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 completely 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/export.h>
19 #include <linux/nodemask.h>
20 #include <linux/swap.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/agent.h>
29 #include <asm/sn/klconfig.h>
30
31 #include "ip27-common.h"
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_NUMNODES];
37
38 EXPORT_SYMBOL(__node_data);
39
gen_region_mask(void)40 static u64 gen_region_mask(void)
41 {
42 int region_shift;
43 u64 region_mask;
44 nasid_t nasid;
45
46 region_shift = get_region_shift();
47 region_mask = 0;
48 for_each_online_node(nasid)
49 region_mask |= BIT_ULL(nasid >> region_shift);
50
51 return region_mask;
52 }
53
54 #define rou_rflag rou_flags
55
56 static int router_distance;
57
router_recurse(klrou_t * router_a,klrou_t * router_b,int depth)58 static void router_recurse(klrou_t *router_a, klrou_t *router_b, int depth)
59 {
60 klrou_t *router;
61 lboard_t *brd;
62 int port;
63
64 if (router_a->rou_rflag == 1)
65 return;
66
67 if (depth >= router_distance)
68 return;
69
70 router_a->rou_rflag = 1;
71
72 for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
73 if (router_a->rou_port[port].port_nasid == INVALID_NASID)
74 continue;
75
76 brd = (lboard_t *)NODE_OFFSET_TO_K0(
77 router_a->rou_port[port].port_nasid,
78 router_a->rou_port[port].port_offset);
79
80 if (brd->brd_type == KLTYPE_ROUTER) {
81 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
82 if (router == router_b) {
83 if (depth < router_distance)
84 router_distance = depth;
85 }
86 else
87 router_recurse(router, router_b, depth + 1);
88 }
89 }
90
91 router_a->rou_rflag = 0;
92 }
93
94 unsigned char __node_distances[MAX_NUMNODES][MAX_NUMNODES];
95 EXPORT_SYMBOL(__node_distances);
96
compute_node_distance(nasid_t nasid_a,nasid_t nasid_b)97 static int __init compute_node_distance(nasid_t nasid_a, nasid_t nasid_b)
98 {
99 klrou_t *router, *router_a = NULL, *router_b = NULL;
100 lboard_t *brd, *dest_brd;
101 nasid_t nasid;
102 int port;
103
104 /* Figure out which routers nodes in question are connected to */
105 for_each_online_node(nasid) {
106 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
107 KLTYPE_ROUTER);
108
109 if (!brd)
110 continue;
111
112 do {
113 if (brd->brd_flags & DUPLICATE_BOARD)
114 continue;
115
116 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
117 router->rou_rflag = 0;
118
119 for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
120 if (router->rou_port[port].port_nasid == INVALID_NASID)
121 continue;
122
123 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
124 router->rou_port[port].port_nasid,
125 router->rou_port[port].port_offset);
126
127 if (dest_brd->brd_type == KLTYPE_IP27) {
128 if (dest_brd->brd_nasid == nasid_a)
129 router_a = router;
130 if (dest_brd->brd_nasid == nasid_b)
131 router_b = router;
132 }
133 }
134
135 } while ((brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)));
136 }
137
138 if (nasid_a == nasid_b)
139 return LOCAL_DISTANCE;
140
141 if (router_a == router_b)
142 return LOCAL_DISTANCE + 1;
143
144 if (router_a == NULL) {
145 pr_info("node_distance: router_a NULL\n");
146 return 255;
147 }
148 if (router_b == NULL) {
149 pr_info("node_distance: router_b NULL\n");
150 return 255;
151 }
152
153 router_distance = 100;
154 router_recurse(router_a, router_b, 2);
155
156 return LOCAL_DISTANCE + router_distance;
157 }
158
init_topology_matrix(void)159 static void __init init_topology_matrix(void)
160 {
161 nasid_t row, col;
162
163 for (row = 0; row < MAX_NUMNODES; row++)
164 for (col = 0; col < MAX_NUMNODES; col++)
165 __node_distances[row][col] = -1;
166
167 for_each_online_node(row) {
168 for_each_online_node(col) {
169 __node_distances[row][col] =
170 compute_node_distance(row, col);
171 }
172 }
173 }
174
dump_topology(void)175 static void __init dump_topology(void)
176 {
177 nasid_t nasid;
178 lboard_t *brd, *dest_brd;
179 int port;
180 int router_num = 0;
181 klrou_t *router;
182 nasid_t row, col;
183
184 pr_info("************** Topology ********************\n");
185
186 pr_info(" ");
187 for_each_online_node(col)
188 pr_cont("%02d ", col);
189 pr_cont("\n");
190 for_each_online_node(row) {
191 pr_info("%02d ", row);
192 for_each_online_node(col)
193 pr_cont("%2d ", node_distance(row, col));
194 pr_cont("\n");
195 }
196
197 for_each_online_node(nasid) {
198 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
199 KLTYPE_ROUTER);
200
201 if (!brd)
202 continue;
203
204 do {
205 if (brd->brd_flags & DUPLICATE_BOARD)
206 continue;
207 pr_cont("Router %d:", router_num);
208 router_num++;
209
210 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
211
212 for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
213 if (router->rou_port[port].port_nasid == INVALID_NASID)
214 continue;
215
216 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
217 router->rou_port[port].port_nasid,
218 router->rou_port[port].port_offset);
219
220 if (dest_brd->brd_type == KLTYPE_IP27)
221 pr_cont(" %d", dest_brd->brd_nasid);
222 if (dest_brd->brd_type == KLTYPE_ROUTER)
223 pr_cont(" r");
224 }
225 pr_cont("\n");
226
227 } while ( (brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)) );
228 }
229 }
230
slot_getbasepfn(nasid_t nasid,int slot)231 static unsigned long __init slot_getbasepfn(nasid_t nasid, int slot)
232 {
233 return ((unsigned long)nasid << PFN_NASIDSHFT) | (slot << SLOT_PFNSHIFT);
234 }
235
slot_psize_compute(nasid_t nasid,int slot)236 static unsigned long __init slot_psize_compute(nasid_t nasid, int slot)
237 {
238 lboard_t *brd;
239 klmembnk_t *banks;
240 unsigned long size;
241
242 /* Find the node board */
243 brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27);
244 if (!brd)
245 return 0;
246
247 /* Get the memory bank structure */
248 banks = (klmembnk_t *) find_first_component(brd, KLSTRUCT_MEMBNK);
249 if (!banks)
250 return 0;
251
252 /* Size in _Megabytes_ */
253 size = (unsigned long)banks->membnk_bnksz[slot/4];
254
255 /* hack for 128 dimm banks */
256 if (size <= 128) {
257 if (slot % 4 == 0) {
258 size <<= 20; /* size in bytes */
259 return size >> PAGE_SHIFT;
260 } else
261 return 0;
262 } else {
263 size /= 4;
264 size <<= 20;
265 return size >> PAGE_SHIFT;
266 }
267 }
268
mlreset(void)269 static void __init mlreset(void)
270 {
271 u64 region_mask;
272 nasid_t nasid;
273
274 master_nasid = get_nasid();
275
276 /*
277 * Probe for all CPUs - this creates the cpumask and sets up the
278 * mapping tables. We need to do this as early as possible.
279 */
280 #ifdef CONFIG_SMP
281 cpu_node_probe();
282 #endif
283
284 init_topology_matrix();
285 dump_topology();
286
287 region_mask = gen_region_mask();
288
289 setup_replication_mask();
290
291 /*
292 * Set all nodes' calias sizes to 8k
293 */
294 for_each_online_node(nasid) {
295 /*
296 * Always have node 0 in the region mask, otherwise
297 * CALIAS accesses get exceptions since the hub
298 * thinks it is a node 0 address.
299 */
300 REMOTE_HUB_S(nasid, PI_REGION_PRESENT, (region_mask | 1));
301 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_0);
302
303 #ifdef LATER
304 /*
305 * Set up all hubs to have a big window pointing at
306 * widget 0. Memory mode, widget 0, offset 0
307 */
308 REMOTE_HUB_S(nasid, IIO_ITTE(SWIN0_BIGWIN),
309 ((HUB_PIO_MAP_TO_MEM << IIO_ITTE_IOSP_SHIFT) |
310 (0 << IIO_ITTE_WIDGET_SHIFT)));
311 #endif
312 }
313 }
314
szmem(void)315 static void __init szmem(void)
316 {
317 unsigned long slot_psize, slot0sz = 0, nodebytes; /* Hack to detect problem configs */
318 int slot;
319 nasid_t node;
320
321 for_each_online_node(node) {
322 nodebytes = 0;
323 for (slot = 0; slot < MAX_MEM_SLOTS; slot++) {
324 slot_psize = slot_psize_compute(node, slot);
325 if (slot == 0)
326 slot0sz = slot_psize;
327 /*
328 * We need to refine the hack when we have replicated
329 * kernel text.
330 */
331 nodebytes += (1LL << SLOT_SHIFT);
332
333 if (!slot_psize)
334 continue;
335
336 if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) >
337 (slot0sz << PAGE_SHIFT)) {
338 pr_info("Ignoring slot %d onwards on node %d\n",
339 slot, node);
340 slot = MAX_MEM_SLOTS;
341 continue;
342 }
343 memblock_add_node(PFN_PHYS(slot_getbasepfn(node, slot)),
344 PFN_PHYS(slot_psize), node,
345 MEMBLOCK_NONE);
346 }
347 }
348 }
349
node_mem_init(nasid_t node)350 static void __init node_mem_init(nasid_t node)
351 {
352 unsigned long slot_firstpfn = slot_getbasepfn(node, 0);
353 unsigned long slot_freepfn = node_getfirstfree(node);
354 unsigned long start_pfn, end_pfn;
355
356 get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
357
358 /*
359 * Allocate the node data structures on the node first.
360 */
361 __node_data[node] = __va(slot_freepfn << PAGE_SHIFT);
362 memset(__node_data[node], 0, PAGE_SIZE);
363
364 NODE_DATA(node)->node_start_pfn = start_pfn;
365 NODE_DATA(node)->node_spanned_pages = end_pfn - start_pfn;
366
367 cpumask_clear(&hub_data(node)->h_cpus);
368
369 slot_freepfn += PFN_UP(sizeof(struct pglist_data) +
370 sizeof(struct hub_data));
371
372 memblock_reserve(slot_firstpfn << PAGE_SHIFT,
373 ((slot_freepfn - slot_firstpfn) << PAGE_SHIFT));
374 }
375
376 /*
377 * A node with nothing. We use it to avoid any special casing in
378 * cpumask_of_node
379 */
380 static struct node_data null_node = {
381 .hub = {
382 .h_cpus = CPU_MASK_NONE
383 }
384 };
385
386 /*
387 * Currently, the intranode memory hole support assumes that each slot
388 * contains at least 32 MBytes of memory. We assume all bootmem data
389 * fits on the first slot.
390 */
prom_meminit(void)391 void __init prom_meminit(void)
392 {
393 nasid_t node;
394
395 mlreset();
396 szmem();
397 max_low_pfn = PHYS_PFN(memblock_end_of_DRAM());
398
399 for (node = 0; node < MAX_NUMNODES; node++) {
400 if (node_online(node)) {
401 node_mem_init(node);
402 continue;
403 }
404 __node_data[node] = &null_node;
405 }
406 }
407
408 extern void setup_zero_pages(void);
409
paging_init(void)410 void __init paging_init(void)
411 {
412 unsigned long zones_size[MAX_NR_ZONES] = {0, };
413
414 pagetable_init();
415 zones_size[ZONE_NORMAL] = max_low_pfn;
416 free_area_init(zones_size);
417 }
418
mem_init(void)419 void __init mem_init(void)
420 {
421 high_memory = (void *) __va(get_num_physpages() << PAGE_SHIFT);
422 memblock_free_all();
423 setup_zero_pages(); /* This comes from node 0 */
424 }
425
arch_alloc_nodedata(int nid)426 pg_data_t * __init arch_alloc_nodedata(int nid)
427 {
428 return memblock_alloc(sizeof(pg_data_t), SMP_CACHE_BYTES);
429 }
430
arch_refresh_nodedata(int nid,pg_data_t * pgdat)431 void arch_refresh_nodedata(int nid, pg_data_t *pgdat)
432 {
433 __node_data[nid] = (struct node_data *)pgdat;
434 }
435