xref: /openbmc/linux/arch/powerpc/kernel/setup_64.c (revision 78560d41)
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 
34 #include <asm/debugfs.h>
35 #include <asm/io.h>
36 #include <asm/kdump.h>
37 #include <asm/prom.h>
38 #include <asm/processor.h>
39 #include <asm/pgtable.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 
70 #include "setup.h"
71 
72 int spinning_secondaries;
73 u64 ppc64_pft_size;
74 
75 struct ppc64_caches ppc64_caches = {
76 	.l1d = {
77 		.block_size = 0x40,
78 		.log_block_size = 6,
79 	},
80 	.l1i = {
81 		.block_size = 0x40,
82 		.log_block_size = 6
83 	},
84 };
85 EXPORT_SYMBOL_GPL(ppc64_caches);
86 
87 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
88 void __init setup_tlb_core_data(void)
89 {
90 	int cpu;
91 
92 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
93 
94 	for_each_possible_cpu(cpu) {
95 		int first = cpu_first_thread_sibling(cpu);
96 
97 		/*
98 		 * If we boot via kdump on a non-primary thread,
99 		 * make sure we point at the thread that actually
100 		 * set up this TLB.
101 		 */
102 		if (cpu_first_thread_sibling(boot_cpuid) == first)
103 			first = boot_cpuid;
104 
105 		paca_ptrs[cpu]->tcd_ptr = &paca_ptrs[first]->tcd;
106 
107 		/*
108 		 * If we have threads, we need either tlbsrx.
109 		 * or e6500 tablewalk mode, or else TLB handlers
110 		 * will be racy and could produce duplicate entries.
111 		 * Should we panic instead?
112 		 */
113 		WARN_ONCE(smt_enabled_at_boot >= 2 &&
114 			  !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
115 			  book3e_htw_mode != PPC_HTW_E6500,
116 			  "%s: unsupported MMU configuration\n", __func__);
117 	}
118 }
119 #endif
120 
121 #ifdef CONFIG_SMP
122 
123 static char *smt_enabled_cmdline;
124 
125 /* Look for ibm,smt-enabled OF option */
126 void __init check_smt_enabled(void)
127 {
128 	struct device_node *dn;
129 	const char *smt_option;
130 
131 	/* Default to enabling all threads */
132 	smt_enabled_at_boot = threads_per_core;
133 
134 	/* Allow the command line to overrule the OF option */
135 	if (smt_enabled_cmdline) {
136 		if (!strcmp(smt_enabled_cmdline, "on"))
137 			smt_enabled_at_boot = threads_per_core;
138 		else if (!strcmp(smt_enabled_cmdline, "off"))
139 			smt_enabled_at_boot = 0;
140 		else {
141 			int smt;
142 			int rc;
143 
144 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
145 			if (!rc)
146 				smt_enabled_at_boot =
147 					min(threads_per_core, smt);
148 		}
149 	} else {
150 		dn = of_find_node_by_path("/options");
151 		if (dn) {
152 			smt_option = of_get_property(dn, "ibm,smt-enabled",
153 						     NULL);
154 
155 			if (smt_option) {
156 				if (!strcmp(smt_option, "on"))
157 					smt_enabled_at_boot = threads_per_core;
158 				else if (!strcmp(smt_option, "off"))
159 					smt_enabled_at_boot = 0;
160 			}
161 
162 			of_node_put(dn);
163 		}
164 	}
165 }
166 
167 /* Look for smt-enabled= cmdline option */
168 static int __init early_smt_enabled(char *p)
169 {
170 	smt_enabled_cmdline = p;
171 	return 0;
172 }
173 early_param("smt-enabled", early_smt_enabled);
174 
175 #endif /* CONFIG_SMP */
176 
177 /** Fix up paca fields required for the boot cpu */
178 static void __init fixup_boot_paca(void)
179 {
180 	/* The boot cpu is started */
181 	get_paca()->cpu_start = 1;
182 	/* Allow percpu accesses to work until we setup percpu data */
183 	get_paca()->data_offset = 0;
184 	/* Mark interrupts disabled in PACA */
185 	irq_soft_mask_set(IRQS_DISABLED);
186 }
187 
188 static void __init configure_exceptions(void)
189 {
190 	/*
191 	 * Setup the trampolines from the lowmem exception vectors
192 	 * to the kdump kernel when not using a relocatable kernel.
193 	 */
194 	setup_kdump_trampoline();
195 
196 	/* Under a PAPR hypervisor, we need hypercalls */
197 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
198 		/* Enable AIL if possible */
199 		pseries_enable_reloc_on_exc();
200 
201 		/*
202 		 * Tell the hypervisor that we want our exceptions to
203 		 * be taken in little endian mode.
204 		 *
205 		 * We don't call this for big endian as our calling convention
206 		 * makes us always enter in BE, and the call may fail under
207 		 * some circumstances with kdump.
208 		 */
209 #ifdef __LITTLE_ENDIAN__
210 		pseries_little_endian_exceptions();
211 #endif
212 	} else {
213 		/* Set endian mode using OPAL */
214 		if (firmware_has_feature(FW_FEATURE_OPAL))
215 			opal_configure_cores();
216 
217 		/* AIL on native is done in cpu_ready_for_interrupts() */
218 	}
219 }
220 
221 static void cpu_ready_for_interrupts(void)
222 {
223 	/*
224 	 * Enable AIL if supported, and we are in hypervisor mode. This
225 	 * is called once for every processor.
226 	 *
227 	 * If we are not in hypervisor mode the job is done once for
228 	 * the whole partition in configure_exceptions().
229 	 */
230 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
231 	    cpu_has_feature(CPU_FTR_ARCH_207S)) {
232 		unsigned long lpcr = mfspr(SPRN_LPCR);
233 		mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
234 	}
235 
236 	/*
237 	 * Set HFSCR:TM based on CPU features:
238 	 * In the special case of TM no suspend (P9N DD2.1), Linux is
239 	 * told TM is off via the dt-ftrs but told to (partially) use
240 	 * it via OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED. So HFSCR[TM]
241 	 * will be off from dt-ftrs but we need to turn it on for the
242 	 * no suspend case.
243 	 */
244 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
245 		if (cpu_has_feature(CPU_FTR_TM_COMP))
246 			mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) | HFSCR_TM);
247 		else
248 			mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) & ~HFSCR_TM);
249 	}
250 
251 	/* Set IR and DR in PACA MSR */
252 	get_paca()->kernel_msr = MSR_KERNEL;
253 }
254 
255 unsigned long spr_default_dscr = 0;
256 
257 void __init record_spr_defaults(void)
258 {
259 	if (early_cpu_has_feature(CPU_FTR_DSCR))
260 		spr_default_dscr = mfspr(SPRN_DSCR);
261 }
262 
263 /*
264  * Early initialization entry point. This is called by head.S
265  * with MMU translation disabled. We rely on the "feature" of
266  * the CPU that ignores the top 2 bits of the address in real
267  * mode so we can access kernel globals normally provided we
268  * only toy with things in the RMO region. From here, we do
269  * some early parsing of the device-tree to setup out MEMBLOCK
270  * data structures, and allocate & initialize the hash table
271  * and segment tables so we can start running with translation
272  * enabled.
273  *
274  * It is this function which will call the probe() callback of
275  * the various platform types and copy the matching one to the
276  * global ppc_md structure. Your platform can eventually do
277  * some very early initializations from the probe() routine, but
278  * this is not recommended, be very careful as, for example, the
279  * device-tree is not accessible via normal means at this point.
280  */
281 
282 void __init early_setup(unsigned long dt_ptr)
283 {
284 	static __initdata struct paca_struct boot_paca;
285 
286 	/* -------- printk is _NOT_ safe to use here ! ------- */
287 
288 	/* Try new device tree based feature discovery ... */
289 	if (!dt_cpu_ftrs_init(__va(dt_ptr)))
290 		/* Otherwise use the old style CPU table */
291 		identify_cpu(0, mfspr(SPRN_PVR));
292 
293 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
294 	initialise_paca(&boot_paca, 0);
295 	setup_paca(&boot_paca);
296 	fixup_boot_paca();
297 
298 	/* -------- printk is now safe to use ------- */
299 
300 	/* Enable early debugging if any specified (see udbg.h) */
301 	udbg_early_init();
302 
303 	udbg_printf(" -> %s(), dt_ptr: 0x%lx\n", __func__, dt_ptr);
304 
305 	/*
306 	 * Do early initialization using the flattened device
307 	 * tree, such as retrieving the physical memory map or
308 	 * calculating/retrieving the hash table size.
309 	 */
310 	early_init_devtree(__va(dt_ptr));
311 
312 	/* Now we know the logical id of our boot cpu, setup the paca. */
313 	if (boot_cpuid != 0) {
314 		/* Poison paca_ptrs[0] again if it's not the boot cpu */
315 		memset(&paca_ptrs[0], 0x88, sizeof(paca_ptrs[0]));
316 	}
317 	setup_paca(paca_ptrs[boot_cpuid]);
318 	fixup_boot_paca();
319 
320 	/*
321 	 * Configure exception handlers. This include setting up trampolines
322 	 * if needed, setting exception endian mode, etc...
323 	 */
324 	configure_exceptions();
325 
326 	/*
327 	 * Configure Kernel Userspace Protection. This needs to happen before
328 	 * feature fixups for platforms that implement this using features.
329 	 */
330 	setup_kup();
331 
332 	/* Apply all the dynamic patching */
333 	apply_feature_fixups();
334 	setup_feature_keys();
335 
336 	early_ioremap_setup();
337 
338 	/* Initialize the hash table or TLB handling */
339 	early_init_mmu();
340 
341 	/*
342 	 * After firmware and early platform setup code has set things up,
343 	 * we note the SPR values for configurable control/performance
344 	 * registers, and use those as initial defaults.
345 	 */
346 	record_spr_defaults();
347 
348 	/*
349 	 * At this point, we can let interrupts switch to virtual mode
350 	 * (the MMU has been setup), so adjust the MSR in the PACA to
351 	 * have IR and DR set and enable AIL if it exists
352 	 */
353 	cpu_ready_for_interrupts();
354 
355 	/*
356 	 * We enable ftrace here, but since we only support DYNAMIC_FTRACE, it
357 	 * will only actually get enabled on the boot cpu much later once
358 	 * ftrace itself has been initialized.
359 	 */
360 	this_cpu_enable_ftrace();
361 
362 	udbg_printf(" <- %s()\n", __func__);
363 
364 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
365 	/*
366 	 * This needs to be done *last* (after the above udbg_printf() even)
367 	 *
368 	 * Right after we return from this function, we turn on the MMU
369 	 * which means the real-mode access trick that btext does will
370 	 * no longer work, it needs to switch to using a real MMU
371 	 * mapping. This call will ensure that it does
372 	 */
373 	btext_map();
374 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
375 }
376 
377 #ifdef CONFIG_SMP
378 void early_setup_secondary(void)
379 {
380 	/* Mark interrupts disabled in PACA */
381 	irq_soft_mask_set(IRQS_DISABLED);
382 
383 	/* Initialize the hash table or TLB handling */
384 	early_init_mmu_secondary();
385 
386 	/* Perform any KUP setup that is per-cpu */
387 	setup_kup();
388 
389 	/*
390 	 * At this point, we can let interrupts switch to virtual mode
391 	 * (the MMU has been setup), so adjust the MSR in the PACA to
392 	 * have IR and DR set.
393 	 */
394 	cpu_ready_for_interrupts();
395 }
396 
397 #endif /* CONFIG_SMP */
398 
399 void panic_smp_self_stop(void)
400 {
401 	hard_irq_disable();
402 	spin_begin();
403 	while (1)
404 		spin_cpu_relax();
405 }
406 
407 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
408 static bool use_spinloop(void)
409 {
410 	if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
411 		/*
412 		 * See comments in head_64.S -- not all platforms insert
413 		 * secondaries at __secondary_hold and wait at the spin
414 		 * loop.
415 		 */
416 		if (firmware_has_feature(FW_FEATURE_OPAL))
417 			return false;
418 		return true;
419 	}
420 
421 	/*
422 	 * When book3e boots from kexec, the ePAPR spin table does
423 	 * not get used.
424 	 */
425 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
426 }
427 
428 void smp_release_cpus(void)
429 {
430 	unsigned long *ptr;
431 	int i;
432 
433 	if (!use_spinloop())
434 		return;
435 
436 	/* All secondary cpus are spinning on a common spinloop, release them
437 	 * all now so they can start to spin on their individual paca
438 	 * spinloops. For non SMP kernels, the secondary cpus never get out
439 	 * of the common spinloop.
440 	 */
441 
442 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
443 			- PHYSICAL_START);
444 	*ptr = ppc_function_entry(generic_secondary_smp_init);
445 
446 	/* And wait a bit for them to catch up */
447 	for (i = 0; i < 100000; i++) {
448 		mb();
449 		HMT_low();
450 		if (spinning_secondaries == 0)
451 			break;
452 		udelay(1);
453 	}
454 	pr_debug("spinning_secondaries = %d\n", spinning_secondaries);
455 }
456 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
457 
458 /*
459  * Initialize some remaining members of the ppc64_caches and systemcfg
460  * structures
461  * (at least until we get rid of them completely). This is mostly some
462  * cache informations about the CPU that will be used by cache flush
463  * routines and/or provided to userland
464  */
465 
466 static void init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
467 			    u32 bsize, u32 sets)
468 {
469 	info->size = size;
470 	info->sets = sets;
471 	info->line_size = lsize;
472 	info->block_size = bsize;
473 	info->log_block_size = __ilog2(bsize);
474 	if (bsize)
475 		info->blocks_per_page = PAGE_SIZE / bsize;
476 	else
477 		info->blocks_per_page = 0;
478 
479 	if (sets == 0)
480 		info->assoc = 0xffff;
481 	else
482 		info->assoc = size / (sets * lsize);
483 }
484 
485 static bool __init parse_cache_info(struct device_node *np,
486 				    bool icache,
487 				    struct ppc_cache_info *info)
488 {
489 	static const char *ipropnames[] __initdata = {
490 		"i-cache-size",
491 		"i-cache-sets",
492 		"i-cache-block-size",
493 		"i-cache-line-size",
494 	};
495 	static const char *dpropnames[] __initdata = {
496 		"d-cache-size",
497 		"d-cache-sets",
498 		"d-cache-block-size",
499 		"d-cache-line-size",
500 	};
501 	const char **propnames = icache ? ipropnames : dpropnames;
502 	const __be32 *sizep, *lsizep, *bsizep, *setsp;
503 	u32 size, lsize, bsize, sets;
504 	bool success = true;
505 
506 	size = 0;
507 	sets = -1u;
508 	lsize = bsize = cur_cpu_spec->dcache_bsize;
509 	sizep = of_get_property(np, propnames[0], NULL);
510 	if (sizep != NULL)
511 		size = be32_to_cpu(*sizep);
512 	setsp = of_get_property(np, propnames[1], NULL);
513 	if (setsp != NULL)
514 		sets = be32_to_cpu(*setsp);
515 	bsizep = of_get_property(np, propnames[2], NULL);
516 	lsizep = of_get_property(np, propnames[3], NULL);
517 	if (bsizep == NULL)
518 		bsizep = lsizep;
519 	if (lsizep != NULL)
520 		lsize = be32_to_cpu(*lsizep);
521 	if (bsizep != NULL)
522 		bsize = be32_to_cpu(*bsizep);
523 	if (sizep == NULL || bsizep == NULL || lsizep == NULL)
524 		success = false;
525 
526 	/*
527 	 * OF is weird .. it represents fully associative caches
528 	 * as "1 way" which doesn't make much sense and doesn't
529 	 * leave room for direct mapped. We'll assume that 0
530 	 * in OF means direct mapped for that reason.
531 	 */
532 	if (sets == 1)
533 		sets = 0;
534 	else if (sets == 0)
535 		sets = 1;
536 
537 	init_cache_info(info, size, lsize, bsize, sets);
538 
539 	return success;
540 }
541 
542 void __init initialize_cache_info(void)
543 {
544 	struct device_node *cpu = NULL, *l2, *l3 = NULL;
545 	u32 pvr;
546 
547 	/*
548 	 * All shipping POWER8 machines have a firmware bug that
549 	 * puts incorrect information in the device-tree. This will
550 	 * be (hopefully) fixed for future chips but for now hard
551 	 * code the values if we are running on one of these
552 	 */
553 	pvr = PVR_VER(mfspr(SPRN_PVR));
554 	if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
555 	    pvr == PVR_POWER8NVL) {
556 						/* size    lsize   blk  sets */
557 		init_cache_info(&ppc64_caches.l1i, 0x8000,   128,  128, 32);
558 		init_cache_info(&ppc64_caches.l1d, 0x10000,  128,  128, 64);
559 		init_cache_info(&ppc64_caches.l2,  0x80000,  128,  0,   512);
560 		init_cache_info(&ppc64_caches.l3,  0x800000, 128,  0,   8192);
561 	} else
562 		cpu = of_find_node_by_type(NULL, "cpu");
563 
564 	/*
565 	 * We're assuming *all* of the CPUs have the same
566 	 * d-cache and i-cache sizes... -Peter
567 	 */
568 	if (cpu) {
569 		if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
570 			pr_warn("Argh, can't find dcache properties !\n");
571 
572 		if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
573 			pr_warn("Argh, can't find icache properties !\n");
574 
575 		/*
576 		 * Try to find the L2 and L3 if any. Assume they are
577 		 * unified and use the D-side properties.
578 		 */
579 		l2 = of_find_next_cache_node(cpu);
580 		of_node_put(cpu);
581 		if (l2) {
582 			parse_cache_info(l2, false, &ppc64_caches.l2);
583 			l3 = of_find_next_cache_node(l2);
584 			of_node_put(l2);
585 		}
586 		if (l3) {
587 			parse_cache_info(l3, false, &ppc64_caches.l3);
588 			of_node_put(l3);
589 		}
590 	}
591 
592 	/* For use by binfmt_elf */
593 	dcache_bsize = ppc64_caches.l1d.block_size;
594 	icache_bsize = ppc64_caches.l1i.block_size;
595 
596 	cur_cpu_spec->dcache_bsize = dcache_bsize;
597 	cur_cpu_spec->icache_bsize = icache_bsize;
598 }
599 
600 /*
601  * This returns the limit below which memory accesses to the linear
602  * mapping are guarnateed not to cause an architectural exception (e.g.,
603  * TLB or SLB miss fault).
604  *
605  * This is used to allocate PACAs and various interrupt stacks that
606  * that are accessed early in interrupt handlers that must not cause
607  * re-entrant interrupts.
608  */
609 __init u64 ppc64_bolted_size(void)
610 {
611 #ifdef CONFIG_PPC_BOOK3E
612 	/* Freescale BookE bolts the entire linear mapping */
613 	/* XXX: BookE ppc64_rma_limit setup seems to disagree? */
614 	if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E))
615 		return linear_map_top;
616 	/* Other BookE, we assume the first GB is bolted */
617 	return 1ul << 30;
618 #else
619 	/* BookS radix, does not take faults on linear mapping */
620 	if (early_radix_enabled())
621 		return ULONG_MAX;
622 
623 	/* BookS hash, the first segment is bolted */
624 	if (early_mmu_has_feature(MMU_FTR_1T_SEGMENT))
625 		return 1UL << SID_SHIFT_1T;
626 	return 1UL << SID_SHIFT;
627 #endif
628 }
629 
630 static void *__init alloc_stack(unsigned long limit, int cpu)
631 {
632 	void *ptr;
633 
634 	BUILD_BUG_ON(STACK_INT_FRAME_SIZE % 16);
635 
636 	ptr = memblock_alloc_try_nid(THREAD_SIZE, THREAD_ALIGN,
637 				     MEMBLOCK_LOW_LIMIT, limit,
638 				     early_cpu_to_node(cpu));
639 	if (!ptr)
640 		panic("cannot allocate stacks");
641 
642 	return ptr;
643 }
644 
645 void __init irqstack_early_init(void)
646 {
647 	u64 limit = ppc64_bolted_size();
648 	unsigned int i;
649 
650 	/*
651 	 * Interrupt stacks must be in the first segment since we
652 	 * cannot afford to take SLB misses on them. They are not
653 	 * accessed in realmode.
654 	 */
655 	for_each_possible_cpu(i) {
656 		softirq_ctx[i] = alloc_stack(limit, i);
657 		hardirq_ctx[i] = alloc_stack(limit, i);
658 	}
659 }
660 
661 #ifdef CONFIG_PPC_BOOK3E
662 void __init exc_lvl_early_init(void)
663 {
664 	unsigned int i;
665 
666 	for_each_possible_cpu(i) {
667 		void *sp;
668 
669 		sp = alloc_stack(ULONG_MAX, i);
670 		critirq_ctx[i] = sp;
671 		paca_ptrs[i]->crit_kstack = sp + THREAD_SIZE;
672 
673 		sp = alloc_stack(ULONG_MAX, i);
674 		dbgirq_ctx[i] = sp;
675 		paca_ptrs[i]->dbg_kstack = sp + THREAD_SIZE;
676 
677 		sp = alloc_stack(ULONG_MAX, i);
678 		mcheckirq_ctx[i] = sp;
679 		paca_ptrs[i]->mc_kstack = sp + THREAD_SIZE;
680 	}
681 
682 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
683 		patch_exception(0x040, exc_debug_debug_book3e);
684 }
685 #endif
686 
687 /*
688  * Stack space used when we detect a bad kernel stack pointer, and
689  * early in SMP boots before relocation is enabled. Exclusive emergency
690  * stack for machine checks.
691  */
692 void __init emergency_stack_init(void)
693 {
694 	u64 limit;
695 	unsigned int i;
696 
697 	/*
698 	 * Emergency stacks must be under 256MB, we cannot afford to take
699 	 * SLB misses on them. The ABI also requires them to be 128-byte
700 	 * aligned.
701 	 *
702 	 * Since we use these as temporary stacks during secondary CPU
703 	 * bringup, machine check, system reset, and HMI, we need to get
704 	 * at them in real mode. This means they must also be within the RMO
705 	 * region.
706 	 *
707 	 * The IRQ stacks allocated elsewhere in this file are zeroed and
708 	 * initialized in kernel/irq.c. These are initialized here in order
709 	 * to have emergency stacks available as early as possible.
710 	 */
711 	limit = min(ppc64_bolted_size(), ppc64_rma_size);
712 
713 	for_each_possible_cpu(i) {
714 		paca_ptrs[i]->emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
715 
716 #ifdef CONFIG_PPC_BOOK3S_64
717 		/* emergency stack for NMI exception handling. */
718 		paca_ptrs[i]->nmi_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
719 
720 		/* emergency stack for machine check exception handling. */
721 		paca_ptrs[i]->mc_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
722 #endif
723 	}
724 }
725 
726 #ifdef CONFIG_SMP
727 #define PCPU_DYN_SIZE		()
728 
729 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
730 {
731 	return memblock_alloc_try_nid(size, align, __pa(MAX_DMA_ADDRESS),
732 				      MEMBLOCK_ALLOC_ACCESSIBLE,
733 				      early_cpu_to_node(cpu));
734 
735 }
736 
737 static void __init pcpu_fc_free(void *ptr, size_t size)
738 {
739 	memblock_free(__pa(ptr), size);
740 }
741 
742 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
743 {
744 	if (early_cpu_to_node(from) == early_cpu_to_node(to))
745 		return LOCAL_DISTANCE;
746 	else
747 		return REMOTE_DISTANCE;
748 }
749 
750 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
751 EXPORT_SYMBOL(__per_cpu_offset);
752 
753 void __init setup_per_cpu_areas(void)
754 {
755 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
756 	size_t atom_size;
757 	unsigned long delta;
758 	unsigned int cpu;
759 	int rc;
760 
761 	/*
762 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
763 	 * to group units.  For larger mappings, use 1M atom which
764 	 * should be large enough to contain a number of units.
765 	 */
766 	if (mmu_linear_psize == MMU_PAGE_4K)
767 		atom_size = PAGE_SIZE;
768 	else
769 		atom_size = 1 << 20;
770 
771 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
772 				    pcpu_fc_alloc, pcpu_fc_free);
773 	if (rc < 0)
774 		panic("cannot initialize percpu area (err=%d)", rc);
775 
776 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
777 	for_each_possible_cpu(cpu) {
778                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
779 		paca_ptrs[cpu]->data_offset = __per_cpu_offset[cpu];
780 	}
781 }
782 #endif
783 
784 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
785 unsigned long memory_block_size_bytes(void)
786 {
787 	if (ppc_md.memory_block_size)
788 		return ppc_md.memory_block_size();
789 
790 	return MIN_MEMORY_BLOCK_SIZE;
791 }
792 #endif
793 
794 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
795 struct ppc_pci_io ppc_pci_io;
796 EXPORT_SYMBOL(ppc_pci_io);
797 #endif
798 
799 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
800 u64 hw_nmi_get_sample_period(int watchdog_thresh)
801 {
802 	return ppc_proc_freq * watchdog_thresh;
803 }
804 #endif
805 
806 /*
807  * The perf based hardlockup detector breaks PMU event based branches, so
808  * disable it by default. Book3S has a soft-nmi hardlockup detector based
809  * on the decrementer interrupt, so it does not suffer from this problem.
810  *
811  * It is likely to get false positives in VM guests, so disable it there
812  * by default too.
813  */
814 static int __init disable_hardlockup_detector(void)
815 {
816 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
817 	hardlockup_detector_disable();
818 #else
819 	if (firmware_has_feature(FW_FEATURE_LPAR))
820 		hardlockup_detector_disable();
821 #endif
822 
823 	return 0;
824 }
825 early_initcall(disable_hardlockup_detector);
826 
827 #ifdef CONFIG_PPC_BOOK3S_64
828 static enum l1d_flush_type enabled_flush_types;
829 static void *l1d_flush_fallback_area;
830 static bool no_rfi_flush;
831 bool rfi_flush;
832 
833 static int __init handle_no_rfi_flush(char *p)
834 {
835 	pr_info("rfi-flush: disabled on command line.");
836 	no_rfi_flush = true;
837 	return 0;
838 }
839 early_param("no_rfi_flush", handle_no_rfi_flush);
840 
841 /*
842  * The RFI flush is not KPTI, but because users will see doco that says to use
843  * nopti we hijack that option here to also disable the RFI flush.
844  */
845 static int __init handle_no_pti(char *p)
846 {
847 	pr_info("rfi-flush: disabling due to 'nopti' on command line.\n");
848 	handle_no_rfi_flush(NULL);
849 	return 0;
850 }
851 early_param("nopti", handle_no_pti);
852 
853 static void do_nothing(void *unused)
854 {
855 	/*
856 	 * We don't need to do the flush explicitly, just enter+exit kernel is
857 	 * sufficient, the RFI exit handlers will do the right thing.
858 	 */
859 }
860 
861 void rfi_flush_enable(bool enable)
862 {
863 	if (enable) {
864 		do_rfi_flush_fixups(enabled_flush_types);
865 		on_each_cpu(do_nothing, NULL, 1);
866 	} else
867 		do_rfi_flush_fixups(L1D_FLUSH_NONE);
868 
869 	rfi_flush = enable;
870 }
871 
872 static void __ref init_fallback_flush(void)
873 {
874 	u64 l1d_size, limit;
875 	int cpu;
876 
877 	/* Only allocate the fallback flush area once (at boot time). */
878 	if (l1d_flush_fallback_area)
879 		return;
880 
881 	l1d_size = ppc64_caches.l1d.size;
882 
883 	/*
884 	 * If there is no d-cache-size property in the device tree, l1d_size
885 	 * could be zero. That leads to the loop in the asm wrapping around to
886 	 * 2^64-1, and then walking off the end of the fallback area and
887 	 * eventually causing a page fault which is fatal. Just default to
888 	 * something vaguely sane.
889 	 */
890 	if (!l1d_size)
891 		l1d_size = (64 * 1024);
892 
893 	limit = min(ppc64_bolted_size(), ppc64_rma_size);
894 
895 	/*
896 	 * Align to L1d size, and size it at 2x L1d size, to catch possible
897 	 * hardware prefetch runoff. We don't have a recipe for load patterns to
898 	 * reliably avoid the prefetcher.
899 	 */
900 	l1d_flush_fallback_area = memblock_alloc_try_nid(l1d_size * 2,
901 						l1d_size, MEMBLOCK_LOW_LIMIT,
902 						limit, NUMA_NO_NODE);
903 	if (!l1d_flush_fallback_area)
904 		panic("%s: Failed to allocate %llu bytes align=0x%llx max_addr=%pa\n",
905 		      __func__, l1d_size * 2, l1d_size, &limit);
906 
907 
908 	for_each_possible_cpu(cpu) {
909 		struct paca_struct *paca = paca_ptrs[cpu];
910 		paca->rfi_flush_fallback_area = l1d_flush_fallback_area;
911 		paca->l1d_flush_size = l1d_size;
912 	}
913 }
914 
915 void setup_rfi_flush(enum l1d_flush_type types, bool enable)
916 {
917 	if (types & L1D_FLUSH_FALLBACK) {
918 		pr_info("rfi-flush: fallback displacement flush available\n");
919 		init_fallback_flush();
920 	}
921 
922 	if (types & L1D_FLUSH_ORI)
923 		pr_info("rfi-flush: ori type flush available\n");
924 
925 	if (types & L1D_FLUSH_MTTRIG)
926 		pr_info("rfi-flush: mttrig type flush available\n");
927 
928 	enabled_flush_types = types;
929 
930 	if (!no_rfi_flush && !cpu_mitigations_off())
931 		rfi_flush_enable(enable);
932 }
933 
934 #ifdef CONFIG_DEBUG_FS
935 static int rfi_flush_set(void *data, u64 val)
936 {
937 	bool enable;
938 
939 	if (val == 1)
940 		enable = true;
941 	else if (val == 0)
942 		enable = false;
943 	else
944 		return -EINVAL;
945 
946 	/* Only do anything if we're changing state */
947 	if (enable != rfi_flush)
948 		rfi_flush_enable(enable);
949 
950 	return 0;
951 }
952 
953 static int rfi_flush_get(void *data, u64 *val)
954 {
955 	*val = rfi_flush ? 1 : 0;
956 	return 0;
957 }
958 
959 DEFINE_SIMPLE_ATTRIBUTE(fops_rfi_flush, rfi_flush_get, rfi_flush_set, "%llu\n");
960 
961 static __init int rfi_flush_debugfs_init(void)
962 {
963 	debugfs_create_file("rfi_flush", 0600, powerpc_debugfs_root, NULL, &fops_rfi_flush);
964 	return 0;
965 }
966 device_initcall(rfi_flush_debugfs_init);
967 #endif
968 #endif /* CONFIG_PPC_BOOK3S_64 */
969