1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Author: Xiang Gao <gaoxiang@loongson.cn>
4 * Huacai Chen <chenhuacai@loongson.cn>
5 *
6 * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
7 */
8#include <linux/init.h>
9#include <linux/kernel.h>
10#include <linux/mm.h>
11#include <linux/mmzone.h>
12#include <linux/export.h>
13#include <linux/nodemask.h>
14#include <linux/swap.h>
15#include <linux/memblock.h>
16#include <linux/pfn.h>
17#include <linux/acpi.h>
18#include <linux/efi.h>
19#include <linux/irq.h>
20#include <linux/pci.h>
21#include <asm/bootinfo.h>
22#include <asm/loongson.h>
23#include <asm/numa.h>
24#include <asm/page.h>
25#include <asm/pgalloc.h>
26#include <asm/sections.h>
27#include <asm/time.h>
28
29int numa_off;
30struct pglist_data *node_data[MAX_NUMNODES];
31unsigned char node_distances[MAX_NUMNODES][MAX_NUMNODES];
32
33EXPORT_SYMBOL(node_data);
34EXPORT_SYMBOL(node_distances);
35
36static struct numa_meminfo numa_meminfo;
37cpumask_t cpus_on_node[MAX_NUMNODES];
38cpumask_t phys_cpus_on_node[MAX_NUMNODES];
39EXPORT_SYMBOL(cpus_on_node);
40
41/*
42 * apicid, cpu, node mappings
43 */
44s16 __cpuid_to_node[CONFIG_NR_CPUS] = {
45 [0 ... CONFIG_NR_CPUS - 1] = NUMA_NO_NODE
46};
47EXPORT_SYMBOL(__cpuid_to_node);
48
49nodemask_t numa_nodes_parsed __initdata;
50
51#ifdef CONFIG_HAVE_SETUP_PER_CPU_AREA
52unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
53EXPORT_SYMBOL(__per_cpu_offset);
54
55static int __init pcpu_cpu_to_node(int cpu)
56{
57 return early_cpu_to_node(cpu);
58}
59
60static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
61{
62 if (early_cpu_to_node(cpu: from) == early_cpu_to_node(cpu: to))
63 return LOCAL_DISTANCE;
64 else
65 return REMOTE_DISTANCE;
66}
67
68void __init pcpu_populate_pte(unsigned long addr)
69{
70 populate_kernel_pte(addr);
71}
72
73void __init setup_per_cpu_areas(void)
74{
75 unsigned long delta;
76 unsigned int cpu;
77 int rc = -EINVAL;
78
79 if (pcpu_chosen_fc == PCPU_FC_AUTO) {
80 if (nr_node_ids >= 8)
81 pcpu_chosen_fc = PCPU_FC_PAGE;
82 else
83 pcpu_chosen_fc = PCPU_FC_EMBED;
84 }
85
86 /*
87 * Always reserve area for module percpu variables. That's
88 * what the legacy allocator did.
89 */
90 if (pcpu_chosen_fc != PCPU_FC_PAGE) {
91 rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
92 PERCPU_DYNAMIC_RESERVE, PMD_SIZE,
93 cpu_distance_fn: pcpu_cpu_distance, cpu_to_nd_fn: pcpu_cpu_to_node);
94 if (rc < 0)
95 pr_warn("%s allocator failed (%d), falling back to page size\n",
96 pcpu_fc_names[pcpu_chosen_fc], rc);
97 }
98 if (rc < 0)
99 rc = pcpu_page_first_chunk(PERCPU_MODULE_RESERVE, cpu_to_nd_fn: pcpu_cpu_to_node);
100 if (rc < 0)
101 panic(fmt: "cannot initialize percpu area (err=%d)", rc);
102
103 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
104 for_each_possible_cpu(cpu)
105 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
106}
107#endif
108
109/*
110 * Get nodeid by logical cpu number.
111 * __cpuid_to_node maps phyical cpu id to node, so we
112 * should use cpu_logical_map(cpu) to index it.
113 *
114 * This routine is only used in early phase during
115 * booting, after setup_per_cpu_areas calling and numa_node
116 * initialization, cpu_to_node will be used instead.
117 */
118int early_cpu_to_node(int cpu)
119{
120 int physid = cpu_logical_map(cpu);
121
122 if (physid < 0)
123 return NUMA_NO_NODE;
124
125 return __cpuid_to_node[physid];
126}
127
128void __init early_numa_add_cpu(int cpuid, s16 node)
129{
130 int cpu = __cpu_number_map[cpuid];
131
132 if (cpu < 0)
133 return;
134
135 cpumask_set_cpu(cpu, dstp: &cpus_on_node[node]);
136 cpumask_set_cpu(cpu: cpuid, dstp: &phys_cpus_on_node[node]);
137}
138
139void numa_add_cpu(unsigned int cpu)
140{
141 int nid = cpu_to_node(cpu);
142 cpumask_set_cpu(cpu, dstp: &cpus_on_node[nid]);
143}
144
145void numa_remove_cpu(unsigned int cpu)
146{
147 int nid = cpu_to_node(cpu);
148 cpumask_clear_cpu(cpu, dstp: &cpus_on_node[nid]);
149}
150
151static int __init numa_add_memblk_to(int nid, u64 start, u64 end,
152 struct numa_meminfo *mi)
153{
154 /* ignore zero length blks */
155 if (start == end)
156 return 0;
157
158 /* whine about and ignore invalid blks */
159 if (start > end || nid < 0 || nid >= MAX_NUMNODES) {
160 pr_warn("NUMA: Warning: invalid memblk node %d [mem %#010Lx-%#010Lx]\n",
161 nid, start, end - 1);
162 return 0;
163 }
164
165 if (mi->nr_blks >= NR_NODE_MEMBLKS) {
166 pr_err("NUMA: too many memblk ranges\n");
167 return -EINVAL;
168 }
169
170 mi->blk[mi->nr_blks].start = PFN_ALIGN(start);
171 mi->blk[mi->nr_blks].end = PFN_ALIGN(end - PAGE_SIZE + 1);
172 mi->blk[mi->nr_blks].nid = nid;
173 mi->nr_blks++;
174 return 0;
175}
176
177/**
178 * numa_add_memblk - Add one numa_memblk to numa_meminfo
179 * @nid: NUMA node ID of the new memblk
180 * @start: Start address of the new memblk
181 * @end: End address of the new memblk
182 *
183 * Add a new memblk to the default numa_meminfo.
184 *
185 * RETURNS:
186 * 0 on success, -errno on failure.
187 */
188int __init numa_add_memblk(int nid, u64 start, u64 end)
189{
190 return numa_add_memblk_to(nid, start, end, mi: &numa_meminfo);
191}
192
193static void __init alloc_node_data(int nid)
194{
195 void *nd;
196 unsigned long nd_pa;
197 size_t nd_sz = roundup(sizeof(pg_data_t), PAGE_SIZE);
198
199 nd_pa = memblock_phys_alloc_try_nid(size: nd_sz, SMP_CACHE_BYTES, nid);
200 if (!nd_pa) {
201 pr_err("Cannot find %zu Byte for node_data (initial node: %d)\n", nd_sz, nid);
202 return;
203 }
204
205 nd = __va(nd_pa);
206
207 node_data[nid] = nd;
208 memset(nd, 0, sizeof(pg_data_t));
209}
210
211static void __init node_mem_init(unsigned int node)
212{
213 unsigned long start_pfn, end_pfn;
214 unsigned long node_addrspace_offset;
215
216 node_addrspace_offset = nid_to_addrbase(node);
217 pr_info("Node%d's addrspace_offset is 0x%lx\n",
218 node, node_addrspace_offset);
219
220 get_pfn_range_for_nid(nid: node, start_pfn: &start_pfn, end_pfn: &end_pfn);
221 pr_info("Node%d: start_pfn=0x%lx, end_pfn=0x%lx\n",
222 node, start_pfn, end_pfn);
223
224 alloc_node_data(nid: node);
225}
226
227#ifdef CONFIG_ACPI_NUMA
228
229static void __init add_node_intersection(u32 node, u64 start, u64 size, u32 type)
230{
231 static unsigned long num_physpages;
232
233 num_physpages += (size >> PAGE_SHIFT);
234 pr_info("Node%d: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx Bytes\n",
235 node, type, start, size);
236 pr_info(" start_pfn:0x%llx, end_pfn:0x%llx, num_physpages:0x%lx\n",
237 start >> PAGE_SHIFT, (start + size) >> PAGE_SHIFT, num_physpages);
238 memblock_set_node(base: start, size, type: &memblock.memory, nid: node);
239}
240
241/*
242 * add_numamem_region
243 *
244 * Add a uasable memory region described by BIOS. The
245 * routine gets each intersection between BIOS's region
246 * and node's region, and adds them into node's memblock
247 * pool.
248 *
249 */
250static void __init add_numamem_region(u64 start, u64 end, u32 type)
251{
252 u32 i;
253 u64 ofs = start;
254
255 if (start >= end) {
256 pr_debug("Invalid region: %016llx-%016llx\n", start, end);
257 return;
258 }
259
260 for (i = 0; i < numa_meminfo.nr_blks; i++) {
261 struct numa_memblk *mb = &numa_meminfo.blk[i];
262
263 if (ofs > mb->end)
264 continue;
265
266 if (end > mb->end) {
267 add_node_intersection(node: mb->nid, start: ofs, size: mb->end - ofs, type);
268 ofs = mb->end;
269 } else {
270 add_node_intersection(node: mb->nid, start: ofs, size: end - ofs, type);
271 break;
272 }
273 }
274}
275
276static void __init init_node_memblock(void)
277{
278 u32 mem_type;
279 u64 mem_end, mem_start, mem_size;
280 efi_memory_desc_t *md;
281
282 /* Parse memory information and activate */
283 for_each_efi_memory_desc(md) {
284 mem_type = md->type;
285 mem_start = md->phys_addr;
286 mem_size = md->num_pages << EFI_PAGE_SHIFT;
287 mem_end = mem_start + mem_size;
288
289 switch (mem_type) {
290 case EFI_LOADER_CODE:
291 case EFI_LOADER_DATA:
292 case EFI_BOOT_SERVICES_CODE:
293 case EFI_BOOT_SERVICES_DATA:
294 case EFI_PERSISTENT_MEMORY:
295 case EFI_CONVENTIONAL_MEMORY:
296 add_numamem_region(start: mem_start, end: mem_end, type: mem_type);
297 break;
298 case EFI_PAL_CODE:
299 case EFI_UNUSABLE_MEMORY:
300 case EFI_ACPI_RECLAIM_MEMORY:
301 add_numamem_region(start: mem_start, end: mem_end, type: mem_type);
302 fallthrough;
303 case EFI_RESERVED_TYPE:
304 case EFI_RUNTIME_SERVICES_CODE:
305 case EFI_RUNTIME_SERVICES_DATA:
306 case EFI_MEMORY_MAPPED_IO:
307 case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
308 pr_info("Resvd: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx Bytes\n",
309 mem_type, mem_start, mem_size);
310 break;
311 }
312 }
313}
314
315static void __init numa_default_distance(void)
316{
317 int row, col;
318
319 for (row = 0; row < MAX_NUMNODES; row++)
320 for (col = 0; col < MAX_NUMNODES; col++) {
321 if (col == row)
322 node_distances[row][col] = LOCAL_DISTANCE;
323 else
324 /* We assume that one node per package here!
325 *
326 * A SLIT should be used for multiple nodes
327 * per package to override default setting.
328 */
329 node_distances[row][col] = REMOTE_DISTANCE;
330 }
331}
332
333/*
334 * fake_numa_init() - For Non-ACPI systems
335 * Return: 0 on success, -errno on failure.
336 */
337static int __init fake_numa_init(void)
338{
339 phys_addr_t start = memblock_start_of_DRAM();
340 phys_addr_t end = memblock_end_of_DRAM() - 1;
341
342 node_set(0, numa_nodes_parsed);
343 pr_info("Faking a node at [mem %pap-%pap]\n", &start, &end);
344
345 return numa_add_memblk(nid: 0, start, end: end + 1);
346}
347
348int __init init_numa_memory(void)
349{
350 int i;
351 int ret;
352 int node;
353
354 for (i = 0; i < NR_CPUS; i++)
355 set_cpuid_to_node(i, NUMA_NO_NODE);
356
357 numa_default_distance();
358 nodes_clear(numa_nodes_parsed);
359 nodes_clear(node_possible_map);
360 nodes_clear(node_online_map);
361 memset(&numa_meminfo, 0, sizeof(numa_meminfo));
362
363 /* Parse SRAT and SLIT if provided by firmware. */
364 ret = acpi_disabled ? fake_numa_init() : acpi_numa_init();
365 if (ret < 0)
366 return ret;
367
368 node_possible_map = numa_nodes_parsed;
369 if (WARN_ON(nodes_empty(node_possible_map)))
370 return -EINVAL;
371
372 init_node_memblock();
373 if (!memblock_validate_numa_coverage(SZ_1M))
374 return -EINVAL;
375
376 for_each_node_mask(node, node_possible_map) {
377 node_mem_init(node);
378 node_set_online(nid: node);
379 }
380 max_low_pfn = PHYS_PFN(memblock_end_of_DRAM());
381
382 setup_nr_node_ids();
383 loongson_sysconf.nr_nodes = nr_node_ids;
384 loongson_sysconf.cores_per_node = cpumask_weight(&phys_cpus_on_node[0]);
385
386 return 0;
387}
388
389#endif
390
391void __init paging_init(void)
392{
393 unsigned int node;
394 unsigned long zones_size[MAX_NR_ZONES] = {0, };
395
396 for_each_online_node(node) {
397 unsigned long start_pfn, end_pfn;
398
399 get_pfn_range_for_nid(nid: node, start_pfn: &start_pfn, end_pfn: &end_pfn);
400
401 if (end_pfn > max_low_pfn)
402 max_low_pfn = end_pfn;
403 }
404#ifdef CONFIG_ZONE_DMA32
405 zones_size[ZONE_DMA32] = MAX_DMA32_PFN;
406#endif
407 zones_size[ZONE_NORMAL] = max_low_pfn;
408 free_area_init(max_zone_pfn: zones_size);
409}
410
411void __init mem_init(void)
412{
413 high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
414 memblock_free_all();
415}
416
417int pcibus_to_node(struct pci_bus *bus)
418{
419 return dev_to_node(dev: &bus->dev);
420}
421EXPORT_SYMBOL(pcibus_to_node);
422

source code of linux/arch/loongarch/kernel/numa.c