blob: 8786268053693cac4ed3114becc62a488a905999 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved.
3 * Copyright (c) 2001 Intel Corp.
4 * Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
5 * Copyright (c) 2002 NEC Corp.
6 * Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
7 * Copyright (c) 2004 Silicon Graphics, Inc
8 * Russ Anderson <rja@sgi.com>
9 * Jesse Barnes <jbarnes@sgi.com>
10 * Jack Steiner <steiner@sgi.com>
11 */
12
13/*
14 * Platform initialization for Discontig Memory
15 */
16
17#include <linux/kernel.h>
18#include <linux/mm.h>
Prarit Bhargava99a19cf2007-08-22 14:34:38 -040019#include <linux/nmi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/swap.h>
21#include <linux/bootmem.h>
22#include <linux/acpi.h>
23#include <linux/efi.h>
24#include <linux/nodemask.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/pgalloc.h>
27#include <asm/tlb.h>
28#include <asm/meminit.h>
29#include <asm/numa.h>
30#include <asm/sections.h>
31
32/*
33 * Track per-node information needed to setup the boot memory allocator, the
34 * per-node areas, and the real VM.
35 */
36struct early_node_data {
37 struct ia64_node_data *node_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 unsigned long pernode_addr;
39 unsigned long pernode_size;
Christoph Lameter09ae1f52007-02-10 01:43:11 -080040#ifdef CONFIG_ZONE_DMA
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 unsigned long num_dma_physpages;
Christoph Lameter09ae1f52007-02-10 01:43:11 -080042#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 unsigned long min_pfn;
44 unsigned long max_pfn;
45};
46
47static struct early_node_data mem_data[MAX_NUMNODES] __initdata;
bob.picco564601a2005-06-30 09:52:00 -070048static nodemask_t memory_less_mask __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -070050pg_data_t *pgdat_list[MAX_NUMNODES];
Yasunori Gotoae5a2c12006-06-27 02:53:38 -070051
Linus Torvalds1da177e2005-04-16 15:20:36 -070052/*
53 * To prevent cache aliasing effects, align per-node structures so that they
54 * start at addresses that are strided by node number.
55 */
Jack Steineracb7f672005-12-05 13:56:50 -060056#define MAX_NODE_ALIGN_OFFSET (32 * 1024 * 1024)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#define NODEDATA_ALIGN(addr, node) \
Jack Steineracb7f672005-12-05 13:56:50 -060058 ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + \
59 (((node)*PERCPU_PAGE_SIZE) & (MAX_NODE_ALIGN_OFFSET - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/**
62 * build_node_maps - callback to setup bootmem structs for each node
63 * @start: physical start of range
64 * @len: length of range
65 * @node: node where this range resides
66 *
67 * We allocate a struct bootmem_data for each piece of memory that we wish to
68 * treat as a virtually contiguous block (i.e. each node). Each such block
69 * must start on an %IA64_GRANULE_SIZE boundary, so we round the address down
70 * if necessary. Any non-existent pages will simply be part of the virtual
71 * memmap. We also update min_low_pfn and max_low_pfn here as we receive
72 * memory ranges from the caller.
73 */
74static int __init build_node_maps(unsigned long start, unsigned long len,
75 int node)
76{
Johannes Weiner3560e242008-07-23 21:28:09 -070077 unsigned long spfn, epfn, end = start + len;
Johannes Weinerb61bfa32008-07-23 21:26:55 -070078 struct bootmem_data *bdp = &bootmem_node_data[node];
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80 epfn = GRANULEROUNDUP(end) >> PAGE_SHIFT;
Johannes Weiner3560e242008-07-23 21:28:09 -070081 spfn = GRANULEROUNDDOWN(start) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83 if (!bdp->node_low_pfn) {
Johannes Weiner3560e242008-07-23 21:28:09 -070084 bdp->node_min_pfn = spfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 bdp->node_low_pfn = epfn;
86 } else {
Johannes Weiner3560e242008-07-23 21:28:09 -070087 bdp->node_min_pfn = min(spfn, bdp->node_min_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 bdp->node_low_pfn = max(epfn, bdp->node_low_pfn);
89 }
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 return 0;
92}
93
94/**
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * early_nr_cpus_node - return number of cpus on a given node
96 * @node: node to check
97 *
98 * Count the number of cpus on @node. We can't use nr_cpus_node() yet because
99 * acpi_boot_init() (which builds the node_to_cpu_mask array) hasn't been
100 * called yet. Note that node 0 will also count all non-existent cpus.
101 */
Yasunori Gotodd0932d2006-06-27 02:53:40 -0700102static int __meminit early_nr_cpus_node(int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
104 int cpu, n = 0;
105
holt@sgi.com2c6e6db2008-04-03 15:17:13 -0500106 for_each_possible_early_cpu(cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 if (node == node_cpuid[cpu].nid)
108 n++;
109
110 return n;
111}
112
113/**
bob.picco564601a2005-06-30 09:52:00 -0700114 * compute_pernodesize - compute size of pernode data
115 * @node: the node id.
116 */
Yasunori Gotodd0932d2006-06-27 02:53:40 -0700117static unsigned long __meminit compute_pernodesize(int node)
bob.picco564601a2005-06-30 09:52:00 -0700118{
119 unsigned long pernodesize = 0, cpus;
120
121 cpus = early_nr_cpus_node(node);
122 pernodesize += PERCPU_PAGE_SIZE * cpus;
123 pernodesize += node * L1_CACHE_BYTES;
124 pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
125 pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
holt@sgi.com41bd26d2008-04-03 15:17:12 -0500126 pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
bob.picco564601a2005-06-30 09:52:00 -0700127 pernodesize = PAGE_ALIGN(pernodesize);
128 return pernodesize;
129}
130
131/**
Tony Luck8d7e3512005-07-06 18:18:10 -0700132 * per_cpu_node_setup - setup per-cpu areas on each node
133 * @cpu_data: per-cpu area on this node
134 * @node: node to setup
135 *
136 * Copy the static per-cpu data into the region we just set aside and then
137 * setup __per_cpu_offset for each CPU on this node. Return a pointer to
138 * the end of the area.
139 */
140static void *per_cpu_node_setup(void *cpu_data, int node)
141{
142#ifdef CONFIG_SMP
143 int cpu;
144
holt@sgi.com2c6e6db2008-04-03 15:17:13 -0500145 for_each_possible_early_cpu(cpu) {
Tejun Heo36886472009-10-02 13:28:56 +0900146 void *src = cpu == 0 ? __cpu0_per_cpu : __phys_per_cpu_start;
147
148 if (node != node_cpuid[cpu].nid)
149 continue;
150
151 memcpy(__va(cpu_data), src, __per_cpu_end - __per_cpu_start);
152 __per_cpu_offset[cpu] = (char *)__va(cpu_data) -
153 __per_cpu_start;
154
155 /*
156 * percpu area for cpu0 is moved from the __init area
157 * which is setup by head.S and used till this point.
158 * Update ar.k3. This move is ensures that percpu
159 * area for cpu0 is on the correct node and its
160 * virtual address isn't insanely far from other
161 * percpu areas which is important for congruent
162 * percpu allocator.
163 */
164 if (cpu == 0)
165 ia64_set_kr(IA64_KR_PER_CPU_DATA,
166 (unsigned long)cpu_data -
167 (unsigned long)__per_cpu_start);
168
169 cpu_data += PERCPU_PAGE_SIZE;
Tony Luck8d7e3512005-07-06 18:18:10 -0700170 }
171#endif
172 return cpu_data;
173}
174
Tejun Heo525947622009-10-02 13:28:56 +0900175#ifdef CONFIG_SMP
176/**
177 * setup_per_cpu_areas - setup percpu areas
178 *
179 * Arch code has already allocated and initialized percpu areas. All
180 * this function has to do is to teach the determined layout to the
181 * dynamic percpu allocator, which happens to be more complex than
182 * creating whole new ones using helpers.
183 */
184void __init setup_per_cpu_areas(void)
185{
186 struct pcpu_alloc_info *ai;
187 struct pcpu_group_info *uninitialized_var(gi);
188 unsigned int *cpu_map;
189 void *base;
190 unsigned long base_offset;
191 unsigned int cpu;
192 ssize_t static_size, reserved_size, dyn_size;
193 int node, prev_node, unit, nr_units, rc;
194
195 ai = pcpu_alloc_alloc_info(MAX_NUMNODES, nr_cpu_ids);
196 if (!ai)
197 panic("failed to allocate pcpu_alloc_info");
198 cpu_map = ai->groups[0].cpu_map;
199
200 /* determine base */
201 base = (void *)ULONG_MAX;
202 for_each_possible_cpu(cpu)
203 base = min(base,
204 (void *)(__per_cpu_offset[cpu] + __per_cpu_start));
205 base_offset = (void *)__per_cpu_start - base;
206
207 /* build cpu_map, units are grouped by node */
208 unit = 0;
209 for_each_node(node)
210 for_each_possible_cpu(cpu)
211 if (node == node_cpuid[cpu].nid)
212 cpu_map[unit++] = cpu;
213 nr_units = unit;
214
215 /* set basic parameters */
216 static_size = __per_cpu_end - __per_cpu_start;
217 reserved_size = PERCPU_MODULE_RESERVE;
218 dyn_size = PERCPU_PAGE_SIZE - static_size - reserved_size;
219 if (dyn_size < 0)
220 panic("percpu area overflow static=%zd reserved=%zd\n",
221 static_size, reserved_size);
222
223 ai->static_size = static_size;
224 ai->reserved_size = reserved_size;
225 ai->dyn_size = dyn_size;
226 ai->unit_size = PERCPU_PAGE_SIZE;
227 ai->atom_size = PAGE_SIZE;
228 ai->alloc_size = PERCPU_PAGE_SIZE;
229
230 /*
231 * CPUs are put into groups according to node. Walk cpu_map
232 * and create new groups at node boundaries.
233 */
234 prev_node = -1;
235 ai->nr_groups = 0;
236 for (unit = 0; unit < nr_units; unit++) {
237 cpu = cpu_map[unit];
238 node = node_cpuid[cpu].nid;
239
240 if (node == prev_node) {
241 gi->nr_units++;
242 continue;
243 }
244 prev_node = node;
245
246 gi = &ai->groups[ai->nr_groups++];
247 gi->nr_units = 1;
248 gi->base_offset = __per_cpu_offset[cpu] + base_offset;
249 gi->cpu_map = &cpu_map[unit];
250 }
251
252 rc = pcpu_setup_first_chunk(ai, base);
253 if (rc)
254 panic("failed to setup percpu area (err=%d)", rc);
255
256 pcpu_free_alloc_info(ai);
257}
258#endif
259
Tony Luck8d7e3512005-07-06 18:18:10 -0700260/**
bob.picco564601a2005-06-30 09:52:00 -0700261 * fill_pernode - initialize pernode data.
262 * @node: the node id.
263 * @pernode: physical address of pernode data
264 * @pernodesize: size of the pernode data
265 */
266static void __init fill_pernode(int node, unsigned long pernode,
267 unsigned long pernodesize)
268{
269 void *cpu_data;
Tony Luck8d7e3512005-07-06 18:18:10 -0700270 int cpus = early_nr_cpus_node(node);
Johannes Weinerb61bfa32008-07-23 21:26:55 -0700271 struct bootmem_data *bdp = &bootmem_node_data[node];
bob.picco564601a2005-06-30 09:52:00 -0700272
273 mem_data[node].pernode_addr = pernode;
274 mem_data[node].pernode_size = pernodesize;
275 memset(__va(pernode), 0, pernodesize);
276
277 cpu_data = (void *)pernode;
278 pernode += PERCPU_PAGE_SIZE * cpus;
279 pernode += node * L1_CACHE_BYTES;
280
Yasunori Gotoae5a2c12006-06-27 02:53:38 -0700281 pgdat_list[node] = __va(pernode);
bob.picco564601a2005-06-30 09:52:00 -0700282 pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
283
284 mem_data[node].node_data = __va(pernode);
285 pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
286
Yasunori Gotoae5a2c12006-06-27 02:53:38 -0700287 pgdat_list[node]->bdata = bdp;
bob.picco564601a2005-06-30 09:52:00 -0700288 pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
289
Tony Luck8d7e3512005-07-06 18:18:10 -0700290 cpu_data = per_cpu_node_setup(cpu_data, node);
bob.picco564601a2005-06-30 09:52:00 -0700291
292 return;
293}
Tony Luck8d7e3512005-07-06 18:18:10 -0700294
bob.picco564601a2005-06-30 09:52:00 -0700295/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 * find_pernode_space - allocate memory for memory map and per-node structures
297 * @start: physical start of range
298 * @len: length of range
299 * @node: node where this range resides
300 *
301 * This routine reserves space for the per-cpu data struct, the list of
302 * pg_data_ts and the per-node data struct. Each node will have something like
303 * the following in the first chunk of addr. space large enough to hold it.
304 *
305 * ________________________
306 * | |
307 * |~~~~~~~~~~~~~~~~~~~~~~~~| <-- NODEDATA_ALIGN(start, node) for the first
308 * | PERCPU_PAGE_SIZE * | start and length big enough
309 * | cpus_on_this_node | Node 0 will also have entries for all non-existent cpus.
310 * |------------------------|
311 * | local pg_data_t * |
312 * |------------------------|
313 * | local ia64_node_data |
314 * |------------------------|
315 * | ??? |
316 * |________________________|
317 *
318 * Once this space has been set aside, the bootmem maps are initialized. We
319 * could probably move the allocation of the per-cpu and ia64_node_data space
320 * outside of this function and use alloc_bootmem_node(), but doing it here
321 * is straightforward and we get the alignments we want so...
322 */
323static int __init find_pernode_space(unsigned long start, unsigned long len,
324 int node)
325{
Johannes Weiner3560e242008-07-23 21:28:09 -0700326 unsigned long spfn, epfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 unsigned long pernodesize = 0, pernode, pages, mapsize;
Johannes Weinerb61bfa32008-07-23 21:26:55 -0700328 struct bootmem_data *bdp = &bootmem_node_data[node];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Johannes Weiner3560e242008-07-23 21:28:09 -0700330 spfn = start >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 epfn = (start + len) >> PAGE_SHIFT;
332
Johannes Weiner3560e242008-07-23 21:28:09 -0700333 pages = bdp->node_low_pfn - bdp->node_min_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
335
336 /*
337 * Make sure this memory falls within this node's usable memory
338 * since we may have thrown some away in build_maps().
339 */
Johannes Weiner3560e242008-07-23 21:28:09 -0700340 if (spfn < bdp->node_min_pfn || epfn > bdp->node_low_pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return 0;
342
343 /* Don't setup this node's local space twice... */
344 if (mem_data[node].pernode_addr)
345 return 0;
346
347 /*
348 * Calculate total size needed, incl. what's necessary
349 * for good alignment and alias prevention.
350 */
bob.picco564601a2005-06-30 09:52:00 -0700351 pernodesize = compute_pernodesize(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 pernode = NODEDATA_ALIGN(start, node);
353
354 /* Is this range big enough for what we want to store here? */
bob.picco564601a2005-06-30 09:52:00 -0700355 if (start + len > (pernode + pernodesize + mapsize))
356 fill_pernode(node, pernode, pernodesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 return 0;
359}
360
361/**
362 * free_node_bootmem - free bootmem allocator memory for use
363 * @start: physical start of range
364 * @len: length of range
365 * @node: node where this range resides
366 *
367 * Simply calls the bootmem allocator to free the specified ranged from
368 * the given pg_data_t's bdata struct. After this function has been called
369 * for all the entries in the EFI memory map, the bootmem allocator will
370 * be ready to service allocation requests.
371 */
372static int __init free_node_bootmem(unsigned long start, unsigned long len,
373 int node)
374{
Yasunori Gotoae5a2c12006-06-27 02:53:38 -0700375 free_bootmem_node(pgdat_list[node], start, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
377 return 0;
378}
379
380/**
381 * reserve_pernode_space - reserve memory for per-node space
382 *
383 * Reserve the space used by the bootmem maps & per-node space in the boot
384 * allocator so that when we actually create the real mem maps we don't
385 * use their memory.
386 */
387static void __init reserve_pernode_space(void)
388{
389 unsigned long base, size, pages;
390 struct bootmem_data *bdp;
391 int node;
392
393 for_each_online_node(node) {
Yasunori Gotoae5a2c12006-06-27 02:53:38 -0700394 pg_data_t *pdp = pgdat_list[node];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
bob.picco564601a2005-06-30 09:52:00 -0700396 if (node_isset(node, memory_less_mask))
397 continue;
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 bdp = pdp->bdata;
400
401 /* First the bootmem_map itself */
Johannes Weiner3560e242008-07-23 21:28:09 -0700402 pages = bdp->node_low_pfn - bdp->node_min_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
404 base = __pa(bdp->node_bootmem_map);
Bernhard Walle72a7fe32008-02-07 00:15:17 -0800405 reserve_bootmem_node(pdp, base, size, BOOTMEM_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 /* Now the per-node space */
408 size = mem_data[node].pernode_size;
409 base = __pa(mem_data[node].pernode_addr);
Bernhard Walle72a7fe32008-02-07 00:15:17 -0800410 reserve_bootmem_node(pdp, base, size, BOOTMEM_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 }
412}
413
Yasunori Goto70490272006-06-27 02:53:39 -0700414static void __meminit scatter_node_data(void)
415{
416 pg_data_t **dst;
417 int node;
418
Yasunori Gotodd8041f2006-07-04 02:57:51 -0700419 /*
420 * for_each_online_node() can't be used at here.
421 * node_online_map is not set for hot-added nodes at this time,
422 * because we are halfway through initialization of the new node's
423 * structures. If for_each_online_node() is used, a new node's
Simon Arlott72fdbdc2007-05-11 14:55:43 -0700424 * pg_data_ptrs will be not initialized. Instead of using it,
Yasunori Gotodd8041f2006-07-04 02:57:51 -0700425 * pgdat_list[] is checked.
426 */
427 for_each_node(node) {
428 if (pgdat_list[node]) {
429 dst = LOCAL_DATA_ADDR(pgdat_list[node])->pg_data_ptrs;
430 memcpy(dst, pgdat_list, sizeof(pgdat_list));
431 }
Yasunori Goto70490272006-06-27 02:53:39 -0700432 }
433}
434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435/**
436 * initialize_pernode_data - fixup per-cpu & per-node pointers
437 *
438 * Each node's per-node area has a copy of the global pg_data_t list, so
439 * we copy that to each node here, as well as setting the per-cpu pointer
440 * to the local node data structure. The active_cpus field of the per-node
441 * structure gets setup by the platform_cpu_init() function later.
442 */
443static void __init initialize_pernode_data(void)
444{
Tony Luck8d7e3512005-07-06 18:18:10 -0700445 int cpu, node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Yasunori Goto70490272006-06-27 02:53:39 -0700447 scatter_node_data();
448
Tony Luck8d7e3512005-07-06 18:18:10 -0700449#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 /* Set the node_data pointer for each per-cpu struct */
holt@sgi.com2c6e6db2008-04-03 15:17:13 -0500451 for_each_possible_early_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 node = node_cpuid[cpu].nid;
Tejun Heo877105c2009-10-29 22:34:14 +0900453 per_cpu(ia64_cpu_info, cpu).node_data =
454 mem_data[node].node_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Tony Luck8d7e3512005-07-06 18:18:10 -0700456#else
457 {
458 struct cpuinfo_ia64 *cpu0_cpu_info;
459 cpu = 0;
460 node = node_cpuid[cpu].nid;
461 cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start +
Rusty Russelldd17c8f2009-10-29 22:34:15 +0900462 ((char *)&ia64_cpu_info - __per_cpu_start));
Tony Luck8d7e3512005-07-06 18:18:10 -0700463 cpu0_cpu_info->node_data = mem_data[node].node_data;
464 }
465#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/**
bob.picco564601a2005-06-30 09:52:00 -0700469 * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit
470 * node but fall back to any other node when __alloc_bootmem_node fails
471 * for best.
472 * @nid: node id
473 * @pernodesize: size of this node's pernode data
bob.picco564601a2005-06-30 09:52:00 -0700474 */
Bob Picco97835242005-10-29 17:23:05 -0400475static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize)
bob.picco564601a2005-06-30 09:52:00 -0700476{
477 void *ptr = NULL;
478 u8 best = 0xff;
Bob Picco97835242005-10-29 17:23:05 -0400479 int bestnode = -1, node, anynode = 0;
bob.picco564601a2005-06-30 09:52:00 -0700480
481 for_each_online_node(node) {
482 if (node_isset(node, memory_less_mask))
483 continue;
484 else if (node_distance(nid, node) < best) {
485 best = node_distance(nid, node);
486 bestnode = node;
487 }
Bob Picco97835242005-10-29 17:23:05 -0400488 anynode = node;
bob.picco564601a2005-06-30 09:52:00 -0700489 }
490
Bob Picco97835242005-10-29 17:23:05 -0400491 if (bestnode == -1)
492 bestnode = anynode;
bob.picco564601a2005-06-30 09:52:00 -0700493
Yasunori Gotoae5a2c12006-06-27 02:53:38 -0700494 ptr = __alloc_bootmem_node(pgdat_list[bestnode], pernodesize,
Bob Picco97835242005-10-29 17:23:05 -0400495 PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
496
bob.picco564601a2005-06-30 09:52:00 -0700497 return ptr;
498}
499
500/**
bob.picco564601a2005-06-30 09:52:00 -0700501 * memory_less_nodes - allocate and initialize CPU only nodes pernode
502 * information.
503 */
504static void __init memory_less_nodes(void)
505{
506 unsigned long pernodesize;
507 void *pernode;
508 int node;
509
510 for_each_node_mask(node, memory_less_mask) {
511 pernodesize = compute_pernodesize(node);
Bob Picco97835242005-10-29 17:23:05 -0400512 pernode = memory_less_node_alloc(node, pernodesize);
bob.picco564601a2005-06-30 09:52:00 -0700513 fill_pernode(node, __pa(pernode), pernodesize);
514 }
515
516 return;
517}
518
519/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 * find_memory - walk the EFI memory map and setup the bootmem allocator
521 *
522 * Called early in boot to setup the bootmem allocator, and to
523 * allocate the per-cpu and per-node structures.
524 */
525void __init find_memory(void)
526{
527 int node;
528
529 reserve_memory();
530
531 if (num_online_nodes() == 0) {
532 printk(KERN_ERR "node info missing!\n");
533 node_set_online(0);
534 }
535
bob.picco564601a2005-06-30 09:52:00 -0700536 nodes_or(memory_less_mask, memory_less_mask, node_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 min_low_pfn = -1;
538 max_low_pfn = 0;
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 /* These actually end up getting called by call_pernode_memory() */
541 efi_memmap_walk(filter_rsvd_memory, build_node_maps);
542 efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
Zou Nan haia3f5c332007-03-20 13:41:57 -0700543 efi_memmap_walk(find_max_min_low_pfn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
bob.picco564601a2005-06-30 09:52:00 -0700545 for_each_online_node(node)
Johannes Weinerb61bfa32008-07-23 21:26:55 -0700546 if (bootmem_node_data[node].node_low_pfn) {
bob.picco564601a2005-06-30 09:52:00 -0700547 node_clear(node, memory_less_mask);
548 mem_data[node].min_pfn = ~0UL;
549 }
Bob Picco139b8302007-01-30 02:11:09 -0800550
Zoltan Menyhart98075d22008-04-11 15:21:35 -0700551 efi_memmap_walk(filter_memory, register_active_ranges);
Bob Picco139b8302007-01-30 02:11:09 -0800552
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 /*
554 * Initialize the boot memory maps in reverse order since that's
555 * what the bootmem allocator expects
556 */
557 for (node = MAX_NUMNODES - 1; node >= 0; node--) {
558 unsigned long pernode, pernodesize, map;
559 struct bootmem_data *bdp;
560
561 if (!node_online(node))
562 continue;
bob.picco564601a2005-06-30 09:52:00 -0700563 else if (node_isset(node, memory_less_mask))
564 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Johannes Weinerb61bfa32008-07-23 21:26:55 -0700566 bdp = &bootmem_node_data[node];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 pernode = mem_data[node].pernode_addr;
568 pernodesize = mem_data[node].pernode_size;
569 map = pernode + pernodesize;
570
Yasunori Gotoae5a2c12006-06-27 02:53:38 -0700571 init_bootmem_node(pgdat_list[node],
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 map>>PAGE_SHIFT,
Johannes Weiner3560e242008-07-23 21:28:09 -0700573 bdp->node_min_pfn,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 bdp->node_low_pfn);
575 }
576
577 efi_memmap_walk(filter_rsvd_memory, free_node_bootmem);
578
579 reserve_pernode_space();
bob.picco564601a2005-06-30 09:52:00 -0700580 memory_less_nodes();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 initialize_pernode_data();
582
583 max_pfn = max_low_pfn;
584
585 find_initrd();
586}
587
Tony Luck8d7e3512005-07-06 18:18:10 -0700588#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589/**
590 * per_cpu_init - setup per-cpu variables
591 *
592 * find_pernode_space() does most of this already, we just need to set
593 * local_per_cpu_offset
594 */
Paul Gortmakerccce9bb2013-06-17 15:51:20 -0400595void *per_cpu_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
597 int cpu;
Ashok Rajff741902005-11-11 14:32:40 -0800598 static int first_time = 1;
599
Ashok Rajff741902005-11-11 14:32:40 -0800600 if (first_time) {
601 first_time = 0;
holt@sgi.com2c6e6db2008-04-03 15:17:13 -0500602 for_each_possible_early_cpu(cpu)
Ashok Rajff741902005-11-11 14:32:40 -0800603 per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 return __per_cpu_start + __per_cpu_offset[smp_processor_id()];
607}
Tony Luck8d7e3512005-07-06 18:18:10 -0700608#endif /* CONFIG_SMP */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 * call_pernode_memory - use SRAT to call callback functions with node info
612 * @start: physical start of range
613 * @len: length of range
614 * @arg: function to call for each range
615 *
616 * efi_memmap_walk() knows nothing about layout of memory across nodes. Find
617 * out to which node a block of memory belongs. Ignore memory that we cannot
618 * identify, and split blocks that run across multiple nodes.
619 *
620 * Take this opportunity to round the start address up and the end address
621 * down to page boundaries.
622 */
623void call_pernode_memory(unsigned long start, unsigned long len, void *arg)
624{
625 unsigned long rs, re, end = start + len;
626 void (*func)(unsigned long, unsigned long, int);
627 int i;
628
629 start = PAGE_ALIGN(start);
630 end &= PAGE_MASK;
631 if (start >= end)
632 return;
633
634 func = arg;
635
636 if (!num_node_memblks) {
637 /* No SRAT table, so assume one node (node 0) */
638 if (start < end)
639 (*func)(start, end - start, 0);
640 return;
641 }
642
643 for (i = 0; i < num_node_memblks; i++) {
644 rs = max(start, node_memblk[i].start_paddr);
645 re = min(end, node_memblk[i].start_paddr +
646 node_memblk[i].size);
647
648 if (rs < re)
649 (*func)(rs, re - rs, node_memblk[i].nid);
650
651 if (re == end)
652 break;
653 }
654}
655
656/**
657 * count_node_pages - callback to build per-node memory info structures
658 * @start: physical start of range
659 * @len: length of range
660 * @node: node where this range resides
661 *
662 * Each node has it's own number of physical pages, DMAable pages, start, and
663 * end page frame number. This routine will be called by call_pernode_memory()
664 * for each piece of usable memory and will setup these values for each node.
665 * Very similar to build_maps().
666 */
667static __init int count_node_pages(unsigned long start, unsigned long len, int node)
668{
669 unsigned long end = start + len;
670
Christoph Lameter09ae1f52007-02-10 01:43:11 -0800671#ifdef CONFIG_ZONE_DMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (start <= __pa(MAX_DMA_ADDRESS))
673 mem_data[node].num_dma_physpages +=
674 (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT;
Christoph Lameter09ae1f52007-02-10 01:43:11 -0800675#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 start = GRANULEROUNDDOWN(start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 end = GRANULEROUNDUP(end);
678 mem_data[node].max_pfn = max(mem_data[node].max_pfn,
679 end >> PAGE_SHIFT);
680 mem_data[node].min_pfn = min(mem_data[node].min_pfn,
681 start >> PAGE_SHIFT);
682
683 return 0;
684}
685
686/**
687 * paging_init - setup page tables
688 *
689 * paging_init() sets up the page tables for each node of the system and frees
690 * the bootmem allocator memory for general use.
691 */
692void __init paging_init(void)
693{
694 unsigned long max_dma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 unsigned long pfn_offset = 0;
Mel Gorman05e0caa2006-09-27 01:49:54 -0700696 unsigned long max_pfn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 int node;
Mel Gorman05e0caa2006-09-27 01:49:54 -0700698 unsigned long max_zone_pfns[MAX_NR_ZONES];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700 max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 efi_memmap_walk(filter_rsvd_memory, count_node_pages);
703
Bob Picco524fd982007-02-05 16:20:08 -0800704 sparse_memory_present_with_active_regions(MAX_NUMNODES);
705 sparse_init();
706
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400707#ifdef CONFIG_VIRTUAL_MEM_MAP
Tejun Heo126b3fc2009-10-02 13:28:55 +0900708 VMALLOC_END -= PAGE_ALIGN(ALIGN(max_low_pfn, MAX_ORDER_NR_PAGES) *
Bob Picco921eea12006-06-28 12:54:55 -0400709 sizeof(struct page));
Tejun Heo126b3fc2009-10-02 13:28:55 +0900710 vmem_map = (struct page *) VMALLOC_END;
bob.picco564601a2005-06-30 09:52:00 -0700711 efi_memmap_walk(create_mem_map_page_table, NULL);
712 printk("Virtual mem_map starts at 0x%p\n", vmem_map);
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400713#endif
bob.picco564601a2005-06-30 09:52:00 -0700714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 for_each_online_node(node) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 pfn_offset = mem_data[node].min_pfn;
717
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400718#ifdef CONFIG_VIRTUAL_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
Bob Picco2d4b1fa2005-10-04 15:13:57 -0400720#endif
Mel Gorman05e0caa2006-09-27 01:49:54 -0700721 if (mem_data[node].max_pfn > max_pfn)
722 max_pfn = mem_data[node].max_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
724
Mel Gorman6391af12006-10-11 01:20:39 -0700725 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
Christoph Lameter09ae1f52007-02-10 01:43:11 -0800726#ifdef CONFIG_ZONE_DMA
Mel Gorman05e0caa2006-09-27 01:49:54 -0700727 max_zone_pfns[ZONE_DMA] = max_dma;
Christoph Lameter09ae1f52007-02-10 01:43:11 -0800728#endif
Mel Gorman05e0caa2006-09-27 01:49:54 -0700729 max_zone_pfns[ZONE_NORMAL] = max_pfn;
730 free_area_init_nodes(max_zone_pfns);
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
733}
Yasunori Goto70490272006-06-27 02:53:39 -0700734
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700735#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Gotodd0932d2006-06-27 02:53:40 -0700736pg_data_t *arch_alloc_nodedata(int nid)
737{
738 unsigned long size = compute_pernodesize(nid);
739
740 return kzalloc(size, GFP_KERNEL);
741}
742
743void arch_free_nodedata(pg_data_t *pgdat)
744{
745 kfree(pgdat);
746}
747
Yasunori Goto70490272006-06-27 02:53:39 -0700748void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat)
749{
750 pgdat_list[update_node] = update_pgdat;
751 scatter_node_data();
752}
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700753#endif
Christoph Lameteref229c52007-10-16 01:24:15 -0700754
755#ifdef CONFIG_SPARSEMEM_VMEMMAP
Johannes Weiner0aad8182013-04-29 15:07:50 -0700756int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
Christoph Lameteref229c52007-10-16 01:24:15 -0700757{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700758 return vmemmap_populate_basepages(start, end, node);
Christoph Lameteref229c52007-10-16 01:24:15 -0700759}
Yasuaki Ishimatsu46723bf2013-02-22 16:33:00 -0800760
Johannes Weiner0aad8182013-04-29 15:07:50 -0700761void vmemmap_free(unsigned long start, unsigned long end)
Tang Chen01975182013-02-22 16:33:08 -0800762{
763}
Christoph Lameteref229c52007-10-16 01:24:15 -0700764#endif