Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Some of the code in this file has been gleaned from the 64 bit | |
3 | * discontigmem support code base. | |
4 | * | |
5 | * Copyright (C) 2002, IBM Corp. | |
6 | * | |
7 | * All rights reserved. | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, but | |
15 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
17 | * NON INFRINGEMENT. See the GNU General Public License for more | |
18 | * details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with this program; if not, write to the Free Software | |
22 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | * | |
24 | * Send feedback to Pat Gaughen <gone@us.ibm.com> | |
25 | */ | |
26 | #include <linux/config.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/bootmem.h> | |
29 | #include <linux/mmzone.h> | |
30 | #include <linux/acpi.h> | |
31 | #include <linux/nodemask.h> | |
32 | #include <asm/srat.h> | |
33 | #include <asm/topology.h> | |
34 | ||
35 | /* | |
36 | * proximity macros and definitions | |
37 | */ | |
38 | #define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */ | |
39 | #define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */ | |
40 | #define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit)) | |
41 | #define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit))) | |
42 | #define MAX_PXM_DOMAINS 256 /* 1 byte and no promises about values */ | |
43 | /* bitmap length; _PXM is at most 255 */ | |
44 | #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8) | |
45 | static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */ | |
46 | ||
47 | #define MAX_CHUNKS_PER_NODE 4 | |
48 | #define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES) | |
49 | struct node_memory_chunk_s { | |
50 | unsigned long start_pfn; | |
51 | unsigned long end_pfn; | |
52 | u8 pxm; // proximity domain of node | |
53 | u8 nid; // which cnode contains this chunk? | |
54 | u8 bank; // which mem bank on this node | |
55 | }; | |
56 | static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS]; | |
57 | ||
58 | static int num_memory_chunks; /* total number of memory chunks */ | |
59 | static int zholes_size_init; | |
60 | static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES]; | |
61 | ||
62 | extern void * boot_ioremap(unsigned long, unsigned long); | |
63 | ||
64 | /* Identify CPU proximity domains */ | |
65 | static void __init parse_cpu_affinity_structure(char *p) | |
66 | { | |
67 | struct acpi_table_processor_affinity *cpu_affinity = | |
68 | (struct acpi_table_processor_affinity *) p; | |
69 | ||
70 | if (!cpu_affinity->flags.enabled) | |
71 | return; /* empty entry */ | |
72 | ||
73 | /* mark this node as "seen" in node bitmap */ | |
74 | BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain); | |
75 | ||
76 | printk("CPU 0x%02X in proximity domain 0x%02X\n", | |
77 | cpu_affinity->apic_id, cpu_affinity->proximity_domain); | |
78 | } | |
79 | ||
80 | /* | |
81 | * Identify memory proximity domains and hot-remove capabilities. | |
82 | * Fill node memory chunk list structure. | |
83 | */ | |
84 | static void __init parse_memory_affinity_structure (char *sratp) | |
85 | { | |
86 | unsigned long long paddr, size; | |
87 | unsigned long start_pfn, end_pfn; | |
88 | u8 pxm; | |
89 | struct node_memory_chunk_s *p, *q, *pend; | |
90 | struct acpi_table_memory_affinity *memory_affinity = | |
91 | (struct acpi_table_memory_affinity *) sratp; | |
92 | ||
93 | if (!memory_affinity->flags.enabled) | |
94 | return; /* empty entry */ | |
95 | ||
96 | /* mark this node as "seen" in node bitmap */ | |
97 | BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain); | |
98 | ||
99 | /* calculate info for memory chunk structure */ | |
100 | paddr = memory_affinity->base_addr_hi; | |
101 | paddr = (paddr << 32) | memory_affinity->base_addr_lo; | |
102 | size = memory_affinity->length_hi; | |
103 | size = (size << 32) | memory_affinity->length_lo; | |
104 | ||
105 | start_pfn = paddr >> PAGE_SHIFT; | |
106 | end_pfn = (paddr + size) >> PAGE_SHIFT; | |
107 | ||
108 | pxm = memory_affinity->proximity_domain; | |
109 | ||
110 | if (num_memory_chunks >= MAXCHUNKS) { | |
111 | printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n", | |
112 | size/(1024*1024), paddr); | |
113 | return; | |
114 | } | |
115 | ||
116 | /* Insertion sort based on base address */ | |
117 | pend = &node_memory_chunk[num_memory_chunks]; | |
118 | for (p = &node_memory_chunk[0]; p < pend; p++) { | |
119 | if (start_pfn < p->start_pfn) | |
120 | break; | |
121 | } | |
122 | if (p < pend) { | |
123 | for (q = pend; q >= p; q--) | |
124 | *(q + 1) = *q; | |
125 | } | |
126 | p->start_pfn = start_pfn; | |
127 | p->end_pfn = end_pfn; | |
128 | p->pxm = pxm; | |
129 | ||
130 | num_memory_chunks++; | |
131 | ||
132 | printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n", | |
133 | start_pfn, end_pfn, | |
134 | memory_affinity->memory_type, | |
135 | memory_affinity->proximity_domain, | |
136 | (memory_affinity->flags.hot_pluggable ? | |
137 | "enabled and removable" : "enabled" ) ); | |
138 | } | |
139 | ||
140 | #if MAX_NR_ZONES != 3 | |
141 | #error "MAX_NR_ZONES != 3, chunk_to_zone requires review" | |
142 | #endif | |
143 | /* Take a chunk of pages from page frame cstart to cend and count the number | |
144 | * of pages in each zone, returned via zones[]. | |
145 | */ | |
146 | static __init void chunk_to_zones(unsigned long cstart, unsigned long cend, | |
147 | unsigned long *zones) | |
148 | { | |
149 | unsigned long max_dma; | |
150 | extern unsigned long max_low_pfn; | |
151 | ||
152 | int z; | |
153 | unsigned long rend; | |
154 | ||
155 | /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide | |
156 | * similarly scoped information and should be handled in a consistant | |
157 | * manner. | |
158 | */ | |
159 | max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
160 | ||
161 | /* Split the hole into the zones in which it falls. Repeatedly | |
162 | * take the segment in which the remaining hole starts, round it | |
163 | * to the end of that zone. | |
164 | */ | |
165 | memset(zones, 0, MAX_NR_ZONES * sizeof(long)); | |
166 | while (cstart < cend) { | |
167 | if (cstart < max_dma) { | |
168 | z = ZONE_DMA; | |
169 | rend = (cend < max_dma)? cend : max_dma; | |
170 | ||
171 | } else if (cstart < max_low_pfn) { | |
172 | z = ZONE_NORMAL; | |
173 | rend = (cend < max_low_pfn)? cend : max_low_pfn; | |
174 | ||
175 | } else { | |
176 | z = ZONE_HIGHMEM; | |
177 | rend = cend; | |
178 | } | |
179 | zones[z] += rend - cstart; | |
180 | cstart = rend; | |
181 | } | |
182 | } | |
183 | ||
184 | /* | |
185 | * The SRAT table always lists ascending addresses, so can always | |
186 | * assume that the first "start" address that you see is the real | |
187 | * start of the node, and that the current "end" address is after | |
188 | * the previous one. | |
189 | */ | |
190 | static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk) | |
191 | { | |
192 | /* | |
193 | * Only add present memory as told by the e820. | |
194 | * There is no guarantee from the SRAT that the memory it | |
195 | * enumerates is present at boot time because it represents | |
196 | * *possible* memory hotplug areas the same as normal RAM. | |
197 | */ | |
198 | if (memory_chunk->start_pfn >= max_pfn) { | |
199 | printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n", | |
200 | memory_chunk->start_pfn, memory_chunk->end_pfn); | |
201 | return; | |
202 | } | |
203 | if (memory_chunk->nid != nid) | |
204 | return; | |
205 | ||
206 | if (!node_has_online_mem(nid)) | |
207 | node_start_pfn[nid] = memory_chunk->start_pfn; | |
208 | ||
209 | if (node_start_pfn[nid] > memory_chunk->start_pfn) | |
210 | node_start_pfn[nid] = memory_chunk->start_pfn; | |
211 | ||
212 | if (node_end_pfn[nid] < memory_chunk->end_pfn) | |
213 | node_end_pfn[nid] = memory_chunk->end_pfn; | |
214 | } | |
215 | ||
69e1a33f AK |
216 | static u8 pxm_to_nid_map[MAX_PXM_DOMAINS];/* _PXM to logical node ID map */ |
217 | ||
218 | int pxm_to_node(int pxm) | |
219 | { | |
220 | return pxm_to_nid_map[pxm]; | |
221 | } | |
222 | ||
1da177e4 LT |
223 | /* Parse the ACPI Static Resource Affinity Table */ |
224 | static int __init acpi20_parse_srat(struct acpi_table_srat *sratp) | |
225 | { | |
226 | u8 *start, *end, *p; | |
227 | int i, j, nid; | |
1da177e4 LT |
228 | u8 nid_to_pxm_map[MAX_NUMNODES];/* logical node ID to _PXM map */ |
229 | ||
230 | start = (u8 *)(&(sratp->reserved) + 1); /* skip header */ | |
231 | p = start; | |
232 | end = (u8 *)sratp + sratp->header.length; | |
233 | ||
234 | memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */ | |
235 | memset(node_memory_chunk, 0, sizeof(node_memory_chunk)); | |
236 | memset(zholes_size, 0, sizeof(zholes_size)); | |
237 | ||
238 | /* -1 in these maps means not available */ | |
239 | memset(pxm_to_nid_map, -1, sizeof(pxm_to_nid_map)); | |
240 | memset(nid_to_pxm_map, -1, sizeof(nid_to_pxm_map)); | |
241 | ||
242 | num_memory_chunks = 0; | |
243 | while (p < end) { | |
244 | switch (*p) { | |
245 | case ACPI_SRAT_PROCESSOR_AFFINITY: | |
246 | parse_cpu_affinity_structure(p); | |
247 | break; | |
248 | case ACPI_SRAT_MEMORY_AFFINITY: | |
249 | parse_memory_affinity_structure(p); | |
250 | break; | |
251 | default: | |
252 | printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]); | |
253 | break; | |
254 | } | |
255 | p += p[1]; | |
256 | if (p[1] == 0) { | |
257 | printk("acpi20_parse_srat: Entry length value is zero;" | |
258 | " can't parse any further!\n"); | |
259 | break; | |
260 | } | |
261 | } | |
262 | ||
263 | if (num_memory_chunks == 0) { | |
264 | printk("could not finy any ACPI SRAT memory areas.\n"); | |
265 | goto out_fail; | |
266 | } | |
267 | ||
268 | /* Calculate total number of nodes in system from PXM bitmap and create | |
269 | * a set of sequential node IDs starting at zero. (ACPI doesn't seem | |
270 | * to specify the range of _PXM values.) | |
271 | */ | |
272 | /* | |
273 | * MCD - we no longer HAVE to number nodes sequentially. PXM domain | |
274 | * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically | |
275 | * 32, so we will continue numbering them in this manner until MAX_NUMNODES | |
276 | * approaches MAX_PXM_DOMAINS for i386. | |
277 | */ | |
278 | nodes_clear(node_online_map); | |
279 | for (i = 0; i < MAX_PXM_DOMAINS; i++) { | |
280 | if (BMAP_TEST(pxm_bitmap, i)) { | |
281 | nid = num_online_nodes(); | |
282 | pxm_to_nid_map[i] = nid; | |
283 | nid_to_pxm_map[nid] = i; | |
284 | node_set_online(nid); | |
285 | } | |
286 | } | |
287 | BUG_ON(num_online_nodes() == 0); | |
288 | ||
289 | /* set cnode id in memory chunk structure */ | |
290 | for (i = 0; i < num_memory_chunks; i++) | |
291 | node_memory_chunk[i].nid = pxm_to_nid_map[node_memory_chunk[i].pxm]; | |
292 | ||
293 | printk("pxm bitmap: "); | |
294 | for (i = 0; i < sizeof(pxm_bitmap); i++) { | |
295 | printk("%02X ", pxm_bitmap[i]); | |
296 | } | |
297 | printk("\n"); | |
298 | printk("Number of logical nodes in system = %d\n", num_online_nodes()); | |
299 | printk("Number of memory chunks in system = %d\n", num_memory_chunks); | |
300 | ||
301 | for (j = 0; j < num_memory_chunks; j++){ | |
302 | struct node_memory_chunk_s * chunk = &node_memory_chunk[j]; | |
303 | printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n", | |
304 | j, chunk->nid, chunk->start_pfn, chunk->end_pfn); | |
305 | node_read_chunk(chunk->nid, chunk); | |
306 | } | |
307 | ||
308 | for_each_online_node(nid) { | |
309 | unsigned long start = node_start_pfn[nid]; | |
310 | unsigned long end = node_end_pfn[nid]; | |
311 | ||
312 | memory_present(nid, start, end); | |
313 | node_remap_size[nid] = node_memmap_size_bytes(nid, start, end); | |
314 | } | |
315 | return 1; | |
316 | out_fail: | |
317 | return 0; | |
318 | } | |
319 | ||
320 | int __init get_memcfg_from_srat(void) | |
321 | { | |
322 | struct acpi_table_header *header = NULL; | |
323 | struct acpi_table_rsdp *rsdp = NULL; | |
324 | struct acpi_table_rsdt *rsdt = NULL; | |
325 | struct acpi_pointer *rsdp_address = NULL; | |
326 | struct acpi_table_rsdt saved_rsdt; | |
327 | int tables = 0; | |
328 | int i = 0; | |
329 | ||
330 | acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING, rsdp_address); | |
331 | ||
332 | if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) { | |
333 | printk("%s: assigning address to rsdp\n", __FUNCTION__); | |
334 | rsdp = (struct acpi_table_rsdp *) | |
335 | (u32)rsdp_address->pointer.physical; | |
336 | } else { | |
337 | printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__); | |
338 | goto out_err; | |
339 | } | |
340 | if (!rsdp) { | |
341 | printk("%s: Didn't find ACPI root!\n", __FUNCTION__); | |
342 | goto out_err; | |
343 | } | |
344 | ||
345 | printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision, | |
346 | rsdp->oem_id); | |
347 | ||
348 | if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) { | |
349 | printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__); | |
350 | goto out_err; | |
351 | } | |
352 | ||
353 | rsdt = (struct acpi_table_rsdt *) | |
354 | boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt)); | |
355 | ||
356 | if (!rsdt) { | |
357 | printk(KERN_WARNING | |
358 | "%s: ACPI: Invalid root system description tables (RSDT)\n", | |
359 | __FUNCTION__); | |
360 | goto out_err; | |
361 | } | |
362 | ||
363 | header = & rsdt->header; | |
364 | ||
365 | if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) { | |
366 | printk(KERN_WARNING "ACPI: RSDT signature incorrect\n"); | |
367 | goto out_err; | |
368 | } | |
369 | ||
370 | /* | |
371 | * The number of tables is computed by taking the | |
372 | * size of all entries (header size minus total | |
373 | * size of RSDT) divided by the size of each entry | |
374 | * (4-byte table pointers). | |
375 | */ | |
376 | tables = (header->length - sizeof(struct acpi_table_header)) / 4; | |
377 | ||
378 | if (!tables) | |
379 | goto out_err; | |
380 | ||
381 | memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt)); | |
382 | ||
383 | if (saved_rsdt.header.length > sizeof(saved_rsdt)) { | |
384 | printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n", | |
385 | saved_rsdt.header.length); | |
386 | goto out_err; | |
387 | } | |
388 | ||
389 | printk("Begin SRAT table scan....\n"); | |
390 | ||
391 | for (i = 0; i < tables; i++) { | |
392 | /* Map in header, then map in full table length. */ | |
393 | header = (struct acpi_table_header *) | |
394 | boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header)); | |
395 | if (!header) | |
396 | break; | |
397 | header = (struct acpi_table_header *) | |
398 | boot_ioremap(saved_rsdt.entry[i], header->length); | |
399 | if (!header) | |
400 | break; | |
401 | ||
402 | if (strncmp((char *) &header->signature, "SRAT", 4)) | |
403 | continue; | |
404 | ||
405 | /* we've found the srat table. don't need to look at any more tables */ | |
406 | return acpi20_parse_srat((struct acpi_table_srat *)header); | |
407 | } | |
408 | out_err: | |
409 | printk("failed to get NUMA memory information from SRAT table\n"); | |
410 | return 0; | |
411 | } | |
412 | ||
413 | /* For each node run the memory list to determine whether there are | |
414 | * any memory holes. For each hole determine which ZONE they fall | |
415 | * into. | |
416 | * | |
417 | * NOTE#1: this requires knowledge of the zone boundries and so | |
418 | * _cannot_ be performed before those are calculated in setup_memory. | |
419 | * | |
420 | * NOTE#2: we rely on the fact that the memory chunks are ordered by | |
421 | * start pfn number during setup. | |
422 | */ | |
423 | static void __init get_zholes_init(void) | |
424 | { | |
425 | int nid; | |
426 | int c; | |
427 | int first; | |
428 | unsigned long end = 0; | |
429 | ||
430 | for_each_online_node(nid) { | |
431 | first = 1; | |
432 | for (c = 0; c < num_memory_chunks; c++){ | |
433 | if (node_memory_chunk[c].nid == nid) { | |
434 | if (first) { | |
435 | end = node_memory_chunk[c].end_pfn; | |
436 | first = 0; | |
437 | ||
438 | } else { | |
439 | /* Record any gap between this chunk | |
440 | * and the previous chunk on this node | |
441 | * against the zones it spans. | |
442 | */ | |
443 | chunk_to_zones(end, | |
444 | node_memory_chunk[c].start_pfn, | |
445 | &zholes_size[nid * MAX_NR_ZONES]); | |
446 | } | |
447 | } | |
448 | } | |
449 | } | |
450 | } | |
451 | ||
452 | unsigned long * __init get_zholes_size(int nid) | |
453 | { | |
454 | if (!zholes_size_init) { | |
455 | zholes_size_init++; | |
456 | get_zholes_init(); | |
457 | } | |
458 | if (nid >= MAX_NUMNODES || !node_online(nid)) | |
459 | printk("%s: nid = %d is invalid/offline. num_online_nodes = %d", | |
460 | __FUNCTION__, nid, num_online_nodes()); | |
461 | return &zholes_size[nid * MAX_NR_ZONES]; | |
462 | } |