acpi/numa/srat.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  *  acpi_numa.c - ACPI NUMA support
0004  *
0005  *  Copyright (C) 2002 Takayoshi Kochi <t-kochi@bq.jp.nec.com>
0006  */
0007
0008 #define pr_fmt(fmt) "ACPI: " fmt
0009
0010 #include <linux/module.h>
0011 #include <linux/init.h>
0012 #include <linux/kernel.h>
0013 #include <linux/types.h>
0014 #include <linux/errno.h>
0015 #include <linux/acpi.h>
0016 #include <linux/memblock.h>
0017 #include <linux/numa.h>
0018 #include <linux/nodemask.h>
0019 #include <linux/topology.h>
0020
0021 static nodemask_t nodes_found_map = NODE_MASK_NONE;
0022
0023 /* maps to convert between proximity domain and logical node ID */
0024 static int pxm_to_node_map[MAX_PXM_DOMAINS]
0025             = { [0 ... MAX_PXM_DOMAINS - 1] = NUMA_NO_NODE };
0026 static int node_to_pxm_map[MAX_NUMNODES]
0027             = { [0 ... MAX_NUMNODES - 1] = PXM_INVAL };
0028
0029 unsigned char acpi_srat_revision __initdata;
0030 static int acpi_numa __initdata;
0031
0032 void __init disable_srat(void)
0033 {
0034     acpi_numa = -1;
0035 }
0036
0037 int pxm_to_node(int pxm)
0038 {
0039     if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
0040         return NUMA_NO_NODE;
0041     return pxm_to_node_map[pxm];
0042 }
0043 EXPORT_SYMBOL(pxm_to_node);
0044
0045 int node_to_pxm(int node)
0046 {
0047     if (node < 0)
0048         return PXM_INVAL;
0049     return node_to_pxm_map[node];
0050 }
0051
0052 static void __acpi_map_pxm_to_node(int pxm, int node)
0053 {
0054     if (pxm_to_node_map[pxm] == NUMA_NO_NODE || node < pxm_to_node_map[pxm])
0055         pxm_to_node_map[pxm] = node;
0056     if (node_to_pxm_map[node] == PXM_INVAL || pxm < node_to_pxm_map[node])
0057         node_to_pxm_map[node] = pxm;
0058 }
0059
0060 int acpi_map_pxm_to_node(int pxm)
0061 {
0062     int node;
0063
0064     if (pxm < 0 || pxm >= MAX_PXM_DOMAINS || numa_off)
0065         return NUMA_NO_NODE;
0066
0067     node = pxm_to_node_map[pxm];
0068
0069     if (node == NUMA_NO_NODE) {
0070         if (nodes_weight(nodes_found_map) >= MAX_NUMNODES)
0071             return NUMA_NO_NODE;
0072         node = first_unset_node(nodes_found_map);
0073         __acpi_map_pxm_to_node(pxm, node);
0074         node_set(node, nodes_found_map);
0075     }
0076
0077     return node;
0078 }
0079 EXPORT_SYMBOL(acpi_map_pxm_to_node);
0080
0081 static void __init
0082 acpi_table_print_srat_entry(struct acpi_subtable_header *header)
0083 {
0084     switch (header->type) {
0085     case ACPI_SRAT_TYPE_CPU_AFFINITY:
0086         {
0087             struct acpi_srat_cpu_affinity *p =
0088                 (struct acpi_srat_cpu_affinity *)header;
0089             pr_debug("SRAT Processor (id[0x%02x] eid[0x%02x]) in proximity domain %d %s\n",
0090                  p->apic_id, p->local_sapic_eid,
0091                  p->proximity_domain_lo,
0092                  (p->flags & ACPI_SRAT_CPU_ENABLED) ?
0093                  "enabled" : "disabled");
0094         }
0095         break;
0096
0097     case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
0098         {
0099             struct acpi_srat_mem_affinity *p =
0100                 (struct acpi_srat_mem_affinity *)header;
0101             pr_debug("SRAT Memory (0x%llx length 0x%llx) in proximity domain %d %s%s%s\n",
0102                  (unsigned long long)p->base_address,
0103                  (unsigned long long)p->length,
0104                  p->proximity_domain,
0105                  (p->flags & ACPI_SRAT_MEM_ENABLED) ?
0106                  "enabled" : "disabled",
0107                  (p->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
0108                  " hot-pluggable" : "",
0109                  (p->flags & ACPI_SRAT_MEM_NON_VOLATILE) ?
0110                  " non-volatile" : "");
0111         }
0112         break;
0113
0114     case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
0115         {
0116             struct acpi_srat_x2apic_cpu_affinity *p =
0117                 (struct acpi_srat_x2apic_cpu_affinity *)header;
0118             pr_debug("SRAT Processor (x2apicid[0x%08x]) in proximity domain %d %s\n",
0119                  p->apic_id,
0120                  p->proximity_domain,
0121                  (p->flags & ACPI_SRAT_CPU_ENABLED) ?
0122                  "enabled" : "disabled");
0123         }
0124         break;
0125
0126     case ACPI_SRAT_TYPE_GICC_AFFINITY:
0127         {
0128             struct acpi_srat_gicc_affinity *p =
0129                 (struct acpi_srat_gicc_affinity *)header;
0130             pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
0131                  p->acpi_processor_uid,
0132                  p->proximity_domain,
0133                  (p->flags & ACPI_SRAT_GICC_ENABLED) ?
0134                  "enabled" : "disabled");
0135         }
0136         break;
0137
0138     case ACPI_SRAT_TYPE_GENERIC_AFFINITY:
0139     {
0140         struct acpi_srat_generic_affinity *p =
0141             (struct acpi_srat_generic_affinity *)header;
0142
0143         if (p->device_handle_type == 0) {
0144             /*
0145              * For pci devices this may be the only place they
0146              * are assigned a proximity domain
0147              */
0148             pr_debug("SRAT Generic Initiator(Seg:%u BDF:%u) in proximity domain %d %s\n",
0149                  *(u16 *)(&p->device_handle[0]),
0150                  *(u16 *)(&p->device_handle[2]),
0151                  p->proximity_domain,
0152                  (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
0153                 "enabled" : "disabled");
0154         } else {
0155             /*
0156              * In this case we can rely on the device having a
0157              * proximity domain reference
0158              */
0159             pr_debug("SRAT Generic Initiator(HID=%.8s UID=%.4s) in proximity domain %d %s\n",
0160                 (char *)(&p->device_handle[0]),
0161                 (char *)(&p->device_handle[8]),
0162                 p->proximity_domain,
0163                 (p->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED) ?
0164                 "enabled" : "disabled");
0165         }
0166     }
0167     break;
0168     default:
0169         pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
0170             header->type);
0171         break;
0172     }
0173 }
0174
0175 /*
0176  * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
0177  * up the NUMA heuristics which wants the local node to have a smaller
0178  * distance than the others.
0179  * Do some quick checks here and only use the SLIT if it passes.
0180  */
0181 static int __init slit_valid(struct acpi_table_slit *slit)
0182 {
0183     int i, j;
0184     int d = slit->locality_count;
0185     for (i = 0; i < d; i++) {
0186         for (j = 0; j < d; j++)  {
0187             u8 val = slit->entry[d*i + j];
0188             if (i == j) {
0189                 if (val != LOCAL_DISTANCE)
0190                     return 0;
0191             } else if (val <= LOCAL_DISTANCE)
0192                 return 0;
0193         }
0194     }
0195     return 1;
0196 }
0197
0198 void __init bad_srat(void)
0199 {
0200     pr_err("SRAT: SRAT not used.\n");
0201     disable_srat();
0202 }
0203
0204 int __init srat_disabled(void)
0205 {
0206     return acpi_numa < 0;
0207 }
0208
0209 #if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
0210 /*
0211  * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
0212  * I/O localities since SRAT does not list them.  I/O localities are
0213  * not supported at this point.
0214  */
0215 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
0216 {
0217     int i, j;
0218
0219     for (i = 0; i < slit->locality_count; i++) {
0220         const int from_node = pxm_to_node(i);
0221
0222         if (from_node == NUMA_NO_NODE)
0223             continue;
0224
0225         for (j = 0; j < slit->locality_count; j++) {
0226             const int to_node = pxm_to_node(j);
0227
0228             if (to_node == NUMA_NO_NODE)
0229                 continue;
0230
0231             numa_set_distance(from_node, to_node,
0232                 slit->entry[slit->locality_count * i + j]);
0233         }
0234     }
0235 }
0236
0237 /*
0238  * Default callback for parsing of the Proximity Domain <-> Memory
0239  * Area mappings
0240  */
0241 int __init
0242 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
0243 {
0244     u64 start, end;
0245     u32 hotpluggable;
0246     int node, pxm;
0247
0248     if (srat_disabled())
0249         goto out_err;
0250     if (ma->header.length < sizeof(struct acpi_srat_mem_affinity)) {
0251         pr_err("SRAT: Unexpected header length: %d\n",
0252                ma->header.length);
0253         goto out_err_bad_srat;
0254     }
0255     if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
0256         goto out_err;
0257     hotpluggable = IS_ENABLED(CONFIG_MEMORY_HOTPLUG) &&
0258         (ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE);
0259
0260     start = ma->base_address;
0261     end = start + ma->length;
0262     pxm = ma->proximity_domain;
0263     if (acpi_srat_revision <= 1)
0264         pxm &= 0xff;
0265
0266     node = acpi_map_pxm_to_node(pxm);
0267     if (node == NUMA_NO_NODE) {
0268         pr_err("SRAT: Too many proximity domains.\n");
0269         goto out_err_bad_srat;
0270     }
0271
0272     if (numa_add_memblk(node, start, end) < 0) {
0273         pr_err("SRAT: Failed to add memblk to node %u [mem %#010Lx-%#010Lx]\n",
0274                node, (unsigned long long) start,
0275                (unsigned long long) end - 1);
0276         goto out_err_bad_srat;
0277     }
0278
0279     node_set(node, numa_nodes_parsed);
0280
0281     pr_info("SRAT: Node %u PXM %u [mem %#010Lx-%#010Lx]%s%s\n",
0282         node, pxm,
0283         (unsigned long long) start, (unsigned long long) end - 1,
0284         hotpluggable ? " hotplug" : "",
0285         ma->flags & ACPI_SRAT_MEM_NON_VOLATILE ? " non-volatile" : "");
0286
0287     /* Mark hotplug range in memblock. */
0288     if (hotpluggable && memblock_mark_hotplug(start, ma->length))
0289         pr_warn("SRAT: Failed to mark hotplug range [mem %#010Lx-%#010Lx] in memblock\n",
0290             (unsigned long long)start, (unsigned long long)end - 1);
0291
0292     max_possible_pfn = max(max_possible_pfn, PFN_UP(end - 1));
0293
0294     return 0;
0295 out_err_bad_srat:
0296     bad_srat();
0297 out_err:
0298     return -EINVAL;
0299 }
0300
0301 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header,
0302                    void *arg, const unsigned long table_end)
0303 {
0304     struct acpi_cedt_cfmws *cfmws;
0305     int *fake_pxm = arg;
0306     u64 start, end;
0307     int node;
0308
0309     cfmws = (struct acpi_cedt_cfmws *)header;
0310     start = cfmws->base_hpa;
0311     end = cfmws->base_hpa + cfmws->window_size;
0312
0313     /* Skip if the SRAT already described the NUMA details for this HPA */
0314     node = phys_to_target_node(start);
0315     if (node != NUMA_NO_NODE)
0316         return 0;
0317
0318     node = acpi_map_pxm_to_node(*fake_pxm);
0319
0320     if (node == NUMA_NO_NODE) {
0321         pr_err("ACPI NUMA: Too many proximity domains while processing CFMWS.\n");
0322         return -EINVAL;
0323     }
0324
0325     if (numa_add_memblk(node, start, end) < 0) {
0326         /* CXL driver must handle the NUMA_NO_NODE case */
0327         pr_warn("ACPI NUMA: Failed to add memblk for CFMWS node %d [mem %#llx-%#llx]\n",
0328             node, start, end);
0329     }
0330
0331     /* Set the next available fake_pxm value */
0332     (*fake_pxm)++;
0333     return 0;
0334 }
0335 #else
0336 static int __init acpi_parse_cfmws(union acpi_subtable_headers *header,
0337                    void *arg, const unsigned long table_end)
0338 {
0339     return 0;
0340 }
0341 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
0342
0343 static int __init acpi_parse_slit(struct acpi_table_header *table)
0344 {
0345     struct acpi_table_slit *slit = (struct acpi_table_slit *)table;
0346
0347     if (!slit_valid(slit)) {
0348         pr_info("SLIT table looks invalid. Not used.\n");
0349         return -EINVAL;
0350     }
0351     acpi_numa_slit_init(slit);
0352
0353     return 0;
0354 }
0355
0356 void __init __weak
0357 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
0358 {
0359     pr_warn("Found unsupported x2apic [0x%08x] SRAT entry\n", pa->apic_id);
0360 }
0361
0362 static int __init
0363 acpi_parse_x2apic_affinity(union acpi_subtable_headers *header,
0364                const unsigned long end)
0365 {
0366     struct acpi_srat_x2apic_cpu_affinity *processor_affinity;
0367
0368     processor_affinity = (struct acpi_srat_x2apic_cpu_affinity *)header;
0369
0370     acpi_table_print_srat_entry(&header->common);
0371
0372     /* let architecture-dependent part to do it */
0373     acpi_numa_x2apic_affinity_init(processor_affinity);
0374
0375     return 0;
0376 }
0377
0378 static int __init
0379 acpi_parse_processor_affinity(union acpi_subtable_headers *header,
0380                   const unsigned long end)
0381 {
0382     struct acpi_srat_cpu_affinity *processor_affinity;
0383
0384     processor_affinity = (struct acpi_srat_cpu_affinity *)header;
0385
0386     acpi_table_print_srat_entry(&header->common);
0387
0388     /* let architecture-dependent part to do it */
0389     acpi_numa_processor_affinity_init(processor_affinity);
0390
0391     return 0;
0392 }
0393
0394 static int __init
0395 acpi_parse_gicc_affinity(union acpi_subtable_headers *header,
0396              const unsigned long end)
0397 {
0398     struct acpi_srat_gicc_affinity *processor_affinity;
0399
0400     processor_affinity = (struct acpi_srat_gicc_affinity *)header;
0401
0402     acpi_table_print_srat_entry(&header->common);
0403
0404     /* let architecture-dependent part to do it */
0405     acpi_numa_gicc_affinity_init(processor_affinity);
0406
0407     return 0;
0408 }
0409
0410 #if defined(CONFIG_X86) || defined(CONFIG_ARM64)
0411 static int __init
0412 acpi_parse_gi_affinity(union acpi_subtable_headers *header,
0413                const unsigned long end)
0414 {
0415     struct acpi_srat_generic_affinity *gi_affinity;
0416     int node;
0417
0418     gi_affinity = (struct acpi_srat_generic_affinity *)header;
0419     if (!gi_affinity)
0420         return -EINVAL;
0421     acpi_table_print_srat_entry(&header->common);
0422
0423     if (!(gi_affinity->flags & ACPI_SRAT_GENERIC_AFFINITY_ENABLED))
0424         return -EINVAL;
0425
0426     node = acpi_map_pxm_to_node(gi_affinity->proximity_domain);
0427     if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
0428         pr_err("SRAT: Too many proximity domains.\n");
0429         return -EINVAL;
0430     }
0431     node_set(node, numa_nodes_parsed);
0432     node_set_state(node, N_GENERIC_INITIATOR);
0433
0434     return 0;
0435 }
0436 #else
0437 static int __init
0438 acpi_parse_gi_affinity(union acpi_subtable_headers *header,
0439                const unsigned long end)
0440 {
0441     return 0;
0442 }
0443 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
0444
0445 static int __initdata parsed_numa_memblks;
0446
0447 static int __init
0448 acpi_parse_memory_affinity(union acpi_subtable_headers * header,
0449                const unsigned long end)
0450 {
0451     struct acpi_srat_mem_affinity *memory_affinity;
0452
0453     memory_affinity = (struct acpi_srat_mem_affinity *)header;
0454
0455     acpi_table_print_srat_entry(&header->common);
0456
0457     /* let architecture-dependent part to do it */
0458     if (!acpi_numa_memory_affinity_init(memory_affinity))
0459         parsed_numa_memblks++;
0460     return 0;
0461 }
0462
0463 static int __init acpi_parse_srat(struct acpi_table_header *table)
0464 {
0465     struct acpi_table_srat *srat = (struct acpi_table_srat *)table;
0466
0467     acpi_srat_revision = srat->header.revision;
0468
0469     /* Real work done in acpi_table_parse_srat below. */
0470
0471     return 0;
0472 }
0473
0474 static int __init
0475 acpi_table_parse_srat(enum acpi_srat_type id,
0476               acpi_tbl_entry_handler handler, unsigned int max_entries)
0477 {
0478     return acpi_table_parse_entries(ACPI_SIG_SRAT,
0479                         sizeof(struct acpi_table_srat), id,
0480                         handler, max_entries);
0481 }
0482
0483 int __init acpi_numa_init(void)
0484 {
0485     int i, fake_pxm, cnt = 0;
0486
0487     if (acpi_disabled)
0488         return -EINVAL;
0489
0490     /*
0491      * Should not limit number with cpu num that is from NR_CPUS or nr_cpus=
0492      * SRAT cpu entries could have different order with that in MADT.
0493      * So go over all cpu entries in SRAT to get apicid to node mapping.
0494      */
0495
0496     /* SRAT: System Resource Affinity Table */
0497     if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
0498         struct acpi_subtable_proc srat_proc[4];
0499
0500         memset(srat_proc, 0, sizeof(srat_proc));
0501         srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;
0502         srat_proc[0].handler = acpi_parse_processor_affinity;
0503         srat_proc[1].id = ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY;
0504         srat_proc[1].handler = acpi_parse_x2apic_affinity;
0505         srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
0506         srat_proc[2].handler = acpi_parse_gicc_affinity;
0507         srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
0508         srat_proc[3].handler = acpi_parse_gi_affinity;
0509
0510         acpi_table_parse_entries_array(ACPI_SIG_SRAT,
0511                     sizeof(struct acpi_table_srat),
0512                     srat_proc, ARRAY_SIZE(srat_proc), 0);
0513
0514         cnt = acpi_table_parse_srat(ACPI_SRAT_TYPE_MEMORY_AFFINITY,
0515                         acpi_parse_memory_affinity, 0);
0516     }
0517
0518     /* SLIT: System Locality Information Table */
0519     acpi_table_parse(ACPI_SIG_SLIT, acpi_parse_slit);
0520
0521     /*
0522      * CXL Fixed Memory Window Structures (CFMWS) must be parsed
0523      * after the SRAT. Create NUMA Nodes for CXL memory ranges that
0524      * are defined in the CFMWS and not already defined in the SRAT.
0525      * Initialize a fake_pxm as the first available PXM to emulate.
0526      */
0527
0528     /* fake_pxm is the next unused PXM value after SRAT parsing */
0529     for (i = 0, fake_pxm = -1; i < MAX_NUMNODES - 1; i++) {
0530         if (node_to_pxm_map[i] > fake_pxm)
0531             fake_pxm = node_to_pxm_map[i];
0532     }
0533     fake_pxm++;
0534     acpi_table_parse_cedt(ACPI_CEDT_TYPE_CFMWS, acpi_parse_cfmws,
0535                   &fake_pxm);
0536
0537     if (cnt < 0)
0538         return cnt;
0539     else if (!parsed_numa_memblks)
0540         return -ENOENT;
0541     return 0;
0542 }
0543
0544 static int acpi_get_pxm(acpi_handle h)
0545 {
0546     unsigned long long pxm;
0547     acpi_status status;
0548     acpi_handle handle;
0549     acpi_handle phandle = h;
0550
0551     do {
0552         handle = phandle;
0553         status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
0554         if (ACPI_SUCCESS(status))
0555             return pxm;
0556         status = acpi_get_parent(handle, &phandle);
0557     } while (ACPI_SUCCESS(status));
0558     return -1;
0559 }
0560
0561 int acpi_get_node(acpi_handle handle)
0562 {
0563     int pxm;
0564
0565     pxm = acpi_get_pxm(handle);
0566
0567     return pxm_to_node(pxm);
0568 }
0569 EXPORT_SYMBOL(acpi_get_node);