2 * Copyright (c) 2010 Advanced Computing Technologies LLC
3 * Written by: John H. Baldwin <jhb@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
33 #include <sys/kernel.h>
37 #include <vm/vm_param.h>
38 #include <vm/vm_phys.h>
40 #include <contrib/dev/acpica/include/acpi.h>
41 #include <contrib/dev/acpica/include/actables.h>
43 #include <machine/intr_machdep.h>
44 #include <machine/apicvar.h>
46 #include <dev/acpica/acpivar.h>
53 } cpus[MAX_APIC_ID + 1];
55 struct mem_affinity mem_info[VM_PHYSSEG_MAX + 1];
58 static ACPI_TABLE_SRAT *srat;
59 static vm_paddr_t srat_physaddr;
61 static void srat_walk_table(acpi_subtable_handler *handler, void *arg);
64 * Returns true if a memory range overlaps with at least one range in
68 overlaps_phys_avail(vm_paddr_t start, vm_paddr_t end)
72 for (i = 0; phys_avail[i] != 0 && phys_avail[i + 1] != 0; i += 2) {
73 if (phys_avail[i + 1] < start)
75 if (phys_avail[i] < end)
84 srat_parse_entry(ACPI_SUBTABLE_HEADER *entry, void *arg)
86 ACPI_SRAT_CPU_AFFINITY *cpu;
87 ACPI_SRAT_X2APIC_CPU_AFFINITY *x2apic;
88 ACPI_SRAT_MEM_AFFINITY *mem;
91 switch (entry->Type) {
92 case ACPI_SRAT_TYPE_CPU_AFFINITY:
93 cpu = (ACPI_SRAT_CPU_AFFINITY *)entry;
94 domain = cpu->ProximityDomainLo |
95 cpu->ProximityDomainHi[0] << 8 |
96 cpu->ProximityDomainHi[1] << 16 |
97 cpu->ProximityDomainHi[2] << 24;
99 printf("SRAT: Found CPU APIC ID %u domain %d: %s\n",
101 (cpu->Flags & ACPI_SRAT_CPU_ENABLED) ?
102 "enabled" : "disabled");
103 if (!(cpu->Flags & ACPI_SRAT_CPU_ENABLED))
105 KASSERT(!cpus[cpu->ApicId].enabled,
106 ("Duplicate local APIC ID %u", cpu->ApicId));
107 cpus[cpu->ApicId].domain = domain;
108 cpus[cpu->ApicId].enabled = 1;
110 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
111 x2apic = (ACPI_SRAT_X2APIC_CPU_AFFINITY *)entry;
113 printf("SRAT: Found CPU APIC ID %u domain %d: %s\n",
114 x2apic->ApicId, x2apic->ProximityDomain,
115 (x2apic->Flags & ACPI_SRAT_CPU_ENABLED) ?
116 "enabled" : "disabled");
117 if (!(x2apic->Flags & ACPI_SRAT_CPU_ENABLED))
119 KASSERT(!cpus[x2apic->ApicId].enabled,
120 ("Duplicate local APIC ID %u", x2apic->ApicId));
121 cpus[x2apic->ApicId].domain = x2apic->ProximityDomain;
122 cpus[x2apic->ApicId].enabled = 1;
124 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
125 mem = (ACPI_SRAT_MEM_AFFINITY *)entry;
128 "SRAT: Found memory domain %d addr %jx len %jx: %s\n",
129 mem->ProximityDomain, (uintmax_t)mem->BaseAddress,
130 (uintmax_t)mem->Length,
131 (mem->Flags & ACPI_SRAT_MEM_ENABLED) ?
132 "enabled" : "disabled");
133 if (!(mem->Flags & ACPI_SRAT_MEM_ENABLED))
135 if (!overlaps_phys_avail(mem->BaseAddress,
136 mem->BaseAddress + mem->Length)) {
137 printf("SRAT: Ignoring memory at addr %jx\n",
138 (uintmax_t)mem->BaseAddress);
141 if (num_mem == VM_PHYSSEG_MAX) {
142 printf("SRAT: Too many memory regions\n");
147 for (i = 0; i < num_mem; i++) {
148 if (mem_info[i].end <= mem->BaseAddress)
150 if (mem_info[i].start <
151 (mem->BaseAddress + mem->Length)) {
152 printf("SRAT: Overlapping memory entries\n");
158 for (i = num_mem; i > slot; i--)
159 mem_info[i] = mem_info[i - 1];
160 mem_info[slot].start = mem->BaseAddress;
161 mem_info[slot].end = mem->BaseAddress + mem->Length;
162 mem_info[slot].domain = mem->ProximityDomain;
169 * Ensure each memory domain has at least one CPU and that each CPU
170 * has at least one memory domain.
177 for (i = 0; i < num_mem; i++) {
179 for (j = 0; j <= MAX_APIC_ID; j++)
180 if (cpus[j].enabled &&
181 cpus[j].domain == mem_info[i].domain) {
182 cpus[j].has_memory = 1;
186 printf("SRAT: No CPU found for memory domain %d\n",
191 for (i = 0; i <= MAX_APIC_ID; i++)
192 if (cpus[i].enabled && !cpus[i].has_memory) {
193 printf("SRAT: No memory found for CPU %d\n", i);
200 * Check that the SRAT memory regions cover all of the regions in
204 check_phys_avail(void)
209 /* j is the current offset into phys_avail[]. */
210 address = phys_avail[0];
212 for (i = 0; i < num_mem; i++) {
214 * Consume as many phys_avail[] entries as fit in this
217 while (address >= mem_info[i].start &&
218 address <= mem_info[i].end) {
220 * If we cover the rest of this phys_avail[] entry,
221 * advance to the next entry.
223 if (phys_avail[j + 1] <= mem_info[i].end) {
225 if (phys_avail[j] == 0 &&
226 phys_avail[j + 1] == 0) {
229 address = phys_avail[j];
231 address = mem_info[i].end + 1;
234 printf("SRAT: No memory region found for %jx - %jx\n",
235 (uintmax_t)phys_avail[j], (uintmax_t)phys_avail[j + 1]);
240 * Renumber the memory domains to be compact and zero-based if not
241 * already. Returns an error if there are too many domains.
244 renumber_domains(void)
246 int domains[VM_PHYSSEG_MAX];
249 /* Enumerate all the domains. */
251 for (i = 0; i < num_mem; i++) {
252 /* See if this domain is already known. */
253 for (j = 0; j < vm_ndomains; j++) {
254 if (domains[j] >= mem_info[i].domain)
257 if (j < vm_ndomains && domains[j] == mem_info[i].domain)
260 /* Insert the new domain at slot 'j'. */
262 for (j = vm_ndomains; j > slot; j--)
263 domains[j] = domains[j - 1];
264 domains[slot] = mem_info[i].domain;
266 if (vm_ndomains > MAXMEMDOM) {
268 printf("SRAT: Too many memory domains\n");
273 /* Renumber each domain to its index in the sorted 'domains' list. */
274 for (i = 0; i < vm_ndomains; i++) {
276 * If the domain is already the right value, no need
282 /* Walk the cpu[] and mem_info[] arrays to renumber. */
283 for (j = 0; j < num_mem; j++)
284 if (mem_info[j].domain == domains[i])
285 mem_info[j].domain = i;
286 for (j = 0; j <= MAX_APIC_ID; j++)
287 if (cpus[j].enabled && cpus[j].domain == domains[i])
290 KASSERT(vm_ndomains > 0,
291 ("renumber_domains: invalid final vm_ndomains setup"));
297 * Look for an ACPI System Resource Affinity Table ("SRAT")
300 parse_srat(void *dummy)
304 if (resource_disabled("srat", 0))
307 srat_physaddr = acpi_find_table(ACPI_SIG_SRAT);
308 if (srat_physaddr == 0)
312 * Make a pass over the table to populate the cpus[] and
315 srat = acpi_map_table(srat_physaddr, ACPI_SIG_SRAT);
317 srat_walk_table(srat_parse_entry, &error);
318 acpi_unmap_table(srat);
320 if (error || check_domains() != 0 || check_phys_avail() != 0 ||
321 renumber_domains() != 0) {
326 /* Point vm_phys at our memory affinity table. */
327 mem_affinity = mem_info;
329 SYSINIT(parse_srat, SI_SUB_VM - 1, SI_ORDER_FIRST, parse_srat, NULL);
332 srat_walk_table(acpi_subtable_handler *handler, void *arg)
335 acpi_walk_subtables(srat + 1, (char *)srat + srat->Header.Length,
340 * Setup per-CPU ACPI IDs.
343 srat_set_cpus(void *dummy)
345 struct cpu_info *cpu;
349 if (srat_physaddr == 0)
351 for (i = 0; i < MAXCPU; i++) {
355 KASSERT(pc != NULL, ("no pcpu data for CPU %u", i));
356 cpu = &cpus[pc->pc_apic_id];
358 panic("SRAT: CPU with APIC ID %u is not known",
360 pc->pc_domain = cpu->domain;
362 printf("SRAT: CPU %u has memory domain %d\n", i,
366 SYSINIT(srat_set_cpus, SI_SUB_CPU, SI_ORDER_ANY, srat_set_cpus, NULL);
367 #endif /* MAXMEMDOM > 1 */