return -1;
159 }
160 if (router_b == NULL) {
161 printk("node_distance: router_b NULL\n");
162 return -1;
163 }
164
165 if (nasid_a == nasid_b)
166 return 0;
167
168 if (router_a == router_b)
169 return 1;
170
171 router_distance = 100;
172 router_recurse(router_a, router_b, 2);
173
174 return router_distance;
175 }
176
init_topology_matrix(void)177 static void __init init_topology_matrix(void)
178 {
179 nasid_t nasid, nasid2;
180 cnodeid_t row, col;
181
182 for (row = 0; row < MAX_COMPACT_NODES; row++)
183 for (col = 0; col < MAX_COMPACT_NODES; col++)
184 __node_distances[row][col] = -1;
185
186 for_each_online_node(row) {
187 nasid = COMPACT_TO_NASID_NODEID(row);
188 for_each_online_node(col) {
189 nasid2 = COMPACT_TO_NASID_NODEID(col);
190 __node_distances[row][col] =
191 compute_node_distance(nasid, nasid2);
192 }
193 }
194 }
195
dump_topology(void)196 static void __init dump_topology(void)
197 {
198 nasid_t nasid;
199 cnodeid_t cnode;
200 lboard_t *brd, *dest_brd;
201 int port;
202 int router_num = 0;
203 klrou_t *router;
204 cnodeid_t row, col;
205
206 printk("************** Topology ********************\n");
207
208 printk(" ");
209 for_each_online_node(col)
210 printk("%02d ", col);
211 printk("\n");
212 for_each_online_node(row) {
213 printk("%02d ", row);
214 for_each_online_node(col)
215 printk("%2d ", node_distance(row, col));
216 printk("\n");
217 }
218
219 for_each_online_node(cnode) {
220 nasid = COMPACT_TO_NASID_NODEID(cnode);
221
222 if (nasid == -1) continue;
223
224 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
225 KLTYPE_ROUTER);
226
227 if (!brd)
228 continue;
229
230 do {
231 if (brd->brd_flags & DUPLICATE_BOARD)
232 continue;
233 printk("Router %d:", router_num);
234 router_num++;
235
236 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
237
238 for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
239 if (router->rou_port[port].port_nasid == INVALID_NASID)
240 continue;
241
242 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
243 router->rou_port[port].port_nasid,
244 router->rou_port[port].port_offset);
245
246 if (dest_brd->brd_type == KLTYPE_IP27)
247 printk(" %d", dest_brd->brd_nasid);
248 if (dest_brd->brd_type == KLTYPE_ROUTER)
249 printk(" r");
250 }
251 printk("\n");
252
253 } while ( (brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)) );
254 }
255 }
256
slot_getbasepfn(cnodeid_t cnode,int slot)257 static pfn_t __init slot_getbasepfn(cnodeid_t cnode, int slot)
258 {
259 nasid_t nasid = COMPACT_TO_NASID_NODEID(cnode);
260
261 return ((pfn_t)nasid << PFN_NASIDSHFT) | (slot << SLOT_PFNSHIFT);
262 }
263
slot_psize_compute(cnodeid_t node,int slot)264 static pfn_t __init slot_psize_compute(cnodeid_t node, int slot)
265 {
266 nasid_t nasid;
267 lboard_t *brd;
268 klmembnk_t *banks;
269 unsigned long size;
270
271 nasid = COMPACT_TO_NASID_NODEID(node);
272 /* Find the node board */
273 brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27);
274 if (!brd)
275 return 0;
276
277 /* Get the memory bank structure */
278 banks = (klmembnk_t *) find_first_component(brd, KLSTRUCT_MEMBNK);
279 if (!banks)
280 return 0;
281
282 /* Size in _Megabytes_ */
283 size = (unsigned long)banks->membnk_bnksz[slot/4];
284
285 /* hack for 128 dimm banks */
286 if (size <= 128) {
287 if (slot % 4 == 0) {
288 size <<= 20; /* size in bytes */
289 return(size >> PAGE_SHIFT);
290 } else
291 return 0;
292 } else {
293 size /= 4;
294 size <<= 20;
295 return size >> PAGE_SHIFT;
296 }
297 }
298
mlreset(void)299 static void __init mlreset(void)
300 {
301 int i;
302
303 master_nasid = get_nasid();
304 fine_mode = is_fine_dirmode();
305
306 /*
307 * Probe for all CPUs - this creates the cpumask and sets up the
308 * mapping tables. We need to do this as early as possible.
309 */
310 #ifdef CONFIG_SMP
311 cpu_node_probe();
312 #endif
313
314 init_topology_matrix();
315 dump_topology();
316
317 gen_region_mask(®ion_mask);
318
319 setup_replication_mask();
320
321 /*
322 * Set all nodes' calias sizes to 8k
323 */
324 for_each_online_node(i) {
325 nasid_t nasid;
326
327 nasid = COMPACT_TO_NASID_NODEID(i);
328
329 /*
330 * Always have node 0 in the region mask, otherwise
331 * CALIAS accesses get exceptions since the hub
332 * thinks it is a node 0 address.
333 */
334 REMOTE_HUB_S(nasid, PI_REGION_PRESENT, (region_mask | 1));
335 #ifdef CONFIG_REPLICATE_EXHANDLERS
336 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_8K);
337 #else
338 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_0);
339 #endif
340
341 #ifdef LATER
342 /*
343 * Set up all hubs to have a big window pointing at
344 * widget 0. Memory mode, widget 0, offset 0
345 */
346 REMOTE_HUB_S(nasid, IIO_ITTE(SWIN0_BIGWIN),
347 ((HUB_PIO_MAP_TO_MEM << IIO_ITTE_IOSP_SHIFT) |
348 (0 << IIO_ITTE_WIDGET_SHIFT)));
349 #endif
350 }
351 }
352
szmem(void)353 static void __init szmem(void)
354 {
355 pfn_t slot_psize, slot0sz = 0, nodebytes; /* Hack to detect problem configs */
356 int slot;
357 cnodeid_t node;
358
359 num_physpages = 0;
360
361 for_each_online_node(node) {
362 nodebytes = 0;
363 for (slot = 0; slot < MAX_MEM_SLOTS; slot++) {
364 slot_psize = slot_psize_compute(node, slot);
365 if (slot == 0)
366 slot0sz = slot_psize;
367 /*
368 * We need to refine the hack when we have replicated
369 * kernel text.
370 */
371 nodebytes += (1LL << SLOT_SHIFT);
372
373 if (!slot_psize)
374 continue;
375
376 if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) >
377 (slot0sz << PAGE_SHIFT)) {
378 printk("Ignoring slot %d onwards on node %d\n",
379 slot, node);
380 slot = MAX_MEM_SLOTS;
381 continue;
382 }
383 num_physpages += slot_psize;
384 add_active_range(node, slot_getbasepfn(node, slot),
385 slot_getbasepfn(node, slot) + slot_psize);
386 }
387 }
388 }
389
node_mem_init(cnodeid_t node)390 static void __init node_mem_init(cnodeid_t node)
391 {
392 pfn_t slot_firstpfn = slot_getbasepfn(node, 0);
393 pfn_t slot_freepfn = node_getfirstfree(node);
394 unsigned long bootmap_size;
395 pfn_t start_pfn, end_pfn;
396
397 get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
398
399 /*
400 * Allocate the node data structures on the node first.
401 */
402 __node_data[node] = __va(slot_freepfn << PAGE_SHIFT);
403
404 NODE_DATA(node)->bdata = &bootmem_node_data[node];
405 NODE_DATA(node)->node_start_pfn = start_pfn;
406 NODE_DATA(node)->node_spanned_pages = end_pfn - start_pfn;
407
408 cpus_clear(hub_data(node)->h_cpus);
409
410 slot_freepfn += PFN_UP(sizeof(struct pglist_data) +
411 sizeof(struct hub_data));
412
413 bootmap_size = init_bootmem_node(NODE_DATA(node), slot_freepfn,
414 start_pfn, end_pfn);
415 free_bootmem_with_active_regions(node, end_pfn);
416 reserve_bootmem_node(NODE_DATA(node), slot_firstpfn << PAGE_SHIFT,
417 ((slot_freepfn - slot_firstpfn) << PAGE_SHIFT) + bootmap_size,
418 BOOTMEM_DEFAULT);
419 sparse_memory_present_with_active_regions(node);
420 }
421
422 /*
423 * A node with nothing. We use it to avoid any special casing in
424 * cpumask_of_node
425 */
426 static struct node_data null_node = {
427 .hub = {
428 .h_cpus = CPU_MASK_NONE
429 }
430 };
431
432 /*
433 * Currently, the intranode memory hole support assumes that each slot
434 * contains at least 32 MBytes of memory. We assume all bootmem data
435 * fits on the first slot.
436 */
prom_meminit(void)437 void __init prom_meminit(void)
438 {
439 cnodeid_t node;
440
441 mlreset();
442 szmem();
443
444 for (node = 0; node < MAX_COMPACT_NODES; node++) {
445 if (node_online(node)) {
446 node_mem_init(node);
447 continue;
448 }
449 __node_data[node] = &null_node;
450 }
451 }
452
prom_free_prom_memory(void)453 void __init prom_free_prom_memory(void)
454 {
455 /* We got nothing to free here ... */
456 }
457
458 extern unsigned long setup_zero_pages(void);
459
paging_init(void)460 void __init paging_init(void)
461 {
462 unsigned long zones_size[MAX_NR_ZONES] = {0, };
463 unsigned node;
464
465 pagetable_init();
466
467 for_each_online_node(node) {
468 pfn_t start_pfn, end_pfn;
469
470 get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
471
472 if (end_pfn > max_low_pfn)
473 max_low_pfn = end_pfn;
474 }
475 zones_size[ZONE_NORMAL] = max_low_pfn;
476 free_area_init_nodes(zones_size);
477 }
478
mem_init(void)479 void __init mem_init(void)
480 {
481 unsigned long codesize, datasize, initsize, tmp;
482 unsigned node;
483
484 high_memory = (void *) __va(num_physpages << PAGE_SHIFT);
485
486 for_each_online_node(node) {
487 /*
488 * This will free up the bootmem, ie, slot 0 memory.
489 */
490 totalram_pages += free_all_bootmem_node(NODE_DATA(node));
491 }
492
493 totalram_pages -= setup_zero_pages(); /* This comes from node 0 */
494
495 codesize = (unsigned long) &_etext - (unsigned long) &_text;
496 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
497 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
498
499 tmp = nr_free_pages();
500 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
501 "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
502 tmp << (PAGE_SHIFT-10),
503 num_physpages << (PAGE_SHIFT-10),
504 codesize >> 10,
505 (num_physpages - tmp) << (PAGE_SHIFT-10),
506 datasize >> 10,
507 initsize >> 10,
508 totalhigh_pages << (PAGE_SHIFT-10));
509 }
510