Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Common boot and setup code for both 32-bit and 64-bit.
4 * Extracted from arch/powerpc/kernel/setup_64.c.
5 *
6 * Copyright (C) 2001 PPC64 Team, IBM Corp
7 */
8
9#undef DEBUG
10
11#include <linux/export.h>
12#include <linux/panic_notifier.h>
13#include <linux/string.h>
14#include <linux/sched.h>
15#include <linux/init.h>
16#include <linux/kernel.h>
17#include <linux/reboot.h>
18#include <linux/delay.h>
19#include <linux/initrd.h>
20#include <linux/platform_device.h>
21#include <linux/printk.h>
22#include <linux/seq_file.h>
23#include <linux/ioport.h>
24#include <linux/console.h>
25#include <linux/root_dev.h>
26#include <linux/cpu.h>
27#include <linux/unistd.h>
28#include <linux/seq_buf.h>
29#include <linux/serial.h>
30#include <linux/serial_8250.h>
31#include <linux/percpu.h>
32#include <linux/memblock.h>
33#include <linux/of.h>
34#include <linux/of_fdt.h>
35#include <linux/of_irq.h>
36#include <linux/hugetlb.h>
37#include <linux/pgtable.h>
38#include <asm/io.h>
39#include <asm/paca.h>
40#include <asm/processor.h>
41#include <asm/vdso_datapage.h>
42#include <asm/smp.h>
43#include <asm/elf.h>
44#include <asm/machdep.h>
45#include <asm/time.h>
46#include <asm/cputable.h>
47#include <asm/sections.h>
48#include <asm/firmware.h>
49#include <asm/btext.h>
50#include <asm/nvram.h>
51#include <asm/setup.h>
52#include <asm/rtas.h>
53#include <asm/iommu.h>
54#include <asm/serial.h>
55#include <asm/cache.h>
56#include <asm/page.h>
57#include <asm/mmu.h>
58#include <asm/xmon.h>
59#include <asm/cputhreads.h>
60#include <mm/mmu_decl.h>
61#include <asm/archrandom.h>
62#include <asm/fadump.h>
63#include <asm/udbg.h>
64#include <asm/hugetlb.h>
65#include <asm/livepatch.h>
66#include <asm/mmu_context.h>
67#include <asm/cpu_has_feature.h>
68#include <asm/kasan.h>
69#include <asm/mce.h>
70
71#include "setup.h"
72
73#ifdef DEBUG
74#define DBG(fmt...) udbg_printf(fmt)
75#else
76#define DBG(fmt...)
77#endif
78
79/* The main machine-dep calls structure
80 */
81struct machdep_calls ppc_md;
82EXPORT_SYMBOL(ppc_md);
83struct machdep_calls *machine_id;
84EXPORT_SYMBOL(machine_id);
85
86int boot_cpuid = -1;
87EXPORT_SYMBOL_GPL(boot_cpuid);
88
89#ifdef CONFIG_PPC64
90int boot_cpu_hwid = -1;
91#endif
92
93/*
94 * These are used in binfmt_elf.c to put aux entries on the stack
95 * for each elf executable being started.
96 */
97int dcache_bsize;
98int icache_bsize;
99
100/* Variables required to store legacy IO irq routing */
101int of_i8042_kbd_irq;
102EXPORT_SYMBOL_GPL(of_i8042_kbd_irq);
103int of_i8042_aux_irq;
104EXPORT_SYMBOL_GPL(of_i8042_aux_irq);
105
106#ifdef __DO_IRQ_CANON
107/* XXX should go elsewhere eventually */
108int ppc_do_canonicalize_irqs;
109EXPORT_SYMBOL(ppc_do_canonicalize_irqs);
110#endif
111
112#ifdef CONFIG_CRASH_CORE
113/* This keeps a track of which one is the crashing cpu. */
114int crashing_cpu = -1;
115#endif
116
117/* also used by kexec */
118void machine_shutdown(void)
119{
120 /*
121 * if fadump is active, cleanup the fadump registration before we
122 * shutdown.
123 */
124 fadump_cleanup();
125
126 if (ppc_md.machine_shutdown)
127 ppc_md.machine_shutdown();
128}
129
130static void machine_hang(void)
131{
132 pr_emerg("System Halted, OK to turn off power\n");
133 local_irq_disable();
134 while (1)
135 ;
136}
137
138void machine_restart(char *cmd)
139{
140 machine_shutdown();
141 if (ppc_md.restart)
142 ppc_md.restart(cmd);
143
144 smp_send_stop();
145
146 do_kernel_restart(cmd);
147 mdelay(1000);
148
149 machine_hang();
150}
151
152void machine_power_off(void)
153{
154 machine_shutdown();
155 do_kernel_power_off();
156 smp_send_stop();
157 machine_hang();
158}
159/* Used by the G5 thermal driver */
160EXPORT_SYMBOL_GPL(machine_power_off);
161
162void (*pm_power_off)(void);
163EXPORT_SYMBOL_GPL(pm_power_off);
164
165size_t __must_check arch_get_random_seed_longs(unsigned long *v, size_t max_longs)
166{
167 if (max_longs && ppc_md.get_random_seed && ppc_md.get_random_seed(v))
168 return 1;
169 return 0;
170}
171EXPORT_SYMBOL(arch_get_random_seed_longs);
172
173void machine_halt(void)
174{
175 machine_shutdown();
176 if (ppc_md.halt)
177 ppc_md.halt();
178
179 smp_send_stop();
180 machine_hang();
181}
182
183#ifdef CONFIG_SMP
184DEFINE_PER_CPU(unsigned int, cpu_pvr);
185#endif
186
187static void show_cpuinfo_summary(struct seq_file *m)
188{
189 struct device_node *root;
190 const char *model = NULL;
191 unsigned long bogosum = 0;
192 int i;
193
194 if (IS_ENABLED(CONFIG_SMP) && IS_ENABLED(CONFIG_PPC32)) {
195 for_each_online_cpu(i)
196 bogosum += loops_per_jiffy;
197 seq_printf(m, "total bogomips\t: %lu.%02lu\n",
198 bogosum / (500000 / HZ), bogosum / (5000 / HZ) % 100);
199 }
200 seq_printf(m, "timebase\t: %lu\n", ppc_tb_freq);
201 if (ppc_md.name)
202 seq_printf(m, "platform\t: %s\n", ppc_md.name);
203 root = of_find_node_by_path("/");
204 if (root)
205 model = of_get_property(root, "model", NULL);
206 if (model)
207 seq_printf(m, "model\t\t: %s\n", model);
208 of_node_put(root);
209
210 if (ppc_md.show_cpuinfo != NULL)
211 ppc_md.show_cpuinfo(m);
212
213 /* Display the amount of memory */
214 if (IS_ENABLED(CONFIG_PPC32))
215 seq_printf(m, "Memory\t\t: %d MB\n",
216 (unsigned int)(total_memory / (1024 * 1024)));
217}
218
219static int show_cpuinfo(struct seq_file *m, void *v)
220{
221 unsigned long cpu_id = (unsigned long)v - 1;
222 unsigned int pvr;
223 unsigned long proc_freq;
224 unsigned short maj;
225 unsigned short min;
226
227#ifdef CONFIG_SMP
228 pvr = per_cpu(cpu_pvr, cpu_id);
229#else
230 pvr = mfspr(SPRN_PVR);
231#endif
232 maj = (pvr >> 8) & 0xFF;
233 min = pvr & 0xFF;
234
235 seq_printf(m, "processor\t: %lu\ncpu\t\t: ", cpu_id);
236
237 if (cur_cpu_spec->pvr_mask && cur_cpu_spec->cpu_name)
238 seq_puts(m, cur_cpu_spec->cpu_name);
239 else
240 seq_printf(m, "unknown (%08x)", pvr);
241
242 if (cpu_has_feature(CPU_FTR_ALTIVEC))
243 seq_puts(m, ", altivec supported");
244
245 seq_putc(m, '\n');
246
247#ifdef CONFIG_TAU
248 if (cpu_has_feature(CPU_FTR_TAU)) {
249 if (IS_ENABLED(CONFIG_TAU_AVERAGE)) {
250 /* more straightforward, but potentially misleading */
251 seq_printf(m, "temperature \t: %u C (uncalibrated)\n",
252 cpu_temp(cpu_id));
253 } else {
254 /* show the actual temp sensor range */
255 u32 temp;
256 temp = cpu_temp_both(cpu_id);
257 seq_printf(m, "temperature \t: %u-%u C (uncalibrated)\n",
258 temp & 0xff, temp >> 16);
259 }
260 }
261#endif /* CONFIG_TAU */
262
263 /*
264 * Platforms that have variable clock rates, should implement
265 * the method ppc_md.get_proc_freq() that reports the clock
266 * rate of a given cpu. The rest can use ppc_proc_freq to
267 * report the clock rate that is same across all cpus.
268 */
269 if (ppc_md.get_proc_freq)
270 proc_freq = ppc_md.get_proc_freq(cpu_id);
271 else
272 proc_freq = ppc_proc_freq;
273
274 if (proc_freq)
275 seq_printf(m, "clock\t\t: %lu.%06luMHz\n",
276 proc_freq / 1000000, proc_freq % 1000000);
277
278 /* If we are a Freescale core do a simple check so
279 * we don't have to keep adding cases in the future */
280 if (PVR_VER(pvr) & 0x8000) {
281 switch (PVR_VER(pvr)) {
282 case 0x8000: /* 7441/7450/7451, Voyager */
283 case 0x8001: /* 7445/7455, Apollo 6 */
284 case 0x8002: /* 7447/7457, Apollo 7 */
285 case 0x8003: /* 7447A, Apollo 7 PM */
286 case 0x8004: /* 7448, Apollo 8 */
287 case 0x800c: /* 7410, Nitro */
288 maj = ((pvr >> 8) & 0xF);
289 min = PVR_MIN(pvr);
290 break;
291 default: /* e500/book-e */
292 maj = PVR_MAJ(pvr);
293 min = PVR_MIN(pvr);
294 break;
295 }
296 } else {
297 switch (PVR_VER(pvr)) {
298 case 0x1008: /* 740P/750P ?? */
299 maj = ((pvr >> 8) & 0xFF) - 1;
300 min = pvr & 0xFF;
301 break;
302 case 0x004e: /* POWER9 bits 12-15 give chip type */
303 case 0x0080: /* POWER10 bit 12 gives SMT8/4 */
304 maj = (pvr >> 8) & 0x0F;
305 min = pvr & 0xFF;
306 break;
307 default:
308 maj = (pvr >> 8) & 0xFF;
309 min = pvr & 0xFF;
310 break;
311 }
312 }
313
314 seq_printf(m, "revision\t: %hd.%hd (pvr %04x %04x)\n",
315 maj, min, PVR_VER(pvr), PVR_REV(pvr));
316
317 if (IS_ENABLED(CONFIG_PPC32))
318 seq_printf(m, "bogomips\t: %lu.%02lu\n", loops_per_jiffy / (500000 / HZ),
319 (loops_per_jiffy / (5000 / HZ)) % 100);
320
321 seq_putc(m, '\n');
322
323 /* If this is the last cpu, print the summary */
324 if (cpumask_next(cpu_id, cpu_online_mask) >= nr_cpu_ids)
325 show_cpuinfo_summary(m);
326
327 return 0;
328}
329
330static void *c_start(struct seq_file *m, loff_t *pos)
331{
332 if (*pos == 0) /* just in case, cpu 0 is not the first */
333 *pos = cpumask_first(cpu_online_mask);
334 else
335 *pos = cpumask_next(*pos - 1, cpu_online_mask);
336 if ((*pos) < nr_cpu_ids)
337 return (void *)(unsigned long)(*pos + 1);
338 return NULL;
339}
340
341static void *c_next(struct seq_file *m, void *v, loff_t *pos)
342{
343 (*pos)++;
344 return c_start(m, pos);
345}
346
347static void c_stop(struct seq_file *m, void *v)
348{
349}
350
351const struct seq_operations cpuinfo_op = {
352 .start = c_start,
353 .next = c_next,
354 .stop = c_stop,
355 .show = show_cpuinfo,
356};
357
358void __init check_for_initrd(void)
359{
360#ifdef CONFIG_BLK_DEV_INITRD
361 DBG(" -> check_for_initrd() initrd_start=0x%lx initrd_end=0x%lx\n",
362 initrd_start, initrd_end);
363
364 /* If we were passed an initrd, set the ROOT_DEV properly if the values
365 * look sensible. If not, clear initrd reference.
366 */
367 if (is_kernel_addr(initrd_start) && is_kernel_addr(initrd_end) &&
368 initrd_end > initrd_start)
369 ROOT_DEV = Root_RAM0;
370 else
371 initrd_start = initrd_end = 0;
372
373 if (initrd_start)
374 pr_info("Found initrd at 0x%lx:0x%lx\n", initrd_start, initrd_end);
375
376 DBG(" <- check_for_initrd()\n");
377#endif /* CONFIG_BLK_DEV_INITRD */
378}
379
380#ifdef CONFIG_SMP
381
382int threads_per_core, threads_per_subcore, threads_shift __read_mostly;
383cpumask_t threads_core_mask __read_mostly;
384EXPORT_SYMBOL_GPL(threads_per_core);
385EXPORT_SYMBOL_GPL(threads_per_subcore);
386EXPORT_SYMBOL_GPL(threads_shift);
387EXPORT_SYMBOL_GPL(threads_core_mask);
388
389static void __init cpu_init_thread_core_maps(int tpc)
390{
391 int i;
392
393 threads_per_core = tpc;
394 threads_per_subcore = tpc;
395 cpumask_clear(&threads_core_mask);
396
397 /* This implementation only supports power of 2 number of threads
398 * for simplicity and performance
399 */
400 threads_shift = ilog2(tpc);
401 BUG_ON(tpc != (1 << threads_shift));
402
403 for (i = 0; i < tpc; i++)
404 cpumask_set_cpu(i, &threads_core_mask);
405
406 printk(KERN_INFO "CPU maps initialized for %d thread%s per core\n",
407 tpc, tpc > 1 ? "s" : "");
408 printk(KERN_DEBUG " (thread shift is %d)\n", threads_shift);
409}
410
411
412u32 *cpu_to_phys_id = NULL;
413
414/**
415 * setup_cpu_maps - initialize the following cpu maps:
416 * cpu_possible_mask
417 * cpu_present_mask
418 *
419 * Having the possible map set up early allows us to restrict allocations
420 * of things like irqstacks to nr_cpu_ids rather than NR_CPUS.
421 *
422 * We do not initialize the online map here; cpus set their own bits in
423 * cpu_online_mask as they come up.
424 *
425 * This function is valid only for Open Firmware systems. finish_device_tree
426 * must be called before using this.
427 *
428 * While we're here, we may as well set the "physical" cpu ids in the paca.
429 *
430 * NOTE: This must match the parsing done in early_init_dt_scan_cpus.
431 */
432void __init smp_setup_cpu_maps(void)
433{
434 struct device_node *dn;
435 int cpu = 0;
436 int nthreads = 1;
437
438 DBG("smp_setup_cpu_maps()\n");
439
440 cpu_to_phys_id = memblock_alloc(nr_cpu_ids * sizeof(u32),
441 __alignof__(u32));
442 if (!cpu_to_phys_id)
443 panic("%s: Failed to allocate %zu bytes align=0x%zx\n",
444 __func__, nr_cpu_ids * sizeof(u32), __alignof__(u32));
445
446 for_each_node_by_type(dn, "cpu") {
447 const __be32 *intserv;
448 __be32 cpu_be;
449 int j, len;
450
451 DBG(" * %pOF...\n", dn);
452
453 intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s",
454 &len);
455 if (intserv) {
456 DBG(" ibm,ppc-interrupt-server#s -> %lu threads\n",
457 (len / sizeof(int)));
458 } else {
459 DBG(" no ibm,ppc-interrupt-server#s -> 1 thread\n");
460 intserv = of_get_property(dn, "reg", &len);
461 if (!intserv) {
462 cpu_be = cpu_to_be32(cpu);
463 /* XXX: what is this? uninitialized?? */
464 intserv = &cpu_be; /* assume logical == phys */
465 len = 4;
466 }
467 }
468
469 nthreads = len / sizeof(int);
470
471 for (j = 0; j < nthreads && cpu < nr_cpu_ids; j++) {
472 bool avail;
473
474 DBG(" thread %d -> cpu %d (hard id %d)\n",
475 j, cpu, be32_to_cpu(intserv[j]));
476
477 avail = of_device_is_available(dn);
478 if (!avail)
479 avail = !of_property_match_string(dn,
480 "enable-method", "spin-table");
481
482 set_cpu_present(cpu, avail);
483 set_cpu_possible(cpu, true);
484 cpu_to_phys_id[cpu] = be32_to_cpu(intserv[j]);
485 cpu++;
486 }
487
488 if (cpu >= nr_cpu_ids) {
489 of_node_put(dn);
490 break;
491 }
492 }
493
494 /* If no SMT supported, nthreads is forced to 1 */
495 if (!cpu_has_feature(CPU_FTR_SMT)) {
496 DBG(" SMT disabled ! nthreads forced to 1\n");
497 nthreads = 1;
498 }
499
500#ifdef CONFIG_PPC64
501 /*
502 * On pSeries LPAR, we need to know how many cpus
503 * could possibly be added to this partition.
504 */
505 if (firmware_has_feature(FW_FEATURE_LPAR) &&
506 (dn = of_find_node_by_path("/rtas"))) {
507 int num_addr_cell, num_size_cell, maxcpus;
508 const __be32 *ireg;
509
510 num_addr_cell = of_n_addr_cells(dn);
511 num_size_cell = of_n_size_cells(dn);
512
513 ireg = of_get_property(dn, "ibm,lrdr-capacity", NULL);
514
515 if (!ireg)
516 goto out;
517
518 maxcpus = be32_to_cpup(ireg + num_addr_cell + num_size_cell);
519
520 /* Double maxcpus for processors which have SMT capability */
521 if (cpu_has_feature(CPU_FTR_SMT))
522 maxcpus *= nthreads;
523
524 if (maxcpus > nr_cpu_ids) {
525 printk(KERN_WARNING
526 "Partition configured for %d cpus, "
527 "operating system maximum is %u.\n",
528 maxcpus, nr_cpu_ids);
529 maxcpus = nr_cpu_ids;
530 } else
531 printk(KERN_INFO "Partition configured for %d cpus.\n",
532 maxcpus);
533
534 for (cpu = 0; cpu < maxcpus; cpu++)
535 set_cpu_possible(cpu, true);
536 out:
537 of_node_put(dn);
538 }
539 vdso_data->processorCount = num_present_cpus();
540#endif /* CONFIG_PPC64 */
541
542 /* Initialize CPU <=> thread mapping/
543 *
544 * WARNING: We assume that the number of threads is the same for
545 * every CPU in the system. If that is not the case, then some code
546 * here will have to be reworked
547 */
548 cpu_init_thread_core_maps(nthreads);
549
550 /* Now that possible cpus are set, set nr_cpu_ids for later use */
551 setup_nr_cpu_ids();
552
553 free_unused_pacas();
554}
555#endif /* CONFIG_SMP */
556
557#ifdef CONFIG_PCSPKR_PLATFORM
558static __init int add_pcspkr(void)
559{
560 struct device_node *np;
561 struct platform_device *pd;
562 int ret;
563
564 np = of_find_compatible_node(NULL, NULL, "pnpPNP,100");
565 of_node_put(np);
566 if (!np)
567 return -ENODEV;
568
569 pd = platform_device_alloc("pcspkr", -1);
570 if (!pd)
571 return -ENOMEM;
572
573 ret = platform_device_add(pd);
574 if (ret)
575 platform_device_put(pd);
576
577 return ret;
578}
579device_initcall(add_pcspkr);
580#endif /* CONFIG_PCSPKR_PLATFORM */
581
582static char ppc_hw_desc_buf[128] __initdata;
583
584struct seq_buf ppc_hw_desc __initdata = {
585 .buffer = ppc_hw_desc_buf,
586 .size = sizeof(ppc_hw_desc_buf),
587 .len = 0,
588};
589
590static __init void probe_machine(void)
591{
592 extern struct machdep_calls __machine_desc_start;
593 extern struct machdep_calls __machine_desc_end;
594 unsigned int i;
595
596 /*
597 * Iterate all ppc_md structures until we find the proper
598 * one for the current machine type
599 */
600 DBG("Probing machine type ...\n");
601
602 /*
603 * Check ppc_md is empty, if not we have a bug, ie, we setup an
604 * entry before probe_machine() which will be overwritten
605 */
606 for (i = 0; i < (sizeof(ppc_md) / sizeof(void *)); i++) {
607 if (((void **)&ppc_md)[i]) {
608 printk(KERN_ERR "Entry %d in ppc_md non empty before"
609 " machine probe !\n", i);
610 }
611 }
612
613 for (machine_id = &__machine_desc_start;
614 machine_id < &__machine_desc_end;
615 machine_id++) {
616 DBG(" %s ...\n", machine_id->name);
617 if (machine_id->compatible && !of_machine_is_compatible(machine_id->compatible))
618 continue;
619 memcpy(&ppc_md, machine_id, sizeof(struct machdep_calls));
620 if (ppc_md.probe && !ppc_md.probe())
621 continue;
622 DBG(" %s match !\n", machine_id->name);
623 break;
624 }
625 /* What can we do if we didn't find ? */
626 if (machine_id >= &__machine_desc_end) {
627 pr_err("No suitable machine description found !\n");
628 for (;;);
629 }
630
631 // Append the machine name to other info we've gathered
632 seq_buf_puts(&ppc_hw_desc, ppc_md.name);
633
634 // Set the generic hardware description shown in oopses
635 dump_stack_set_arch_desc(ppc_hw_desc.buffer);
636
637 pr_info("Hardware name: %s\n", ppc_hw_desc.buffer);
638}
639
640/* Match a class of boards, not a specific device configuration. */
641int check_legacy_ioport(unsigned long base_port)
642{
643 struct device_node *parent, *np = NULL;
644 int ret = -ENODEV;
645
646 switch(base_port) {
647 case I8042_DATA_REG:
648 if (!(np = of_find_compatible_node(NULL, NULL, "pnpPNP,303")))
649 np = of_find_compatible_node(NULL, NULL, "pnpPNP,f03");
650 if (np) {
651 parent = of_get_parent(np);
652
653 of_i8042_kbd_irq = irq_of_parse_and_map(parent, 0);
654 if (!of_i8042_kbd_irq)
655 of_i8042_kbd_irq = 1;
656
657 of_i8042_aux_irq = irq_of_parse_and_map(parent, 1);
658 if (!of_i8042_aux_irq)
659 of_i8042_aux_irq = 12;
660
661 of_node_put(np);
662 np = parent;
663 break;
664 }
665 np = of_find_node_by_type(NULL, "8042");
666 /* Pegasos has no device_type on its 8042 node, look for the
667 * name instead */
668 if (!np)
669 np = of_find_node_by_name(NULL, "8042");
670 if (np) {
671 of_i8042_kbd_irq = 1;
672 of_i8042_aux_irq = 12;
673 }
674 break;
675 case FDC_BASE: /* FDC1 */
676 np = of_find_node_by_type(NULL, "fdc");
677 break;
678 default:
679 /* ipmi is supposed to fail here */
680 break;
681 }
682 if (!np)
683 return ret;
684 parent = of_get_parent(np);
685 if (parent) {
686 if (of_node_is_type(parent, "isa"))
687 ret = 0;
688 of_node_put(parent);
689 }
690 of_node_put(np);
691 return ret;
692}
693EXPORT_SYMBOL(check_legacy_ioport);
694
695/*
696 * Panic notifiers setup
697 *
698 * We have 3 notifiers for powerpc, each one from a different "nature":
699 *
700 * - ppc_panic_fadump_handler() is a hypervisor notifier, which hard-disables
701 * IRQs and deal with the Firmware-Assisted dump, when it is configured;
702 * should run early in the panic path.
703 *
704 * - dump_kernel_offset() is an informative notifier, just showing the KASLR
705 * offset if we have RANDOMIZE_BASE set.
706 *
707 * - ppc_panic_platform_handler() is a low-level handler that's registered
708 * only if the platform wishes to perform final actions in the panic path,
709 * hence it should run late and might not even return. Currently, only
710 * pseries and ps3 platforms register callbacks.
711 */
712static int ppc_panic_fadump_handler(struct notifier_block *this,
713 unsigned long event, void *ptr)
714{
715 /*
716 * panic does a local_irq_disable, but we really
717 * want interrupts to be hard disabled.
718 */
719 hard_irq_disable();
720
721 /*
722 * If firmware-assisted dump has been registered then trigger
723 * its callback and let the firmware handles everything else.
724 */
725 crash_fadump(NULL, ptr);
726
727 return NOTIFY_DONE;
728}
729
730static int dump_kernel_offset(struct notifier_block *self, unsigned long v,
731 void *p)
732{
733 pr_emerg("Kernel Offset: 0x%lx from 0x%lx\n",
734 kaslr_offset(), KERNELBASE);
735
736 return NOTIFY_DONE;
737}
738
739static int ppc_panic_platform_handler(struct notifier_block *this,
740 unsigned long event, void *ptr)
741{
742 /*
743 * This handler is only registered if we have a panic callback
744 * on ppc_md, hence NULL check is not needed.
745 * Also, it may not return, so it runs really late on panic path.
746 */
747 ppc_md.panic(ptr);
748
749 return NOTIFY_DONE;
750}
751
752static struct notifier_block ppc_fadump_block = {
753 .notifier_call = ppc_panic_fadump_handler,
754 .priority = INT_MAX, /* run early, to notify the firmware ASAP */
755};
756
757static struct notifier_block kernel_offset_notifier = {
758 .notifier_call = dump_kernel_offset,
759};
760
761static struct notifier_block ppc_panic_block = {
762 .notifier_call = ppc_panic_platform_handler,
763 .priority = INT_MIN, /* may not return; must be done last */
764};
765
766void __init setup_panic(void)
767{
768 /* Hard-disables IRQs + deal with FW-assisted dump (fadump) */
769 atomic_notifier_chain_register(&panic_notifier_list,
770 &ppc_fadump_block);
771
772 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && kaslr_offset() > 0)
773 atomic_notifier_chain_register(&panic_notifier_list,
774 &kernel_offset_notifier);
775
776 /* Low-level platform-specific routines that should run on panic */
777 if (ppc_md.panic)
778 atomic_notifier_chain_register(&panic_notifier_list,
779 &ppc_panic_block);
780}
781
782#ifdef CONFIG_CHECK_CACHE_COHERENCY
783/*
784 * For platforms that have configurable cache-coherency. This function
785 * checks that the cache coherency setting of the kernel matches the setting
786 * left by the firmware, as indicated in the device tree. Since a mismatch
787 * will eventually result in DMA failures, we print * and error and call
788 * BUG() in that case.
789 */
790
791#define KERNEL_COHERENCY (!IS_ENABLED(CONFIG_NOT_COHERENT_CACHE))
792
793static int __init check_cache_coherency(void)
794{
795 struct device_node *np;
796 const void *prop;
797 bool devtree_coherency;
798
799 np = of_find_node_by_path("/");
800 prop = of_get_property(np, "coherency-off", NULL);
801 of_node_put(np);
802
803 devtree_coherency = prop ? false : true;
804
805 if (devtree_coherency != KERNEL_COHERENCY) {
806 printk(KERN_ERR
807 "kernel coherency:%s != device tree_coherency:%s\n",
808 KERNEL_COHERENCY ? "on" : "off",
809 devtree_coherency ? "on" : "off");
810 BUG();
811 }
812
813 return 0;
814}
815
816late_initcall(check_cache_coherency);
817#endif /* CONFIG_CHECK_CACHE_COHERENCY */
818
819void ppc_printk_progress(char *s, unsigned short hex)
820{
821 pr_info("%s\n", s);
822}
823
824static __init void print_system_info(void)
825{
826 pr_info("-----------------------------------------------------\n");
827 pr_info("phys_mem_size = 0x%llx\n",
828 (unsigned long long)memblock_phys_mem_size());
829
830 pr_info("dcache_bsize = 0x%x\n", dcache_bsize);
831 pr_info("icache_bsize = 0x%x\n", icache_bsize);
832
833 pr_info("cpu_features = 0x%016lx\n", cur_cpu_spec->cpu_features);
834 pr_info(" possible = 0x%016lx\n",
835 (unsigned long)CPU_FTRS_POSSIBLE);
836 pr_info(" always = 0x%016lx\n",
837 (unsigned long)CPU_FTRS_ALWAYS);
838 pr_info("cpu_user_features = 0x%08x 0x%08x\n",
839 cur_cpu_spec->cpu_user_features,
840 cur_cpu_spec->cpu_user_features2);
841 pr_info("mmu_features = 0x%08x\n", cur_cpu_spec->mmu_features);
842#ifdef CONFIG_PPC64
843 pr_info("firmware_features = 0x%016lx\n", powerpc_firmware_features);
844#ifdef CONFIG_PPC_BOOK3S
845 pr_info("vmalloc start = 0x%lx\n", KERN_VIRT_START);
846 pr_info("IO start = 0x%lx\n", KERN_IO_START);
847 pr_info("vmemmap start = 0x%lx\n", (unsigned long)vmemmap);
848#endif
849#endif
850
851 if (!early_radix_enabled())
852 print_system_hash_info();
853
854 if (PHYSICAL_START > 0)
855 pr_info("physical_start = 0x%llx\n",
856 (unsigned long long)PHYSICAL_START);
857 pr_info("-----------------------------------------------------\n");
858}
859
860#ifdef CONFIG_SMP
861static void __init smp_setup_pacas(void)
862{
863 int cpu;
864
865 for_each_possible_cpu(cpu) {
866 if (cpu == smp_processor_id())
867 continue;
868 allocate_paca(cpu);
869 set_hard_smp_processor_id(cpu, cpu_to_phys_id[cpu]);
870 }
871
872 memblock_free(cpu_to_phys_id, nr_cpu_ids * sizeof(u32));
873 cpu_to_phys_id = NULL;
874}
875#endif
876
877/*
878 * Called into from start_kernel this initializes memblock, which is used
879 * to manage page allocation until mem_init is called.
880 */
881void __init setup_arch(char **cmdline_p)
882{
883 kasan_init();
884
885 *cmdline_p = boot_command_line;
886
887 /* Set a half-reasonable default so udelay does something sensible */
888 loops_per_jiffy = 500000000 / HZ;
889
890 /* Unflatten the device-tree passed by prom_init or kexec */
891 unflatten_device_tree();
892
893 /*
894 * Initialize cache line/block info from device-tree (on ppc64) or
895 * just cputable (on ppc32).
896 */
897 initialize_cache_info();
898
899 /* Initialize RTAS if available. */
900 rtas_initialize();
901
902 /* Check if we have an initrd provided via the device-tree. */
903 check_for_initrd();
904
905 /* Probe the machine type, establish ppc_md. */
906 probe_machine();
907
908 /* Setup panic notifier if requested by the platform. */
909 setup_panic();
910
911 /*
912 * Configure ppc_md.power_save (ppc32 only, 64-bit machines do
913 * it from their respective probe() function.
914 */
915 setup_power_save();
916
917 /* Discover standard serial ports. */
918 find_legacy_serial_ports();
919
920 /* Register early console with the printk subsystem. */
921 register_early_udbg_console();
922
923 /* Setup the various CPU maps based on the device-tree. */
924 smp_setup_cpu_maps();
925
926 /* Initialize xmon. */
927 xmon_setup();
928
929 /* Check the SMT related command line arguments (ppc64). */
930 check_smt_enabled();
931
932 /* Parse memory topology */
933 mem_topology_setup();
934 /* Set max_mapnr before paging_init() */
935 set_max_mapnr(max_pfn);
936
937 /*
938 * Release secondary cpus out of their spinloops at 0x60 now that
939 * we can map physical -> logical CPU ids.
940 *
941 * Freescale Book3e parts spin in a loop provided by firmware,
942 * so smp_release_cpus() does nothing for them.
943 */
944#ifdef CONFIG_SMP
945 smp_setup_pacas();
946
947 /* On BookE, setup per-core TLB data structures. */
948 setup_tlb_core_data();
949#endif
950
951 /* Print various info about the machine that has been gathered so far. */
952 print_system_info();
953
954 klp_init_thread_info(&init_task);
955
956 setup_initial_init_mm(_stext, _etext, _edata, _end);
957 /* sched_init() does the mmgrab(&init_mm) for the primary CPU */
958 VM_WARN_ON(cpumask_test_cpu(smp_processor_id(), mm_cpumask(&init_mm)));
959 cpumask_set_cpu(smp_processor_id(), mm_cpumask(&init_mm));
960 inc_mm_active_cpus(&init_mm);
961 mm_iommu_init(&init_mm);
962
963 irqstack_early_init();
964 exc_lvl_early_init();
965 emergency_stack_init();
966
967 mce_init();
968 smp_release_cpus();
969
970 initmem_init();
971
972 /*
973 * Reserve large chunks of memory for use by CMA for KVM and hugetlb. These must
974 * be called after initmem_init(), so that pageblock_order is initialised.
975 */
976 kvm_cma_reserve();
977 gigantic_hugetlb_cma_reserve();
978
979 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
980
981 if (ppc_md.setup_arch)
982 ppc_md.setup_arch();
983
984 setup_barrier_nospec();
985 setup_spectre_v2();
986
987 paging_init();
988
989 /* Initialize the MMU context management stuff. */
990 mmu_context_init();
991
992 /* Interrupt code needs to be 64K-aligned. */
993 if (IS_ENABLED(CONFIG_PPC64) && (unsigned long)_stext & 0xffff)
994 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
995 (unsigned long)_stext);
996}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Common boot and setup code for both 32-bit and 64-bit.
4 * Extracted from arch/powerpc/kernel/setup_64.c.
5 *
6 * Copyright (C) 2001 PPC64 Team, IBM Corp
7 */
8
9#undef DEBUG
10
11#include <linux/export.h>
12#include <linux/panic_notifier.h>
13#include <linux/string.h>
14#include <linux/sched.h>
15#include <linux/init.h>
16#include <linux/kernel.h>
17#include <linux/reboot.h>
18#include <linux/delay.h>
19#include <linux/initrd.h>
20#include <linux/platform_device.h>
21#include <linux/printk.h>
22#include <linux/seq_file.h>
23#include <linux/ioport.h>
24#include <linux/console.h>
25#include <linux/root_dev.h>
26#include <linux/cpu.h>
27#include <linux/unistd.h>
28#include <linux/seq_buf.h>
29#include <linux/serial.h>
30#include <linux/serial_8250.h>
31#include <linux/percpu.h>
32#include <linux/memblock.h>
33#include <linux/of.h>
34#include <linux/of_fdt.h>
35#include <linux/of_irq.h>
36#include <linux/hugetlb.h>
37#include <linux/pgtable.h>
38#include <asm/io.h>
39#include <asm/paca.h>
40#include <asm/processor.h>
41#include <asm/vdso_datapage.h>
42#include <asm/smp.h>
43#include <asm/elf.h>
44#include <asm/machdep.h>
45#include <asm/time.h>
46#include <asm/cputable.h>
47#include <asm/sections.h>
48#include <asm/firmware.h>
49#include <asm/btext.h>
50#include <asm/nvram.h>
51#include <asm/setup.h>
52#include <asm/rtas.h>
53#include <asm/iommu.h>
54#include <asm/serial.h>
55#include <asm/cache.h>
56#include <asm/page.h>
57#include <asm/mmu.h>
58#include <asm/xmon.h>
59#include <asm/cputhreads.h>
60#include <mm/mmu_decl.h>
61#include <asm/archrandom.h>
62#include <asm/fadump.h>
63#include <asm/udbg.h>
64#include <asm/hugetlb.h>
65#include <asm/livepatch.h>
66#include <asm/mmu_context.h>
67#include <asm/cpu_has_feature.h>
68#include <asm/kasan.h>
69#include <asm/mce.h>
70
71#include "setup.h"
72
73#ifdef DEBUG
74#define DBG(fmt...) udbg_printf(fmt)
75#else
76#define DBG(fmt...)
77#endif
78
79/* The main machine-dep calls structure
80 */
81struct machdep_calls ppc_md;
82EXPORT_SYMBOL(ppc_md);
83struct machdep_calls *machine_id;
84EXPORT_SYMBOL(machine_id);
85
86int boot_cpuid = -1;
87EXPORT_SYMBOL_GPL(boot_cpuid);
88int __initdata boot_core_hwid = -1;
89
90#ifdef CONFIG_PPC64
91int boot_cpu_hwid = -1;
92#endif
93
94/*
95 * These are used in binfmt_elf.c to put aux entries on the stack
96 * for each elf executable being started.
97 */
98int dcache_bsize;
99int icache_bsize;
100
101/* Variables required to store legacy IO irq routing */
102int of_i8042_kbd_irq;
103EXPORT_SYMBOL_GPL(of_i8042_kbd_irq);
104int of_i8042_aux_irq;
105EXPORT_SYMBOL_GPL(of_i8042_aux_irq);
106
107#ifdef __DO_IRQ_CANON
108/* XXX should go elsewhere eventually */
109int ppc_do_canonicalize_irqs;
110EXPORT_SYMBOL(ppc_do_canonicalize_irqs);
111#endif
112
113#ifdef CONFIG_CRASH_DUMP
114/* This keeps a track of which one is the crashing cpu. */
115int crashing_cpu = -1;
116#endif
117
118/* also used by kexec */
119void machine_shutdown(void)
120{
121 /*
122 * if fadump is active, cleanup the fadump registration before we
123 * shutdown.
124 */
125 fadump_cleanup();
126
127 if (ppc_md.machine_shutdown)
128 ppc_md.machine_shutdown();
129}
130
131static void machine_hang(void)
132{
133 pr_emerg("System Halted, OK to turn off power\n");
134 local_irq_disable();
135 while (1)
136 ;
137}
138
139void machine_restart(char *cmd)
140{
141 machine_shutdown();
142 if (ppc_md.restart)
143 ppc_md.restart(cmd);
144
145 smp_send_stop();
146
147 do_kernel_restart(cmd);
148 mdelay(1000);
149
150 machine_hang();
151}
152
153void machine_power_off(void)
154{
155 machine_shutdown();
156 do_kernel_power_off();
157 smp_send_stop();
158 machine_hang();
159}
160/* Used by the G5 thermal driver */
161EXPORT_SYMBOL_GPL(machine_power_off);
162
163void (*pm_power_off)(void);
164EXPORT_SYMBOL_GPL(pm_power_off);
165
166size_t __must_check arch_get_random_seed_longs(unsigned long *v, size_t max_longs)
167{
168 if (max_longs && ppc_md.get_random_seed && ppc_md.get_random_seed(v))
169 return 1;
170 return 0;
171}
172EXPORT_SYMBOL(arch_get_random_seed_longs);
173
174void machine_halt(void)
175{
176 machine_shutdown();
177 if (ppc_md.halt)
178 ppc_md.halt();
179
180 smp_send_stop();
181 machine_hang();
182}
183
184#ifdef CONFIG_SMP
185DEFINE_PER_CPU(unsigned int, cpu_pvr);
186#endif
187
188static void show_cpuinfo_summary(struct seq_file *m)
189{
190 struct device_node *root;
191 const char *model = NULL;
192 unsigned long bogosum = 0;
193 int i;
194
195 if (IS_ENABLED(CONFIG_SMP) && IS_ENABLED(CONFIG_PPC32)) {
196 for_each_online_cpu(i)
197 bogosum += loops_per_jiffy;
198 seq_printf(m, "total bogomips\t: %lu.%02lu\n",
199 bogosum / (500000 / HZ), bogosum / (5000 / HZ) % 100);
200 }
201 seq_printf(m, "timebase\t: %lu\n", ppc_tb_freq);
202 if (ppc_md.name)
203 seq_printf(m, "platform\t: %s\n", ppc_md.name);
204 root = of_find_node_by_path("/");
205 if (root)
206 model = of_get_property(root, "model", NULL);
207 if (model)
208 seq_printf(m, "model\t\t: %s\n", model);
209 of_node_put(root);
210
211 if (ppc_md.show_cpuinfo != NULL)
212 ppc_md.show_cpuinfo(m);
213
214 /* Display the amount of memory */
215 if (IS_ENABLED(CONFIG_PPC32))
216 seq_printf(m, "Memory\t\t: %d MB\n",
217 (unsigned int)(total_memory / (1024 * 1024)));
218}
219
220static int show_cpuinfo(struct seq_file *m, void *v)
221{
222 unsigned long cpu_id = (unsigned long)v - 1;
223 unsigned int pvr;
224 unsigned long proc_freq;
225 unsigned short maj;
226 unsigned short min;
227
228#ifdef CONFIG_SMP
229 pvr = per_cpu(cpu_pvr, cpu_id);
230#else
231 pvr = mfspr(SPRN_PVR);
232#endif
233 maj = (pvr >> 8) & 0xFF;
234 min = pvr & 0xFF;
235
236 seq_printf(m, "processor\t: %lu\ncpu\t\t: ", cpu_id);
237
238 if (cur_cpu_spec->pvr_mask && cur_cpu_spec->cpu_name)
239 seq_puts(m, cur_cpu_spec->cpu_name);
240 else
241 seq_printf(m, "unknown (%08x)", pvr);
242
243 if (cpu_has_feature(CPU_FTR_ALTIVEC))
244 seq_puts(m, ", altivec supported");
245
246 seq_putc(m, '\n');
247
248#ifdef CONFIG_TAU
249 if (cpu_has_feature(CPU_FTR_TAU)) {
250 if (IS_ENABLED(CONFIG_TAU_AVERAGE)) {
251 /* more straightforward, but potentially misleading */
252 seq_printf(m, "temperature \t: %u C (uncalibrated)\n",
253 cpu_temp(cpu_id));
254 } else {
255 /* show the actual temp sensor range */
256 u32 temp;
257 temp = cpu_temp_both(cpu_id);
258 seq_printf(m, "temperature \t: %u-%u C (uncalibrated)\n",
259 temp & 0xff, temp >> 16);
260 }
261 }
262#endif /* CONFIG_TAU */
263
264 /*
265 * Platforms that have variable clock rates, should implement
266 * the method ppc_md.get_proc_freq() that reports the clock
267 * rate of a given cpu. The rest can use ppc_proc_freq to
268 * report the clock rate that is same across all cpus.
269 */
270 if (ppc_md.get_proc_freq)
271 proc_freq = ppc_md.get_proc_freq(cpu_id);
272 else
273 proc_freq = ppc_proc_freq;
274
275 if (proc_freq)
276 seq_printf(m, "clock\t\t: %lu.%06luMHz\n",
277 proc_freq / 1000000, proc_freq % 1000000);
278
279 /* If we are a Freescale core do a simple check so
280 * we don't have to keep adding cases in the future */
281 if (PVR_VER(pvr) & 0x8000) {
282 switch (PVR_VER(pvr)) {
283 case 0x8000: /* 7441/7450/7451, Voyager */
284 case 0x8001: /* 7445/7455, Apollo 6 */
285 case 0x8002: /* 7447/7457, Apollo 7 */
286 case 0x8003: /* 7447A, Apollo 7 PM */
287 case 0x8004: /* 7448, Apollo 8 */
288 case 0x800c: /* 7410, Nitro */
289 maj = ((pvr >> 8) & 0xF);
290 min = PVR_MIN(pvr);
291 break;
292 default: /* e500/book-e */
293 maj = PVR_MAJ(pvr);
294 min = PVR_MIN(pvr);
295 break;
296 }
297 } else {
298 switch (PVR_VER(pvr)) {
299 case 0x1008: /* 740P/750P ?? */
300 maj = ((pvr >> 8) & 0xFF) - 1;
301 min = pvr & 0xFF;
302 break;
303 case 0x004e: /* POWER9 bits 12-15 give chip type */
304 case 0x0080: /* POWER10 bit 12 gives SMT8/4 */
305 maj = (pvr >> 8) & 0x0F;
306 min = pvr & 0xFF;
307 break;
308 default:
309 maj = (pvr >> 8) & 0xFF;
310 min = pvr & 0xFF;
311 break;
312 }
313 }
314
315 seq_printf(m, "revision\t: %hd.%hd (pvr %04x %04x)\n",
316 maj, min, PVR_VER(pvr), PVR_REV(pvr));
317
318 if (IS_ENABLED(CONFIG_PPC32))
319 seq_printf(m, "bogomips\t: %lu.%02lu\n", loops_per_jiffy / (500000 / HZ),
320 (loops_per_jiffy / (5000 / HZ)) % 100);
321
322 seq_putc(m, '\n');
323
324 /* If this is the last cpu, print the summary */
325 if (cpumask_next(cpu_id, cpu_online_mask) >= nr_cpu_ids)
326 show_cpuinfo_summary(m);
327
328 return 0;
329}
330
331static void *c_start(struct seq_file *m, loff_t *pos)
332{
333 if (*pos == 0) /* just in case, cpu 0 is not the first */
334 *pos = cpumask_first(cpu_online_mask);
335 else
336 *pos = cpumask_next(*pos - 1, cpu_online_mask);
337 if ((*pos) < nr_cpu_ids)
338 return (void *)(unsigned long)(*pos + 1);
339 return NULL;
340}
341
342static void *c_next(struct seq_file *m, void *v, loff_t *pos)
343{
344 (*pos)++;
345 return c_start(m, pos);
346}
347
348static void c_stop(struct seq_file *m, void *v)
349{
350}
351
352const struct seq_operations cpuinfo_op = {
353 .start = c_start,
354 .next = c_next,
355 .stop = c_stop,
356 .show = show_cpuinfo,
357};
358
359void __init check_for_initrd(void)
360{
361#ifdef CONFIG_BLK_DEV_INITRD
362 DBG(" -> check_for_initrd() initrd_start=0x%lx initrd_end=0x%lx\n",
363 initrd_start, initrd_end);
364
365 /* If we were passed an initrd, set the ROOT_DEV properly if the values
366 * look sensible. If not, clear initrd reference.
367 */
368 if (is_kernel_addr(initrd_start) && is_kernel_addr(initrd_end) &&
369 initrd_end > initrd_start)
370 ROOT_DEV = Root_RAM0;
371 else
372 initrd_start = initrd_end = 0;
373
374 if (initrd_start)
375 pr_info("Found initrd at 0x%lx:0x%lx\n", initrd_start, initrd_end);
376
377 DBG(" <- check_for_initrd()\n");
378#endif /* CONFIG_BLK_DEV_INITRD */
379}
380
381#ifdef CONFIG_SMP
382
383int threads_per_core, threads_per_subcore, threads_shift __read_mostly;
384cpumask_t threads_core_mask __read_mostly;
385EXPORT_SYMBOL_GPL(threads_per_core);
386EXPORT_SYMBOL_GPL(threads_per_subcore);
387EXPORT_SYMBOL_GPL(threads_shift);
388EXPORT_SYMBOL_GPL(threads_core_mask);
389
390static void __init cpu_init_thread_core_maps(int tpc)
391{
392 int i;
393
394 threads_per_core = tpc;
395 threads_per_subcore = tpc;
396 cpumask_clear(&threads_core_mask);
397
398 /* This implementation only supports power of 2 number of threads
399 * for simplicity and performance
400 */
401 threads_shift = ilog2(tpc);
402 BUG_ON(tpc != (1 << threads_shift));
403
404 for (i = 0; i < tpc; i++)
405 cpumask_set_cpu(i, &threads_core_mask);
406
407 printk(KERN_INFO "CPU maps initialized for %d thread%s per core\n",
408 tpc, tpc > 1 ? "s" : "");
409 printk(KERN_DEBUG " (thread shift is %d)\n", threads_shift);
410}
411
412
413u32 *cpu_to_phys_id = NULL;
414
415static int assign_threads(unsigned int cpu, unsigned int nthreads, bool present,
416 const __be32 *hw_ids)
417{
418 for (int i = 0; i < nthreads && cpu < nr_cpu_ids; i++) {
419 __be32 hwid;
420
421 hwid = be32_to_cpu(hw_ids[i]);
422
423 DBG(" thread %d -> cpu %d (hard id %d)\n", i, cpu, hwid);
424
425 set_cpu_present(cpu, present);
426 set_cpu_possible(cpu, true);
427 cpu_to_phys_id[cpu] = hwid;
428 cpu++;
429 }
430
431 return cpu;
432}
433
434/**
435 * setup_cpu_maps - initialize the following cpu maps:
436 * cpu_possible_mask
437 * cpu_present_mask
438 *
439 * Having the possible map set up early allows us to restrict allocations
440 * of things like irqstacks to nr_cpu_ids rather than NR_CPUS.
441 *
442 * We do not initialize the online map here; cpus set their own bits in
443 * cpu_online_mask as they come up.
444 *
445 * This function is valid only for Open Firmware systems. finish_device_tree
446 * must be called before using this.
447 *
448 * While we're here, we may as well set the "physical" cpu ids in the paca.
449 *
450 * NOTE: This must match the parsing done in early_init_dt_scan_cpus.
451 */
452void __init smp_setup_cpu_maps(void)
453{
454 struct device_node *dn;
455 int cpu = 0;
456 int nthreads = 1;
457
458 DBG("smp_setup_cpu_maps()\n");
459
460 cpu_to_phys_id = memblock_alloc(nr_cpu_ids * sizeof(u32),
461 __alignof__(u32));
462 if (!cpu_to_phys_id)
463 panic("%s: Failed to allocate %zu bytes align=0x%zx\n",
464 __func__, nr_cpu_ids * sizeof(u32), __alignof__(u32));
465
466 for_each_node_by_type(dn, "cpu") {
467 const __be32 *intserv;
468 __be32 cpu_be;
469 int len;
470
471 DBG(" * %pOF...\n", dn);
472
473 intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s",
474 &len);
475 if (intserv) {
476 DBG(" ibm,ppc-interrupt-server#s -> %lu threads\n",
477 (len / sizeof(int)));
478 } else {
479 DBG(" no ibm,ppc-interrupt-server#s -> 1 thread\n");
480 intserv = of_get_property(dn, "reg", &len);
481 if (!intserv) {
482 cpu_be = cpu_to_be32(cpu);
483 /* XXX: what is this? uninitialized?? */
484 intserv = &cpu_be; /* assume logical == phys */
485 len = 4;
486 }
487 }
488
489 nthreads = len / sizeof(int);
490
491 bool avail = of_device_is_available(dn);
492 if (!avail)
493 avail = !of_property_match_string(dn,
494 "enable-method", "spin-table");
495
496 if (boot_core_hwid >= 0) {
497 if (cpu == 0) {
498 pr_info("Skipping CPU node %pOF to allow for boot core.\n", dn);
499 cpu = nthreads;
500 continue;
501 }
502
503 if (be32_to_cpu(intserv[0]) == boot_core_hwid) {
504 pr_info("Renumbered boot core %pOF to logical 0\n", dn);
505 assign_threads(0, nthreads, avail, intserv);
506 of_node_put(dn);
507 break;
508 }
509 } else if (cpu >= nr_cpu_ids) {
510 of_node_put(dn);
511 break;
512 }
513
514 if (cpu < nr_cpu_ids)
515 cpu = assign_threads(cpu, nthreads, avail, intserv);
516 }
517
518 /* If no SMT supported, nthreads is forced to 1 */
519 if (!cpu_has_feature(CPU_FTR_SMT)) {
520 DBG(" SMT disabled ! nthreads forced to 1\n");
521 nthreads = 1;
522 }
523
524#ifdef CONFIG_PPC64
525 /*
526 * On pSeries LPAR, we need to know how many cpus
527 * could possibly be added to this partition.
528 */
529 if (firmware_has_feature(FW_FEATURE_LPAR) &&
530 (dn = of_find_node_by_path("/rtas"))) {
531 int num_addr_cell, num_size_cell, maxcpus;
532 const __be32 *ireg;
533
534 num_addr_cell = of_n_addr_cells(dn);
535 num_size_cell = of_n_size_cells(dn);
536
537 ireg = of_get_property(dn, "ibm,lrdr-capacity", NULL);
538
539 if (!ireg)
540 goto out;
541
542 maxcpus = be32_to_cpup(ireg + num_addr_cell + num_size_cell);
543
544 /* Double maxcpus for processors which have SMT capability */
545 if (cpu_has_feature(CPU_FTR_SMT))
546 maxcpus *= nthreads;
547
548 if (maxcpus > nr_cpu_ids) {
549 printk(KERN_WARNING
550 "Partition configured for %d cpus, "
551 "operating system maximum is %u.\n",
552 maxcpus, nr_cpu_ids);
553 maxcpus = nr_cpu_ids;
554 } else
555 printk(KERN_INFO "Partition configured for %d cpus.\n",
556 maxcpus);
557
558 for (cpu = 0; cpu < maxcpus; cpu++)
559 set_cpu_possible(cpu, true);
560 out:
561 of_node_put(dn);
562 }
563 vdso_data->processorCount = num_present_cpus();
564#endif /* CONFIG_PPC64 */
565
566 /* Initialize CPU <=> thread mapping/
567 *
568 * WARNING: We assume that the number of threads is the same for
569 * every CPU in the system. If that is not the case, then some code
570 * here will have to be reworked
571 */
572 cpu_init_thread_core_maps(nthreads);
573
574 /* Now that possible cpus are set, set nr_cpu_ids for later use */
575 setup_nr_cpu_ids();
576
577 free_unused_pacas();
578}
579#endif /* CONFIG_SMP */
580
581#ifdef CONFIG_PCSPKR_PLATFORM
582static __init int add_pcspkr(void)
583{
584 struct device_node *np;
585 struct platform_device *pd;
586 int ret;
587
588 np = of_find_compatible_node(NULL, NULL, "pnpPNP,100");
589 of_node_put(np);
590 if (!np)
591 return -ENODEV;
592
593 pd = platform_device_alloc("pcspkr", -1);
594 if (!pd)
595 return -ENOMEM;
596
597 ret = platform_device_add(pd);
598 if (ret)
599 platform_device_put(pd);
600
601 return ret;
602}
603device_initcall(add_pcspkr);
604#endif /* CONFIG_PCSPKR_PLATFORM */
605
606static char ppc_hw_desc_buf[128] __initdata;
607
608struct seq_buf ppc_hw_desc __initdata = {
609 .buffer = ppc_hw_desc_buf,
610 .size = sizeof(ppc_hw_desc_buf),
611 .len = 0,
612};
613
614static __init void probe_machine(void)
615{
616 extern struct machdep_calls __machine_desc_start;
617 extern struct machdep_calls __machine_desc_end;
618 unsigned int i;
619
620 /*
621 * Iterate all ppc_md structures until we find the proper
622 * one for the current machine type
623 */
624 DBG("Probing machine type ...\n");
625
626 /*
627 * Check ppc_md is empty, if not we have a bug, ie, we setup an
628 * entry before probe_machine() which will be overwritten
629 */
630 for (i = 0; i < (sizeof(ppc_md) / sizeof(void *)); i++) {
631 if (((void **)&ppc_md)[i]) {
632 printk(KERN_ERR "Entry %d in ppc_md non empty before"
633 " machine probe !\n", i);
634 }
635 }
636
637 for (machine_id = &__machine_desc_start;
638 machine_id < &__machine_desc_end;
639 machine_id++) {
640 DBG(" %s ...\n", machine_id->name);
641 if (machine_id->compatible && !of_machine_is_compatible(machine_id->compatible))
642 continue;
643 if (machine_id->compatibles && !of_machine_compatible_match(machine_id->compatibles))
644 continue;
645 memcpy(&ppc_md, machine_id, sizeof(struct machdep_calls));
646 if (ppc_md.probe && !ppc_md.probe())
647 continue;
648 DBG(" %s match !\n", machine_id->name);
649 break;
650 }
651 /* What can we do if we didn't find ? */
652 if (machine_id >= &__machine_desc_end) {
653 pr_err("No suitable machine description found !\n");
654 for (;;);
655 }
656
657 // Append the machine name to other info we've gathered
658 seq_buf_puts(&ppc_hw_desc, ppc_md.name);
659
660 // Set the generic hardware description shown in oopses
661 dump_stack_set_arch_desc(ppc_hw_desc.buffer);
662
663 pr_info("Hardware name: %s\n", ppc_hw_desc.buffer);
664}
665
666/* Match a class of boards, not a specific device configuration. */
667int check_legacy_ioport(unsigned long base_port)
668{
669 struct device_node *parent, *np = NULL;
670 int ret = -ENODEV;
671
672 switch(base_port) {
673 case I8042_DATA_REG:
674 if (!(np = of_find_compatible_node(NULL, NULL, "pnpPNP,303")))
675 np = of_find_compatible_node(NULL, NULL, "pnpPNP,f03");
676 if (np) {
677 parent = of_get_parent(np);
678
679 of_i8042_kbd_irq = irq_of_parse_and_map(parent, 0);
680 if (!of_i8042_kbd_irq)
681 of_i8042_kbd_irq = 1;
682
683 of_i8042_aux_irq = irq_of_parse_and_map(parent, 1);
684 if (!of_i8042_aux_irq)
685 of_i8042_aux_irq = 12;
686
687 of_node_put(np);
688 np = parent;
689 break;
690 }
691 np = of_find_node_by_type(NULL, "8042");
692 /* Pegasos has no device_type on its 8042 node, look for the
693 * name instead */
694 if (!np)
695 np = of_find_node_by_name(NULL, "8042");
696 if (np) {
697 of_i8042_kbd_irq = 1;
698 of_i8042_aux_irq = 12;
699 }
700 break;
701 case FDC_BASE: /* FDC1 */
702 np = of_find_node_by_type(NULL, "fdc");
703 break;
704 default:
705 /* ipmi is supposed to fail here */
706 break;
707 }
708 if (!np)
709 return ret;
710 parent = of_get_parent(np);
711 if (parent) {
712 if (of_node_is_type(parent, "isa"))
713 ret = 0;
714 of_node_put(parent);
715 }
716 of_node_put(np);
717 return ret;
718}
719EXPORT_SYMBOL(check_legacy_ioport);
720
721/*
722 * Panic notifiers setup
723 *
724 * We have 3 notifiers for powerpc, each one from a different "nature":
725 *
726 * - ppc_panic_fadump_handler() is a hypervisor notifier, which hard-disables
727 * IRQs and deal with the Firmware-Assisted dump, when it is configured;
728 * should run early in the panic path.
729 *
730 * - dump_kernel_offset() is an informative notifier, just showing the KASLR
731 * offset if we have RANDOMIZE_BASE set.
732 *
733 * - ppc_panic_platform_handler() is a low-level handler that's registered
734 * only if the platform wishes to perform final actions in the panic path,
735 * hence it should run late and might not even return. Currently, only
736 * pseries and ps3 platforms register callbacks.
737 */
738static int ppc_panic_fadump_handler(struct notifier_block *this,
739 unsigned long event, void *ptr)
740{
741 /*
742 * panic does a local_irq_disable, but we really
743 * want interrupts to be hard disabled.
744 */
745 hard_irq_disable();
746
747 /*
748 * If firmware-assisted dump has been registered then trigger
749 * its callback and let the firmware handles everything else.
750 */
751 crash_fadump(NULL, ptr);
752
753 return NOTIFY_DONE;
754}
755
756static int dump_kernel_offset(struct notifier_block *self, unsigned long v,
757 void *p)
758{
759 pr_emerg("Kernel Offset: 0x%lx from 0x%lx\n",
760 kaslr_offset(), KERNELBASE);
761
762 return NOTIFY_DONE;
763}
764
765static int ppc_panic_platform_handler(struct notifier_block *this,
766 unsigned long event, void *ptr)
767{
768 /*
769 * This handler is only registered if we have a panic callback
770 * on ppc_md, hence NULL check is not needed.
771 * Also, it may not return, so it runs really late on panic path.
772 */
773 ppc_md.panic(ptr);
774
775 return NOTIFY_DONE;
776}
777
778static struct notifier_block ppc_fadump_block = {
779 .notifier_call = ppc_panic_fadump_handler,
780 .priority = INT_MAX, /* run early, to notify the firmware ASAP */
781};
782
783static struct notifier_block kernel_offset_notifier = {
784 .notifier_call = dump_kernel_offset,
785};
786
787static struct notifier_block ppc_panic_block = {
788 .notifier_call = ppc_panic_platform_handler,
789 .priority = INT_MIN, /* may not return; must be done last */
790};
791
792void __init setup_panic(void)
793{
794 /* Hard-disables IRQs + deal with FW-assisted dump (fadump) */
795 atomic_notifier_chain_register(&panic_notifier_list,
796 &ppc_fadump_block);
797
798 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && kaslr_offset() > 0)
799 atomic_notifier_chain_register(&panic_notifier_list,
800 &kernel_offset_notifier);
801
802 /* Low-level platform-specific routines that should run on panic */
803 if (ppc_md.panic)
804 atomic_notifier_chain_register(&panic_notifier_list,
805 &ppc_panic_block);
806}
807
808#ifdef CONFIG_CHECK_CACHE_COHERENCY
809/*
810 * For platforms that have configurable cache-coherency. This function
811 * checks that the cache coherency setting of the kernel matches the setting
812 * left by the firmware, as indicated in the device tree. Since a mismatch
813 * will eventually result in DMA failures, we print * and error and call
814 * BUG() in that case.
815 */
816
817#define KERNEL_COHERENCY (!IS_ENABLED(CONFIG_NOT_COHERENT_CACHE))
818
819static int __init check_cache_coherency(void)
820{
821 struct device_node *np;
822 const void *prop;
823 bool devtree_coherency;
824
825 np = of_find_node_by_path("/");
826 prop = of_get_property(np, "coherency-off", NULL);
827 of_node_put(np);
828
829 devtree_coherency = prop ? false : true;
830
831 if (devtree_coherency != KERNEL_COHERENCY) {
832 printk(KERN_ERR
833 "kernel coherency:%s != device tree_coherency:%s\n",
834 KERNEL_COHERENCY ? "on" : "off",
835 devtree_coherency ? "on" : "off");
836 BUG();
837 }
838
839 return 0;
840}
841
842late_initcall(check_cache_coherency);
843#endif /* CONFIG_CHECK_CACHE_COHERENCY */
844
845void ppc_printk_progress(char *s, unsigned short hex)
846{
847 pr_info("%s\n", s);
848}
849
850static __init void print_system_info(void)
851{
852 pr_info("-----------------------------------------------------\n");
853 pr_info("phys_mem_size = 0x%llx\n",
854 (unsigned long long)memblock_phys_mem_size());
855
856 pr_info("dcache_bsize = 0x%x\n", dcache_bsize);
857 pr_info("icache_bsize = 0x%x\n", icache_bsize);
858
859 pr_info("cpu_features = 0x%016lx\n", cur_cpu_spec->cpu_features);
860 pr_info(" possible = 0x%016lx\n",
861 (unsigned long)CPU_FTRS_POSSIBLE);
862 pr_info(" always = 0x%016lx\n",
863 (unsigned long)CPU_FTRS_ALWAYS);
864 pr_info("cpu_user_features = 0x%08x 0x%08x\n",
865 cur_cpu_spec->cpu_user_features,
866 cur_cpu_spec->cpu_user_features2);
867 pr_info("mmu_features = 0x%08x\n", cur_cpu_spec->mmu_features);
868#ifdef CONFIG_PPC64
869 pr_info("firmware_features = 0x%016lx\n", powerpc_firmware_features);
870#ifdef CONFIG_PPC_BOOK3S
871 pr_info("vmalloc start = 0x%lx\n", KERN_VIRT_START);
872 pr_info("IO start = 0x%lx\n", KERN_IO_START);
873 pr_info("vmemmap start = 0x%lx\n", (unsigned long)vmemmap);
874#endif
875#endif
876
877 if (!early_radix_enabled())
878 print_system_hash_info();
879
880 if (PHYSICAL_START > 0)
881 pr_info("physical_start = 0x%llx\n",
882 (unsigned long long)PHYSICAL_START);
883 pr_info("-----------------------------------------------------\n");
884}
885
886#ifdef CONFIG_SMP
887static void __init smp_setup_pacas(void)
888{
889 int cpu;
890
891 for_each_possible_cpu(cpu) {
892 if (cpu == smp_processor_id())
893 continue;
894 allocate_paca(cpu);
895 set_hard_smp_processor_id(cpu, cpu_to_phys_id[cpu]);
896 }
897
898 memblock_free(cpu_to_phys_id, nr_cpu_ids * sizeof(u32));
899 cpu_to_phys_id = NULL;
900}
901#endif
902
903/*
904 * Called into from start_kernel this initializes memblock, which is used
905 * to manage page allocation until mem_init is called.
906 */
907void __init setup_arch(char **cmdline_p)
908{
909 kasan_init();
910
911 *cmdline_p = boot_command_line;
912
913 /* Set a half-reasonable default so udelay does something sensible */
914 loops_per_jiffy = 500000000 / HZ;
915
916 /* Unflatten the device-tree passed by prom_init or kexec */
917 unflatten_device_tree();
918
919 /*
920 * Initialize cache line/block info from device-tree (on ppc64) or
921 * just cputable (on ppc32).
922 */
923 initialize_cache_info();
924
925 /* Initialize RTAS if available. */
926 rtas_initialize();
927
928 /* Check if we have an initrd provided via the device-tree. */
929 check_for_initrd();
930
931 /* Probe the machine type, establish ppc_md. */
932 probe_machine();
933
934 /* Setup panic notifier if requested by the platform. */
935 setup_panic();
936
937 /*
938 * Configure ppc_md.power_save (ppc32 only, 64-bit machines do
939 * it from their respective probe() function.
940 */
941 setup_power_save();
942
943 /* Discover standard serial ports. */
944 find_legacy_serial_ports();
945
946 /* Register early console with the printk subsystem. */
947 register_early_udbg_console();
948
949 /* Setup the various CPU maps based on the device-tree. */
950 smp_setup_cpu_maps();
951
952 /* Initialize xmon. */
953 xmon_setup();
954
955 /* Check the SMT related command line arguments (ppc64). */
956 check_smt_enabled();
957
958 /* Parse memory topology */
959 mem_topology_setup();
960 /* Set max_mapnr before paging_init() */
961 set_max_mapnr(max_pfn);
962
963 /*
964 * Release secondary cpus out of their spinloops at 0x60 now that
965 * we can map physical -> logical CPU ids.
966 *
967 * Freescale Book3e parts spin in a loop provided by firmware,
968 * so smp_release_cpus() does nothing for them.
969 */
970#ifdef CONFIG_SMP
971 smp_setup_pacas();
972
973 /* On BookE, setup per-core TLB data structures. */
974 setup_tlb_core_data();
975#endif
976
977 /* Print various info about the machine that has been gathered so far. */
978 print_system_info();
979
980 klp_init_thread_info(&init_task);
981
982 setup_initial_init_mm(_stext, _etext, _edata, _end);
983 /* sched_init() does the mmgrab(&init_mm) for the primary CPU */
984 VM_WARN_ON(cpumask_test_cpu(smp_processor_id(), mm_cpumask(&init_mm)));
985 cpumask_set_cpu(smp_processor_id(), mm_cpumask(&init_mm));
986 inc_mm_active_cpus(&init_mm);
987 mm_iommu_init(&init_mm);
988
989 irqstack_early_init();
990 exc_lvl_early_init();
991 emergency_stack_init();
992
993 mce_init();
994 smp_release_cpus();
995
996 initmem_init();
997
998 /*
999 * Reserve large chunks of memory for use by CMA for KVM and hugetlb. These must
1000 * be called after initmem_init(), so that pageblock_order is initialised.
1001 */
1002 kvm_cma_reserve();
1003 gigantic_hugetlb_cma_reserve();
1004
1005 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1006
1007 if (ppc_md.setup_arch)
1008 ppc_md.setup_arch();
1009
1010 setup_barrier_nospec();
1011 setup_spectre_v2();
1012
1013 paging_init();
1014
1015 /* Initialize the MMU context management stuff. */
1016 mmu_context_init();
1017
1018 /* Interrupt code needs to be 64K-aligned. */
1019 if (IS_ENABLED(CONFIG_PPC64) && (unsigned long)_stext & 0xffff)
1020 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
1021 (unsigned long)_stext);
1022}