2 * Copyright (c) 2015 Nathan Whitehorn
3 * Copyright (c) 2017-2018 Semihalf
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
41 #include <machine/bus.h>
42 #include <machine/cpu.h>
43 #include <machine/hid.h>
44 #include <machine/platformvar.h>
45 #include <machine/pmap.h>
46 #include <machine/rtas.h>
47 #include <machine/smp.h>
48 #include <machine/spr.h>
49 #include <machine/trap.h>
51 #include <dev/ofw/openfirm.h>
52 #include <machine/ofw_machdep.h>
53 #include <powerpc/aim/mmu_oea64.h>
55 #include "platform_if.h"
62 void (*powernv_smp_ap_extra_init)(void);
64 static int powernv_probe(platform_t);
65 static int powernv_attach(platform_t);
66 void powernv_mem_regions(platform_t, struct mem_region *phys, int *physsz,
67 struct mem_region *avail, int *availsz);
68 static void powernv_numa_mem_regions(platform_t plat, struct numa_mem_region *phys, int *physsz);
69 static u_long powernv_timebase_freq(platform_t, struct cpuref *cpuref);
70 static int powernv_smp_first_cpu(platform_t, struct cpuref *cpuref);
71 static int powernv_smp_next_cpu(platform_t, struct cpuref *cpuref);
72 static int powernv_smp_get_bsp(platform_t, struct cpuref *cpuref);
73 static void powernv_smp_ap_init(platform_t);
75 static int powernv_smp_start_cpu(platform_t, struct pcpu *cpu);
76 static void powernv_smp_probe_threads(platform_t);
77 static struct cpu_group *powernv_smp_topo(platform_t plat);
79 static void powernv_reset(platform_t);
80 static void powernv_cpu_idle(sbintime_t sbt);
81 static int powernv_cpuref_init(void);
82 static int powernv_node_numa_domain(platform_t platform, phandle_t node);
84 static platform_method_t powernv_methods[] = {
85 PLATFORMMETHOD(platform_probe, powernv_probe),
86 PLATFORMMETHOD(platform_attach, powernv_attach),
87 PLATFORMMETHOD(platform_mem_regions, powernv_mem_regions),
88 PLATFORMMETHOD(platform_numa_mem_regions, powernv_numa_mem_regions),
89 PLATFORMMETHOD(platform_timebase_freq, powernv_timebase_freq),
91 PLATFORMMETHOD(platform_smp_ap_init, powernv_smp_ap_init),
92 PLATFORMMETHOD(platform_smp_first_cpu, powernv_smp_first_cpu),
93 PLATFORMMETHOD(platform_smp_next_cpu, powernv_smp_next_cpu),
94 PLATFORMMETHOD(platform_smp_get_bsp, powernv_smp_get_bsp),
96 PLATFORMMETHOD(platform_smp_start_cpu, powernv_smp_start_cpu),
97 PLATFORMMETHOD(platform_smp_probe_threads, powernv_smp_probe_threads),
98 PLATFORMMETHOD(platform_smp_topo, powernv_smp_topo),
100 PLATFORMMETHOD(platform_node_numa_domain, powernv_node_numa_domain),
102 PLATFORMMETHOD(platform_reset, powernv_reset),
107 static platform_def_t powernv_platform = {
113 static struct cpuref platform_cpuref[MAXCPU];
114 static int platform_cpuref_cnt;
115 static int platform_cpuref_valid;
116 static int platform_associativity;
118 PLATFORM_DEF(powernv_platform);
120 static uint64_t powernv_boot_pir;
123 powernv_probe(platform_t plat)
125 if (opal_check() == 0)
126 return (BUS_PROBE_SPECIFIC);
132 powernv_attach(platform_t plat)
134 uint32_t nptlp, shift = 0, slb_encoding = 0;
135 int32_t lp_size, lp_encoding;
137 pcell_t refpoints[3];
144 /* Ping OPAL again just to make sure */
147 #if BYTE_ORDER == LITTLE_ENDIAN
148 opal_call(OPAL_REINIT_CPUS, 2 /* Little endian */);
150 opal_call(OPAL_REINIT_CPUS, 1 /* Big endian */);
152 opal = OF_finddevice("/ibm,opal");
154 platform_associativity = 4; /* Skiboot default. */
155 if (OF_getencprop(opal, "ibm,associativity-reference-points", refpoints,
156 sizeof(refpoints)) > 0) {
157 platform_associativity = refpoints[0];
160 if (cpu_idle_hook == NULL)
161 cpu_idle_hook = powernv_cpu_idle;
163 powernv_boot_pir = mfspr(SPR_PIR);
165 /* LPID must not be altered when PSL_DR or PSL_IR is set */
167 mtmsr(msr & ~(PSL_DR | PSL_IR));
169 /* Direct interrupts to SRR instead of HSRR and reset LPCR otherwise */
173 if (cpu_features2 & PPC_FEATURE2_ARCH_3_00)
176 mtspr(SPR_LPCR, lpcr);
181 powernv_cpuref_init();
183 /* Set SLB count from device tree */
187 res = OF_getprop(cpu, "name", buf, sizeof(buf));
188 if (res > 0 && strcmp(buf, "cpus") == 0)
197 res = OF_getprop(cpu, "device_type", buf, sizeof(buf));
198 if (res > 0 && strcmp(buf, "cpu") == 0)
205 res = OF_getencprop(cpu, "ibm,slb-size", &prop, sizeof(prop));
210 * Scan the large page size property for PAPR compatible machines.
211 * See PAPR D.5 Changes to Section 5.1.4, 'CPU Node Properties'
212 * for the encoding of the property.
215 len = OF_getproplen(cpu, "ibm,segment-page-sizes");
218 * We have to use a variable length array on the stack
219 * since we have very limited stack space.
221 pcell_t arr[len/sizeof(cell_t)];
222 res = OF_getencprop(cpu, "ibm,segment-page-sizes", arr,
228 slb_encoding = arr[idx + 1];
229 nptlp = arr[idx + 2];
232 while (len > 0 && nptlp) {
234 lp_encoding = arr[idx+1];
235 if (slb_encoding == SLBV_L && lp_encoding == 0)
242 if (nptlp && slb_encoding == SLBV_L && lp_encoding == 0)
247 panic("Standard large pages (SLB[L] = 1, PTE[LP] = 0) "
248 "not supported by this system.");
250 moea64_large_page_shift = shift;
251 moea64_large_page_size = 1ULL << lp_size;
260 powernv_mem_regions(platform_t plat, struct mem_region *phys, int *physsz,
261 struct mem_region *avail, int *availsz)
264 ofw_mem_regions(phys, physsz, avail, availsz);
268 powernv_numa_mem_regions(platform_t plat, struct numa_mem_region *phys, int *physsz)
271 ofw_numa_mem_regions(phys, physsz);
275 powernv_timebase_freq(platform_t plat, struct cpuref *cpuref)
278 phandle_t cpu, dev, root;
283 dev = OF_child(root);
285 res = OF_getprop(dev, "name", buf, sizeof(buf));
286 if (res > 0 && strcmp(buf, "cpus") == 0)
291 for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) {
292 res = OF_getprop(cpu, "device_type", buf, sizeof(buf));
293 if (res > 0 && strcmp(buf, "cpu") == 0)
299 OF_getencprop(cpu, "timebase-frequency", &ticks, sizeof(ticks));
302 panic("Unable to determine timebase frequency!");
309 powernv_cpuref_init(void)
313 int a, res, tmp_cpuref_cnt;
314 static struct cpuref tmp_cpuref[MAXCPU];
315 cell_t interrupt_servers[32];
318 if (platform_cpuref_valid)
324 res = OF_getprop(dev, "name", buf, sizeof(buf));
325 if (res > 0 && strcmp(buf, "cpus") == 0)
332 for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) {
333 res = OF_getprop(cpu, "device_type", buf, sizeof(buf));
334 if (res > 0 && strcmp(buf, "cpu") == 0) {
335 res = OF_getproplen(cpu, "ibm,ppc-interrupt-server#s");
337 OF_getencprop(cpu, "ibm,ppc-interrupt-server#s",
338 interrupt_servers, res);
340 for (a = 0; a < res/sizeof(cell_t); a++) {
341 tmp_cpuref[tmp_cpuref_cnt].cr_hwref = interrupt_servers[a];
342 tmp_cpuref[tmp_cpuref_cnt].cr_cpuid = tmp_cpuref_cnt;
343 tmp_cpuref[tmp_cpuref_cnt].cr_domain =
344 powernv_node_numa_domain(NULL, cpu);
345 if (interrupt_servers[a] == (uint32_t)powernv_boot_pir)
346 bsp = tmp_cpuref_cnt;
354 /* Map IDs, so BSP has CPUID 0 regardless of hwref */
355 for (a = bsp; a < tmp_cpuref_cnt; a++) {
356 platform_cpuref[platform_cpuref_cnt].cr_hwref = tmp_cpuref[a].cr_hwref;
357 platform_cpuref[platform_cpuref_cnt].cr_cpuid = platform_cpuref_cnt;
358 platform_cpuref[platform_cpuref_cnt].cr_domain = tmp_cpuref[a].cr_domain;
359 platform_cpuref_cnt++;
361 for (a = 0; a < bsp; a++) {
362 platform_cpuref[platform_cpuref_cnt].cr_hwref = tmp_cpuref[a].cr_hwref;
363 platform_cpuref[platform_cpuref_cnt].cr_cpuid = platform_cpuref_cnt;
364 platform_cpuref[platform_cpuref_cnt].cr_domain = tmp_cpuref[a].cr_domain;
365 platform_cpuref_cnt++;
368 platform_cpuref_valid = 1;
374 powernv_smp_first_cpu(platform_t plat, struct cpuref *cpuref)
376 if (platform_cpuref_valid == 0)
379 cpuref->cr_cpuid = 0;
380 cpuref->cr_hwref = platform_cpuref[0].cr_hwref;
381 cpuref->cr_domain = platform_cpuref[0].cr_domain;
387 powernv_smp_next_cpu(platform_t plat, struct cpuref *cpuref)
391 if (platform_cpuref_valid == 0)
394 id = cpuref->cr_cpuid + 1;
395 if (id >= platform_cpuref_cnt)
398 cpuref->cr_cpuid = platform_cpuref[id].cr_cpuid;
399 cpuref->cr_hwref = platform_cpuref[id].cr_hwref;
400 cpuref->cr_domain = platform_cpuref[id].cr_domain;
406 powernv_smp_get_bsp(platform_t plat, struct cpuref *cpuref)
409 cpuref->cr_cpuid = platform_cpuref[0].cr_cpuid;
410 cpuref->cr_hwref = platform_cpuref[0].cr_hwref;
411 cpuref->cr_domain = platform_cpuref[0].cr_domain;
417 powernv_smp_start_cpu(platform_t plat, struct pcpu *pc)
424 result = opal_call(OPAL_START_CPU, pc->pc_hwref, EXC_RST);
425 if (result != OPAL_SUCCESS) {
426 printf("OPAL error (%d): unable to start AP %d\n",
427 result, (int)pc->pc_hwref);
435 powernv_smp_probe_threads(platform_t plat)
438 phandle_t cpu, dev, root;
443 dev = OF_child(root);
445 res = OF_getprop(dev, "name", buf, sizeof(buf));
446 if (res > 0 && strcmp(buf, "cpus") == 0)
452 for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) {
453 res = OF_getprop(cpu, "device_type", buf, sizeof(buf));
454 if (res <= 0 || strcmp(buf, "cpu") != 0)
457 res = OF_getproplen(cpu, "ibm,ppc-interrupt-server#s");
460 nthreads = res / sizeof(cell_t);
466 smp_threads_per_core = nthreads;
467 if (mp_ncpus % nthreads == 0)
468 mp_ncores = mp_ncpus / nthreads;
471 static struct cpu_group *
472 powernv_smp_topo(platform_t plat)
474 if (mp_ncpus % smp_threads_per_core != 0) {
475 printf("WARNING: Irregular SMP topology. Performance may be "
476 "suboptimal (%d threads, %d on first core)\n",
477 mp_ncpus, smp_threads_per_core);
478 return (smp_topo_none());
481 /* Don't do anything fancier for non-threaded SMP */
482 if (smp_threads_per_core == 1)
483 return (smp_topo_none());
485 return (smp_topo_1level(CG_SHARE_L1, smp_threads_per_core,
492 powernv_reset(platform_t platform)
495 opal_call(OPAL_CEC_REBOOT);
499 powernv_smp_ap_init(platform_t platform)
502 if (powernv_smp_ap_extra_init != NULL)
503 powernv_smp_ap_extra_init();
507 powernv_cpu_idle(sbintime_t sbt)
512 powernv_node_numa_domain(platform_t platform, phandle_t node)
514 /* XXX: Is locking necessary in here? */
515 static int numa_domains[MAXMEMDOM];
516 static int numa_max_domain;
517 cell_t associativity[5];
523 if (vm_ndomains == 1)
526 res = OF_getencprop(node, "ibm,associativity",
527 associativity, sizeof(associativity));
530 * If this node doesn't have associativity, or if there are not
531 * enough elements in it, check its parent.
533 if (res < (int)(sizeof(cell_t) * (platform_associativity + 1))) {
534 node = OF_parent(node);
535 /* If already at the root, use default domain. */
538 return (powernv_node_numa_domain(platform, node));
541 for (i = 0; i < numa_max_domain; i++) {
542 if (numa_domains[i] == associativity[platform_associativity])
546 numa_domains[numa_max_domain++] =
547 associativity[platform_associativity];
554 /* Set up the Nest MMU on POWER9 relatively early, but after pmap is setup. */
556 powernv_setup_nmmu(void *unused)
558 if (opal_check() != 0)
560 opal_call(OPAL_NMMU_SET_PTCR, -1, mfspr(SPR_PTCR));
563 SYSINIT(powernv_setup_nmmu, SI_SUB_CPU, SI_ORDER_ANY, powernv_setup_nmmu, NULL);