srat.c revision 299485 
1/*-
2 * Copyright (c) 2010 Hudson River Trading LLC
3 * Written by: John H. Baldwin <jhb@FreeBSD.org>
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
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.
14 *
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
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: stable/10/sys/x86/acpica/srat.c 299485 2016-05-11 22:06:28Z vangyzen $");
30
31#include <sys/param.h>
32#include <sys/bus.h>
33#include <sys/kernel.h>
34#include <sys/lock.h>
35#include <sys/mutex.h>
36#include <sys/smp.h>
37#include <sys/vmmeter.h>
38#include <vm/vm.h>
39#include <vm/pmap.h>
40#include <vm/vm_param.h>
41#include <vm/vm_page.h>
42#include <vm/vm_phys.h>
43
44#include <contrib/dev/acpica/include/acpi.h>
45#include <contrib/dev/acpica/include/actables.h>
46
47#include <machine/intr_machdep.h>
48#include <machine/apicvar.h>
49
50#include <dev/acpica/acpivar.h>
51
52#if MAXMEMDOM > 1
53struct cpu_info {
54	int enabled:1;
55	int has_memory:1;
56	int domain;
57} cpus[MAX_APIC_ID + 1];
58
59struct mem_affinity mem_info[VM_PHYSSEG_MAX + 1];
60int num_mem;
61
62static ACPI_TABLE_SRAT *srat;
63static vm_paddr_t srat_physaddr;
64
65static int vm_domains[VM_PHYSSEG_MAX];
66
67static void	srat_walk_table(acpi_subtable_handler *handler, void *arg);
68
69/*
70 * Returns true if a memory range overlaps with at least one range in
71 * phys_avail[].
72 */
73static int
74overlaps_phys_avail(vm_paddr_t start, vm_paddr_t end)
75{
76	int i;
77
78	for (i = 0; phys_avail[i] != 0 && phys_avail[i + 1] != 0; i += 2) {
79		if (phys_avail[i + 1] < start)
80			continue;
81		if (phys_avail[i] < end)
82			return (1);
83		break;
84	}
85	return (0);
86
87}
88
89static void
90srat_parse_entry(ACPI_SUBTABLE_HEADER *entry, void *arg)
91{
92	ACPI_SRAT_CPU_AFFINITY *cpu;
93	ACPI_SRAT_X2APIC_CPU_AFFINITY *x2apic;
94	ACPI_SRAT_MEM_AFFINITY *mem;
95	int domain, i, slot;
96
97	switch (entry->Type) {
98	case ACPI_SRAT_TYPE_CPU_AFFINITY:
99		cpu = (ACPI_SRAT_CPU_AFFINITY *)entry;
100		domain = cpu->ProximityDomainLo |
101		    cpu->ProximityDomainHi[0] << 8 |
102		    cpu->ProximityDomainHi[1] << 16 |
103		    cpu->ProximityDomainHi[2] << 24;
104		if (bootverbose)
105			printf("SRAT: Found CPU APIC ID %u domain %d: %s\n",
106			    cpu->ApicId, domain,
107			    (cpu->Flags & ACPI_SRAT_CPU_ENABLED) ?
108			    "enabled" : "disabled");
109		if (!(cpu->Flags & ACPI_SRAT_CPU_ENABLED))
110			break;
111		if (cpus[cpu->ApicId].enabled) {
112			printf("SRAT: Duplicate local APIC ID %u\n",
113			    cpu->ApicId);
114			*(int *)arg = ENXIO;
115			break;
116		}
117		cpus[cpu->ApicId].domain = domain;
118		cpus[cpu->ApicId].enabled = 1;
119		break;
120	case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY:
121		x2apic = (ACPI_SRAT_X2APIC_CPU_AFFINITY *)entry;
122		if (bootverbose)
123			printf("SRAT: Found CPU APIC ID %u domain %d: %s\n",
124			    x2apic->ApicId, x2apic->ProximityDomain,
125			    (x2apic->Flags & ACPI_SRAT_CPU_ENABLED) ?
126			    "enabled" : "disabled");
127		if (!(x2apic->Flags & ACPI_SRAT_CPU_ENABLED))
128			break;
129		KASSERT(!cpus[x2apic->ApicId].enabled,
130		    ("Duplicate local APIC ID %u", x2apic->ApicId));
131		cpus[x2apic->ApicId].domain = x2apic->ProximityDomain;
132		cpus[x2apic->ApicId].enabled = 1;
133		break;
134	case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
135		mem = (ACPI_SRAT_MEM_AFFINITY *)entry;
136		if (bootverbose)
137			printf(
138		    "SRAT: Found memory domain %d addr %jx len %jx: %s\n",
139			    mem->ProximityDomain, (uintmax_t)mem->BaseAddress,
140			    (uintmax_t)mem->Length,
141			    (mem->Flags & ACPI_SRAT_MEM_ENABLED) ?
142			    "enabled" : "disabled");
143		if (!(mem->Flags & ACPI_SRAT_MEM_ENABLED))
144			break;
145		if (!overlaps_phys_avail(mem->BaseAddress,
146		    mem->BaseAddress + mem->Length)) {
147			printf("SRAT: Ignoring memory at addr %jx\n",
148			    (uintmax_t)mem->BaseAddress);
149			break;
150		}
151		if (num_mem == VM_PHYSSEG_MAX) {
152			printf("SRAT: Too many memory regions\n");
153			*(int *)arg = ENXIO;
154			break;
155		}
156		slot = num_mem;
157		for (i = 0; i < num_mem; i++) {
158			if (mem_info[i].end <= mem->BaseAddress)
159				continue;
160			if (mem_info[i].start <
161			    (mem->BaseAddress + mem->Length)) {
162				printf("SRAT: Overlapping memory entries\n");
163				*(int *)arg = ENXIO;
164				return;
165			}
166			slot = i;
167		}
168		for (i = num_mem; i > slot; i--)
169			mem_info[i] = mem_info[i - 1];
170		mem_info[slot].start = mem->BaseAddress;
171		mem_info[slot].end = mem->BaseAddress + mem->Length;
172		mem_info[slot].domain = mem->ProximityDomain;
173		num_mem++;
174		break;
175	}
176}
177
178/*
179 * Ensure each memory domain has at least one CPU and that each CPU
180 * has at least one memory domain.
181 */
182static int
183check_domains(void)
184{
185	int found, i, j;
186
187	for (i = 0; i < num_mem; i++) {
188		found = 0;
189		for (j = 0; j <= MAX_APIC_ID; j++)
190			if (cpus[j].enabled &&
191			    cpus[j].domain == mem_info[i].domain) {
192				cpus[j].has_memory = 1;
193				found++;
194			}
195		if (!found) {
196			printf("SRAT: No CPU found for memory domain %d\n",
197			    mem_info[i].domain);
198			return (ENXIO);
199		}
200	}
201	for (i = 0; i <= MAX_APIC_ID; i++)
202		if (cpus[i].enabled && !cpus[i].has_memory) {
203			printf("SRAT: No memory found for CPU %d\n", i);
204			return (ENXIO);
205		}
206	return (0);
207}
208
209/*
210 * Check that the SRAT memory regions cover all of the regions in
211 * phys_avail[].
212 */
213static int
214check_phys_avail(void)
215{
216	vm_paddr_t address;
217	int i, j;
218
219	/* j is the current offset into phys_avail[]. */
220	address = phys_avail[0];
221	j = 0;
222	for (i = 0; i < num_mem; i++) {
223		/*
224		 * Consume as many phys_avail[] entries as fit in this
225		 * region.
226		 */
227		while (address >= mem_info[i].start &&
228		    address <= mem_info[i].end) {
229			/*
230			 * If we cover the rest of this phys_avail[] entry,
231			 * advance to the next entry.
232			 */
233			if (phys_avail[j + 1] <= mem_info[i].end) {
234				j += 2;
235				if (phys_avail[j] == 0 &&
236				    phys_avail[j + 1] == 0) {
237					return (0);
238				}
239				address = phys_avail[j];
240			} else
241				address = mem_info[i].end + 1;
242		}
243	}
244	printf("SRAT: No memory region found for %jx - %jx\n",
245	    (uintmax_t)phys_avail[j], (uintmax_t)phys_avail[j + 1]);
246	return (ENXIO);
247}
248
249/*
250 * Renumber the memory domains to be compact and zero-based if not
251 * already.  Returns an error if there are too many domains.
252 */
253static int
254renumber_domains(void)
255{
256	int i, j, slot;
257
258	/* Enumerate all the domains. */
259	vm_ndomains = 0;
260	for (i = 0; i < num_mem; i++) {
261		/* See if this domain is already known. */
262		for (j = 0; j < vm_ndomains; j++) {
263			if (vm_domains[j] >= mem_info[i].domain)
264				break;
265		}
266		if (j < vm_ndomains && vm_domains[j] == mem_info[i].domain)
267			continue;
268
269		/* Insert the new domain at slot 'j'. */
270		slot = j;
271		for (j = vm_ndomains; j > slot; j--)
272			vm_domains[j] = vm_domains[j - 1];
273		vm_domains[slot] = mem_info[i].domain;
274		vm_ndomains++;
275		if (vm_ndomains > MAXMEMDOM) {
276			vm_ndomains = 1;
277			printf("SRAT: Too many memory domains\n");
278			return (EFBIG);
279		}
280	}
281
282	/* Renumber each domain to its index in the sorted 'domains' list. */
283	for (i = 0; i < vm_ndomains; i++) {
284		/*
285		 * If the domain is already the right value, no need
286		 * to renumber.
287		 */
288		if (vm_domains[i] == i)
289			continue;
290
291		/* Walk the cpu[] and mem_info[] arrays to renumber. */
292		for (j = 0; j < num_mem; j++)
293			if (mem_info[j].domain == vm_domains[i])
294				mem_info[j].domain = i;
295		for (j = 0; j <= MAX_APIC_ID; j++)
296			if (cpus[j].enabled && cpus[j].domain == vm_domains[i])
297				cpus[j].domain = i;
298	}
299	KASSERT(vm_ndomains > 0,
300	    ("renumber_domains: invalid final vm_ndomains setup"));
301
302	return (0);
303}
304
305/*
306 * Look for an ACPI System Resource Affinity Table ("SRAT")
307 */
308static void
309parse_srat(void *dummy)
310{
311	int error;
312
313	if (resource_disabled("srat", 0))
314		return;
315
316	srat_physaddr = acpi_find_table(ACPI_SIG_SRAT);
317	if (srat_physaddr == 0)
318		return;
319
320	/*
321	 * Make a pass over the table to populate the cpus[] and
322	 * mem_info[] tables.
323	 */
324	srat = acpi_map_table(srat_physaddr, ACPI_SIG_SRAT);
325	error = 0;
326	srat_walk_table(srat_parse_entry, &error);
327	acpi_unmap_table(srat);
328	srat = NULL;
329	if (error || check_domains() != 0 || check_phys_avail() != 0 ||
330	    renumber_domains() != 0) {
331		srat_physaddr = 0;
332		return;
333	}
334
335	/* Point vm_phys at our memory affinity table. */
336	mem_affinity = mem_info;
337}
338SYSINIT(parse_srat, SI_SUB_VM - 1, SI_ORDER_FIRST, parse_srat, NULL);
339
340static void
341srat_walk_table(acpi_subtable_handler *handler, void *arg)
342{
343
344	acpi_walk_subtables(srat + 1, (char *)srat + srat->Header.Length,
345	    handler, arg);
346}
347
348/*
349 * Setup per-CPU ACPI IDs.
350 */
351static void
352srat_set_cpus(void *dummy)
353{
354	struct cpu_info *cpu;
355	struct pcpu *pc;
356	u_int i;
357
358	if (srat_physaddr == 0)
359		return;
360	for (i = 0; i < MAXCPU; i++) {
361		if (CPU_ABSENT(i))
362			continue;
363		pc = pcpu_find(i);
364		KASSERT(pc != NULL, ("no pcpu data for CPU %u", i));
365		cpu = &cpus[pc->pc_apic_id];
366		if (!cpu->enabled)
367			panic("SRAT: CPU with APIC ID %u is not known",
368			    pc->pc_apic_id);
369		pc->pc_domain = cpu->domain;
370		if (bootverbose)
371			printf("SRAT: CPU %u has memory domain %d\n", i,
372			    cpu->domain);
373	}
374}
375SYSINIT(srat_set_cpus, SI_SUB_CPU, SI_ORDER_ANY, srat_set_cpus, NULL);
376
377/*
378 * Map a _PXM value to a VM domain ID.
379 *
380 * Returns the domain ID, or -1 if no domain ID was found.
381 */
382int
383acpi_map_pxm_to_vm_domainid(int pxm)
384{
385	int i;
386
387	for (i = 0; i < vm_ndomains; i++) {
388		if (vm_domains[i] == pxm)
389			return (i);
390	}
391
392	return (-1);
393}
394
395#endif /* MAXMEMDOM > 1 */
396