xref: /openbmc/linux/arch/powerpc/kernel/setup_64.c (revision 1d3b5aaa)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *
4  * Common boot and setup code.
5  *
6  * Copyright (C) 2001 PPC64 Team, IBM Corp
7  */
8 
9 #include <linux/export.h>
10 #include <linux/string.h>
11 #include <linux/sched.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/reboot.h>
15 #include <linux/delay.h>
16 #include <linux/initrd.h>
17 #include <linux/seq_file.h>
18 #include <linux/ioport.h>
19 #include <linux/console.h>
20 #include <linux/utsname.h>
21 #include <linux/tty.h>
22 #include <linux/root_dev.h>
23 #include <linux/notifier.h>
24 #include <linux/cpu.h>
25 #include <linux/unistd.h>
26 #include <linux/serial.h>
27 #include <linux/serial_8250.h>
28 #include <linux/memblock.h>
29 #include <linux/pci.h>
30 #include <linux/lockdep.h>
31 #include <linux/memory.h>
32 #include <linux/nmi.h>
33 #include <linux/pgtable.h>
34 
35 #include <asm/kvm_guest.h>
36 #include <asm/io.h>
37 #include <asm/kdump.h>
38 #include <asm/prom.h>
39 #include <asm/processor.h>
40 #include <asm/smp.h>
41 #include <asm/elf.h>
42 #include <asm/machdep.h>
43 #include <asm/paca.h>
44 #include <asm/time.h>
45 #include <asm/cputable.h>
46 #include <asm/dt_cpu_ftrs.h>
47 #include <asm/sections.h>
48 #include <asm/btext.h>
49 #include <asm/nvram.h>
50 #include <asm/setup.h>
51 #include <asm/rtas.h>
52 #include <asm/iommu.h>
53 #include <asm/serial.h>
54 #include <asm/cache.h>
55 #include <asm/page.h>
56 #include <asm/mmu.h>
57 #include <asm/firmware.h>
58 #include <asm/xmon.h>
59 #include <asm/udbg.h>
60 #include <asm/kexec.h>
61 #include <asm/code-patching.h>
62 #include <asm/livepatch.h>
63 #include <asm/opal.h>
64 #include <asm/cputhreads.h>
65 #include <asm/hw_irq.h>
66 #include <asm/feature-fixups.h>
67 #include <asm/kup.h>
68 #include <asm/early_ioremap.h>
69 #include <asm/pgalloc.h>
70 
71 #include "setup.h"
72 
73 int spinning_secondaries;
74 u64 ppc64_pft_size;
75 
76 struct ppc64_caches ppc64_caches = {
77 	.l1d = {
78 		.block_size = 0x40,
79 		.log_block_size = 6,
80 	},
81 	.l1i = {
82 		.block_size = 0x40,
83 		.log_block_size = 6
84 	},
85 };
86 EXPORT_SYMBOL_GPL(ppc64_caches);
87 
88 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
89 void __init setup_tlb_core_data(void)
90 {
91 	int cpu;
92 
93 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
94 
95 	for_each_possible_cpu(cpu) {
96 		int first = cpu_first_thread_sibling(cpu);
97 
98 		/*
99 		 * If we boot via kdump on a non-primary thread,
100 		 * make sure we point at the thread that actually
101 		 * set up this TLB.
102 		 */
103 		if (cpu_first_thread_sibling(boot_cpuid) == first)
104 			first = boot_cpuid;
105 
106 		paca_ptrs[cpu]->tcd_ptr = &paca_ptrs[first]->tcd;
107 
108 		/*
109 		 * If we have threads, we need either tlbsrx.
110 		 * or e6500 tablewalk mode, or else TLB handlers
111 		 * will be racy and could produce duplicate entries.
112 		 * Should we panic instead?
113 		 */
114 		WARN_ONCE(smt_enabled_at_boot >= 2 &&
115 			  !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
116 			  book3e_htw_mode != PPC_HTW_E6500,
117 			  "%s: unsupported MMU configuration\n", __func__);
118 	}
119 }
120 #endif
121 
122 #ifdef CONFIG_SMP
123 
124 static char *smt_enabled_cmdline;
125 
126 /* Look for ibm,smt-enabled OF option */
127 void __init check_smt_enabled(void)
128 {
129 	struct device_node *dn;
130 	const char *smt_option;
131 
132 	/* Default to enabling all threads */
133 	smt_enabled_at_boot = threads_per_core;
134 
135 	/* Allow the command line to overrule the OF option */
136 	if (smt_enabled_cmdline) {
137 		if (!strcmp(smt_enabled_cmdline, "on"))
138 			smt_enabled_at_boot = threads_per_core;
139 		else if (!strcmp(smt_enabled_cmdline, "off"))
140 			smt_enabled_at_boot = 0;
141 		else {
142 			int smt;
143 			int rc;
144 
145 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
146 			if (!rc)
147 				smt_enabled_at_boot =
148 					min(threads_per_core, smt);
149 		}
150 	} else {
151 		dn = of_find_node_by_path("/options");
152 		if (dn) {
153 			smt_option = of_get_property(dn, "ibm,smt-enabled",
154 						     NULL);
155 
156 			if (smt_option) {
157 				if (!strcmp(smt_option, "on"))
158 					smt_enabled_at_boot = threads_per_core;
159 				else if (!strcmp(smt_option, "off"))
160 					smt_enabled_at_boot = 0;
161 			}
162 
163 			of_node_put(dn);
164 		}
165 	}
166 }
167 
168 /* Look for smt-enabled= cmdline option */
169 static int __init early_smt_enabled(char *p)
170 {
171 	smt_enabled_cmdline = p;
172 	return 0;
173 }
174 early_param("smt-enabled", early_smt_enabled);
175 
176 #endif /* CONFIG_SMP */
177 
178 /** Fix up paca fields required for the boot cpu */
179 static void __init fixup_boot_paca(void)
180 {
181 	/* The boot cpu is started */
182 	get_paca()->cpu_start = 1;
183 	/* Allow percpu accesses to work until we setup percpu data */
184 	get_paca()->data_offset = 0;
185 	/* Mark interrupts disabled in PACA */
186 	irq_soft_mask_set(IRQS_DISABLED);
187 }
188 
189 static void __init configure_exceptions(void)
190 {
191 	/*
192 	 * Setup the trampolines from the lowmem exception vectors
193 	 * to the kdump kernel when not using a relocatable kernel.
194 	 */
195 	setup_kdump_trampoline();
196 
197 	/* Under a PAPR hypervisor, we need hypercalls */
198 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
199 		/* Enable AIL if possible */
200 		if (!pseries_enable_reloc_on_exc()) {
201 			init_task.thread.fscr &= ~FSCR_SCV;
202 			cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_SCV;
203 		}
204 
205 		/*
206 		 * Tell the hypervisor that we want our exceptions to
207 		 * be taken in little endian mode.
208 		 *
209 		 * We don't call this for big endian as our calling convention
210 		 * makes us always enter in BE, and the call may fail under
211 		 * some circumstances with kdump.
212 		 */
213 #ifdef __LITTLE_ENDIAN__
214 		pseries_little_endian_exceptions();
215 #endif
216 	} else {
217 		/* Set endian mode using OPAL */
218 		if (firmware_has_feature(FW_FEATURE_OPAL))
219 			opal_configure_cores();
220 
221 		/* AIL on native is done in cpu_ready_for_interrupts() */
222 	}
223 }
224 
225 static void cpu_ready_for_interrupts(void)
226 {
227 	/*
228 	 * Enable AIL if supported, and we are in hypervisor mode. This
229 	 * is called once for every processor.
230 	 *
231 	 * If we are not in hypervisor mode the job is done once for
232 	 * the whole partition in configure_exceptions().
233 	 */
234 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
235 		unsigned long lpcr = mfspr(SPRN_LPCR);
236 		unsigned long new_lpcr = lpcr;
237 
238 		if (cpu_has_feature(CPU_FTR_ARCH_31)) {
239 			/* P10 DD1 does not have HAIL */
240 			if (pvr_version_is(PVR_POWER10) &&
241 					(mfspr(SPRN_PVR) & 0xf00) == 0x100)
242 				new_lpcr |= LPCR_AIL_3;
243 			else
244 				new_lpcr |= LPCR_HAIL;
245 		} else if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
246 			new_lpcr |= LPCR_AIL_3;
247 		}
248 
249 		if (new_lpcr != lpcr)
250 			mtspr(SPRN_LPCR, new_lpcr);
251 	}
252 
253 	/*
254 	 * Set HFSCR:TM based on CPU features:
255 	 * In the special case of TM no suspend (P9N DD2.1), Linux is
256 	 * told TM is off via the dt-ftrs but told to (partially) use
257 	 * it via OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED. So HFSCR[TM]
258 	 * will be off from dt-ftrs but we need to turn it on for the
259 	 * no suspend case.
260 	 */
261 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
262 		if (cpu_has_feature(CPU_FTR_TM_COMP))
263 			mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) | HFSCR_TM);
264 		else
265 			mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) & ~HFSCR_TM);
266 	}
267 
268 	/* Set IR and DR in PACA MSR */
269 	get_paca()->kernel_msr = MSR_KERNEL;
270 }
271 
272 unsigned long spr_default_dscr = 0;
273 
274 static void __init record_spr_defaults(void)
275 {
276 	if (early_cpu_has_feature(CPU_FTR_DSCR))
277 		spr_default_dscr = mfspr(SPRN_DSCR);
278 }
279 
280 /*
281  * Early initialization entry point. This is called by head.S
282  * with MMU translation disabled. We rely on the "feature" of
283  * the CPU that ignores the top 2 bits of the address in real
284  * mode so we can access kernel globals normally provided we
285  * only toy with things in the RMO region. From here, we do
286  * some early parsing of the device-tree to setup out MEMBLOCK
287  * data structures, and allocate & initialize the hash table
288  * and segment tables so we can start running with translation
289  * enabled.
290  *
291  * It is this function which will call the probe() callback of
292  * the various platform types and copy the matching one to the
293  * global ppc_md structure. Your platform can eventually do
294  * some very early initializations from the probe() routine, but
295  * this is not recommended, be very careful as, for example, the
296  * device-tree is not accessible via normal means at this point.
297  */
298 
299 void __init early_setup(unsigned long dt_ptr)
300 {
301 	static __initdata struct paca_struct boot_paca;
302 
303 	/* -------- printk is _NOT_ safe to use here ! ------- */
304 
305 	/*
306 	 * Assume we're on cpu 0 for now.
307 	 *
308 	 * We need to load a PACA very early for a few reasons.
309 	 *
310 	 * The stack protector canary is stored in the paca, so as soon as we
311 	 * call any stack protected code we need r13 pointing somewhere valid.
312 	 *
313 	 * If we are using kcov it will call in_task() in its instrumentation,
314 	 * which relies on the current task from the PACA.
315 	 *
316 	 * dt_cpu_ftrs_init() calls into generic OF/fdt code, as well as
317 	 * printk(), which can trigger both stack protector and kcov.
318 	 *
319 	 * percpu variables and spin locks also use the paca.
320 	 *
321 	 * So set up a temporary paca. It will be replaced below once we know
322 	 * what CPU we are on.
323 	 */
324 	initialise_paca(&boot_paca, 0);
325 	setup_paca(&boot_paca);
326 	fixup_boot_paca();
327 
328 	/* -------- printk is now safe to use ------- */
329 
330 	/* Try new device tree based feature discovery ... */
331 	if (!dt_cpu_ftrs_init(__va(dt_ptr)))
332 		/* Otherwise use the old style CPU table */
333 		identify_cpu(0, mfspr(SPRN_PVR));
334 
335 	/* Enable early debugging if any specified (see udbg.h) */
336 	udbg_early_init();
337 
338 	udbg_printf(" -> %s(), dt_ptr: 0x%lx\n", __func__, dt_ptr);
339 
340 	/*
341 	 * Do early initialization using the flattened device
342 	 * tree, such as retrieving the physical memory map or
343 	 * calculating/retrieving the hash table size.
344 	 */
345 	early_init_devtree(__va(dt_ptr));
346 
347 	/* Now we know the logical id of our boot cpu, setup the paca. */
348 	if (boot_cpuid != 0) {
349 		/* Poison paca_ptrs[0] again if it's not the boot cpu */
350 		memset(&paca_ptrs[0], 0x88, sizeof(paca_ptrs[0]));
351 	}
352 	setup_paca(paca_ptrs[boot_cpuid]);
353 	fixup_boot_paca();
354 
355 	/*
356 	 * Configure exception handlers. This include setting up trampolines
357 	 * if needed, setting exception endian mode, etc...
358 	 */
359 	configure_exceptions();
360 
361 	/*
362 	 * Configure Kernel Userspace Protection. This needs to happen before
363 	 * feature fixups for platforms that implement this using features.
364 	 */
365 	setup_kup();
366 
367 	/* Apply all the dynamic patching */
368 	apply_feature_fixups();
369 	setup_feature_keys();
370 
371 	/* Initialize the hash table or TLB handling */
372 	early_init_mmu();
373 
374 	early_ioremap_setup();
375 
376 	/*
377 	 * After firmware and early platform setup code has set things up,
378 	 * we note the SPR values for configurable control/performance
379 	 * registers, and use those as initial defaults.
380 	 */
381 	record_spr_defaults();
382 
383 	/*
384 	 * At this point, we can let interrupts switch to virtual mode
385 	 * (the MMU has been setup), so adjust the MSR in the PACA to
386 	 * have IR and DR set and enable AIL if it exists
387 	 */
388 	cpu_ready_for_interrupts();
389 
390 	/*
391 	 * We enable ftrace here, but since we only support DYNAMIC_FTRACE, it
392 	 * will only actually get enabled on the boot cpu much later once
393 	 * ftrace itself has been initialized.
394 	 */
395 	this_cpu_enable_ftrace();
396 
397 	udbg_printf(" <- %s()\n", __func__);
398 
399 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
400 	/*
401 	 * This needs to be done *last* (after the above udbg_printf() even)
402 	 *
403 	 * Right after we return from this function, we turn on the MMU
404 	 * which means the real-mode access trick that btext does will
405 	 * no longer work, it needs to switch to using a real MMU
406 	 * mapping. This call will ensure that it does
407 	 */
408 	btext_map();
409 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
410 }
411 
412 #ifdef CONFIG_SMP
413 void early_setup_secondary(void)
414 {
415 	/* Mark interrupts disabled in PACA */
416 	irq_soft_mask_set(IRQS_DISABLED);
417 
418 	/* Initialize the hash table or TLB handling */
419 	early_init_mmu_secondary();
420 
421 	/* Perform any KUP setup that is per-cpu */
422 	setup_kup();
423 
424 	/*
425 	 * At this point, we can let interrupts switch to virtual mode
426 	 * (the MMU has been setup), so adjust the MSR in the PACA to
427 	 * have IR and DR set.
428 	 */
429 	cpu_ready_for_interrupts();
430 }
431 
432 #endif /* CONFIG_SMP */
433 
434 void panic_smp_self_stop(void)
435 {
436 	hard_irq_disable();
437 	spin_begin();
438 	while (1)
439 		spin_cpu_relax();
440 }
441 
442 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
443 static bool use_spinloop(void)
444 {
445 	if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
446 		/*
447 		 * See comments in head_64.S -- not all platforms insert
448 		 * secondaries at __secondary_hold and wait at the spin
449 		 * loop.
450 		 */
451 		if (firmware_has_feature(FW_FEATURE_OPAL))
452 			return false;
453 		return true;
454 	}
455 
456 	/*
457 	 * When book3e boots from kexec, the ePAPR spin table does
458 	 * not get used.
459 	 */
460 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
461 }
462 
463 void smp_release_cpus(void)
464 {
465 	unsigned long *ptr;
466 	int i;
467 
468 	if (!use_spinloop())
469 		return;
470 
471 	/* All secondary cpus are spinning on a common spinloop, release them
472 	 * all now so they can start to spin on their individual paca
473 	 * spinloops. For non SMP kernels, the secondary cpus never get out
474 	 * of the common spinloop.
475 	 */
476 
477 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
478 			- PHYSICAL_START);
479 	*ptr = ppc_function_entry(generic_secondary_smp_init);
480 
481 	/* And wait a bit for them to catch up */
482 	for (i = 0; i < 100000; i++) {
483 		mb();
484 		HMT_low();
485 		if (spinning_secondaries == 0)
486 			break;
487 		udelay(1);
488 	}
489 	pr_debug("spinning_secondaries = %d\n", spinning_secondaries);
490 }
491 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
492 
493 /*
494  * Initialize some remaining members of the ppc64_caches and systemcfg
495  * structures
496  * (at least until we get rid of them completely). This is mostly some
497  * cache informations about the CPU that will be used by cache flush
498  * routines and/or provided to userland
499  */
500 
501 static void __init init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
502 			    u32 bsize, u32 sets)
503 {
504 	info->size = size;
505 	info->sets = sets;
506 	info->line_size = lsize;
507 	info->block_size = bsize;
508 	info->log_block_size = __ilog2(bsize);
509 	if (bsize)
510 		info->blocks_per_page = PAGE_SIZE / bsize;
511 	else
512 		info->blocks_per_page = 0;
513 
514 	if (sets == 0)
515 		info->assoc = 0xffff;
516 	else
517 		info->assoc = size / (sets * lsize);
518 }
519 
520 static bool __init parse_cache_info(struct device_node *np,
521 				    bool icache,
522 				    struct ppc_cache_info *info)
523 {
524 	static const char *ipropnames[] __initdata = {
525 		"i-cache-size",
526 		"i-cache-sets",
527 		"i-cache-block-size",
528 		"i-cache-line-size",
529 	};
530 	static const char *dpropnames[] __initdata = {
531 		"d-cache-size",
532 		"d-cache-sets",
533 		"d-cache-block-size",
534 		"d-cache-line-size",
535 	};
536 	const char **propnames = icache ? ipropnames : dpropnames;
537 	const __be32 *sizep, *lsizep, *bsizep, *setsp;
538 	u32 size, lsize, bsize, sets;
539 	bool success = true;
540 
541 	size = 0;
542 	sets = -1u;
543 	lsize = bsize = cur_cpu_spec->dcache_bsize;
544 	sizep = of_get_property(np, propnames[0], NULL);
545 	if (sizep != NULL)
546 		size = be32_to_cpu(*sizep);
547 	setsp = of_get_property(np, propnames[1], NULL);
548 	if (setsp != NULL)
549 		sets = be32_to_cpu(*setsp);
550 	bsizep = of_get_property(np, propnames[2], NULL);
551 	lsizep = of_get_property(np, propnames[3], NULL);
552 	if (bsizep == NULL)
553 		bsizep = lsizep;
554 	if (lsizep == NULL)
555 		lsizep = bsizep;
556 	if (lsizep != NULL)
557 		lsize = be32_to_cpu(*lsizep);
558 	if (bsizep != NULL)
559 		bsize = be32_to_cpu(*bsizep);
560 	if (sizep == NULL || bsizep == NULL || lsizep == NULL)
561 		success = false;
562 
563 	/*
564 	 * OF is weird .. it represents fully associative caches
565 	 * as "1 way" which doesn't make much sense and doesn't
566 	 * leave room for direct mapped. We'll assume that 0
567 	 * in OF means direct mapped for that reason.
568 	 */
569 	if (sets == 1)
570 		sets = 0;
571 	else if (sets == 0)
572 		sets = 1;
573 
574 	init_cache_info(info, size, lsize, bsize, sets);
575 
576 	return success;
577 }
578 
579 void __init initialize_cache_info(void)
580 {
581 	struct device_node *cpu = NULL, *l2, *l3 = NULL;
582 	u32 pvr;
583 
584 	/*
585 	 * All shipping POWER8 machines have a firmware bug that
586 	 * puts incorrect information in the device-tree. This will
587 	 * be (hopefully) fixed for future chips but for now hard
588 	 * code the values if we are running on one of these
589 	 */
590 	pvr = PVR_VER(mfspr(SPRN_PVR));
591 	if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
592 	    pvr == PVR_POWER8NVL) {
593 						/* size    lsize   blk  sets */
594 		init_cache_info(&ppc64_caches.l1i, 0x8000,   128,  128, 32);
595 		init_cache_info(&ppc64_caches.l1d, 0x10000,  128,  128, 64);
596 		init_cache_info(&ppc64_caches.l2,  0x80000,  128,  0,   512);
597 		init_cache_info(&ppc64_caches.l3,  0x800000, 128,  0,   8192);
598 	} else
599 		cpu = of_find_node_by_type(NULL, "cpu");
600 
601 	/*
602 	 * We're assuming *all* of the CPUs have the same
603 	 * d-cache and i-cache sizes... -Peter
604 	 */
605 	if (cpu) {
606 		if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
607 			pr_warn("Argh, can't find dcache properties !\n");
608 
609 		if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
610 			pr_warn("Argh, can't find icache properties !\n");
611 
612 		/*
613 		 * Try to find the L2 and L3 if any. Assume they are
614 		 * unified and use the D-side properties.
615 		 */
616 		l2 = of_find_next_cache_node(cpu);
617 		of_node_put(cpu);
618 		if (l2) {
619 			parse_cache_info(l2, false, &ppc64_caches.l2);
620 			l3 = of_find_next_cache_node(l2);
621 			of_node_put(l2);
622 		}
623 		if (l3) {
624 			parse_cache_info(l3, false, &ppc64_caches.l3);
625 			of_node_put(l3);
626 		}
627 	}
628 
629 	/* For use by binfmt_elf */
630 	dcache_bsize = ppc64_caches.l1d.block_size;
631 	icache_bsize = ppc64_caches.l1i.block_size;
632 
633 	cur_cpu_spec->dcache_bsize = dcache_bsize;
634 	cur_cpu_spec->icache_bsize = icache_bsize;
635 }
636 
637 /*
638  * This returns the limit below which memory accesses to the linear
639  * mapping are guarnateed not to cause an architectural exception (e.g.,
640  * TLB or SLB miss fault).
641  *
642  * This is used to allocate PACAs and various interrupt stacks that
643  * that are accessed early in interrupt handlers that must not cause
644  * re-entrant interrupts.
645  */
646 __init u64 ppc64_bolted_size(void)
647 {
648 #ifdef CONFIG_PPC_BOOK3E
649 	/* Freescale BookE bolts the entire linear mapping */
650 	/* XXX: BookE ppc64_rma_limit setup seems to disagree? */
651 	if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E))
652 		return linear_map_top;
653 	/* Other BookE, we assume the first GB is bolted */
654 	return 1ul << 30;
655 #else
656 	/* BookS radix, does not take faults on linear mapping */
657 	if (early_radix_enabled())
658 		return ULONG_MAX;
659 
660 	/* BookS hash, the first segment is bolted */
661 	if (early_mmu_has_feature(MMU_FTR_1T_SEGMENT))
662 		return 1UL << SID_SHIFT_1T;
663 	return 1UL << SID_SHIFT;
664 #endif
665 }
666 
667 static void *__init alloc_stack(unsigned long limit, int cpu)
668 {
669 	void *ptr;
670 
671 	BUILD_BUG_ON(STACK_INT_FRAME_SIZE % 16);
672 
673 	ptr = memblock_alloc_try_nid(THREAD_SIZE, THREAD_ALIGN,
674 				     MEMBLOCK_LOW_LIMIT, limit,
675 				     early_cpu_to_node(cpu));
676 	if (!ptr)
677 		panic("cannot allocate stacks");
678 
679 	return ptr;
680 }
681 
682 void __init irqstack_early_init(void)
683 {
684 	u64 limit = ppc64_bolted_size();
685 	unsigned int i;
686 
687 	/*
688 	 * Interrupt stacks must be in the first segment since we
689 	 * cannot afford to take SLB misses on them. They are not
690 	 * accessed in realmode.
691 	 */
692 	for_each_possible_cpu(i) {
693 		softirq_ctx[i] = alloc_stack(limit, i);
694 		hardirq_ctx[i] = alloc_stack(limit, i);
695 	}
696 }
697 
698 #ifdef CONFIG_PPC_BOOK3E
699 void __init exc_lvl_early_init(void)
700 {
701 	unsigned int i;
702 
703 	for_each_possible_cpu(i) {
704 		void *sp;
705 
706 		sp = alloc_stack(ULONG_MAX, i);
707 		critirq_ctx[i] = sp;
708 		paca_ptrs[i]->crit_kstack = sp + THREAD_SIZE;
709 
710 		sp = alloc_stack(ULONG_MAX, i);
711 		dbgirq_ctx[i] = sp;
712 		paca_ptrs[i]->dbg_kstack = sp + THREAD_SIZE;
713 
714 		sp = alloc_stack(ULONG_MAX, i);
715 		mcheckirq_ctx[i] = sp;
716 		paca_ptrs[i]->mc_kstack = sp + THREAD_SIZE;
717 	}
718 
719 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
720 		patch_exception(0x040, exc_debug_debug_book3e);
721 }
722 #endif
723 
724 /*
725  * Stack space used when we detect a bad kernel stack pointer, and
726  * early in SMP boots before relocation is enabled. Exclusive emergency
727  * stack for machine checks.
728  */
729 void __init emergency_stack_init(void)
730 {
731 	u64 limit, mce_limit;
732 	unsigned int i;
733 
734 	/*
735 	 * Emergency stacks must be under 256MB, we cannot afford to take
736 	 * SLB misses on them. The ABI also requires them to be 128-byte
737 	 * aligned.
738 	 *
739 	 * Since we use these as temporary stacks during secondary CPU
740 	 * bringup, machine check, system reset, and HMI, we need to get
741 	 * at them in real mode. This means they must also be within the RMO
742 	 * region.
743 	 *
744 	 * The IRQ stacks allocated elsewhere in this file are zeroed and
745 	 * initialized in kernel/irq.c. These are initialized here in order
746 	 * to have emergency stacks available as early as possible.
747 	 */
748 	limit = mce_limit = min(ppc64_bolted_size(), ppc64_rma_size);
749 
750 	/*
751 	 * Machine check on pseries calls rtas, but can't use the static
752 	 * rtas_args due to a machine check hitting while the lock is held.
753 	 * rtas args have to be under 4GB, so the machine check stack is
754 	 * limited to 4GB so args can be put on stack.
755 	 */
756 	if (firmware_has_feature(FW_FEATURE_LPAR) && mce_limit > SZ_4G)
757 		mce_limit = SZ_4G;
758 
759 	for_each_possible_cpu(i) {
760 		paca_ptrs[i]->emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
761 
762 #ifdef CONFIG_PPC_BOOK3S_64
763 		/* emergency stack for NMI exception handling. */
764 		paca_ptrs[i]->nmi_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
765 
766 		/* emergency stack for machine check exception handling. */
767 		paca_ptrs[i]->mc_emergency_sp = alloc_stack(mce_limit, i) + THREAD_SIZE;
768 #endif
769 	}
770 }
771 
772 #ifdef CONFIG_SMP
773 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
774 {
775 	if (early_cpu_to_node(from) == early_cpu_to_node(to))
776 		return LOCAL_DISTANCE;
777 	else
778 		return REMOTE_DISTANCE;
779 }
780 
781 static __init int pcpu_cpu_to_node(int cpu)
782 {
783 	return early_cpu_to_node(cpu);
784 }
785 
786 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
787 EXPORT_SYMBOL(__per_cpu_offset);
788 
789 void __init setup_per_cpu_areas(void)
790 {
791 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
792 	size_t atom_size;
793 	unsigned long delta;
794 	unsigned int cpu;
795 	int rc = -EINVAL;
796 
797 	/*
798 	 * BookE and BookS radix are historical values and should be revisited.
799 	 */
800 	if (IS_ENABLED(CONFIG_PPC_BOOK3E)) {
801 		atom_size = SZ_1M;
802 	} else if (radix_enabled()) {
803 		atom_size = PAGE_SIZE;
804 	} else if (IS_ENABLED(CONFIG_PPC_64S_HASH_MMU)) {
805 		/*
806 		 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
807 		 * to group units.  For larger mappings, use 1M atom which
808 		 * should be large enough to contain a number of units.
809 		 */
810 		if (mmu_linear_psize == MMU_PAGE_4K)
811 			atom_size = PAGE_SIZE;
812 		else
813 			atom_size = SZ_1M;
814 	}
815 
816 	if (pcpu_chosen_fc != PCPU_FC_PAGE) {
817 		rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
818 					    pcpu_cpu_to_node);
819 		if (rc)
820 			pr_warn("PERCPU: %s allocator failed (%d), "
821 				"falling back to page size\n",
822 				pcpu_fc_names[pcpu_chosen_fc], rc);
823 	}
824 
825 	if (rc < 0)
826 		rc = pcpu_page_first_chunk(0, pcpu_cpu_to_node);
827 	if (rc < 0)
828 		panic("cannot initialize percpu area (err=%d)", rc);
829 
830 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
831 	for_each_possible_cpu(cpu) {
832                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
833 		paca_ptrs[cpu]->data_offset = __per_cpu_offset[cpu];
834 	}
835 }
836 #endif
837 
838 #ifdef CONFIG_MEMORY_HOTPLUG
839 unsigned long memory_block_size_bytes(void)
840 {
841 	if (ppc_md.memory_block_size)
842 		return ppc_md.memory_block_size();
843 
844 	return MIN_MEMORY_BLOCK_SIZE;
845 }
846 #endif
847 
848 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
849 struct ppc_pci_io ppc_pci_io;
850 EXPORT_SYMBOL(ppc_pci_io);
851 #endif
852 
853 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
854 u64 hw_nmi_get_sample_period(int watchdog_thresh)
855 {
856 	return ppc_proc_freq * watchdog_thresh;
857 }
858 #endif
859 
860 /*
861  * The perf based hardlockup detector breaks PMU event based branches, so
862  * disable it by default. Book3S has a soft-nmi hardlockup detector based
863  * on the decrementer interrupt, so it does not suffer from this problem.
864  *
865  * It is likely to get false positives in KVM guests, so disable it there
866  * by default too. PowerVM will not stop or arbitrarily oversubscribe
867  * CPUs, but give a minimum regular allotment even with SPLPAR, so enable
868  * the detector for non-KVM guests, assume PowerVM.
869  */
870 static int __init disable_hardlockup_detector(void)
871 {
872 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
873 	hardlockup_detector_disable();
874 #else
875 	if (firmware_has_feature(FW_FEATURE_LPAR)) {
876 		if (is_kvm_guest())
877 			hardlockup_detector_disable();
878 	}
879 #endif
880 
881 	return 0;
882 }
883 early_initcall(disable_hardlockup_detector);
884