xref: /openbmc/linux/arch/powerpc/kernel/setup_64.c (revision 4ce94eab)
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/debugfs.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 #include <asm/asm-prototypes.h>
71 
72 #include "setup.h"
73 
74 int spinning_secondaries;
75 u64 ppc64_pft_size;
76 
77 struct ppc64_caches ppc64_caches = {
78 	.l1d = {
79 		.block_size = 0x40,
80 		.log_block_size = 6,
81 	},
82 	.l1i = {
83 		.block_size = 0x40,
84 		.log_block_size = 6
85 	},
86 };
87 EXPORT_SYMBOL_GPL(ppc64_caches);
88 
89 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
90 void __init setup_tlb_core_data(void)
91 {
92 	int cpu;
93 
94 	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
95 
96 	for_each_possible_cpu(cpu) {
97 		int first = cpu_first_thread_sibling(cpu);
98 
99 		/*
100 		 * If we boot via kdump on a non-primary thread,
101 		 * make sure we point at the thread that actually
102 		 * set up this TLB.
103 		 */
104 		if (cpu_first_thread_sibling(boot_cpuid) == first)
105 			first = boot_cpuid;
106 
107 		paca_ptrs[cpu]->tcd_ptr = &paca_ptrs[first]->tcd;
108 
109 		/*
110 		 * If we have threads, we need either tlbsrx.
111 		 * or e6500 tablewalk mode, or else TLB handlers
112 		 * will be racy and could produce duplicate entries.
113 		 * Should we panic instead?
114 		 */
115 		WARN_ONCE(smt_enabled_at_boot >= 2 &&
116 			  !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
117 			  book3e_htw_mode != PPC_HTW_E6500,
118 			  "%s: unsupported MMU configuration\n", __func__);
119 	}
120 }
121 #endif
122 
123 #ifdef CONFIG_SMP
124 
125 static char *smt_enabled_cmdline;
126 
127 /* Look for ibm,smt-enabled OF option */
128 void __init check_smt_enabled(void)
129 {
130 	struct device_node *dn;
131 	const char *smt_option;
132 
133 	/* Default to enabling all threads */
134 	smt_enabled_at_boot = threads_per_core;
135 
136 	/* Allow the command line to overrule the OF option */
137 	if (smt_enabled_cmdline) {
138 		if (!strcmp(smt_enabled_cmdline, "on"))
139 			smt_enabled_at_boot = threads_per_core;
140 		else if (!strcmp(smt_enabled_cmdline, "off"))
141 			smt_enabled_at_boot = 0;
142 		else {
143 			int smt;
144 			int rc;
145 
146 			rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
147 			if (!rc)
148 				smt_enabled_at_boot =
149 					min(threads_per_core, smt);
150 		}
151 	} else {
152 		dn = of_find_node_by_path("/options");
153 		if (dn) {
154 			smt_option = of_get_property(dn, "ibm,smt-enabled",
155 						     NULL);
156 
157 			if (smt_option) {
158 				if (!strcmp(smt_option, "on"))
159 					smt_enabled_at_boot = threads_per_core;
160 				else if (!strcmp(smt_option, "off"))
161 					smt_enabled_at_boot = 0;
162 			}
163 
164 			of_node_put(dn);
165 		}
166 	}
167 }
168 
169 /* Look for smt-enabled= cmdline option */
170 static int __init early_smt_enabled(char *p)
171 {
172 	smt_enabled_cmdline = p;
173 	return 0;
174 }
175 early_param("smt-enabled", early_smt_enabled);
176 
177 #endif /* CONFIG_SMP */
178 
179 /** Fix up paca fields required for the boot cpu */
180 static void __init fixup_boot_paca(void)
181 {
182 	/* The boot cpu is started */
183 	get_paca()->cpu_start = 1;
184 	/* Allow percpu accesses to work until we setup percpu data */
185 	get_paca()->data_offset = 0;
186 	/* Mark interrupts disabled in PACA */
187 	irq_soft_mask_set(IRQS_DISABLED);
188 }
189 
190 static void __init configure_exceptions(void)
191 {
192 	/*
193 	 * Setup the trampolines from the lowmem exception vectors
194 	 * to the kdump kernel when not using a relocatable kernel.
195 	 */
196 	setup_kdump_trampoline();
197 
198 	/* Under a PAPR hypervisor, we need hypercalls */
199 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
200 		/* Enable AIL if possible */
201 		if (!pseries_enable_reloc_on_exc()) {
202 			init_task.thread.fscr &= ~FSCR_SCV;
203 			cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_SCV;
204 		}
205 
206 		/*
207 		 * Tell the hypervisor that we want our exceptions to
208 		 * be taken in little endian mode.
209 		 *
210 		 * We don't call this for big endian as our calling convention
211 		 * makes us always enter in BE, and the call may fail under
212 		 * some circumstances with kdump.
213 		 */
214 #ifdef __LITTLE_ENDIAN__
215 		pseries_little_endian_exceptions();
216 #endif
217 	} else {
218 		/* Set endian mode using OPAL */
219 		if (firmware_has_feature(FW_FEATURE_OPAL))
220 			opal_configure_cores();
221 
222 		/* AIL on native is done in cpu_ready_for_interrupts() */
223 	}
224 }
225 
226 static void cpu_ready_for_interrupts(void)
227 {
228 	/*
229 	 * Enable AIL if supported, and we are in hypervisor mode. This
230 	 * is called once for every processor.
231 	 *
232 	 * If we are not in hypervisor mode the job is done once for
233 	 * the whole partition in configure_exceptions().
234 	 */
235 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
236 	    cpu_has_feature(CPU_FTR_ARCH_207S)) {
237 		unsigned long lpcr = mfspr(SPRN_LPCR);
238 		mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
239 	}
240 
241 	/*
242 	 * Set HFSCR:TM based on CPU features:
243 	 * In the special case of TM no suspend (P9N DD2.1), Linux is
244 	 * told TM is off via the dt-ftrs but told to (partially) use
245 	 * it via OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED. So HFSCR[TM]
246 	 * will be off from dt-ftrs but we need to turn it on for the
247 	 * no suspend case.
248 	 */
249 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
250 		if (cpu_has_feature(CPU_FTR_TM_COMP))
251 			mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) | HFSCR_TM);
252 		else
253 			mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) & ~HFSCR_TM);
254 	}
255 
256 	/* Set IR and DR in PACA MSR */
257 	get_paca()->kernel_msr = MSR_KERNEL;
258 }
259 
260 unsigned long spr_default_dscr = 0;
261 
262 static void __init record_spr_defaults(void)
263 {
264 	if (early_cpu_has_feature(CPU_FTR_DSCR))
265 		spr_default_dscr = mfspr(SPRN_DSCR);
266 }
267 
268 /*
269  * Early initialization entry point. This is called by head.S
270  * with MMU translation disabled. We rely on the "feature" of
271  * the CPU that ignores the top 2 bits of the address in real
272  * mode so we can access kernel globals normally provided we
273  * only toy with things in the RMO region. From here, we do
274  * some early parsing of the device-tree to setup out MEMBLOCK
275  * data structures, and allocate & initialize the hash table
276  * and segment tables so we can start running with translation
277  * enabled.
278  *
279  * It is this function which will call the probe() callback of
280  * the various platform types and copy the matching one to the
281  * global ppc_md structure. Your platform can eventually do
282  * some very early initializations from the probe() routine, but
283  * this is not recommended, be very careful as, for example, the
284  * device-tree is not accessible via normal means at this point.
285  */
286 
287 void __init early_setup(unsigned long dt_ptr)
288 {
289 	static __initdata struct paca_struct boot_paca;
290 
291 	/* -------- printk is _NOT_ safe to use here ! ------- */
292 
293 	/*
294 	 * Assume we're on cpu 0 for now.
295 	 *
296 	 * We need to load a PACA very early for a few reasons.
297 	 *
298 	 * The stack protector canary is stored in the paca, so as soon as we
299 	 * call any stack protected code we need r13 pointing somewhere valid.
300 	 *
301 	 * If we are using kcov it will call in_task() in its instrumentation,
302 	 * which relies on the current task from the PACA.
303 	 *
304 	 * dt_cpu_ftrs_init() calls into generic OF/fdt code, as well as
305 	 * printk(), which can trigger both stack protector and kcov.
306 	 *
307 	 * percpu variables and spin locks also use the paca.
308 	 *
309 	 * So set up a temporary paca. It will be replaced below once we know
310 	 * what CPU we are on.
311 	 */
312 	initialise_paca(&boot_paca, 0);
313 	setup_paca(&boot_paca);
314 	fixup_boot_paca();
315 
316 	/* -------- printk is now safe to use ------- */
317 
318 	/* Try new device tree based feature discovery ... */
319 	if (!dt_cpu_ftrs_init(__va(dt_ptr)))
320 		/* Otherwise use the old style CPU table */
321 		identify_cpu(0, mfspr(SPRN_PVR));
322 
323 	/* Enable early debugging if any specified (see udbg.h) */
324 	udbg_early_init();
325 
326 	udbg_printf(" -> %s(), dt_ptr: 0x%lx\n", __func__, dt_ptr);
327 
328 	/*
329 	 * Do early initialization using the flattened device
330 	 * tree, such as retrieving the physical memory map or
331 	 * calculating/retrieving the hash table size.
332 	 */
333 	early_init_devtree(__va(dt_ptr));
334 
335 	/* Now we know the logical id of our boot cpu, setup the paca. */
336 	if (boot_cpuid != 0) {
337 		/* Poison paca_ptrs[0] again if it's not the boot cpu */
338 		memset(&paca_ptrs[0], 0x88, sizeof(paca_ptrs[0]));
339 	}
340 	setup_paca(paca_ptrs[boot_cpuid]);
341 	fixup_boot_paca();
342 
343 	/*
344 	 * Configure exception handlers. This include setting up trampolines
345 	 * if needed, setting exception endian mode, etc...
346 	 */
347 	configure_exceptions();
348 
349 	/*
350 	 * Configure Kernel Userspace Protection. This needs to happen before
351 	 * feature fixups for platforms that implement this using features.
352 	 */
353 	setup_kup();
354 
355 	/* Apply all the dynamic patching */
356 	apply_feature_fixups();
357 	setup_feature_keys();
358 
359 	early_ioremap_setup();
360 
361 	/* Initialize the hash table or TLB handling */
362 	early_init_mmu();
363 
364 	/*
365 	 * After firmware and early platform setup code has set things up,
366 	 * we note the SPR values for configurable control/performance
367 	 * registers, and use those as initial defaults.
368 	 */
369 	record_spr_defaults();
370 
371 	/*
372 	 * At this point, we can let interrupts switch to virtual mode
373 	 * (the MMU has been setup), so adjust the MSR in the PACA to
374 	 * have IR and DR set and enable AIL if it exists
375 	 */
376 	cpu_ready_for_interrupts();
377 
378 	/*
379 	 * We enable ftrace here, but since we only support DYNAMIC_FTRACE, it
380 	 * will only actually get enabled on the boot cpu much later once
381 	 * ftrace itself has been initialized.
382 	 */
383 	this_cpu_enable_ftrace();
384 
385 	udbg_printf(" <- %s()\n", __func__);
386 
387 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
388 	/*
389 	 * This needs to be done *last* (after the above udbg_printf() even)
390 	 *
391 	 * Right after we return from this function, we turn on the MMU
392 	 * which means the real-mode access trick that btext does will
393 	 * no longer work, it needs to switch to using a real MMU
394 	 * mapping. This call will ensure that it does
395 	 */
396 	btext_map();
397 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
398 }
399 
400 #ifdef CONFIG_SMP
401 void early_setup_secondary(void)
402 {
403 	/* Mark interrupts disabled in PACA */
404 	irq_soft_mask_set(IRQS_DISABLED);
405 
406 	/* Initialize the hash table or TLB handling */
407 	early_init_mmu_secondary();
408 
409 	/* Perform any KUP setup that is per-cpu */
410 	setup_kup();
411 
412 	/*
413 	 * At this point, we can let interrupts switch to virtual mode
414 	 * (the MMU has been setup), so adjust the MSR in the PACA to
415 	 * have IR and DR set.
416 	 */
417 	cpu_ready_for_interrupts();
418 }
419 
420 #endif /* CONFIG_SMP */
421 
422 void panic_smp_self_stop(void)
423 {
424 	hard_irq_disable();
425 	spin_begin();
426 	while (1)
427 		spin_cpu_relax();
428 }
429 
430 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
431 static bool use_spinloop(void)
432 {
433 	if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
434 		/*
435 		 * See comments in head_64.S -- not all platforms insert
436 		 * secondaries at __secondary_hold and wait at the spin
437 		 * loop.
438 		 */
439 		if (firmware_has_feature(FW_FEATURE_OPAL))
440 			return false;
441 		return true;
442 	}
443 
444 	/*
445 	 * When book3e boots from kexec, the ePAPR spin table does
446 	 * not get used.
447 	 */
448 	return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
449 }
450 
451 void smp_release_cpus(void)
452 {
453 	unsigned long *ptr;
454 	int i;
455 
456 	if (!use_spinloop())
457 		return;
458 
459 	/* All secondary cpus are spinning on a common spinloop, release them
460 	 * all now so they can start to spin on their individual paca
461 	 * spinloops. For non SMP kernels, the secondary cpus never get out
462 	 * of the common spinloop.
463 	 */
464 
465 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
466 			- PHYSICAL_START);
467 	*ptr = ppc_function_entry(generic_secondary_smp_init);
468 
469 	/* And wait a bit for them to catch up */
470 	for (i = 0; i < 100000; i++) {
471 		mb();
472 		HMT_low();
473 		if (spinning_secondaries == 0)
474 			break;
475 		udelay(1);
476 	}
477 	pr_debug("spinning_secondaries = %d\n", spinning_secondaries);
478 }
479 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
480 
481 /*
482  * Initialize some remaining members of the ppc64_caches and systemcfg
483  * structures
484  * (at least until we get rid of them completely). This is mostly some
485  * cache informations about the CPU that will be used by cache flush
486  * routines and/or provided to userland
487  */
488 
489 static void init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
490 			    u32 bsize, u32 sets)
491 {
492 	info->size = size;
493 	info->sets = sets;
494 	info->line_size = lsize;
495 	info->block_size = bsize;
496 	info->log_block_size = __ilog2(bsize);
497 	if (bsize)
498 		info->blocks_per_page = PAGE_SIZE / bsize;
499 	else
500 		info->blocks_per_page = 0;
501 
502 	if (sets == 0)
503 		info->assoc = 0xffff;
504 	else
505 		info->assoc = size / (sets * lsize);
506 }
507 
508 static bool __init parse_cache_info(struct device_node *np,
509 				    bool icache,
510 				    struct ppc_cache_info *info)
511 {
512 	static const char *ipropnames[] __initdata = {
513 		"i-cache-size",
514 		"i-cache-sets",
515 		"i-cache-block-size",
516 		"i-cache-line-size",
517 	};
518 	static const char *dpropnames[] __initdata = {
519 		"d-cache-size",
520 		"d-cache-sets",
521 		"d-cache-block-size",
522 		"d-cache-line-size",
523 	};
524 	const char **propnames = icache ? ipropnames : dpropnames;
525 	const __be32 *sizep, *lsizep, *bsizep, *setsp;
526 	u32 size, lsize, bsize, sets;
527 	bool success = true;
528 
529 	size = 0;
530 	sets = -1u;
531 	lsize = bsize = cur_cpu_spec->dcache_bsize;
532 	sizep = of_get_property(np, propnames[0], NULL);
533 	if (sizep != NULL)
534 		size = be32_to_cpu(*sizep);
535 	setsp = of_get_property(np, propnames[1], NULL);
536 	if (setsp != NULL)
537 		sets = be32_to_cpu(*setsp);
538 	bsizep = of_get_property(np, propnames[2], NULL);
539 	lsizep = of_get_property(np, propnames[3], NULL);
540 	if (bsizep == NULL)
541 		bsizep = lsizep;
542 	if (lsizep == NULL)
543 		lsizep = bsizep;
544 	if (lsizep != NULL)
545 		lsize = be32_to_cpu(*lsizep);
546 	if (bsizep != NULL)
547 		bsize = be32_to_cpu(*bsizep);
548 	if (sizep == NULL || bsizep == NULL || lsizep == NULL)
549 		success = false;
550 
551 	/*
552 	 * OF is weird .. it represents fully associative caches
553 	 * as "1 way" which doesn't make much sense and doesn't
554 	 * leave room for direct mapped. We'll assume that 0
555 	 * in OF means direct mapped for that reason.
556 	 */
557 	if (sets == 1)
558 		sets = 0;
559 	else if (sets == 0)
560 		sets = 1;
561 
562 	init_cache_info(info, size, lsize, bsize, sets);
563 
564 	return success;
565 }
566 
567 void __init initialize_cache_info(void)
568 {
569 	struct device_node *cpu = NULL, *l2, *l3 = NULL;
570 	u32 pvr;
571 
572 	/*
573 	 * All shipping POWER8 machines have a firmware bug that
574 	 * puts incorrect information in the device-tree. This will
575 	 * be (hopefully) fixed for future chips but for now hard
576 	 * code the values if we are running on one of these
577 	 */
578 	pvr = PVR_VER(mfspr(SPRN_PVR));
579 	if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
580 	    pvr == PVR_POWER8NVL) {
581 						/* size    lsize   blk  sets */
582 		init_cache_info(&ppc64_caches.l1i, 0x8000,   128,  128, 32);
583 		init_cache_info(&ppc64_caches.l1d, 0x10000,  128,  128, 64);
584 		init_cache_info(&ppc64_caches.l2,  0x80000,  128,  0,   512);
585 		init_cache_info(&ppc64_caches.l3,  0x800000, 128,  0,   8192);
586 	} else
587 		cpu = of_find_node_by_type(NULL, "cpu");
588 
589 	/*
590 	 * We're assuming *all* of the CPUs have the same
591 	 * d-cache and i-cache sizes... -Peter
592 	 */
593 	if (cpu) {
594 		if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
595 			pr_warn("Argh, can't find dcache properties !\n");
596 
597 		if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
598 			pr_warn("Argh, can't find icache properties !\n");
599 
600 		/*
601 		 * Try to find the L2 and L3 if any. Assume they are
602 		 * unified and use the D-side properties.
603 		 */
604 		l2 = of_find_next_cache_node(cpu);
605 		of_node_put(cpu);
606 		if (l2) {
607 			parse_cache_info(l2, false, &ppc64_caches.l2);
608 			l3 = of_find_next_cache_node(l2);
609 			of_node_put(l2);
610 		}
611 		if (l3) {
612 			parse_cache_info(l3, false, &ppc64_caches.l3);
613 			of_node_put(l3);
614 		}
615 	}
616 
617 	/* For use by binfmt_elf */
618 	dcache_bsize = ppc64_caches.l1d.block_size;
619 	icache_bsize = ppc64_caches.l1i.block_size;
620 
621 	cur_cpu_spec->dcache_bsize = dcache_bsize;
622 	cur_cpu_spec->icache_bsize = icache_bsize;
623 }
624 
625 /*
626  * This returns the limit below which memory accesses to the linear
627  * mapping are guarnateed not to cause an architectural exception (e.g.,
628  * TLB or SLB miss fault).
629  *
630  * This is used to allocate PACAs and various interrupt stacks that
631  * that are accessed early in interrupt handlers that must not cause
632  * re-entrant interrupts.
633  */
634 __init u64 ppc64_bolted_size(void)
635 {
636 #ifdef CONFIG_PPC_BOOK3E
637 	/* Freescale BookE bolts the entire linear mapping */
638 	/* XXX: BookE ppc64_rma_limit setup seems to disagree? */
639 	if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E))
640 		return linear_map_top;
641 	/* Other BookE, we assume the first GB is bolted */
642 	return 1ul << 30;
643 #else
644 	/* BookS radix, does not take faults on linear mapping */
645 	if (early_radix_enabled())
646 		return ULONG_MAX;
647 
648 	/* BookS hash, the first segment is bolted */
649 	if (early_mmu_has_feature(MMU_FTR_1T_SEGMENT))
650 		return 1UL << SID_SHIFT_1T;
651 	return 1UL << SID_SHIFT;
652 #endif
653 }
654 
655 static void *__init alloc_stack(unsigned long limit, int cpu)
656 {
657 	void *ptr;
658 
659 	BUILD_BUG_ON(STACK_INT_FRAME_SIZE % 16);
660 
661 	ptr = memblock_alloc_try_nid(THREAD_SIZE, THREAD_ALIGN,
662 				     MEMBLOCK_LOW_LIMIT, limit,
663 				     early_cpu_to_node(cpu));
664 	if (!ptr)
665 		panic("cannot allocate stacks");
666 
667 	return ptr;
668 }
669 
670 void __init irqstack_early_init(void)
671 {
672 	u64 limit = ppc64_bolted_size();
673 	unsigned int i;
674 
675 	/*
676 	 * Interrupt stacks must be in the first segment since we
677 	 * cannot afford to take SLB misses on them. They are not
678 	 * accessed in realmode.
679 	 */
680 	for_each_possible_cpu(i) {
681 		softirq_ctx[i] = alloc_stack(limit, i);
682 		hardirq_ctx[i] = alloc_stack(limit, i);
683 	}
684 }
685 
686 #ifdef CONFIG_PPC_BOOK3E
687 void __init exc_lvl_early_init(void)
688 {
689 	unsigned int i;
690 
691 	for_each_possible_cpu(i) {
692 		void *sp;
693 
694 		sp = alloc_stack(ULONG_MAX, i);
695 		critirq_ctx[i] = sp;
696 		paca_ptrs[i]->crit_kstack = sp + THREAD_SIZE;
697 
698 		sp = alloc_stack(ULONG_MAX, i);
699 		dbgirq_ctx[i] = sp;
700 		paca_ptrs[i]->dbg_kstack = sp + THREAD_SIZE;
701 
702 		sp = alloc_stack(ULONG_MAX, i);
703 		mcheckirq_ctx[i] = sp;
704 		paca_ptrs[i]->mc_kstack = sp + THREAD_SIZE;
705 	}
706 
707 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
708 		patch_exception(0x040, exc_debug_debug_book3e);
709 }
710 #endif
711 
712 /*
713  * Stack space used when we detect a bad kernel stack pointer, and
714  * early in SMP boots before relocation is enabled. Exclusive emergency
715  * stack for machine checks.
716  */
717 void __init emergency_stack_init(void)
718 {
719 	u64 limit, mce_limit;
720 	unsigned int i;
721 
722 	/*
723 	 * Emergency stacks must be under 256MB, we cannot afford to take
724 	 * SLB misses on them. The ABI also requires them to be 128-byte
725 	 * aligned.
726 	 *
727 	 * Since we use these as temporary stacks during secondary CPU
728 	 * bringup, machine check, system reset, and HMI, we need to get
729 	 * at them in real mode. This means they must also be within the RMO
730 	 * region.
731 	 *
732 	 * The IRQ stacks allocated elsewhere in this file are zeroed and
733 	 * initialized in kernel/irq.c. These are initialized here in order
734 	 * to have emergency stacks available as early as possible.
735 	 */
736 	limit = mce_limit = min(ppc64_bolted_size(), ppc64_rma_size);
737 
738 	/*
739 	 * Machine check on pseries calls rtas, but can't use the static
740 	 * rtas_args due to a machine check hitting while the lock is held.
741 	 * rtas args have to be under 4GB, so the machine check stack is
742 	 * limited to 4GB so args can be put on stack.
743 	 */
744 	if (firmware_has_feature(FW_FEATURE_LPAR) && mce_limit > SZ_4G)
745 		mce_limit = SZ_4G;
746 
747 	for_each_possible_cpu(i) {
748 		paca_ptrs[i]->emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
749 
750 #ifdef CONFIG_PPC_BOOK3S_64
751 		/* emergency stack for NMI exception handling. */
752 		paca_ptrs[i]->nmi_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE;
753 
754 		/* emergency stack for machine check exception handling. */
755 		paca_ptrs[i]->mc_emergency_sp = alloc_stack(mce_limit, i) + THREAD_SIZE;
756 #endif
757 	}
758 }
759 
760 #ifdef CONFIG_SMP
761 /**
762  * pcpu_alloc_bootmem - NUMA friendly alloc_bootmem wrapper for percpu
763  * @cpu: cpu to allocate for
764  * @size: size allocation in bytes
765  * @align: alignment
766  *
767  * Allocate @size bytes aligned at @align for cpu @cpu.  This wrapper
768  * does the right thing for NUMA regardless of the current
769  * configuration.
770  *
771  * RETURNS:
772  * Pointer to the allocated area on success, NULL on failure.
773  */
774 static void * __init pcpu_alloc_bootmem(unsigned int cpu, size_t size,
775 					size_t align)
776 {
777 	const unsigned long goal = __pa(MAX_DMA_ADDRESS);
778 #ifdef CONFIG_NEED_MULTIPLE_NODES
779 	int node = early_cpu_to_node(cpu);
780 	void *ptr;
781 
782 	if (!node_online(node) || !NODE_DATA(node)) {
783 		ptr = memblock_alloc_from(size, align, goal);
784 		pr_info("cpu %d has no node %d or node-local memory\n",
785 			cpu, node);
786 		pr_debug("per cpu data for cpu%d %lu bytes at %016lx\n",
787 			 cpu, size, __pa(ptr));
788 	} else {
789 		ptr = memblock_alloc_try_nid(size, align, goal,
790 					     MEMBLOCK_ALLOC_ACCESSIBLE, node);
791 		pr_debug("per cpu data for cpu%d %lu bytes on node%d at "
792 			 "%016lx\n", cpu, size, node, __pa(ptr));
793 	}
794 	return ptr;
795 #else
796 	return memblock_alloc_from(size, align, goal);
797 #endif
798 }
799 
800 static void __init pcpu_free_bootmem(void *ptr, size_t size)
801 {
802 	memblock_free(__pa(ptr), size);
803 }
804 
805 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
806 {
807 	if (early_cpu_to_node(from) == early_cpu_to_node(to))
808 		return LOCAL_DISTANCE;
809 	else
810 		return REMOTE_DISTANCE;
811 }
812 
813 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
814 EXPORT_SYMBOL(__per_cpu_offset);
815 
816 static void __init pcpu_populate_pte(unsigned long addr)
817 {
818 	pgd_t *pgd = pgd_offset_k(addr);
819 	p4d_t *p4d;
820 	pud_t *pud;
821 	pmd_t *pmd;
822 
823 	p4d = p4d_offset(pgd, addr);
824 	if (p4d_none(*p4d)) {
825 		pud_t *new;
826 
827 		new = memblock_alloc(PUD_TABLE_SIZE, PUD_TABLE_SIZE);
828 		if (!new)
829 			goto err_alloc;
830 		p4d_populate(&init_mm, p4d, new);
831 	}
832 
833 	pud = pud_offset(p4d, addr);
834 	if (pud_none(*pud)) {
835 		pmd_t *new;
836 
837 		new = memblock_alloc(PMD_TABLE_SIZE, PMD_TABLE_SIZE);
838 		if (!new)
839 			goto err_alloc;
840 		pud_populate(&init_mm, pud, new);
841 	}
842 
843 	pmd = pmd_offset(pud, addr);
844 	if (!pmd_present(*pmd)) {
845 		pte_t *new;
846 
847 		new = memblock_alloc(PTE_TABLE_SIZE, PTE_TABLE_SIZE);
848 		if (!new)
849 			goto err_alloc;
850 		pmd_populate_kernel(&init_mm, pmd, new);
851 	}
852 
853 	return;
854 
855 err_alloc:
856 	panic("%s: Failed to allocate %lu bytes align=%lx from=%lx\n",
857 	      __func__, PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
858 }
859 
860 
861 void __init setup_per_cpu_areas(void)
862 {
863 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
864 	size_t atom_size;
865 	unsigned long delta;
866 	unsigned int cpu;
867 	int rc = -EINVAL;
868 
869 	/*
870 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
871 	 * to group units.  For larger mappings, use 1M atom which
872 	 * should be large enough to contain a number of units.
873 	 */
874 	if (mmu_linear_psize == MMU_PAGE_4K)
875 		atom_size = PAGE_SIZE;
876 	else
877 		atom_size = 1 << 20;
878 
879 	if (pcpu_chosen_fc != PCPU_FC_PAGE) {
880 		rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
881 					    pcpu_alloc_bootmem, pcpu_free_bootmem);
882 		if (rc)
883 			pr_warn("PERCPU: %s allocator failed (%d), "
884 				"falling back to page size\n",
885 				pcpu_fc_names[pcpu_chosen_fc], rc);
886 	}
887 
888 	if (rc < 0)
889 		rc = pcpu_page_first_chunk(0, pcpu_alloc_bootmem, pcpu_free_bootmem,
890 					   pcpu_populate_pte);
891 	if (rc < 0)
892 		panic("cannot initialize percpu area (err=%d)", rc);
893 
894 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
895 	for_each_possible_cpu(cpu) {
896                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
897 		paca_ptrs[cpu]->data_offset = __per_cpu_offset[cpu];
898 	}
899 }
900 #endif
901 
902 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
903 unsigned long memory_block_size_bytes(void)
904 {
905 	if (ppc_md.memory_block_size)
906 		return ppc_md.memory_block_size();
907 
908 	return MIN_MEMORY_BLOCK_SIZE;
909 }
910 #endif
911 
912 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
913 struct ppc_pci_io ppc_pci_io;
914 EXPORT_SYMBOL(ppc_pci_io);
915 #endif
916 
917 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
918 u64 hw_nmi_get_sample_period(int watchdog_thresh)
919 {
920 	return ppc_proc_freq * watchdog_thresh;
921 }
922 #endif
923 
924 /*
925  * The perf based hardlockup detector breaks PMU event based branches, so
926  * disable it by default. Book3S has a soft-nmi hardlockup detector based
927  * on the decrementer interrupt, so it does not suffer from this problem.
928  *
929  * It is likely to get false positives in VM guests, so disable it there
930  * by default too.
931  */
932 static int __init disable_hardlockup_detector(void)
933 {
934 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF
935 	hardlockup_detector_disable();
936 #else
937 	if (firmware_has_feature(FW_FEATURE_LPAR))
938 		hardlockup_detector_disable();
939 #endif
940 
941 	return 0;
942 }
943 early_initcall(disable_hardlockup_detector);
944 
945 #ifdef CONFIG_PPC_BOOK3S_64
946 static enum l1d_flush_type enabled_flush_types;
947 static void *l1d_flush_fallback_area;
948 static bool no_rfi_flush;
949 static bool no_entry_flush;
950 static bool no_uaccess_flush;
951 bool rfi_flush;
952 bool entry_flush;
953 bool uaccess_flush;
954 DEFINE_STATIC_KEY_FALSE(uaccess_flush_key);
955 EXPORT_SYMBOL(uaccess_flush_key);
956 
957 static int __init handle_no_rfi_flush(char *p)
958 {
959 	pr_info("rfi-flush: disabled on command line.");
960 	no_rfi_flush = true;
961 	return 0;
962 }
963 early_param("no_rfi_flush", handle_no_rfi_flush);
964 
965 static int __init handle_no_entry_flush(char *p)
966 {
967 	pr_info("entry-flush: disabled on command line.");
968 	no_entry_flush = true;
969 	return 0;
970 }
971 early_param("no_entry_flush", handle_no_entry_flush);
972 
973 static int __init handle_no_uaccess_flush(char *p)
974 {
975 	pr_info("uaccess-flush: disabled on command line.");
976 	no_uaccess_flush = true;
977 	return 0;
978 }
979 early_param("no_uaccess_flush", handle_no_uaccess_flush);
980 
981 /*
982  * The RFI flush is not KPTI, but because users will see doco that says to use
983  * nopti we hijack that option here to also disable the RFI flush.
984  */
985 static int __init handle_no_pti(char *p)
986 {
987 	pr_info("rfi-flush: disabling due to 'nopti' on command line.\n");
988 	handle_no_rfi_flush(NULL);
989 	return 0;
990 }
991 early_param("nopti", handle_no_pti);
992 
993 static void do_nothing(void *unused)
994 {
995 	/*
996 	 * We don't need to do the flush explicitly, just enter+exit kernel is
997 	 * sufficient, the RFI exit handlers will do the right thing.
998 	 */
999 }
1000 
1001 void rfi_flush_enable(bool enable)
1002 {
1003 	if (enable) {
1004 		do_rfi_flush_fixups(enabled_flush_types);
1005 		on_each_cpu(do_nothing, NULL, 1);
1006 	} else
1007 		do_rfi_flush_fixups(L1D_FLUSH_NONE);
1008 
1009 	rfi_flush = enable;
1010 }
1011 
1012 static void entry_flush_enable(bool enable)
1013 {
1014 	if (enable) {
1015 		do_entry_flush_fixups(enabled_flush_types);
1016 		on_each_cpu(do_nothing, NULL, 1);
1017 	} else {
1018 		do_entry_flush_fixups(L1D_FLUSH_NONE);
1019 	}
1020 
1021 	entry_flush = enable;
1022 }
1023 
1024 static void uaccess_flush_enable(bool enable)
1025 {
1026 	if (enable) {
1027 		do_uaccess_flush_fixups(enabled_flush_types);
1028 		static_branch_enable(&uaccess_flush_key);
1029 		on_each_cpu(do_nothing, NULL, 1);
1030 	} else {
1031 		static_branch_disable(&uaccess_flush_key);
1032 		do_uaccess_flush_fixups(L1D_FLUSH_NONE);
1033 	}
1034 
1035 	uaccess_flush = enable;
1036 }
1037 
1038 static void __ref init_fallback_flush(void)
1039 {
1040 	u64 l1d_size, limit;
1041 	int cpu;
1042 
1043 	/* Only allocate the fallback flush area once (at boot time). */
1044 	if (l1d_flush_fallback_area)
1045 		return;
1046 
1047 	l1d_size = ppc64_caches.l1d.size;
1048 
1049 	/*
1050 	 * If there is no d-cache-size property in the device tree, l1d_size
1051 	 * could be zero. That leads to the loop in the asm wrapping around to
1052 	 * 2^64-1, and then walking off the end of the fallback area and
1053 	 * eventually causing a page fault which is fatal. Just default to
1054 	 * something vaguely sane.
1055 	 */
1056 	if (!l1d_size)
1057 		l1d_size = (64 * 1024);
1058 
1059 	limit = min(ppc64_bolted_size(), ppc64_rma_size);
1060 
1061 	/*
1062 	 * Align to L1d size, and size it at 2x L1d size, to catch possible
1063 	 * hardware prefetch runoff. We don't have a recipe for load patterns to
1064 	 * reliably avoid the prefetcher.
1065 	 */
1066 	l1d_flush_fallback_area = memblock_alloc_try_nid(l1d_size * 2,
1067 						l1d_size, MEMBLOCK_LOW_LIMIT,
1068 						limit, NUMA_NO_NODE);
1069 	if (!l1d_flush_fallback_area)
1070 		panic("%s: Failed to allocate %llu bytes align=0x%llx max_addr=%pa\n",
1071 		      __func__, l1d_size * 2, l1d_size, &limit);
1072 
1073 
1074 	for_each_possible_cpu(cpu) {
1075 		struct paca_struct *paca = paca_ptrs[cpu];
1076 		paca->rfi_flush_fallback_area = l1d_flush_fallback_area;
1077 		paca->l1d_flush_size = l1d_size;
1078 	}
1079 }
1080 
1081 void setup_rfi_flush(enum l1d_flush_type types, bool enable)
1082 {
1083 	if (types & L1D_FLUSH_FALLBACK) {
1084 		pr_info("rfi-flush: fallback displacement flush available\n");
1085 		init_fallback_flush();
1086 	}
1087 
1088 	if (types & L1D_FLUSH_ORI)
1089 		pr_info("rfi-flush: ori type flush available\n");
1090 
1091 	if (types & L1D_FLUSH_MTTRIG)
1092 		pr_info("rfi-flush: mttrig type flush available\n");
1093 
1094 	enabled_flush_types = types;
1095 
1096 	if (!cpu_mitigations_off() && !no_rfi_flush)
1097 		rfi_flush_enable(enable);
1098 }
1099 
1100 void setup_entry_flush(bool enable)
1101 {
1102 	if (cpu_mitigations_off())
1103 		return;
1104 
1105 	if (!no_entry_flush)
1106 		entry_flush_enable(enable);
1107 }
1108 
1109 void setup_uaccess_flush(bool enable)
1110 {
1111 	if (cpu_mitigations_off())
1112 		return;
1113 
1114 	if (!no_uaccess_flush)
1115 		uaccess_flush_enable(enable);
1116 }
1117 
1118 #ifdef CONFIG_DEBUG_FS
1119 static int rfi_flush_set(void *data, u64 val)
1120 {
1121 	bool enable;
1122 
1123 	if (val == 1)
1124 		enable = true;
1125 	else if (val == 0)
1126 		enable = false;
1127 	else
1128 		return -EINVAL;
1129 
1130 	/* Only do anything if we're changing state */
1131 	if (enable != rfi_flush)
1132 		rfi_flush_enable(enable);
1133 
1134 	return 0;
1135 }
1136 
1137 static int rfi_flush_get(void *data, u64 *val)
1138 {
1139 	*val = rfi_flush ? 1 : 0;
1140 	return 0;
1141 }
1142 
1143 DEFINE_SIMPLE_ATTRIBUTE(fops_rfi_flush, rfi_flush_get, rfi_flush_set, "%llu\n");
1144 
1145 static int entry_flush_set(void *data, u64 val)
1146 {
1147 	bool enable;
1148 
1149 	if (val == 1)
1150 		enable = true;
1151 	else if (val == 0)
1152 		enable = false;
1153 	else
1154 		return -EINVAL;
1155 
1156 	/* Only do anything if we're changing state */
1157 	if (enable != entry_flush)
1158 		entry_flush_enable(enable);
1159 
1160 	return 0;
1161 }
1162 
1163 static int entry_flush_get(void *data, u64 *val)
1164 {
1165 	*val = entry_flush ? 1 : 0;
1166 	return 0;
1167 }
1168 
1169 DEFINE_SIMPLE_ATTRIBUTE(fops_entry_flush, entry_flush_get, entry_flush_set, "%llu\n");
1170 
1171 static int uaccess_flush_set(void *data, u64 val)
1172 {
1173 	bool enable;
1174 
1175 	if (val == 1)
1176 		enable = true;
1177 	else if (val == 0)
1178 		enable = false;
1179 	else
1180 		return -EINVAL;
1181 
1182 	/* Only do anything if we're changing state */
1183 	if (enable != uaccess_flush)
1184 		uaccess_flush_enable(enable);
1185 
1186 	return 0;
1187 }
1188 
1189 static int uaccess_flush_get(void *data, u64 *val)
1190 {
1191 	*val = uaccess_flush ? 1 : 0;
1192 	return 0;
1193 }
1194 
1195 DEFINE_SIMPLE_ATTRIBUTE(fops_uaccess_flush, uaccess_flush_get, uaccess_flush_set, "%llu\n");
1196 
1197 static __init int rfi_flush_debugfs_init(void)
1198 {
1199 	debugfs_create_file("rfi_flush", 0600, powerpc_debugfs_root, NULL, &fops_rfi_flush);
1200 	debugfs_create_file("entry_flush", 0600, powerpc_debugfs_root, NULL, &fops_entry_flush);
1201 	debugfs_create_file("uaccess_flush", 0600, powerpc_debugfs_root, NULL, &fops_uaccess_flush);
1202 	return 0;
1203 }
1204 device_initcall(rfi_flush_debugfs_init);
1205 #endif
1206 #endif /* CONFIG_PPC_BOOK3S_64 */
1207